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, Version 1.0 only 6 * (the "License"). You may not use this file except in compliance 7 * with the License. 8 * 9 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 10 * or http://www.opensolaris.org/os/licensing. 11 * See the License for the specific language governing permissions 12 * and limitations under the License. 13 * 14 * When distributing Covered Code, include this CDDL HEADER in each 15 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 16 * If applicable, add the following below this CDDL HEADER, with the 17 * fields enclosed by brackets "[]" replaced with your own identifying 18 * information: Portions Copyright [yyyy] [name of copyright owner] 19 * 20 * CDDL HEADER END 21 */ 22 /* 23 * Copyright 2006 Sun Microsystems, Inc. All rights reserved. 24 * Use is subject to license terms. 25 */ 26 27 #pragma ident "%Z%%M% %I% %E% SMI" 28 29 #include <sys/types.h> 30 #include <sys/param.h> 31 #include <sys/time.h> 32 #include <sys/systm.h> 33 #include <sys/sysmacros.h> 34 #include <sys/resource.h> 35 #include <sys/vfs.h> 36 #include <sys/vnode.h> 37 #include <sys/file.h> 38 #include <sys/stat.h> 39 #include <sys/kmem.h> 40 #include <sys/taskq.h> 41 #include <sys/uio.h> 42 #include <sys/vmsystm.h> 43 #include <sys/atomic.h> 44 #include <vm/seg_vn.h> 45 #include <vm/pvn.h> 46 #include <vm/as.h> 47 #include <sys/mman.h> 48 #include <sys/pathname.h> 49 #include <sys/cmn_err.h> 50 #include <sys/errno.h> 51 #include <sys/unistd.h> 52 #include <sys/zfs_vfsops.h> 53 #include <sys/zfs_dir.h> 54 #include <sys/zfs_acl.h> 55 #include <sys/zfs_ioctl.h> 56 #include <sys/fs/zfs.h> 57 #include <sys/dmu.h> 58 #include <sys/spa.h> 59 #include <sys/txg.h> 60 #include <sys/refcount.h> /* temporary for debugging purposes */ 61 #include <sys/dbuf.h> 62 #include <sys/zap.h> 63 #include <sys/dirent.h> 64 #include <sys/policy.h> 65 #include <sys/sunddi.h> 66 #include <sys/filio.h> 67 #include "fs/fs_subr.h" 68 #include <sys/zfs_ctldir.h> 69 70 /* 71 * Programming rules. 72 * 73 * Each vnode op performs some logical unit of work. To do this, the ZPL must 74 * properly lock its in-core state, create a DMU transaction, do the work, 75 * record this work in the intent log (ZIL), commit the DMU transaction, 76 * and wait the the intent log to commit if it's is a synchronous operation. 77 * Morover, the vnode ops must work in both normal and log replay context. 78 * The ordering of events is important to avoid deadlocks and references 79 * to freed memory. The example below illustrates the following Big Rules: 80 * 81 * (1) A check must be made in each zfs thread for a mounted file system. 82 * This is done avoiding races using ZFS_ENTER(zfsvfs). 83 * A ZFS_EXIT(zfsvfs) is needed before all returns. 84 * 85 * (2) VN_RELE() should always be the last thing except for zil_commit() 86 * and ZFS_EXIT(). This is for 3 reasons: 87 * First, if it's the last reference, the vnode/znode 88 * can be freed, so the zp may point to freed memory. Second, the last 89 * reference will call zfs_zinactive(), which may induce a lot of work -- 90 * pushing cached pages (which requires z_grow_lock) and syncing out 91 * cached atime changes. Third, zfs_zinactive() may require a new tx, 92 * which could deadlock the system if you were already holding one. 93 * 94 * (3) Always pass zfsvfs->z_assign as the second argument to dmu_tx_assign(). 95 * In normal operation, this will be TXG_NOWAIT. During ZIL replay, 96 * it will be a specific txg. Either way, dmu_tx_assign() never blocks. 97 * This is critical because we don't want to block while holding locks. 98 * Note, in particular, that if a lock is sometimes acquired before 99 * the tx assigns, and sometimes after (e.g. z_lock), then failing to 100 * use a non-blocking assign can deadlock the system. The scenario: 101 * 102 * Thread A has grabbed a lock before calling dmu_tx_assign(). 103 * Thread B is in an already-assigned tx, and blocks for this lock. 104 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open() 105 * forever, because the previous txg can't quiesce until B's tx commits. 106 * 107 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT, 108 * then drop all locks, call txg_wait_open(), and try again. 109 * 110 * (4) If the operation succeeded, generate the intent log entry for it 111 * before dropping locks. This ensures that the ordering of events 112 * in the intent log matches the order in which they actually occurred. 113 * 114 * (5) At the end of each vnode op, the DMU tx must always commit, 115 * regardless of whether there were any errors. 116 * 117 * (6) After dropping all locks, invoke zil_commit(zilog, seq, ioflag) 118 * to ensure that synchronous semantics are provided when necessary. 119 * 120 * In general, this is how things should be ordered in each vnode op: 121 * 122 * ZFS_ENTER(zfsvfs); // exit if unmounted 123 * top: 124 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD()) 125 * rw_enter(...); // grab any other locks you need 126 * tx = dmu_tx_create(...); // get DMU tx 127 * dmu_tx_hold_*(); // hold each object you might modify 128 * error = dmu_tx_assign(tx, zfsvfs->z_assign); // try to assign 129 * if (error) { 130 * dmu_tx_abort(tx); // abort DMU tx 131 * rw_exit(...); // drop locks 132 * zfs_dirent_unlock(dl); // unlock directory entry 133 * VN_RELE(...); // release held vnodes 134 * if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 135 * txg_wait_open(dmu_objset_pool(os), 0); 136 * goto top; 137 * } 138 * ZFS_EXIT(zfsvfs); // finished in zfs 139 * return (error); // really out of space 140 * } 141 * error = do_real_work(); // do whatever this VOP does 142 * if (error == 0) 143 * seq = zfs_log_*(...); // on success, make ZIL entry 144 * dmu_tx_commit(tx); // commit DMU tx -- error or not 145 * rw_exit(...); // drop locks 146 * zfs_dirent_unlock(dl); // unlock directory entry 147 * VN_RELE(...); // release held vnodes 148 * zil_commit(zilog, seq, ioflag); // synchronous when necessary 149 * ZFS_EXIT(zfsvfs); // finished in zfs 150 * return (error); // done, report error 151 */ 152 153 /* ARGSUSED */ 154 static int 155 zfs_open(vnode_t **vpp, int flag, cred_t *cr) 156 { 157 return (0); 158 } 159 160 /* ARGSUSED */ 161 static int 162 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr) 163 { 164 /* 165 * Clean up any locks held by this process on the vp. 166 */ 167 cleanlocks(vp, ddi_get_pid(), 0); 168 cleanshares(vp, ddi_get_pid()); 169 170 return (0); 171 } 172 173 /* 174 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and 175 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter. 176 */ 177 static int 178 zfs_holey(vnode_t *vp, int cmd, offset_t *off) 179 { 180 znode_t *zp = VTOZ(vp); 181 uint64_t noff = (uint64_t)*off; /* new offset */ 182 uint64_t file_sz; 183 int error; 184 boolean_t hole; 185 186 rw_enter(&zp->z_grow_lock, RW_READER); 187 file_sz = zp->z_phys->zp_size; 188 if (noff >= file_sz) { 189 rw_exit(&zp->z_grow_lock); 190 return (ENXIO); 191 } 192 193 if (cmd == _FIO_SEEK_HOLE) 194 hole = B_TRUE; 195 else 196 hole = B_FALSE; 197 198 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff); 199 rw_exit(&zp->z_grow_lock); 200 201 /* end of file? */ 202 if ((error == ESRCH) || (noff > file_sz)) { 203 /* 204 * Handle the virtual hole at the end of file. 205 */ 206 if (hole) { 207 *off = file_sz; 208 return (0); 209 } 210 return (ENXIO); 211 } 212 213 if (noff < *off) 214 return (error); 215 *off = noff; 216 return (error); 217 } 218 219 /* ARGSUSED */ 220 static int 221 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred, 222 int *rvalp) 223 { 224 offset_t off; 225 int error; 226 zfsvfs_t *zfsvfs; 227 228 switch (com) { 229 case _FIOFFS: 230 return (zfs_sync(vp->v_vfsp, 0, cred)); 231 232 case _FIO_SEEK_DATA: 233 case _FIO_SEEK_HOLE: 234 if (ddi_copyin((void *)data, &off, sizeof (off), flag)) 235 return (EFAULT); 236 237 zfsvfs = VTOZ(vp)->z_zfsvfs; 238 ZFS_ENTER(zfsvfs); 239 240 /* offset parameter is in/out */ 241 error = zfs_holey(vp, com, &off); 242 ZFS_EXIT(zfsvfs); 243 if (error) 244 return (error); 245 if (ddi_copyout(&off, (void *)data, sizeof (off), flag)) 246 return (EFAULT); 247 return (0); 248 } 249 return (ENOTTY); 250 } 251 252 /* 253 * When a file is memory mapped, we must keep the IO data synchronized 254 * between the DMU cache and the memory mapped pages. What this means: 255 * 256 * On Write: If we find a memory mapped page, we write to *both* 257 * the page and the dmu buffer. 