1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright 2006 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 #pragma ident "%Z%%M% %I% %E% SMI" 27 28 #include <sys/types.h> 29 #include <sys/param.h> 30 #include <sys/time.h> 31 #include <sys/systm.h> 32 #include <sys/sysmacros.h> 33 #include <sys/resource.h> 34 #include <sys/vfs.h> 35 #include <sys/vnode.h> 36 #include <sys/file.h> 37 #include <sys/stat.h> 38 #include <sys/kmem.h> 39 #include <sys/taskq.h> 40 #include <sys/uio.h> 41 #include <sys/vmsystm.h> 42 #include <sys/atomic.h> 43 #include <vm/seg_vn.h> 44 #include <vm/pvn.h> 45 #include <vm/as.h> 46 #include <sys/mman.h> 47 #include <sys/pathname.h> 48 #include <sys/cmn_err.h> 49 #include <sys/errno.h> 50 #include <sys/unistd.h> 51 #include <sys/zfs_vfsops.h> 52 #include <sys/zfs_dir.h> 53 #include <sys/zfs_acl.h> 54 #include <sys/zfs_ioctl.h> 55 #include <sys/fs/zfs.h> 56 #include <sys/dmu.h> 57 #include <sys/spa.h> 58 #include <sys/txg.h> 59 #include <sys/dbuf.h> 60 #include <sys/zap.h> 61 #include <sys/dirent.h> 62 #include <sys/policy.h> 63 #include <sys/sunddi.h> 64 #include <sys/filio.h> 65 #include "fs/fs_subr.h" 66 #include <sys/zfs_ctldir.h> 67 #include <sys/dnlc.h> 68 #include <sys/zfs_rlock.h> 69 70 /* 71 * Programming rules. 72 * 73 * Each vnode op performs some logical unit of work. To do this, the ZPL must 74 * properly lock its in-core state, create a DMU transaction, do the work, 75 * record this work in the intent log (ZIL), commit the DMU transaction, 76 * and wait the the intent log to commit if it's is a synchronous operation. 77 * Morover, the vnode ops must work in both normal and log replay context. 78 * The ordering of events is important to avoid deadlocks and references 79 * to freed memory. The example below illustrates the following Big Rules: 80 * 81 * (1) A check must be made in each zfs thread for a mounted file system. 82 * This is done avoiding races using ZFS_ENTER(zfsvfs). 83 * A ZFS_EXIT(zfsvfs) is needed before all returns. 84 * 85 * (2) VN_RELE() should always be the last thing except for zil_commit() 86 * and ZFS_EXIT(). This is for 3 reasons: 87 * First, if it's the last reference, the vnode/znode 88 * can be freed, so the zp may point to freed memory. Second, the last 89 * reference will call zfs_zinactive(), which may induce a lot of work -- 90 * pushing cached pages (which acquires range locks) and syncing out 91 * cached atime changes. Third, zfs_zinactive() may require a new tx, 92 * which could deadlock the system if you were already holding one. 93 * 94 * (3) All range locks must be grabbed before calling dmu_tx_assign(), 95 * as they can span dmu_tx_assign() calls. 96 * 97 * (4) Always pass zfsvfs->z_assign as the second argument to dmu_tx_assign(). 98 * In normal operation, this will be TXG_NOWAIT. During ZIL replay, 99 * it will be a specific txg. Either way, dmu_tx_assign() never blocks. 100 * This is critical because we don't want to block while holding locks. 101 * Note, in particular, that if a lock is sometimes acquired before 102 * the tx assigns, and sometimes after (e.g. z_lock), then failing to 103 * use a non-blocking assign can deadlock the system. The scenario: 104 * 105 * Thread A has grabbed a lock before calling dmu_tx_assign(). 106 * Thread B is in an already-assigned tx, and blocks for this lock. 107 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open() 108 * forever, because the previous txg can't quiesce until B's tx commits. 109 * 110 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT, 111 * then drop all locks, call txg_wait_open(), and try again. 112 * 113 * (5) If the operation succeeded, generate the intent log entry for it 114 * before dropping locks. This ensures that the ordering of events 115 * in the intent log matches the order in which they actually occurred. 116 * 117 * (6) At the end of each vnode op, the DMU tx must always commit, 118 * regardless of whether there were any errors. 119 * 120 * (7) After dropping all locks, invoke zil_commit(zilog, seq, ioflag) 121 * to ensure that synchronous semantics are provided when necessary. 122 * 123 * In general, this is how things should be ordered in each vnode op: 124 * 125 * ZFS_ENTER(zfsvfs); // exit if unmounted 126 * top: 127 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD()) 128 * rw_enter(...); // grab any other locks you need 129 * tx = dmu_tx_create(...); // get DMU tx 130 * dmu_tx_hold_*(); // hold each object you might modify 131 * error = dmu_tx_assign(tx, zfsvfs->z_assign); // try to assign 132 * if (error) { 133 * dmu_tx_abort(tx); // abort DMU tx 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 * txg_wait_open(dmu_objset_pool(os), 0); 139 * goto top; 140 * } 141 * ZFS_EXIT(zfsvfs); // finished in zfs 142 * return (error); // really out of space 143 * } 144 * error = do_real_work(); // do whatever this VOP does 145 * if (error == 0) 146 * seq = zfs_log_*(...); // on success, make ZIL entry 147 * dmu_tx_commit(tx); // commit DMU tx -- error or not 148 * rw_exit(...); // drop locks 149 * zfs_dirent_unlock(dl); // unlock directory entry 150 * VN_RELE(...); // release held vnodes 151 * zil_commit(zilog, seq, ioflag); // synchronous when necessary 152 * ZFS_EXIT(zfsvfs); // finished in zfs 153 * return (error); // done, report error 154 */ 155 156 /* ARGSUSED */ 157 static int 158 zfs_open(vnode_t **vpp, int flag, cred_t *cr) 159 { 160 return (0); 161 } 162 163 /* ARGSUSED */ 164 static int 165 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr) 166 { 167 /* 168 * Clean up any locks held by this process on the vp. 169 */ 170 cleanlocks(vp, ddi_get_pid(), 0); 171 cleanshares(vp, ddi_get_pid()); 172 173 return (0); 174 } 175 176 /* 177 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and 178 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter. 179 */ 180 static int 181 zfs_holey(vnode_t *vp, int cmd, offset_t *off) 182 { 183 znode_t *zp = VTOZ(vp); 184 uint64_t noff = (uint64_t)*off; /* new offset */ 185 uint64_t file_sz; 186 int error; 187 boolean_t hole; 188 189 file_sz = zp->z_phys->zp_size; 190 if (noff >= file_sz) { 191 return (ENXIO); 192 } 193 194 if (cmd == _FIO_SEEK_HOLE) 195 hole = B_TRUE; 196 else 197 hole = B_FALSE; 198 199 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff); 200 201 /* end of file? */ 202 if ((error == ESRCH) || (noff > file_sz)) { 203 /* 204 * Handle the virtual hole at the end of file. 205 */ 206 if (hole) { 207 *off = file_sz; 208 return (0); 209 } 210 return (ENXIO); 211 } 212 213 if (noff < *off) 214 return (error); 215 *off = noff; 216 return (error); 217 } 218 219 /* ARGSUSED */ 220 static int 221 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred, 222 int *rvalp) 223 { 224 offset_t off; 225 int error; 226 zfsvfs_t *zfsvfs; 227 228 switch (com) { 229 case _FIOFFS: 230 return (zfs_sync(vp->v_vfsp, 0, cred)); 231 232 /* 233 * The following two ioctls are used by bfu. Faking out, 234 * necessary to avoid bfu errors. 235 */ 236 case _FIOGDIO: 237 case _FIOSDIO: 238 return (0); 239 240 case _FIO_SEEK_DATA: 241 case _FIO_SEEK_HOLE: 242 if (ddi_copyin((void *)data, &off, sizeof (off), flag)) 243 return (EFAULT); 244 245 zfsvfs = VTOZ(vp)->z_zfsvfs; 246 ZFS_ENTER(zfsvfs); 247 248 /* offset parameter is in/out */ 249 error = zfs_holey(vp, com, &off); 250 ZFS_EXIT(zfsvfs); 251 if (error) 252 return (error); 253 if (ddi_copyout(&off, (void *)data, sizeof (off), flag)) 254 return (EFAULT); 255 return (0); 256 } 257 return (ENOTTY); 258 } 259 260 /* 261 * When a file is memory mapped, we must keep the IO data synchronized 262 * between the DMU cache and the memory mapped pages. What this means: 263 * 264 * On Write: If we find a memory mapped page, we write to *both* 265 * the page and the dmu buffer. 266 * 267 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 268 * the file is memory mapped. 269 */ 270 static int 271 mappedwrite(vnode_t *vp, uint64_t woff, int nbytes, uio_t *uio, dmu_tx_t *tx) 272 { 273 znode_t *zp = VTOZ(vp); 274 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 275 int64_t start, off; 276 int len = nbytes; 277 int error = 0; 278 279 start = uio->uio_loffset; 280 off = start & PAGEOFFSET; 281 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 282 page_t *pp; 283 uint64_t bytes = MIN(PAGESIZE - off, len); 284 285 /* 286 * We don't want a new page to "appear" in the middle of 287 * the file update (because it may not get the write 288 * update data), so we grab a lock to block 289 * zfs_getpage(). 290 */ 291 rw_enter(&zp->z_map_lock, RW_WRITER); 292 if (pp = page_lookup(vp, start, SE_SHARED)) { 293 caddr_t va; 294 295 rw_exit(&zp->z_map_lock); 296 va = ppmapin(pp, PROT_READ | PROT_WRITE, (caddr_t)-1L); 297 error = uiomove(va+off, bytes, UIO_WRITE, uio); 298 if (error == 0) { 299 dmu_write(zfsvfs->z_os, zp->z_id, 300 woff, bytes, va+off, tx); 301 } 302 ppmapout(va); 303 page_unlock(pp); 304 } else { 305 error = dmu_write_uio(zfsvfs->z_os, zp->z_id, 306 woff, bytes, uio, tx); 307 rw_exit(&zp->z_map_lock); 308 } 309 len -= bytes; 310 woff += bytes; 311 off = 0; 312 if (error) 313 break; 314 } 315 return (error); 316 } 317 318 /* 319 * When a file is memory mapped, we must keep the IO data synchronized 320 * between the DMU cache and the memory mapped pages. What this means: 321 * 322 * On Read: We "read" preferentially from memory mapped pages, 323 * else we default from the dmu buffer. 324 * 325 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 326 * the file is memory mapped. 327 */ 328 static int 329 mappedread(vnode_t *vp, char *addr, int nbytes, uio_t *uio) 330 { 331 int64_t start, off, bytes; 332 int len = nbytes; 333 int error = 0; 334 335 start = uio->uio_loffset; 336 off = start & PAGEOFFSET; 337 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 338 page_t *pp; 339 340 bytes = MIN(PAGESIZE - off, len); 341 if (pp = page_lookup(vp, start, SE_SHARED)) { 342 caddr_t va; 343 344 va = ppmapin(pp, PROT_READ | PROT_WRITE, (caddr_t)-1L); 345 error = uiomove(va + off, bytes, UIO_READ, uio); 346 ppmapout(va); 347 page_unlock(pp); 348 } else { 349 /* XXX use dmu_read here? */ 350 error = uiomove(addr, bytes, UIO_READ, uio); 351 } 352 len -= bytes; 353 addr += bytes; 354 off = 0; 355 if (error) 356 break; 357 } 358 return (error); 359 } 360 361 uint_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ 362 363 /* 364 * Read bytes from specified file into supplied buffer. 365 * 366 * IN: vp - vnode of file to be read from. 367 * uio - structure supplying read location, range info, 368 * and return buffer. 369 * ioflag - SYNC flags; used to provide FRSYNC semantics. 370 * cr - credentials of caller. 371 * 372 * OUT: uio - updated offset and range, buffer filled. 373 * 374 * RETURN: 0 if success 375 * error code if failure 376 * 377 * Side Effects: 378 * vp - atime updated if byte count > 0 379 */ 380 /* ARGSUSED */ 381 static int 382 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 383 { 384 znode_t *zp = VTOZ(vp); 385 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 386 uint64_t delta; 387 ssize_t n, size, cnt, ndone; 388 int error, i, numbufs; 389 dmu_buf_t *dbp, **dbpp; 390 rl_t *rl; 391 392 ZFS_ENTER(zfsvfs); 393 394 /* 395 * Validate file offset 396 */ 397 if (uio->uio_loffset < (offset_t)0) { 398 ZFS_EXIT(zfsvfs); 399 return (EINVAL); 400 } 401 402 /* 403 * Fasttrack empty reads 404 */ 405 if (uio->uio_resid == 0) { 406 ZFS_EXIT(zfsvfs); 407 return (0); 408 } 409 410 /* 411 * Check for mandatory locks 412 */ 413 if (MANDMODE((mode_t)zp->z_phys->zp_mode)) { 414 if (error = chklock(vp, FREAD, 415 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { 416 ZFS_EXIT(zfsvfs); 417 return (error); 418 } 419 } 420 421 /* 422 * If we're in FRSYNC mode, sync out this znode before reading it. 423 */ 424 zil_commit(zfsvfs->z_log, zp->z_last_itx, ioflag & FRSYNC); 425 426 /* 427 * Lock the range against changes. 428 */ 429 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER); 430 431 /* 432 * If we are reading past end-of-file we can skip 433 * to the end; but we might still need to set atime. 434 */ 435 if (uio->uio_loffset >= zp->z_phys->zp_size) { 436 cnt = 0; 437 error = 0; 438 goto out; 439 } 440 441 cnt = MIN(uio->uio_resid, zp->z_phys->zp_size - uio->uio_loffset); 442 443 for (ndone = 0; ndone < cnt; ndone += zfs_read_chunk_size) { 444 ASSERT(uio->uio_loffset < zp->z_phys->zp_size); 445 n = MIN(zfs_read_chunk_size, 446 zp->z_phys->zp_size - uio->uio_loffset); 447 n = MIN(n, cnt); 448 error = dmu_buf_hold_array(zfsvfs->z_os, zp->z_id, 449 uio->uio_loffset, n, TRUE, FTAG, &numbufs, &dbpp); 450 if (error) 451 goto out; 452 /* 453 * Compute the adjustment to align the dmu buffers 454 * with the uio buffer. 