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 <sys/vm.h> 44 #include <vm/seg_vn.h> 45 #include <vm/pvn.h> 46 #include <vm/as.h> 47 #include <sys/mman.h> 48 #include <sys/pathname.h> 49 #include <sys/cmn_err.h> 50 #include <sys/errno.h> 51 #include <sys/unistd.h> 52 #include <sys/zfs_vfsops.h> 53 #include <sys/zfs_dir.h> 54 #include <sys/zfs_acl.h> 55 #include <sys/zfs_ioctl.h> 56 #include <sys/fs/zfs.h> 57 #include <sys/dmu.h> 58 #include <sys/spa.h> 59 #include <sys/txg.h> 60 #include <sys/dbuf.h> 61 #include <sys/zap.h> 62 #include <sys/dirent.h> 63 #include <sys/policy.h> 64 #include <sys/sunddi.h> 65 #include <sys/filio.h> 66 #include "fs/fs_subr.h" 67 #include <sys/zfs_ctldir.h> 68 #include <sys/dnlc.h> 69 #include <sys/zfs_rlock.h> 70 71 /* 72 * Programming rules. 73 * 74 * Each vnode op performs some logical unit of work. To do this, the ZPL must 75 * properly lock its in-core state, create a DMU transaction, do the work, 76 * record this work in the intent log (ZIL), commit the DMU transaction, 77 * and wait the the intent log to commit if it's is a synchronous operation. 78 * Morover, the vnode ops must work in both normal and log replay context. 79 * The ordering of events is important to avoid deadlocks and references 80 * to freed memory. The example below illustrates the following Big Rules: 81 * 82 * (1) A check must be made in each zfs thread for a mounted file system. 83 * This is done avoiding races using ZFS_ENTER(zfsvfs). 84 * A ZFS_EXIT(zfsvfs) is needed before all returns. 85 * 86 * (2) VN_RELE() should always be the last thing except for zil_commit() 87 * (if necessary) and ZFS_EXIT(). This is for 3 reasons: 88 * First, if it's the last reference, the vnode/znode 89 * can be freed, so the zp may point to freed memory. Second, the last 90 * reference will call zfs_zinactive(), which may induce a lot of work -- 91 * pushing cached pages (which acquires range locks) and syncing out 92 * cached atime changes. Third, zfs_zinactive() may require a new tx, 93 * which could deadlock the system if you were already holding one. 94 * 95 * (3) All range locks must be grabbed before calling dmu_tx_assign(), 96 * as they can span dmu_tx_assign() calls. 97 * 98 * (4) Always pass zfsvfs->z_assign as the second argument to dmu_tx_assign(). 99 * In normal operation, this will be TXG_NOWAIT. During ZIL replay, 100 * it will be a specific txg. Either way, dmu_tx_assign() never blocks. 101 * This is critical because we don't want to block while holding locks. 102 * Note, in particular, that if a lock is sometimes acquired before 103 * the tx assigns, and sometimes after (e.g. z_lock), then failing to 104 * use a non-blocking assign can deadlock the system. The scenario: 105 * 106 * Thread A has grabbed a lock before calling dmu_tx_assign(). 107 * Thread B is in an already-assigned tx, and blocks for this lock. 108 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open() 109 * forever, because the previous txg can't quiesce until B's tx commits. 110 * 111 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT, 112 * then drop all locks, call dmu_tx_wait(), and try again. 113 * 114 * (5) If the operation succeeded, generate the intent log entry for it 115 * before dropping locks. This ensures that the ordering of events 116 * in the intent log matches the order in which they actually occurred. 117 * 118 * (6) At the end of each vnode op, the DMU tx must always commit, 119 * regardless of whether there were any errors. 120 * 121 * (7) After dropping all locks, invoke zil_commit(zilog, seq, foid) 122 * to ensure that synchronous semantics are provided when necessary. 123 * 124 * In general, this is how things should be ordered in each vnode op: 125 * 126 * ZFS_ENTER(zfsvfs); // exit if unmounted 127 * top: 128 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD()) 129 * rw_enter(...); // grab any other locks you need 130 * tx = dmu_tx_create(...); // get DMU tx 131 * dmu_tx_hold_*(); // hold each object you might modify 132 * error = dmu_tx_assign(tx, zfsvfs->z_assign); // try to assign 133 * if (error) { 134 * rw_exit(...); // drop locks 135 * zfs_dirent_unlock(dl); // unlock directory entry 136 * VN_RELE(...); // release held vnodes 137 * if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 138 * dmu_tx_wait(tx); 139 * dmu_tx_abort(tx); 140 * goto top; 141 * } 142 * dmu_tx_abort(tx); // abort DMU tx 143 * ZFS_EXIT(zfsvfs); // finished in zfs 144 * return (error); // really out of space 145 * } 146 * error = do_real_work(); // do whatever this VOP does 147 * if (error == 0) 148 * zfs_log_*(...); // on success, make ZIL entry 149 * dmu_tx_commit(tx); // commit DMU tx -- error or not 150 * rw_exit(...); // drop locks 151 * zfs_dirent_unlock(dl); // unlock directory entry 152 * VN_RELE(...); // release held vnodes 153 * zil_commit(zilog, seq, foid); // synchronous when necessary 154 * ZFS_EXIT(zfsvfs); // finished in zfs 155 * return (error); // done, report error 156 */ 157 158 /* ARGSUSED */ 159 static int 160 zfs_open(vnode_t **vpp, int flag, cred_t *cr) 161 { 162 return (0); 163 } 164 165 /* ARGSUSED */ 166 static int 167 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr) 168 { 169 /* 170 * Clean up any locks held by this process on the vp. 171 */ 172 cleanlocks(vp, ddi_get_pid(), 0); 173 cleanshares(vp, ddi_get_pid()); 174 175 return (0); 176 } 177 178 /* 179 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and 180 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter. 181 */ 182 static int 183 zfs_holey(vnode_t *vp, int cmd, offset_t *off) 184 { 185 znode_t *zp = VTOZ(vp); 186 uint64_t noff = (uint64_t)*off; /* new offset */ 187 uint64_t file_sz; 188 int error; 189 boolean_t hole; 190 191 file_sz = zp->z_phys->zp_size; 192 if (noff >= file_sz) { 193 return (ENXIO); 194 } 195 196 if (cmd == _FIO_SEEK_HOLE) 197 hole = B_TRUE; 198 else 199 hole = B_FALSE; 200 201 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff); 202 203 /* end of file? */ 204 if ((error == ESRCH) || (noff > file_sz)) { 205 /* 206 * Handle the virtual hole at the end of file. 207 */ 208 if (hole) { 209 *off = file_sz; 210 return (0); 211 } 212 return (ENXIO); 213 } 214 215 if (noff < *off) 216 return (error); 217 *off = noff; 218 return (error); 219 } 220 221 /* ARGSUSED */ 222 static int 223 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred, 224 int *rvalp) 225 { 226 offset_t off; 227 int error; 228 zfsvfs_t *zfsvfs; 229 230 switch (com) { 231 case _FIOFFS: 232 return (zfs_sync(vp->v_vfsp, 0, cred)); 233 234 /* 235 * The following two ioctls are used by bfu. Faking out, 236 * necessary to avoid bfu errors. 237 */ 238 case _FIOGDIO: 239 case _FIOSDIO: 240 return (0); 241 242 case _FIO_SEEK_DATA: 243 case _FIO_SEEK_HOLE: 244 if (ddi_copyin((void *)data, &off, sizeof (off), flag)) 245 return (EFAULT); 246 247 zfsvfs = VTOZ(vp)->z_zfsvfs; 248 ZFS_ENTER(zfsvfs); 249 250 /* offset parameter is in/out */ 251 error = zfs_holey(vp, com, &off); 252 ZFS_EXIT(zfsvfs); 253 if (error) 254 return (error); 255 if (ddi_copyout(&off, (void *)data, sizeof (off), flag)) 256 return (EFAULT); 257 return (0); 258 } 259 return (ENOTTY); 260 } 261 262 /* 263 * When a file is memory mapped, we must keep the IO data synchronized 264 * between the DMU cache and the memory mapped pages. What this means: 265 * 266 * On Write: If we find a memory mapped page, we write to *both* 267 * the page and the dmu buffer. 268 * 269 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 270 * the file is memory mapped. 271 */ 272 static int 273 mappedwrite(vnode_t *vp, uint64_t woff, int nbytes, uio_t *uio, dmu_tx_t *tx) 274 { 275 znode_t *zp = VTOZ(vp); 276 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 277 int64_t start, off; 278 int len = nbytes; 279 int error = 0; 280 281 start = uio->uio_loffset; 282 off = start & PAGEOFFSET; 283 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 284 page_t *pp; 285 uint64_t bytes = MIN(PAGESIZE - off, len); 286 287 /* 288 * We don't want a new page to "appear" in the middle of 289 * the file update (because it may not get the write 290 * update data), so we grab a lock to block 291 * zfs_getpage(). 292 */ 293 rw_enter(&zp->z_map_lock, RW_WRITER); 294 if (pp = page_lookup(vp, start, SE_SHARED)) { 295 caddr_t va; 296 297 rw_exit(&zp->z_map_lock); 298 va = ppmapin(pp, PROT_READ | PROT_WRITE, (caddr_t)-1L); 299 error = uiomove(va+off, bytes, UIO_WRITE, uio); 300 if (error == 0) { 301 dmu_write(zfsvfs->z_os, zp->z_id, 302 woff, bytes, va+off, tx); 303 } 304 ppmapout(va); 305 page_unlock(pp); 306 } else { 307 error = dmu_write_uio(zfsvfs->z_os, zp->z_id, 308 woff, bytes, uio, tx); 309 rw_exit(&zp->z_map_lock); 310 } 311 len -= bytes; 312 woff += bytes; 313 off = 0; 314 if (error) 315 break; 316 } 317 return (error); 318 } 319 320 /* 321 * When a file is memory mapped, we must keep the IO data synchronized 322 * between the DMU cache and the memory mapped pages. What this means: 323 * 324 * On Read: We "read" preferentially from memory mapped pages, 325 * else we default from the dmu buffer. 326 * 327 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 328 * the file is memory mapped. 329 */ 330 static int 331 mappedread(vnode_t *vp, char *addr, int nbytes, uio_t *uio) 332 { 333 int64_t start, off, bytes; 334 int len = nbytes; 335 int error = 0; 336 337 start = uio->uio_loffset; 338 off = start & PAGEOFFSET; 339 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 340 page_t *pp; 341 342 bytes = MIN(PAGESIZE - off, len); 343 if (pp = page_lookup(vp, start, SE_SHARED)) { 344 caddr_t va; 345 346 va = ppmapin(pp, PROT_READ, (caddr_t)-1L); 347 error = uiomove(va + off, bytes, UIO_READ, uio); 348 ppmapout(va); 349 page_unlock(pp); 350 } else { 351 /* XXX use dmu_read here? */ 352 error = uiomove(addr, bytes, UIO_READ, uio); 353 } 354 len -= bytes; 355 addr += bytes; 356 off = 0; 357 if (error) 358 break; 359 } 360 return (error); 361 } 362 363 uint_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ 364 365 /* 366 * Read bytes from specified file into supplied buffer. 367 * 368 * IN: vp - vnode of file to be read from. 369 * uio - structure supplying read location, range info, 370 * and return buffer. 371 * ioflag - SYNC flags; used to provide FRSYNC semantics. 372 * cr - credentials of caller. 373 * 374 * OUT: uio - updated offset and range, buffer filled. 375 * 376 * RETURN: 0 if success 377 * error code if failure 378 * 379 * Side Effects: 380 * vp - atime updated if byte count > 0 381 */ 382 /* ARGSUSED */ 383 static int 384 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 385 { 386 znode_t *zp = VTOZ(vp); 387 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 388 uint64_t delta; 389 ssize_t n, size, cnt, ndone; 390 int error, i, numbufs; 391 dmu_buf_t *dbp, **dbpp; 392 rl_t *rl; 393 394 ZFS_ENTER(zfsvfs); 395 396 /* 397 * Validate file offset 398 */ 399 if (uio->uio_loffset < (offset_t)0) { 400 ZFS_EXIT(zfsvfs); 401 return (EINVAL); 402 } 403 404 /* 405 * Fasttrack empty reads 406 */ 407 if (uio->uio_resid == 0) { 408 ZFS_EXIT(zfsvfs); 409 return (0); 410 } 411 412 /* 413 * Check for mandatory locks 414 */ 415 if (MANDMODE((mode_t)zp->z_phys->zp_mode)) { 416 if (error = chklock(vp, FREAD, 417 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { 418 ZFS_EXIT(zfsvfs); 419 return (error); 420 } 421 } 422 423 /* 424 * If we're in FRSYNC mode, sync out this znode before reading it. 425 */ 426 if (ioflag & FRSYNC) 427 zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id); 428 429 /* 430 * Lock the range against changes. 431 */ 432 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER); 433 434 /* 435 * If we are reading past end-of-file we can skip 436 * to the end; but we might still need to set atime. 437 */ 438 if (uio->uio_loffset >= zp->z_phys->zp_size) { 439 cnt = 0; 440 error = 0; 441 goto out; 442 } 443 444 cnt = MIN(uio->uio_resid, zp->z_phys->zp_size - uio->uio_loffset); 445 446 for (ndone = 0; ndone < cnt; ndone += zfs_read_chunk_size) { 447 ASSERT(uio->uio_loffset < zp->z_phys->zp_size); 448 n = MIN(zfs_read_chunk_size, 449 zp->z_phys->zp_size - uio->uio_loffset); 450 n = MIN(n, cnt); 451 error = dmu_buf_hold_array_by_bonus(zp->z_dbuf, 452 uio->uio_loffset, n, TRUE, FTAG, &numbufs, &dbpp); 453 if (error) 454 goto out; 455 /* 456 * Compute the adjustment to align the dmu buffers 457 * with the uio buffer. 458 */ 459 delta = uio->uio_loffset - dbpp[0]->db_offset; 460 461 for (i = 0; i < numbufs; i++) { 462 if (n < 0) 463 break; 464 dbp = dbpp[i]; 465 size = dbp->db_size - delta; 466 /* 467 * XXX -- this is correct, but may be suboptimal. 