258 * 259 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 260 * the file is memory mapped. 261 */ 262 static int 263 mappedwrite(vnode_t *vp, uint64_t woff, int nbytes, uio_t *uio, dmu_tx_t *tx) 264 { 265 znode_t *zp = VTOZ(vp); 266 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 267 int64_t start, off; 268 int len = nbytes; 269 int error = 0; 270 271 start = uio->uio_loffset; 272 off = start & PAGEOFFSET; 273 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 274 page_t *pp; 275 uint64_t bytes = MIN(PAGESIZE - off, len); 276 277 /* 278 * We don't want a new page to "appear" in the middle of 279 * the file update (because it may not get the write 280 * update data), so we grab a lock to block 281 * zfs_getpage(). 282 */ 283 rw_enter(&zp->z_map_lock, RW_WRITER); 284 if (pp = page_lookup(vp, start, SE_SHARED)) { 285 caddr_t va; 286 287 rw_exit(&zp->z_map_lock); 288 va = ppmapin(pp, PROT_READ | PROT_WRITE, (caddr_t)-1L); 289 error = uiomove(va+off, bytes, UIO_WRITE, uio); 290 if (error == 0) { 291 dmu_write(zfsvfs->z_os, zp->z_id, 292 woff, bytes, va+off, tx); 293 } 294 ppmapout(va); 295 page_unlock(pp); 296 } else { 297 error = dmu_write_uio(zfsvfs->z_os, zp->z_id, 298 woff, bytes, uio, tx); 299 rw_exit(&zp->z_map_lock); 300 } 301 len -= bytes; 302 woff += bytes; 303 off = 0; 304 if (error) 305 break; 306 } 307 return (error); 308 } 309 310 /* 311 * When a file is memory mapped, we must keep the IO data synchronized 312 * between the DMU cache and the memory mapped pages. What this means: 313 * 314 * On Read: We "read" preferentially from memory mapped pages, 315 * else we default from the dmu buffer. 316 * 317 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 318 * the file is memory mapped. 319 */ 320 static int 321 mappedread(vnode_t *vp, char *addr, int nbytes, uio_t *uio) 322 { 323 int64_t start, off, bytes; 324 int len = nbytes; 325 int error = 0; 326 327 start = uio->uio_loffset; 328 off = start & PAGEOFFSET; 329 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 330 page_t *pp; 331 332 bytes = MIN(PAGESIZE - off, len); 333 if (pp = page_lookup(vp, start, SE_SHARED)) { 334 caddr_t va; 335 336 va = ppmapin(pp, PROT_READ | PROT_WRITE, (caddr_t)-1L); 337 error = uiomove(va + off, bytes, UIO_READ, uio); 338 ppmapout(va); 339 page_unlock(pp); 340 } else { 341 /* XXX use dmu_read here? */ 342 error = uiomove(addr, bytes, UIO_READ, uio); 343 } 344 len -= bytes; 345 addr += bytes; 346 off = 0; 347 if (error) 348 break; 349 } 350 return (error); 351 } 352 353 uint_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ 354 355 /* 356 * Read bytes from specified file into supplied buffer. 357 * 358 * IN: vp - vnode of file to be read from. 359 * uio - structure supplying read location, range info, 360 * and return buffer. 361 * ioflag - SYNC flags; used to provide FRSYNC semantics. 362 * cr - credentials of caller. 363 * 364 * OUT: uio - updated offset and range, buffer filled. 365 * 366 * RETURN: 0 if success 367 * error code if failure 368 * 369 * Side Effects: 370 * vp - atime updated if byte count > 0 371 */ 372 /* ARGSUSED */ 373 static int 374 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 375 { 376 znode_t *zp = VTOZ(vp); 377 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 378 uint64_t delta; 379 ssize_t n, size, cnt, ndone; 380 int error, i, numbufs; 381 dmu_buf_t *dbp, **dbpp; 382 383 ZFS_ENTER(zfsvfs); 384 385 /* 386 * Validate file offset 387 */ 388 if (uio->uio_loffset < (offset_t)0) { 389 ZFS_EXIT(zfsvfs); 390 return (EINVAL); 391 } 392 393 /* 394 * Fasttrack empty reads 395 */ 396 if (uio->uio_resid == 0) { 397 ZFS_EXIT(zfsvfs); 398 return (0); 399 } 400 401 /* 402 * Check for region locks 403 */ 404 if (MANDMODE((mode_t)zp->z_phys->zp_mode)) { 405 if (error = chklock(vp, FREAD, 406 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { 407 ZFS_EXIT(zfsvfs); 408 return (error); 409 } 410 } 411 412 /* 413 * If we're in FRSYNC mode, sync out this znode before reading it. 414 */ 415 zil_commit(zfsvfs->z_log, zp->z_last_itx, ioflag & FRSYNC); 416 417 /* 418 * Make sure nobody restructures the file (changes block size) 419 * in the middle of the read. 420 */ 421 rw_enter(&zp->z_grow_lock, RW_READER); 422 /* 423 * If we are reading past end-of-file we can skip 424 * to the end; but we might still need to set atime. 425 */ 426 if (uio->uio_loffset >= zp->z_phys->zp_size) { 427 cnt = 0; 428 error = 0; 429 goto out; 430 } 431 432 cnt = MIN(uio->uio_resid, zp->z_phys->zp_size - uio->uio_loffset); 433 434 for (ndone = 0; ndone < cnt; ndone += zfs_read_chunk_size) { 435 ASSERT(uio->uio_loffset < zp->z_phys->zp_size); 436 n = MIN(zfs_read_chunk_size, 437 zp->z_phys->zp_size - uio->uio_loffset); 438 n = MIN(n, cnt); 439 dbpp = dmu_buf_hold_array(zfsvfs->z_os, zp->z_id, 440 uio->uio_loffset, n, &numbufs); 441 if (error = dmu_buf_read_array_canfail(dbpp, numbufs)) { 442 dmu_buf_rele_array(dbpp, numbufs); 443 goto out; 444 } 445 /* 446 * Compute the adjustment to align the dmu buffers 447 * with the uio buffer. 448 */ 449 delta = uio->uio_loffset - dbpp[0]->db_offset; 450 451 for (i = 0; i < numbufs; i++) { 452 if (n < 0) 453 break; 454 dbp = dbpp[i]; 455 size = dbp->db_size - delta; 456 /* 457 * XXX -- this is correct, but may be suboptimal. 458 * If the pages are all clean, we don't need to 459 * go through mappedread(). Maybe the VMODSORT 460 * stuff can help us here. 461 */ 462 if (vn_has_cached_data(vp)) { 463 error = mappedread(vp, (caddr_t)dbp->db_data + 464 delta, (n < size ? n : size), uio); 465 } else { 466 error = uiomove((caddr_t)dbp->db_data + delta, 467 (n < size ? n : size), UIO_READ, uio); 468 } 469 if (error) { 470 dmu_buf_rele_array(dbpp, numbufs); 471 goto out; 472 } 473 n -= dbp->db_size; 474 if (delta) { 475 n += delta; 476 delta = 0; 477 } 478 } 479 dmu_buf_rele_array(dbpp, numbufs); 480 } 481 out: 482 rw_exit(&zp->z_grow_lock); 483 484 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 485 ZFS_EXIT(zfsvfs); 486 return (error); 487 } 488 489 /* 490 * Fault in the pages of the first n bytes specified by the uio structure. 491 * 1 byte in each page is touched and the uio struct is unmodified. 492 * Any error will exit this routine as this is only a best 493 * attempt to get the pages resident. This is a copy of ufs_trans_touch(). 494 */ 495 static void 496 zfs_prefault_write(ssize_t n, struct uio *uio) 497 { 498 struct iovec *iov; 499 ulong_t cnt, incr; 500 caddr_t p; 501 uint8_t tmp; 502 503 iov = uio->uio_iov; 504 505 while (n) { 506 cnt = MIN(iov->iov_len, n); 507 if (cnt == 0) { 508 /* empty iov entry */ 509 iov++; 510 continue; 511 } 512 n -= cnt; 513 /* 514 * touch each page in this segment. 515 */ 516 p = iov->iov_base; 517 while (cnt) { 518 switch (uio->uio_segflg) { 519 case UIO_USERSPACE: 520 case UIO_USERISPACE: 521 if (fuword8(p, &tmp)) 522 return; 523 break; 524 case UIO_SYSSPACE: 525 if (kcopy(p, &tmp, 1)) 526 return; 527 break; 528 } 529 incr = MIN(cnt, PAGESIZE); 530 p += incr; 531 cnt -= incr; 532 } 533 /* 534 * touch the last byte in case it straddles a page. 535 */ 536 p--; 537 switch (uio->uio_segflg) { 538 case UIO_USERSPACE: 539 case UIO_USERISPACE: 540 if (fuword8(p, &tmp)) 541 return; 542 break; 543 case UIO_SYSSPACE: 544 if (kcopy(p, &tmp, 1)) 545 return; 546 break; 547 } 548 iov++; 549 } 550 } 551 552 /* 553 * Write the bytes to a file. 554 * 555 * IN: vp - vnode of file to be written to. 556 * uio - structure supplying write location, range info, 557 * and data buffer. 558 * ioflag - FAPPEND flag set if in append mode. 559 * cr - credentials of caller. 560 * 561 * OUT: uio - updated offset and range. 562 * 563 * RETURN: 0 if success 564 * error code if failure 565 * 566 * Timestamps: 567 * vp - ctime|mtime updated if byte count > 0 568 * 569 * Note: zfs_write() holds z_append_lock across calls to txg_wait_open(). 