455 */ 456 delta = uio->uio_loffset - dbpp[0]->db_offset; 457 458 for (i = 0; i < numbufs; i++) { 459 if (n < 0) 460 break; 461 dbp = dbpp[i]; 462 size = dbp->db_size - delta; 463 /* 464 * XXX -- this is correct, but may be suboptimal. 465 * If the pages are all clean, we don't need to 466 * go through mappedread(). Maybe the VMODSORT 467 * stuff can help us here. 468 */ 469 if (vn_has_cached_data(vp)) { 470 error = mappedread(vp, (caddr_t)dbp->db_data + 471 delta, (n < size ? n : size), uio); 472 } else { 473 error = uiomove((caddr_t)dbp->db_data + delta, 474 (n < size ? n : size), UIO_READ, uio); 475 } 476 if (error) { 477 dmu_buf_rele_array(dbpp, numbufs, FTAG); 478 goto out; 479 } 480 n -= dbp->db_size; 481 if (delta) { 482 n += delta; 483 delta = 0; 484 } 485 } 486 dmu_buf_rele_array(dbpp, numbufs, FTAG); 487 } 488 out: 489 zfs_range_unlock(zp, rl); 490 491 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 492 ZFS_EXIT(zfsvfs); 493 return (error); 494 } 495 496 /* 497 * Fault in the pages of the first n bytes specified by the uio structure. 498 * 1 byte in each page is touched and the uio struct is unmodified. 499 * Any error will exit this routine as this is only a best 500 * attempt to get the pages resident. This is a copy of ufs_trans_touch(). 501 */ 502 static void 503 zfs_prefault_write(ssize_t n, struct uio *uio) 504 { 505 struct iovec *iov; 506 ulong_t cnt, incr; 507 caddr_t p; 508 uint8_t tmp; 509 510 iov = uio->uio_iov; 511 512 while (n) { 513 cnt = MIN(iov->iov_len, n); 514 if (cnt == 0) { 515 /* empty iov entry */ 516 iov++; 517 continue; 518 } 519 n -= cnt; 520 /* 521 * touch each page in this segment. 522 */ 523 p = iov->iov_base; 524 while (cnt) { 525 switch (uio->uio_segflg) { 526 case UIO_USERSPACE: 527 case UIO_USERISPACE: 528 if (fuword8(p, &tmp)) 529 return; 530 break; 531 case UIO_SYSSPACE: 532 if (kcopy(p, &tmp, 1)) 533 return; 534 break; 535 } 536 incr = MIN(cnt, PAGESIZE); 537 p += incr; 538 cnt -= incr; 539 } 540 /* 541 * touch the last byte in case it straddles a page. 542 */ 543 p--; 544 switch (uio->uio_segflg) { 545 case UIO_USERSPACE: 546 case UIO_USERISPACE: 547 if (fuword8(p, &tmp)) 548 return; 549 break; 550 case UIO_SYSSPACE: 551 if (kcopy(p, &tmp, 1)) 552 return; 553 break; 554 } 555 iov++; 556 } 557 } 558 559 /* 560 * Write the bytes to a file. 561 * 562 * IN: vp - vnode of file to be written to. 563 * uio - structure supplying write location, range info, 564 * and data buffer. 565 * ioflag - FAPPEND flag set if in append mode. 566 * cr - credentials of caller. 567 * 568 * OUT: uio - updated offset and range. 569 * 570 * RETURN: 0 if success 571 * error code if failure 572 * 573 * Timestamps: 574 * vp - ctime|mtime updated if byte count > 0 575 */ 576 /* ARGSUSED */ 577 static int 578 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 579 { 580 znode_t *zp = VTOZ(vp); 581 rlim64_t limit = uio->uio_llimit; 582 ssize_t start_resid = uio->uio_resid; 583 ssize_t tx_bytes; 584 uint64_t end_size; 585 dmu_tx_t *tx; 586 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 587 zilog_t *zilog = zfsvfs->z_log; 588 uint64_t seq = 0; 589 offset_t woff; 590 ssize_t n, nbytes; 591 rl_t *rl; 592 int max_blksz = zfsvfs->z_max_blksz; 593 int error; 594 595 /* 596 * Fasttrack empty write 597 */ 598 n = start_resid; 599 if (n == 0) 600 return (0); 601 602 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 603 limit = MAXOFFSET_T; 604 605 ZFS_ENTER(zfsvfs); 606 607 /* 608 * Pre-fault the initial pages to ensure slow (eg NFS) pages 609 * don't hold up txg. 610 */ 611 zfs_prefault_write(MIN(start_resid, SPA_MAXBLOCKSIZE), uio); 612 613 /* 614 * If in append mode, set the io offset pointer to eof. 615 */ 616 if (ioflag & FAPPEND) { 617 /* 618 * Range lock for a file append: 619 * The value for the start of range will be determined by 620 * zfs_range_lock() (to guarantee append semantics). 621 * If this write will cause the block size to increase, 622 * zfs_range_lock() will lock the entire file, so we must 623 * later reduce the range after we grow the block size. 624 */ 625 rl = zfs_range_lock(zp, 0, n, RL_APPEND); 626 if (rl->r_len == UINT64_MAX) { 627 /* overlocked, zp_size can't change */ 628 woff = uio->uio_loffset = zp->z_phys->zp_size; 629 } else { 630 woff = uio->uio_loffset = rl->r_off; 631 } 632 } else { 633 woff = uio->uio_loffset; 634 /* 635 * Validate file offset 636 */ 637 if (woff < 0) { 638 ZFS_EXIT(zfsvfs); 639 return (EINVAL); 640 } 641 642 /* 643 * If we need to grow the block size then zfs_range_lock() 644 * will lock a wider range than we request here. 645 * Later after growing the block size we reduce the range. 646 */ 647 rl = zfs_range_lock(zp, woff, n, RL_WRITER); 648 } 649 650 if (woff >= limit) { 651 error = EFBIG; 652 goto no_tx_done; 653 } 654 655 if ((woff + n) > limit || woff > (limit - n)) 656 n = limit - woff; 657 658 /* 659 * Check for mandatory locks 660 */ 661 if (MANDMODE((mode_t)zp->z_phys->zp_mode) && 662 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) 663 goto no_tx_done; 664 end_size = MAX(zp->z_phys->zp_size, woff + n); 665 top: 666 tx = dmu_tx_create(zfsvfs->z_os); 667 dmu_tx_hold_bonus(tx, zp->z_id); 668 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 669 error = dmu_tx_assign(tx, zfsvfs->z_assign); 670 if (error) { 671 dmu_tx_abort(tx); 672 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 673 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 674 goto top; 675 } 676 goto no_tx_done; 677 } 678 679 /* 680 * If zfs_range_lock() over-locked we grow the blocksize 681 * and then reduce the lock range. 682 */ 683 if (rl->r_len == UINT64_MAX) { 684 uint64_t new_blksz; 685 686 if (zp->z_blksz > max_blksz) { 687 ASSERT(!ISP2(zp->z_blksz)); 688 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE); 689 } else { 690 new_blksz = MIN(end_size, max_blksz); 691 } 692 zfs_grow_blocksize(zp, new_blksz, tx); 693 zfs_range_reduce(zp, rl, woff, n); 694 } 695 696 /* 697 * The file data does not fit in the znode "cache", so we 698 * will be writing to the file block data buffers. 699 * Each buffer will be written in a separate transaction; 700 * this keeps the intent log records small and allows us 701 * to do more fine-grained space accounting. 702 */ 703 while (n > 0) { 704 /* 705 * XXX - should we really limit each write to z_max_blksz? 706 * Perhaps we should use SPA_MAXBLOCKSIZE chunks? 707 */ 708 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz)); 709 rw_enter(&zp->z_map_lock, RW_READER); 710 711 tx_bytes = uio->uio_resid; 712 if (vn_has_cached_data(vp)) { 713 rw_exit(&zp->z_map_lock); 714 error = mappedwrite(vp, woff, nbytes, uio, tx); 715 } else { 716 error = dmu_write_uio(zfsvfs->z_os, zp->z_id, 717 woff, nbytes, uio, tx); 718 rw_exit(&zp->z_map_lock); 719 } 720 tx_bytes -= uio->uio_resid; 721 722 if (error) { 723 /* XXX - do we need to "clean up" the dmu buffer? */ 724 break; 725 } 726 727 ASSERT(tx_bytes == nbytes); 728 729 /* 730 * Clear Set-UID/Set-GID bits on successful write if not 731 * privileged and at least one of the excute bits is set. 732 * 733 * It would be nice to to this after all writes have 734 * been done, but that would still expose the ISUID/ISGID 735 * to another app after the partial write is committed. 736 */ 737 738 mutex_enter(&zp->z_acl_lock); 739 if ((zp->z_phys->zp_mode & (S_IXUSR | (S_IXUSR >> 3) | 740 (S_IXUSR >> 6))) != 0 && 741 (zp->z_phys->zp_mode & (S_ISUID | S_ISGID)) != 0 && 742 secpolicy_vnode_setid_retain(cr, 743 (zp->z_phys->zp_mode & S_ISUID) != 0 && 744 zp->z_phys->zp_uid == 0) != 0) { 745 zp->z_phys->zp_mode &= ~(S_ISUID | S_ISGID); 746 } 747 mutex_exit(&zp->z_acl_lock); 748 749 n -= nbytes; 750 if (n <= 0) 751 break; 752 753 /* 754 * We have more work ahead of us, so wrap up this transaction 755 * and start another. Exact same logic as tx_done below. 756 */ 757 while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset) { 758 dmu_buf_will_dirty(zp->z_dbuf, tx); 759 (void) atomic_cas_64(&zp->z_phys->zp_size, end_size, 760 uio->uio_loffset); 761 } 762 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 763 seq = zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, 764 ioflag, uio); 765 dmu_tx_commit(tx); 766 767 /* Pre-fault the next set of pages */ 768 zfs_prefault_write(MIN(n, SPA_MAXBLOCKSIZE), uio); 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 dmu_tx_abort(tx); 780 if (error == ERESTART && 781 zfsvfs->z_assign == TXG_NOWAIT) { 782 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 783 goto top; 784 } 785 goto no_tx_done; 786 } 787 } 788 789 tx_done: 790 791 if (tx_bytes != 0) { 792 /* 793 * Update the file size if it has changed; account 794 * for possible concurrent updates. 795 */ 796 while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset) { 797 dmu_buf_will_dirty(zp->z_dbuf, tx); 798 (void) atomic_cas_64(&zp->z_phys->zp_size, end_size, 799 uio->uio_loffset); 800 } 801 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 802 seq = zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, 803 ioflag, uio); 804 } 805 dmu_tx_commit(tx); 806 807 808 no_tx_done: 809 810 zfs_range_unlock(zp, rl); 811 812 /* 813 * If we're in replay mode, or we made no progress, return error. 814 * Otherwise, it's at least a partial write, so it's successful. 815 */ 816 if (zfsvfs->z_assign >= TXG_INITIAL || uio->uio_resid == start_resid) { 817 ZFS_EXIT(zfsvfs); 818 return (error); 819 } 820 821 zil_commit(zilog, seq, ioflag & (FSYNC | FDSYNC)); 822 823 ZFS_EXIT(zfsvfs); 824 return (0); 825 } 826 827 /* 828 * Get data to generate a TX_WRITE intent log record. 829 */ 830 int 831 zfs_get_data(void *arg, lr_write_t *lr, char *buf) 832 { 833 zfsvfs_t *zfsvfs = arg; 834 objset_t *os = zfsvfs->z_os; 835 znode_t *zp; 836 uint64_t off = lr->lr_offset; 837 rl_t *rl; 838 int dlen = lr->lr_length; /* length of user data */ 839 int error = 0; 840 841 ASSERT(dlen != 0); 842 843 /* 844 * Nothing to do if the file has been removed 845 */ 846 if (zfs_zget(zfsvfs, lr->lr_foid, &zp) != 0) 847 return (ENOENT); 848 if (zp->z_reap) { 849 VN_RELE(ZTOV(zp)); 850 return (ENOENT); 851 } 852 853 /* 854 * Write records come in two flavors: immediate and indirect. 855 * For small writes it's cheaper to store the data with the 856 * log record (immediate); for large writes it's cheaper to 857 * sync the data and get a pointer to it (indirect) so that 858 * we don't have to write the data twice. 859 */ 860 if (buf != NULL) { /* immediate write */ 861 dmu_buf_t *db; 862 863 rl = zfs_range_lock(zp, off, dlen, RL_READER); 864 /* test for truncation needs to be done while range locked */ 865 if (off >= zp->z_phys->zp_size) { 866 error = ENOENT; 867 goto out; 868 } 869 VERIFY(0 == dmu_buf_hold(os, lr->lr_foid, off, FTAG, &db)); 870 bcopy((char *)db->db_data + off - db->db_offset, buf, dlen); 871 dmu_buf_rele(db, FTAG); 872 } else { /* indirect write */ 873 uint64_t boff; /* block starting offset */ 874 875 /* 876 * Have to lock the whole block to ensure when it's 877 * written out and it's checksum is being calculated 878 * that no one can change the data. We need to re-check 879 * blocksize after we get the lock in case it's changed! 880 */ 881 for (;;) { 882 boff = off & ~(zp->z_blksz - 1); 883 dlen = zp->z_blksz; 884 rl = zfs_range_lock(zp, boff, dlen, RL_READER); 885 if (zp->z_blksz == dlen) 886 break; 887 zfs_range_unlock(zp, rl); 888 } 889 /* test for truncation needs to be done while range locked */ 890 if (off >= zp->z_phys->zp_size) { 891 error = ENOENT; 892 goto out; 893 } 894 txg_suspend(dmu_objset_pool(os)); 895 error = dmu_sync(os, lr->lr_foid, off, &lr->lr_blkoff, 896 &lr->lr_blkptr, lr->lr_common.lrc_txg); 897 txg_resume(dmu_objset_pool(os)); 898 } 899 out: 900 zfs_range_unlock(zp, rl); 901 VN_RELE(ZTOV(zp)); 902 return (error); 903 } 904 905 /*ARGSUSED*/ 906 static int 907 zfs_access(vnode_t *vp, int mode, int flags, cred_t *cr) 908 { 909 znode_t *zp = VTOZ(vp); 910 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 911 int error; 912 913 ZFS_ENTER(zfsvfs); 914 error = zfs_zaccess_rwx(zp, mode, cr); 915 ZFS_EXIT(zfsvfs); 916 return (error); 917 } 918 919 /* 920 * Lookup an entry in a directory, or an extended attribute directory. 921 * If it exists, return a held vnode reference for it. 922 * 923 * IN: dvp - vnode of directory to search. 924 * nm - name of entry to lookup. 925 * pnp - full pathname to lookup [UNUSED]. 926 * flags - LOOKUP_XATTR set if looking for an attribute. 927 * rdir - root directory vnode [UNUSED]. 