468 * If the pages are all clean, we don't need to 469 * go through mappedread(). Maybe the VMODSORT 470 * stuff can help us here. 471 */ 472 if (vn_has_cached_data(vp)) { 473 error = mappedread(vp, (caddr_t)dbp->db_data + 474 delta, (n < size ? n : size), uio); 475 } else { 476 error = uiomove((caddr_t)dbp->db_data + delta, 477 (n < size ? n : size), UIO_READ, uio); 478 } 479 if (error) { 480 dmu_buf_rele_array(dbpp, numbufs, FTAG); 481 goto out; 482 } 483 n -= dbp->db_size; 484 if (delta) { 485 n += delta; 486 delta = 0; 487 } 488 } 489 dmu_buf_rele_array(dbpp, numbufs, FTAG); 490 } 491 out: 492 zfs_range_unlock(rl); 493 494 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 495 ZFS_EXIT(zfsvfs); 496 return (error); 497 } 498 499 /* 500 * Fault in the pages of the first n bytes specified by the uio structure. 501 * 1 byte in each page is touched and the uio struct is unmodified. 502 * Any error will exit this routine as this is only a best 503 * attempt to get the pages resident. This is a copy of ufs_trans_touch(). 504 */ 505 static void 506 zfs_prefault_write(ssize_t n, struct uio *uio) 507 { 508 struct iovec *iov; 509 ulong_t cnt, incr; 510 caddr_t p; 511 uint8_t tmp; 512 513 iov = uio->uio_iov; 514 515 while (n) { 516 cnt = MIN(iov->iov_len, n); 517 if (cnt == 0) { 518 /* empty iov entry */ 519 iov++; 520 continue; 521 } 522 n -= cnt; 523 /* 524 * touch each page in this segment. 525 */ 526 p = iov->iov_base; 527 while (cnt) { 528 switch (uio->uio_segflg) { 529 case UIO_USERSPACE: 530 case UIO_USERISPACE: 531 if (fuword8(p, &tmp)) 532 return; 533 break; 534 case UIO_SYSSPACE: 535 if (kcopy(p, &tmp, 1)) 536 return; 537 break; 538 } 539 incr = MIN(cnt, PAGESIZE); 540 p += incr; 541 cnt -= incr; 542 } 543 /* 544 * touch the last byte in case it straddles a page. 545 */ 546 p--; 547 switch (uio->uio_segflg) { 548 case UIO_USERSPACE: 549 case UIO_USERISPACE: 550 if (fuword8(p, &tmp)) 551 return; 552 break; 553 case UIO_SYSSPACE: 554 if (kcopy(p, &tmp, 1)) 555 return; 556 break; 557 } 558 iov++; 559 } 560 } 561 562 /* 563 * Write the bytes to a file. 564 * 565 * IN: vp - vnode of file to be written to. 566 * uio - structure supplying write location, range info, 567 * and data buffer. 568 * ioflag - FAPPEND flag set if in append mode. 569 * cr - credentials of caller. 570 * 571 * OUT: uio - updated offset and range. 572 * 573 * RETURN: 0 if success 574 * error code if failure 575 * 576 * Timestamps: 577 * vp - ctime|mtime updated if byte count > 0 578 */ 579 /* ARGSUSED */ 580 static int 581 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 582 { 583 znode_t *zp = VTOZ(vp); 584 rlim64_t limit = uio->uio_llimit; 585 ssize_t start_resid = uio->uio_resid; 586 ssize_t tx_bytes; 587 uint64_t end_size; 588 dmu_tx_t *tx; 589 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 590 zilog_t *zilog = zfsvfs->z_log; 591 offset_t woff; 592 ssize_t n, nbytes; 593 rl_t *rl; 594 int max_blksz = zfsvfs->z_max_blksz; 595 int error; 596 597 /* 598 * Fasttrack empty write 599 */ 600 n = start_resid; 601 if (n == 0) 602 return (0); 603 604 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 605 limit = MAXOFFSET_T; 606 607 ZFS_ENTER(zfsvfs); 608 609 /* 610 * Pre-fault the pages to ensure slow (eg NFS) pages 611 * don't hold up txg. 612 */ 613 zfs_prefault_write(n, uio); 614 615 /* 616 * If in append mode, set the io offset pointer to eof. 617 */ 618 if (ioflag & FAPPEND) { 619 /* 620 * Range lock for a file append: 621 * The value for the start of range will be determined by 622 * zfs_range_lock() (to guarantee append semantics). 623 * If this write will cause the block size to increase, 624 * zfs_range_lock() will lock the entire file, so we must 625 * later reduce the range after we grow the block size. 626 */ 627 rl = zfs_range_lock(zp, 0, n, RL_APPEND); 628 if (rl->r_len == UINT64_MAX) { 629 /* overlocked, zp_size can't change */ 630 woff = uio->uio_loffset = zp->z_phys->zp_size; 631 } else { 632 woff = uio->uio_loffset = rl->r_off; 633 } 634 } else { 635 woff = uio->uio_loffset; 636 /* 637 * Validate file offset 638 */ 639 if (woff < 0) { 640 ZFS_EXIT(zfsvfs); 641 return (EINVAL); 642 } 643 644 /* 645 * If we need to grow the block size then zfs_range_lock() 646 * will lock a wider range than we request here. 647 * Later after growing the block size we reduce the range. 648 */ 649 rl = zfs_range_lock(zp, woff, n, RL_WRITER); 650 } 651 652 if (woff >= limit) { 653 error = EFBIG; 654 goto no_tx_done; 655 } 656 657 if ((woff + n) > limit || woff > (limit - n)) 658 n = limit - woff; 659 660 /* 661 * Check for mandatory locks 662 */ 663 if (MANDMODE((mode_t)zp->z_phys->zp_mode) && 664 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) 665 goto no_tx_done; 666 end_size = MAX(zp->z_phys->zp_size, woff + n); 667 top: 668 tx = dmu_tx_create(zfsvfs->z_os); 669 dmu_tx_hold_bonus(tx, zp->z_id); 670 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 671 error = dmu_tx_assign(tx, zfsvfs->z_assign); 672 if (error) { 673 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 674 dmu_tx_wait(tx); 675 dmu_tx_abort(tx); 676 goto top; 677 } 678 dmu_tx_abort(tx); 679 goto no_tx_done; 680 } 681 682 /* 683 * If zfs_range_lock() over-locked we grow the blocksize 684 * and then reduce the lock range. 685 */ 686 if (rl->r_len == UINT64_MAX) { 687 uint64_t new_blksz; 688 689 if (zp->z_blksz > max_blksz) { 690 ASSERT(!ISP2(zp->z_blksz)); 691 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE); 692 } else { 693 new_blksz = MIN(end_size, max_blksz); 694 } 695 zfs_grow_blocksize(zp, new_blksz, tx); 696 zfs_range_reduce(rl, woff, n); 697 } 698 699 /* 700 * The file data does not fit in the znode "cache", so we 701 * will be writing to the file block data buffers. 702 * Each buffer will be written in a separate transaction; 703 * this keeps the intent log records small and allows us 704 * to do more fine-grained space accounting. 705 */ 706 while (n > 0) { 707 /* 708 * XXX - should we really limit each write to z_max_blksz? 709 * Perhaps we should use SPA_MAXBLOCKSIZE chunks? 710 */ 711 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz)); 712 rw_enter(&zp->z_map_lock, RW_READER); 713 714 tx_bytes = uio->uio_resid; 715 if (vn_has_cached_data(vp)) { 716 rw_exit(&zp->z_map_lock); 717 error = mappedwrite(vp, woff, nbytes, uio, tx); 718 } else { 719 error = dmu_write_uio(zfsvfs->z_os, zp->z_id, 720 woff, nbytes, uio, tx); 721 rw_exit(&zp->z_map_lock); 722 } 723 tx_bytes -= uio->uio_resid; 724 725 if (error) { 726 /* XXX - do we need to "clean up" the dmu buffer? */ 727 break; 728 } 729 730 ASSERT(tx_bytes == nbytes); 731 732 /* 733 * Clear Set-UID/Set-GID bits on successful write if not 734 * privileged and at least one of the excute bits is set. 735 * 736 * It would be nice to to this after all writes have 737 * been done, but that would still expose the ISUID/ISGID 738 * to another app after the partial write is committed. 739 */ 740 741 mutex_enter(&zp->z_acl_lock); 742 if ((zp->z_phys->zp_mode & (S_IXUSR | (S_IXUSR >> 3) | 743 (S_IXUSR >> 6))) != 0 && 744 (zp->z_phys->zp_mode & (S_ISUID | S_ISGID)) != 0 && 745 secpolicy_vnode_setid_retain(cr, 746 (zp->z_phys->zp_mode & S_ISUID) != 0 && 747 zp->z_phys->zp_uid == 0) != 0) { 748 zp->z_phys->zp_mode &= ~(S_ISUID | S_ISGID); 749 } 750 mutex_exit(&zp->z_acl_lock); 751 752 n -= nbytes; 753 if (n <= 0) 754 break; 755 756 /* 757 * We have more work ahead of us, so wrap up this transaction 758 * and start another. Exact same logic as tx_done below. 759 */ 760 while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset) { 761 dmu_buf_will_dirty(zp->z_dbuf, tx); 762 (void) atomic_cas_64(&zp->z_phys->zp_size, end_size, 763 uio->uio_loffset); 764 } 765 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 766 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, 767 ioflag, uio); 768 dmu_tx_commit(tx); 769 770 /* 771 * Start another transaction. 772 */ 773 woff = uio->uio_loffset; 774 tx = dmu_tx_create(zfsvfs->z_os); 775 dmu_tx_hold_bonus(tx, zp->z_id); 776 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 777 error = dmu_tx_assign(tx, zfsvfs->z_assign); 778 if (error) { 779 if (error == ERESTART && 780 zfsvfs->z_assign == TXG_NOWAIT) { 781 dmu_tx_wait(tx); 782 dmu_tx_abort(tx); 783 goto top; 784 } 785 dmu_tx_abort(tx); 786 goto no_tx_done; 787 } 788 } 789 790 tx_done: 791 792 if (tx_bytes != 0) { 793 /* 794 * Update the file size if it has changed; account 795 * for possible concurrent updates. 796 */ 797 while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset) { 798 dmu_buf_will_dirty(zp->z_dbuf, tx); 799 (void) atomic_cas_64(&zp->z_phys->zp_size, end_size, 800 uio->uio_loffset); 801 } 802 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 803 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, 804 ioflag, uio); 805 } 806 dmu_tx_commit(tx); 807 808 809 no_tx_done: 810 811 zfs_range_unlock(rl); 812 813 /* 814 * If we're in replay mode, or we made no progress, return error. 815 * Otherwise, it's at least a partial write, so it's successful. 816 */ 817 if (zfsvfs->z_assign >= TXG_INITIAL || uio->uio_resid == start_resid) { 818 ZFS_EXIT(zfsvfs); 819 return (error); 820 } 821 822 if (ioflag & (FSYNC | FDSYNC)) 823 zil_commit(zilog, zp->z_last_itx, zp->z_id); 824 825 ZFS_EXIT(zfsvfs); 826 return (0); 827 } 828 829 void 830 zfs_get_done(dmu_buf_t *db, void *vrl) 831 { 832 rl_t *rl = (rl_t *)vrl; 833 vnode_t *vp = ZTOV(rl->r_zp); 834 835 dmu_buf_rele(db, rl); 836 zfs_range_unlock(rl); 837 VN_RELE(vp); 838 } 839 840 /* 841 * Get data to generate a TX_WRITE intent log record. 842 */ 843 int 844 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 845 { 846 zfsvfs_t *zfsvfs = arg; 847 objset_t *os = zfsvfs->z_os; 848 znode_t *zp; 849 uint64_t off = lr->lr_offset; 850 dmu_buf_t *db; 851 rl_t *rl; 852 int dlen = lr->lr_length; /* length of user data */ 853 int error = 0; 854 855 ASSERT(dlen != 0); 856 857 /* 858 * Nothing to do if the file has been removed 859 */ 860 if (zfs_zget(zfsvfs, lr->lr_foid, &zp) != 0) 861 return (ENOENT); 862 if (zp->z_reap) { 863 VN_RELE(ZTOV(zp)); 864 return (ENOENT); 865 } 866 867 /* 868 * Write records come in two flavors: immediate and indirect. 869 * For small writes it's cheaper to store the data with the 870 * log record (immediate); for large writes it's cheaper to 871 * sync the data and get a pointer to it (indirect) so that 872 * we don't have to write the data twice. 873 */ 874 if (buf != NULL) { /* immediate write */ 875 rl = zfs_range_lock(zp, off, dlen, RL_READER); 876 /* test for truncation needs to be done while range locked */ 877 if (off >= zp->z_phys->zp_size) { 878 error = ENOENT; 879 goto out; 880 } 881 VERIFY(0 == dmu_read(os, lr->lr_foid, off, dlen, buf)); 882 } else { /* indirect write */ 883 uint64_t boff; /* block starting offset */ 884 885 ASSERT3U(dlen, <=, zp->z_blksz); 886 /* 887 * Have to lock the whole block to ensure when it's 888 * written out and it's checksum is being calculated 889 * that no one can change the data. We need to re-check 890 * blocksize after we get the lock in case it's changed! 891 */ 892 for (;;) { 893 if (ISP2(zp->z_blksz)) { 894 boff = P2ALIGN_TYPED(off, zp->z_blksz, 895 uint64_t); 896 } else { 897 boff = 0; 898 } 899 dlen = zp->z_blksz; 900 rl = zfs_range_lock(zp, boff, dlen, RL_READER); 901 if (zp->z_blksz == dlen) 902 break; 903 zfs_range_unlock(rl); 904 } 905 /* test for truncation needs to be done while range locked */ 906 if (off >= zp->z_phys->zp_size) { 907 error = ENOENT; 908 goto out; 909 } 910 VERIFY(0 == dmu_buf_hold(os, lr->lr_foid, boff, rl, &db)); 911 ASSERT(boff == db->db_offset); 912 lr->lr_blkoff = off - boff; 913 error = dmu_sync(zio, db, &lr->lr_blkptr, 914 lr->lr_common.lrc_txg, zio ? zfs_get_done : NULL, rl); 915 /* 916 * If we get EINPROGRESS, then we need to wait for a 917 * write IO initiated by dmu_sync() to complete before 918 * we can release this dbuf. We will finish everything 919 * up in the zfs_get_done() callback. 920 */ 921 if (error == EINPROGRESS) 922 return (0); 923 dmu_buf_rele(db, rl); 924 } 925 out: 926 zfs_range_unlock(rl); 927 VN_RELE(ZTOV(zp)); 928 return (error); 929 } 930 931 /*ARGSUSED*/ 932 static int 933 zfs_access(vnode_t *vp, int mode, int flags, cred_t *cr) 934 { 935 znode_t *zp = VTOZ(vp); 936 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 937 int error; 938 939 ZFS_ENTER(zfsvfs); 940 error = zfs_zaccess_rwx(zp, mode, cr); 941 ZFS_EXIT(zfsvfs); 942 return (error); 943 } 944 945 /* 946 * Lookup an entry in a directory, or an extended attribute directory. 