570 * It has to because of the semantics of FAPPEND. The implication is that 571 * we must never grab z_append_lock while in an assigned tx. 572 */ 573 /* ARGSUSED */ 574 static int 575 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 576 { 577 znode_t *zp = VTOZ(vp); 578 rlim64_t limit = uio->uio_llimit; 579 ssize_t start_resid = uio->uio_resid; 580 ssize_t tx_bytes; 581 uint64_t end_size; 582 dmu_tx_t *tx; 583 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 584 zilog_t *zilog = zfsvfs->z_log; 585 uint64_t seq = 0; 586 offset_t woff; 587 ssize_t n, nbytes; 588 int max_blksz = zfsvfs->z_max_blksz; 589 int need_append_lock, error; 590 krw_t grow_rw = RW_READER; 591 592 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 593 limit = MAXOFFSET_T; 594 595 n = start_resid; 596 597 /* 598 * Fasttrack empty write 599 */ 600 if (n == 0) 601 return (0); 602 603 ZFS_ENTER(zfsvfs); 604 605 /* 606 * Pre-fault the pages to ensure slow (eg NFS) pages don't hold up txg 607 */ 608 zfs_prefault_write(MIN(start_resid, SPA_MAXBLOCKSIZE), uio); 609 610 /* 611 * If in append mode, set the io offset pointer to eof. 612 */ 613 need_append_lock = ioflag & FAPPEND; 614 if (need_append_lock) { 615 rw_enter(&zp->z_append_lock, RW_WRITER); 616 woff = uio->uio_loffset = zp->z_phys->zp_size; 617 } else { 618 woff = uio->uio_loffset; 619 /* 620 * Validate file offset 621 */ 622 if (woff < 0) { 623 ZFS_EXIT(zfsvfs); 624 return (EINVAL); 625 } 626 627 /* 628 * If this write could change the file length, 629 * we need to synchronize with "appenders". 630 */ 631 if (woff < limit - n && woff + n > zp->z_phys->zp_size) { 632 need_append_lock = TRUE; 633 rw_enter(&zp->z_append_lock, RW_READER); 634 } 635 } 636 637 if (woff >= limit) { 638 error = EFBIG; 639 goto no_tx_done; 640 } 641 642 if ((woff + n) > limit || woff > (limit - n)) 643 n = limit - woff; 644 645 /* 646 * Check for region locks 647 */ 648 if (MANDMODE((mode_t)zp->z_phys->zp_mode) && 649 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) 650 goto no_tx_done; 651 top: 652 /* 653 * Make sure nobody restructures the file (changes block size) 654 * in the middle of the write. 655 */ 656 rw_enter(&zp->z_grow_lock, grow_rw); 657 658 end_size = MAX(zp->z_phys->zp_size, woff + n); 659 tx = dmu_tx_create(zfsvfs->z_os); 660 dmu_tx_hold_bonus(tx, zp->z_id); 661 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 662 error = dmu_tx_assign(tx, zfsvfs->z_assign); 663 if (error) { 664 dmu_tx_abort(tx); 665 rw_exit(&zp->z_grow_lock); 666 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 667 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 668 goto top; 669 } 670 goto no_tx_done; 671 } 672 673 if (end_size > zp->z_blksz && 674 (!ISP2(zp->z_blksz) || zp->z_blksz < max_blksz)) { 675 uint64_t new_blksz; 676 /* 677 * This write will increase the file size beyond 678 * the current block size so increase the block size. 679 */ 680 if (grow_rw == RW_READER && !rw_tryupgrade(&zp->z_grow_lock)) { 681 dmu_tx_commit(tx); 682 rw_exit(&zp->z_grow_lock); 683 grow_rw = RW_WRITER; 684 goto top; 685 } 686 if (zp->z_blksz > max_blksz) { 687 ASSERT(!ISP2(zp->z_blksz)); 688 new_blksz = MIN(end_size, SPA_MAXBLOCKSIZE); 689 } else { 690 new_blksz = MIN(end_size, max_blksz); 691 } 692 error = zfs_grow_blocksize(zp, new_blksz, tx); 693 if (error) { 694 tx_bytes = 0; 695 goto tx_done; 696 } 697 } 698 699 if (grow_rw == RW_WRITER) { 700 rw_downgrade(&zp->z_grow_lock); 701 grow_rw = RW_READER; 702 } 703 704 /* 705 * The file data does not fit in the znode "cache", so we 706 * will be writing to the file block data buffers. 707 * Each buffer will be written in a separate transaction; 708 * this keeps the intent log records small and allows us 709 * to do more fine-grained space accounting. 710 */ 711 while (n > 0) { 712 /* 713 * XXX - should we really limit each write to z_max_blksz? 714 * Perhaps we should use SPA_MAXBLOCKSIZE chunks? 715 */ 716 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz)); 717 rw_enter(&zp->z_map_lock, RW_READER); 718 719 tx_bytes = uio->uio_resid; 720 if (vn_has_cached_data(vp)) { 721 rw_exit(&zp->z_map_lock); 722 error = mappedwrite(vp, woff, nbytes, uio, tx); 723 } else { 724 error = dmu_write_uio(zfsvfs->z_os, zp->z_id, 725 woff, nbytes, uio, tx); 726 rw_exit(&zp->z_map_lock); 727 } 728 tx_bytes -= uio->uio_resid; 729 730 if (error) { 731 /* XXX - do we need to "clean up" the dmu buffer? */ 732 break; 733 } 734 735 ASSERT(tx_bytes == nbytes); 736 737 n -= nbytes; 738 if (n <= 0) 739 break; 740 741 /* 742 * We have more work ahead of us, so wrap up this transaction 743 * and start another. Exact same logic as tx_done below. 744 */ 745 while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset) { 746 dmu_buf_will_dirty(zp->z_dbuf, tx); 747 (void) atomic_cas_64(&zp->z_phys->zp_size, end_size, 748 uio->uio_loffset); 749 } 750 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 751 seq = zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, 752 ioflag, uio); 753 dmu_tx_commit(tx); 754 755 /* Pre-fault the next set of pages */ 756 zfs_prefault_write(MIN(n, SPA_MAXBLOCKSIZE), uio); 757 758 /* 759 * Start another transaction. 760 */ 761 woff = uio->uio_loffset; 762 tx = dmu_tx_create(zfsvfs->z_os); 763 dmu_tx_hold_bonus(tx, zp->z_id); 764 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 765 error = dmu_tx_assign(tx, zfsvfs->z_assign); 766 if (error) { 767 dmu_tx_abort(tx); 768 rw_exit(&zp->z_grow_lock); 769 if (error == ERESTART && 770 zfsvfs->z_assign == TXG_NOWAIT) { 771 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 772 goto top; 773 } 774 goto no_tx_done; 775 } 776 } 777 778 tx_done: 779 780 if (tx_bytes != 0) { 781 /* 782 * Update the file size if it has changed; account 783 * for possible concurrent updates. 784 */ 785 while ((end_size = zp->z_phys->zp_size) < uio->uio_loffset) { 786 dmu_buf_will_dirty(zp->z_dbuf, tx); 787 (void) atomic_cas_64(&zp->z_phys->zp_size, end_size, 788 uio->uio_loffset); 789 } 790 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 791 seq = zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, 792 ioflag, uio); 793 } 794 dmu_tx_commit(tx); 795 796 rw_exit(&zp->z_grow_lock); 797 798 no_tx_done: 799 800 if (need_append_lock) 801 rw_exit(&zp->z_append_lock); 802 803 /* 804 * If we're in replay mode, or we made no progress, return error. 805 * Otherwise, it's at least a partial write, so it's successful. 806 */ 807 if (zfsvfs->z_assign >= TXG_INITIAL || uio->uio_resid == start_resid) { 808 ZFS_EXIT(zfsvfs); 809 return (error); 810 } 811 812 zil_commit(zilog, seq, ioflag & (FSYNC | FDSYNC)); 813 814 ZFS_EXIT(zfsvfs); 815 return (0); 816 } 817 818 /* 819 * Get data to generate a TX_WRITE intent log record. 820 */ 821 int 822 zfs_get_data(void *arg, lr_write_t *lr) 823 { 824 zfsvfs_t *zfsvfs = arg; 825 objset_t *os = zfsvfs->z_os; 826 znode_t *zp; 827 uint64_t off = lr->lr_offset; 828 int dlen = lr->lr_length; /* length of user data */ 829 int reclen = lr->lr_common.lrc_reclen; 830 int error = 0; 831 832 ASSERT(dlen != 0); 833 834 /* 835 * Nothing to do if the file has been removed or truncated. 836 */ 837 if (zfs_zget(zfsvfs, lr->lr_foid, &zp) != 0) 838 return (ENOENT); 839 if (off >= zp->z_phys->zp_size || zp->z_reap) { 840 VN_RELE(ZTOV(zp)); 841 return (ENOENT); 842 } 843 844 /* 845 * Write records come in two flavors: immediate and indirect. 846 * For small writes it's cheaper to store the data with the 847 * log record (immediate); for large writes it's cheaper to 848 * sync the data and get a pointer to it (indirect) so that 849 * we don't have to write the data twice. 850 */ 851 if (sizeof (lr_write_t) + dlen <= reclen) { /* immediate write */ 852 rw_enter(&zp->z_grow_lock, RW_READER); 853 dmu_buf_t *db = dmu_buf_hold(os, lr->lr_foid, off); 854 dmu_buf_read(db); 855 bcopy((char *)db->db_data + off - db->db_offset, lr + 1, dlen); 856 dmu_buf_rele(db); 857 rw_exit(&zp->z_grow_lock); 858 } else { 859 /* 860 * We have to grab z_grow_lock as RW_WRITER because 861 * dmu_sync() can't handle concurrent dbuf_dirty() (6313856). 862 * z_grow_lock will be replaced with a range lock soon, 863 * which will eliminate the concurrency hit, but dmu_sync() 864 * really needs more thought. It shouldn't have to rely on 865 * the caller to provide MT safety. 