928 * cr - credentials of caller. 929 * 930 * OUT: vpp - vnode of located entry, NULL if not found. 931 * 932 * RETURN: 0 if success 933 * error code if failure 934 * 935 * Timestamps: 936 * NA 937 */ 938 /* ARGSUSED */ 939 static int 940 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp, 941 int flags, vnode_t *rdir, cred_t *cr) 942 { 943 944 znode_t *zdp = VTOZ(dvp); 945 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 946 int error; 947 948 ZFS_ENTER(zfsvfs); 949 950 *vpp = NULL; 951 952 if (flags & LOOKUP_XATTR) { 953 /* 954 * We don't allow recursive attributes.. 955 * Maybe someday we will. 956 */ 957 if (zdp->z_phys->zp_flags & ZFS_XATTR) { 958 ZFS_EXIT(zfsvfs); 959 return (EINVAL); 960 } 961 962 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr)) { 963 ZFS_EXIT(zfsvfs); 964 return (error); 965 } 966 967 /* 968 * Do we have permission to get into attribute directory? 969 */ 970 971 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, cr)) { 972 VN_RELE(*vpp); 973 } 974 975 ZFS_EXIT(zfsvfs); 976 return (error); 977 } 978 979 if (dvp->v_type != VDIR) { 980 ZFS_EXIT(zfsvfs); 981 return (ENOTDIR); 982 } 983 984 /* 985 * Check accessibility of directory. 986 */ 987 988 if (error = zfs_zaccess(zdp, ACE_EXECUTE, cr)) { 989 ZFS_EXIT(zfsvfs); 990 return (error); 991 } 992 993 if ((error = zfs_dirlook(zdp, nm, vpp)) == 0) { 994 995 /* 996 * Convert device special files 997 */ 998 if (IS_DEVVP(*vpp)) { 999 vnode_t *svp; 1000 1001 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1002 VN_RELE(*vpp); 1003 if (svp == NULL) 1004 error = ENOSYS; 1005 else 1006 *vpp = svp; 1007 } 1008 } 1009 1010 ZFS_EXIT(zfsvfs); 1011 return (error); 1012 } 1013 1014 /* 1015 * Attempt to create a new entry in a directory. If the entry 1016 * already exists, truncate the file if permissible, else return 1017 * an error. Return the vp of the created or trunc'd file. 1018 * 1019 * IN: dvp - vnode of directory to put new file entry in. 1020 * name - name of new file entry. 1021 * vap - attributes of new file. 1022 * excl - flag indicating exclusive or non-exclusive mode. 1023 * mode - mode to open file with. 1024 * cr - credentials of caller. 1025 * flag - large file flag [UNUSED]. 1026 * 1027 * OUT: vpp - vnode of created or trunc'd entry. 1028 * 1029 * RETURN: 0 if success 1030 * error code if failure 1031 * 1032 * Timestamps: 1033 * dvp - ctime|mtime updated if new entry created 1034 * vp - ctime|mtime always, atime if new 1035 */ 1036 /* ARGSUSED */ 1037 static int 1038 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl, 1039 int mode, vnode_t **vpp, cred_t *cr, int flag) 1040 { 1041 znode_t *zp, *dzp = VTOZ(dvp); 1042 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1043 zilog_t *zilog = zfsvfs->z_log; 1044 uint64_t seq = 0; 1045 objset_t *os = zfsvfs->z_os; 1046 zfs_dirlock_t *dl; 1047 dmu_tx_t *tx; 1048 rl_t *rl; 1049 int error; 1050 uint64_t zoid; 1051 1052 ZFS_ENTER(zfsvfs); 1053 1054 top: 1055 *vpp = NULL; 1056 1057 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr)) 1058 vap->va_mode &= ~VSVTX; 1059 1060 if (*name == '\0') { 1061 /* 1062 * Null component name refers to the directory itself. 1063 */ 1064 VN_HOLD(dvp); 1065 zp = dzp; 1066 dl = NULL; 1067 error = 0; 1068 } else { 1069 /* possible VN_HOLD(zp) */ 1070 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, 0)) { 1071 if (strcmp(name, "..") == 0) 1072 error = EISDIR; 1073 ZFS_EXIT(zfsvfs); 1074 return (error); 1075 } 1076 } 1077 1078 zoid = zp ? zp->z_id : -1ULL; 1079 1080 if (zp == NULL) { 1081 /* 1082 * Create a new file object and update the directory 1083 * to reference it. 1084 */ 1085 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 1086 goto out; 1087 } 1088 1089 /* 1090 * We only support the creation of regular files in 1091 * extended attribute directories. 1092 */ 1093 if ((dzp->z_phys->zp_flags & ZFS_XATTR) && 1094 (vap->va_type != VREG)) { 1095 error = EINVAL; 1096 goto out; 1097 } 1098 1099 tx = dmu_tx_create(os); 1100 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1101 dmu_tx_hold_bonus(tx, dzp->z_id); 1102 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1103 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 1104 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1105 0, SPA_MAXBLOCKSIZE); 1106 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1107 if (error) { 1108 dmu_tx_abort(tx); 1109 zfs_dirent_unlock(dl); 1110 if (error == ERESTART && 1111 zfsvfs->z_assign == TXG_NOWAIT) { 1112 txg_wait_open(dmu_objset_pool(os), 0); 1113 goto top; 1114 } 1115 ZFS_EXIT(zfsvfs); 1116 return (error); 1117 } 1118 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 1119 ASSERT(zp->z_id == zoid); 1120 (void) zfs_link_create(dl, zp, tx, ZNEW); 1121 seq = zfs_log_create(zilog, tx, TX_CREATE, dzp, zp, name); 1122 dmu_tx_commit(tx); 1123 } else { 1124 /* 1125 * A directory entry already exists for this name. 1126 */ 1127 /* 1128 * Can't truncate an existing file if in exclusive mode. 1129 */ 1130 if (excl == EXCL) { 1131 error = EEXIST; 1132 goto out; 1133 } 1134 /* 1135 * Can't open a directory for writing. 1136 */ 1137 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1138 error = EISDIR; 1139 goto out; 1140 } 1141 /* 1142 * Verify requested access to file. 1143 */ 1144 if (mode && (error = zfs_zaccess_rwx(zp, mode, cr))) { 1145 goto out; 1146 } 1147 /* 1148 * Truncate regular files if requested. 1149 */ 1150 1151 /* 1152 * Need to update dzp->z_seq? 1153 */ 1154 1155 mutex_enter(&dzp->z_lock); 1156 dzp->z_seq++; 1157 mutex_exit(&dzp->z_lock); 1158 1159 if ((ZTOV(zp)->v_type == VREG) && (zp->z_phys->zp_size != 0) && 1160 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1161 /* 1162 * Truncate the file. 1163 */ 1164 tx = dmu_tx_create(os); 1165 dmu_tx_hold_bonus(tx, zoid); 1166 dmu_tx_hold_free(tx, zoid, 0, DMU_OBJECT_END); 1167 /* Lock the whole range of the file */ 1168 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 1169 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1170 if (error) { 1171 dmu_tx_abort(tx); 1172 zfs_range_unlock(zp, rl); 1173 if (dl) 1174 zfs_dirent_unlock(dl); 1175 VN_RELE(ZTOV(zp)); 1176 if (error == ERESTART && 1177 zfsvfs->z_assign == TXG_NOWAIT) { 1178 txg_wait_open(dmu_objset_pool(os), 0); 1179 goto top; 1180 } 1181 ZFS_EXIT(zfsvfs); 1182 return (error); 1183 } 1184 error = zfs_freesp(zp, 0, 0, mode, tx, cr); 1185 if (error == 0) { 1186 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 1187 seq = zfs_log_truncate(zilog, tx, 1188 TX_TRUNCATE, zp, 0, 0); 1189 } 1190 zfs_range_unlock(zp, rl); 1191 dmu_tx_commit(tx); 1192 } 1193 } 1194 out: 1195 1196 if (dl) 1197 zfs_dirent_unlock(dl); 1198 1199 if (error) { 1200 if (zp) 1201 VN_RELE(ZTOV(zp)); 1202 } else { 1203 *vpp = ZTOV(zp); 1204 /* 1205 * If vnode is for a device return a specfs vnode instead. 1206 */ 1207 if (IS_DEVVP(*vpp)) { 1208 struct vnode *svp; 1209 1210 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1211 VN_RELE(*vpp); 1212 if (svp == NULL) { 1213 error = ENOSYS; 1214 } 1215 *vpp = svp; 1216 } 1217 } 1218 1219 zil_commit(zilog, seq, 0); 1220 1221 ZFS_EXIT(zfsvfs); 1222 return (error); 1223 } 1224 1225 /* 1226 * Remove an entry from a directory. 1227 * 1228 * IN: dvp - vnode of directory to remove entry from. 1229 * name - name of entry to remove. 1230 * cr - credentials of caller. 1231 * 1232 * RETURN: 0 if success 1233 * error code if failure 1234 * 1235 * Timestamps: 1236 * dvp - ctime|mtime 1237 * vp - ctime (if nlink > 0) 1238 */ 1239 static int 1240 zfs_remove(vnode_t *dvp, char *name, cred_t *cr) 1241 { 1242 znode_t *zp, *dzp = VTOZ(dvp); 1243 znode_t *xzp = NULL; 1244 vnode_t *vp; 1245 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1246 zilog_t *zilog = zfsvfs->z_log; 1247 uint64_t seq = 0; 1248 uint64_t acl_obj, xattr_obj; 1249 zfs_dirlock_t *dl; 1250 dmu_tx_t *tx; 1251 int may_delete_now, delete_now = FALSE; 1252 int reaped; 1253 int error; 1254 1255 ZFS_ENTER(zfsvfs); 1256 1257 top: 1258 /* 1259 * Attempt to lock directory; fail if entry doesn't exist. 1260 */ 1261 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) { 1262 ZFS_EXIT(zfsvfs); 1263 return (error); 1264 } 1265 1266 vp = ZTOV(zp); 1267 1268 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1269 goto out; 1270 } 1271 1272 /* 1273 * Need to use rmdir for removing directories. 1274 */ 1275 if (vp->v_type == VDIR) { 1276 error = EPERM; 1277 goto out; 1278 } 1279 1280 vnevent_remove(vp); 1281 1282 dnlc_remove(dvp, name); 1283 1284 mutex_enter(&vp->v_lock); 1285 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp); 1286 mutex_exit(&vp->v_lock); 1287 1288 /* 1289 * We may delete the znode now, or we may put it on the delete queue; 1290 * it depends on whether we're the last link, and on whether there are 1291 * other holds on the vnode. So we dmu_tx_hold() the right things to 1292 * allow for either case. 1293 */ 1294 tx = dmu_tx_create(zfsvfs->z_os); 1295 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1296 dmu_tx_hold_bonus(tx, zp->z_id); 1297 if (may_delete_now) 1298 dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END); 1299 1300 /* are there any extended attributes? */ 1301 if ((xattr_obj = zp->z_phys->zp_xattr) != 0) { 1302 /* 1303 * XXX - There is a possibility that the delete 1304 * of the parent file could succeed, but then we get 1305 * an ENOSPC when we try to delete the xattrs... 1306 * so we would need to re-try the deletes periodically 1307 */ 1308 /* XXX - do we need this if we are deleting? */ 1309 dmu_tx_hold_bonus(tx, xattr_obj); 1310 } 1311 1312 /* are there any additional acls */ 1313 if ((acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj) != 0 && 1314 may_delete_now) 1315 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 1316 1317 /* charge as an update -- would be nice not to charge at all */ 1318 dmu_tx_hold_zap(tx, zfsvfs->z_dqueue, FALSE, NULL); 1319 1320 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1321 if (error) { 1322 dmu_tx_abort(tx); 1323 zfs_dirent_unlock(dl); 1324 VN_RELE(vp); 1325 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1326 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 1327 goto top; 1328 } 1329 ZFS_EXIT(zfsvfs); 1330 return (error); 1331 } 1332 1333 /* 1334 * Remove the directory entry. 1335 */ 1336 error = zfs_link_destroy(dl, zp, tx, 0, &reaped); 1337 1338 if (error) { 1339 dmu_tx_commit(tx); 1340 goto out; 1341 } 1342 1343 if (reaped) { 1344 mutex_enter(&vp->v_lock); 1345 delete_now = may_delete_now && 1346 vp->v_count == 1 && !vn_has_cached_data(vp) && 1347 zp->z_phys->zp_xattr == xattr_obj && 1348 zp->z_phys->zp_acl.z_acl_extern_obj == acl_obj; 1349 mutex_exit(&vp->v_lock); 1350 } 1351 1352 if (delete_now) { 1353 if (zp->z_phys->zp_xattr) { 1354 error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp); 1355 ASSERT3U(error, ==, 0); 1356 ASSERT3U(xzp->z_phys->zp_links, ==, 2); 1357 dmu_buf_will_dirty(xzp->z_dbuf, tx); 1358 mutex_enter(&xzp->z_lock); 1359 xzp->z_reap = 1; 1360 xzp->z_phys->zp_links = 0; 1361 mutex_exit(&xzp->z_lock); 1362 zfs_dq_add(xzp, tx); 1363 zp->z_phys->zp_xattr = 0; /* probably unnecessary */ 1364 } 1365 mutex_enter(&zp->z_lock); 1366 mutex_enter(&vp->v_lock); 1367 vp->v_count--; 1368 ASSERT3U(vp->v_count, ==, 0); 1369 mutex_exit(&vp->v_lock); 1370 zp->z_active = 0; 1371 mutex_exit(&zp->z_lock); 1372 zfs_znode_delete(zp, tx); 1373 VFS_RELE(zfsvfs->z_vfs); 1374 } else if (reaped) { 1375 zfs_dq_add(zp, tx); 1376 } 1377 1378 seq = zfs_log_remove(zilog, tx, TX_REMOVE, dzp, name); 1379 1380 dmu_tx_commit(tx); 1381 out: 1382 zfs_dirent_unlock(dl); 1383 1384 if (!delete_now) { 1385 VN_RELE(vp); 1386 } else if (xzp) { 1387 /* this rele delayed to prevent nesting transactions */ 1388 VN_RELE(ZTOV(xzp)); 1389 } 1390 1391 zil_commit(zilog, seq, 0); 1392 1393 ZFS_EXIT(zfsvfs); 1394 return (error); 1395 } 1396 1397 /* 1398 * Create a new directory and insert it into dvp using the name 1399 * provided. Return a pointer to the inserted directory. 1400 * 1401 * IN: dvp - vnode of directory to add subdir to. 1402 * dirname - name of new directory. 1403 * vap - attributes of new directory. 1404 * cr - credentials of caller. 1405 * 1406 * OUT: vpp - vnode of created directory. 