947 * If it exists, return a held vnode reference for it. 948 * 949 * IN: dvp - vnode of directory to search. 950 * nm - name of entry to lookup. 951 * pnp - full pathname to lookup [UNUSED]. 952 * flags - LOOKUP_XATTR set if looking for an attribute. 953 * rdir - root directory vnode [UNUSED]. 954 * cr - credentials of caller. 955 * 956 * OUT: vpp - vnode of located entry, NULL if not found. 957 * 958 * RETURN: 0 if success 959 * error code if failure 960 * 961 * Timestamps: 962 * NA 963 */ 964 /* ARGSUSED */ 965 static int 966 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp, 967 int flags, vnode_t *rdir, cred_t *cr) 968 { 969 970 znode_t *zdp = VTOZ(dvp); 971 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 972 int error; 973 974 ZFS_ENTER(zfsvfs); 975 976 *vpp = NULL; 977 978 if (flags & LOOKUP_XATTR) { 979 /* 980 * We don't allow recursive attributes.. 981 * Maybe someday we will. 982 */ 983 if (zdp->z_phys->zp_flags & ZFS_XATTR) { 984 ZFS_EXIT(zfsvfs); 985 return (EINVAL); 986 } 987 988 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr)) { 989 ZFS_EXIT(zfsvfs); 990 return (error); 991 } 992 993 /* 994 * Do we have permission to get into attribute directory? 995 */ 996 997 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, cr)) { 998 VN_RELE(*vpp); 999 } 1000 1001 ZFS_EXIT(zfsvfs); 1002 return (error); 1003 } 1004 1005 if (dvp->v_type != VDIR) { 1006 ZFS_EXIT(zfsvfs); 1007 return (ENOTDIR); 1008 } 1009 1010 /* 1011 * Check accessibility of directory. 1012 */ 1013 1014 if (error = zfs_zaccess(zdp, ACE_EXECUTE, cr)) { 1015 ZFS_EXIT(zfsvfs); 1016 return (error); 1017 } 1018 1019 if ((error = zfs_dirlook(zdp, nm, vpp)) == 0) { 1020 1021 /* 1022 * Convert device special files 1023 */ 1024 if (IS_DEVVP(*vpp)) { 1025 vnode_t *svp; 1026 1027 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1028 VN_RELE(*vpp); 1029 if (svp == NULL) 1030 error = ENOSYS; 1031 else 1032 *vpp = svp; 1033 } 1034 } 1035 1036 ZFS_EXIT(zfsvfs); 1037 return (error); 1038 } 1039 1040 /* 1041 * Attempt to create a new entry in a directory. If the entry 1042 * already exists, truncate the file if permissible, else return 1043 * an error. Return the vp of the created or trunc'd file. 1044 * 1045 * IN: dvp - vnode of directory to put new file entry in. 1046 * name - name of new file entry. 1047 * vap - attributes of new file. 1048 * excl - flag indicating exclusive or non-exclusive mode. 1049 * mode - mode to open file with. 1050 * cr - credentials of caller. 1051 * flag - large file flag [UNUSED]. 1052 * 1053 * OUT: vpp - vnode of created or trunc'd entry. 1054 * 1055 * RETURN: 0 if success 1056 * error code if failure 1057 * 1058 * Timestamps: 1059 * dvp - ctime|mtime updated if new entry created 1060 * vp - ctime|mtime always, atime if new 1061 */ 1062 /* ARGSUSED */ 1063 static int 1064 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl, 1065 int mode, vnode_t **vpp, cred_t *cr, int flag) 1066 { 1067 znode_t *zp, *dzp = VTOZ(dvp); 1068 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1069 zilog_t *zilog = zfsvfs->z_log; 1070 objset_t *os = zfsvfs->z_os; 1071 zfs_dirlock_t *dl; 1072 dmu_tx_t *tx; 1073 int error; 1074 uint64_t zoid; 1075 1076 ZFS_ENTER(zfsvfs); 1077 1078 top: 1079 *vpp = NULL; 1080 1081 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr)) 1082 vap->va_mode &= ~VSVTX; 1083 1084 if (*name == '\0') { 1085 /* 1086 * Null component name refers to the directory itself. 1087 */ 1088 VN_HOLD(dvp); 1089 zp = dzp; 1090 dl = NULL; 1091 error = 0; 1092 } else { 1093 /* possible VN_HOLD(zp) */ 1094 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, 0)) { 1095 if (strcmp(name, "..") == 0) 1096 error = EISDIR; 1097 ZFS_EXIT(zfsvfs); 1098 return (error); 1099 } 1100 } 1101 1102 zoid = zp ? zp->z_id : -1ULL; 1103 1104 if (zp == NULL) { 1105 /* 1106 * Create a new file object and update the directory 1107 * to reference it. 1108 */ 1109 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 1110 goto out; 1111 } 1112 1113 /* 1114 * We only support the creation of regular files in 1115 * extended attribute directories. 1116 */ 1117 if ((dzp->z_phys->zp_flags & ZFS_XATTR) && 1118 (vap->va_type != VREG)) { 1119 error = EINVAL; 1120 goto out; 1121 } 1122 1123 tx = dmu_tx_create(os); 1124 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1125 dmu_tx_hold_bonus(tx, dzp->z_id); 1126 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1127 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 1128 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1129 0, SPA_MAXBLOCKSIZE); 1130 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1131 if (error) { 1132 zfs_dirent_unlock(dl); 1133 if (error == ERESTART && 1134 zfsvfs->z_assign == TXG_NOWAIT) { 1135 dmu_tx_wait(tx); 1136 dmu_tx_abort(tx); 1137 goto top; 1138 } 1139 dmu_tx_abort(tx); 1140 ZFS_EXIT(zfsvfs); 1141 return (error); 1142 } 1143 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 1144 ASSERT(zp->z_id == zoid); 1145 (void) zfs_link_create(dl, zp, tx, ZNEW); 1146 zfs_log_create(zilog, tx, TX_CREATE, dzp, zp, name); 1147 dmu_tx_commit(tx); 1148 } else { 1149 /* 1150 * A directory entry already exists for this name. 1151 */ 1152 /* 1153 * Can't truncate an existing file if in exclusive mode. 1154 */ 1155 if (excl == EXCL) { 1156 error = EEXIST; 1157 goto out; 1158 } 1159 /* 1160 * Can't open a directory for writing. 1161 */ 1162 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1163 error = EISDIR; 1164 goto out; 1165 } 1166 /* 1167 * Verify requested access to file. 1168 */ 1169 if (mode && (error = zfs_zaccess_rwx(zp, mode, cr))) { 1170 goto out; 1171 } 1172 1173 mutex_enter(&dzp->z_lock); 1174 dzp->z_seq++; 1175 mutex_exit(&dzp->z_lock); 1176 1177 /* 1178 * Truncate regular files if requested. 1179 */ 1180 if ((ZTOV(zp)->v_type == VREG) && 1181 (zp->z_phys->zp_size != 0) && 1182 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1183 error = zfs_freesp(zp, 0, 0, mode, TRUE); 1184 if (error == ERESTART && 1185 zfsvfs->z_assign == TXG_NOWAIT) { 1186 /* NB: we already did dmu_tx_wait() */ 1187 zfs_dirent_unlock(dl); 1188 VN_RELE(ZTOV(zp)); 1189 goto top; 1190 } 1191 } 1192 } 1193 out: 1194 1195 if (dl) 1196 zfs_dirent_unlock(dl); 1197 1198 if (error) { 1199 if (zp) 1200 VN_RELE(ZTOV(zp)); 1201 } else { 1202 *vpp = ZTOV(zp); 1203 /* 1204 * If vnode is for a device return a specfs vnode instead. 1205 */ 1206 if (IS_DEVVP(*vpp)) { 1207 struct vnode *svp; 1208 1209 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1210 VN_RELE(*vpp); 1211 if (svp == NULL) { 1212 error = ENOSYS; 1213 } 1214 *vpp = svp; 1215 } 1216 } 1217 1218 ZFS_EXIT(zfsvfs); 1219 return (error); 1220 } 1221 1222 /* 1223 * Remove an entry from a directory. 1224 * 1225 * IN: dvp - vnode of directory to remove entry from. 1226 * name - name of entry to remove. 1227 * cr - credentials of caller. 1228 * 1229 * RETURN: 0 if success 1230 * error code if failure 1231 * 1232 * Timestamps: 1233 * dvp - ctime|mtime 1234 * vp - ctime (if nlink > 0) 1235 */ 1236 static int 1237 zfs_remove(vnode_t *dvp, char *name, cred_t *cr) 1238 { 1239 znode_t *zp, *dzp = VTOZ(dvp); 1240 znode_t *xzp = NULL; 1241 vnode_t *vp; 1242 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1243 zilog_t *zilog = zfsvfs->z_log; 1244 uint64_t acl_obj, xattr_obj; 1245 zfs_dirlock_t *dl; 1246 dmu_tx_t *tx; 1247 int may_delete_now, delete_now = FALSE; 1248 int reaped; 1249 int error; 1250 1251 ZFS_ENTER(zfsvfs); 1252 1253 top: 1254 /* 1255 * Attempt to lock directory; fail if entry doesn't exist. 1256 */ 1257 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) { 1258 ZFS_EXIT(zfsvfs); 1259 return (error); 1260 } 1261 1262 vp = ZTOV(zp); 1263 1264 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1265 goto out; 1266 } 1267 1268 /* 1269 * Need to use rmdir for removing directories. 1270 */ 1271 if (vp->v_type == VDIR) { 1272 error = EPERM; 1273 goto out; 1274 } 1275 1276 vnevent_remove(vp); 1277 1278 dnlc_remove(dvp, name); 1279 1280 mutex_enter(&vp->v_lock); 1281 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp); 1282 mutex_exit(&vp->v_lock); 1283 1284 /* 1285 * We may delete the znode now, or we may put it on the delete queue; 1286 * it depends on whether we're the last link, and on whether there are 1287 * other holds on the vnode. So we dmu_tx_hold() the right things to 1288 * allow for either case. 1289 */ 1290 tx = dmu_tx_create(zfsvfs->z_os); 1291 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1292 dmu_tx_hold_bonus(tx, zp->z_id); 1293 if (may_delete_now) 1294 dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END); 1295 1296 /* are there any extended attributes? */ 1297 if ((xattr_obj = zp->z_phys->zp_xattr) != 0) { 1298 /* 1299 * XXX - There is a possibility that the delete 1300 * of the parent file could succeed, but then we get 1301 * an ENOSPC when we try to delete the xattrs... 1302 * so we would need to re-try the deletes periodically 1303 */ 1304 /* XXX - do we need this if we are deleting? */ 1305 dmu_tx_hold_bonus(tx, xattr_obj); 1306 } 1307 1308 /* are there any additional acls */ 1309 if ((acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj) != 0 && 1310 may_delete_now) 1311 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 1312 1313 /* charge as an update -- would be nice not to charge at all */ 1314 dmu_tx_hold_zap(tx, zfsvfs->z_dqueue, FALSE, NULL); 1315 1316 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1317 if (error) { 1318 zfs_dirent_unlock(dl); 1319 VN_RELE(vp); 1320 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1321 dmu_tx_wait(tx); 1322 dmu_tx_abort(tx); 1323 goto top; 1324 } 1325 dmu_tx_abort(tx); 1326 ZFS_EXIT(zfsvfs); 1327 return (error); 1328 } 1329 1330 /* 1331 * Remove the directory entry. 1332 */ 1333 error = zfs_link_destroy(dl, zp, tx, 0, &reaped); 1334 1335 if (error) { 1336 dmu_tx_commit(tx); 1337 goto out; 1338 } 1339 1340 if (reaped) { 1341 mutex_enter(&vp->v_lock); 1342 delete_now = may_delete_now && 1343 vp->v_count == 1 && !vn_has_cached_data(vp) && 1344 zp->z_phys->zp_xattr == xattr_obj && 1345 zp->z_phys->zp_acl.z_acl_extern_obj == acl_obj; 1346 mutex_exit(&vp->v_lock); 1347 } 1348 1349 if (delete_now) { 1350 if (zp->z_phys->zp_xattr) { 1351 error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp); 1352 ASSERT3U(error, ==, 0); 1353 ASSERT3U(xzp->z_phys->zp_links, ==, 2); 1354 dmu_buf_will_dirty(xzp->z_dbuf, tx); 1355 mutex_enter(&xzp->z_lock); 1356 xzp->z_reap = 1; 1357 xzp->z_phys->zp_links = 0; 1358 mutex_exit(&xzp->z_lock); 1359 zfs_dq_add(xzp, tx); 1360 zp->z_phys->zp_xattr = 0; /* probably unnecessary */ 1361 } 1362 mutex_enter(&zp->z_lock); 1363 mutex_enter(&vp->v_lock); 1364 vp->v_count--; 1365 ASSERT3U(vp->v_count, ==, 0); 1366 mutex_exit(&vp->v_lock); 1367 mutex_exit(&zp->z_lock); 1368 zfs_znode_delete(zp, tx); 1369 VFS_RELE(zfsvfs->z_vfs); 1370 } else if (reaped) { 1371 zfs_dq_add(zp, tx); 1372 } 1373 1374 zfs_log_remove(zilog, tx, TX_REMOVE, dzp, name); 1375 1376 dmu_tx_commit(tx); 1377 out: 1378 zfs_dirent_unlock(dl); 1379 1380 if (!delete_now) { 1381 VN_RELE(vp); 1382 } else if (xzp) { 1383 /* this rele delayed to prevent nesting transactions */ 1384 VN_RELE(ZTOV(xzp)); 1385 } 1386 1387 ZFS_EXIT(zfsvfs); 1388 return (error); 1389 } 1390 1391 /* 1392 * Create a new directory and insert it into dvp using the name 1393 * provided. Return a pointer to the inserted directory. 1394 * 1395 * IN: dvp - vnode of directory to add subdir to. 1396 * dirname - name of new directory. 1397 * vap - attributes of new directory. 1398 * cr - credentials of caller. 1399 * 1400 * OUT: vpp - vnode of created directory. 1401 * 1402 * RETURN: 0 if success 1403 * error code if failure 1404 * 1405 * Timestamps: 1406 * dvp - ctime|mtime updated 1407 * vp - ctime|mtime|atime updated 1408 */ 1409 static int 1410 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr) 1411 { 1412 znode_t *zp, *dzp = VTOZ(dvp); 1413 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1414 zilog_t *zilog = zfsvfs->z_log; 1415 zfs_dirlock_t *dl; 1416 uint64_t zoid = 0; 1417 dmu_tx_t *tx; 1418 int error; 1419 1420 ASSERT(vap->va_type == VDIR); 1421 1422 ZFS_ENTER(zfsvfs); 1423 1424 if (dzp->z_phys->zp_flags & ZFS_XATTR) { 1425 ZFS_EXIT(zfsvfs); 1426 return (EINVAL); 1427 } 1428 top: 1429 *vpp = NULL; 1430 1431 /* 1432 * First make sure the new directory doesn't exist. 1433 */ 1434 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, ZNEW)) { 1435 ZFS_EXIT(zfsvfs); 1436 return (error); 1437 } 1438 1439 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, cr)) { 1440 zfs_dirent_unlock(dl); 1441 ZFS_EXIT(zfsvfs); 1442 return (error); 1443 } 1444 1445 /* 1446 * Add a new entry to the directory. 