866 */ 867 rw_enter(&zp->z_grow_lock, RW_WRITER); 868 txg_suspend(dmu_objset_pool(os)); 869 error = dmu_sync(os, lr->lr_foid, off, &lr->lr_blkoff, 870 &lr->lr_blkptr, lr->lr_common.lrc_txg); 871 txg_resume(dmu_objset_pool(os)); 872 rw_exit(&zp->z_grow_lock); 873 } 874 VN_RELE(ZTOV(zp)); 875 return (error); 876 } 877 878 /*ARGSUSED*/ 879 static int 880 zfs_access(vnode_t *vp, int mode, int flags, cred_t *cr) 881 { 882 znode_t *zp = VTOZ(vp); 883 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 884 int error; 885 886 ZFS_ENTER(zfsvfs); 887 error = zfs_zaccess_rwx(zp, mode, cr); 888 ZFS_EXIT(zfsvfs); 889 return (error); 890 } 891 892 /* 893 * Lookup an entry in a directory, or an extended attribute directory. 894 * If it exists, return a held vnode reference for it. 895 * 896 * IN: dvp - vnode of directory to search. 897 * nm - name of entry to lookup. 898 * pnp - full pathname to lookup [UNUSED]. 899 * flags - LOOKUP_XATTR set if looking for an attribute. 900 * rdir - root directory vnode [UNUSED]. 901 * cr - credentials of caller. 902 * 903 * OUT: vpp - vnode of located entry, NULL if not found. 904 * 905 * RETURN: 0 if success 906 * error code if failure 907 * 908 * Timestamps: 909 * NA 910 */ 911 /* ARGSUSED */ 912 static int 913 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp, 914 int flags, vnode_t *rdir, cred_t *cr) 915 { 916 917 znode_t *zdp = VTOZ(dvp); 918 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 919 int error; 920 921 ZFS_ENTER(zfsvfs); 922 923 *vpp = NULL; 924 925 if (flags & LOOKUP_XATTR) { 926 /* 927 * We don't allow recursive attributes.. 928 * Maybe someday we will. 929 */ 930 if (zdp->z_phys->zp_flags & ZFS_XATTR) { 931 ZFS_EXIT(zfsvfs); 932 return (EINVAL); 933 } 934 935 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr)) { 936 ZFS_EXIT(zfsvfs); 937 return (error); 938 } 939 940 /* 941 * Do we have permission to get into attribute directory? 942 */ 943 944 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, cr)) { 945 VN_RELE(*vpp); 946 } 947 948 ZFS_EXIT(zfsvfs); 949 return (error); 950 } 951 952 /* 953 * Check accessibility of directory. 954 */ 955 956 if (error = zfs_zaccess(zdp, ACE_EXECUTE, cr)) { 957 ZFS_EXIT(zfsvfs); 958 return (error); 959 } 960 961 if ((error = zfs_dirlook(zdp, nm, vpp)) == 0) { 962 963 /* 964 * Convert device special files 965 */ 966 if (IS_DEVVP(*vpp)) { 967 vnode_t *svp; 968 969 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 970 VN_RELE(*vpp); 971 if (svp == NULL) 972 error = ENOSYS; 973 else 974 *vpp = svp; 975 } 976 } 977 978 ZFS_EXIT(zfsvfs); 979 return (error); 980 } 981 982 /* 983 * Attempt to create a new entry in a directory. If the entry 984 * already exists, truncate the file if permissible, else return 985 * an error. Return the vp of the created or trunc'd file. 986 * 987 * IN: dvp - vnode of directory to put new file entry in. 988 * name - name of new file entry. 989 * vap - attributes of new file. 990 * excl - flag indicating exclusive or non-exclusive mode. 991 * mode - mode to open file with. 992 * cr - credentials of caller. 993 * flag - large file flag [UNUSED]. 994 * 995 * OUT: vpp - vnode of created or trunc'd entry. 996 * 997 * RETURN: 0 if success 998 * error code if failure 999 * 1000 * Timestamps: 1001 * dvp - ctime|mtime updated if new entry created 1002 * vp - ctime|mtime always, atime if new 1003 */ 1004 /* ARGSUSED */ 1005 static int 1006 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl, 1007 int mode, vnode_t **vpp, cred_t *cr, int flag) 1008 { 1009 znode_t *zp, *dzp = VTOZ(dvp); 1010 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1011 zilog_t *zilog = zfsvfs->z_log; 1012 uint64_t seq = 0; 1013 objset_t *os = zfsvfs->z_os; 1014 zfs_dirlock_t *dl; 1015 dmu_tx_t *tx; 1016 int error; 1017 uint64_t zoid; 1018 1019 ZFS_ENTER(zfsvfs); 1020 1021 top: 1022 *vpp = NULL; 1023 1024 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr)) 1025 vap->va_mode &= ~VSVTX; 1026 1027 if (*name == '\0') { 1028 /* 1029 * Null component name refers to the directory itself. 1030 */ 1031 VN_HOLD(dvp); 1032 zp = dzp; 1033 dl = NULL; 1034 error = 0; 1035 } else { 1036 /* possible VN_HOLD(zp) */ 1037 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, 0)) { 1038 if (strcmp(name, "..") == 0) 1039 error = EISDIR; 1040 ZFS_EXIT(zfsvfs); 1041 return (error); 1042 } 1043 } 1044 1045 zoid = zp ? zp->z_id : -1ULL; 1046 1047 if (zp == NULL) { 1048 /* 1049 * Create a new file object and update the directory 1050 * to reference it. 1051 */ 1052 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 1053 goto out; 1054 } 1055 1056 /* 1057 * We only support the creation of regular files in 1058 * extended attribute directories. 1059 */ 1060 if ((dzp->z_phys->zp_flags & ZFS_XATTR) && 1061 (vap->va_type != VREG)) { 1062 error = EINVAL; 1063 goto out; 1064 } 1065 1066 tx = dmu_tx_create(os); 1067 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 1068 dmu_tx_hold_bonus(tx, dzp->z_id); 1069 dmu_tx_hold_zap(tx, dzp->z_id, 1); 1070 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 1071 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1072 0, SPA_MAXBLOCKSIZE); 1073 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1074 if (error) { 1075 dmu_tx_abort(tx); 1076 zfs_dirent_unlock(dl); 1077 if (error == ERESTART && 1078 zfsvfs->z_assign == TXG_NOWAIT) { 1079 txg_wait_open(dmu_objset_pool(os), 0); 1080 goto top; 1081 } 1082 ZFS_EXIT(zfsvfs); 1083 return (error); 1084 } 1085 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 1086 ASSERT(zp->z_id == zoid); 1087 (void) zfs_link_create(dl, zp, tx, ZNEW); 1088 seq = zfs_log_create(zilog, tx, TX_CREATE, dzp, zp, name); 1089 dmu_tx_commit(tx); 1090 } else { 1091 /* 1092 * A directory entry already exists for this name. 1093 */ 1094 /* 1095 * Can't truncate an existing file if in exclusive mode. 1096 */ 1097 if (excl == EXCL) { 1098 error = EEXIST; 1099 goto out; 1100 } 1101 /* 1102 * Can't open a directory for writing. 1103 */ 1104 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1105 error = EISDIR; 1106 goto out; 1107 } 1108 /* 1109 * Verify requested access to file. 1110 */ 1111 if (mode && (error = zfs_zaccess_rwx(zp, mode, cr))) { 1112 goto out; 1113 } 1114 /* 1115 * Truncate regular files if requested. 1116 */ 1117 1118 /* 1119 * Need to update dzp->z_seq? 1120 */ 1121 1122 mutex_enter(&dzp->z_lock); 1123 dzp->z_seq++; 1124 mutex_exit(&dzp->z_lock); 1125 1126 if ((ZTOV(zp)->v_type == VREG) && (zp->z_phys->zp_size != 0) && 1127 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1128 /* 1129 * Truncate the file. 1130 */ 1131 tx = dmu_tx_create(os); 1132 dmu_tx_hold_bonus(tx, zoid); 1133 dmu_tx_hold_free(tx, zoid, 0, DMU_OBJECT_END); 1134 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1135 if (error) { 1136 dmu_tx_abort(tx); 1137 if (dl) 1138 zfs_dirent_unlock(dl); 1139 VN_RELE(ZTOV(zp)); 1140 if (error == ERESTART && 1141 zfsvfs->z_assign == TXG_NOWAIT) { 1142 txg_wait_open(dmu_objset_pool(os), 0); 1143 goto top; 1144 } 1145 ZFS_EXIT(zfsvfs); 1146 return (error); 1147 } 1148 /* 1149 * Grab the grow_lock to serialize this change with 1150 * respect to other file manipulations. 1151 */ 1152 rw_enter(&zp->z_grow_lock, RW_WRITER); 1153 error = zfs_freesp(zp, 0, 0, mode, tx, cr); 1154 if (error == 0) { 1155 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 1156 seq = zfs_log_truncate(zilog, tx, 1157 TX_TRUNCATE, zp, 0, 0); 1158 } 1159 rw_exit(&zp->z_grow_lock); 1160 dmu_tx_commit(tx); 1161 } 1162 } 1163 out: 1164 1165 if (dl) 1166 zfs_dirent_unlock(dl); 1167 1168 if (error) { 1169 if (zp) 1170 VN_RELE(ZTOV(zp)); 1171 } else { 1172 *vpp = ZTOV(zp); 1173 /* 1174 * If vnode is for a device return a specfs vnode instead. 1175 */ 1176 if (IS_DEVVP(*vpp)) { 1177 struct vnode *svp; 1178 1179 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1180 VN_RELE(*vpp); 1181 if (svp == NULL) { 1182 error = ENOSYS; 1183 } 1184 *vpp = svp; 1185 } 1186 } 1187 1188 zil_commit(zilog, seq, 0); 1189 1190 ZFS_EXIT(zfsvfs); 1191 return (error); 1192 } 1193 1194 /* 1195 * Remove an entry from a directory. 1196 * 1197 * IN: dvp - vnode of directory to remove entry from. 1198 * name - name of entry to remove. 1199 * cr - credentials of caller. 1200 * 1201 * RETURN: 0 if success 1202 * error code if failure 1203 * 1204 * Timestamps: 1205 * dvp - ctime|mtime 1206 * vp - ctime (if nlink > 0) 1207 */ 1208 static int 1209 zfs_remove(vnode_t *dvp, char *name, cred_t *cr) 1210 { 1211 znode_t *zp, *dzp = VTOZ(dvp); 1212 znode_t *xzp = NULL; 1213 vnode_t *vp; 1214 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1215 zilog_t *zilog = zfsvfs->z_log; 1216 uint64_t seq = 0; 1217 uint64_t acl_obj, xattr_obj; 1218 zfs_dirlock_t *dl; 1219 dmu_tx_t *tx; 1220 int may_delete_now, delete_now = FALSE; 1221 int reaped; 1222 int error; 1223 1224 ZFS_ENTER(zfsvfs); 1225 1226 top: 1227 /* 1228 * Attempt to lock directory; fail if entry doesn't exist. 