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 2007 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26 /* Portions Copyright 2007 Jeremy Teo */ 27 28 #pragma ident "%Z%%M% %I% %E% SMI" 29 30 #include <sys/types.h> 31 #include <sys/param.h> 32 #include <sys/time.h> 33 #include <sys/systm.h> 34 #include <sys/sysmacros.h> 35 #include <sys/resource.h> 36 #include <sys/vfs.h> 37 #include <sys/vfs_opreg.h> 38 #include <sys/vnode.h> 39 #include <sys/file.h> 40 #include <sys/stat.h> 41 #include <sys/kmem.h> 42 #include <sys/taskq.h> 43 #include <sys/uio.h> 44 #include <sys/vmsystm.h> 45 #include <sys/atomic.h> 46 #include <sys/vm.h> 47 #include <vm/seg_vn.h> 48 #include <vm/pvn.h> 49 #include <vm/as.h> 50 #include <sys/mman.h> 51 #include <sys/pathname.h> 52 #include <sys/cmn_err.h> 53 #include <sys/errno.h> 54 #include <sys/unistd.h> 55 #include <sys/zfs_vfsops.h> 56 #include <sys/zfs_dir.h> 57 #include <sys/zfs_acl.h> 58 #include <sys/zfs_ioctl.h> 59 #include <sys/fs/zfs.h> 60 #include <sys/dmu.h> 61 #include <sys/spa.h> 62 #include <sys/txg.h> 63 #include <sys/dbuf.h> 64 #include <sys/zap.h> 65 #include <sys/dirent.h> 66 #include <sys/policy.h> 67 #include <sys/sunddi.h> 68 #include <sys/filio.h> 69 #include "fs/fs_subr.h" 70 #include <sys/zfs_ctldir.h> 71 #include <sys/dnlc.h> 72 #include <sys/zfs_rlock.h> 73 74 /* 75 * Programming rules. 76 * 77 * Each vnode op performs some logical unit of work. To do this, the ZPL must 78 * properly lock its in-core state, create a DMU transaction, do the work, 79 * record this work in the intent log (ZIL), commit the DMU transaction, 80 * and wait the the intent log to commit if it's is a synchronous operation. 81 * Morover, the vnode ops must work in both normal and log replay context. 82 * The ordering of events is important to avoid deadlocks and references 83 * to freed memory. The example below illustrates the following Big Rules: 84 * 85 * (1) A check must be made in each zfs thread for a mounted file system. 86 * This is done avoiding races using ZFS_ENTER(zfsvfs). 87 * A ZFS_EXIT(zfsvfs) is needed before all returns. 88 * 89 * (2) VN_RELE() should always be the last thing except for zil_commit() 90 * (if necessary) and ZFS_EXIT(). This is for 3 reasons: 91 * First, if it's the last reference, the vnode/znode 92 * can be freed, so the zp may point to freed memory. Second, the last 93 * reference will call zfs_zinactive(), which may induce a lot of work -- 94 * pushing cached pages (which acquires range locks) and syncing out 95 * cached atime changes. Third, zfs_zinactive() may require a new tx, 96 * which could deadlock the system if you were already holding one. 97 * 98 * (3) All range locks must be grabbed before calling dmu_tx_assign(), 99 * as they can span dmu_tx_assign() calls. 100 * 101 * (4) Always pass zfsvfs->z_assign as the second argument to dmu_tx_assign(). 102 * In normal operation, this will be TXG_NOWAIT. During ZIL replay, 103 * it will be a specific txg. Either way, dmu_tx_assign() never blocks. 104 * This is critical because we don't want to block while holding locks. 105 * Note, in particular, that if a lock is sometimes acquired before 106 * the tx assigns, and sometimes after (e.g. z_lock), then failing to 107 * use a non-blocking assign can deadlock the system. The scenario: 108 * 109 * Thread A has grabbed a lock before calling dmu_tx_assign(). 110 * Thread B is in an already-assigned tx, and blocks for this lock. 111 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open() 112 * forever, because the previous txg can't quiesce until B's tx commits. 113 * 114 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT, 115 * then drop all locks, call dmu_tx_wait(), and try again. 116 * 117 * (5) If the operation succeeded, generate the intent log entry for it 118 * before dropping locks. This ensures that the ordering of events 119 * in the intent log matches the order in which they actually occurred. 120 * 121 * (6) At the end of each vnode op, the DMU tx must always commit, 122 * regardless of whether there were any errors. 123 * 124 * (7) After dropping all locks, invoke zil_commit(zilog, seq, foid) 125 * to ensure that synchronous semantics are provided when necessary. 126 * 127 * In general, this is how things should be ordered in each vnode op: 128 * 129 * ZFS_ENTER(zfsvfs); // exit if unmounted 130 * top: 131 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD()) 132 * rw_enter(...); // grab any other locks you need 133 * tx = dmu_tx_create(...); // get DMU tx 134 * dmu_tx_hold_*(); // hold each object you might modify 135 * error = dmu_tx_assign(tx, zfsvfs->z_assign); // try to assign 136 * if (error) { 137 * rw_exit(...); // drop locks 138 * zfs_dirent_unlock(dl); // unlock directory entry 139 * VN_RELE(...); // release held vnodes 140 * if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 141 * dmu_tx_wait(tx); 142 * dmu_tx_abort(tx); 143 * goto top; 144 * } 145 * dmu_tx_abort(tx); // abort DMU tx 146 * ZFS_EXIT(zfsvfs); // finished in zfs 147 * return (error); // really out of space 148 * } 149 * error = do_real_work(); // do whatever this VOP does 150 * if (error == 0) 151 * zfs_log_*(...); // on success, make ZIL entry 152 * dmu_tx_commit(tx); // commit DMU tx -- error or not 153 * rw_exit(...); // drop locks 154 * zfs_dirent_unlock(dl); // unlock directory entry 155 * VN_RELE(...); // release held vnodes 156 * zil_commit(zilog, seq, foid); // synchronous when necessary 157 * ZFS_EXIT(zfsvfs); // finished in zfs 158 * return (error); // done, report error 159 */ 160 /* ARGSUSED */ 161 static int 162 zfs_open(vnode_t **vpp, int flag, cred_t *cr) 163 { 164 znode_t *zp = VTOZ(*vpp); 165 166 /* Keep a count of the synchronous opens in the znode */ 167 if (flag & (FSYNC | FDSYNC)) 168 atomic_inc_32(&zp->z_sync_cnt); 169 return (0); 170 } 171 172 /* ARGSUSED */ 173 static int 174 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr) 175 { 176 znode_t *zp = VTOZ(vp); 177 178 /* Decrement the synchronous opens in the znode */ 179 if (flag & (FSYNC | FDSYNC)) 180 atomic_dec_32(&zp->z_sync_cnt); 181 182 /* 183 * Clean up any locks held by this process on the vp. 184 */ 185 cleanlocks(vp, ddi_get_pid(), 0); 186 cleanshares(vp, ddi_get_pid()); 187 188 return (0); 189 } 190 191 /* 192 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and 193 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter. 194 */ 195 static int 196 zfs_holey(vnode_t *vp, int cmd, offset_t *off) 197 { 198 znode_t *zp = VTOZ(vp); 199 uint64_t noff = (uint64_t)*off; /* new offset */ 200 uint64_t file_sz; 201 int error; 202 boolean_t hole; 203 204 file_sz = zp->z_phys->zp_size; 205 if (noff >= file_sz) { 206 return (ENXIO); 207 } 208 209 if (cmd == _FIO_SEEK_HOLE) 210 hole = B_TRUE; 211 else 212 hole = B_FALSE; 213 214 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff); 215 216 /* end of file? */ 217 if ((error == ESRCH) || (noff > file_sz)) { 218 /* 219 * Handle the virtual hole at the end of file. 220 */ 221 if (hole) { 222 *off = file_sz; 223 return (0); 224 } 225 return (ENXIO); 226 } 227 228 if (noff < *off) 229 return (error); 230 *off = noff; 231 return (error); 232 } 233 234 /* ARGSUSED */ 235 static int 236 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred, 237 int *rvalp) 238 { 239 offset_t off; 240 int error; 241 zfsvfs_t *zfsvfs; 242 243 switch (com) { 244 case _FIOFFS: 245 return (zfs_sync(vp->v_vfsp, 0, cred)); 246 247 /* 248 * The following two ioctls are used by bfu. Faking out, 249 * necessary to avoid bfu errors. 250 */ 251 case _FIOGDIO: 252 case _FIOSDIO: 253 return (0); 254 255 case _FIO_SEEK_DATA: 256 case _FIO_SEEK_HOLE: 257 if (ddi_copyin((void *)data, &off, sizeof (off), flag)) 258 return (EFAULT); 259 260 zfsvfs = VTOZ(vp)->z_zfsvfs; 261 ZFS_ENTER(zfsvfs); 262 263 /* offset parameter is in/out */ 264 error = zfs_holey(vp, com, &off); 265 ZFS_EXIT(zfsvfs); 266 if (error) 267 return (error); 268 if (ddi_copyout(&off, (void *)data, sizeof (off), flag)) 269 return (EFAULT); 270 return (0); 271 } 272 return (ENOTTY); 273 } 274 275 /* 276 * When a file is memory mapped, we must keep the IO data synchronized 277 * between the DMU cache and the memory mapped pages. What this means: 278 * 279 * On Write: If we find a memory mapped page, we write to *both* 280 * the page and the dmu buffer. 281 * 282 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 283 * the file is memory mapped. 284 */ 285 static int 286 mappedwrite(vnode_t *vp, int nbytes, uio_t *uio, dmu_tx_t *tx) 287 { 288 znode_t *zp = VTOZ(vp); 289 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 290 int64_t start, off; 291 int len = nbytes; 292 int error = 0; 293 294 start = uio->uio_loffset; 295 off = start & PAGEOFFSET; 296 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 297 page_t *pp; 298 uint64_t bytes = MIN(PAGESIZE - off, len); 299 uint64_t woff = uio->uio_loffset; 300 301 /* 302 * We don't want a new page to "appear" in the middle of 303 * the file update (because it may not get the write 304 * update data), so we grab a lock to block 305 * zfs_getpage(). 306 */ 307 rw_enter(&zp->z_map_lock, RW_WRITER); 308 if (pp = page_lookup(vp, start, SE_SHARED)) { 309 caddr_t va; 310 311 rw_exit(&zp->z_map_lock); 312 va = ppmapin(pp, PROT_READ | PROT_WRITE, (caddr_t)-1L); 313 error = uiomove(va+off, bytes, UIO_WRITE, uio); 314 if (error == 0) { 315 dmu_write(zfsvfs->z_os, zp->z_id, 316 woff, bytes, va+off, tx); 317 } 318 ppmapout(va); 319 page_unlock(pp); 320 } else { 321 error = dmu_write_uio(zfsvfs->z_os, zp->z_id, 322 uio, bytes, tx); 323 rw_exit(&zp->z_map_lock); 324 } 325 len -= bytes; 326 off = 0; 327 if (error) 328 break; 329 } 330 return (error); 331 } 332 333 /* 334 * When a file is memory mapped, we must keep the IO data synchronized 335 * between the DMU cache and the memory mapped pages. What this means: 336 * 337 * On Read: We "read" preferentially from memory mapped pages, 338 * else we default from the dmu buffer. 339 * 340 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 341 * the file is memory mapped. 342 */ 343 static int 344 mappedread(vnode_t *vp, int nbytes, uio_t *uio) 345 { 346 znode_t *zp = VTOZ(vp); 347 objset_t *os = zp->z_zfsvfs->z_os; 348 int64_t start, off; 349 int len = nbytes; 350 int error = 0; 351 352 start = uio->uio_loffset; 353 off = start & PAGEOFFSET; 354 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 355 page_t *pp; 356 uint64_t bytes = MIN(PAGESIZE - off, len); 357 358 if (pp = page_lookup(vp, start, SE_SHARED)) { 359 caddr_t va; 360 361 va = ppmapin(pp, PROT_READ, (caddr_t)-1L); 362 error = uiomove(va + off, bytes, UIO_READ, uio); 363 ppmapout(va); 364 page_unlock(pp); 365 } else { 366 error = dmu_read_uio(os, zp->z_id, uio, bytes); 367 } 368 len -= bytes; 369 off = 0; 370 if (error) 371 break; 372 } 373 return (error); 374 } 375 376 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ 377 378 /* 379 * Read bytes from specified file into supplied buffer. 380 * 381 * IN: vp - vnode of file to be read from. 382 * uio - structure supplying read location, range info, 383 * and return buffer. 384 * ioflag - SYNC flags; used to provide FRSYNC semantics. 385 * cr - credentials of caller. 386 * 387 * OUT: uio - updated offset and range, buffer filled. 388 * 389 * RETURN: 0 if success 390 * error code if failure 391 * 392 * Side Effects: 393 * vp - atime updated if byte count > 0 394 */ 395 /* ARGSUSED */ 396 static int 397 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 398 { 399 znode_t *zp = VTOZ(vp); 400 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 401 objset_t *os = zfsvfs->z_os; 402 ssize_t n, nbytes; 403 int error; 404 rl_t *rl; 405 406 ZFS_ENTER(zfsvfs); 407 408 /* 409 * Validate file offset 410 */ 411 if (uio->uio_loffset < (offset_t)0) { 412 ZFS_EXIT(zfsvfs); 413 return (EINVAL); 414 } 415 416 /* 417 * Fasttrack empty reads 418 */ 419 if (uio->uio_resid == 0) { 420 ZFS_EXIT(zfsvfs); 421 return (0); 422 } 423 424 /* 425 * Check for mandatory locks 426 */ 427 if (MANDMODE((mode_t)zp->z_phys->zp_mode)) { 428 if (error = chklock(vp, FREAD, 429 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { 430 ZFS_EXIT(zfsvfs); 431 return (error); 432 } 433 } 434 435 /* 436 * If we're in FRSYNC mode, sync out this znode before reading it. 437 */ 438 if (ioflag & FRSYNC) 439 zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id); 440 441 /* 442 * Lock the range against changes. 443 */ 444 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER); 445 446 /* 447 * If we are reading past end-of-file we can skip 448 * to the end; but we might still need to set atime. 