1447 */ 1448 tx = dmu_tx_create(zfsvfs->z_os); 1449 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 1450 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 1451 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 1452 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1453 0, SPA_MAXBLOCKSIZE); 1454 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1455 if (error) { 1456 zfs_dirent_unlock(dl); 1457 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1458 dmu_tx_wait(tx); 1459 dmu_tx_abort(tx); 1460 goto top; 1461 } 1462 dmu_tx_abort(tx); 1463 ZFS_EXIT(zfsvfs); 1464 return (error); 1465 } 1466 1467 /* 1468 * Create new node. 1469 */ 1470 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 1471 1472 /* 1473 * Now put new name in parent dir. 1474 */ 1475 (void) zfs_link_create(dl, zp, tx, ZNEW); 1476 1477 *vpp = ZTOV(zp); 1478 1479 zfs_log_create(zilog, tx, TX_MKDIR, dzp, zp, dirname); 1480 dmu_tx_commit(tx); 1481 1482 zfs_dirent_unlock(dl); 1483 1484 ZFS_EXIT(zfsvfs); 1485 return (0); 1486 } 1487 1488 /* 1489 * Remove a directory subdir entry. If the current working 1490 * directory is the same as the subdir to be removed, the 1491 * remove will fail. 1492 * 1493 * IN: dvp - vnode of directory to remove from. 1494 * name - name of directory to be removed. 1495 * cwd - vnode of current working directory. 1496 * cr - credentials of caller. 1497 * 1498 * RETURN: 0 if success 1499 * error code if failure 1500 * 1501 * Timestamps: 1502 * dvp - ctime|mtime updated 1503 */ 1504 static int 1505 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr) 1506 { 1507 znode_t *dzp = VTOZ(dvp); 1508 znode_t *zp; 1509 vnode_t *vp; 1510 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1511 zilog_t *zilog = zfsvfs->z_log; 1512 zfs_dirlock_t *dl; 1513 dmu_tx_t *tx; 1514 int error; 1515 1516 ZFS_ENTER(zfsvfs); 1517 1518 top: 1519 zp = NULL; 1520 1521 /* 1522 * Attempt to lock directory; fail if entry doesn't exist. 1523 */ 1524 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) { 1525 ZFS_EXIT(zfsvfs); 1526 return (error); 1527 } 1528 1529 vp = ZTOV(zp); 1530 1531 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1532 goto out; 1533 } 1534 1535 if (vp->v_type != VDIR) { 1536 error = ENOTDIR; 1537 goto out; 1538 } 1539 1540 if (vp == cwd) { 1541 error = EINVAL; 1542 goto out; 1543 } 1544 1545 vnevent_rmdir(vp); 1546 1547 /* 1548 * Grab a lock on the parent pointer make sure we play well 1549 * with the treewalk and directory rename code. 1550 */ 1551 rw_enter(&zp->z_parent_lock, RW_WRITER); 1552 1553 tx = dmu_tx_create(zfsvfs->z_os); 1554 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1555 dmu_tx_hold_bonus(tx, zp->z_id); 1556 dmu_tx_hold_zap(tx, zfsvfs->z_dqueue, FALSE, NULL); 1557 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1558 if (error) { 1559 rw_exit(&zp->z_parent_lock); 1560 zfs_dirent_unlock(dl); 1561 VN_RELE(vp); 1562 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1563 dmu_tx_wait(tx); 1564 dmu_tx_abort(tx); 1565 goto top; 1566 } 1567 dmu_tx_abort(tx); 1568 ZFS_EXIT(zfsvfs); 1569 return (error); 1570 } 1571 1572 error = zfs_link_destroy(dl, zp, tx, 0, NULL); 1573 1574 if (error == 0) 1575 zfs_log_remove(zilog, tx, TX_RMDIR, dzp, name); 1576 1577 dmu_tx_commit(tx); 1578 1579 rw_exit(&zp->z_parent_lock); 1580 out: 1581 zfs_dirent_unlock(dl); 1582 1583 VN_RELE(vp); 1584 1585 ZFS_EXIT(zfsvfs); 1586 return (error); 1587 } 1588 1589 /* 1590 * Read as many directory entries as will fit into the provided 1591 * buffer from the given directory cursor position (specified in 1592 * the uio structure. 1593 * 1594 * IN: vp - vnode of directory to read. 1595 * uio - structure supplying read location, range info, 1596 * and return buffer. 1597 * cr - credentials of caller. 1598 * 1599 * OUT: uio - updated offset and range, buffer filled. 1600 * eofp - set to true if end-of-file detected. 1601 * 1602 * RETURN: 0 if success 1603 * error code if failure 1604 * 1605 * Timestamps: 1606 * vp - atime updated 1607 * 1608 * Note that the low 4 bits of the cookie returned by zap is always zero. 1609 * This allows us to use the low range for "special" directory entries: 1610 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 1611 * we use the offset 2 for the '.zfs' directory. 1612 */ 1613 /* ARGSUSED */ 1614 static int 1615 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp) 1616 { 1617 znode_t *zp = VTOZ(vp); 1618 iovec_t *iovp; 1619 dirent64_t *odp; 1620 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1621 objset_t *os; 1622 caddr_t outbuf; 1623 size_t bufsize; 1624 zap_cursor_t zc; 1625 zap_attribute_t zap; 1626 uint_t bytes_wanted; 1627 ushort_t this_reclen; 1628 uint64_t offset; /* must be unsigned; checks for < 1 */ 1629 off64_t *next; 1630 int local_eof; 1631 int outcount; 1632 int error; 1633 uint8_t prefetch; 1634 1635 ZFS_ENTER(zfsvfs); 1636 1637 /* 1638 * If we are not given an eof variable, 1639 * use a local one. 1640 */ 1641 if (eofp == NULL) 1642 eofp = &local_eof; 1643 1644 /* 1645 * Check for valid iov_len. 1646 */ 1647 if (uio->uio_iov->iov_len <= 0) { 1648 ZFS_EXIT(zfsvfs); 1649 return (EINVAL); 1650 } 1651 1652 /* 1653 * Quit if directory has been removed (posix) 1654 */ 1655 if ((*eofp = zp->z_reap) != 0) { 1656 ZFS_EXIT(zfsvfs); 1657 return (0); 1658 } 1659 1660 error = 0; 1661 os = zfsvfs->z_os; 1662 offset = uio->uio_loffset; 1663 prefetch = zp->z_zn_prefetch; 1664 1665 /* 1666 * Initialize the iterator cursor. 1667 */ 1668 if (offset <= 3) { 1669 /* 1670 * Start iteration from the beginning of the directory. 1671 */ 1672 zap_cursor_init(&zc, os, zp->z_id); 1673 } else { 1674 /* 1675 * The offset is a serialized cursor. 1676 */ 1677 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 1678 } 1679 1680 /* 1681 * Get space to change directory entries into fs independent format. 1682 */ 1683 iovp = uio->uio_iov; 1684 bytes_wanted = iovp->iov_len; 1685 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 1686 bufsize = bytes_wanted; 1687 outbuf = kmem_alloc(bufsize, KM_SLEEP); 1688 odp = (struct dirent64 *)outbuf; 1689 } else { 1690 bufsize = bytes_wanted; 1691 odp = (struct dirent64 *)iovp->iov_base; 1692 } 1693 1694 /* 1695 * Transform to file-system independent format 1696 */ 1697 outcount = 0; 1698 while (outcount < bytes_wanted) { 1699 /* 1700 * Special case `.', `..', and `.zfs'. 1701 */ 1702 if (offset == 0) { 1703 (void) strcpy(zap.za_name, "."); 1704 zap.za_first_integer = zp->z_id; 1705 this_reclen = DIRENT64_RECLEN(1); 1706 } else if (offset == 1) { 1707 (void) strcpy(zap.za_name, ".."); 1708 zap.za_first_integer = zp->z_phys->zp_parent; 1709 this_reclen = DIRENT64_RECLEN(2); 1710 } else if (offset == 2 && zfs_show_ctldir(zp)) { 1711 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 1712 zap.za_first_integer = ZFSCTL_INO_ROOT; 1713 this_reclen = 1714 DIRENT64_RECLEN(sizeof (ZFS_CTLDIR_NAME) - 1); 1715 } else { 1716 /* 1717 * Grab next entry. 1718 */ 1719 if (error = zap_cursor_retrieve(&zc, &zap)) { 1720 if ((*eofp = (error == ENOENT)) != 0) 1721 break; 1722 else 1723 goto update; 1724 } 1725 1726 if (zap.za_integer_length != 8 || 1727 zap.za_num_integers != 1) { 1728 cmn_err(CE_WARN, "zap_readdir: bad directory " 1729 "entry, obj = %lld, offset = %lld\n", 1730 (u_longlong_t)zp->z_id, 1731 (u_longlong_t)offset); 1732 error = ENXIO; 1733 goto update; 1734 } 1735 this_reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 1736 } 1737 1738 /* 1739 * Will this entry fit in the buffer? 1740 */ 1741 if (outcount + this_reclen > bufsize) { 1742 /* 1743 * Did we manage to fit anything in the buffer? 1744 */ 1745 if (!outcount) { 1746 error = EINVAL; 1747 goto update; 1748 } 1749 break; 1750 } 1751 /* 1752 * Add this entry: 1753 */ 1754 odp->d_ino = (ino64_t)zap.za_first_integer; 1755 odp->d_reclen = (ushort_t)this_reclen; 1756 /* NOTE: d_off is the offset for the *next* entry */ 1757 next = &(odp->d_off); 1758 (void) strncpy(odp->d_name, zap.za_name, 1759 DIRENT64_NAMELEN(this_reclen)); 1760 outcount += this_reclen; 1761 odp = (dirent64_t *)((intptr_t)odp + this_reclen); 1762 1763 ASSERT(outcount <= bufsize); 1764 1765 /* Prefetch znode */ 1766 if (prefetch) 1767 dmu_prefetch(os, zap.za_first_integer, 0, 0); 1768 1769 /* 1770 * Move to the next entry, fill in the previous offset. 1771 */ 1772 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 1773 zap_cursor_advance(&zc); 1774 offset = zap_cursor_serialize(&zc); 1775 } else { 1776 offset += 1; 1777 } 1778 *next = offset; 1779 } 1780 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 1781 1782 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 1783 iovp->iov_base += outcount; 1784 iovp->iov_len -= outcount; 1785 uio->uio_resid -= outcount; 1786 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 1787 /* 1788 * Reset the pointer. 1789 */ 1790 offset = uio->uio_loffset; 1791 } 1792 1793 update: 1794 zap_cursor_fini(&zc); 1795 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 1796 kmem_free(outbuf, bufsize); 1797 1798 if (error == ENOENT) 1799 error = 0; 1800 1801 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 1802 1803 uio->uio_loffset = offset; 1804 ZFS_EXIT(zfsvfs); 1805 return (error); 1806 } 1807 1808 static int 1809 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr) 1810 { 1811 znode_t *zp = VTOZ(vp); 1812 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1813 1814 /* 1815 * Regardless of whether this is required for standards conformance, 1816 * this is the logical behavior when fsync() is called on a file with 1817 * dirty pages. We use B_ASYNC since the ZIL transactions are already 1818 * going to be pushed out as part of the zil_commit(). 1819 */ 1820 if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) && 1821 (vp->v_type == VREG) && !(IS_SWAPVP(vp))) 1822 (void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr); 1823 1824 ZFS_ENTER(zfsvfs); 1825 zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id); 1826 ZFS_EXIT(zfsvfs); 1827 return (0); 1828 } 1829 1830 /* 1831 * Get the requested file attributes and place them in the provided 1832 * vattr structure. 1833 * 1834 * IN: vp - vnode of file. 1835 * vap - va_mask identifies requested attributes. 1836 * flags - [UNUSED] 1837 * cr - credentials of caller. 1838 * 1839 * OUT: vap - attribute values. 1840 * 1841 * RETURN: 0 (always succeeds) 1842 */ 1843 /* ARGSUSED */ 1844 static int 1845 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr) 1846 { 1847 znode_t *zp = VTOZ(vp); 1848 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1849 znode_phys_t *pzp = zp->z_phys; 1850 int error; 1851 1852 ZFS_ENTER(zfsvfs); 1853 1854 /* 1855 * Return all attributes. It's cheaper to provide the answer 1856 * than to determine whether we were asked the question. 1857 */ 1858 mutex_enter(&zp->z_lock); 1859 1860 vap->va_type = vp->v_type; 1861 vap->va_mode = pzp->zp_mode & MODEMASK; 1862 vap->va_uid = zp->z_phys->zp_uid; 1863 vap->va_gid = zp->z_phys->zp_gid; 1864 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev; 1865 vap->va_nodeid = zp->z_id; 1866 vap->va_nlink = MIN(pzp->zp_links, UINT32_MAX); /* nlink_t limit! */ 1867 vap->va_size = pzp->zp_size; 1868 vap->va_rdev = vp->v_rdev; 1869 vap->va_seq = zp->z_seq; 1870 1871 ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime); 1872 ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime); 1873 ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime); 1874 1875 /* 1876 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 1877 * Also, if we are the owner don't bother, since owner should 1878 * always be allowed to read basic attributes of file. 1879 */ 1880 if (!(zp->z_phys->zp_flags & ZFS_ACL_TRIVIAL) && 1881 (zp->z_phys->zp_uid != crgetuid(cr))) { 1882 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, cr)) { 1883 mutex_exit(&zp->z_lock); 1884 ZFS_EXIT(zfsvfs); 1885 return (error); 1886 } 1887 } 1888 1889 mutex_exit(&zp->z_lock); 1890 1891 dmu_object_size_from_db(zp->z_dbuf, &vap->va_blksize, &vap->va_nblocks); 1892 1893 if (zp->z_blksz == 0) { 1894 /* 1895 * Block size hasn't been set; suggest maximal I/O transfers. 1896 */ 1897 vap->va_blksize = zfsvfs->z_max_blksz; 1898 } 1899 1900 ZFS_EXIT(zfsvfs); 1901 return (0); 1902 } 1903 1904 /* 1905 * Set the file attributes to the values contained in the 1906 * vattr structure. 1907 * 1908 * IN: vp - vnode of file to be modified. 1909 * vap - new attribute values. 1910 * flags - ATTR_UTIME set if non-default time values provided. 1911 * cr - credentials of caller. 