1229 */ 1230 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) { 1231 ZFS_EXIT(zfsvfs); 1232 return (error); 1233 } 1234 1235 vp = ZTOV(zp); 1236 1237 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1238 goto out; 1239 } 1240 1241 /* 1242 * Need to use rmdir for removing directories. 1243 */ 1244 if (vp->v_type == VDIR) { 1245 error = EPERM; 1246 goto out; 1247 } 1248 1249 vnevent_remove(vp); 1250 1251 mutex_enter(&vp->v_lock); 1252 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp); 1253 mutex_exit(&vp->v_lock); 1254 1255 /* 1256 * We may delete the znode now, or we may put it on the delete queue; 1257 * it depends on whether we're the last link, and on whether there are 1258 * other holds on the vnode. So we dmu_tx_hold() the right things to 1259 * allow for either case. 1260 */ 1261 tx = dmu_tx_create(zfsvfs->z_os); 1262 dmu_tx_hold_zap(tx, dzp->z_id, -1); 1263 dmu_tx_hold_bonus(tx, zp->z_id); 1264 if (may_delete_now) 1265 dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END); 1266 1267 /* are there any extended attributes? */ 1268 if ((xattr_obj = zp->z_phys->zp_xattr) != 0) { 1269 /* 1270 * XXX - There is a possibility that the delete 1271 * of the parent file could succeed, but then we get 1272 * an ENOSPC when we try to delete the xattrs... 1273 * so we would need to re-try the deletes periodically 1274 */ 1275 /* XXX - do we need this if we are deleting? */ 1276 dmu_tx_hold_bonus(tx, xattr_obj); 1277 } 1278 1279 /* are there any additional acls */ 1280 if ((acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj) != 0 && 1281 may_delete_now) 1282 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 1283 1284 /* charge as an update -- would be nice not to charge at all */ 1285 dmu_tx_hold_zap(tx, zfsvfs->z_dqueue, -1); 1286 1287 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1288 if (error) { 1289 dmu_tx_abort(tx); 1290 zfs_dirent_unlock(dl); 1291 VN_RELE(vp); 1292 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1293 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 1294 goto top; 1295 } 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, &reaped); 1304 1305 if (error) { 1306 dmu_tx_commit(tx); 1307 goto out; 1308 } 1309 1310 if (reaped) { 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_reap = 1; 1327 xzp->z_phys->zp_links = 0; 1328 mutex_exit(&xzp->z_lock); 1329 zfs_dq_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 zp->z_active = 0; 1338 mutex_exit(&zp->z_lock); 1339 zfs_znode_delete(zp, tx); 1340 VFS_RELE(zfsvfs->z_vfs); 1341 } else if (reaped) { 1342 zfs_dq_add(zp, tx); 1343 } 1344 1345 seq = zfs_log_remove(zilog, tx, TX_REMOVE, dzp, name); 1346 1347 dmu_tx_commit(tx); 1348 out: 1349 zfs_dirent_unlock(dl); 1350 1351 if (!delete_now) { 1352 VN_RELE(vp); 1353 } else if (xzp) { 1354 /* this rele delayed to prevent nesting transactions */ 1355 VN_RELE(ZTOV(xzp)); 1356 } 1357 1358 zil_commit(zilog, seq, 0); 1359 1360 ZFS_EXIT(zfsvfs); 1361 return (error); 1362 } 1363 1364 /* 1365 * Create a new directory and insert it into dvp using the name 1366 * provided. Return a pointer to the inserted directory. 1367 * 1368 * IN: dvp - vnode of directory to add subdir to. 1369 * dirname - name of new directory. 1370 * vap - attributes of new directory. 1371 * cr - credentials of caller. 1372 * 1373 * OUT: vpp - vnode of created directory. 1374 * 1375 * RETURN: 0 if success 1376 * error code if failure 1377 * 1378 * Timestamps: 1379 * dvp - ctime|mtime updated 1380 * vp - ctime|mtime|atime updated 1381 */ 1382 static int 1383 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr) 1384 { 1385 znode_t *zp, *dzp = VTOZ(dvp); 1386 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1387 zilog_t *zilog = zfsvfs->z_log; 1388 uint64_t seq = 0; 1389 zfs_dirlock_t *dl; 1390 uint64_t zoid = 0; 1391 dmu_tx_t *tx; 1392 int error; 1393 1394 ASSERT(vap->va_type == VDIR); 1395 1396 ZFS_ENTER(zfsvfs); 1397 1398 if (dzp->z_phys->zp_flags & ZFS_XATTR) { 1399 ZFS_EXIT(zfsvfs); 1400 return (EINVAL); 1401 } 1402 top: 1403 *vpp = NULL; 1404 1405 /* 1406 * First make sure the new directory doesn't exist. 1407 */ 1408 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, ZNEW)) { 1409 ZFS_EXIT(zfsvfs); 1410 return (error); 1411 } 1412 1413 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, cr)) { 1414 zfs_dirent_unlock(dl); 1415 ZFS_EXIT(zfsvfs); 1416 return (error); 1417 } 1418 1419 /* 1420 * Add a new entry to the directory. 1421 */ 1422 tx = dmu_tx_create(zfsvfs->z_os); 1423 dmu_tx_hold_zap(tx, dzp->z_id, 1); 1424 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, 0); 1425 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 1426 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1427 0, SPA_MAXBLOCKSIZE); 1428 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1429 if (error) { 1430 dmu_tx_abort(tx); 1431 zfs_dirent_unlock(dl); 1432 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1433 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 1434 goto top; 1435 } 1436 ZFS_EXIT(zfsvfs); 1437 return (error); 1438 } 1439 1440 /* 1441 * Create new node. 1442 */ 1443 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 1444 1445 /* 1446 * Now put new name in parent dir. 1447 */ 1448 (void) zfs_link_create(dl, zp, tx, ZNEW); 1449 1450 *vpp = ZTOV(zp); 1451 1452 seq = zfs_log_create(zilog, tx, TX_MKDIR, dzp, zp, dirname); 1453 dmu_tx_commit(tx); 1454 1455 zfs_dirent_unlock(dl); 1456 1457 zil_commit(zilog, seq, 0); 1458 1459 ZFS_EXIT(zfsvfs); 1460 return (0); 1461 } 1462 1463 /* 1464 * Remove a directory subdir entry. If the current working 1465 * directory is the same as the subdir to be removed, the 1466 * remove will fail. 1467 * 1468 * IN: dvp - vnode of directory to remove from. 1469 * name - name of directory to be removed. 1470 * cwd - vnode of current working directory. 1471 * cr - credentials of caller. 1472 * 1473 * RETURN: 0 if success 1474 * error code if failure 1475 * 1476 * Timestamps: 1477 * dvp - ctime|mtime updated 1478 */ 1479 static int 1480 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr) 1481 { 1482 znode_t *dzp = VTOZ(dvp); 1483 znode_t *zp; 1484 vnode_t *vp; 1485 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1486 zilog_t *zilog = zfsvfs->z_log; 1487 uint64_t seq = 0; 1488 zfs_dirlock_t *dl; 1489 dmu_tx_t *tx; 1490 int error; 1491 1492 ZFS_ENTER(zfsvfs); 1493 1494 top: 1495 zp = NULL; 1496 1497 /* 1498 * Attempt to lock directory; fail if entry doesn't exist. 1499 */ 1500 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZEXISTS)) { 1501 ZFS_EXIT(zfsvfs); 1502 return (error); 1503 } 1504 1505 vp = ZTOV(zp); 1506 1507 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1508 goto out; 1509 } 1510 1511 if (vp->v_type != VDIR) { 1512 error = ENOTDIR; 1513 goto out; 1514 } 1515 1516 if (vp == cwd) { 1517 error = EINVAL; 1518 goto out; 1519 } 1520 1521 vnevent_rmdir(vp); 1522 1523 /* 1524 * Grab a lock on the parent pointer 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, 1); 1531 dmu_tx_hold_bonus(tx, zp->z_id); 1532 dmu_tx_hold_zap(tx, zfsvfs->z_dqueue, 1); 1533 error = dmu_tx_assign(tx, zfsvfs->z_assign); 1534 if (error) { 1535 dmu_tx_abort(tx); 1536 rw_exit(&zp->z_parent_lock); 1537 zfs_dirent_unlock(dl); 1538 VN_RELE(vp); 1539 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 1540 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 1541 goto top; 1542 } 1543 ZFS_EXIT(zfsvfs); 1544 return (error); 1545 } 1546 1547 error = zfs_link_destroy(dl, zp, tx, 0, NULL); 1548 1549 if (error == 0) 1550 seq = zfs_log_remove(zilog, tx, TX_RMDIR, dzp, name); 1551 1552 dmu_tx_commit(tx); 1553 1554 rw_exit(&zp->z_parent_lock); 1555 out: 1556 zfs_dirent_unlock(dl); 1557 1558 VN_RELE(vp); 1559 1560 zil_commit(zilog, seq, 0); 1561 1562 ZFS_EXIT(zfsvfs); 1563 return (error); 1564 } 1565 1566 /* 1567 * Read as many directory entries as will fit into the provided 1568 * buffer from the given directory cursor position (specified in 1569 * the uio structure. 1570 * 1571 * IN: vp - vnode of directory to read. 1572 * uio - structure supplying read location, range info, 1573 * and return buffer. 