449 */ 450 if (uio->uio_loffset >= zp->z_phys->zp_size) { 451 error = 0; 452 goto out; 453 } 454 455 ASSERT(uio->uio_loffset < zp->z_phys->zp_size); 456 n = MIN(uio->uio_resid, zp->z_phys->zp_size - uio->uio_loffset); 457 458 while (n > 0) { 459 nbytes = MIN(n, zfs_read_chunk_size - 460 P2PHASE(uio->uio_loffset, zfs_read_chunk_size)); 461 462 if (vn_has_cached_data(vp)) 463 error = mappedread(vp, nbytes, uio); 464 else 465 error = dmu_read_uio(os, zp->z_id, uio, nbytes); 466 if (error) 467 break; 468 469 n -= nbytes; 470 } 471 472 out: 473 zfs_range_unlock(rl); 474 475 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 476 ZFS_EXIT(zfsvfs); 477 return (error); 478 } 479 480 /* 481 * Fault in the pages of the first n bytes specified by the uio structure. 482 * 1 byte in each page is touched and the uio struct is unmodified. 483 * Any error will exit this routine as this is only a best 484 * attempt to get the pages resident. This is a copy of ufs_trans_touch(). 485 */ 486 static void 487 zfs_prefault_write(ssize_t n, struct uio *uio) 488 { 489 struct iovec *iov; 490 ulong_t cnt, incr; 491 caddr_t p; 492 uint8_t tmp; 493 494 iov = uio->uio_iov; 495 496 while (n) { 497 cnt = MIN(iov->iov_len, n); 498 if (cnt == 0) { 499 /* empty iov entry */ 500 iov++; 501 continue; 502 } 503 n -= cnt; 504 /* 505 * touch each page in this segment. 506 */ 507 p = iov->iov_base; 508 while (cnt) { 509 switch (uio->uio_segflg) { 510 case UIO_USERSPACE: 511 case UIO_USERISPACE: 512 if (fuword8(p, &tmp)) 513 return; 514 break; 515 case UIO_SYSSPACE: 516 if (kcopy(p, &tmp, 1)) 517 return; 518 break; 519 } 520 incr = MIN(cnt, PAGESIZE); 521 p += incr; 522 cnt -= incr; 523 } 524 /* 525 * touch the last byte in case it straddles a page. 526 */ 527 p--; 528 switch (uio->uio_segflg) { 529 case UIO_USERSPACE: 530 case UIO_USERISPACE: 531 if (fuword8(p, &tmp)) 532 return; 533 break; 534 case UIO_SYSSPACE: 535 if (kcopy(p, &tmp, 1)) 536 return; 537 break; 538 } 539 iov++; 540 } 541 } 542 543 /* 544 * Write the bytes to a file. 545 * 546 * IN: vp - vnode of file to be written to. 547 * uio - structure supplying write location, range info, 548 * and data buffer. 549 * ioflag - FAPPEND flag set if in append mode. 550 * cr - credentials of caller. 551 * 552 * OUT: uio - updated offset and range. 553 * 554 * RETURN: 0 if success 555 * error code if failure 556 * 557 * Timestamps: 558 * vp - ctime|mtime updated if byte count > 0 559 */ 560 /* ARGSUSED */ 561 static int 562 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 563 { 564 znode_t *zp = VTOZ(vp); 565 rlim64_t limit = uio->uio_llimit; 566 ssize_t start_resid = uio->uio_resid; 567 ssize_t tx_bytes; 568 uint64_t end_size; 569 dmu_tx_t *tx; 570 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 571 zilog_t *zilog = zfsvfs->z_log; 572 offset_t woff; 573 ssize_t n, nbytes; 574 rl_t *rl; 575 int max_blksz = zfsvfs->z_max_blksz; 576 int error; 577 578 /* 579 * Fasttrack empty write 580 */ 581 n = start_resid; 582 if (n == 0) 583 return (0); 584 585 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 586 limit = MAXOFFSET_T; 587 588 ZFS_ENTER(zfsvfs); 589 590 /* 591 * Pre-fault the pages to ensure slow (eg NFS) pages 592 * don't hold up txg. 593 */ 594 zfs_prefault_write(n, uio); 595 596 /* 597 * If in append mode, set the io offset pointer to eof. 598 */ 599 if (ioflag & FAPPEND) { 600 /* 601 * Range lock for a file append: 602 * The value for the start of range will be determined by 603 * zfs_range_lock() (to guarantee append semantics). 604 * If this write will cause the block size to increase, 605 * zfs_range_lock() will lock the entire file, so we must 606 * later reduce the range after we grow the block size. 607 */ 608 rl = zfs_range_lock(zp, 0, n, RL_APPEND); 609 if (rl->r_len == UINT64_MAX) { 610 /* overlocked, zp_size can't change */ 611 woff = uio->uio_loffset = zp->z_phys->zp_size; 612 } else { 613 woff = uio->uio_loffset = rl->r_off; 614 } 615 } else { 616 woff = uio->uio_loffset; 617 /* 618 * Validate file offset 619 */ 620 if (woff < 0) { 621 ZFS_EXIT(zfsvfs); 622 return (EINVAL); 623 } 624 625 /* 626 * If we need to grow the block size then zfs_range_lock() 627 * will lock a wider range than we request here. 628 * Later after growing the block size we reduce the range. 629 */ 630 rl = zfs_range_lock(zp, woff, n, RL_WRITER); 631 } 632 633 if (woff >= limit) { 634 zfs_range_unlock(rl); 635 ZFS_EXIT(zfsvfs); 636 return (EFBIG); 637 } 638 639 if ((woff + n) > limit || woff > (limit - n)) 640 n = limit - woff; 641 642 /* 643 * Check for mandatory locks 644 */ 645 if (MANDMODE((mode_t)zp->z_phys->zp_mode) && 646 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) { 647 zfs_range_unlock(rl); 648 ZFS_EXIT(zfsvfs); 649 return (error); 650 } 651 end_size = MAX(zp->z_phys->zp_size, woff + n); 652 653 /* 654 * Write the file in reasonable size chunks. Each chunk is written 655 * in a separate transaction; this keeps the intent log records small 656 * and allows us to do more fine-grained space accounting. 657 */ 658 while (n > 0) { 659 /* 660 * Start a transaction. 661 */ 662 woff = uio->uio_loffset; 663 tx = dmu_tx_create(zfsvfs->z_os); 664 dmu_tx_hold_bonus(tx, zp->z_id); 665 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 666 error = dmu_tx_assign(tx, zfsvfs->z_assign); 667 if (error) { 668 if (error == ERESTART && 669 zfsvfs->z_assign == TXG_NOWAIT) { 670 dmu_tx_wait(tx); 671 dmu_tx_abort(tx); 672 continue; 673 } 674 dmu_tx_abort(tx); 675 break; 676 } 677 678 /* 679 * If zfs_range_lock() over-locked we grow the blocksize 680 * and then reduce the lock range. This will only happen 681 * on the first iteration since zfs_range_reduce() will 682 * shrink down r_len to the appropriate size. 683 */ 684 if (rl->r_len == UINT64_MAX) { 685 uint64_t new_blksz; 686 687 if (zp->z_blksz > max_blksz) { 688 ASSERT(!ISP2(zp->z_blksz)); 689 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE); 690 } else { 691 new_blksz = MIN(end_size, max_blksz); 692 } 693 zfs_grow_blocksize(zp, new_blksz, tx); 694 zfs_range_reduce(rl, woff, n); 695 } 696 697 /* 698 * XXX - should we really limit each write to z_max_blksz? 699 * Perhaps we should use SPA_MAXBLOCKSIZE chunks? 700 */ 701 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz)); 702 rw_enter(&zp->z_map_lock, RW_READER); 703 704 tx_bytes = uio->uio_resid; 705 if (vn_has_cached_data(vp)) { 706 rw_exit(&zp->z_map_lock); 707 error = mappedwrite(vp, nbytes, uio, tx); 708 } else { 709 error = dmu_write_uio(zfsvfs->z_os, zp->z_id, 710 uio, nbytes, tx); 711 rw_exit(&zp->z_map_lock); 712 } 713 tx_bytes -= uio->uio_resid; 714 715 /* 716 * If we made no progress, we're done. If we made even 717 * partial progress, update the znode and ZIL accordingly. 718 */ 719 if (tx_bytes == 0) { 720 dmu_tx_commit(tx); 721 ASSERT(error != 0); 722 break; 723 } 724 725 /* 726 * Clear Set-UID/Set-GID bits on successful write if not 727 * privileged and at least one of the excute bits is set. 728 * 729 * It would be nice to to this after all writes have 730 * been done, but that would still expose the ISUID/ISGID 731 * to another app after the partial write is committed. 732 */ 733 mutex_enter(&zp->z_acl_lock); 734 if ((zp->z_phys->zp_mode & (S_IXUSR | (S_IXUSR >> 3) | 735 (S_IXUSR >> 6))) != 0 && 736 (zp->z_phys->zp_mode & (S_ISUID | S_ISGID)) != 0 && 737 secpolicy_vnode_setid_retain(cr, 738 (zp->z_phys->zp_mode & S_ISUID) != 0 && 739 zp->z_phys->zp_uid == 0) != 0) { 740 zp->z_phys->zp_mode &= ~(S_ISUID | S_ISGID); 741 } 742 mutex_exit(&zp->z_acl_lock); 743 744 /* 745 * Update time stamp. NOTE: This marks the bonus buffer as 746 * dirty, so we don't have to do it again for zp_size. 747 */ 748 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 749 750 /* 751 * Update the file size (zp_size) if it has changed; 752 * account for possible concurrent updates. 753 */ 754 while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset) 755 (void) atomic_cas_64(&zp->z_phys->zp_size, end_size, 756 uio->uio_loffset); 757 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag); 758 dmu_tx_commit(tx); 759 760 if (error != 0) 761 break; 762 ASSERT(tx_bytes == nbytes); 763 n -= nbytes; 764 } 765 766 zfs_range_unlock(rl); 767 768 /* 769 * If we're in replay mode, or we made no progress, return error. 770 * Otherwise, it's at least a partial write, so it's successful. 771 */ 772 if (zfsvfs->z_assign >= TXG_INITIAL || uio->uio_resid == start_resid) { 773 ZFS_EXIT(zfsvfs); 774 return (error); 775 } 776 777 if (ioflag & (FSYNC | FDSYNC)) 778 zil_commit(zilog, zp->z_last_itx, zp->z_id); 779 780 ZFS_EXIT(zfsvfs); 781 return (0); 782 } 783 784 void 785 zfs_get_done(dmu_buf_t *db, void *vzgd) 786 { 787 zgd_t *zgd = (zgd_t *)vzgd; 788 rl_t *rl = zgd->zgd_rl; 789 vnode_t *vp = ZTOV(rl->r_zp); 790 791 dmu_buf_rele(db, vzgd); 792 zfs_range_unlock(rl); 793 VN_RELE(vp); 794 zil_add_vdev(zgd->zgd_zilog, DVA_GET_VDEV(BP_IDENTITY(zgd->zgd_bp))); 795 kmem_free(zgd, sizeof (zgd_t)); 796 } 797 798 /* 799 * Get data to generate a TX_WRITE intent log record. 800 */ 801 int 802 zfs_get_data(void *arg, lr_write_t *lr, char *buf, zio_t *zio) 803 { 804 zfsvfs_t *zfsvfs = arg; 805 objset_t *os = zfsvfs->z_os; 806 znode_t *zp; 807 uint64_t off = lr->lr_offset; 808 dmu_buf_t *db; 809 rl_t *rl; 810 zgd_t *zgd; 811 int dlen = lr->lr_length; /* length of user data */ 812 int error = 0; 813 814 ASSERT(zio); 815 ASSERT(dlen != 0); 816 817 /* 818 * Nothing to do if the file has been removed 819 */ 820 if (zfs_zget(zfsvfs, lr->lr_foid, &zp) != 0) 821 return (ENOENT); 822 if (zp->z_unlinked) { 823 VN_RELE(ZTOV(zp)); 824 return (ENOENT); 825 } 826 827 /* 828 * Write records come in two flavors: immediate and indirect. 829 * For small writes it's cheaper to store the data with the 830 * log record (immediate); for large writes it's cheaper to 831 * sync the data and get a pointer to it (indirect) so that 832 * we don't have to write the data twice. 833 */ 834 if (buf != NULL) { /* immediate write */ 835 rl = zfs_range_lock(zp, off, dlen, RL_READER); 836 /* test for truncation needs to be done while range locked */ 837 if (off >= zp->z_phys->zp_size) { 838 error = ENOENT; 839 goto out; 840 } 841 VERIFY(0 == dmu_read(os, lr->lr_foid, off, dlen, buf)); 842 } else { /* indirect write */ 843 uint64_t boff; /* block starting offset */ 844 845 /* 846 * Have to lock the whole block to ensure when it's 847 * written out and it's checksum is being calculated 848 * that no one can change the data. We need to re-check 849 * blocksize after we get the lock in case it's changed! 850 */ 851 for (;;) { 852 if (ISP2(zp->z_blksz)) { 853 boff = P2ALIGN_TYPED(off, zp->z_blksz, 854 uint64_t); 855 } else { 856 boff = 0; 857 } 858 dlen = zp->z_blksz; 859 rl = zfs_range_lock(zp, boff, dlen, RL_READER); 860 if (zp->z_blksz == dlen) 861 break; 862 zfs_range_unlock(rl); 863 } 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 zgd = (zgd_t *)kmem_alloc(sizeof (zgd_t), KM_SLEEP); 870 zgd->zgd_rl = rl; 871 zgd->zgd_zilog = zfsvfs->z_log; 872 zgd->zgd_bp = &lr->lr_blkptr; 873 VERIFY(0 == dmu_buf_hold(os, lr->lr_foid, boff, zgd, &db)); 874 ASSERT(boff == db->db_offset); 875 lr->lr_blkoff = off - boff; 876 error = dmu_sync(zio, db, &lr->lr_blkptr, 877 lr->lr_common.lrc_txg, zfs_get_done, zgd); 878 ASSERT(error == EEXIST || lr->lr_length <= zp->z_blksz); 879 if (error == 0) { 880 zil_add_vdev(zfsvfs->z_log, 881 DVA_GET_VDEV(BP_IDENTITY(&lr->lr_blkptr))); 882 } 883 /* 884 * If we get EINPROGRESS, then we need to wait for a 885 * write IO initiated by dmu_sync() to complete before 886 * we can release this dbuf. We will finish everything 887 * up in the zfs_get_done() callback. 888 */ 889 if (error == EINPROGRESS) 890 return (0); 891 dmu_buf_rele(db, zgd); 892 kmem_free(zgd, sizeof (zgd_t)); 893 } 894 out: 895 zfs_range_unlock(rl); 896 VN_RELE(ZTOV(zp)); 897 return (error); 898 } 899 900 /*ARGSUSED*/ 901 static int 902 zfs_access(vnode_t *vp, int mode, int flags, cred_t *cr) 903 { 904 znode_t *zp = VTOZ(vp); 905 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 906 int error; 907 908 ZFS_ENTER(zfsvfs); 909 error = zfs_zaccess_rwx(zp, mode, cr); 910 ZFS_EXIT(zfsvfs); 911 return (error); 912 } 913 914 /* 915 * Lookup an entry in a directory, or an extended attribute directory. 916 * If it exists, return a held vnode reference for it. 917 * 918 * IN: dvp - vnode of directory to search. 919 * nm - name of entry to lookup. 920 * pnp - full pathname to lookup [UNUSED]. 921 * flags - LOOKUP_XATTR set if looking for an attribute. 922 * rdir - root directory vnode [UNUSED]. 923 * cr - credentials of caller. 924 * 925 * OUT: vpp - vnode of located entry, NULL if not found. 926 * 927 * RETURN: 0 if success 928 * error code if failure 929 * 930 * Timestamps: 931 * NA 932 */ 933 /* ARGSUSED */ 934 static int 935 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp, 936 int flags, vnode_t *rdir, cred_t *cr) 937 { 938 939 znode_t *zdp = VTOZ(dvp); 940 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 941 int error; 942 943 ZFS_ENTER(zfsvfs); 944 945 *vpp = NULL; 946 947 if (flags & LOOKUP_XATTR) { 948 /* 949 * If the xattr property is off, refuse the lookup request. 