1912 * 1913 * RETURN: 0 if success 1914 * error code if failure 1915 * 1916 * Timestamps: 1917 * vp - ctime updated, mtime updated if size changed. 1918 */ 1919 /* ARGSUSED */ 1920 static int 1921 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 1922 caller_context_t *ct) 1923 { 1924 struct znode *zp = VTOZ(vp); 1925 znode_phys_t *pzp = zp->z_phys; 1926 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1927 zilog_t *zilog = zfsvfs->z_log; 1928 dmu_tx_t *tx; 1929 vattr_t oldva; 1930 uint_t mask = vap->va_mask; 1931 uint_t saved_mask; 1932 int trim_mask = 0; 1933 uint64_t new_mode; 1934 znode_t *attrzp; 1935 int need_policy = FALSE; 1936 int err; 1937 1938 if (mask == 0) 1939 return (0); 1940 1941 if (mask & AT_NOSET) 1942 return (EINVAL); 1943 1944 if (mask & AT_SIZE && vp->v_type == VDIR) 1945 return (EISDIR); 1946 1947 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) 1948 return (EINVAL); 1949 1950 ZFS_ENTER(zfsvfs); 1951 1952 top: 1953 attrzp = NULL; 1954 1955 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 1956 ZFS_EXIT(zfsvfs); 1957 return (EROFS); 1958 } 1959 1960 /* 1961 * First validate permissions 1962 */ 1963 1964 if (mask & AT_SIZE) { 1965 err = zfs_zaccess(zp, ACE_WRITE_DATA, cr); 1966 if (err) { 1967 ZFS_EXIT(zfsvfs); 1968 return (err); 1969 } 1970 /* 1971 * XXX - Note, we are not providing any open 1972 * mode flags here (like FNDELAY), so we may 1973 * block if there are locks present... this 1974 * should be addressed in openat(). 1975 */ 1976 do { 1977 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 1978 /* NB: we already did dmu_tx_wait() if necessary */ 1979 } while (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT); 1980 if (err) { 1981 ZFS_EXIT(zfsvfs); 1982 return (err); 1983 } 1984 } 1985 1986 if (mask & (AT_ATIME|AT_MTIME)) 1987 need_policy = zfs_zaccess_v4_perm(zp, ACE_WRITE_ATTRIBUTES, cr); 1988 1989 if (mask & (AT_UID|AT_GID)) { 1990 int idmask = (mask & (AT_UID|AT_GID)); 1991 int take_owner; 1992 int take_group; 1993 1994 /* 1995 * NOTE: even if a new mode is being set, 1996 * we may clear S_ISUID/S_ISGID bits. 1997 */ 1998 1999 if (!(mask & AT_MODE)) 2000 vap->va_mode = pzp->zp_mode; 2001 2002 /* 2003 * Take ownership or chgrp to group we are a member of 2004 */ 2005 2006 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 2007 take_group = (mask & AT_GID) && groupmember(vap->va_gid, cr); 2008 2009 /* 2010 * If both AT_UID and AT_GID are set then take_owner and 2011 * take_group must both be set in order to allow taking 2012 * ownership. 2013 * 2014 * Otherwise, send the check through secpolicy_vnode_setattr() 2015 * 2016 */ 2017 2018 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 2019 ((idmask == AT_UID) && take_owner) || 2020 ((idmask == AT_GID) && take_group)) { 2021 if (zfs_zaccess_v4_perm(zp, ACE_WRITE_OWNER, cr) == 0) { 2022 /* 2023 * Remove setuid/setgid for non-privileged users 2024 */ 2025 secpolicy_setid_clear(vap, cr); 2026 trim_mask = (mask & (AT_UID|AT_GID)); 2027 } else { 2028 need_policy = TRUE; 2029 } 2030 } else { 2031 need_policy = TRUE; 2032 } 2033 } 2034 2035 mutex_enter(&zp->z_lock); 2036 oldva.va_mode = pzp->zp_mode; 2037 oldva.va_uid = zp->z_phys->zp_uid; 2038 oldva.va_gid = zp->z_phys->zp_gid; 2039 mutex_exit(&zp->z_lock); 2040 2041 if (mask & AT_MODE) { 2042 if (zfs_zaccess_v4_perm(zp, ACE_WRITE_ACL, cr) == 0) { 2043 err = secpolicy_setid_setsticky_clear(vp, vap, 2044 &oldva, cr); 2045 if (err) { 2046 ZFS_EXIT(zfsvfs); 2047 return (err); 2048 } 2049 trim_mask |= AT_MODE; 2050 } else { 2051 need_policy = TRUE; 2052 } 2053 } 2054 2055 if (need_policy) { 2056 /* 2057 * If trim_mask is set then take ownership 2058 * has been granted or write_acl is present and user 2059 * has the ability to modify mode. In that case remove 2060 * UID|GID and or MODE from mask so that 2061 * secpolicy_vnode_setattr() doesn't revoke it. 2062 */ 2063 2064 if (trim_mask) { 2065 saved_mask = vap->va_mask; 2066 vap->va_mask &= ~trim_mask; 2067 2068 } 2069 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 2070 (int (*)(void *, int, cred_t *))zfs_zaccess_rwx, zp); 2071 if (err) { 2072 ZFS_EXIT(zfsvfs); 2073 return (err); 2074 } 2075 2076 if (trim_mask) 2077 vap->va_mask |= saved_mask; 2078 } 2079 2080 /* 2081 * secpolicy_vnode_setattr, or take ownership may have 2082 * changed va_mask 2083 */ 2084 mask = vap->va_mask; 2085 2086 tx = dmu_tx_create(zfsvfs->z_os); 2087 dmu_tx_hold_bonus(tx, zp->z_id); 2088 2089 if (mask & AT_MODE) { 2090 uint64_t pmode = pzp->zp_mode; 2091 2092 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 2093 2094 if (zp->z_phys->zp_acl.z_acl_extern_obj) 2095 dmu_tx_hold_write(tx, 2096 pzp->zp_acl.z_acl_extern_obj, 0, SPA_MAXBLOCKSIZE); 2097 else 2098 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2099 0, ZFS_ACL_SIZE(MAX_ACL_SIZE)); 2100 } 2101 2102 if ((mask & (AT_UID | AT_GID)) && zp->z_phys->zp_xattr != 0) { 2103 err = zfs_zget(zp->z_zfsvfs, zp->z_phys->zp_xattr, &attrzp); 2104 if (err) { 2105 dmu_tx_abort(tx); 2106 ZFS_EXIT(zfsvfs); 2107 return (err); 2108 } 2109 dmu_tx_hold_bonus(tx, attrzp->z_id); 2110 } 2111 2112 err = dmu_tx_assign(tx, zfsvfs->z_assign); 2113 if (err) { 2114 if (attrzp) 2115 VN_RELE(ZTOV(attrzp)); 2116 if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2117 dmu_tx_wait(tx); 2118 dmu_tx_abort(tx); 2119 goto top; 2120 } 2121 dmu_tx_abort(tx); 2122 ZFS_EXIT(zfsvfs); 2123 return (err); 2124 } 2125 2126 dmu_buf_will_dirty(zp->z_dbuf, tx); 2127 2128 /* 2129 * Set each attribute requested. 2130 * We group settings according to the locks they need to acquire. 2131 * 2132 * Note: you cannot set ctime directly, although it will be 2133 * updated as a side-effect of calling this function. 2134 */ 2135 2136 mutex_enter(&zp->z_lock); 2137 2138 if (mask & AT_MODE) { 2139 err = zfs_acl_chmod_setattr(zp, new_mode, tx); 2140 ASSERT3U(err, ==, 0); 2141 } 2142 2143 if (attrzp) 2144 mutex_enter(&attrzp->z_lock); 2145 2146 if (mask & AT_UID) { 2147 zp->z_phys->zp_uid = (uint64_t)vap->va_uid; 2148 if (attrzp) { 2149 attrzp->z_phys->zp_uid = (uint64_t)vap->va_uid; 2150 } 2151 } 2152 2153 if (mask & AT_GID) { 2154 zp->z_phys->zp_gid = (uint64_t)vap->va_gid; 2155 if (attrzp) 2156 attrzp->z_phys->zp_gid = (uint64_t)vap->va_gid; 2157 } 2158 2159 if (attrzp) 2160 mutex_exit(&attrzp->z_lock); 2161 2162 if (mask & AT_ATIME) 2163 ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime); 2164 2165 if (mask & AT_MTIME) 2166 ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime); 2167 2168 if (mask & AT_SIZE) 2169 zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx); 2170 else if (mask != 0) 2171 zfs_time_stamper_locked(zp, STATE_CHANGED, tx); 2172 2173 if (mask != 0) 2174 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask); 2175 2176 mutex_exit(&zp->z_lock); 2177 2178 if (attrzp) 2179 VN_RELE(ZTOV(attrzp)); 2180 2181 dmu_tx_commit(tx); 2182 2183 ZFS_EXIT(zfsvfs); 2184 return (err); 2185 } 2186 2187 /* 2188 * Search back through the directory tree, using the ".." entries. 2189 * Lock each directory in the chain to prevent concurrent renames. 2190 * Fail any attempt to move a directory into one of its own descendants. 2191 * XXX - z_parent_lock can overlap with map or grow locks 2192 */ 2193 typedef struct zfs_zlock { 2194 krwlock_t *zl_rwlock; /* lock we acquired */ 2195 znode_t *zl_znode; /* znode we held */ 2196 struct zfs_zlock *zl_next; /* next in list */ 2197 } zfs_zlock_t; 2198 2199 static int 2200 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 2201 { 2202 zfs_zlock_t *zl; 2203 znode_t *zp = tdzp; 2204 uint64_t rootid = zp->z_zfsvfs->z_root; 2205 uint64_t *oidp = &zp->z_id; 2206 krwlock_t *rwlp = &szp->z_parent_lock; 2207 krw_t rw = RW_WRITER; 2208 2209 /* 2210 * First pass write-locks szp and compares to zp->z_id. 2211 * Later passes read-lock zp and compare to zp->z_parent. 2212 */ 2213 do { 2214 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 2215 zl->zl_rwlock = rwlp; 2216 zl->zl_znode = NULL; 2217 zl->zl_next = *zlpp; 2218 *zlpp = zl; 2219 2220 rw_enter(rwlp, rw); 2221 2222 if (*oidp == szp->z_id) /* We're a descendant of szp */ 2223 return (EINVAL); 2224 2225 if (*oidp == rootid) /* We've hit the top */ 2226 return (0); 2227 2228 if (rw == RW_READER) { /* i.e. not the first pass */ 2229 int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp); 2230 if (error) 2231 return (error); 2232 zl->zl_znode = zp; 2233 } 2234 oidp = &zp->z_phys->zp_parent; 2235 rwlp = &zp->z_parent_lock; 2236 rw = RW_READER; 2237 2238 } while (zp->z_id != sdzp->z_id); 2239 2240 return (0); 2241 } 2242 2243 /* 2244 * Drop locks and release vnodes that were held by zfs_rename_lock(). 2245 */ 2246 static void 2247 zfs_rename_unlock(zfs_zlock_t **zlpp) 2248 { 2249 zfs_zlock_t *zl; 2250 2251 while ((zl = *zlpp) != NULL) { 2252 if (zl->zl_znode != NULL) 2253 VN_RELE(ZTOV(zl->zl_znode)); 2254 rw_exit(zl->zl_rwlock); 2255 *zlpp = zl->zl_next; 2256 kmem_free(zl, sizeof (*zl)); 2257 } 2258 } 2259 2260 /* 2261 * Move an entry from the provided source directory to the target 2262 * directory. Change the entry name as indicated. 2263 * 2264 * IN: sdvp - Source directory containing the "old entry". 2265 * snm - Old entry name. 2266 * tdvp - Target directory to contain the "new entry". 2267 * tnm - New entry name. 2268 * cr - credentials of caller. 2269 * 2270 * RETURN: 0 if success 2271 * error code if failure 2272 * 2273 * Timestamps: 2274 * sdvp,tdvp - ctime|mtime updated 2275 */ 2276 static int 2277 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr) 2278 { 2279 znode_t *tdzp, *szp, *tzp; 2280 znode_t *sdzp = VTOZ(sdvp); 2281 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs; 2282 zilog_t *zilog = zfsvfs->z_log; 2283 vnode_t *realvp; 2284 zfs_dirlock_t *sdl, *tdl; 2285 dmu_tx_t *tx; 2286 zfs_zlock_t *zl; 2287 int cmp, serr, terr, error; 2288 2289 ZFS_ENTER(zfsvfs); 2290 2291 /* 2292 * Make sure we have the real vp for the target directory. 2293 */ 2294 if (VOP_REALVP(tdvp, &realvp) == 0) 2295 tdvp = realvp; 2296 2297 if (tdvp->v_vfsp != sdvp->v_vfsp) { 2298 ZFS_EXIT(zfsvfs); 2299 return (EXDEV); 2300 } 2301 2302 tdzp = VTOZ(tdvp); 2303 top: 2304 szp = NULL; 2305 tzp = NULL; 2306 zl = NULL; 2307 2308 /* 2309 * This is to prevent the creation of links into attribute space 2310 * by renaming a linked file into/outof an attribute directory. 2311 * See the comment in zfs_link() for why this is considered bad. 2312 */ 2313 if ((tdzp->z_phys->zp_flags & ZFS_XATTR) != 2314 (sdzp->z_phys->zp_flags & ZFS_XATTR)) { 2315 ZFS_EXIT(zfsvfs); 2316 return (EINVAL); 2317 } 2318 2319 /* 2320 * Lock source and target directory entries. To prevent deadlock, 2321 * a lock ordering must be defined. We lock the directory with 2322 * the smallest object id first, or if it's a tie, the one with 2323 * the lexically first name. 2324 */ 2325 if (sdzp->z_id < tdzp->z_id) { 2326 cmp = -1; 2327 } else if (sdzp->z_id > tdzp->z_id) { 2328 cmp = 1; 2329 } else { 2330 cmp = strcmp(snm, tnm); 2331 if (cmp == 0) { 2332 /* 2333 * POSIX: "If the old argument and the new argument 2334 * both refer to links to the same existing file, 2335 * the rename() function shall return successfully 2336 * and perform no other action." 2337 */ 2338 ZFS_EXIT(zfsvfs); 2339 return (0); 2340 } 2341 } 2342 if (cmp < 0) { 2343 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS); 2344 terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0); 2345 } else { 2346 terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0); 2347 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS); 2348 } 2349 2350 if (serr) { 2351 /* 2352 * Source entry invalid or not there. 2353 */ 2354 if (!terr) { 2355 zfs_dirent_unlock(tdl); 2356 if (tzp) 2357 VN_RELE(ZTOV(tzp)); 2358 } 2359 if (strcmp(snm, "..") == 0) 2360 serr = EINVAL; 2361 ZFS_EXIT(zfsvfs); 2362 return (serr); 2363 } 2364 if (terr) { 2365 zfs_dirent_unlock(sdl); 2366 VN_RELE(ZTOV(szp)); 2367 if (strcmp(tnm, "..") == 0) 2368 terr = EINVAL; 2369 ZFS_EXIT(zfsvfs); 2370 return (terr); 2371 } 2372 2373 /* 2374 * Must have write access at the source to remove the old entry 2375 * and write access at the target to create the new entry. 2376 * Note that if target and source are the same, this can be 2377 * done in a single check. 2378 */ 2379 2380 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 2381 goto out; 2382 2383 if (ZTOV(szp)->v_type == VDIR) { 2384 /* 2385 * Check to make sure rename is valid. 2386 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 2387 */ 2388 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 2389 goto out; 2390 } 2391 2392 /* 2393 * Does target exist? 