1407 * 1408 * RETURN: 0 if success 1409 * error code if failure 1410 * 1411 * Timestamps: 1412 * dvp - ctime|mtime updated 1413 * vp - ctime|mtime|atime updated 1414 */ 1415 static int 1416 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr) 1417 { 1418 znode_t *zp, *dzp = VTOZ(dvp); 1419 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1420 zilog_t *zilog = zfsvfs->z_log; 1421 uint64_t seq = 0; 1422 zfs_dirlock_t *dl; 1423 uint64_t zoid = 0; 1424 dmu_tx_t *tx; 1425 int error; 1426 1427 ASSERT(vap->va_type == VDIR); 1428 1429 ZFS_ENTER(zfsvfs); 1430 1431 if (dzp->z_phys->zp_flags & ZFS_XATTR) { 1432 ZFS_EXIT(zfsvfs); 1433 return (EINVAL); 1434 } 1435 top: 1436 *vpp = NULL; 1437 1438 /* 1439 * First make sure the new directory doesn't exist. 1440 */ 1441 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, ZNEW)) { 1442 ZFS_EXIT(zfsvfs); 1443 return (error); 1444 } 1445 1446 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, cr)) { 1447 zfs_dirent_unlock(dl); 1448 ZFS_EXIT(zfsvfs); 1449 return (error); 1450 } 1451 1452 /* 1453 * Add a new entry to the directory. 1454 */ 1455 tx = dmu_tx_create(zfsvfs->z_os); 1456 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 1457 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 1458 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 1459 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1460 0, SPA_MAXBLOCKSIZE); 1461 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1462 if (error) { 1463 dmu_tx_abort(tx); 1464 zfs_dirent_unlock(dl); 1465 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1466 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 1467 goto top; 1468 } 1469 ZFS_EXIT(zfsvfs); 1470 return (error); 1471 } 1472 1473 /* 1474 * Create new node. 1475 */ 1476 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 1477 1478 /* 1479 * Now put new name in parent dir. 1480 */ 1481 (void) zfs_link_create(dl, zp, tx, ZNEW); 1482 1483 *vpp = ZTOV(zp); 1484 1485 seq = zfs_log_create(zilog, tx, TX_MKDIR, dzp, zp, dirname); 1486 dmu_tx_commit(tx); 1487 1488 zfs_dirent_unlock(dl); 1489 1490 zil_commit(zilog, seq, 0); 1491 1492 ZFS_EXIT(zfsvfs); 1493 return (0); 1494 } 1495 1496 /* 1497 * Remove a directory subdir entry. If the current working 1498 * directory is the same as the subdir to be removed, the 1499 * remove will fail. 1500 * 1501 * IN: dvp - vnode of directory to remove from. 1502 * name - name of directory to be removed. 1503 * cwd - vnode of current working directory. 1504 * cr - credentials of caller. 1505 * 1506 * RETURN: 0 if success 1507 * error code if failure 1508 * 1509 * Timestamps: 1510 * dvp - ctime|mtime updated 1511 */ 1512 static int 1513 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr) 1514 { 1515 znode_t *dzp = VTOZ(dvp); 1516 znode_t *zp; 1517 vnode_t *vp; 1518 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1519 zilog_t *zilog = zfsvfs->z_log; 1520 uint64_t seq = 0; 1521 zfs_dirlock_t *dl; 1522 dmu_tx_t *tx; 1523 int error; 1524 1525 ZFS_ENTER(zfsvfs); 1526 1527 top: 1528 zp = NULL; 1529 1530 /* 1531 * Attempt to lock directory; fail if entry doesn't exist. 1532 */ 1533 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) { 1534 ZFS_EXIT(zfsvfs); 1535 return (error); 1536 } 1537 1538 vp = ZTOV(zp); 1539 1540 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1541 goto out; 1542 } 1543 1544 if (vp->v_type != VDIR) { 1545 error = ENOTDIR; 1546 goto out; 1547 } 1548 1549 if (vp == cwd) { 1550 error = EINVAL; 1551 goto out; 1552 } 1553 1554 vnevent_rmdir(vp); 1555 1556 /* 1557 * Grab a lock on the parent pointer make sure we play well 1558 * with the treewalk and directory rename code. 1559 */ 1560 rw_enter(&zp->z_parent_lock, RW_WRITER); 1561 1562 tx = dmu_tx_create(zfsvfs->z_os); 1563 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1564 dmu_tx_hold_bonus(tx, zp->z_id); 1565 dmu_tx_hold_zap(tx, zfsvfs->z_dqueue, FALSE, NULL); 1566 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1567 if (error) { 1568 dmu_tx_abort(tx); 1569 rw_exit(&zp->z_parent_lock); 1570 zfs_dirent_unlock(dl); 1571 VN_RELE(vp); 1572 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1573 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 1574 goto top; 1575 } 1576 ZFS_EXIT(zfsvfs); 1577 return (error); 1578 } 1579 1580 error = zfs_link_destroy(dl, zp, tx, 0, NULL); 1581 1582 if (error == 0) 1583 seq = zfs_log_remove(zilog, tx, TX_RMDIR, dzp, name); 1584 1585 dmu_tx_commit(tx); 1586 1587 rw_exit(&zp->z_parent_lock); 1588 out: 1589 zfs_dirent_unlock(dl); 1590 1591 VN_RELE(vp); 1592 1593 zil_commit(zilog, seq, 0); 1594 1595 ZFS_EXIT(zfsvfs); 1596 return (error); 1597 } 1598 1599 /* 1600 * Read as many directory entries as will fit into the provided 1601 * buffer from the given directory cursor position (specified in 1602 * the uio structure. 1603 * 1604 * IN: vp - vnode of directory to read. 1605 * uio - structure supplying read location, range info, 1606 * and return buffer. 1607 * cr - credentials of caller. 1608 * 1609 * OUT: uio - updated offset and range, buffer filled. 1610 * eofp - set to true if end-of-file detected. 1611 * 1612 * RETURN: 0 if success 1613 * error code if failure 1614 * 1615 * Timestamps: 1616 * vp - atime updated 1617 * 1618 * Note that the low 4 bits of the cookie returned by zap is always zero. 1619 * This allows us to use the low range for "special" directory entries: 1620 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 1621 * we use the offset 2 for the '.zfs' directory. 1622 */ 1623 /* ARGSUSED */ 1624 static int 1625 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp) 1626 { 1627 znode_t *zp = VTOZ(vp); 1628 iovec_t *iovp; 1629 dirent64_t *odp; 1630 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1631 objset_t *os; 1632 caddr_t outbuf; 1633 size_t bufsize; 1634 zap_cursor_t zc; 1635 zap_attribute_t zap; 1636 uint_t bytes_wanted; 1637 ushort_t this_reclen; 1638 uint64_t offset; /* must be unsigned; checks for < 1 */ 1639 off64_t *next; 1640 int local_eof; 1641 int outcount; 1642 int error; 1643 uint8_t prefetch; 1644 1645 ZFS_ENTER(zfsvfs); 1646 1647 /* 1648 * If we are not given an eof variable, 1649 * use a local one. 1650 */ 1651 if (eofp == NULL) 1652 eofp = &local_eof; 1653 1654 /* 1655 * Check for valid iov_len. 1656 */ 1657 if (uio->uio_iov->iov_len <= 0) { 1658 ZFS_EXIT(zfsvfs); 1659 return (EINVAL); 1660 } 1661 1662 /* 1663 * Quit if directory has been removed (posix) 1664 */ 1665 if ((*eofp = zp->z_reap) != 0) { 1666 ZFS_EXIT(zfsvfs); 1667 return (0); 1668 } 1669 1670 error = 0; 1671 os = zfsvfs->z_os; 1672 offset = uio->uio_loffset; 1673 prefetch = zp->z_zn_prefetch; 1674 1675 /* 1676 * Initialize the iterator cursor. 1677 */ 1678 if (offset <= 3) { 1679 /* 1680 * Start iteration from the beginning of the directory. 1681 */ 1682 zap_cursor_init(&zc, os, zp->z_id); 1683 } else { 1684 /* 1685 * The offset is a serialized cursor. 1686 */ 1687 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 1688 } 1689 1690 /* 1691 * Get space to change directory entries into fs independent format. 1692 */ 1693 iovp = uio->uio_iov; 1694 bytes_wanted = iovp->iov_len; 1695 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 1696 bufsize = bytes_wanted; 1697 outbuf = kmem_alloc(bufsize, KM_SLEEP); 1698 odp = (struct dirent64 *)outbuf; 1699 } else { 1700 bufsize = bytes_wanted; 1701 odp = (struct dirent64 *)iovp->iov_base; 1702 } 1703 1704 /* 1705 * Transform to file-system independent format 1706 */ 1707 outcount = 0; 1708 while (outcount < bytes_wanted) { 1709 /* 1710 * Special case `.', `..', and `.zfs'. 1711 */ 1712 if (offset == 0) { 1713 (void) strcpy(zap.za_name, "."); 1714 zap.za_first_integer = zp->z_id; 1715 this_reclen = DIRENT64_RECLEN(1); 1716 } else if (offset == 1) { 1717 (void) strcpy(zap.za_name, ".."); 1718 zap.za_first_integer = zp->z_phys->zp_parent; 1719 this_reclen = DIRENT64_RECLEN(2); 1720 } else if (offset == 2 && zfs_show_ctldir(zp)) { 1721 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 1722 zap.za_first_integer = ZFSCTL_INO_ROOT; 1723 this_reclen = 1724 DIRENT64_RECLEN(sizeof (ZFS_CTLDIR_NAME) - 1); 1725 } else { 1726 /* 1727 * Grab next entry. 1728 */ 1729 if (error = zap_cursor_retrieve(&zc, &zap)) { 1730 if ((*eofp = (error == ENOENT)) != 0) 1731 break; 1732 else 1733 goto update; 1734 } 1735 1736 if (zap.za_integer_length != 8 || 1737 zap.za_num_integers != 1) { 1738 cmn_err(CE_WARN, "zap_readdir: bad directory " 1739 "entry, obj = %lld, offset = %lld\n", 1740 (u_longlong_t)zp->z_id, 1741 (u_longlong_t)offset); 1742 error = ENXIO; 1743 goto update; 1744 } 1745 this_reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 1746 } 1747 1748 /* 1749 * Will this entry fit in the buffer? 1750 */ 1751 if (outcount + this_reclen > bufsize) { 1752 /* 1753 * Did we manage to fit anything in the buffer? 1754 */ 1755 if (!outcount) { 1756 error = EINVAL; 1757 goto update; 1758 } 1759 break; 1760 } 1761 /* 1762 * Add this entry: 1763 */ 1764 odp->d_ino = (ino64_t)zap.za_first_integer; 1765 odp->d_reclen = (ushort_t)this_reclen; 1766 /* NOTE: d_off is the offset for the *next* entry */ 1767 next = &(odp->d_off); 1768 (void) strncpy(odp->d_name, zap.za_name, 1769 DIRENT64_NAMELEN(this_reclen)); 1770 outcount += this_reclen; 1771 odp = (dirent64_t *)((intptr_t)odp + this_reclen); 1772 1773 ASSERT(outcount <= bufsize); 1774 1775 /* Prefetch znode */ 1776 if (prefetch) 1777 dmu_prefetch(os, zap.za_first_integer, 0, 0); 1778 1779 /* 1780 * Move to the next entry, fill in the previous offset. 1781 */ 1782 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 1783 zap_cursor_advance(&zc); 1784 offset = zap_cursor_serialize(&zc); 1785 } else { 1786 offset += 1; 1787 } 1788 *next = offset; 1789 } 1790 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 1791 1792 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 1793 iovp->iov_base += outcount; 1794 iovp->iov_len -= outcount; 1795 uio->uio_resid -= outcount; 1796 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 1797 /* 1798 * Reset the pointer. 1799 */ 1800 offset = uio->uio_loffset; 1801 } 1802 1803 update: 1804 zap_cursor_fini(&zc); 1805 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 1806 kmem_free(outbuf, bufsize); 1807 1808 if (error == ENOENT) 1809 error = 0; 1810 1811 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 1812 1813 uio->uio_loffset = offset; 1814 ZFS_EXIT(zfsvfs); 1815 return (error); 1816 } 1817 1818 static int 1819 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr) 1820 { 1821 znode_t *zp = VTOZ(vp); 1822 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1823 1824 /* 1825 * Regardless of whether this is required for standards conformance, 1826 * this is the logical behavior when fsync() is called on a file with 1827 * dirty pages. We use B_ASYNC since the ZIL transactions are already 1828 * going to be pushed out as part of the zil_commit(). 1829 */ 1830 if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) && 1831 (vp->v_type == VREG) && !(IS_SWAPVP(vp))) 1832 (void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr); 1833 1834 ZFS_ENTER(zfsvfs); 1835 zil_commit(zfsvfs->z_log, zp->z_last_itx, FSYNC); 1836 ZFS_EXIT(zfsvfs); 1837 return (0); 1838 } 1839 1840 /* 1841 * Get the requested file attributes and place them in the provided 1842 * vattr structure. 1843 * 1844 * IN: vp - vnode of file. 1845 * vap - va_mask identifies requested attributes. 1846 * flags - [UNUSED] 1847 * cr - credentials of caller. 1848 * 1849 * OUT: vap - attribute values. 1850 * 1851 * RETURN: 0 (always succeeds) 1852 */ 1853 /* ARGSUSED */ 1854 static int 1855 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr) 1856 { 1857 znode_t *zp = VTOZ(vp); 1858 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1859 znode_phys_t *pzp = zp->z_phys; 1860 int error; 1861 1862 ZFS_ENTER(zfsvfs); 1863 1864 /* 1865 * Return all attributes. It's cheaper to provide the answer 1866 * than to determine whether we were asked the question. 1867 */ 1868 mutex_enter(&zp->z_lock); 1869 1870 vap->va_type = vp->v_type; 1871 vap->va_mode = pzp->zp_mode & MODEMASK; 1872 vap->va_uid = zp->z_phys->zp_uid; 1873 vap->va_gid = zp->z_phys->zp_gid; 1874 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev; 1875 vap->va_nodeid = zp->z_id; 1876 vap->va_nlink = MIN(pzp->zp_links, UINT32_MAX); /* nlink_t limit! */ 1877 vap->va_size = pzp->zp_size; 1878 vap->va_rdev = pzp->zp_rdev; 1879 vap->va_seq = zp->z_seq; 1880 1881 ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime); 1882 ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime); 1883 ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime); 1884 1885 /* 1886 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 1887 * Also, if we are the owner don't bother, since owner should 1888 * always be allowed to read basic attributes of file. 1889 */ 1890 if (!