950 */ 951 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) { 952 ZFS_EXIT(zfsvfs); 953 return (EINVAL); 954 } 955 956 /* 957 * We don't allow recursive attributes.. 958 * Maybe someday we will. 959 */ 960 if (zdp->z_phys->zp_flags & ZFS_XATTR) { 961 ZFS_EXIT(zfsvfs); 962 return (EINVAL); 963 } 964 965 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) { 966 ZFS_EXIT(zfsvfs); 967 return (error); 968 } 969 970 /* 971 * Do we have permission to get into attribute directory? 972 */ 973 974 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, cr)) { 975 VN_RELE(*vpp); 976 } 977 978 ZFS_EXIT(zfsvfs); 979 return (error); 980 } 981 982 if (dvp->v_type != VDIR) { 983 ZFS_EXIT(zfsvfs); 984 return (ENOTDIR); 985 } 986 987 /* 988 * Check accessibility of directory. 989 */ 990 991 if (error = zfs_zaccess(zdp, ACE_EXECUTE, cr)) { 992 ZFS_EXIT(zfsvfs); 993 return (error); 994 } 995 996 if ((error = zfs_dirlook(zdp, nm, vpp)) == 0) { 997 998 /* 999 * Convert device special files 1000 */ 1001 if (IS_DEVVP(*vpp)) { 1002 vnode_t *svp; 1003 1004 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1005 VN_RELE(*vpp); 1006 if (svp == NULL) 1007 error = ENOSYS; 1008 else 1009 *vpp = svp; 1010 } 1011 } 1012 1013 ZFS_EXIT(zfsvfs); 1014 return (error); 1015 } 1016 1017 /* 1018 * Attempt to create a new entry in a directory. If the entry 1019 * already exists, truncate the file if permissible, else return 1020 * an error. Return the vp of the created or trunc'd file. 1021 * 1022 * IN: dvp - vnode of directory to put new file entry in. 1023 * name - name of new file entry. 1024 * vap - attributes of new file. 1025 * excl - flag indicating exclusive or non-exclusive mode. 1026 * mode - mode to open file with. 1027 * cr - credentials of caller. 1028 * flag - large file flag [UNUSED]. 1029 * 1030 * OUT: vpp - vnode of created or trunc'd entry. 1031 * 1032 * RETURN: 0 if success 1033 * error code if failure 1034 * 1035 * Timestamps: 1036 * dvp - ctime|mtime updated if new entry created 1037 * vp - ctime|mtime always, atime if new 1038 */ 1039 /* ARGSUSED */ 1040 static int 1041 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl, 1042 int mode, vnode_t **vpp, cred_t *cr, int flag) 1043 { 1044 znode_t *zp, *dzp = VTOZ(dvp); 1045 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1046 zilog_t *zilog = zfsvfs->z_log; 1047 objset_t *os = zfsvfs->z_os; 1048 zfs_dirlock_t *dl; 1049 dmu_tx_t *tx; 1050 int error; 1051 uint64_t zoid; 1052 1053 ZFS_ENTER(zfsvfs); 1054 1055 top: 1056 *vpp = NULL; 1057 1058 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr)) 1059 vap->va_mode &= ~VSVTX; 1060 1061 if (*name == '\0') { 1062 /* 1063 * Null component name refers to the directory itself. 1064 */ 1065 VN_HOLD(dvp); 1066 zp = dzp; 1067 dl = NULL; 1068 error = 0; 1069 } else { 1070 /* possible VN_HOLD(zp) */ 1071 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, 0)) { 1072 if (strcmp(name, "..") == 0) 1073 error = EISDIR; 1074 ZFS_EXIT(zfsvfs); 1075 return (error); 1076 } 1077 } 1078 1079 zoid = zp ? zp->z_id : -1ULL; 1080 1081 if (zp == NULL) { 1082 /* 1083 * Create a new file object and update the directory 1084 * to reference it. 1085 */ 1086 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 1087 goto out; 1088 } 1089 1090 /* 1091 * We only support the creation of regular files in 1092 * extended attribute directories. 1093 */ 1094 if ((dzp->z_phys->zp_flags & ZFS_XATTR) && 1095 (vap->va_type != VREG)) { 1096 error = EINVAL; 1097 goto out; 1098 } 1099 1100 tx = dmu_tx_create(os); 1101 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1102 dmu_tx_hold_bonus(tx, dzp->z_id); 1103 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1104 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 1105 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1106 0, SPA_MAXBLOCKSIZE); 1107 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1108 if (error) { 1109 zfs_dirent_unlock(dl); 1110 if (error == ERESTART && 1111 zfsvfs->z_assign == TXG_NOWAIT) { 1112 dmu_tx_wait(tx); 1113 dmu_tx_abort(tx); 1114 goto top; 1115 } 1116 dmu_tx_abort(tx); 1117 ZFS_EXIT(zfsvfs); 1118 return (error); 1119 } 1120 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 1121 ASSERT(zp->z_id == zoid); 1122 (void) zfs_link_create(dl, zp, tx, ZNEW); 1123 zfs_log_create(zilog, tx, TX_CREATE, dzp, zp, name); 1124 dmu_tx_commit(tx); 1125 } else { 1126 /* 1127 * A directory entry already exists for this name. 1128 */ 1129 /* 1130 * Can't truncate an existing file if in exclusive mode. 1131 */ 1132 if (excl == EXCL) { 1133 error = EEXIST; 1134 goto out; 1135 } 1136 /* 1137 * Can't open a directory for writing. 1138 */ 1139 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1140 error = EISDIR; 1141 goto out; 1142 } 1143 /* 1144 * Verify requested access to file. 1145 */ 1146 if (mode && (error = zfs_zaccess_rwx(zp, mode, cr))) { 1147 goto out; 1148 } 1149 1150 mutex_enter(&dzp->z_lock); 1151 dzp->z_seq++; 1152 mutex_exit(&dzp->z_lock); 1153 1154 /* 1155 * Truncate regular files if requested. 1156 */ 1157 if ((ZTOV(zp)->v_type == VREG) && 1158 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1159 error = zfs_freesp(zp, 0, 0, mode, TRUE); 1160 if (error == ERESTART && 1161 zfsvfs->z_assign == TXG_NOWAIT) { 1162 /* NB: we already did dmu_tx_wait() */ 1163 zfs_dirent_unlock(dl); 1164 VN_RELE(ZTOV(zp)); 1165 goto top; 1166 } 1167 } 1168 } 1169 out: 1170 1171 if (dl) 1172 zfs_dirent_unlock(dl); 1173 1174 if (error) { 1175 if (zp) 1176 VN_RELE(ZTOV(zp)); 1177 } else { 1178 *vpp = ZTOV(zp); 1179 /* 1180 * If vnode is for a device return a specfs vnode instead. 1181 */ 1182 if (IS_DEVVP(*vpp)) { 1183 struct vnode *svp; 1184 1185 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1186 VN_RELE(*vpp); 1187 if (svp == NULL) { 1188 error = ENOSYS; 1189 } 1190 *vpp = svp; 1191 } 1192 } 1193 1194 ZFS_EXIT(zfsvfs); 1195 return (error); 1196 } 1197 1198 /* 1199 * Remove an entry from a directory. 1200 * 1201 * IN: dvp - vnode of directory to remove entry from. 1202 * name - name of entry to remove. 1203 * cr - credentials of caller. 1204 * 1205 * RETURN: 0 if success 1206 * error code if failure 1207 * 1208 * Timestamps: 1209 * dvp - ctime|mtime 1210 * vp - ctime (if nlink > 0) 1211 */ 1212 static int 1213 zfs_remove(vnode_t *dvp, char *name, cred_t *cr) 1214 { 1215 znode_t *zp, *dzp = VTOZ(dvp); 1216 znode_t *xzp = NULL; 1217 vnode_t *vp; 1218 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1219 zilog_t *zilog = zfsvfs->z_log; 1220 uint64_t acl_obj, xattr_obj; 1221 zfs_dirlock_t *dl; 1222 dmu_tx_t *tx; 1223 boolean_t may_delete_now, delete_now = FALSE; 1224 boolean_t unlinked; 1225 int error; 1226 1227 ZFS_ENTER(zfsvfs); 1228 1229 top: 1230 /* 1231 * Attempt to lock directory; fail if entry doesn't exist. 1232 */ 1233 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) { 1234 ZFS_EXIT(zfsvfs); 1235 return (error); 1236 } 1237 1238 vp = ZTOV(zp); 1239 1240 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1241 goto out; 1242 } 1243 1244 /* 1245 * Need to use rmdir for removing directories. 1246 */ 1247 if (vp->v_type == VDIR) { 1248 error = EPERM; 1249 goto out; 1250 } 1251 1252 vnevent_remove(vp); 1253 1254 dnlc_remove(dvp, name); 1255 1256 mutex_enter(&vp->v_lock); 1257 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp); 1258 mutex_exit(&vp->v_lock); 1259 1260 /* 1261 * We may delete the znode now, or we may put it in the unlinked set; 1262 * it depends on whether we're the last link, and on whether there are 1263 * other holds on the vnode. So we dmu_tx_hold() the right things to 1264 * allow for either case. 1265 */ 1266 tx = dmu_tx_create(zfsvfs->z_os); 1267 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1268 dmu_tx_hold_bonus(tx, zp->z_id); 1269 if (may_delete_now) 1270 dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END); 1271 1272 /* are there any extended attributes? */ 1273 if ((xattr_obj = zp->z_phys->zp_xattr) != 0) { 1274 /* XXX - do we need this if we are deleting? */ 1275 dmu_tx_hold_bonus(tx, xattr_obj); 1276 } 1277 1278 /* are there any additional acls */ 1279 if ((acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj) != 0 && 1280 may_delete_now) 1281 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 1282 1283 /* charge as an update -- would be nice not to charge at all */ 1284 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1285 1286 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1287 if (error) { 1288 zfs_dirent_unlock(dl); 1289 VN_RELE(vp); 1290 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1291 dmu_tx_wait(tx); 1292 dmu_tx_abort(tx); 1293 goto top; 1294 } 1295 dmu_tx_abort(tx); 1296 ZFS_EXIT(zfsvfs); 1297 return (error); 1298 } 1299 1300 /* 1301 * Remove the directory entry. 1302 */ 1303 error = zfs_link_destroy(dl, zp, tx, 0, &unlinked); 1304 1305 if (error) { 1306 dmu_tx_commit(tx); 1307 goto out; 1308 } 1309 1310 if (unlinked) { 1311 mutex_enter(&vp->v_lock); 1312 delete_now = may_delete_now && 1313 vp->v_count == 1 && !vn_has_cached_data(vp) && 1314 zp->z_phys->zp_xattr == xattr_obj && 1315 zp->z_phys->zp_acl.z_acl_extern_obj == acl_obj; 1316 mutex_exit(&vp->v_lock); 1317 } 1318 1319 if (delete_now) { 1320 if (zp->z_phys->zp_xattr) { 1321 error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp); 1322 ASSERT3U(error, ==, 0); 1323 ASSERT3U(xzp->z_phys->zp_links, ==, 2); 1324 dmu_buf_will_dirty(xzp->z_dbuf, tx); 1325 mutex_enter(&xzp->z_lock); 1326 xzp->z_unlinked = 1; 1327 xzp->z_phys->zp_links = 0; 1328 mutex_exit(&xzp->z_lock); 1329 zfs_unlinked_add(xzp, tx); 1330 zp->z_phys->zp_xattr = 0; /* probably unnecessary */ 1331 } 1332 mutex_enter(&zp->z_lock); 1333 mutex_enter(&vp->v_lock); 1334 vp->v_count--; 1335 ASSERT3U(vp->v_count, ==, 0); 1336 mutex_exit(&vp->v_lock); 1337 mutex_exit(&zp->z_lock); 1338 zfs_znode_delete(zp, tx); 1339 VFS_RELE(zfsvfs->z_vfs); 1340 } else if (unlinked) { 1341 zfs_unlinked_add(zp, tx); 1342 } 1343 1344 zfs_log_remove(zilog, tx, TX_REMOVE, dzp, name); 1345 1346 dmu_tx_commit(tx); 1347 out: 1348 zfs_dirent_unlock(dl); 1349 1350 if (!delete_now) { 1351 VN_RELE(vp); 1352 } else if (xzp) { 1353 /* this rele delayed to prevent nesting transactions */ 1354 VN_RELE(ZTOV(xzp)); 1355 } 1356 1357 ZFS_EXIT(zfsvfs); 1358 return (error); 1359 } 1360 1361 /* 1362 * Create a new directory and insert it into dvp using the name 1363 * provided. Return a pointer to the inserted directory. 1364 * 1365 * IN: dvp - vnode of directory to add subdir to. 1366 * dirname - name of new directory. 1367 * vap - attributes of new directory. 1368 * cr - credentials of caller. 1369 * 1370 * OUT: vpp - vnode of created directory. 1371 * 1372 * RETURN: 0 if success 1373 * error code if failure 1374 * 1375 * Timestamps: 1376 * dvp - ctime|mtime updated 1377 * vp - ctime|mtime|atime updated 1378 */ 1379 static int 1380 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr) 1381 { 1382 znode_t *zp, *dzp = VTOZ(dvp); 1383 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1384 zilog_t *zilog = zfsvfs->z_log; 1385 zfs_dirlock_t *dl; 1386 uint64_t zoid = 0; 1387 dmu_tx_t *tx; 1388 int error; 1389 1390 ASSERT(vap->va_type == VDIR); 1391 1392 ZFS_ENTER(zfsvfs); 1393 1394 if (dzp->z_phys->zp_flags & ZFS_XATTR) { 1395 ZFS_EXIT(zfsvfs); 1396 return (EINVAL); 1397 } 1398 top: 1399 *vpp = NULL; 1400 1401 /* 1402 * First make sure the new directory doesn't exist. 1403 */ 1404 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, ZNEW)) { 1405 ZFS_EXIT(zfsvfs); 1406 return (error); 1407 } 1408 1409 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, cr)) { 1410 zfs_dirent_unlock(dl); 1411 ZFS_EXIT(zfsvfs); 1412 return (error); 1413 } 1414 1415 /* 1416 * Add a new entry to the directory. 1417 */ 1418 tx = dmu_tx_create(zfsvfs->z_os); 1419 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 1420 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 1421 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 1422 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1423 0, SPA_MAXBLOCKSIZE); 1424 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1425 if (error) { 1426 zfs_dirent_unlock(dl); 1427 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1428 dmu_tx_wait(tx); 1429 dmu_tx_abort(tx); 1430 goto top; 1431 } 1432 dmu_tx_abort(tx); 1433 ZFS_EXIT(zfsvfs); 1434 return (error); 1435 } 1436 1437 /* 1438 * Create new node. 1439 */ 1440 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 1441 1442 /* 1443 * Now put new name in parent dir. 1444 */ 1445 (void) zfs_link_create(dl, zp, tx, ZNEW); 1446 1447 *vpp = ZTOV(zp); 1448 1449 zfs_log_create(zilog, tx, TX_MKDIR, dzp, zp, dirname); 1450 dmu_tx_commit(tx); 1451 1452 zfs_dirent_unlock(dl); 1453 1454 ZFS_EXIT(zfsvfs); 1455 return (0); 1456 } 1457 1458 /* 1459 * Remove a directory subdir entry. If the current working 1460 * directory is the same as the subdir to be removed, the 1461 * remove will fail. 1462 * 1463 * IN: dvp - vnode of directory to remove from. 1464 * name - name of directory to be removed. 