2394 */ 2395 if (tzp) { 2396 /* 2397 * Source and target must be the same type. 2398 */ 2399 if (ZTOV(szp)->v_type == VDIR) { 2400 if (ZTOV(tzp)->v_type != VDIR) { 2401 error = ENOTDIR; 2402 goto out; 2403 } 2404 } else { 2405 if (ZTOV(tzp)->v_type == VDIR) { 2406 error = EISDIR; 2407 goto out; 2408 } 2409 } 2410 /* 2411 * POSIX dictates that when the source and target 2412 * entries refer to the same file object, rename 2413 * must do nothing and exit without error. 2414 */ 2415 if (szp->z_id == tzp->z_id) { 2416 error = 0; 2417 goto out; 2418 } 2419 } 2420 2421 vnevent_rename_src(ZTOV(szp)); 2422 if (tzp) 2423 vnevent_rename_dest(ZTOV(tzp)); 2424 2425 tx = dmu_tx_create(zfsvfs->z_os); 2426 dmu_tx_hold_bonus(tx, szp->z_id); /* nlink changes */ 2427 dmu_tx_hold_bonus(tx, sdzp->z_id); /* nlink changes */ 2428 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 2429 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 2430 if (sdzp != tdzp) 2431 dmu_tx_hold_bonus(tx, tdzp->z_id); /* nlink changes */ 2432 if (tzp) 2433 dmu_tx_hold_bonus(tx, tzp->z_id); /* parent changes */ 2434 dmu_tx_hold_zap(tx, zfsvfs->z_dqueue, FALSE, NULL); 2435 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2436 if (error) { 2437 if (zl != NULL) 2438 zfs_rename_unlock(&zl); 2439 zfs_dirent_unlock(sdl); 2440 zfs_dirent_unlock(tdl); 2441 VN_RELE(ZTOV(szp)); 2442 if (tzp) 2443 VN_RELE(ZTOV(tzp)); 2444 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2445 dmu_tx_wait(tx); 2446 dmu_tx_abort(tx); 2447 goto top; 2448 } 2449 dmu_tx_abort(tx); 2450 ZFS_EXIT(zfsvfs); 2451 return (error); 2452 } 2453 2454 if (tzp) /* Attempt to remove the existing target */ 2455 error = zfs_link_destroy(tdl, tzp, tx, 0, NULL); 2456 2457 if (error == 0) { 2458 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 2459 if (error == 0) { 2460 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 2461 ASSERT(error == 0); 2462 zfs_log_rename(zilog, tx, TX_RENAME, sdzp, 2463 sdl->dl_name, tdzp, tdl->dl_name, szp); 2464 } 2465 } 2466 2467 dmu_tx_commit(tx); 2468 out: 2469 if (zl != NULL) 2470 zfs_rename_unlock(&zl); 2471 2472 zfs_dirent_unlock(sdl); 2473 zfs_dirent_unlock(tdl); 2474 2475 VN_RELE(ZTOV(szp)); 2476 if (tzp) 2477 VN_RELE(ZTOV(tzp)); 2478 2479 ZFS_EXIT(zfsvfs); 2480 return (error); 2481 } 2482 2483 /* 2484 * Insert the indicated symbolic reference entry into the directory. 2485 * 2486 * IN: dvp - Directory to contain new symbolic link. 2487 * link - Name for new symlink entry. 2488 * vap - Attributes of new entry. 2489 * target - Target path of new symlink. 2490 * cr - credentials of caller. 2491 * 2492 * RETURN: 0 if success 2493 * error code if failure 2494 * 2495 * Timestamps: 2496 * dvp - ctime|mtime updated 2497 */ 2498 static int 2499 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr) 2500 { 2501 znode_t *zp, *dzp = VTOZ(dvp); 2502 zfs_dirlock_t *dl; 2503 dmu_tx_t *tx; 2504 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2505 zilog_t *zilog = zfsvfs->z_log; 2506 uint64_t zoid; 2507 int len = strlen(link); 2508 int error; 2509 2510 ASSERT(vap->va_type == VLNK); 2511 2512 ZFS_ENTER(zfsvfs); 2513 top: 2514 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 2515 ZFS_EXIT(zfsvfs); 2516 return (error); 2517 } 2518 2519 if (len > MAXPATHLEN) { 2520 ZFS_EXIT(zfsvfs); 2521 return (ENAMETOOLONG); 2522 } 2523 2524 /* 2525 * Attempt to lock directory; fail if entry already exists. 2526 */ 2527 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZNEW)) { 2528 ZFS_EXIT(zfsvfs); 2529 return (error); 2530 } 2531 2532 tx = dmu_tx_create(zfsvfs->z_os); 2533 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 2534 dmu_tx_hold_bonus(tx, dzp->z_id); 2535 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 2536 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 2537 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE); 2538 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2539 if (error) { 2540 zfs_dirent_unlock(dl); 2541 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2542 dmu_tx_wait(tx); 2543 dmu_tx_abort(tx); 2544 goto top; 2545 } 2546 dmu_tx_abort(tx); 2547 ZFS_EXIT(zfsvfs); 2548 return (error); 2549 } 2550 2551 dmu_buf_will_dirty(dzp->z_dbuf, tx); 2552 2553 /* 2554 * Create a new object for the symlink. 2555 * Put the link content into bonus buffer if it will fit; 2556 * otherwise, store it just like any other file data. 2557 */ 2558 zoid = 0; 2559 if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) { 2560 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, len); 2561 if (len != 0) 2562 bcopy(link, zp->z_phys + 1, len); 2563 } else { 2564 dmu_buf_t *dbp; 2565 2566 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 2567 2568 /* 2569 * Nothing can access the znode yet so no locking needed 2570 * for growing the znode's blocksize. 2571 */ 2572 zfs_grow_blocksize(zp, len, tx); 2573 2574 VERIFY(0 == dmu_buf_hold(zfsvfs->z_os, zoid, 0, FTAG, &dbp)); 2575 dmu_buf_will_dirty(dbp, tx); 2576 2577 ASSERT3U(len, <=, dbp->db_size); 2578 bcopy(link, dbp->db_data, len); 2579 dmu_buf_rele(dbp, FTAG); 2580 } 2581 zp->z_phys->zp_size = len; 2582 2583 /* 2584 * Insert the new object into the directory. 2585 */ 2586 (void) zfs_link_create(dl, zp, tx, ZNEW); 2587 out: 2588 if (error == 0) 2589 zfs_log_symlink(zilog, tx, TX_SYMLINK, dzp, zp, name, link); 2590 2591 dmu_tx_commit(tx); 2592 2593 zfs_dirent_unlock(dl); 2594 2595 VN_RELE(ZTOV(zp)); 2596 2597 ZFS_EXIT(zfsvfs); 2598 return (error); 2599 } 2600 2601 /* 2602 * Return, in the buffer contained in the provided uio structure, 2603 * the symbolic path referred to by vp. 2604 * 2605 * IN: vp - vnode of symbolic link. 2606 * uoip - structure to contain the link path. 2607 * cr - credentials of caller. 2608 * 2609 * OUT: uio - structure to contain the link path. 2610 * 2611 * RETURN: 0 if success 2612 * error code if failure 2613 * 2614 * Timestamps: 2615 * vp - atime updated 2616 */ 2617 /* ARGSUSED */ 2618 static int 2619 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr) 2620 { 2621 znode_t *zp = VTOZ(vp); 2622 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2623 size_t bufsz; 2624 int error; 2625 2626 ZFS_ENTER(zfsvfs); 2627 2628 bufsz = (size_t)zp->z_phys->zp_size; 2629 if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) { 2630 error = uiomove(zp->z_phys + 1, 2631 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 2632 } else { 2633 dmu_buf_t *dbp; 2634 error = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0, FTAG, &dbp); 2635 if (error) { 2636 ZFS_EXIT(zfsvfs); 2637 return (error); 2638 } 2639 error = uiomove(dbp->db_data, 2640 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 2641 dmu_buf_rele(dbp, FTAG); 2642 } 2643 2644 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2645 ZFS_EXIT(zfsvfs); 2646 return (error); 2647 } 2648 2649 /* 2650 * Insert a new entry into directory tdvp referencing svp. 2651 * 2652 * IN: tdvp - Directory to contain new entry. 2653 * svp - vnode of new entry. 2654 * name - name of new entry. 2655 * cr - credentials of caller. 2656 * 2657 * RETURN: 0 if success 2658 * error code if failure 2659 * 2660 * Timestamps: 2661 * tdvp - ctime|mtime updated 2662 * svp - ctime updated 2663 */ 2664 /* ARGSUSED */ 2665 static int 2666 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr) 2667 { 2668 znode_t *dzp = VTOZ(tdvp); 2669 znode_t *tzp, *szp; 2670 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2671 zilog_t *zilog = zfsvfs->z_log; 2672 zfs_dirlock_t *dl; 2673 dmu_tx_t *tx; 2674 vnode_t *realvp; 2675 int error; 2676 2677 ASSERT(tdvp->v_type == VDIR); 2678 2679 ZFS_ENTER(zfsvfs); 2680 2681 if (VOP_REALVP(svp, &realvp) == 0) 2682 svp = realvp; 2683 2684 if (svp->v_vfsp != tdvp->v_vfsp) { 2685 ZFS_EXIT(zfsvfs); 2686 return (EXDEV); 2687 } 2688 2689 szp = VTOZ(svp); 2690 top: 2691 /* 2692 * We do not support links between attributes and non-attributes 2693 * because of the potential security risk of creating links 2694 * into "normal" file space in order to circumvent restrictions 2695 * imposed in attribute space. 2696 */ 2697 if ((szp->z_phys->zp_flags & ZFS_XATTR) != 2698 (dzp->z_phys->zp_flags & ZFS_XATTR)) { 2699 ZFS_EXIT(zfsvfs); 2700 return (EINVAL); 2701 } 2702 2703 /* 2704 * POSIX dictates that we return EPERM here. 2705 * Better choices include ENOTSUP or EISDIR. 2706 */ 2707 if (svp->v_type == VDIR) { 2708 ZFS_EXIT(zfsvfs); 2709 return (EPERM); 2710 } 2711 2712 if ((uid_t)szp->z_phys->zp_uid != crgetuid(cr) && 2713 secpolicy_basic_link(cr) != 0) { 2714 ZFS_EXIT(zfsvfs); 2715 return (EPERM); 2716 } 2717 2718 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 2719 ZFS_EXIT(zfsvfs); 2720 return (error); 2721 } 2722 2723 /* 2724 * Attempt to lock directory; fail if entry already exists. 2725 */ 2726 if (error = zfs_dirent_lock(&dl, dzp, name, &tzp, ZNEW)) { 2727 ZFS_EXIT(zfsvfs); 2728 return (error); 2729 } 2730 2731 tx = dmu_tx_create(zfsvfs->z_os); 2732 dmu_tx_hold_bonus(tx, szp->z_id); 2733 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 2734 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2735 if (error) { 2736 zfs_dirent_unlock(dl); 2737 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2738 dmu_tx_wait(tx); 2739 dmu_tx_abort(tx); 2740 goto top; 2741 } 2742 dmu_tx_abort(tx); 2743 ZFS_EXIT(zfsvfs); 2744 return (error); 2745 } 2746 2747 error = zfs_link_create(dl, szp, tx, 0); 2748 2749 if (error == 0) 2750 zfs_log_link(zilog, tx, TX_LINK, dzp, szp, name); 2751 2752 dmu_tx_commit(tx); 2753 2754 zfs_dirent_unlock(dl); 2755 2756 ZFS_EXIT(zfsvfs); 2757 return (error); 2758 } 2759 2760 /* 2761 * zfs_null_putapage() is used when the file system has been force 2762 * unmounted. It just drops the pages. 2763 */ 2764 /* ARGSUSED */ 2765 static int 2766 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 2767 size_t *lenp, int flags, cred_t *cr) 2768 { 2769 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR); 2770 return (0); 2771 } 2772 2773 /* 2774 * Push a page out to disk, klustering if possible. 2775 * 2776 * IN: vp - file to push page to. 2777 * pp - page to push. 2778 * flags - additional flags. 2779 * cr - credentials of caller. 2780 * 2781 * OUT: offp - start of range pushed. 2782 * lenp - len of range pushed. 2783 * 2784 * RETURN: 0 if success 2785 * error code if failure 2786 * 2787 * NOTE: callers must have locked the page to be pushed. On 2788 * exit, the page (and all other pages in the kluster) must be 2789 * unlocked. 2790 */ 2791 /* ARGSUSED */ 2792 static int 2793 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 2794 size_t *lenp, int flags, cred_t *cr) 2795 { 2796 znode_t *zp = VTOZ(vp); 2797 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2798 zilog_t *zilog = zfsvfs->z_log; 2799 dmu_tx_t *tx; 2800 rl_t *rl; 2801 u_offset_t off, koff; 2802 size_t len, klen; 2803 int err; 2804 2805 off = pp->p_offset; 2806 len = PAGESIZE; 2807 /* 2808 * If our blocksize is bigger than the page size, try to kluster 2809 * muiltiple pages so that we write a full block (thus avoiding 2810 * a read-modify-write). 2811 */ 2812 if (zp->z_blksz > PAGESIZE) { 2813 uint64_t filesz = zp->z_phys->zp_size; 2814 2815 if (!ISP2(zp->z_blksz)) { 2816 /* Only one block in the file. */ 2817 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 2818 koff = 0; 2819 } else { 2820 klen = zp->z_blksz; 2821 koff = P2ALIGN(off, (u_offset_t)klen); 2822 } 2823 ASSERT(koff <= filesz); 2824 if (koff + klen > filesz) 2825 klen = P2ROUNDUP(filesz - koff, (uint64_t)PAGESIZE); 2826 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags); 2827 } 2828 ASSERT3U(btop(len), ==, btopr(len)); 2829 top: 2830 rl = zfs_range_lock(zp, off, len, RL_WRITER); 2831 /* 2832 * Can't push pages past end-of-file. 2833 */ 2834 if (off >= zp->z_phys->zp_size) { 2835 /* discard all pages */ 2836 flags |= B_INVAL; 2837 err = 0; 2838 goto out; 2839 } else if (off + len > zp->z_phys->zp_size) { 2840 int npages = btopr(zp->z_phys->zp_size - off); 2841 page_t *trunc; 2842 2843 page_list_break(&pp, &trunc, npages); 2844 /* discard pages past end of file */ 2845 if (trunc) 2846 pvn_write_done(trunc, B_INVAL | flags); 2847 len = zp->z_phys->zp_size - off; 2848 } 2849 2850 tx = dmu_tx_create(zfsvfs->z_os); 2851 dmu_tx_hold_write(tx, zp->z_id, off, len); 2852 dmu_tx_hold_bonus(tx, zp->z_id); 2853 err = dmu_tx_assign(tx, zfsvfs->z_assign); 2854 if (err != 0) { 2855 if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2856 zfs_range_unlock(rl); 2857 dmu_tx_wait(tx); 2858 dmu_tx_abort(tx); 2859 err = 0; 2860 goto top; 2861 } 2862 dmu_tx_abort(tx); 2863 goto out; 2864 } 2865 2866 if (zp->z_blksz <= PAGESIZE) { 2867 caddr_t va = ppmapin(pp, PROT_READ, (caddr_t)-1); 2868 ASSERT3U(len, <=, PAGESIZE); 2869 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx); 2870 ppmapout(va); 2871 } else { 2872 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx); 2873 } 2874 2875 if (err == 0) { 2876 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 2877 (void) zfs_log_write( 2878 zilog, tx, TX_WRITE, zp, off, len, 0, NULL); 2879 dmu_tx_commit(tx); 2880 } 2881 2882 out: 2883 zfs_range_unlock(rl); 2884 pvn_write_done(pp, (err ? B_ERROR : 0) | B_WRITE | flags); 2885 if (offp) 2886 *offp = off; 2887 if (lenp) 2888 *lenp = len; 2889 2890 return (err); 2891 } 2892 2893 /* 2894 * Copy the portion of the file indicated from pages into the file. 2895 * The pages are stored in a page list attached to the files vnode. 