(zp->z_phys->zp_flags & ZFS_ACL_TRIVIAL) && 1891 (zp->z_phys->zp_uid != crgetuid(cr))) { 1892 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, cr)) { 1893 mutex_exit(&zp->z_lock); 1894 ZFS_EXIT(zfsvfs); 1895 return (error); 1896 } 1897 } 1898 1899 mutex_exit(&zp->z_lock); 1900 1901 dmu_object_size_from_db(zp->z_dbuf, &vap->va_blksize, &vap->va_nblocks); 1902 1903 if (zp->z_blksz == 0) { 1904 /* 1905 * Block size hasn't been set; suggest maximal I/O transfers. 1906 */ 1907 vap->va_blksize = zfsvfs->z_max_blksz; 1908 } 1909 1910 ZFS_EXIT(zfsvfs); 1911 return (0); 1912 } 1913 1914 /* 1915 * Set the file attributes to the values contained in the 1916 * vattr structure. 1917 * 1918 * IN: vp - vnode of file to be modified. 1919 * vap - new attribute values. 1920 * flags - ATTR_UTIME set if non-default time values provided. 1921 * cr - credentials of caller. 1922 * 1923 * RETURN: 0 if success 1924 * error code if failure 1925 * 1926 * Timestamps: 1927 * vp - ctime updated, mtime updated if size changed. 1928 */ 1929 /* ARGSUSED */ 1930 static int 1931 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 1932 caller_context_t *ct) 1933 { 1934 struct znode *zp = VTOZ(vp); 1935 znode_phys_t *pzp = zp->z_phys; 1936 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1937 zilog_t *zilog = zfsvfs->z_log; 1938 uint64_t seq = 0; 1939 dmu_tx_t *tx; 1940 rl_t *rl; 1941 uint_t mask = vap->va_mask; 1942 uint_t mask_applied = 0; 1943 vattr_t oldva; 1944 int trim_mask = FALSE; 1945 int saved_mask; 1946 uint64_t new_mode; 1947 znode_t *attrzp; 1948 int need_policy = FALSE; 1949 int err; 1950 1951 if (mask == 0) 1952 return (0); 1953 1954 if (mask & AT_NOSET) 1955 return (EINVAL); 1956 1957 if (mask & AT_SIZE && vp->v_type == VDIR) 1958 return (EISDIR); 1959 1960 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) 1961 return (EINVAL); 1962 1963 ZFS_ENTER(zfsvfs); 1964 1965 top: 1966 rl = NULL; 1967 attrzp = NULL; 1968 1969 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 1970 ZFS_EXIT(zfsvfs); 1971 return (EROFS); 1972 } 1973 1974 /* 1975 * First validate permissions 1976 */ 1977 1978 if (mask & AT_SIZE) { 1979 err = zfs_zaccess(zp, ACE_WRITE_DATA, cr); 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 = TRUE; 2027 saved_mask = vap->va_mask; 2028 } else { 2029 need_policy = TRUE; 2030 } 2031 } else { 2032 need_policy = TRUE; 2033 } 2034 } 2035 2036 if (mask & AT_MODE) 2037 need_policy = TRUE; 2038 2039 if (need_policy) { 2040 mutex_enter(&zp->z_lock); 2041 oldva.va_mode = pzp->zp_mode; 2042 oldva.va_uid = zp->z_phys->zp_uid; 2043 oldva.va_gid = zp->z_phys->zp_gid; 2044 mutex_exit(&zp->z_lock); 2045 2046 /* 2047 * If trim_mask is set then take ownership 2048 * has been granted. In that case remove 2049 * UID|GID from mask so that 2050 * secpolicy_vnode_setattr() doesn't revoke it. 2051 */ 2052 if (trim_mask) 2053 vap->va_mask &= ~(AT_UID|AT_GID); 2054 2055 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 2056 (int (*)(void *, int, cred_t *))zfs_zaccess_rwx, zp); 2057 if (err) { 2058 ZFS_EXIT(zfsvfs); 2059 return (err); 2060 } 2061 2062 if (trim_mask) 2063 vap->va_mask |= (saved_mask & (AT_UID|AT_GID)); 2064 } 2065 2066 /* 2067 * secpolicy_vnode_setattr, or take ownership may have 2068 * changed va_mask 2069 */ 2070 mask = vap->va_mask; 2071 2072 tx = dmu_tx_create(zfsvfs->z_os); 2073 dmu_tx_hold_bonus(tx, zp->z_id); 2074 2075 if (mask & AT_MODE) { 2076 uint64_t pmode = pzp->zp_mode; 2077 2078 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 2079 2080 if (zp->z_phys->zp_acl.z_acl_extern_obj) 2081 dmu_tx_hold_write(tx, 2082 pzp->zp_acl.z_acl_extern_obj, 0, SPA_MAXBLOCKSIZE); 2083 else 2084 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2085 0, ZFS_ACL_SIZE(MAX_ACL_SIZE)); 2086 } 2087 2088 if (mask & AT_SIZE) { 2089 uint64_t off = vap->va_size; 2090 /* 2091 * Range lock the entire file, to ensure the truncate 2092 * is serialised. 2093 */ 2094 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 2095 ASSERT(rl != NULL); 2096 if (off < zp->z_phys->zp_size) 2097 dmu_tx_hold_free(tx, zp->z_id, off, DMU_OBJECT_END); 2098 else if (zp->z_blksz < zfsvfs->z_max_blksz && off > zp->z_blksz) 2099 /* we will rewrite this block if we grow */ 2100 dmu_tx_hold_write(tx, zp->z_id, 0, zp->z_phys->zp_size); 2101 } 2102 2103 if ((mask & (AT_UID | AT_GID)) && zp->z_phys->zp_xattr != 0) { 2104 err = zfs_zget(zp->z_zfsvfs, zp->z_phys->zp_xattr, &attrzp); 2105 if (err) { 2106 dmu_tx_abort(tx); 2107 if (rl != NULL) 2108 zfs_range_unlock(zp, rl); 2109 ZFS_EXIT(zfsvfs); 2110 return (err); 2111 } 2112 dmu_tx_hold_bonus(tx, attrzp->z_id); 2113 } 2114 2115 err = dmu_tx_assign(tx, zfsvfs->z_assign); 2116 if (err) { 2117 if (attrzp) 2118 VN_RELE(ZTOV(attrzp)); 2119 dmu_tx_abort(tx); 2120 if (rl != NULL) 2121 zfs_range_unlock(zp, rl); 2122 if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2123 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 2124 goto top; 2125 } 2126 ZFS_EXIT(zfsvfs); 2127 return (err); 2128 } 2129 2130 dmu_buf_will_dirty(zp->z_dbuf, tx); 2131 2132 /* 2133 * Set each attribute requested. 2134 * We group settings according to the locks they need to acquire. 2135 * 2136 * Note: you cannot set ctime directly, although it will be 2137 * updated as a side-effect of calling this function. 2138 */ 2139 if (mask & AT_SIZE) { 2140 /* 2141 * XXX - Note, we are not providing any open 2142 * mode flags here (like FNDELAY), so we may 2143 * block if there are locks present... this 2144 * should be addressed in openat(). 2145 */ 2146 err = zfs_freesp(zp, vap->va_size, 0, 0, tx, cr); 2147 if (err) { 2148 mutex_enter(&zp->z_lock); 2149 goto out; 2150 } 2151 mask_applied |= AT_SIZE; 2152 } 2153 2154 mask_applied = mask; /* no errors after this point */ 2155 2156 mutex_enter(&zp->z_lock); 2157 2158 if (mask & AT_MODE) { 2159 err = zfs_acl_chmod_setattr(zp, new_mode, tx); 2160 ASSERT3U(err, ==, 0); 2161 } 2162 2163 if (attrzp) 2164 mutex_enter(&attrzp->z_lock); 2165 2166 if (mask & AT_UID) { 2167 zp->z_phys->zp_uid = (uint64_t)vap->va_uid; 2168 if (attrzp) { 2169 attrzp->z_phys->zp_uid = (uint64_t)vap->va_uid; 2170 } 2171 } 2172 2173 if (mask & AT_GID) { 2174 zp->z_phys->zp_gid = (uint64_t)vap->va_gid; 2175 if (attrzp) 2176 attrzp->z_phys->zp_gid = (uint64_t)vap->va_gid; 2177 } 2178 2179 if (attrzp) 2180 mutex_exit(&attrzp->z_lock); 2181 2182 if (mask & AT_ATIME) 2183 ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime); 2184 2185 if (mask & AT_MTIME) 2186 ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime); 2187 2188 if (mask_applied & AT_SIZE) 2189 zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx); 2190 else if (mask_applied != 0) 2191 zfs_time_stamper_locked(zp, STATE_CHANGED, tx); 2192 2193 out: 2194 2195 if (mask_applied != 0) 2196 seq = zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, 2197 mask_applied); 2198 2199 mutex_exit(&zp->z_lock); 2200 2201 if (attrzp) 2202 VN_RELE(ZTOV(attrzp)); 2203 2204 if (rl != NULL) 2205 zfs_range_unlock(zp, rl); 2206 2207 dmu_tx_commit(tx); 2208 2209 zil_commit(zilog, seq, 0); 2210 2211 ZFS_EXIT(zfsvfs); 2212 return (err); 2213 } 2214 2215 /* 2216 * Search back through the directory tree, using the ".." entries. 2217 * Lock each directory in the chain to prevent concurrent renames. 2218 * Fail any attempt to move a directory into one of its own descendants. 2219 * XXX - z_parent_lock can overlap with map or grow locks 2220 */ 2221 typedef struct zfs_zlock { 2222 krwlock_t *zl_rwlock; /* lock we acquired */ 2223 znode_t *zl_znode; /* znode we held */ 2224 struct zfs_zlock *zl_next; /* next in list */ 2225 } zfs_zlock_t; 2226 2227 static int 2228 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 2229 { 2230 zfs_zlock_t *zl; 2231 znode_t *zp = tdzp; 2232 uint64_t rootid = zp->z_zfsvfs->z_root; 2233 uint64_t *oidp = &zp->z_id; 2234 krwlock_t *rwlp = &szp->z_parent_lock; 2235 krw_t rw = RW_WRITER; 2236 2237 /* 2238 * First pass write-locks szp and compares to zp->z_id. 2239 * Later passes read-lock zp and compare to zp->z_parent. 2240 */ 2241 do { 2242 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 2243 zl->zl_rwlock = rwlp; 2244 zl->zl_znode = NULL; 2245 zl->zl_next = *zlpp; 2246 *zlpp = zl; 2247 2248 rw_enter(rwlp, rw); 2249 2250 if (*oidp == szp->z_id) /* We're a descendant of szp */ 2251 return (EINVAL); 2252 2253 if (*oidp == rootid) /* We've hit the top */ 2254 return (0); 2255 2256 if (rw == RW_READER) { /* i.e. not the first pass */ 2257 int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp); 2258 if (error) 2259 return (error); 2260 zl->zl_znode = zp; 2261 } 2262 oidp = &zp->z_phys->zp_parent; 2263 rwlp = &zp->z_parent_lock; 2264 rw = RW_READER; 2265 2266 } while (zp->z_id != sdzp->z_id); 2267 2268 return (0); 2269 } 2270 2271 /* 2272 * Drop locks and release vnodes that were held by zfs_rename_lock(). 2273 */ 2274 static void 2275 zfs_rename_unlock(zfs_zlock_t **zlpp) 2276 { 2277 zfs_zlock_t *zl; 2278 2279 while ((zl = *zlpp) != NULL) { 2280 if (zl->zl_znode != NULL) 2281 VN_RELE(ZTOV(zl->zl_znode)); 2282 rw_exit(zl->zl_rwlock); 2283 *zlpp = zl->zl_next; 2284 kmem_free(zl, sizeof (*zl)); 2285 } 2286 } 2287 2288 /* 2289 * Move an entry from the provided source directory to the target 2290 * directory. Change the entry name as indicated. 2291 * 2292 * IN: sdvp - Source directory containing the "old entry". 2293 * snm - Old entry name. 2294 * tdvp - Target directory to contain the "new entry". 2295 * tnm - New entry name. 2296 * cr - credentials of caller. 2297 * 2298 * RETURN: 0 if success 2299 * error code if failure 2300 * 2301 * Timestamps: 2302 * sdvp,tdvp - ctime|mtime updated 2303 */ 2304 static int 2305 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr) 2306 { 2307 znode_t *tdzp, *szp, *tzp; 2308 znode_t *sdzp = VTOZ(sdvp); 2309 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs; 2310 zilog_t *zilog = zfsvfs->z_log; 2311 uint64_t seq = 0; 2312 vnode_t *realvp; 2313 zfs_dirlock_t *sdl, *tdl; 2314 dmu_tx_t *tx; 2315 zfs_zlock_t *zl; 2316 int cmp, serr, terr, error; 2317 2318 ZFS_ENTER(zfsvfs); 2319 2320 /* 2321 * Make sure we have the real vp for the target directory. 2322 */ 2323 if (VOP_REALVP(tdvp, &realvp) == 0) 2324 tdvp = realvp; 2325 2326 if (tdvp->v_vfsp != sdvp->v_vfsp) { 2327 ZFS_EXIT(zfsvfs); 2328 return (EXDEV); 2329 } 2330 2331 tdzp = VTOZ(tdvp); 2332 top: 2333 szp = NULL; 2334 tzp = NULL; 2335 zl = NULL; 2336 2337 /* 2338 * This is to prevent the creation of links into attribute space 2339 * by renaming a linked file into/outof an attribute directory. 2340 * See the comment in zfs_link() for why this is considered bad. 2341 */ 2342 if ((tdzp->z_phys->zp_flags & ZFS_XATTR) != 2343 (sdzp->z_phys->zp_flags & ZFS_XATTR)) { 2344 ZFS_EXIT(zfsvfs); 2345 return (EINVAL); 2346 } 2347 2348 /* 2349 * Lock source and target directory entries. To prevent deadlock, 2350 * a lock ordering must be defined. We lock the directory with 2351 * the smallest object id first, or if it's a tie, the one with 2352 * the lexically first name. 2353 */ 2354 if (sdzp->z_id < tdzp->z_id) { 2355 cmp = -1; 2356 } else if (sdzp->z_id > tdzp->z_id) { 2357 cmp = 1; 2358 } else { 2359 cmp = strcmp(snm, tnm); 2360 if (cmp == 0) { 2361 /* 2362 * POSIX: "If the old argument and the new argument 2363 * both refer to links to the same existing file, 2364 * the rename() function shall return successfully 2365 * and perform no other action." 2366 */ 2367 ZFS_EXIT(zfsvfs); 2368 return (0); 2369 } 2370 } 2371 if (cmp < 0) { 2372 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS); 2373 terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0); 2374 } else { 2375 terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0); 2376 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS); 2377 } 2378 2379 if (serr) { 2380 /* 2381 * Source entry invalid or not there. 2382 */ 2383 if (!terr) { 2384 zfs_dirent_unlock(tdl); 2385 if (tzp) 2386 VN_RELE(ZTOV(tzp)); 2387 } 2388 if (strcmp(snm, "..") == 0) 2389 serr = EINVAL; 2390 ZFS_EXIT(zfsvfs); 2391 return (serr); 2392 } 2393 if (terr) { 2394 zfs_dirent_unlock(sdl); 2395 VN_RELE(ZTOV(szp)); 2396 if (strcmp(tnm, "..") == 0) 2397 terr = EINVAL; 2398 ZFS_EXIT(zfsvfs); 2399 return (terr); 2400 } 2401 2402 /* 2403 * Must have write access at the source to remove the old entry 2404 * and write access at the target to create the new entry. 2405 * Note that if target and source are the same, this can be 2406 * done in a single check. 2407 */ 2408 2409 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 2410 goto out; 2411 2412 if (ZTOV(szp)->v_type == VDIR) { 2413 /* 2414 * Check to make sure rename is valid. 