1465 * cwd - vnode of current working directory. 1466 * cr - credentials of caller. 1467 * 1468 * RETURN: 0 if success 1469 * error code if failure 1470 * 1471 * Timestamps: 1472 * dvp - ctime|mtime updated 1473 */ 1474 static int 1475 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr) 1476 { 1477 znode_t *dzp = VTOZ(dvp); 1478 znode_t *zp; 1479 vnode_t *vp; 1480 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1481 zilog_t *zilog = zfsvfs->z_log; 1482 zfs_dirlock_t *dl; 1483 dmu_tx_t *tx; 1484 int error; 1485 1486 ZFS_ENTER(zfsvfs); 1487 1488 top: 1489 zp = NULL; 1490 1491 /* 1492 * Attempt to lock directory; fail if entry doesn't exist. 1493 */ 1494 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) { 1495 ZFS_EXIT(zfsvfs); 1496 return (error); 1497 } 1498 1499 vp = ZTOV(zp); 1500 1501 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1502 goto out; 1503 } 1504 1505 if (vp->v_type != VDIR) { 1506 error = ENOTDIR; 1507 goto out; 1508 } 1509 1510 if (vp == cwd) { 1511 error = EINVAL; 1512 goto out; 1513 } 1514 1515 vnevent_rmdir(vp); 1516 1517 /* 1518 * Grab a lock on the directory to make sure that noone is 1519 * trying to add (or lookup) entries while we are removing it. 1520 */ 1521 rw_enter(&zp->z_name_lock, RW_WRITER); 1522 1523 /* 1524 * Grab a lock on the parent pointer to make sure we play well 1525 * with the treewalk and directory rename code. 1526 */ 1527 rw_enter(&zp->z_parent_lock, RW_WRITER); 1528 1529 tx = dmu_tx_create(zfsvfs->z_os); 1530 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1531 dmu_tx_hold_bonus(tx, zp->z_id); 1532 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1533 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1534 if (error) { 1535 rw_exit(&zp->z_parent_lock); 1536 rw_exit(&zp->z_name_lock); 1537 zfs_dirent_unlock(dl); 1538 VN_RELE(vp); 1539 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1540 dmu_tx_wait(tx); 1541 dmu_tx_abort(tx); 1542 goto top; 1543 } 1544 dmu_tx_abort(tx); 1545 ZFS_EXIT(zfsvfs); 1546 return (error); 1547 } 1548 1549 error = zfs_link_destroy(dl, zp, tx, 0, NULL); 1550 1551 if (error == 0) 1552 zfs_log_remove(zilog, tx, TX_RMDIR, dzp, name); 1553 1554 dmu_tx_commit(tx); 1555 1556 rw_exit(&zp->z_parent_lock); 1557 rw_exit(&zp->z_name_lock); 1558 out: 1559 zfs_dirent_unlock(dl); 1560 1561 VN_RELE(vp); 1562 1563 ZFS_EXIT(zfsvfs); 1564 return (error); 1565 } 1566 1567 /* 1568 * Read as many directory entries as will fit into the provided 1569 * buffer from the given directory cursor position (specified in 1570 * the uio structure. 1571 * 1572 * IN: vp - vnode of directory to read. 1573 * uio - structure supplying read location, range info, 1574 * and return buffer. 1575 * cr - credentials of caller. 1576 * 1577 * OUT: uio - updated offset and range, buffer filled. 1578 * eofp - set to true if end-of-file detected. 1579 * 1580 * RETURN: 0 if success 1581 * error code if failure 1582 * 1583 * Timestamps: 1584 * vp - atime updated 1585 * 1586 * Note that the low 4 bits of the cookie returned by zap is always zero. 1587 * This allows us to use the low range for "special" directory entries: 1588 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 1589 * we use the offset 2 for the '.zfs' directory. 1590 */ 1591 /* ARGSUSED */ 1592 static int 1593 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp) 1594 { 1595 znode_t *zp = VTOZ(vp); 1596 iovec_t *iovp; 1597 dirent64_t *odp; 1598 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1599 objset_t *os; 1600 caddr_t outbuf; 1601 size_t bufsize; 1602 zap_cursor_t zc; 1603 zap_attribute_t zap; 1604 uint_t bytes_wanted; 1605 uint64_t offset; /* must be unsigned; checks for < 1 */ 1606 int local_eof; 1607 int outcount; 1608 int error; 1609 uint8_t prefetch; 1610 1611 ZFS_ENTER(zfsvfs); 1612 1613 /* 1614 * If we are not given an eof variable, 1615 * use a local one. 1616 */ 1617 if (eofp == NULL) 1618 eofp = &local_eof; 1619 1620 /* 1621 * Check for valid iov_len. 1622 */ 1623 if (uio->uio_iov->iov_len <= 0) { 1624 ZFS_EXIT(zfsvfs); 1625 return (EINVAL); 1626 } 1627 1628 /* 1629 * Quit if directory has been removed (posix) 1630 */ 1631 if ((*eofp = zp->z_unlinked) != 0) { 1632 ZFS_EXIT(zfsvfs); 1633 return (0); 1634 } 1635 1636 error = 0; 1637 os = zfsvfs->z_os; 1638 offset = uio->uio_loffset; 1639 prefetch = zp->z_zn_prefetch; 1640 1641 /* 1642 * Initialize the iterator cursor. 1643 */ 1644 if (offset <= 3) { 1645 /* 1646 * Start iteration from the beginning of the directory. 1647 */ 1648 zap_cursor_init(&zc, os, zp->z_id); 1649 } else { 1650 /* 1651 * The offset is a serialized cursor. 1652 */ 1653 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 1654 } 1655 1656 /* 1657 * Get space to change directory entries into fs independent format. 1658 */ 1659 iovp = uio->uio_iov; 1660 bytes_wanted = iovp->iov_len; 1661 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 1662 bufsize = bytes_wanted; 1663 outbuf = kmem_alloc(bufsize, KM_SLEEP); 1664 odp = (struct dirent64 *)outbuf; 1665 } else { 1666 bufsize = bytes_wanted; 1667 odp = (struct dirent64 *)iovp->iov_base; 1668 } 1669 1670 /* 1671 * Transform to file-system independent format 1672 */ 1673 outcount = 0; 1674 while (outcount < bytes_wanted) { 1675 ino64_t objnum; 1676 ushort_t reclen; 1677 off64_t *next; 1678 1679 /* 1680 * Special case `.', `..', and `.zfs'. 1681 */ 1682 if (offset == 0) { 1683 (void) strcpy(zap.za_name, "."); 1684 objnum = zp->z_id; 1685 } else if (offset == 1) { 1686 (void) strcpy(zap.za_name, ".."); 1687 objnum = zp->z_phys->zp_parent; 1688 } else if (offset == 2 && zfs_show_ctldir(zp)) { 1689 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 1690 objnum = ZFSCTL_INO_ROOT; 1691 } else { 1692 /* 1693 * Grab next entry. 1694 */ 1695 if (error = zap_cursor_retrieve(&zc, &zap)) { 1696 if ((*eofp = (error == ENOENT)) != 0) 1697 break; 1698 else 1699 goto update; 1700 } 1701 1702 if (zap.za_integer_length != 8 || 1703 zap.za_num_integers != 1) { 1704 cmn_err(CE_WARN, "zap_readdir: bad directory " 1705 "entry, obj = %lld, offset = %lld\n", 1706 (u_longlong_t)zp->z_id, 1707 (u_longlong_t)offset); 1708 error = ENXIO; 1709 goto update; 1710 } 1711 1712 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); 1713 /* 1714 * MacOS X can extract the object type here such as: 1715 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); 1716 */ 1717 } 1718 reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 1719 1720 /* 1721 * Will this entry fit in the buffer? 1722 */ 1723 if (outcount + reclen > bufsize) { 1724 /* 1725 * Did we manage to fit anything in the buffer? 1726 */ 1727 if (!outcount) { 1728 error = EINVAL; 1729 goto update; 1730 } 1731 break; 1732 } 1733 /* 1734 * Add this entry: 1735 */ 1736 odp->d_ino = objnum; 1737 odp->d_reclen = reclen; 1738 /* NOTE: d_off is the offset for the *next* entry */ 1739 next = &(odp->d_off); 1740 (void) strncpy(odp->d_name, zap.za_name, 1741 DIRENT64_NAMELEN(reclen)); 1742 outcount += reclen; 1743 odp = (dirent64_t *)((intptr_t)odp + reclen); 1744 1745 ASSERT(outcount <= bufsize); 1746 1747 /* Prefetch znode */ 1748 if (prefetch) 1749 dmu_prefetch(os, objnum, 0, 0); 1750 1751 /* 1752 * Move to the next entry, fill in the previous offset. 1753 */ 1754 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 1755 zap_cursor_advance(&zc); 1756 offset = zap_cursor_serialize(&zc); 1757 } else { 1758 offset += 1; 1759 } 1760 *next = offset; 1761 } 1762 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 1763 1764 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 1765 iovp->iov_base += outcount; 1766 iovp->iov_len -= outcount; 1767 uio->uio_resid -= outcount; 1768 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 1769 /* 1770 * Reset the pointer. 1771 */ 1772 offset = uio->uio_loffset; 1773 } 1774 1775 update: 1776 zap_cursor_fini(&zc); 1777 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 1778 kmem_free(outbuf, bufsize); 1779 1780 if (error == ENOENT) 1781 error = 0; 1782 1783 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 1784 1785 uio->uio_loffset = offset; 1786 ZFS_EXIT(zfsvfs); 1787 return (error); 1788 } 1789 1790 static int 1791 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr) 1792 { 1793 znode_t *zp = VTOZ(vp); 1794 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1795 1796 /* 1797 * Regardless of whether this is required for standards conformance, 1798 * this is the logical behavior when fsync() is called on a file with 1799 * dirty pages. We use B_ASYNC since the ZIL transactions are already 1800 * going to be pushed out as part of the zil_commit(). 1801 */ 1802 if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) && 1803 (vp->v_type == VREG) && !(IS_SWAPVP(vp))) 1804 (void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr); 1805 1806 ZFS_ENTER(zfsvfs); 1807 zil_commit(zfsvfs->z_log, zp->z_last_itx, zp->z_id); 1808 ZFS_EXIT(zfsvfs); 1809 return (0); 1810 } 1811 1812 /* 1813 * Get the requested file attributes and place them in the provided 1814 * vattr structure. 1815 * 1816 * IN: vp - vnode of file. 1817 * vap - va_mask identifies requested attributes. 1818 * flags - [UNUSED] 1819 * cr - credentials of caller. 1820 * 1821 * OUT: vap - attribute values. 1822 * 1823 * RETURN: 0 (always succeeds) 1824 */ 1825 /* ARGSUSED */ 1826 static int 1827 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr) 1828 { 1829 znode_t *zp = VTOZ(vp); 1830 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1831 znode_phys_t *pzp = zp->z_phys; 1832 int error; 1833 1834 ZFS_ENTER(zfsvfs); 1835 1836 /* 1837 * Return all attributes. It's cheaper to provide the answer 1838 * than to determine whether we were asked the question. 1839 */ 1840 mutex_enter(&zp->z_lock); 1841 1842 vap->va_type = vp->v_type; 1843 vap->va_mode = pzp->zp_mode & MODEMASK; 1844 vap->va_uid = zp->z_phys->zp_uid; 1845 vap->va_gid = zp->z_phys->zp_gid; 1846 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev; 1847 vap->va_nodeid = zp->z_id; 1848 vap->va_nlink = MIN(pzp->zp_links, UINT32_MAX); /* nlink_t limit! */ 1849 vap->va_size = pzp->zp_size; 1850 vap->va_rdev = vp->v_rdev; 1851 vap->va_seq = zp->z_seq; 1852 1853 ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime); 1854 ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime); 1855 ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime); 1856 1857 /* 1858 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 1859 * Also, if we are the owner don't bother, since owner should 1860 * always be allowed to read basic attributes of file. 1861 */ 1862 if (!(zp->z_phys->zp_flags & ZFS_ACL_TRIVIAL) && 1863 (zp->z_phys->zp_uid != crgetuid(cr))) { 1864 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, cr)) { 1865 mutex_exit(&zp->z_lock); 1866 ZFS_EXIT(zfsvfs); 1867 return (error); 1868 } 1869 } 1870 1871 mutex_exit(&zp->z_lock); 1872 1873 dmu_object_size_from_db(zp->z_dbuf, &vap->va_blksize, &vap->va_nblocks); 1874 1875 if (zp->z_blksz == 0) { 1876 /* 1877 * Block size hasn't been set; suggest maximal I/O transfers. 1878 */ 1879 vap->va_blksize = zfsvfs->z_max_blksz; 1880 } 1881 1882 ZFS_EXIT(zfsvfs); 1883 return (0); 1884 } 1885 1886 /* 1887 * Set the file attributes to the values contained in the 1888 * vattr structure. 1889 * 1890 * IN: vp - vnode of file to be modified. 1891 * vap - new attribute values. 1892 * flags - ATTR_UTIME set if non-default time values provided. 1893 * cr - credentials of caller. 1894 * 1895 * RETURN: 0 if success 1896 * error code if failure 1897 * 1898 * Timestamps: 1899 * vp - ctime updated, mtime updated if size changed. 1900 */ 1901 /* ARGSUSED */ 1902 static int 1903 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 1904 caller_context_t *ct) 1905 { 1906 struct znode *zp = VTOZ(vp); 1907 znode_phys_t *pzp = zp->z_phys; 1908 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1909 zilog_t *zilog = zfsvfs->z_log; 1910 dmu_tx_t *tx; 1911 vattr_t oldva; 1912 uint_t mask = vap->va_mask; 1913 uint_t saved_mask; 1914 int trim_mask = 0; 1915 uint64_t new_mode; 1916 znode_t *attrzp; 1917 int need_policy = FALSE; 1918 int err; 1919 1920 if (mask == 0) 1921 return (0); 1922 1923 if (mask & AT_NOSET) 1924 return (EINVAL); 1925 1926 if (mask & AT_SIZE && vp->v_type == VDIR) 1927 return (EISDIR); 1928 1929 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) 1930 return (EINVAL); 1931 1932 ZFS_ENTER(zfsvfs); 1933 1934 top: 1935 attrzp = NULL; 1936 1937 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 1938 ZFS_EXIT(zfsvfs); 1939 return (EROFS); 1940 } 1941 1942 /* 1943 * First validate permissions 1944 */ 1945 1946 if (mask & AT_SIZE) { 1947 err = zfs_zaccess(zp, ACE_WRITE_DATA, cr); 1948 if (err) { 1949 ZFS_EXIT(zfsvfs); 1950 return (err); 1951 } 1952 /* 1953 * XXX - Note, we are not providing any open 1954 * mode flags here (like FNDELAY), so we may 1955 * block if there are locks present... this 1956 * should be addressed in openat(). 