2896 * 2897 * IN: vp - vnode of file to push page data to. 2898 * off - position in file to put data. 2899 * len - amount of data to write. 2900 * flags - flags to control the operation. 2901 * cr - credentials of caller. 2902 * 2903 * RETURN: 0 if success 2904 * error code if failure 2905 * 2906 * Timestamps: 2907 * vp - ctime|mtime updated 2908 */ 2909 static int 2910 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr) 2911 { 2912 znode_t *zp = VTOZ(vp); 2913 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2914 page_t *pp; 2915 size_t io_len; 2916 u_offset_t io_off; 2917 uint64_t filesz; 2918 int error = 0; 2919 2920 ZFS_ENTER(zfsvfs); 2921 2922 ASSERT(zp->z_dbuf_held && zp->z_phys); 2923 2924 if (len == 0) { 2925 /* 2926 * Search the entire vp list for pages >= off. 2927 */ 2928 error = pvn_vplist_dirty(vp, (u_offset_t)off, zfs_putapage, 2929 flags, cr); 2930 goto out; 2931 } 2932 2933 filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */ 2934 if (off > filesz) { 2935 /* past end of file */ 2936 ZFS_EXIT(zfsvfs); 2937 return (0); 2938 } 2939 2940 len = MIN(len, filesz - off); 2941 2942 for (io_off = off; io_off < off + len; io_off += io_len) { 2943 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { 2944 pp = page_lookup(vp, io_off, 2945 (flags & (B_INVAL | B_FREE)) ? 2946 SE_EXCL : SE_SHARED); 2947 } else { 2948 pp = page_lookup_nowait(vp, io_off, 2949 (flags & B_FREE) ? SE_EXCL : SE_SHARED); 2950 } 2951 2952 if (pp != NULL && pvn_getdirty(pp, flags)) { 2953 int err; 2954 2955 /* 2956 * Found a dirty page to push 2957 */ 2958 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr); 2959 if (err) 2960 error = err; 2961 } else { 2962 io_len = PAGESIZE; 2963 } 2964 } 2965 out: 2966 if ((flags & B_ASYNC) == 0) 2967 zil_commit(zfsvfs->z_log, UINT64_MAX, zp->z_id); 2968 ZFS_EXIT(zfsvfs); 2969 return (error); 2970 } 2971 2972 void 2973 zfs_inactive(vnode_t *vp, cred_t *cr) 2974 { 2975 znode_t *zp = VTOZ(vp); 2976 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2977 int error; 2978 2979 rw_enter(&zfsvfs->z_um_lock, RW_READER); 2980 if (zfsvfs->z_unmounted2) { 2981 ASSERT(zp->z_dbuf_held == 0); 2982 2983 if (vn_has_cached_data(vp)) { 2984 (void) pvn_vplist_dirty(vp, 0, zfs_null_putapage, 2985 B_INVAL, cr); 2986 } 2987 2988 mutex_enter(&zp->z_lock); 2989 vp->v_count = 0; /* count arrives as 1 */ 2990 if (zp->z_dbuf == NULL) { 2991 mutex_exit(&zp->z_lock); 2992 zfs_znode_free(zp); 2993 } else { 2994 mutex_exit(&zp->z_lock); 2995 } 2996 rw_exit(&zfsvfs->z_um_lock); 2997 VFS_RELE(zfsvfs->z_vfs); 2998 return; 2999 } 3000 3001 /* 3002 * Attempt to push any data in the page cache. If this fails 3003 * we will get kicked out later in zfs_zinactive(). 3004 */ 3005 if (vn_has_cached_data(vp)) { 3006 (void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC, 3007 cr); 3008 } 3009 3010 if (zp->z_atime_dirty && zp->z_reap == 0) { 3011 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 3012 3013 dmu_tx_hold_bonus(tx, zp->z_id); 3014 error = dmu_tx_assign(tx, TXG_WAIT); 3015 if (error) { 3016 dmu_tx_abort(tx); 3017 } else { 3018 dmu_buf_will_dirty(zp->z_dbuf, tx); 3019 mutex_enter(&zp->z_lock); 3020 zp->z_atime_dirty = 0; 3021 mutex_exit(&zp->z_lock); 3022 dmu_tx_commit(tx); 3023 } 3024 } 3025 3026 zfs_zinactive(zp); 3027 rw_exit(&zfsvfs->z_um_lock); 3028 } 3029 3030 /* 3031 * Bounds-check the seek operation. 3032 * 3033 * IN: vp - vnode seeking within 3034 * ooff - old file offset 3035 * noffp - pointer to new file offset 3036 * 3037 * RETURN: 0 if success 3038 * EINVAL if new offset invalid 3039 */ 3040 /* ARGSUSED */ 3041 static int 3042 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp) 3043 { 3044 if (vp->v_type == VDIR) 3045 return (0); 3046 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0); 3047 } 3048 3049 /* 3050 * Pre-filter the generic locking function to trap attempts to place 3051 * a mandatory lock on a memory mapped file. 3052 */ 3053 static int 3054 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset, 3055 flk_callback_t *flk_cbp, cred_t *cr) 3056 { 3057 znode_t *zp = VTOZ(vp); 3058 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3059 int error; 3060 3061 ZFS_ENTER(zfsvfs); 3062 3063 /* 3064 * We are following the UFS semantics with respect to mapcnt 3065 * here: If we see that the file is mapped already, then we will 3066 * return an error, but we don't worry about races between this 3067 * function and zfs_map(). 3068 */ 3069 if (zp->z_mapcnt > 0 && MANDMODE((mode_t)zp->z_phys->zp_mode)) { 3070 ZFS_EXIT(zfsvfs); 3071 return (EAGAIN); 3072 } 3073 error = fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr); 3074 ZFS_EXIT(zfsvfs); 3075 return (error); 3076 } 3077 3078 /* 3079 * If we can't find a page in the cache, we will create a new page 3080 * and fill it with file data. For efficiency, we may try to fill 3081 * multiple pages at once (klustering). 3082 */ 3083 static int 3084 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, 3085 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw) 3086 { 3087 znode_t *zp = VTOZ(vp); 3088 page_t *pp, *cur_pp; 3089 objset_t *os = zp->z_zfsvfs->z_os; 3090 caddr_t va; 3091 u_offset_t io_off, total; 3092 uint64_t oid = zp->z_id; 3093 size_t io_len; 3094 uint64_t filesz; 3095 int err; 3096 3097 /* 3098 * If we are only asking for a single page don't bother klustering. 3099 */ 3100 filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */ 3101 if (off >= filesz) 3102 return (EFAULT); 3103 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) { 3104 io_off = off; 3105 io_len = PAGESIZE; 3106 pp = page_create_va(vp, io_off, io_len, PG_WAIT, seg, addr); 3107 } else { 3108 /* 3109 * Try to fill a kluster of pages (a blocks worth). 3110 */ 3111 size_t klen; 3112 u_offset_t koff; 3113 3114 if (!ISP2(zp->z_blksz)) { 3115 /* Only one block in the file. */ 3116 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 3117 koff = 0; 3118 } else { 3119 klen = plsz; 3120 koff = P2ALIGN(off, (u_offset_t)klen); 3121 } 3122 ASSERT(koff <= filesz); 3123 if (koff + klen > filesz) 3124 klen = P2ROUNDUP(filesz, (uint64_t)PAGESIZE) - koff; 3125 ASSERT3U(off, >=, koff); 3126 ASSERT3U(off, <, koff + klen); 3127 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 3128 &io_len, koff, klen, 0); 3129 } 3130 if (pp == NULL) { 3131 /* 3132 * Some other thread entered the page before us. 3133 * Return to zfs_getpage to retry the lookup. 3134 */ 3135 *pl = NULL; 3136 return (0); 3137 } 3138 3139 /* 3140 * Fill the pages in the kluster. 3141 */ 3142 cur_pp = pp; 3143 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) { 3144 ASSERT3U(io_off, ==, cur_pp->p_offset); 3145 va = ppmapin(cur_pp, PROT_READ | PROT_WRITE, (caddr_t)-1); 3146 err = dmu_read(os, oid, io_off, PAGESIZE, va); 3147 ppmapout(va); 3148 if (err) { 3149 /* On error, toss the entire kluster */ 3150 pvn_read_done(pp, B_ERROR); 3151 return (err); 3152 } 3153 cur_pp = cur_pp->p_next; 3154 } 3155 out: 3156 /* 3157 * Fill in the page list array from the kluster. If 3158 * there are too many pages in the kluster, return 3159 * as many pages as possible starting from the desired 3160 * offset `off'. 3161 * NOTE: the page list will always be null terminated. 3162 */ 3163 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 3164 3165 return (0); 3166 } 3167 3168 /* 3169 * Return pointers to the pages for the file region [off, off + len] 3170 * in the pl array. If plsz is greater than len, this function may 3171 * also return page pointers from before or after the specified 3172 * region (i.e. some region [off', off' + plsz]). These additional 3173 * pages are only returned if they are already in the cache, or were 3174 * created as part of a klustered read. 3175 * 3176 * IN: vp - vnode of file to get data from. 3177 * off - position in file to get data from. 3178 * len - amount of data to retrieve. 3179 * plsz - length of provided page list. 3180 * seg - segment to obtain pages for. 3181 * addr - virtual address of fault. 3182 * rw - mode of created pages. 3183 * cr - credentials of caller. 3184 * 3185 * OUT: protp - protection mode of created pages. 3186 * pl - list of pages created. 3187 * 3188 * RETURN: 0 if success 3189 * error code if failure 3190 * 3191 * Timestamps: 3192 * vp - atime updated 3193 */ 3194 /* ARGSUSED */ 3195 static int 3196 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 3197 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 3198 enum seg_rw rw, cred_t *cr) 3199 { 3200 znode_t *zp = VTOZ(vp); 3201 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3202 page_t *pp, **pl0 = pl; 3203 int need_unlock = 0, err = 0; 3204 offset_t orig_off; 3205 3206 ZFS_ENTER(zfsvfs); 3207 3208 if (protp) 3209 *protp = PROT_ALL; 3210 3211 ASSERT(zp->z_dbuf_held && zp->z_phys); 3212 3213 /* no faultahead (for now) */ 3214 if (pl == NULL) { 3215 ZFS_EXIT(zfsvfs); 3216 return (0); 3217 } 3218 3219 /* can't fault past EOF */ 3220 if (off >= zp->z_phys->zp_size) { 3221 ZFS_EXIT(zfsvfs); 3222 return (EFAULT); 3223 } 3224 orig_off = off; 3225 3226 /* 3227 * If we already own the lock, then we must be page faulting 3228 * in the middle of a write to this file (i.e., we are writing 3229 * to this file using data from a mapped region of the file). 3230 */ 3231 if (rw_owner(&zp->z_map_lock) != curthread) { 3232 rw_enter(&zp->z_map_lock, RW_WRITER); 3233 need_unlock = TRUE; 3234 } 3235 3236 /* 3237 * Loop through the requested range [off, off + len] looking 3238 * for pages. If we don't find a page, we will need to create 3239 * a new page and fill it with data from the file. 3240 */ 3241 while (len > 0) { 3242 if (plsz < PAGESIZE) 3243 break; 3244 if (pp = page_lookup(vp, off, SE_SHARED)) { 3245 *pl++ = pp; 3246 off += PAGESIZE; 3247 addr += PAGESIZE; 3248 len -= PAGESIZE; 3249 plsz -= PAGESIZE; 3250 } else { 3251 err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw); 3252 if (err) 3253 goto out; 3254 /* 3255 * klustering may have changed our region 3256 * to be block aligned. 3257 */ 3258 if (((pp = *pl) != 0) && (off != pp->p_offset)) { 3259 int delta = off - pp->p_offset; 3260 len += delta; 3261 off -= delta; 3262 addr -= delta; 3263 } 3264 while (*pl) { 3265 pl++; 3266 off += PAGESIZE; 3267 addr += PAGESIZE; 3268 plsz -= PAGESIZE; 3269 if (len > PAGESIZE) 3270 len -= PAGESIZE; 3271 else 3272 len = 0; 3273 } 3274 } 3275 } 3276 3277 /* 3278 * Fill out the page array with any pages already in the cache. 3279 */ 3280 while (plsz > 0) { 3281 pp = page_lookup_nowait(vp, off, SE_SHARED); 3282 if (pp == NULL) 3283 break; 3284 *pl++ = pp; 3285 off += PAGESIZE; 3286 plsz -= PAGESIZE; 3287 } 3288 3289 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 3290 out: 3291 /* 3292 * We can't grab the range lock for the page as reader which would 3293 * stop truncation as this leads to deadlock. So we need to recheck 3294 * the file size. 3295 */ 3296 if (orig_off >= zp->z_phys->zp_size) 3297 err = EFAULT; 3298 if (err) { 3299 /* 3300 * Release any pages we have previously locked. 3301 */ 3302 while (pl > pl0) 3303 page_unlock(*--pl); 3304 } 3305 3306 *pl = NULL; 3307 3308 if (need_unlock) 3309 rw_exit(&zp->z_map_lock); 3310 3311 ZFS_EXIT(zfsvfs); 3312 return (err); 3313 } 3314 3315 /* 3316 * Request a memory map for a section of a file. This code interacts 3317 * with common code and the VM system as follows: 3318 * 3319 * common code calls mmap(), which ends up in smmap_common() 3320 * 3321 * this calls VOP_MAP(), which takes you into (say) zfs 3322 * 3323 * zfs_map() calls as_map(), passing segvn_create() as the callback 3324 * 3325 * segvn_create() creates the new segment and calls VOP_ADDMAP() 3326 * 3327 * zfs_addmap() updates z_mapcnt 3328 */ 3329 static int 3330 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 3331 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr) 3332 { 3333 znode_t *zp = VTOZ(vp); 3334 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3335 segvn_crargs_t vn_a; 3336 int error; 3337 3338 ZFS_ENTER(zfsvfs); 3339 3340 if (vp->v_flag & VNOMAP) { 3341 ZFS_EXIT(zfsvfs); 3342 return (ENOSYS); 3343 } 3344 3345 if (off < 0 || len > MAXOFFSET_T - off) { 3346 ZFS_EXIT(zfsvfs); 3347 return (ENXIO); 3348 } 3349 3350 if (vp->v_type != VREG) { 3351 ZFS_EXIT(zfsvfs); 3352 return (ENODEV); 3353 } 3354 3355 /* 3356 * If file is locked, disallow mapping. 