2415 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 2416 */ 2417 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 2418 goto out; 2419 } 2420 2421 /* 2422 * Does target exist? 2423 */ 2424 if (tzp) { 2425 /* 2426 * Source and target must be the same type. 2427 */ 2428 if (ZTOV(szp)->v_type == VDIR) { 2429 if (ZTOV(tzp)->v_type != VDIR) { 2430 error = ENOTDIR; 2431 goto out; 2432 } 2433 } else { 2434 if (ZTOV(tzp)->v_type == VDIR) { 2435 error = EISDIR; 2436 goto out; 2437 } 2438 } 2439 /* 2440 * POSIX dictates that when the source and target 2441 * entries refer to the same file object, rename 2442 * must do nothing and exit without error. 2443 */ 2444 if (szp->z_id == tzp->z_id) { 2445 error = 0; 2446 goto out; 2447 } 2448 } 2449 2450 vnevent_rename_src(ZTOV(szp)); 2451 if (tzp) 2452 vnevent_rename_dest(ZTOV(tzp)); 2453 2454 tx = dmu_tx_create(zfsvfs->z_os); 2455 dmu_tx_hold_bonus(tx, szp->z_id); /* nlink changes */ 2456 dmu_tx_hold_bonus(tx, sdzp->z_id); /* nlink changes */ 2457 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 2458 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 2459 if (sdzp != tdzp) 2460 dmu_tx_hold_bonus(tx, tdzp->z_id); /* nlink changes */ 2461 if (tzp) 2462 dmu_tx_hold_bonus(tx, tzp->z_id); /* parent changes */ 2463 dmu_tx_hold_zap(tx, zfsvfs->z_dqueue, FALSE, NULL); 2464 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2465 if (error) { 2466 dmu_tx_abort(tx); 2467 if (zl != NULL) 2468 zfs_rename_unlock(&zl); 2469 zfs_dirent_unlock(sdl); 2470 zfs_dirent_unlock(tdl); 2471 VN_RELE(ZTOV(szp)); 2472 if (tzp) 2473 VN_RELE(ZTOV(tzp)); 2474 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2475 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 2476 goto top; 2477 } 2478 ZFS_EXIT(zfsvfs); 2479 return (error); 2480 } 2481 2482 if (tzp) /* Attempt to remove the existing target */ 2483 error = zfs_link_destroy(tdl, tzp, tx, 0, NULL); 2484 2485 if (error == 0) { 2486 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 2487 if (error == 0) { 2488 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 2489 ASSERT(error == 0); 2490 seq = zfs_log_rename(zilog, tx, TX_RENAME, 2491 sdzp, sdl->dl_name, tdzp, tdl->dl_name, szp); 2492 } 2493 } 2494 2495 dmu_tx_commit(tx); 2496 out: 2497 if (zl != NULL) 2498 zfs_rename_unlock(&zl); 2499 2500 zfs_dirent_unlock(sdl); 2501 zfs_dirent_unlock(tdl); 2502 2503 VN_RELE(ZTOV(szp)); 2504 if (tzp) 2505 VN_RELE(ZTOV(tzp)); 2506 2507 zil_commit(zilog, seq, 0); 2508 2509 ZFS_EXIT(zfsvfs); 2510 return (error); 2511 } 2512 2513 /* 2514 * Insert the indicated symbolic reference entry into the directory. 2515 * 2516 * IN: dvp - Directory to contain new symbolic link. 2517 * link - Name for new symlink entry. 2518 * vap - Attributes of new entry. 2519 * target - Target path of new symlink. 2520 * cr - credentials of caller. 2521 * 2522 * RETURN: 0 if success 2523 * error code if failure 2524 * 2525 * Timestamps: 2526 * dvp - ctime|mtime updated 2527 */ 2528 static int 2529 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr) 2530 { 2531 znode_t *zp, *dzp = VTOZ(dvp); 2532 zfs_dirlock_t *dl; 2533 dmu_tx_t *tx; 2534 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2535 zilog_t *zilog = zfsvfs->z_log; 2536 uint64_t seq = 0; 2537 uint64_t zoid; 2538 int len = strlen(link); 2539 int error; 2540 2541 ASSERT(vap->va_type == VLNK); 2542 2543 ZFS_ENTER(zfsvfs); 2544 top: 2545 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 2546 ZFS_EXIT(zfsvfs); 2547 return (error); 2548 } 2549 2550 if (len > MAXPATHLEN) { 2551 ZFS_EXIT(zfsvfs); 2552 return (ENAMETOOLONG); 2553 } 2554 2555 /* 2556 * Attempt to lock directory; fail if entry already exists. 2557 */ 2558 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZNEW)) { 2559 ZFS_EXIT(zfsvfs); 2560 return (error); 2561 } 2562 2563 tx = dmu_tx_create(zfsvfs->z_os); 2564 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 2565 dmu_tx_hold_bonus(tx, dzp->z_id); 2566 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 2567 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 2568 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE); 2569 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2570 if (error) { 2571 dmu_tx_abort(tx); 2572 zfs_dirent_unlock(dl); 2573 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2574 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 2575 goto top; 2576 } 2577 ZFS_EXIT(zfsvfs); 2578 return (error); 2579 } 2580 2581 dmu_buf_will_dirty(dzp->z_dbuf, tx); 2582 2583 /* 2584 * Create a new object for the symlink. 2585 * Put the link content into bonus buffer if it will fit; 2586 * otherwise, store it just like any other file data. 2587 */ 2588 zoid = 0; 2589 if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) { 2590 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, len); 2591 if (len != 0) 2592 bcopy(link, zp->z_phys + 1, len); 2593 } else { 2594 dmu_buf_t *dbp; 2595 2596 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 2597 2598 /* 2599 * Nothing can access the znode yet so no locking needed 2600 * for growing the znode's blocksize. 2601 */ 2602 zfs_grow_blocksize(zp, len, tx); 2603 2604 VERIFY(0 == dmu_buf_hold(zfsvfs->z_os, zoid, 0, FTAG, &dbp)); 2605 dmu_buf_will_dirty(dbp, tx); 2606 2607 ASSERT3U(len, <=, dbp->db_size); 2608 bcopy(link, dbp->db_data, len); 2609 dmu_buf_rele(dbp, FTAG); 2610 } 2611 zp->z_phys->zp_size = len; 2612 2613 /* 2614 * Insert the new object into the directory. 2615 */ 2616 (void) zfs_link_create(dl, zp, tx, ZNEW); 2617 out: 2618 if (error == 0) 2619 seq = zfs_log_symlink(zilog, tx, TX_SYMLINK, 2620 dzp, zp, name, link); 2621 2622 dmu_tx_commit(tx); 2623 2624 zfs_dirent_unlock(dl); 2625 2626 VN_RELE(ZTOV(zp)); 2627 2628 zil_commit(zilog, seq, 0); 2629 2630 ZFS_EXIT(zfsvfs); 2631 return (error); 2632 } 2633 2634 /* 2635 * Return, in the buffer contained in the provided uio structure, 2636 * the symbolic path referred to by vp. 2637 * 2638 * IN: vp - vnode of symbolic link. 2639 * uoip - structure to contain the link path. 2640 * cr - credentials of caller. 2641 * 2642 * OUT: uio - structure to contain the link path. 2643 * 2644 * RETURN: 0 if success 2645 * error code if failure 2646 * 2647 * Timestamps: 2648 * vp - atime updated 2649 */ 2650 /* ARGSUSED */ 2651 static int 2652 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr) 2653 { 2654 znode_t *zp = VTOZ(vp); 2655 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2656 size_t bufsz; 2657 int error; 2658 2659 ZFS_ENTER(zfsvfs); 2660 2661 bufsz = (size_t)zp->z_phys->zp_size; 2662 if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) { 2663 error = uiomove(zp->z_phys + 1, 2664 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 2665 } else { 2666 dmu_buf_t *dbp; 2667 error = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0, FTAG, &dbp); 2668 if (error) { 2669 ZFS_EXIT(zfsvfs); 2670 return (error); 2671 } 2672 error = uiomove(dbp->db_data, 2673 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 2674 dmu_buf_rele(dbp, FTAG); 2675 } 2676 2677 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2678 ZFS_EXIT(zfsvfs); 2679 return (error); 2680 } 2681 2682 /* 2683 * Insert a new entry into directory tdvp referencing svp. 2684 * 2685 * IN: tdvp - Directory to contain new entry. 2686 * svp - vnode of new entry. 2687 * name - name of new entry. 2688 * cr - credentials of caller. 2689 * 2690 * RETURN: 0 if success 2691 * error code if failure 2692 * 2693 * Timestamps: 2694 * tdvp - ctime|mtime updated 2695 * svp - ctime updated 2696 */ 2697 /* ARGSUSED */ 2698 static int 2699 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr) 2700 { 2701 znode_t *dzp = VTOZ(tdvp); 2702 znode_t *tzp, *szp; 2703 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2704 zilog_t *zilog = zfsvfs->z_log; 2705 uint64_t seq = 0; 2706 zfs_dirlock_t *dl; 2707 dmu_tx_t *tx; 2708 vnode_t *realvp; 2709 int error; 2710 2711 ASSERT(tdvp->v_type == VDIR); 2712 2713 ZFS_ENTER(zfsvfs); 2714 2715 if (VOP_REALVP(svp, &realvp) == 0) 2716 svp = realvp; 2717 2718 if (svp->v_vfsp != tdvp->v_vfsp) { 2719 ZFS_EXIT(zfsvfs); 2720 return (EXDEV); 2721 } 2722 2723 szp = VTOZ(svp); 2724 top: 2725 /* 2726 * We do not support links between attributes and non-attributes 2727 * because of the potential security risk of creating links 2728 * into "normal" file space in order to circumvent restrictions 2729 * imposed in attribute space. 2730 */ 2731 if ((szp->z_phys->zp_flags & ZFS_XATTR) != 2732 (dzp->z_phys->zp_flags & ZFS_XATTR)) { 2733 ZFS_EXIT(zfsvfs); 2734 return (EINVAL); 2735 } 2736 2737 /* 2738 * POSIX dictates that we return EPERM here. 2739 * Better choices include ENOTSUP or EISDIR. 2740 */ 2741 if (svp->v_type == VDIR) { 2742 ZFS_EXIT(zfsvfs); 2743 return (EPERM); 2744 } 2745 2746 if ((uid_t)szp->z_phys->zp_uid != crgetuid(cr) && 2747 secpolicy_basic_link(cr) != 0) { 2748 ZFS_EXIT(zfsvfs); 2749 return (EPERM); 2750 } 2751 2752 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 2753 ZFS_EXIT(zfsvfs); 2754 return (error); 2755 } 2756 2757 /* 2758 * Attempt to lock directory; fail if entry already exists. 2759 */ 2760 if (error = zfs_dirent_lock(&dl, dzp, name, &tzp, ZNEW)) { 2761 ZFS_EXIT(zfsvfs); 2762 return (error); 2763 } 2764 2765 tx = dmu_tx_create(zfsvfs->z_os); 2766 dmu_tx_hold_bonus(tx, szp->z_id); 2767 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 2768 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2769 if (error) { 2770 dmu_tx_abort(tx); 2771 zfs_dirent_unlock(dl); 2772 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2773 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 2774 goto top; 2775 } 2776 ZFS_EXIT(zfsvfs); 2777 return (error); 2778 } 2779 2780 error = zfs_link_create(dl, szp, tx, 0); 2781 2782 if (error == 0) 2783 seq = zfs_log_link(zilog, tx, TX_LINK, dzp, szp, name); 2784 2785 dmu_tx_commit(tx); 2786 2787 zfs_dirent_unlock(dl); 2788 2789 zil_commit(zilog, seq, 0); 2790 2791 ZFS_EXIT(zfsvfs); 2792 return (error); 2793 } 2794 2795 /* 2796 * zfs_null_putapage() is used when the file system has been force 2797 * unmounted. It just drops the pages. 2798 */ 2799 /* ARGSUSED */ 2800 static int 2801 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 2802 size_t *lenp, int flags, cred_t *cr) 2803 { 2804 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR); 2805 return (0); 2806 } 2807 2808 /* ARGSUSED */ 2809 static int 2810 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 2811 size_t *lenp, int flags, cred_t *cr) 2812 { 2813 znode_t *zp = VTOZ(vp); 2814 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2815 zilog_t *zilog = zfsvfs->z_log; 2816 dmu_tx_t *tx; 2817 rl_t *rl; 2818 u_offset_t off; 2819 ssize_t len; 2820 caddr_t va; 2821 int err; 2822 2823 top: 2824 off = pp->p_offset; 2825 rl = zfs_range_lock(zp, off, PAGESIZE, RL_WRITER); 2826 len = MIN(PAGESIZE, zp->z_phys->zp_size - off); 2827 2828 tx = dmu_tx_create(zfsvfs->z_os); 2829 dmu_tx_hold_write(tx, zp->z_id, off, len); 2830 dmu_tx_hold_bonus(tx, zp->z_id); 2831 err = dmu_tx_assign(tx, zfsvfs->z_assign); 2832 if (err != 0) { 2833 dmu_tx_abort(tx); 2834 zfs_range_unlock(zp, rl); 2835 if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2836 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 2837 goto top; 2838 } 2839 goto out; 2840 } 2841 2842 va = ppmapin(pp, PROT_READ | PROT_WRITE, (caddr_t)-1); 2843 2844 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx); 2845 2846 ppmapout(va); 2847 2848 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 2849 (void) zfs_log_write(zilog, tx, TX_WRITE, zp, off, len, 0, NULL); 2850 dmu_tx_commit(tx); 2851 2852 zfs_range_unlock(zp, rl); 2853 2854 pvn_write_done(pp, B_WRITE | flags); 2855 if (offp) 2856 *offp = off; 2857 if (lenp) 2858 *lenp = len; 2859 2860 out: 2861 return (err); 2862 } 2863 2864 /* 2865 * Copy the portion of the file indicated from pages into the file. 2866 * The pages are stored in a page list attached to the files vnode. 2867 * 2868 * IN: vp - vnode of file to push page data to. 2869 * off - position in file to put data. 2870 * len - amount of data to write. 2871 * flags - flags to control the operation. 2872 * cr - credentials of caller. 