1957 */ 1958 do { 1959 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 1960 /* NB: we already did dmu_tx_wait() if necessary */ 1961 } while (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT); 1962 if (err) { 1963 ZFS_EXIT(zfsvfs); 1964 return (err); 1965 } 1966 } 1967 1968 if (mask & (AT_ATIME|AT_MTIME)) 1969 need_policy = zfs_zaccess_v4_perm(zp, ACE_WRITE_ATTRIBUTES, cr); 1970 1971 if (mask & (AT_UID|AT_GID)) { 1972 int idmask = (mask & (AT_UID|AT_GID)); 1973 int take_owner; 1974 int take_group; 1975 1976 /* 1977 * NOTE: even if a new mode is being set, 1978 * we may clear S_ISUID/S_ISGID bits. 1979 */ 1980 1981 if (!(mask & AT_MODE)) 1982 vap->va_mode = pzp->zp_mode; 1983 1984 /* 1985 * Take ownership or chgrp to group we are a member of 1986 */ 1987 1988 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 1989 take_group = (mask & AT_GID) && groupmember(vap->va_gid, cr); 1990 1991 /* 1992 * If both AT_UID and AT_GID are set then take_owner and 1993 * take_group must both be set in order to allow taking 1994 * ownership. 1995 * 1996 * Otherwise, send the check through secpolicy_vnode_setattr() 1997 * 1998 */ 1999 2000 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 2001 ((idmask == AT_UID) && take_owner) || 2002 ((idmask == AT_GID) && take_group)) { 2003 if (zfs_zaccess_v4_perm(zp, ACE_WRITE_OWNER, cr) == 0) { 2004 /* 2005 * Remove setuid/setgid for non-privileged users 2006 */ 2007 secpolicy_setid_clear(vap, cr); 2008 trim_mask = (mask & (AT_UID|AT_GID)); 2009 } else { 2010 need_policy = TRUE; 2011 } 2012 } else { 2013 need_policy = TRUE; 2014 } 2015 } 2016 2017 mutex_enter(&zp->z_lock); 2018 oldva.va_mode = pzp->zp_mode; 2019 oldva.va_uid = zp->z_phys->zp_uid; 2020 oldva.va_gid = zp->z_phys->zp_gid; 2021 mutex_exit(&zp->z_lock); 2022 2023 if (mask & AT_MODE) { 2024 if (zfs_zaccess_v4_perm(zp, ACE_WRITE_ACL, cr) == 0) { 2025 err = secpolicy_setid_setsticky_clear(vp, vap, 2026 &oldva, cr); 2027 if (err) { 2028 ZFS_EXIT(zfsvfs); 2029 return (err); 2030 } 2031 trim_mask |= AT_MODE; 2032 } else { 2033 need_policy = TRUE; 2034 } 2035 } 2036 2037 if (need_policy) { 2038 /* 2039 * If trim_mask is set then take ownership 2040 * has been granted or write_acl is present and user 2041 * has the ability to modify mode. In that case remove 2042 * UID|GID and or MODE from mask so that 2043 * secpolicy_vnode_setattr() doesn't revoke it. 2044 */ 2045 2046 if (trim_mask) { 2047 saved_mask = vap->va_mask; 2048 vap->va_mask &= ~trim_mask; 2049 2050 } 2051 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 2052 (int (*)(void *, int, cred_t *))zfs_zaccess_rwx, zp); 2053 if (err) { 2054 ZFS_EXIT(zfsvfs); 2055 return (err); 2056 } 2057 2058 if (trim_mask) 2059 vap->va_mask |= saved_mask; 2060 } 2061 2062 /* 2063 * secpolicy_vnode_setattr, or take ownership may have 2064 * changed va_mask 2065 */ 2066 mask = vap->va_mask; 2067 2068 tx = dmu_tx_create(zfsvfs->z_os); 2069 dmu_tx_hold_bonus(tx, zp->z_id); 2070 2071 if (mask & AT_MODE) { 2072 uint64_t pmode = pzp->zp_mode; 2073 2074 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 2075 2076 if (zp->z_phys->zp_acl.z_acl_extern_obj) 2077 dmu_tx_hold_write(tx, 2078 pzp->zp_acl.z_acl_extern_obj, 0, SPA_MAXBLOCKSIZE); 2079 else 2080 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2081 0, ZFS_ACL_SIZE(MAX_ACL_SIZE)); 2082 } 2083 2084 if ((mask & (AT_UID | AT_GID)) && zp->z_phys->zp_xattr != 0) { 2085 err = zfs_zget(zp->z_zfsvfs, zp->z_phys->zp_xattr, &attrzp); 2086 if (err) { 2087 dmu_tx_abort(tx); 2088 ZFS_EXIT(zfsvfs); 2089 return (err); 2090 } 2091 dmu_tx_hold_bonus(tx, attrzp->z_id); 2092 } 2093 2094 err = dmu_tx_assign(tx, zfsvfs->z_assign); 2095 if (err) { 2096 if (attrzp) 2097 VN_RELE(ZTOV(attrzp)); 2098 if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2099 dmu_tx_wait(tx); 2100 dmu_tx_abort(tx); 2101 goto top; 2102 } 2103 dmu_tx_abort(tx); 2104 ZFS_EXIT(zfsvfs); 2105 return (err); 2106 } 2107 2108 dmu_buf_will_dirty(zp->z_dbuf, tx); 2109 2110 /* 2111 * Set each attribute requested. 2112 * We group settings according to the locks they need to acquire. 2113 * 2114 * Note: you cannot set ctime directly, although it will be 2115 * updated as a side-effect of calling this function. 2116 */ 2117 2118 mutex_enter(&zp->z_lock); 2119 2120 if (mask & AT_MODE) { 2121 err = zfs_acl_chmod_setattr(zp, new_mode, tx); 2122 ASSERT3U(err, ==, 0); 2123 } 2124 2125 if (attrzp) 2126 mutex_enter(&attrzp->z_lock); 2127 2128 if (mask & AT_UID) { 2129 zp->z_phys->zp_uid = (uint64_t)vap->va_uid; 2130 if (attrzp) { 2131 attrzp->z_phys->zp_uid = (uint64_t)vap->va_uid; 2132 } 2133 } 2134 2135 if (mask & AT_GID) { 2136 zp->z_phys->zp_gid = (uint64_t)vap->va_gid; 2137 if (attrzp) 2138 attrzp->z_phys->zp_gid = (uint64_t)vap->va_gid; 2139 } 2140 2141 if (attrzp) 2142 mutex_exit(&attrzp->z_lock); 2143 2144 if (mask & AT_ATIME) 2145 ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime); 2146 2147 if (mask & AT_MTIME) 2148 ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime); 2149 2150 if (mask & AT_SIZE) 2151 zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx); 2152 else if (mask != 0) 2153 zfs_time_stamper_locked(zp, STATE_CHANGED, tx); 2154 2155 if (mask != 0) 2156 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask); 2157 2158 mutex_exit(&zp->z_lock); 2159 2160 if (attrzp) 2161 VN_RELE(ZTOV(attrzp)); 2162 2163 dmu_tx_commit(tx); 2164 2165 ZFS_EXIT(zfsvfs); 2166 return (err); 2167 } 2168 2169 typedef struct zfs_zlock { 2170 krwlock_t *zl_rwlock; /* lock we acquired */ 2171 znode_t *zl_znode; /* znode we held */ 2172 struct zfs_zlock *zl_next; /* next in list */ 2173 } zfs_zlock_t; 2174 2175 /* 2176 * Drop locks and release vnodes that were held by zfs_rename_lock(). 2177 */ 2178 static void 2179 zfs_rename_unlock(zfs_zlock_t **zlpp) 2180 { 2181 zfs_zlock_t *zl; 2182 2183 while ((zl = *zlpp) != NULL) { 2184 if (zl->zl_znode != NULL) 2185 VN_RELE(ZTOV(zl->zl_znode)); 2186 rw_exit(zl->zl_rwlock); 2187 *zlpp = zl->zl_next; 2188 kmem_free(zl, sizeof (*zl)); 2189 } 2190 } 2191 2192 /* 2193 * Search back through the directory tree, using the ".." entries. 2194 * Lock each directory in the chain to prevent concurrent renames. 2195 * Fail any attempt to move a directory into one of its own descendants. 2196 * XXX - z_parent_lock can overlap with map or grow locks 2197 */ 2198 static int 2199 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 2200 { 2201 zfs_zlock_t *zl; 2202 znode_t *zp = tdzp; 2203 uint64_t rootid = zp->z_zfsvfs->z_root; 2204 uint64_t *oidp = &zp->z_id; 2205 krwlock_t *rwlp = &szp->z_parent_lock; 2206 krw_t rw = RW_WRITER; 2207 2208 /* 2209 * First pass write-locks szp and compares to zp->z_id. 2210 * Later passes read-lock zp and compare to zp->z_parent. 2211 */ 2212 do { 2213 if (!rw_tryenter(rwlp, rw)) { 2214 /* 2215 * Another thread is renaming in this path. 2216 * Note that if we are a WRITER, we don't have any 2217 * parent_locks held yet. 2218 */ 2219 if (rw == RW_READER && zp->z_id > szp->z_id) { 2220 /* 2221 * Drop our locks and restart 2222 */ 2223 zfs_rename_unlock(&zl); 2224 *zlpp = NULL; 2225 zp = tdzp; 2226 oidp = &zp->z_id; 2227 rwlp = &szp->z_parent_lock; 2228 rw = RW_WRITER; 2229 continue; 2230 } else { 2231 /* 2232 * Wait for other thread to drop its locks 2233 */ 2234 rw_enter(rwlp, rw); 2235 } 2236 } 2237 2238 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 2239 zl->zl_rwlock = rwlp; 2240 zl->zl_znode = NULL; 2241 zl->zl_next = *zlpp; 2242 *zlpp = zl; 2243 2244 if (*oidp == szp->z_id) /* We're a descendant of szp */ 2245 return (EINVAL); 2246 2247 if (*oidp == rootid) /* We've hit the top */ 2248 return (0); 2249 2250 if (rw == RW_READER) { /* i.e. not the first pass */ 2251 int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp); 2252 if (error) 2253 return (error); 2254 zl->zl_znode = zp; 2255 } 2256 oidp = &zp->z_phys->zp_parent; 2257 rwlp = &zp->z_parent_lock; 2258 rw = RW_READER; 2259 2260 } while (zp->z_id != sdzp->z_id); 2261 2262 return (0); 2263 } 2264 2265 /* 2266 * Move an entry from the provided source directory to the target 2267 * directory. Change the entry name as indicated. 2268 * 2269 * IN: sdvp - Source directory containing the "old entry". 2270 * snm - Old entry name. 2271 * tdvp - Target directory to contain the "new entry". 2272 * tnm - New entry name. 2273 * cr - credentials of caller. 2274 * 2275 * RETURN: 0 if success 2276 * error code if failure 2277 * 2278 * Timestamps: 2279 * sdvp,tdvp - ctime|mtime updated 2280 */ 2281 static int 2282 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr) 2283 { 2284 znode_t *tdzp, *szp, *tzp; 2285 znode_t *sdzp = VTOZ(sdvp); 2286 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs; 2287 zilog_t *zilog = zfsvfs->z_log; 2288 vnode_t *realvp; 2289 zfs_dirlock_t *sdl, *tdl; 2290 dmu_tx_t *tx; 2291 zfs_zlock_t *zl; 2292 int cmp, serr, terr, error; 2293 2294 ZFS_ENTER(zfsvfs); 2295 2296 /* 2297 * Make sure we have the real vp for the target directory. 2298 */ 2299 if (VOP_REALVP(tdvp, &realvp) == 0) 2300 tdvp = realvp; 2301 2302 if (tdvp->v_vfsp != sdvp->v_vfsp) { 2303 ZFS_EXIT(zfsvfs); 2304 return (EXDEV); 2305 } 2306 2307 tdzp = VTOZ(tdvp); 2308 top: 2309 szp = NULL; 2310 tzp = NULL; 2311 zl = NULL; 2312 2313 /* 2314 * This is to prevent the creation of links into attribute space 2315 * by renaming a linked file into/outof an attribute directory. 2316 * See the comment in zfs_link() for why this is considered bad. 2317 */ 2318 if ((tdzp->z_phys->zp_flags & ZFS_XATTR) != 2319 (sdzp->z_phys->zp_flags & ZFS_XATTR)) { 2320 ZFS_EXIT(zfsvfs); 2321 return (EINVAL); 2322 } 2323 2324 /* 2325 * Lock source and target directory entries. To prevent deadlock, 2326 * a lock ordering must be defined. We lock the directory with 2327 * the smallest object id first, or if it's a tie, the one with 2328 * the lexically first name. 2329 */ 2330 if (sdzp->z_id < tdzp->z_id) { 2331 cmp = -1; 2332 } else if (sdzp->z_id > tdzp->z_id) { 2333 cmp = 1; 2334 } else { 2335 cmp = strcmp(snm, tnm); 2336 if (cmp == 0) { 2337 /* 2338 * POSIX: "If the old argument and the new argument 2339 * both refer to links to the same existing file, 2340 * the rename() function shall return successfully 2341 * and perform no other action." 2342 */ 2343 ZFS_EXIT(zfsvfs); 2344 return (0); 2345 } 2346 } 2347 if (cmp < 0) { 2348 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS); 2349 terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0); 2350 } else { 2351 terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0); 2352 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS); 2353 } 2354 2355 if (serr) { 2356 /* 2357 * Source entry invalid or not there. 2358 */ 2359 if (!terr) { 2360 zfs_dirent_unlock(tdl); 2361 if (tzp) 2362 VN_RELE(ZTOV(tzp)); 2363 } 2364 if (strcmp(snm, "..") == 0) 2365 serr = EINVAL; 2366 ZFS_EXIT(zfsvfs); 2367 return (serr); 2368 } 2369 if (terr) { 2370 zfs_dirent_unlock(sdl); 2371 VN_RELE(ZTOV(szp)); 2372 if (strcmp(tnm, "..") == 0) 2373 terr = EINVAL; 2374 ZFS_EXIT(zfsvfs); 2375 return (terr); 2376 } 2377 2378 /* 2379 * Must have write access at the source to remove the old entry 2380 * and write access at the target to create the new entry. 2381 * Note that if target and source are the same, this can be 2382 * done in a single check. 2383 */ 2384 2385 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 2386 goto out; 2387 2388 if (ZTOV(szp)->v_type == VDIR) { 2389 /* 2390 * Check to make sure rename is valid. 2391 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 2392 */ 2393 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 2394 goto out; 2395 } 2396 2397 /* 2398 * Does target exist? 2399 */ 2400 if (tzp) { 2401 /* 2402 * Source and target must be the same type. 2403 */ 2404 if (ZTOV(szp)->v_type == VDIR) { 2405 if (ZTOV(tzp)->v_type != VDIR) { 2406 error = ENOTDIR; 2407 goto out; 2408 } 2409 } else { 2410 if (ZTOV(tzp)->v_type == VDIR) { 2411 error = EISDIR; 2412 goto out; 2413 } 2414 } 2415 /* 2416 * POSIX dictates that when the source and target 2417 * entries refer to the same file object, rename 2418 * must do nothing and exit without error. 