1574 * cr - credentials of caller. 1575 * 1576 * OUT: uio - updated offset and range, buffer filled. 1577 * eofp - set to true if end-of-file detected. 1578 * 1579 * RETURN: 0 if success 1580 * error code if failure 1581 * 1582 * Timestamps: 1583 * vp - atime updated 1584 * 1585 * Note that the low 4 bits of the cookie returned by zap is always zero. 1586 * This allows us to use the low range for "special" directory entries: 1587 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 1588 * we use the offset 2 for the '.zfs' directory. 1589 */ 1590 /* ARGSUSED */ 1591 static int 1592 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp) 1593 { 1594 znode_t *zp = VTOZ(vp); 1595 iovec_t *iovp; 1596 dirent64_t *odp; 1597 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1598 objset_t *os; 1599 caddr_t outbuf; 1600 size_t bufsize; 1601 zap_cursor_t zc; 1602 zap_attribute_t zap; 1603 uint_t bytes_wanted; 1604 ushort_t this_reclen; 1605 uint64_t offset; /* must be unsigned; checks for < 1 */ 1606 off64_t *next; 1607 int local_eof; 1608 int outcount; 1609 int error; 1610 uint8_t prefetch; 1611 1612 ZFS_ENTER(zfsvfs); 1613 1614 /* 1615 * If we are not given an eof variable, 1616 * use a local one. 1617 */ 1618 if (eofp == NULL) 1619 eofp = &local_eof; 1620 1621 /* 1622 * Check for valid iov_len. 1623 */ 1624 if (uio->uio_iov->iov_len <= 0) { 1625 ZFS_EXIT(zfsvfs); 1626 return (EINVAL); 1627 } 1628 1629 /* 1630 * Quit if directory has been removed (posix) 1631 */ 1632 if ((*eofp = zp->z_reap) != 0) { 1633 ZFS_EXIT(zfsvfs); 1634 return (0); 1635 } 1636 1637 error = 0; 1638 os = zfsvfs->z_os; 1639 offset = uio->uio_loffset; 1640 prefetch = zp->z_zn_prefetch; 1641 1642 /* 1643 * Initialize the iterator cursor. 1644 */ 1645 if (offset <= 3) { 1646 /* 1647 * Start iteration from the beginning of the directory. 1648 */ 1649 zap_cursor_init(&zc, os, zp->z_id); 1650 } else { 1651 /* 1652 * The offset is a serialized cursor. 1653 */ 1654 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 1655 } 1656 1657 /* 1658 * Get space to change directory entries into fs independent format. 1659 */ 1660 iovp = uio->uio_iov; 1661 bytes_wanted = iovp->iov_len; 1662 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 1663 bufsize = bytes_wanted; 1664 outbuf = kmem_alloc(bufsize, KM_SLEEP); 1665 odp = (struct dirent64 *)outbuf; 1666 } else { 1667 bufsize = bytes_wanted; 1668 odp = (struct dirent64 *)iovp->iov_base; 1669 } 1670 1671 /* 1672 * Transform to file-system independent format 1673 */ 1674 outcount = 0; 1675 while (outcount < bytes_wanted) { 1676 /* 1677 * Special case `.', `..', and `.zfs'. 1678 */ 1679 if (offset == 0) { 1680 (void) strcpy(zap.za_name, "."); 1681 zap.za_first_integer = zp->z_id; 1682 this_reclen = DIRENT64_RECLEN(1); 1683 } else if (offset == 1) { 1684 (void) strcpy(zap.za_name, ".."); 1685 zap.za_first_integer = zp->z_phys->zp_parent; 1686 this_reclen = DIRENT64_RECLEN(2); 1687 } else if (offset == 2 && zfs_show_ctldir(zp)) { 1688 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 1689 zap.za_first_integer = ZFSCTL_INO_ROOT; 1690 this_reclen = 1691 DIRENT64_RECLEN(sizeof (ZFS_CTLDIR_NAME) - 1); 1692 } else { 1693 /* 1694 * Grab next entry. 1695 */ 1696 if (error = zap_cursor_retrieve(&zc, &zap)) { 1697 if ((*eofp = (error == ENOENT)) != 0) 1698 break; 1699 else 1700 goto update; 1701 } 1702 1703 if (zap.za_integer_length != 8 || 1704 zap.za_num_integers != 1) { 1705 cmn_err(CE_WARN, "zap_readdir: bad directory " 1706 "entry, obj = %lld, offset = %lld\n", 1707 (u_longlong_t)zp->z_id, 1708 (u_longlong_t)offset); 1709 error = ENXIO; 1710 goto update; 1711 } 1712 this_reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 1713 } 1714 1715 /* 1716 * Will this entry fit in the buffer? 1717 */ 1718 if (outcount + this_reclen > bufsize) { 1719 /* 1720 * Did we manage to fit anything in the buffer? 1721 */ 1722 if (!outcount) { 1723 error = EINVAL; 1724 goto update; 1725 } 1726 break; 1727 } 1728 /* 1729 * Add this entry: 1730 */ 1731 odp->d_ino = (ino64_t)zap.za_first_integer; 1732 odp->d_reclen = (ushort_t)this_reclen; 1733 /* NOTE: d_off is the offset for the *next* entry */ 1734 next = &(odp->d_off); 1735 (void) strncpy(odp->d_name, zap.za_name, 1736 DIRENT64_NAMELEN(this_reclen)); 1737 outcount += this_reclen; 1738 odp = (dirent64_t *)((intptr_t)odp + this_reclen); 1739 1740 ASSERT(outcount <= bufsize); 1741 1742 /* Prefetch znode */ 1743 if (prefetch) 1744 dmu_prefetch(os, zap.za_first_integer, 0, 0); 1745 1746 /* 1747 * Move to the next entry, fill in the previous offset. 1748 */ 1749 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 1750 zap_cursor_advance(&zc); 1751 offset = zap_cursor_serialize(&zc); 1752 } else { 1753 offset += 1; 1754 } 1755 *next = offset; 1756 } 1757 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 1758 1759 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 1760 iovp->iov_base += outcount; 1761 iovp->iov_len -= outcount; 1762 uio->uio_resid -= outcount; 1763 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 1764 /* 1765 * Reset the pointer. 1766 */ 1767 offset = uio->uio_loffset; 1768 } 1769 1770 update: 1771 zap_cursor_fini(&zc); 1772 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 1773 kmem_free(outbuf, bufsize); 1774 1775 if (error == ENOENT) 1776 error = 0; 1777 1778 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 1779 1780 uio->uio_loffset = offset; 1781 ZFS_EXIT(zfsvfs); 1782 return (error); 1783 } 1784 1785 /* ARGSUSED */ 1786 static int 1787 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr) 1788 { 1789 znode_t *zp = VTOZ(vp); 1790 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1791 1792 ZFS_ENTER(zfsvfs); 1793 zil_commit(zfsvfs->z_log, zp->z_last_itx, FSYNC); 1794 ZFS_EXIT(zfsvfs); 1795 return (0); 1796 } 1797 1798 /* 1799 * Get the requested file attributes and place them in the provided 1800 * vattr structure. 1801 * 1802 * IN: vp - vnode of file. 1803 * vap - va_mask identifies requested attributes. 1804 * flags - [UNUSED] 1805 * cr - credentials of caller. 1806 * 1807 * OUT: vap - attribute values. 1808 * 1809 * RETURN: 0 (always succeeds) 1810 */ 1811 /* ARGSUSED */ 1812 static int 1813 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr) 1814 { 1815 znode_t *zp = VTOZ(vp); 1816 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1817 znode_phys_t *pzp = zp->z_phys; 1818 int error; 1819 1820 ZFS_ENTER(zfsvfs); 1821 1822 /* 1823 * Return all attributes. It's cheaper to provide the answer 1824 * than to determine whether we were asked the question. 1825 */ 1826 mutex_enter(&zp->z_lock); 1827 1828 vap->va_type = vp->v_type; 1829 vap->va_mode = pzp->zp_mode & MODEMASK; 1830 vap->va_uid = zp->z_phys->zp_uid; 1831 vap->va_gid = zp->z_phys->zp_gid; 1832 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev; 1833 vap->va_nodeid = zp->z_id; 1834 vap->va_nlink = MIN(pzp->zp_links, UINT32_MAX); /* nlink_t limit! */ 1835 vap->va_size = pzp->zp_size; 1836 vap->va_rdev = pzp->zp_rdev; 1837 vap->va_seq = zp->z_seq; 1838 1839 ZFS_TIME_DECODE(&vap->va_atime, pzp->zp_atime); 1840 ZFS_TIME_DECODE(&vap->va_mtime, pzp->zp_mtime); 1841 ZFS_TIME_DECODE(&vap->va_ctime, pzp->zp_ctime); 1842 1843 /* 1844 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 1845 * Also, if we are the owner don't bother, since owner should 1846 * always be allowed to read basic attributes of file. 1847 */ 1848 if (!(zp->z_phys->zp_flags & ZFS_ACL_TRIVIAL) && 1849 (zp->z_phys->zp_uid != crgetuid(cr))) { 1850 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, cr)) { 1851 mutex_exit(&zp->z_lock); 1852 ZFS_EXIT(zfsvfs); 1853 return (error); 1854 } 1855 } 1856 1857 mutex_exit(&zp->z_lock); 1858 1859 dmu_object_size_from_db(zp->z_dbuf, &vap->va_blksize, &vap->va_nblocks); 1860 1861 if (zp->z_blksz == 0) { 1862 /* 1863 * Block size hasn't been set; suggest maximal I/O transfers. 1864 */ 1865 vap->va_blksize = zfsvfs->z_max_blksz; 1866 } 1867 1868 ZFS_EXIT(zfsvfs); 1869 return (0); 1870 } 1871 1872 /* 1873 * Set the file attributes to the values contained in the 1874 * vattr structure. 1875 * 1876 * IN: vp - vnode of file to be modified. 1877 * vap - new attribute values. 1878 * flags - ATTR_UTIME set if non-default time values provided. 1879 * cr - credentials of caller. 