3357 */ 3358 if (MANDMODE((mode_t)zp->z_phys->zp_mode) && vn_has_flocks(vp)) { 3359 ZFS_EXIT(zfsvfs); 3360 return (EAGAIN); 3361 } 3362 3363 as_rangelock(as); 3364 if ((flags & MAP_FIXED) == 0) { 3365 map_addr(addrp, len, off, 1, flags); 3366 if (*addrp == NULL) { 3367 as_rangeunlock(as); 3368 ZFS_EXIT(zfsvfs); 3369 return (ENOMEM); 3370 } 3371 } else { 3372 /* 3373 * User specified address - blow away any previous mappings 3374 */ 3375 (void) as_unmap(as, *addrp, len); 3376 } 3377 3378 vn_a.vp = vp; 3379 vn_a.offset = (u_offset_t)off; 3380 vn_a.type = flags & MAP_TYPE; 3381 vn_a.prot = prot; 3382 vn_a.maxprot = maxprot; 3383 vn_a.cred = cr; 3384 vn_a.amp = NULL; 3385 vn_a.flags = flags & ~MAP_TYPE; 3386 vn_a.szc = 0; 3387 vn_a.lgrp_mem_policy_flags = 0; 3388 3389 error = as_map(as, *addrp, len, segvn_create, &vn_a); 3390 3391 as_rangeunlock(as); 3392 ZFS_EXIT(zfsvfs); 3393 return (error); 3394 } 3395 3396 /* ARGSUSED */ 3397 static int 3398 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 3399 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr) 3400 { 3401 uint64_t pages = btopr(len); 3402 3403 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages); 3404 return (0); 3405 } 3406 3407 /* 3408 * The reason we push dirty pages as part of zfs_delmap() is so that we get a 3409 * more accurate mtime for the associated file. Since we don't have a way of 3410 * detecting when the data was actually modified, we have to resort to 3411 * heuristics. If an explicit msync() is done, then we mark the mtime when the 3412 * last page is pushed. The problem occurs when the msync() call is omitted, 3413 * which by far the most common case: 3414 * 3415 * open() 3416 * mmap() 3417 * <modify memory> 3418 * munmap() 3419 * close() 3420 * <time lapse> 3421 * putpage() via fsflush 3422 * 3423 * If we wait until fsflush to come along, we can have a modification time that 3424 * is some arbitrary point in the future. In order to prevent this in the 3425 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is 3426 * torn down. 3427 */ 3428 /* ARGSUSED */ 3429 static int 3430 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 3431 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr) 3432 { 3433 uint64_t pages = btopr(len); 3434 3435 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages); 3436 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages); 3437 3438 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) && 3439 vn_has_cached_data(vp)) 3440 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr); 3441 3442 return (0); 3443 } 3444 3445 /* 3446 * Free or allocate space in a file. Currently, this function only 3447 * supports the `F_FREESP' command. However, this command is somewhat 3448 * misnamed, as its functionality includes the ability to allocate as 3449 * well as free space. 3450 * 3451 * IN: vp - vnode of file to free data in. 3452 * cmd - action to take (only F_FREESP supported). 3453 * bfp - section of file to free/alloc. 3454 * flag - current file open mode flags. 3455 * offset - current file offset. 3456 * cr - credentials of caller [UNUSED]. 3457 * 3458 * RETURN: 0 if success 3459 * error code if failure 3460 * 3461 * Timestamps: 3462 * vp - ctime|mtime updated 3463 */ 3464 /* ARGSUSED */ 3465 static int 3466 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag, 3467 offset_t offset, cred_t *cr, caller_context_t *ct) 3468 { 3469 znode_t *zp = VTOZ(vp); 3470 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3471 uint64_t off, len; 3472 int error; 3473 3474 ZFS_ENTER(zfsvfs); 3475 3476 top: 3477 if (cmd != F_FREESP) { 3478 ZFS_EXIT(zfsvfs); 3479 return (EINVAL); 3480 } 3481 3482 if (error = convoff(vp, bfp, 0, offset)) { 3483 ZFS_EXIT(zfsvfs); 3484 return (error); 3485 } 3486 3487 if (bfp->l_len < 0) { 3488 ZFS_EXIT(zfsvfs); 3489 return (EINVAL); 3490 } 3491 3492 off = bfp->l_start; 3493 len = bfp->l_len; /* 0 means from off to end of file */ 3494 3495 do { 3496 error = zfs_freesp(zp, off, len, flag, TRUE); 3497 /* NB: we already did dmu_tx_wait() if necessary */ 3498 } while (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT); 3499 3500 ZFS_EXIT(zfsvfs); 3501 return (error); 3502 } 3503 3504 static int 3505 zfs_fid(vnode_t *vp, fid_t *fidp) 3506 { 3507 znode_t *zp = VTOZ(vp); 3508 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3509 uint32_t gen = (uint32_t)zp->z_phys->zp_gen; 3510 uint64_t object = zp->z_id; 3511 zfid_short_t *zfid; 3512 int size, i; 3513 3514 ZFS_ENTER(zfsvfs); 3515 3516 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 3517 if (fidp->fid_len < size) { 3518 fidp->fid_len = size; 3519 ZFS_EXIT(zfsvfs); 3520 return (ENOSPC); 3521 } 3522 3523 zfid = (zfid_short_t *)fidp; 3524 3525 zfid->zf_len = size; 3526 3527 for (i = 0; i < sizeof (zfid->zf_object); i++) 3528 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 3529 3530 /* Must have a non-zero generation number to distinguish from .zfs */ 3531 if (gen == 0) 3532 gen = 1; 3533 for (i = 0; i < sizeof (zfid->zf_gen); i++) 3534 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 3535 3536 if (size == LONG_FID_LEN) { 3537 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 3538 zfid_long_t *zlfid; 3539 3540 zlfid = (zfid_long_t *)fidp; 3541 3542 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 3543 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 3544 3545 /* XXX - this should be the generation number for the objset */ 3546 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 3547 zlfid->zf_setgen[i] = 0; 3548 } 3549 3550 ZFS_EXIT(zfsvfs); 3551 return (0); 3552 } 3553 3554 static int 3555 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr) 3556 { 3557 znode_t *zp, *xzp; 3558 zfsvfs_t *zfsvfs; 3559 zfs_dirlock_t *dl; 3560 int error; 3561 3562 switch (cmd) { 3563 case _PC_LINK_MAX: 3564 *valp = ULONG_MAX; 3565 return (0); 3566 3567 case _PC_FILESIZEBITS: 3568 *valp = 64; 3569 return (0); 3570 3571 case _PC_XATTR_EXISTS: 3572 zp = VTOZ(vp); 3573 zfsvfs = zp->z_zfsvfs; 3574 ZFS_ENTER(zfsvfs); 3575 *valp = 0; 3576 error = zfs_dirent_lock(&dl, zp, "", &xzp, 3577 ZXATTR | ZEXISTS | ZSHARED); 3578 if (error == 0) { 3579 zfs_dirent_unlock(dl); 3580 if (!zfs_dirempty(xzp)) 3581 *valp = 1; 3582 VN_RELE(ZTOV(xzp)); 3583 } else if (error == ENOENT) { 3584 /* 3585 * If there aren't extended attributes, it's the 3586 * same as having zero of them. 3587 */ 3588 error = 0; 3589 } 3590 ZFS_EXIT(zfsvfs); 3591 return (error); 3592 3593 case _PC_ACL_ENABLED: 3594 *valp = _ACL_ACE_ENABLED; 3595 return (0); 3596 3597 case _PC_MIN_HOLE_SIZE: 3598 *valp = (ulong_t)SPA_MINBLOCKSIZE; 3599 return (0); 3600 3601 default: 3602 return (fs_pathconf(vp, cmd, valp, cr)); 3603 } 3604 } 3605 3606 /*ARGSUSED*/ 3607 static int 3608 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr) 3609 { 3610 znode_t *zp = VTOZ(vp); 3611 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3612 int error; 3613 3614 ZFS_ENTER(zfsvfs); 3615 error = zfs_getacl(zp, vsecp, cr); 3616 ZFS_EXIT(zfsvfs); 3617 3618 return (error); 3619 } 3620 3621 /*ARGSUSED*/ 3622 static int 3623 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr) 3624 { 3625 znode_t *zp = VTOZ(vp); 3626 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3627 int error; 3628 3629 ZFS_ENTER(zfsvfs); 3630 error = zfs_setacl(zp, vsecp, cr); 3631 ZFS_EXIT(zfsvfs); 3632 return (error); 3633 } 3634 3635 /* 3636 * Predeclare these here so that the compiler assumes that 3637 * this is an "old style" function declaration that does 3638 * not include arguments => we won't get type mismatch errors 3639 * in the initializations that follow. 3640 */ 3641 static int zfs_inval(); 3642 static int zfs_isdir(); 3643 3644 static int 3645 zfs_inval() 3646 { 3647 return (EINVAL); 3648 } 3649 3650 static int 3651 zfs_isdir() 3652 { 3653 return (EISDIR); 3654 } 3655 /* 3656 * Directory vnode operations template 3657 */ 3658 vnodeops_t *zfs_dvnodeops; 3659 const fs_operation_def_t zfs_dvnodeops_template[] = { 3660 VOPNAME_OPEN, zfs_open, 3661 VOPNAME_CLOSE, zfs_close, 3662 VOPNAME_READ, zfs_isdir, 3663 VOPNAME_WRITE, zfs_isdir, 3664 VOPNAME_IOCTL, zfs_ioctl, 3665 VOPNAME_GETATTR, zfs_getattr, 3666 VOPNAME_SETATTR, zfs_setattr, 3667 VOPNAME_ACCESS, zfs_access, 3668 VOPNAME_LOOKUP, zfs_lookup, 3669 VOPNAME_CREATE, zfs_create, 3670 VOPNAME_REMOVE, zfs_remove, 3671 VOPNAME_LINK, zfs_link, 3672 VOPNAME_RENAME, zfs_rename, 3673 VOPNAME_MKDIR, zfs_mkdir, 3674 VOPNAME_RMDIR, zfs_rmdir, 3675 VOPNAME_READDIR, zfs_readdir, 3676 VOPNAME_SYMLINK, zfs_symlink, 3677 VOPNAME_FSYNC, zfs_fsync, 3678 VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive, 3679 VOPNAME_FID, zfs_fid, 3680 VOPNAME_SEEK, zfs_seek, 3681 VOPNAME_PATHCONF, zfs_pathconf, 3682 VOPNAME_GETSECATTR, zfs_getsecattr, 3683 VOPNAME_SETSECATTR, zfs_setsecattr, 3684 NULL, NULL 3685 }; 3686 3687 /* 3688 * Regular file vnode operations template 3689 */ 3690 vnodeops_t *zfs_fvnodeops; 3691 const fs_operation_def_t zfs_fvnodeops_template[] = { 3692 VOPNAME_OPEN, zfs_open, 3693 VOPNAME_CLOSE, zfs_close, 3694 VOPNAME_READ, zfs_read, 3695 VOPNAME_WRITE, zfs_write, 3696 VOPNAME_IOCTL, zfs_ioctl, 3697 VOPNAME_GETATTR, zfs_getattr, 3698 VOPNAME_SETATTR, zfs_setattr, 3699 VOPNAME_ACCESS, zfs_access, 3700 VOPNAME_LOOKUP, zfs_lookup, 3701 VOPNAME_RENAME, zfs_rename, 3702 VOPNAME_FSYNC, zfs_fsync, 3703 VOPNAME_INACTIVE, (fs_generic_func_p)zfs_inactive, 3704 VOPNAME_FID, zfs_fid, 3705 VOPNAME_SEEK, zfs_seek, 3706 VOPNAME_FRLOCK, zfs_frlock, 3707 VOPNAME_SPACE, zfs_space, 3708 VOPNAME_GETPAGE, zfs_getpage, 3709 VOPNAME_PUTPAGE, zfs_putpage, 3710 VOPNAME_MAP, (fs_generic_func_p) zfs_map, 3711 VOPNAME_ADDMAP, (fs_generic_func_p) zfs_addmap, 3712 VOPNAME_DELMAP, zfs_delmap, 3713 VOPNAME_PATHCONF, zfs_pathconf, 3714 VOPNAME_GETSECATTR, zfs_getsecattr, 3715 VOPNAME_SETSECATTR, zfs_setsecattr, 3716 VOPNAME_VNEVENT, fs_vnevent_support, 3717 NULL, NULL 3718 }; 3719 3720 /* 3721 * Symbolic link vnode operations template 3722 */ 3723 vnodeops_t *zfs_symvnodeops; 3724 const fs_operation_def_t zfs_symvnodeops_template[] = { 3725 VOPNAME_GETATTR, zfs_getattr, 3726 VOPNAME_SETATTR, zfs_setattr, 3727 VOPNAME_ACCESS, zfs_access, 3728 VOPNAME_RENAME, zfs_rename, 3729 VOPNAME_READLINK, zfs_readlink, 3730 VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive, 3731 VOPNAME_FID, zfs_fid, 3732 VOPNAME_PATHCONF, zfs_pathconf, 3733 VOPNAME_VNEVENT, fs_vnevent_support, 3734 NULL, NULL 3735 }; 3736 3737 /* 3738 * Extended attribute directory vnode operations template 3739 * This template is identical to the directory vnodes 3740 * operation template except for restricted operations: 3741 * VOP_MKDIR() 3742 * VOP_SYMLINK() 3743 * Note that there are other restrictions embedded in: 3744 * zfs_create() - restrict type to VREG 3745 * zfs_link() - no links into/out of attribute space 3746 * zfs_rename() - no moves into/out of attribute space 3747 */ 3748 vnodeops_t *zfs_xdvnodeops; 3749 const fs_operation_def_t zfs_xdvnodeops_template[] = { 3750 VOPNAME_OPEN, zfs_open, 3751 VOPNAME_CLOSE, zfs_close, 3752 VOPNAME_IOCTL, zfs_ioctl, 3753 VOPNAME_GETATTR, zfs_getattr, 3754 VOPNAME_SETATTR, zfs_setattr, 3755 VOPNAME_ACCESS, zfs_access, 3756 VOPNAME_LOOKUP, zfs_lookup, 3757 VOPNAME_CREATE, zfs_create, 3758 VOPNAME_REMOVE, zfs_remove, 3759 VOPNAME_LINK, zfs_link, 3760 VOPNAME_RENAME, zfs_rename, 3761 VOPNAME_MKDIR, zfs_inval, 3762 VOPNAME_RMDIR, zfs_rmdir, 3763 VOPNAME_READDIR, zfs_readdir, 3764 VOPNAME_SYMLINK, zfs_inval, 3765 VOPNAME_FSYNC, zfs_fsync, 3766 VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive, 3767 VOPNAME_FID, zfs_fid, 3768 VOPNAME_SEEK, zfs_seek, 3769 VOPNAME_PATHCONF, zfs_pathconf, 3770 VOPNAME_GETSECATTR, zfs_getsecattr, 3771 VOPNAME_SETSECATTR, zfs_setsecattr, 3772 VOPNAME_VNEVENT, fs_vnevent_support, 3773 NULL, NULL 3774 }; 3775 3776 /* 3777 * Error vnode operations template 3778 */ 3779 vnodeops_t *zfs_evnodeops; 3780 const fs_operation_def_t zfs_evnodeops_template[] = { 3781 VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive, 3782 VOPNAME_PATHCONF, zfs_pathconf, 3783 NULL, NULL 3784 }; 3785