2419 */ 2420 if (szp->z_id == tzp->z_id) { 2421 error = 0; 2422 goto out; 2423 } 2424 } 2425 2426 vnevent_rename_src(ZTOV(szp)); 2427 if (tzp) 2428 vnevent_rename_dest(ZTOV(tzp)); 2429 2430 tx = dmu_tx_create(zfsvfs->z_os); 2431 dmu_tx_hold_bonus(tx, szp->z_id); /* nlink changes */ 2432 dmu_tx_hold_bonus(tx, sdzp->z_id); /* nlink changes */ 2433 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 2434 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 2435 if (sdzp != tdzp) 2436 dmu_tx_hold_bonus(tx, tdzp->z_id); /* nlink changes */ 2437 if (tzp) 2438 dmu_tx_hold_bonus(tx, tzp->z_id); /* parent changes */ 2439 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 2440 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2441 if (error) { 2442 if (zl != NULL) 2443 zfs_rename_unlock(&zl); 2444 zfs_dirent_unlock(sdl); 2445 zfs_dirent_unlock(tdl); 2446 VN_RELE(ZTOV(szp)); 2447 if (tzp) 2448 VN_RELE(ZTOV(tzp)); 2449 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2450 dmu_tx_wait(tx); 2451 dmu_tx_abort(tx); 2452 goto top; 2453 } 2454 dmu_tx_abort(tx); 2455 ZFS_EXIT(zfsvfs); 2456 return (error); 2457 } 2458 2459 if (tzp) /* Attempt to remove the existing target */ 2460 error = zfs_link_destroy(tdl, tzp, tx, 0, NULL); 2461 2462 if (error == 0) { 2463 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 2464 if (error == 0) { 2465 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 2466 ASSERT(error == 0); 2467 zfs_log_rename(zilog, tx, TX_RENAME, sdzp, 2468 sdl->dl_name, tdzp, tdl->dl_name, szp); 2469 } 2470 } 2471 2472 dmu_tx_commit(tx); 2473 out: 2474 if (zl != NULL) 2475 zfs_rename_unlock(&zl); 2476 2477 zfs_dirent_unlock(sdl); 2478 zfs_dirent_unlock(tdl); 2479 2480 VN_RELE(ZTOV(szp)); 2481 if (tzp) 2482 VN_RELE(ZTOV(tzp)); 2483 2484 ZFS_EXIT(zfsvfs); 2485 return (error); 2486 } 2487 2488 /* 2489 * Insert the indicated symbolic reference entry into the directory. 2490 * 2491 * IN: dvp - Directory to contain new symbolic link. 2492 * link - Name for new symlink entry. 2493 * vap - Attributes of new entry. 2494 * target - Target path of new symlink. 2495 * cr - credentials of caller. 2496 * 2497 * RETURN: 0 if success 2498 * error code if failure 2499 * 2500 * Timestamps: 2501 * dvp - ctime|mtime updated 2502 */ 2503 static int 2504 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr) 2505 { 2506 znode_t *zp, *dzp = VTOZ(dvp); 2507 zfs_dirlock_t *dl; 2508 dmu_tx_t *tx; 2509 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2510 zilog_t *zilog = zfsvfs->z_log; 2511 uint64_t zoid; 2512 int len = strlen(link); 2513 int error; 2514 2515 ASSERT(vap->va_type == VLNK); 2516 2517 ZFS_ENTER(zfsvfs); 2518 top: 2519 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 2520 ZFS_EXIT(zfsvfs); 2521 return (error); 2522 } 2523 2524 if (len > MAXPATHLEN) { 2525 ZFS_EXIT(zfsvfs); 2526 return (ENAMETOOLONG); 2527 } 2528 2529 /* 2530 * Attempt to lock directory; fail if entry already exists. 2531 */ 2532 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZNEW)) { 2533 ZFS_EXIT(zfsvfs); 2534 return (error); 2535 } 2536 2537 tx = dmu_tx_create(zfsvfs->z_os); 2538 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 2539 dmu_tx_hold_bonus(tx, dzp->z_id); 2540 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 2541 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 2542 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE); 2543 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2544 if (error) { 2545 zfs_dirent_unlock(dl); 2546 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2547 dmu_tx_wait(tx); 2548 dmu_tx_abort(tx); 2549 goto top; 2550 } 2551 dmu_tx_abort(tx); 2552 ZFS_EXIT(zfsvfs); 2553 return (error); 2554 } 2555 2556 dmu_buf_will_dirty(dzp->z_dbuf, tx); 2557 2558 /* 2559 * Create a new object for the symlink. 2560 * Put the link content into bonus buffer if it will fit; 2561 * otherwise, store it just like any other file data. 2562 */ 2563 zoid = 0; 2564 if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) { 2565 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, len); 2566 if (len != 0) 2567 bcopy(link, zp->z_phys + 1, len); 2568 } else { 2569 dmu_buf_t *dbp; 2570 2571 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 2572 2573 /* 2574 * Nothing can access the znode yet so no locking needed 2575 * for growing the znode's blocksize. 2576 */ 2577 zfs_grow_blocksize(zp, len, tx); 2578 2579 VERIFY(0 == dmu_buf_hold(zfsvfs->z_os, zoid, 0, FTAG, &dbp)); 2580 dmu_buf_will_dirty(dbp, tx); 2581 2582 ASSERT3U(len, <=, dbp->db_size); 2583 bcopy(link, dbp->db_data, len); 2584 dmu_buf_rele(dbp, FTAG); 2585 } 2586 zp->z_phys->zp_size = len; 2587 2588 /* 2589 * Insert the new object into the directory. 2590 */ 2591 (void) zfs_link_create(dl, zp, tx, ZNEW); 2592 out: 2593 if (error == 0) 2594 zfs_log_symlink(zilog, tx, TX_SYMLINK, dzp, zp, name, link); 2595 2596 dmu_tx_commit(tx); 2597 2598 zfs_dirent_unlock(dl); 2599 2600 VN_RELE(ZTOV(zp)); 2601 2602 ZFS_EXIT(zfsvfs); 2603 return (error); 2604 } 2605 2606 /* 2607 * Return, in the buffer contained in the provided uio structure, 2608 * the symbolic path referred to by vp. 2609 * 2610 * IN: vp - vnode of symbolic link. 2611 * uoip - structure to contain the link path. 2612 * cr - credentials of caller. 2613 * 2614 * OUT: uio - structure to contain the link path. 2615 * 2616 * RETURN: 0 if success 2617 * error code if failure 2618 * 2619 * Timestamps: 2620 * vp - atime updated 2621 */ 2622 /* ARGSUSED */ 2623 static int 2624 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr) 2625 { 2626 znode_t *zp = VTOZ(vp); 2627 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2628 size_t bufsz; 2629 int error; 2630 2631 ZFS_ENTER(zfsvfs); 2632 2633 bufsz = (size_t)zp->z_phys->zp_size; 2634 if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) { 2635 error = uiomove(zp->z_phys + 1, 2636 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 2637 } else { 2638 dmu_buf_t *dbp; 2639 error = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0, FTAG, &dbp); 2640 if (error) { 2641 ZFS_EXIT(zfsvfs); 2642 return (error); 2643 } 2644 error = uiomove(dbp->db_data, 2645 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 2646 dmu_buf_rele(dbp, FTAG); 2647 } 2648 2649 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2650 ZFS_EXIT(zfsvfs); 2651 return (error); 2652 } 2653 2654 /* 2655 * Insert a new entry into directory tdvp referencing svp. 2656 * 2657 * IN: tdvp - Directory to contain new entry. 2658 * svp - vnode of new entry. 2659 * name - name of new entry. 2660 * cr - credentials of caller. 2661 * 2662 * RETURN: 0 if success 2663 * error code if failure 2664 * 2665 * Timestamps: 2666 * tdvp - ctime|mtime updated 2667 * svp - ctime updated 2668 */ 2669 /* ARGSUSED */ 2670 static int 2671 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr) 2672 { 2673 znode_t *dzp = VTOZ(tdvp); 2674 znode_t *tzp, *szp; 2675 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2676 zilog_t *zilog = zfsvfs->z_log; 2677 zfs_dirlock_t *dl; 2678 dmu_tx_t *tx; 2679 vnode_t *realvp; 2680 int error; 2681 2682 ASSERT(tdvp->v_type == VDIR); 2683 2684 ZFS_ENTER(zfsvfs); 2685 2686 if (VOP_REALVP(svp, &realvp) == 0) 2687 svp = realvp; 2688 2689 if (svp->v_vfsp != tdvp->v_vfsp) { 2690 ZFS_EXIT(zfsvfs); 2691 return (EXDEV); 2692 } 2693 2694 szp = VTOZ(svp); 2695 top: 2696 /* 2697 * We do not support links between attributes and non-attributes 2698 * because of the potential security risk of creating links 2699 * into "normal" file space in order to circumvent restrictions 2700 * imposed in attribute space. 2701 */ 2702 if ((szp->z_phys->zp_flags & ZFS_XATTR) != 2703 (dzp->z_phys->zp_flags & ZFS_XATTR)) { 2704 ZFS_EXIT(zfsvfs); 2705 return (EINVAL); 2706 } 2707 2708 /* 2709 * POSIX dictates that we return EPERM here. 2710 * Better choices include ENOTSUP or EISDIR. 2711 */ 2712 if (svp->v_type == VDIR) { 2713 ZFS_EXIT(zfsvfs); 2714 return (EPERM); 2715 } 2716 2717 if ((uid_t)szp->z_phys->zp_uid != crgetuid(cr) && 2718 secpolicy_basic_link(cr) != 0) { 2719 ZFS_EXIT(zfsvfs); 2720 return (EPERM); 2721 } 2722 2723 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 2724 ZFS_EXIT(zfsvfs); 2725 return (error); 2726 } 2727 2728 /* 2729 * Attempt to lock directory; fail if entry already exists. 2730 */ 2731 if (error = zfs_dirent_lock(&dl, dzp, name, &tzp, ZNEW)) { 2732 ZFS_EXIT(zfsvfs); 2733 return (error); 2734 } 2735 2736 tx = dmu_tx_create(zfsvfs->z_os); 2737 dmu_tx_hold_bonus(tx, szp->z_id); 2738 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 2739 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2740 if (error) { 2741 zfs_dirent_unlock(dl); 2742 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2743 dmu_tx_wait(tx); 2744 dmu_tx_abort(tx); 2745 goto top; 2746 } 2747 dmu_tx_abort(tx); 2748 ZFS_EXIT(zfsvfs); 2749 return (error); 2750 } 2751 2752 error = zfs_link_create(dl, szp, tx, 0); 2753 2754 if (error == 0) 2755 zfs_log_link(zilog, tx, TX_LINK, dzp, szp, name); 2756 2757 dmu_tx_commit(tx); 2758 2759 zfs_dirent_unlock(dl); 2760 2761 ZFS_EXIT(zfsvfs); 2762 return (error); 2763 } 2764 2765 /* 2766 * zfs_null_putapage() is used when the file system has been force 2767 * unmounted. It just drops the pages. 2768 */ 2769 /* ARGSUSED */ 2770 static int 2771 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 2772 size_t *lenp, int flags, cred_t *cr) 2773 { 2774 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR); 2775 return (0); 2776 } 2777 2778 /* 2779 * Push a page out to disk, klustering if possible. 2780 * 2781 * IN: vp - file to push page to. 2782 * pp - page to push. 2783 * flags - additional flags. 2784 * cr - credentials of caller. 2785 * 2786 * OUT: offp - start of range pushed. 2787 * lenp - len of range pushed. 2788 * 2789 * RETURN: 0 if success 2790 * error code if failure 2791 * 2792 * NOTE: callers must have locked the page to be pushed. On 2793 * exit, the page (and all other pages in the kluster) must be 2794 * unlocked. 2795 */ 2796 /* ARGSUSED */ 2797 static int 2798 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 2799 size_t *lenp, int flags, cred_t *cr) 2800 { 2801 znode_t *zp = VTOZ(vp); 2802 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2803 zilog_t *zilog = zfsvfs->z_log; 2804 dmu_tx_t *tx; 2805 rl_t *rl; 2806 u_offset_t off, koff; 2807 size_t len, klen; 2808 int err; 2809 2810 off = pp->p_offset; 2811 len = PAGESIZE; 2812 /* 2813 * If our blocksize is bigger than the page size, try to kluster 2814 * muiltiple pages so that we write a full block (thus avoiding 2815 * a read-modify-write). 2816 */ 2817 if (zp->z_blksz > PAGESIZE) { 2818 uint64_t filesz = zp->z_phys->zp_size; 2819 2820 if (!ISP2(zp->z_blksz)) { 2821 /* Only one block in the file. */ 2822 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 2823 koff = 0; 2824 } else { 2825 klen = zp->z_blksz; 2826 koff = P2ALIGN(off, (u_offset_t)klen); 2827 } 2828 ASSERT(koff <= filesz); 2829 if (koff + klen > filesz) 2830 klen = P2ROUNDUP(filesz - koff, (uint64_t)PAGESIZE); 2831 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags); 2832 } 2833 ASSERT3U(btop(len), ==, btopr(len)); 2834 top: 2835 rl = zfs_range_lock(zp, off, len, RL_WRITER); 2836 /* 2837 * Can't push pages past end-of-file. 2838 */ 2839 if (off >= zp->z_phys->zp_size) { 2840 /* discard all pages */ 2841 flags |= B_INVAL; 2842 err = 0; 2843 goto out; 2844 } else if (off + len > zp->z_phys->zp_size) { 2845 int npages = btopr(zp->z_phys->zp_size - off); 2846 page_t *trunc; 2847 2848 page_list_break(&pp, &trunc, npages); 2849 /* discard pages past end of file */ 2850 if (trunc) 2851 pvn_write_done(trunc, B_INVAL | flags); 2852 len = zp->z_phys->zp_size - off; 2853 } 2854 2855 tx = dmu_tx_create(zfsvfs->z_os); 2856 dmu_tx_hold_write(tx, zp->z_id, off, len); 2857 dmu_tx_hold_bonus(tx, zp->z_id); 2858 err = dmu_tx_assign(tx, zfsvfs->z_assign); 2859 if (err != 0) { 2860 if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2861 zfs_range_unlock(rl); 2862 dmu_tx_wait(tx); 2863 dmu_tx_abort(tx); 2864 err = 0; 2865 goto top; 2866 } 2867 dmu_tx_abort(tx); 2868 goto out; 2869 } 2870 2871 if (zp->z_blksz <= PAGESIZE) { 2872 caddr_t va = ppmapin(pp, PROT_READ, (caddr_t)-1); 2873 ASSERT3U(len, <=, PAGESIZE); 2874 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx); 2875 ppmapout(va); 2876 } else { 2877 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx); 2878 } 2879 2880 if (err == 0) { 2881 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 2882 zfs_log_write(zilog, tx, TX_WRITE, zp, off, len, 0); 2883 dmu_tx_commit(tx); 2884 } 2885 2886 out: 2887 zfs_range_unlock(rl); 2888 pvn_write_done(pp, (err ? B_ERROR : 0) | B_WRITE | flags); 2889 if (offp) 2890 *offp = off; 2891 if (lenp) 2892 *lenp = len; 2893 2894 return (err); 2895 } 2896 2897 /* 2898 * Copy the portion of the file indicated from pages into the file. 2899 * The pages are stored in a page list attached to the files vnode. 2900 * 2901 * IN: vp - vnode of file to push page data to. 2902 * off - position in file to put data. 2903 * len - amount of data to write. 2904 * flags - flags to control the operation. 2905 * cr - credentials of caller. 