2873 * 2874 * RETURN: 0 if success 2875 * error code if failure 2876 * 2877 * Timestamps: 2878 * vp - ctime|mtime updated 2879 */ 2880 static int 2881 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr) 2882 { 2883 znode_t *zp = VTOZ(vp); 2884 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2885 page_t *pp; 2886 size_t io_len; 2887 u_offset_t io_off; 2888 uint64_t filesz; 2889 int error = 0; 2890 2891 ZFS_ENTER(zfsvfs); 2892 2893 ASSERT(zp->z_dbuf_held && zp->z_phys); 2894 2895 if (len == 0) { 2896 /* 2897 * Search the entire vp list for pages >= off. 2898 */ 2899 error = pvn_vplist_dirty(vp, (u_offset_t)off, zfs_putapage, 2900 flags, cr); 2901 goto out; 2902 } 2903 2904 filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */ 2905 if (off > filesz) { 2906 /* past end of file */ 2907 ZFS_EXIT(zfsvfs); 2908 return (0); 2909 } 2910 2911 len = MIN(len, filesz - off); 2912 2913 for (io_off = off; io_off < off + len; io_off += io_len) { 2914 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { 2915 pp = page_lookup(vp, io_off, 2916 (flags & (B_INVAL | B_FREE)) ? 2917 SE_EXCL : SE_SHARED); 2918 } else { 2919 pp = page_lookup_nowait(vp, io_off, 2920 (flags & B_FREE) ? SE_EXCL : SE_SHARED); 2921 } 2922 2923 if (pp != NULL && pvn_getdirty(pp, flags)) { 2924 int err; 2925 2926 /* 2927 * Found a dirty page to push 2928 */ 2929 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr); 2930 if (err) 2931 error = err; 2932 } else { 2933 io_len = PAGESIZE; 2934 } 2935 } 2936 out: 2937 zil_commit(zfsvfs->z_log, UINT64_MAX, (flags & B_ASYNC) ? 0 : FDSYNC); 2938 ZFS_EXIT(zfsvfs); 2939 return (error); 2940 } 2941 2942 void 2943 zfs_inactive(vnode_t *vp, cred_t *cr) 2944 { 2945 znode_t *zp = VTOZ(vp); 2946 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2947 int error; 2948 2949 rw_enter(&zfsvfs->z_um_lock, RW_READER); 2950 if (zfsvfs->z_unmounted2) { 2951 ASSERT(zp->z_dbuf_held == 0); 2952 2953 if (vn_has_cached_data(vp)) { 2954 (void) pvn_vplist_dirty(vp, 0, zfs_null_putapage, 2955 B_INVAL, cr); 2956 } 2957 2958 mutex_enter(&zp->z_lock); 2959 vp->v_count = 0; /* count arrives as 1 */ 2960 if (zp->z_dbuf == NULL) { 2961 mutex_exit(&zp->z_lock); 2962 zfs_znode_free(zp); 2963 } else { 2964 mutex_exit(&zp->z_lock); 2965 } 2966 rw_exit(&zfsvfs->z_um_lock); 2967 VFS_RELE(zfsvfs->z_vfs); 2968 return; 2969 } 2970 2971 /* 2972 * Attempt to push any data in the page cache. If this fails 2973 * we will get kicked out later in zfs_zinactive(). 2974 */ 2975 if (vn_has_cached_data(vp)) { 2976 (void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC, 2977 cr); 2978 } 2979 2980 if (zp->z_atime_dirty && zp->z_reap == 0) { 2981 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 2982 2983 dmu_tx_hold_bonus(tx, zp->z_id); 2984 error = dmu_tx_assign(tx, TXG_WAIT); 2985 if (error) { 2986 dmu_tx_abort(tx); 2987 } else { 2988 dmu_buf_will_dirty(zp->z_dbuf, tx); 2989 mutex_enter(&zp->z_lock); 2990 zp->z_atime_dirty = 0; 2991 mutex_exit(&zp->z_lock); 2992 dmu_tx_commit(tx); 2993 } 2994 } 2995 2996 zfs_zinactive(zp); 2997 rw_exit(&zfsvfs->z_um_lock); 2998 } 2999 3000 /* 3001 * Bounds-check the seek operation. 3002 * 3003 * IN: vp - vnode seeking within 3004 * ooff - old file offset 3005 * noffp - pointer to new file offset 3006 * 3007 * RETURN: 0 if success 3008 * EINVAL if new offset invalid 3009 */ 3010 /* ARGSUSED */ 3011 static int 3012 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp) 3013 { 3014 if (vp->v_type == VDIR) 3015 return (0); 3016 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0); 3017 } 3018 3019 /* 3020 * Pre-filter the generic locking function to trap attempts to place 3021 * a mandatory lock on a memory mapped file. 3022 */ 3023 static int 3024 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset, 3025 flk_callback_t *flk_cbp, cred_t *cr) 3026 { 3027 znode_t *zp = VTOZ(vp); 3028 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3029 int error; 3030 3031 ZFS_ENTER(zfsvfs); 3032 3033 /* 3034 * We are following the UFS semantics with respect to mapcnt 3035 * here: If we see that the file is mapped already, then we will 3036 * return an error, but we don't worry about races between this 3037 * function and zfs_map(). 3038 */ 3039 if (zp->z_mapcnt > 0 && MANDMODE((mode_t)zp->z_phys->zp_mode)) { 3040 ZFS_EXIT(zfsvfs); 3041 return (EAGAIN); 3042 } 3043 error = fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr); 3044 ZFS_EXIT(zfsvfs); 3045 return (error); 3046 } 3047 3048 /* 3049 * If we can't find a page in the cache, we will create a new page 3050 * and fill it with file data. For efficiency, we may try to fill 3051 * multiple pages at once (klustering). 3052 */ 3053 static int 3054 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, 3055 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw) 3056 { 3057 znode_t *zp = VTOZ(vp); 3058 page_t *pp, *cur_pp; 3059 objset_t *os = zp->z_zfsvfs->z_os; 3060 caddr_t va; 3061 u_offset_t io_off, total; 3062 uint64_t oid = zp->z_id; 3063 size_t io_len; 3064 uint64_t filesz; 3065 int err; 3066 3067 /* 3068 * If we are only asking for a single page don't bother klustering. 3069 */ 3070 filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */ 3071 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE || off > filesz) { 3072 io_off = off; 3073 io_len = PAGESIZE; 3074 pp = page_create_va(vp, io_off, io_len, PG_WAIT, seg, addr); 3075 } else { 3076 /* 3077 * Try to fill a kluster of pages (a blocks worth). 3078 */ 3079 size_t klen; 3080 u_offset_t koff; 3081 3082 if (!ISP2(zp->z_blksz)) { 3083 /* Only one block in the file. */ 3084 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 3085 koff = 0; 3086 } else { 3087 klen = plsz; 3088 koff = P2ALIGN(off, (u_offset_t)klen); 3089 } 3090 if (klen > filesz) 3091 klen = P2ROUNDUP(filesz, (uint64_t)PAGESIZE); 3092 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 3093 &io_len, koff, klen, 0); 3094 } 3095 if (pp == NULL) { 3096 /* 3097 * Some other thread entered the page before us. 3098 * Return to zfs_getpage to retry the lookup. 3099 */ 3100 *pl = NULL; 3101 return (0); 3102 } 3103 3104 /* 3105 * Fill the pages in the kluster. 3106 */ 3107 cur_pp = pp; 3108 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) { 3109 ASSERT(io_off == cur_pp->p_offset); 3110 va = ppmapin(cur_pp, PROT_READ | PROT_WRITE, (caddr_t)-1); 3111 err = dmu_read(os, oid, io_off, PAGESIZE, va); 3112 ppmapout(va); 3113 if (err) { 3114 /* On error, toss the entire kluster */ 3115 pvn_read_done(pp, B_ERROR); 3116 return (err); 3117 } 3118 cur_pp = cur_pp->p_next; 3119 } 3120 out: 3121 /* 3122 * Fill in the page list array from the kluster. If 3123 * there are too many pages in the kluster, return 3124 * as many pages as possible starting from the desired 3125 * offset `off'. 3126 * NOTE: the page list will always be null terminated. 3127 */ 3128 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 3129 3130 return (0); 3131 } 3132 3133 /* 3134 * Return pointers to the pages for the file region [off, off + len] 3135 * in the pl array. If plsz is greater than len, this function may 3136 * also return page pointers from before or after the specified 3137 * region (i.e. some region [off', off' + plsz]). These additional 3138 * pages are only returned if they are already in the cache, or were 3139 * created as part of a klustered read. 3140 * 3141 * IN: vp - vnode of file to get data from. 3142 * off - position in file to get data from. 3143 * len - amount of data to retrieve. 3144 * plsz - length of provided page list. 3145 * seg - segment to obtain pages for. 3146 * addr - virtual address of fault. 3147 * rw - mode of created pages. 3148 * cr - credentials of caller. 3149 * 3150 * OUT: protp - protection mode of created pages. 3151 * pl - list of pages created. 3152 * 3153 * RETURN: 0 if success 3154 * error code if failure 3155 * 3156 * Timestamps: 3157 * vp - atime updated 3158 */ 3159 /* ARGSUSED */ 3160 static int 3161 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 3162 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 3163 enum seg_rw rw, cred_t *cr) 3164 { 3165 znode_t *zp = VTOZ(vp); 3166 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3167 page_t *pp, **pl0 = pl; 3168 rl_t *rl; 3169 int cnt = 0, need_unlock = 0, err = 0; 3170 3171 ZFS_ENTER(zfsvfs); 3172 3173 if (protp) 3174 *protp = PROT_ALL; 3175 3176 ASSERT(zp->z_dbuf_held && zp->z_phys); 3177 3178 /* no faultahead (for now) */ 3179 if (pl == NULL) { 3180 ZFS_EXIT(zfsvfs); 3181 return (0); 3182 } 3183 3184 /* 3185 * Make sure nobody restructures the file in the middle of the getpage. 3186 */ 3187 rl = zfs_range_lock(zp, off, len, RL_READER); 3188 3189 /* can't fault past EOF */ 3190 if (off >= zp->z_phys->zp_size) { 3191 zfs_range_unlock(zp, rl); 3192 ZFS_EXIT(zfsvfs); 3193 return (EFAULT); 3194 } 3195 3196 /* 3197 * If we already own the lock, then we must be page faulting 3198 * in the middle of a write to this file (i.e., we are writing 3199 * to this file using data from a mapped region of the file). 3200 */ 3201 if (!rw_owner(&zp->z_map_lock)) { 3202 rw_enter(&zp->z_map_lock, RW_WRITER); 3203 need_unlock = TRUE; 3204 } 3205 3206 /* 3207 * Loop through the requested range [off, off + len] looking 3208 * for pages. If we don't find a page, we will need to create 3209 * a new page and fill it with data from the file. 3210 */ 3211 while (len > 0) { 3212 if (plsz < PAGESIZE) 3213 break; 3214 if (pp = page_lookup(vp, off, SE_SHARED)) { 3215 *pl++ = pp; 3216 off += PAGESIZE; 3217 addr += PAGESIZE; 3218 len -= PAGESIZE; 3219 plsz -= PAGESIZE; 3220 } else { 3221 err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw); 3222 /* 3223 * klustering may have changed our region 3224 * to be block aligned. 3225 */ 3226 if (((pp = *pl) != 0) && (off != pp->p_offset)) { 3227 int delta = off - pp->p_offset; 3228 len += delta; 3229 off -= delta; 3230 addr -= delta; 3231 } 3232 while (*pl) { 3233 pl++; 3234 cnt++; 3235 off += PAGESIZE; 3236 addr += PAGESIZE; 3237 plsz -= PAGESIZE; 3238 if (len > PAGESIZE) 3239 len -= PAGESIZE; 3240 else 3241 len = 0; 3242 } 3243 if (err) { 3244 /* 3245 * Release any pages we have locked. 3246 */ 3247 while (pl > pl0) 3248 page_unlock(*--pl); 3249 goto out; 3250 } 3251 } 3252 } 3253 3254 /* 3255 * Fill out the page array with any pages already in the cache. 3256 */ 3257 while (plsz > 0) { 3258 pp = page_lookup_nowait(vp, off, SE_SHARED); 3259 if (pp == NULL) 3260 break; 3261 *pl++ = pp; 3262 off += PAGESIZE; 3263 plsz -= PAGESIZE; 3264 } 3265 3266 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 3267 out: 3268 *pl = NULL; 3269 3270 if (need_unlock) 3271 rw_exit(&zp->z_map_lock); 3272 zfs_range_unlock(zp, rl); 3273 3274 ZFS_EXIT(zfsvfs); 3275 return (err); 3276 } 3277 3278 /* 3279 * Request a memory map for a section of a file. This code interacts 3280 * with common code and the VM system as follows: 3281 * 3282 * common code calls mmap(), which ends up in smmap_common() 3283 * 3284 * this calls VOP_MAP(), which takes you into (say) zfs 3285 * 3286 * zfs_map() calls as_map(), passing segvn_create() as the callback 3287 * 3288 * segvn_create() creates the new segment and calls VOP_ADDMAP() 3289 * 3290 * zfs_addmap() updates z_mapcnt 3291 */ 3292 static int 3293 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 3294 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr) 3295 { 3296 znode_t *zp = VTOZ(vp); 3297 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3298 segvn_crargs_t vn_a; 3299 int error; 3300 3301 ZFS_ENTER(zfsvfs); 3302 3303 if (vp->v_flag & VNOMAP) { 3304 ZFS_EXIT(zfsvfs); 3305 return (ENOSYS); 3306 } 3307 3308 if (off < 0 || len > MAXOFFSET_T - off) { 3309 ZFS_EXIT(zfsvfs); 3310 return (ENXIO); 3311 } 3312 3313 if (vp->v_type != VREG) { 3314 ZFS_EXIT(zfsvfs); 3315 return (ENODEV); 3316 } 3317 3318 /* 3319 * If file is locked, disallow mapping. 