1880 * 1881 * RETURN: 0 if success 1882 * error code if failure 1883 * 1884 * Timestamps: 1885 * vp - ctime updated, mtime updated if size changed. 1886 */ 1887 /* ARGSUSED */ 1888 static int 1889 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 1890 caller_context_t *ct) 1891 { 1892 struct znode *zp = VTOZ(vp); 1893 znode_phys_t *pzp = zp->z_phys; 1894 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1895 zilog_t *zilog = zfsvfs->z_log; 1896 uint64_t seq = 0; 1897 dmu_tx_t *tx; 1898 uint_t mask = vap->va_mask; 1899 uint_t mask_applied = 0; 1900 vattr_t oldva; 1901 int trim_mask = FALSE; 1902 int saved_mask; 1903 uint64_t new_mode; 1904 znode_t *attrzp; 1905 int have_grow_lock; 1906 int need_policy = FALSE; 1907 int err; 1908 1909 if (mask == 0) 1910 return (0); 1911 1912 if (mask & AT_NOSET) 1913 return (EINVAL); 1914 1915 if (mask & AT_SIZE && vp->v_type == VDIR) 1916 return (EISDIR); 1917 1918 if (mask & AT_SIZE && vp->v_type != VREG) 1919 return (EINVAL); 1920 1921 ZFS_ENTER(zfsvfs); 1922 1923 top: 1924 have_grow_lock = FALSE; 1925 attrzp = NULL; 1926 1927 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 1928 ZFS_EXIT(zfsvfs); 1929 return (EROFS); 1930 } 1931 1932 /* 1933 * First validate permissions 1934 */ 1935 1936 if (mask & AT_SIZE) { 1937 err = zfs_zaccess(zp, ACE_WRITE_DATA, cr); 1938 if (err) { 1939 ZFS_EXIT(zfsvfs); 1940 return (err); 1941 } 1942 } 1943 1944 if (mask & (AT_ATIME|AT_MTIME)) 1945 need_policy = zfs_zaccess_v4_perm(zp, ACE_WRITE_ATTRIBUTES, cr); 1946 1947 if (mask & (AT_UID|AT_GID)) { 1948 int idmask = (mask & (AT_UID|AT_GID)); 1949 int take_owner; 1950 int take_group; 1951 1952 /* 1953 * NOTE: even if a new mode is being set, 1954 * we may clear S_ISUID/S_ISGID bits. 1955 */ 1956 1957 if (!(mask & AT_MODE)) 1958 vap->va_mode = pzp->zp_mode; 1959 1960 /* 1961 * Take ownership or chgrp to group we are a member of 1962 */ 1963 1964 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 1965 take_group = (mask & AT_GID) && groupmember(vap->va_gid, cr); 1966 1967 /* 1968 * If both AT_UID and AT_GID are set then take_owner and 1969 * take_group must both be set in order to allow taking 1970 * ownership. 1971 * 1972 * Otherwise, send the check through secpolicy_vnode_setattr() 1973 * 1974 */ 1975 1976 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 1977 ((idmask == AT_UID) && take_owner) || 1978 ((idmask == AT_GID) && take_group)) { 1979 if (zfs_zaccess_v4_perm(zp, ACE_WRITE_OWNER, cr) == 0) { 1980 /* 1981 * Remove setuid/setgid for non-privileged users 1982 */ 1983 secpolicy_setid_clear(vap, cr); 1984 trim_mask = TRUE; 1985 saved_mask = vap->va_mask; 1986 } else { 1987 need_policy = TRUE; 1988 } 1989 } else { 1990 need_policy = TRUE; 1991 } 1992 } 1993 1994 if (mask & AT_MODE) 1995 need_policy = TRUE; 1996 1997 if (need_policy) { 1998 mutex_enter(&zp->z_lock); 1999 oldva.va_mode = pzp->zp_mode; 2000 oldva.va_uid = zp->z_phys->zp_uid; 2001 oldva.va_gid = zp->z_phys->zp_gid; 2002 mutex_exit(&zp->z_lock); 2003 2004 /* 2005 * If trim_mask is set then take ownership 2006 * has been granted. In that case remove 2007 * UID|GID from mask so that 2008 * secpolicy_vnode_setattr() doesn't revoke it. 2009 */ 2010 if (trim_mask) 2011 vap->va_mask &= ~(AT_UID|AT_GID); 2012 2013 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 2014 (int (*)(void *, int, cred_t *))zfs_zaccess_rwx, zp); 2015 if (err) { 2016 ZFS_EXIT(zfsvfs); 2017 return (err); 2018 } 2019 2020 if (trim_mask) 2021 vap->va_mask |= (saved_mask & (AT_UID|AT_GID)); 2022 } 2023 2024 /* 2025 * secpolicy_vnode_setattr, or take ownership may have 2026 * changed va_mask 2027 */ 2028 mask = vap->va_mask; 2029 2030 tx = dmu_tx_create(zfsvfs->z_os); 2031 dmu_tx_hold_bonus(tx, zp->z_id); 2032 2033 if (mask & AT_MODE) { 2034 2035 new_mode = (pzp->zp_mode & S_IFMT) | (vap->va_mode & ~S_IFMT); 2036 2037 if (zp->z_phys->zp_acl.z_acl_extern_obj) 2038 dmu_tx_hold_write(tx, 2039 pzp->zp_acl.z_acl_extern_obj, 0, SPA_MAXBLOCKSIZE); 2040 else 2041 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2042 0, ZFS_ACL_SIZE(MAX_ACL_SIZE)); 2043 } 2044 2045 if (mask & AT_SIZE) { 2046 uint64_t off = vap->va_size; 2047 /* 2048 * Grab the grow_lock to serialize this change with 2049 * respect to other file manipulations. 2050 */ 2051 rw_enter(&zp->z_grow_lock, RW_WRITER); 2052 have_grow_lock = TRUE; 2053 if (off < zp->z_phys->zp_size) 2054 dmu_tx_hold_free(tx, zp->z_id, off, DMU_OBJECT_END); 2055 else if (zp->z_phys->zp_size && 2056 zp->z_blksz < zfsvfs->z_max_blksz && off > zp->z_blksz) 2057 /* we will rewrite this block if we grow */ 2058 dmu_tx_hold_write(tx, zp->z_id, 0, zp->z_phys->zp_size); 2059 } 2060 2061 if ((mask & (AT_UID | AT_GID)) && zp->z_phys->zp_xattr != 0) { 2062 err = zfs_zget(zp->z_zfsvfs, zp->z_phys->zp_xattr, &attrzp); 2063 if (err) { 2064 dmu_tx_abort(tx); 2065 if (have_grow_lock) 2066 rw_exit(&zp->z_grow_lock); 2067 ZFS_EXIT(zfsvfs); 2068 return (err); 2069 } 2070 dmu_tx_hold_bonus(tx, attrzp->z_id); 2071 } 2072 2073 err = dmu_tx_assign(tx, zfsvfs->z_assign); 2074 if (err) { 2075 if (attrzp) 2076 VN_RELE(ZTOV(attrzp)); 2077 dmu_tx_abort(tx); 2078 if (have_grow_lock) 2079 rw_exit(&zp->z_grow_lock); 2080 if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2081 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 2082 goto top; 2083 } 2084 ZFS_EXIT(zfsvfs); 2085 return (err); 2086 } 2087 2088 dmu_buf_will_dirty(zp->z_dbuf, tx); 2089 2090 /* 2091 * Set each attribute requested. 2092 * We group settings according to the locks they need to acquire. 2093 * 2094 * Note: you cannot set ctime directly, although it will be 2095 * updated as a side-effect of calling this function. 2096 */ 2097 if (mask & AT_SIZE) { 2098 /* 2099 * XXX - Note, we are not providing any open 2100 * mode flags here (like FNDELAY), so we may 2101 * block if there are locks present... this 2102 * should be addressed in openat(). 2103 */ 2104 err = zfs_freesp(zp, vap->va_size, 0, 0, tx, cr); 2105 if (err) { 2106 mutex_enter(&zp->z_lock); 2107 goto out; 2108 } 2109 mask_applied |= AT_SIZE; 2110 } 2111 2112 mask_applied = mask; /* no errors after this point */ 2113 2114 mutex_enter(&zp->z_lock); 2115 2116 if (mask & AT_MODE) { 2117 err = zfs_acl_chmod_setattr(zp, new_mode, tx); 2118 ASSERT3U(err, ==, 0); 2119 } 2120 2121 if (attrzp) 2122 mutex_enter(&attrzp->z_lock); 2123 2124 if (mask & AT_UID) { 2125 zp->z_phys->zp_uid = (uint64_t)vap->va_uid; 2126 if (attrzp) { 2127 attrzp->z_phys->zp_uid = (uint64_t)vap->va_uid; 2128 } 2129 } 2130 2131 if (mask & AT_GID) { 2132 zp->z_phys->zp_gid = (uint64_t)vap->va_gid; 2133 if (attrzp) 2134 attrzp->z_phys->zp_gid = (uint64_t)vap->va_gid; 2135 } 2136 2137 if (attrzp) 2138 mutex_exit(&attrzp->z_lock); 2139 2140 if (mask & AT_ATIME) 2141 ZFS_TIME_ENCODE(&vap->va_atime, pzp->zp_atime); 2142 2143 if (mask & AT_MTIME) 2144 ZFS_TIME_ENCODE(&vap->va_mtime, pzp->zp_mtime); 2145 2146 if (mask_applied & AT_SIZE) 2147 zfs_time_stamper_locked(zp, CONTENT_MODIFIED, tx); 2148 else if (mask_applied != 0) 2149 zfs_time_stamper_locked(zp, STATE_CHANGED, tx); 2150 2151 out: 2152 2153 if (mask_applied != 0) 2154 seq = zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, 2155 mask_applied); 2156 2157 mutex_exit(&zp->z_lock); 2158 2159 if (attrzp) 2160 VN_RELE(ZTOV(attrzp)); 2161 2162 if (have_grow_lock) 2163 rw_exit(&zp->z_grow_lock); 2164 2165 dmu_tx_commit(tx); 2166 2167 zil_commit(zilog, seq, 0); 2168 2169 ZFS_EXIT(zfsvfs); 2170 return (err); 2171 } 2172 2173 /* 2174 * Search back through the directory tree, using the ".." entries. 2175 * Lock each directory in the chain to prevent concurrent renames. 2176 * Fail any attempt to move a directory into one of its own descendants. 2177 * XXX - z_parent_lock can overlap with map or grow locks 2178 */ 2179 typedef struct zfs_zlock { 2180 krwlock_t *zl_rwlock; /* lock we acquired */ 2181 znode_t *zl_znode; /* znode we held */ 2182 struct zfs_zlock *zl_next; /* next in list */ 2183 } zfs_zlock_t; 2184 2185 static int 2186 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 2187 { 2188 zfs_zlock_t *zl; 2189 znode_t *zp = tdzp; 2190 uint64_t rootid = zp->z_zfsvfs->z_root; 2191 uint64_t *oidp = &zp->z_id; 2192 krwlock_t *rwlp = &szp->z_parent_lock; 2193 krw_t rw = RW_WRITER; 2194 2195 /* 2196 * First pass write-locks szp and compares to zp->z_id. 2197 * Later passes read-lock zp and compare to zp->z_parent. 