2906 * 2907 * RETURN: 0 if success 2908 * error code if failure 2909 * 2910 * Timestamps: 2911 * vp - ctime|mtime updated 2912 */ 2913 static int 2914 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr) 2915 { 2916 znode_t *zp = VTOZ(vp); 2917 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2918 page_t *pp; 2919 size_t io_len; 2920 u_offset_t io_off; 2921 uint64_t filesz; 2922 int error = 0; 2923 2924 ZFS_ENTER(zfsvfs); 2925 2926 ASSERT(zp->z_dbuf_held && zp->z_phys); 2927 2928 if (len == 0) { 2929 /* 2930 * Search the entire vp list for pages >= off. 2931 */ 2932 error = pvn_vplist_dirty(vp, (u_offset_t)off, zfs_putapage, 2933 flags, cr); 2934 goto out; 2935 } 2936 2937 filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */ 2938 if (off > filesz) { 2939 /* past end of file */ 2940 ZFS_EXIT(zfsvfs); 2941 return (0); 2942 } 2943 2944 len = MIN(len, filesz - off); 2945 2946 for (io_off = off; io_off < off + len; io_off += io_len) { 2947 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { 2948 pp = page_lookup(vp, io_off, 2949 (flags & (B_INVAL | B_FREE)) ? 2950 SE_EXCL : SE_SHARED); 2951 } else { 2952 pp = page_lookup_nowait(vp, io_off, 2953 (flags & B_FREE) ? SE_EXCL : SE_SHARED); 2954 } 2955 2956 if (pp != NULL && pvn_getdirty(pp, flags)) { 2957 int err; 2958 2959 /* 2960 * Found a dirty page to push 2961 */ 2962 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr); 2963 if (err) 2964 error = err; 2965 } else { 2966 io_len = PAGESIZE; 2967 } 2968 } 2969 out: 2970 if ((flags & B_ASYNC) == 0) 2971 zil_commit(zfsvfs->z_log, UINT64_MAX, zp->z_id); 2972 ZFS_EXIT(zfsvfs); 2973 return (error); 2974 } 2975 2976 void 2977 zfs_inactive(vnode_t *vp, cred_t *cr) 2978 { 2979 znode_t *zp = VTOZ(vp); 2980 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2981 int error; 2982 2983 rw_enter(&zfsvfs->z_um_lock, RW_READER); 2984 if (zfsvfs->z_unmounted2) { 2985 ASSERT(zp->z_dbuf_held == 0); 2986 2987 if (vn_has_cached_data(vp)) { 2988 (void) pvn_vplist_dirty(vp, 0, zfs_null_putapage, 2989 B_INVAL, cr); 2990 } 2991 2992 mutex_enter(&zp->z_lock); 2993 vp->v_count = 0; /* count arrives as 1 */ 2994 if (zp->z_dbuf == NULL) { 2995 mutex_exit(&zp->z_lock); 2996 zfs_znode_free(zp); 2997 } else { 2998 mutex_exit(&zp->z_lock); 2999 } 3000 rw_exit(&zfsvfs->z_um_lock); 3001 VFS_RELE(zfsvfs->z_vfs); 3002 return; 3003 } 3004 3005 /* 3006 * Attempt to push any data in the page cache. If this fails 3007 * we will get kicked out later in zfs_zinactive(). 3008 */ 3009 if (vn_has_cached_data(vp)) { 3010 (void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC, 3011 cr); 3012 } 3013 3014 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 3015 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 3016 3017 dmu_tx_hold_bonus(tx, zp->z_id); 3018 error = dmu_tx_assign(tx, TXG_WAIT); 3019 if (error) { 3020 dmu_tx_abort(tx); 3021 } else { 3022 dmu_buf_will_dirty(zp->z_dbuf, tx); 3023 mutex_enter(&zp->z_lock); 3024 zp->z_atime_dirty = 0; 3025 mutex_exit(&zp->z_lock); 3026 dmu_tx_commit(tx); 3027 } 3028 } 3029 3030 zfs_zinactive(zp); 3031 rw_exit(&zfsvfs->z_um_lock); 3032 } 3033 3034 /* 3035 * Bounds-check the seek operation. 3036 * 3037 * IN: vp - vnode seeking within 3038 * ooff - old file offset 3039 * noffp - pointer to new file offset 3040 * 3041 * RETURN: 0 if success 3042 * EINVAL if new offset invalid 3043 */ 3044 /* ARGSUSED */ 3045 static int 3046 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp) 3047 { 3048 if (vp->v_type == VDIR) 3049 return (0); 3050 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0); 3051 } 3052 3053 /* 3054 * Pre-filter the generic locking function to trap attempts to place 3055 * a mandatory lock on a memory mapped file. 3056 */ 3057 static int 3058 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset, 3059 flk_callback_t *flk_cbp, cred_t *cr) 3060 { 3061 znode_t *zp = VTOZ(vp); 3062 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3063 int error; 3064 3065 ZFS_ENTER(zfsvfs); 3066 3067 /* 3068 * We are following the UFS semantics with respect to mapcnt 3069 * here: If we see that the file is mapped already, then we will 3070 * return an error, but we don't worry about races between this 3071 * function and zfs_map(). 3072 */ 3073 if (zp->z_mapcnt > 0 && MANDMODE((mode_t)zp->z_phys->zp_mode)) { 3074 ZFS_EXIT(zfsvfs); 3075 return (EAGAIN); 3076 } 3077 error = fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr); 3078 ZFS_EXIT(zfsvfs); 3079 return (error); 3080 } 3081 3082 /* 3083 * If we can't find a page in the cache, we will create a new page 3084 * and fill it with file data. For efficiency, we may try to fill 3085 * multiple pages at once (klustering). 3086 */ 3087 static int 3088 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, 3089 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw) 3090 { 3091 znode_t *zp = VTOZ(vp); 3092 page_t *pp, *cur_pp; 3093 objset_t *os = zp->z_zfsvfs->z_os; 3094 caddr_t va; 3095 u_offset_t io_off, total; 3096 uint64_t oid = zp->z_id; 3097 size_t io_len; 3098 uint64_t filesz; 3099 int err; 3100 3101 /* 3102 * If we are only asking for a single page don't bother klustering. 3103 */ 3104 filesz = zp->z_phys->zp_size; /* get consistent copy of zp_size */ 3105 if (off >= filesz) 3106 return (EFAULT); 3107 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) { 3108 io_off = off; 3109 io_len = PAGESIZE; 3110 pp = page_create_va(vp, io_off, io_len, PG_WAIT, seg, addr); 3111 } else { 3112 /* 3113 * Try to fill a kluster of pages (a blocks worth). 3114 */ 3115 size_t klen; 3116 u_offset_t koff; 3117 3118 if (!ISP2(zp->z_blksz)) { 3119 /* Only one block in the file. */ 3120 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 3121 koff = 0; 3122 } else { 3123 /* 3124 * It would be ideal to align our offset to the 3125 * blocksize but doing so has resulted in some 3126 * strange application crashes. For now, we 3127 * leave the offset as is and only adjust the 3128 * length if we are off the end of the file. 3129 */ 3130 koff = off; 3131 klen = plsz; 3132 } 3133 ASSERT(koff <= filesz); 3134 if (koff + klen > filesz) 3135 klen = P2ROUNDUP(filesz, (uint64_t)PAGESIZE) - koff; 3136 ASSERT3U(off, >=, koff); 3137 ASSERT3U(off, <, koff + klen); 3138 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 3139 &io_len, koff, klen, 0); 3140 } 3141 if (pp == NULL) { 3142 /* 3143 * Some other thread entered the page before us. 3144 * Return to zfs_getpage to retry the lookup. 3145 */ 3146 *pl = NULL; 3147 return (0); 3148 } 3149 3150 /* 3151 * Fill the pages in the kluster. 3152 */ 3153 cur_pp = pp; 3154 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) { 3155 ASSERT3U(io_off, ==, cur_pp->p_offset); 3156 va = ppmapin(cur_pp, PROT_READ | PROT_WRITE, (caddr_t)-1); 3157 err = dmu_read(os, oid, io_off, PAGESIZE, va); 3158 ppmapout(va); 3159 if (err) { 3160 /* On error, toss the entire kluster */ 3161 pvn_read_done(pp, B_ERROR); 3162 return (err); 3163 } 3164 cur_pp = cur_pp->p_next; 3165 } 3166 out: 3167 /* 3168 * Fill in the page list array from the kluster. If 3169 * there are too many pages in the kluster, return 3170 * as many pages as possible starting from the desired 3171 * offset `off'. 3172 * NOTE: the page list will always be null terminated. 3173 */ 3174 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 3175 3176 return (0); 3177 } 3178 3179 /* 3180 * Return pointers to the pages for the file region [off, off + len] 3181 * in the pl array. If plsz is greater than len, this function may 3182 * also return page pointers from before or after the specified 3183 * region (i.e. some region [off', off' + plsz]). These additional 3184 * pages are only returned if they are already in the cache, or were 3185 * created as part of a klustered read. 3186 * 3187 * IN: vp - vnode of file to get data from. 3188 * off - position in file to get data from. 3189 * len - amount of data to retrieve. 3190 * plsz - length of provided page list. 3191 * seg - segment to obtain pages for. 3192 * addr - virtual address of fault. 3193 * rw - mode of created pages. 3194 * cr - credentials of caller. 3195 * 3196 * OUT: protp - protection mode of created pages. 3197 * pl - list of pages created. 3198 * 3199 * RETURN: 0 if success 3200 * error code if failure 3201 * 3202 * Timestamps: 3203 * vp - atime updated 3204 */ 3205 /* ARGSUSED */ 3206 static int 3207 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 3208 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 3209 enum seg_rw rw, cred_t *cr) 3210 { 3211 znode_t *zp = VTOZ(vp); 3212 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3213 page_t *pp, **pl0 = pl; 3214 int need_unlock = 0, err = 0; 3215 offset_t orig_off; 3216 3217 ZFS_ENTER(zfsvfs); 3218 3219 if (protp) 3220 *protp = PROT_ALL; 3221 3222 ASSERT(zp->z_dbuf_held && zp->z_phys); 3223 3224 /* no faultahead (for now) */ 3225 if (pl == NULL) { 3226 ZFS_EXIT(zfsvfs); 3227 return (0); 3228 } 3229 3230 /* can't fault past EOF */ 3231 if (off >= zp->z_phys->zp_size) { 3232 ZFS_EXIT(zfsvfs); 3233 return (EFAULT); 3234 } 3235 orig_off = off; 3236 3237 /* 3238 * If we already own the lock, then we must be page faulting 3239 * in the middle of a write to this file (i.e., we are writing 3240 * to this file using data from a mapped region of the file). 3241 */ 3242 if (rw_owner(&zp->z_map_lock) != curthread) { 3243 rw_enter(&zp->z_map_lock, RW_WRITER); 3244 need_unlock = TRUE; 3245 } 3246 3247 /* 3248 * Loop through the requested range [off, off + len] looking 3249 * for pages. If we don't find a page, we will need to create 3250 * a new page and fill it with data from the file. 3251 */ 3252 while (len > 0) { 3253 if (plsz < PAGESIZE) 3254 break; 3255 if (pp = page_lookup(vp, off, SE_SHARED)) { 3256 *pl++ = pp; 3257 off += PAGESIZE; 3258 addr += PAGESIZE; 3259 len -= PAGESIZE; 3260 plsz -= PAGESIZE; 3261 } else { 3262 err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw); 3263 if (err) 3264 goto out; 3265 /* 3266 * klustering may have changed our region 3267 * to be block aligned. 3268 */ 3269 if (((pp = *pl) != 0) && (off != pp->p_offset)) { 3270 int delta = off - pp->p_offset; 3271 len += delta; 3272 off -= delta; 3273 addr -= delta; 3274 } 3275 while (*pl) { 3276 pl++; 3277 off += PAGESIZE; 3278 addr += PAGESIZE; 3279 plsz -= PAGESIZE; 3280 if (len > PAGESIZE) 3281 len -= PAGESIZE; 3282 else 3283 len = 0; 3284 } 3285 } 3286 } 3287 3288 /* 3289 * Fill out the page array with any pages already in the cache. 3290 */ 3291 while (plsz > 0) { 3292 pp = page_lookup_nowait(vp, off, SE_SHARED); 3293 if (pp == NULL) 3294 break; 3295 *pl++ = pp; 3296 off += PAGESIZE; 3297 plsz -= PAGESIZE; 3298 } 3299 3300 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 3301 out: 3302 /* 3303 * We can't grab the range lock for the page as reader which would 3304 * stop truncation as this leads to deadlock. So we need to recheck 3305 * the file size. 3306 */ 3307 if (orig_off >= zp->z_phys->zp_size) 3308 err = EFAULT; 3309 if (err) { 3310 /* 3311 * Release any pages we have previously locked. 3312 */ 3313 while (pl > pl0) 3314 page_unlock(*--pl); 3315 } 3316 3317 *pl = NULL; 3318 3319 if (need_unlock) 3320 rw_exit(&zp->z_map_lock); 3321 3322 ZFS_EXIT(zfsvfs); 3323 return (err); 3324 } 3325 3326 /* 3327 * Request a memory map for a section of a file. This code interacts 3328 * with common code and the VM system as follows: 3329 * 3330 * common code calls mmap(), which ends up in smmap_common() 3331 * 3332 * this calls VOP_MAP(), which takes you into (say) zfs 3333 * 3334 * zfs_map() calls as_map(), passing segvn_create() as the callback 3335 * 3336 * segvn_create() creates the new segment and calls VOP_ADDMAP() 3337 * 3338 * zfs_addmap() updates z_mapcnt 3339 */ 3340 static int 3341 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 3342 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr) 3343 { 3344 znode_t *zp = VTOZ(vp); 3345 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3346 segvn_crargs_t vn_a; 3347 int error; 3348 3349 ZFS_ENTER(zfsvfs); 3350 3351 if (vp->v_flag & VNOMAP) { 3352 ZFS_EXIT(zfsvfs); 3353 return (ENOSYS); 3354 } 3355 3356 if (off < 0 || len > MAXOFFSET_T - off) { 3357 ZFS_EXIT(zfsvfs); 3358 return (ENXIO); 3359 } 3360 3361 if (vp->v_type != VREG) { 3362 ZFS_EXIT(zfsvfs); 3363 return (ENODEV); 3364 } 3365 3366 /* 3367 * If file is locked, disallow mapping. 