3320 */ 3321 if (MANDMODE((mode_t)zp->z_phys->zp_mode) && vn_has_flocks(vp)) { 3322 ZFS_EXIT(zfsvfs); 3323 return (EAGAIN); 3324 } 3325 3326 as_rangelock(as); 3327 if ((flags & MAP_FIXED) == 0) { 3328 map_addr(addrp, len, off, 1, flags); 3329 if (*addrp == NULL) { 3330 as_rangeunlock(as); 3331 ZFS_EXIT(zfsvfs); 3332 return (ENOMEM); 3333 } 3334 } else { 3335 /* 3336 * User specified address - blow away any previous mappings 3337 */ 3338 (void) as_unmap(as, *addrp, len); 3339 } 3340 3341 vn_a.vp = vp; 3342 vn_a.offset = (u_offset_t)off; 3343 vn_a.type = flags & MAP_TYPE; 3344 vn_a.prot = prot; 3345 vn_a.maxprot = maxprot; 3346 vn_a.cred = cr; 3347 vn_a.amp = NULL; 3348 vn_a.flags = flags & ~MAP_TYPE; 3349 vn_a.szc = 0; 3350 vn_a.lgrp_mem_policy_flags = 0; 3351 3352 error = as_map(as, *addrp, len, segvn_create, &vn_a); 3353 3354 as_rangeunlock(as); 3355 ZFS_EXIT(zfsvfs); 3356 return (error); 3357 } 3358 3359 /* ARGSUSED */ 3360 static int 3361 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 3362 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr) 3363 { 3364 uint64_t pages = btopr(len); 3365 3366 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages); 3367 return (0); 3368 } 3369 3370 /* 3371 * The reason we push dirty pages as part of zfs_delmap() is so that we get a 3372 * more accurate mtime for the associated file. Since we don't have a way of 3373 * detecting when the data was actually modified, we have to resort to 3374 * heuristics. If an explicit msync() is done, then we mark the mtime when the 3375 * last page is pushed. The problem occurs when the msync() call is omitted, 3376 * which by far the most common case: 3377 * 3378 * open() 3379 * mmap() 3380 * <modify memory> 3381 * munmap() 3382 * close() 3383 * <time lapse> 3384 * putpage() via fsflush 3385 * 3386 * If we wait until fsflush to come along, we can have a modification time that 3387 * is some arbitrary point in the future. In order to prevent this in the 3388 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is 3389 * torn down. 3390 */ 3391 /* ARGSUSED */ 3392 static int 3393 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 3394 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr) 3395 { 3396 uint64_t pages = btopr(len); 3397 3398 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages); 3399 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages); 3400 3401 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) && 3402 vn_has_cached_data(vp)) 3403 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr); 3404 3405 return (0); 3406 } 3407 3408 /* 3409 * Free or allocate space in a file. Currently, this function only 3410 * supports the `F_FREESP' command. However, this command is somewhat 3411 * misnamed, as its functionality includes the ability to allocate as 3412 * well as free space. 3413 * 3414 * IN: vp - vnode of file to free data in. 3415 * cmd - action to take (only F_FREESP supported). 3416 * bfp - section of file to free/alloc. 3417 * flag - current file open mode flags. 3418 * offset - current file offset. 3419 * cr - credentials of caller [UNUSED]. 3420 * 3421 * RETURN: 0 if success 3422 * error code if failure 3423 * 3424 * Timestamps: 3425 * vp - ctime|mtime updated 3426 */ 3427 /* ARGSUSED */ 3428 static int 3429 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag, 3430 offset_t offset, cred_t *cr, caller_context_t *ct) 3431 { 3432 dmu_tx_t *tx; 3433 znode_t *zp = VTOZ(vp); 3434 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3435 zilog_t *zilog = zfsvfs->z_log; 3436 rl_t *rl; 3437 uint64_t seq = 0; 3438 uint64_t off, len; 3439 int error; 3440 3441 ZFS_ENTER(zfsvfs); 3442 3443 top: 3444 if (cmd != F_FREESP) { 3445 ZFS_EXIT(zfsvfs); 3446 return (EINVAL); 3447 } 3448 3449 if (error = convoff(vp, bfp, 0, offset)) { 3450 ZFS_EXIT(zfsvfs); 3451 return (error); 3452 } 3453 3454 if (bfp->l_len < 0) { 3455 ZFS_EXIT(zfsvfs); 3456 return (EINVAL); 3457 } 3458 3459 off = bfp->l_start; 3460 len = bfp->l_len; /* 0 means from off to end of file */ 3461 tx = dmu_tx_create(zfsvfs->z_os); 3462 dmu_tx_hold_bonus(tx, zp->z_id); 3463 /* 3464 * If we will change zp_size (in zfs_freesp) then lock the whole file, 3465 * otherwise just lock the range being freed. 3466 */ 3467 if (len == 0 || off + len > zp->z_phys->zp_size) { 3468 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 3469 } else { 3470 rl = zfs_range_lock(zp, off, len, RL_WRITER); 3471 /* recheck, in case zp_size changed */ 3472 if (off + len > zp->z_phys->zp_size) { 3473 /* lost race: file size changed, lock whole file */ 3474 zfs_range_unlock(zp, rl); 3475 rl = zfs_range_lock(zp, 0, UINT64_MAX, RL_WRITER); 3476 } 3477 } 3478 3479 if (off + len > zp->z_blksz && zp->z_blksz < zfsvfs->z_max_blksz && 3480 off >= zp->z_phys->zp_size) { 3481 /* 3482 * We are increasing the length of the file, 3483 * and this may mean a block size increase. 3484 */ 3485 dmu_tx_hold_write(tx, zp->z_id, 0, 3486 MIN(off + len, zfsvfs->z_max_blksz)); 3487 } else if (off < zp->z_phys->zp_size) { 3488 /* 3489 * If len == 0, we are truncating the file. 3490 */ 3491 dmu_tx_hold_free(tx, zp->z_id, off, len ? len : DMU_OBJECT_END); 3492 } 3493 3494 error = dmu_tx_assign(tx, zfsvfs->z_assign); 3495 if (error) { 3496 dmu_tx_abort(tx); 3497 zfs_range_unlock(zp, rl); 3498 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 3499 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 3500 goto top; 3501 } 3502 ZFS_EXIT(zfsvfs); 3503 return (error); 3504 } 3505 3506 error = zfs_freesp(zp, off, len, flag, tx, cr); 3507 3508 if (error == 0) { 3509 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 3510 seq = zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len); 3511 } 3512 3513 zfs_range_unlock(zp, rl); 3514 3515 dmu_tx_commit(tx); 3516 3517 zil_commit(zilog, seq, 0); 3518 3519 ZFS_EXIT(zfsvfs); 3520 return (error); 3521 } 3522 3523 static int 3524 zfs_fid(vnode_t *vp, fid_t *fidp) 3525 { 3526 znode_t *zp = VTOZ(vp); 3527 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3528 uint32_t gen = (uint32_t)zp->z_phys->zp_gen; 3529 uint64_t object = zp->z_id; 3530 zfid_short_t *zfid; 3531 int size, i; 3532 3533 ZFS_ENTER(zfsvfs); 3534 3535 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 3536 if (fidp->fid_len < size) { 3537 fidp->fid_len = size; 3538 ZFS_EXIT(zfsvfs); 3539 return (ENOSPC); 3540 } 3541 3542 zfid = (zfid_short_t *)fidp; 3543 3544 zfid->zf_len = size; 3545 3546 for (i = 0; i < sizeof (zfid->zf_object); i++) 3547 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 3548 3549 /* Must have a non-zero generation number to distinguish from .zfs */ 3550 if (gen == 0) 3551 gen = 1; 3552 for (i = 0; i < sizeof (zfid->zf_gen); i++) 3553 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 3554 3555 if (size == LONG_FID_LEN) { 3556 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 3557 zfid_long_t *zlfid; 3558 3559 zlfid = (zfid_long_t *)fidp; 3560 3561 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 3562 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 3563 3564 /* XXX - this should be the generation number for the objset */ 3565 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 3566 zlfid->zf_setgen[i] = 0; 3567 } 3568 3569 ZFS_EXIT(zfsvfs); 3570 return (0); 3571 } 3572 3573 static int 3574 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr) 3575 { 3576 znode_t *zp, *xzp; 3577 zfsvfs_t *zfsvfs; 3578 zfs_dirlock_t *dl; 3579 int error; 3580 3581 switch (cmd) { 3582 case _PC_LINK_MAX: 3583 *valp = ULONG_MAX; 3584 return (0); 3585 3586 case _PC_FILESIZEBITS: 3587 *valp = 64; 3588 return (0); 3589 3590 case _PC_XATTR_EXISTS: 3591 zp = VTOZ(vp); 3592 zfsvfs = zp->z_zfsvfs; 3593 ZFS_ENTER(zfsvfs); 3594 *valp = 0; 3595 error = zfs_dirent_lock(&dl, zp, "", &xzp, 3596 ZXATTR | ZEXISTS | ZSHARED); 3597 if (error == 0) { 3598 zfs_dirent_unlock(dl); 3599 if (!zfs_dirempty(xzp)) 3600 *valp = 1; 3601 VN_RELE(ZTOV(xzp)); 3602 } else if (error == ENOENT) { 3603 /* 3604 * If there aren't extended attributes, it's the 3605 * same as having zero of them. 3606 */ 3607 error = 0; 3608 } 3609 ZFS_EXIT(zfsvfs); 3610 return (error); 3611 3612 case _PC_ACL_ENABLED: 3613 *valp = _ACL_ACE_ENABLED; 3614 return (0); 3615 3616 case _PC_MIN_HOLE_SIZE: 3617 *valp = (ulong_t)SPA_MINBLOCKSIZE; 3618 return (0); 3619 3620 default: 3621 return (fs_pathconf(vp, cmd, valp, cr)); 3622 } 3623 } 3624 3625 /*ARGSUSED*/ 3626 static int 3627 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr) 3628 { 3629 znode_t *zp = VTOZ(vp); 3630 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3631 int error; 3632 3633 ZFS_ENTER(zfsvfs); 3634 error = zfs_getacl(zp, vsecp, cr); 3635 ZFS_EXIT(zfsvfs); 3636 3637 return (error); 3638 } 3639 3640 /*ARGSUSED*/ 3641 static int 3642 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr) 3643 { 3644 znode_t *zp = VTOZ(vp); 3645 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3646 int error; 3647 3648 ZFS_ENTER(zfsvfs); 3649 error = zfs_setacl(zp, vsecp, cr); 3650 ZFS_EXIT(zfsvfs); 3651 return (error); 3652 } 3653 3654 /* 3655 * Predeclare these here so that the compiler assumes that 3656 * this is an "old style" function declaration that does 3657 * not include arguments => we won't get type mismatch errors 3658 * in the initializations that follow. 3659 */ 3660 static int zfs_inval(); 3661 static int zfs_isdir(); 3662 3663 static int 3664 zfs_inval() 3665 { 3666 return (EINVAL); 3667 } 3668 3669 static int 3670 zfs_isdir() 3671 { 3672 return (EISDIR); 3673 } 3674 /* 3675 * Directory vnode operations template 3676 */ 3677 vnodeops_t *zfs_dvnodeops; 3678 const fs_operation_def_t zfs_dvnodeops_template[] = { 3679 VOPNAME_OPEN, zfs_open, 3680 VOPNAME_CLOSE, zfs_close, 3681 VOPNAME_READ, zfs_isdir, 3682 VOPNAME_WRITE, zfs_isdir, 3683 VOPNAME_IOCTL, zfs_ioctl, 3684 VOPNAME_GETATTR, zfs_getattr, 3685 VOPNAME_SETATTR, zfs_setattr, 3686 VOPNAME_ACCESS, zfs_access, 3687 VOPNAME_LOOKUP, zfs_lookup, 3688 VOPNAME_CREATE, zfs_create, 3689 VOPNAME_REMOVE, zfs_remove, 3690 VOPNAME_LINK, zfs_link, 3691 VOPNAME_RENAME, zfs_rename, 3692 VOPNAME_MKDIR, zfs_mkdir, 3693 VOPNAME_RMDIR, zfs_rmdir, 3694 VOPNAME_READDIR, zfs_readdir, 3695 VOPNAME_SYMLINK, zfs_symlink, 3696 VOPNAME_FSYNC, zfs_fsync, 3697 VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive, 3698 VOPNAME_FID, zfs_fid, 3699 VOPNAME_SEEK, zfs_seek, 3700 VOPNAME_PATHCONF, zfs_pathconf, 3701 VOPNAME_GETSECATTR, zfs_getsecattr, 3702 VOPNAME_SETSECATTR, zfs_setsecattr, 3703 NULL, NULL 3704 }; 3705 3706 /* 3707 * Regular file vnode operations template 3708 */ 3709 vnodeops_t *zfs_fvnodeops; 3710 const fs_operation_def_t zfs_fvnodeops_template[] = { 3711 VOPNAME_OPEN, zfs_open, 3712 VOPNAME_CLOSE, zfs_close, 3713 VOPNAME_READ, zfs_read, 3714 VOPNAME_WRITE, zfs_write, 3715 VOPNAME_IOCTL, zfs_ioctl, 3716 VOPNAME_GETATTR, zfs_getattr, 3717 VOPNAME_SETATTR, zfs_setattr, 3718 VOPNAME_ACCESS, zfs_access, 3719 VOPNAME_LOOKUP, zfs_lookup, 3720 VOPNAME_RENAME, zfs_rename, 3721 VOPNAME_FSYNC, zfs_fsync, 3722 VOPNAME_INACTIVE, (fs_generic_func_p)zfs_inactive, 3723 VOPNAME_FID, zfs_fid, 3724 VOPNAME_SEEK, zfs_seek, 3725 VOPNAME_FRLOCK, zfs_frlock, 3726 VOPNAME_SPACE, zfs_space, 3727 VOPNAME_GETPAGE, zfs_getpage, 3728 VOPNAME_PUTPAGE, zfs_putpage, 3729 VOPNAME_MAP, (fs_generic_func_p) zfs_map, 3730 VOPNAME_ADDMAP, (fs_generic_func_p) zfs_addmap, 3731 VOPNAME_DELMAP, zfs_delmap, 3732 VOPNAME_PATHCONF, zfs_pathconf, 3733 VOPNAME_GETSECATTR, zfs_getsecattr, 3734 VOPNAME_SETSECATTR, zfs_setsecattr, 3735 VOPNAME_VNEVENT, fs_vnevent_support, 3736 NULL, NULL 3737 }; 3738 3739 /* 3740 * Symbolic link vnode operations template 3741 */ 3742 vnodeops_t *zfs_symvnodeops; 3743 const fs_operation_def_t zfs_symvnodeops_template[] = { 3744 VOPNAME_GETATTR, zfs_getattr, 3745 VOPNAME_SETATTR, zfs_setattr, 3746 VOPNAME_ACCESS, zfs_access, 3747 VOPNAME_RENAME, zfs_rename, 3748 VOPNAME_READLINK, zfs_readlink, 3749 VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive, 3750 VOPNAME_FID, zfs_fid, 3751 VOPNAME_PATHCONF, zfs_pathconf, 3752 VOPNAME_VNEVENT, fs_vnevent_support, 3753 NULL, NULL 3754 }; 3755 3756 /* 3757 * Extended attribute directory vnode operations template 3758 * This template is identical to the directory vnodes 3759 * operation template except for restricted operations: 3760 * VOP_MKDIR() 3761 * VOP_SYMLINK() 3762 * Note that there are other restrictions embedded in: 3763 * zfs_create() - restrict type to VREG 3764 * zfs_link() - no links into/out of attribute space 3765 * zfs_rename() - no moves into/out of attribute space 3766 */ 3767 vnodeops_t *zfs_xdvnodeops; 3768 const fs_operation_def_t zfs_xdvnodeops_template[] = { 3769 VOPNAME_OPEN, zfs_open, 3770 VOPNAME_CLOSE, zfs_close, 3771 VOPNAME_IOCTL, zfs_ioctl, 3772 VOPNAME_GETATTR, zfs_getattr, 3773 VOPNAME_SETATTR, zfs_setattr, 3774 VOPNAME_ACCESS, zfs_access, 3775 VOPNAME_LOOKUP, zfs_lookup, 3776 VOPNAME_CREATE, zfs_create, 3777 VOPNAME_REMOVE, zfs_remove, 3778 VOPNAME_LINK, zfs_link, 3779 VOPNAME_RENAME, zfs_rename, 3780 VOPNAME_MKDIR, zfs_inval, 3781 VOPNAME_RMDIR, zfs_rmdir, 3782 VOPNAME_READDIR, zfs_readdir, 3783 VOPNAME_SYMLINK, zfs_inval, 3784 VOPNAME_FSYNC, zfs_fsync, 3785 VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive, 3786 VOPNAME_FID, zfs_fid, 3787 VOPNAME_SEEK, zfs_seek, 3788 VOPNAME_PATHCONF, zfs_pathconf, 3789 VOPNAME_GETSECATTR, zfs_getsecattr, 3790 VOPNAME_SETSECATTR, zfs_setsecattr, 3791 VOPNAME_VNEVENT, fs_vnevent_support, 3792 NULL, NULL 3793 }; 3794 3795 /* 3796 * Error vnode operations template 3797 */ 3798 vnodeops_t *zfs_evnodeops; 3799 const fs_operation_def_t zfs_evnodeops_template[] = { 3800 VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive, 3801 VOPNAME_PATHCONF, zfs_pathconf, 3802 NULL, NULL 3803 }; 3804