3368 */ 3369 if (MANDMODE((mode_t)zp->z_phys->zp_mode) && vn_has_flocks(vp)) { 3370 ZFS_EXIT(zfsvfs); 3371 return (EAGAIN); 3372 } 3373 3374 as_rangelock(as); 3375 if ((flags & MAP_FIXED) == 0) { 3376 map_addr(addrp, len, off, 1, flags); 3377 if (*addrp == NULL) { 3378 as_rangeunlock(as); 3379 ZFS_EXIT(zfsvfs); 3380 return (ENOMEM); 3381 } 3382 } else { 3383 /* 3384 * User specified address - blow away any previous mappings 3385 */ 3386 (void) as_unmap(as, *addrp, len); 3387 } 3388 3389 vn_a.vp = vp; 3390 vn_a.offset = (u_offset_t)off; 3391 vn_a.type = flags & MAP_TYPE; 3392 vn_a.prot = prot; 3393 vn_a.maxprot = maxprot; 3394 vn_a.cred = cr; 3395 vn_a.amp = NULL; 3396 vn_a.flags = flags & ~MAP_TYPE; 3397 vn_a.szc = 0; 3398 vn_a.lgrp_mem_policy_flags = 0; 3399 3400 error = as_map(as, *addrp, len, segvn_create, &vn_a); 3401 3402 as_rangeunlock(as); 3403 ZFS_EXIT(zfsvfs); 3404 return (error); 3405 } 3406 3407 /* ARGSUSED */ 3408 static int 3409 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 3410 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr) 3411 { 3412 uint64_t pages = btopr(len); 3413 3414 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages); 3415 return (0); 3416 } 3417 3418 /* 3419 * The reason we push dirty pages as part of zfs_delmap() is so that we get a 3420 * more accurate mtime for the associated file. Since we don't have a way of 3421 * detecting when the data was actually modified, we have to resort to 3422 * heuristics. If an explicit msync() is done, then we mark the mtime when the 3423 * last page is pushed. The problem occurs when the msync() call is omitted, 3424 * which by far the most common case: 3425 * 3426 * open() 3427 * mmap() 3428 * <modify memory> 3429 * munmap() 3430 * close() 3431 * <time lapse> 3432 * putpage() via fsflush 3433 * 3434 * If we wait until fsflush to come along, we can have a modification time that 3435 * is some arbitrary point in the future. In order to prevent this in the 3436 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is 3437 * torn down. 3438 */ 3439 /* ARGSUSED */ 3440 static int 3441 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 3442 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr) 3443 { 3444 uint64_t pages = btopr(len); 3445 3446 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages); 3447 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages); 3448 3449 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) && 3450 vn_has_cached_data(vp)) 3451 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr); 3452 3453 return (0); 3454 } 3455 3456 /* 3457 * Free or allocate space in a file. Currently, this function only 3458 * supports the `F_FREESP' command. However, this command is somewhat 3459 * misnamed, as its functionality includes the ability to allocate as 3460 * well as free space. 3461 * 3462 * IN: vp - vnode of file to free data in. 3463 * cmd - action to take (only F_FREESP supported). 3464 * bfp - section of file to free/alloc. 3465 * flag - current file open mode flags. 3466 * offset - current file offset. 3467 * cr - credentials of caller [UNUSED]. 3468 * 3469 * RETURN: 0 if success 3470 * error code if failure 3471 * 3472 * Timestamps: 3473 * vp - ctime|mtime updated 3474 */ 3475 /* ARGSUSED */ 3476 static int 3477 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag, 3478 offset_t offset, cred_t *cr, caller_context_t *ct) 3479 { 3480 znode_t *zp = VTOZ(vp); 3481 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3482 uint64_t off, len; 3483 int error; 3484 3485 ZFS_ENTER(zfsvfs); 3486 3487 top: 3488 if (cmd != F_FREESP) { 3489 ZFS_EXIT(zfsvfs); 3490 return (EINVAL); 3491 } 3492 3493 if (error = convoff(vp, bfp, 0, offset)) { 3494 ZFS_EXIT(zfsvfs); 3495 return (error); 3496 } 3497 3498 if (bfp->l_len < 0) { 3499 ZFS_EXIT(zfsvfs); 3500 return (EINVAL); 3501 } 3502 3503 off = bfp->l_start; 3504 len = bfp->l_len; /* 0 means from off to end of file */ 3505 3506 do { 3507 error = zfs_freesp(zp, off, len, flag, TRUE); 3508 /* NB: we already did dmu_tx_wait() if necessary */ 3509 } while (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT); 3510 3511 ZFS_EXIT(zfsvfs); 3512 return (error); 3513 } 3514 3515 static int 3516 zfs_fid(vnode_t *vp, fid_t *fidp) 3517 { 3518 znode_t *zp = VTOZ(vp); 3519 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3520 uint32_t gen = (uint32_t)zp->z_phys->zp_gen; 3521 uint64_t object = zp->z_id; 3522 zfid_short_t *zfid; 3523 int size, i; 3524 3525 ZFS_ENTER(zfsvfs); 3526 3527 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 3528 if (fidp->fid_len < size) { 3529 fidp->fid_len = size; 3530 ZFS_EXIT(zfsvfs); 3531 return (ENOSPC); 3532 } 3533 3534 zfid = (zfid_short_t *)fidp; 3535 3536 zfid->zf_len = size; 3537 3538 for (i = 0; i < sizeof (zfid->zf_object); i++) 3539 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 3540 3541 /* Must have a non-zero generation number to distinguish from .zfs */ 3542 if (gen == 0) 3543 gen = 1; 3544 for (i = 0; i < sizeof (zfid->zf_gen); i++) 3545 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 3546 3547 if (size == LONG_FID_LEN) { 3548 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 3549 zfid_long_t *zlfid; 3550 3551 zlfid = (zfid_long_t *)fidp; 3552 3553 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 3554 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 3555 3556 /* XXX - this should be the generation number for the objset */ 3557 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 3558 zlfid->zf_setgen[i] = 0; 3559 } 3560 3561 ZFS_EXIT(zfsvfs); 3562 return (0); 3563 } 3564 3565 static int 3566 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr) 3567 { 3568 znode_t *zp, *xzp; 3569 zfsvfs_t *zfsvfs; 3570 zfs_dirlock_t *dl; 3571 int error; 3572 3573 switch (cmd) { 3574 case _PC_LINK_MAX: 3575 *valp = ULONG_MAX; 3576 return (0); 3577 3578 case _PC_FILESIZEBITS: 3579 *valp = 64; 3580 return (0); 3581 3582 case _PC_XATTR_EXISTS: 3583 zp = VTOZ(vp); 3584 zfsvfs = zp->z_zfsvfs; 3585 ZFS_ENTER(zfsvfs); 3586 *valp = 0; 3587 error = zfs_dirent_lock(&dl, zp, "", &xzp, 3588 ZXATTR | ZEXISTS | ZSHARED); 3589 if (error == 0) { 3590 zfs_dirent_unlock(dl); 3591 if (!zfs_dirempty(xzp)) 3592 *valp = 1; 3593 VN_RELE(ZTOV(xzp)); 3594 } else if (error == ENOENT) { 3595 /* 3596 * If there aren't extended attributes, it's the 3597 * same as having zero of them. 3598 */ 3599 error = 0; 3600 } 3601 ZFS_EXIT(zfsvfs); 3602 return (error); 3603 3604 case _PC_ACL_ENABLED: 3605 *valp = _ACL_ACE_ENABLED; 3606 return (0); 3607 3608 case _PC_MIN_HOLE_SIZE: 3609 *valp = (ulong_t)SPA_MINBLOCKSIZE; 3610 return (0); 3611 3612 default: 3613 return (fs_pathconf(vp, cmd, valp, cr)); 3614 } 3615 } 3616 3617 /*ARGSUSED*/ 3618 static int 3619 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr) 3620 { 3621 znode_t *zp = VTOZ(vp); 3622 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3623 int error; 3624 3625 ZFS_ENTER(zfsvfs); 3626 error = zfs_getacl(zp, vsecp, cr); 3627 ZFS_EXIT(zfsvfs); 3628 3629 return (error); 3630 } 3631 3632 /*ARGSUSED*/ 3633 static int 3634 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr) 3635 { 3636 znode_t *zp = VTOZ(vp); 3637 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3638 int error; 3639 3640 ZFS_ENTER(zfsvfs); 3641 error = zfs_setacl(zp, vsecp, cr); 3642 ZFS_EXIT(zfsvfs); 3643 return (error); 3644 } 3645 3646 /* 3647 * Predeclare these here so that the compiler assumes that 3648 * this is an "old style" function declaration that does 3649 * not include arguments => we won't get type mismatch errors 3650 * in the initializations that follow. 3651 */ 3652 static int zfs_inval(); 3653 static int zfs_isdir(); 3654 3655 static int 3656 zfs_inval() 3657 { 3658 return (EINVAL); 3659 } 3660 3661 static int 3662 zfs_isdir() 3663 { 3664 return (EISDIR); 3665 } 3666 /* 3667 * Directory vnode operations template 3668 */ 3669 vnodeops_t *zfs_dvnodeops; 3670 const fs_operation_def_t zfs_dvnodeops_template[] = { 3671 VOPNAME_OPEN, { .vop_open = zfs_open }, 3672 VOPNAME_CLOSE, { .vop_close = zfs_close }, 3673 VOPNAME_READ, { .error = zfs_isdir }, 3674 VOPNAME_WRITE, { .error = zfs_isdir }, 3675 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 3676 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 3677 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 3678 VOPNAME_ACCESS, { .vop_access = zfs_access }, 3679 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 3680 VOPNAME_CREATE, { .vop_create = zfs_create }, 3681 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 3682 VOPNAME_LINK, { .vop_link = zfs_link }, 3683 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 3684 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir }, 3685 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 3686 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 3687 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink }, 3688 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 3689 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 3690 VOPNAME_FID, { .vop_fid = zfs_fid }, 3691 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 3692 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 3693 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 3694 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 3695 NULL, NULL 3696 }; 3697 3698 /* 3699 * Regular file vnode operations template 3700 */ 3701 vnodeops_t *zfs_fvnodeops; 3702 const fs_operation_def_t zfs_fvnodeops_template[] = { 3703 VOPNAME_OPEN, { .vop_open = zfs_open }, 3704 VOPNAME_CLOSE, { .vop_close = zfs_close }, 3705 VOPNAME_READ, { .vop_read = zfs_read }, 3706 VOPNAME_WRITE, { .vop_write = zfs_write }, 3707 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 3708 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 3709 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 3710 VOPNAME_ACCESS, { .vop_access = zfs_access }, 3711 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 3712 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 3713 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 3714 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 3715 VOPNAME_FID, { .vop_fid = zfs_fid }, 3716 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 3717 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock }, 3718 VOPNAME_SPACE, { .vop_space = zfs_space }, 3719 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage }, 3720 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage }, 3721 VOPNAME_MAP, { .vop_map = zfs_map }, 3722 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap }, 3723 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap }, 3724 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 3725 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 3726 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 3727 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 3728 NULL, NULL 3729 }; 3730 3731 /* 3732 * Symbolic link vnode operations template 3733 */ 3734 vnodeops_t *zfs_symvnodeops; 3735 const fs_operation_def_t zfs_symvnodeops_template[] = { 3736 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 3737 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 3738 VOPNAME_ACCESS, { .vop_access = zfs_access }, 3739 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 3740 VOPNAME_READLINK, { .vop_readlink = zfs_readlink }, 3741 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 3742 VOPNAME_FID, { .vop_fid = zfs_fid }, 3743 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 3744 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 3745 NULL, NULL 3746 }; 3747 3748 /* 3749 * Extended attribute directory vnode operations template 3750 * This template is identical to the directory vnodes 3751 * operation template except for restricted operations: 3752 * VOP_MKDIR() 3753 * VOP_SYMLINK() 3754 * Note that there are other restrictions embedded in: 3755 * zfs_create() - restrict type to VREG 3756 * zfs_link() - no links into/out of attribute space 3757 * zfs_rename() - no moves into/out of attribute space 3758 */ 3759 vnodeops_t *zfs_xdvnodeops; 3760 const fs_operation_def_t zfs_xdvnodeops_template[] = { 3761 VOPNAME_OPEN, { .vop_open = zfs_open }, 3762 VOPNAME_CLOSE, { .vop_close = zfs_close }, 3763 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 3764 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 3765 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 3766 VOPNAME_ACCESS, { .vop_access = zfs_access }, 3767 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 3768 VOPNAME_CREATE, { .vop_create = zfs_create }, 3769 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 3770 VOPNAME_LINK, { .vop_link = zfs_link }, 3771 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 3772 VOPNAME_MKDIR, { .error = zfs_inval }, 3773 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 3774 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 3775 VOPNAME_SYMLINK, { .error = zfs_inval }, 3776 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 3777 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 3778 VOPNAME_FID, { .vop_fid = zfs_fid }, 3779 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 3780 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 3781 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 3782 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 3783 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 3784 NULL, NULL 3785 }; 3786 3787 /* 3788 * Error vnode operations template 3789 */ 3790 vnodeops_t *zfs_evnodeops; 3791 const fs_operation_def_t zfs_evnodeops_template[] = { 3792 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 3793 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 3794 NULL, NULL 3795 }; 3796