2198 */ 2199 do { 2200 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 2201 zl->zl_rwlock = rwlp; 2202 zl->zl_znode = NULL; 2203 zl->zl_next = *zlpp; 2204 *zlpp = zl; 2205 2206 rw_enter(rwlp, rw); 2207 2208 if (*oidp == szp->z_id) /* We're a descendant of szp */ 2209 return (EINVAL); 2210 2211 if (*oidp == rootid) /* We've hit the top */ 2212 return (0); 2213 2214 if (rw == RW_READER) { /* i.e. not the first pass */ 2215 int error = zfs_zget(zp->z_zfsvfs, *oidp, &zp); 2216 if (error) 2217 return (error); 2218 zl->zl_znode = zp; 2219 } 2220 oidp = &zp->z_phys->zp_parent; 2221 rwlp = &zp->z_parent_lock; 2222 rw = RW_READER; 2223 2224 } while (zp->z_id != sdzp->z_id); 2225 2226 return (0); 2227 } 2228 2229 /* 2230 * Drop locks and release vnodes that were held by zfs_rename_lock(). 2231 */ 2232 static void 2233 zfs_rename_unlock(zfs_zlock_t **zlpp) 2234 { 2235 zfs_zlock_t *zl; 2236 2237 while ((zl = *zlpp) != NULL) { 2238 if (zl->zl_znode != NULL) 2239 VN_RELE(ZTOV(zl->zl_znode)); 2240 rw_exit(zl->zl_rwlock); 2241 *zlpp = zl->zl_next; 2242 kmem_free(zl, sizeof (*zl)); 2243 } 2244 } 2245 2246 /* 2247 * Move an entry from the provided source directory to the target 2248 * directory. Change the entry name as indicated. 2249 * 2250 * IN: sdvp - Source directory containing the "old entry". 2251 * snm - Old entry name. 2252 * tdvp - Target directory to contain the "new entry". 2253 * tnm - New entry name. 2254 * cr - credentials of caller. 2255 * 2256 * RETURN: 0 if success 2257 * error code if failure 2258 * 2259 * Timestamps: 2260 * sdvp,tdvp - ctime|mtime updated 2261 */ 2262 static int 2263 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr) 2264 { 2265 znode_t *tdzp, *szp, *tzp; 2266 znode_t *sdzp = VTOZ(sdvp); 2267 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs; 2268 zilog_t *zilog = zfsvfs->z_log; 2269 uint64_t seq = 0; 2270 vnode_t *realvp; 2271 zfs_dirlock_t *sdl, *tdl; 2272 dmu_tx_t *tx; 2273 zfs_zlock_t *zl; 2274 int cmp, serr, terr, error; 2275 2276 ZFS_ENTER(zfsvfs); 2277 2278 /* 2279 * Make sure we have the real vp for the target directory. 2280 */ 2281 if (VOP_REALVP(tdvp, &realvp) == 0) 2282 tdvp = realvp; 2283 2284 if (tdvp->v_vfsp != sdvp->v_vfsp) { 2285 ZFS_EXIT(zfsvfs); 2286 return (EXDEV); 2287 } 2288 2289 tdzp = VTOZ(tdvp); 2290 top: 2291 szp = NULL; 2292 tzp = NULL; 2293 zl = NULL; 2294 2295 /* 2296 * This is to prevent the creation of links into attribute space 2297 * by renaming a linked file into/outof an attribute directory. 2298 * See the comment in zfs_link() for why this is considered bad. 2299 */ 2300 if ((tdzp->z_phys->zp_flags & ZFS_XATTR) != 2301 (sdzp->z_phys->zp_flags & ZFS_XATTR)) { 2302 ZFS_EXIT(zfsvfs); 2303 return (EINVAL); 2304 } 2305 2306 /* 2307 * Lock source and target directory entries. To prevent deadlock, 2308 * a lock ordering must be defined. We lock the directory with 2309 * the smallest object id first, or if it's a tie, the one with 2310 * the lexically first name. 2311 */ 2312 if (sdzp->z_id < tdzp->z_id) { 2313 cmp = -1; 2314 } else if (sdzp->z_id > tdzp->z_id) { 2315 cmp = 1; 2316 } else { 2317 cmp = strcmp(snm, tnm); 2318 if (cmp == 0) { 2319 /* 2320 * POSIX: "If the old argument and the new argument 2321 * both refer to links to the same existing file, 2322 * the rename() function shall return successfully 2323 * and perform no other action." 2324 */ 2325 ZFS_EXIT(zfsvfs); 2326 return (0); 2327 } 2328 } 2329 if (cmp < 0) { 2330 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS); 2331 terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0); 2332 } else { 2333 terr = zfs_dirent_lock(&tdl, tdzp, tnm, &tzp, 0); 2334 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, ZEXISTS); 2335 } 2336 2337 if (serr) { 2338 /* 2339 * Source entry invalid or not there. 2340 */ 2341 if (!terr) { 2342 zfs_dirent_unlock(tdl); 2343 if (tzp) 2344 VN_RELE(ZTOV(tzp)); 2345 } 2346 if (strcmp(snm, "..") == 0) 2347 serr = EINVAL; 2348 ZFS_EXIT(zfsvfs); 2349 return (serr); 2350 } 2351 if (terr) { 2352 zfs_dirent_unlock(sdl); 2353 VN_RELE(ZTOV(szp)); 2354 if (strcmp(tnm, "..") == 0) 2355 terr = EINVAL; 2356 ZFS_EXIT(zfsvfs); 2357 return (terr); 2358 } 2359 2360 /* 2361 * Must have write access at the source to remove the old entry 2362 * and write access at the target to create the new entry. 2363 * Note that if target and source are the same, this can be 2364 * done in a single check. 2365 */ 2366 2367 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 2368 goto out; 2369 2370 if (ZTOV(szp)->v_type == VDIR) { 2371 /* 2372 * Check to make sure rename is valid. 2373 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 2374 */ 2375 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 2376 goto out; 2377 } 2378 2379 /* 2380 * Does target exist? 2381 */ 2382 if (tzp) { 2383 /* 2384 * Source and target must be the same type. 2385 */ 2386 if (ZTOV(szp)->v_type == VDIR) { 2387 if (ZTOV(tzp)->v_type != VDIR) { 2388 error = ENOTDIR; 2389 goto out; 2390 } 2391 } else { 2392 if (ZTOV(tzp)->v_type == VDIR) { 2393 error = EISDIR; 2394 goto out; 2395 } 2396 } 2397 /* 2398 * POSIX dictates that when the source and target 2399 * entries refer to the same file object, rename 2400 * must do nothing and exit without error. 2401 */ 2402 if (szp->z_id == tzp->z_id) { 2403 error = 0; 2404 goto out; 2405 } 2406 } 2407 2408 vnevent_rename_src(ZTOV(szp)); 2409 if (tzp) 2410 vnevent_rename_dest(ZTOV(tzp)); 2411 2412 tx = dmu_tx_create(zfsvfs->z_os); 2413 dmu_tx_hold_bonus(tx, szp->z_id); /* nlink changes */ 2414 dmu_tx_hold_bonus(tx, sdzp->z_id); /* nlink changes */ 2415 if (sdzp != tdzp) { 2416 dmu_tx_hold_zap(tx, sdzp->z_id, 1); 2417 dmu_tx_hold_zap(tx, tdzp->z_id, 1); 2418 dmu_tx_hold_bonus(tx, tdzp->z_id); /* nlink changes */ 2419 } else { 2420 dmu_tx_hold_zap(tx, sdzp->z_id, 2); 2421 } 2422 if (tzp) { 2423 dmu_tx_hold_bonus(tx, tzp->z_id); /* nlink changes */ 2424 } 2425 dmu_tx_hold_zap(tx, zfsvfs->z_dqueue, 1); 2426 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2427 if (error) { 2428 dmu_tx_abort(tx); 2429 if (zl != NULL) 2430 zfs_rename_unlock(&zl); 2431 zfs_dirent_unlock(sdl); 2432 zfs_dirent_unlock(tdl); 2433 VN_RELE(ZTOV(szp)); 2434 if (tzp) 2435 VN_RELE(ZTOV(tzp)); 2436 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2437 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 2438 goto top; 2439 } 2440 ZFS_EXIT(zfsvfs); 2441 return (error); 2442 } 2443 2444 if (tzp) /* Attempt to remove the existing target */ 2445 error = zfs_link_destroy(tdl, tzp, tx, 0, NULL); 2446 2447 if (error == 0) { 2448 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 2449 if (error == 0) { 2450 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 2451 ASSERT(error == 0); 2452 seq = zfs_log_rename(zilog, tx, TX_RENAME, 2453 sdzp, sdl->dl_name, tdzp, tdl->dl_name, szp); 2454 } 2455 } 2456 2457 dmu_tx_commit(tx); 2458 out: 2459 if (zl != NULL) 2460 zfs_rename_unlock(&zl); 2461 2462 zfs_dirent_unlock(sdl); 2463 zfs_dirent_unlock(tdl); 2464 2465 VN_RELE(ZTOV(szp)); 2466 if (tzp) 2467 VN_RELE(ZTOV(tzp)); 2468 2469 zil_commit(zilog, seq, 0); 2470 2471 ZFS_EXIT(zfsvfs); 2472 return (error); 2473 } 2474 2475 /* 2476 * Insert the indicated symbolic reference entry into the directory. 2477 * 2478 * IN: dvp - Directory to contain new symbolic link. 2479 * link - Name for new symlink entry. 2480 * vap - Attributes of new entry. 2481 * target - Target path of new symlink. 2482 * cr - credentials of caller. 2483 * 2484 * RETURN: 0 if success 2485 * error code if failure 2486 * 2487 * Timestamps: 2488 * dvp - ctime|mtime updated 2489 */ 2490 static int 2491 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr) 2492 { 2493 znode_t *zp, *dzp = VTOZ(dvp); 2494 zfs_dirlock_t *dl; 2495 dmu_tx_t *tx; 2496 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2497 zilog_t *zilog = zfsvfs->z_log; 2498 uint64_t seq = 0; 2499 uint64_t zoid; 2500 int len = strlen(link); 2501 int error; 2502 2503 ASSERT(vap->va_type == VLNK); 2504 2505 ZFS_ENTER(zfsvfs); 2506 top: 2507 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 2508 ZFS_EXIT(zfsvfs); 2509 return (error); 2510 } 2511 2512 if (len > MAXPATHLEN) { 2513 ZFS_EXIT(zfsvfs); 2514 return (ENAMETOOLONG); 2515 } 2516 2517 /* 2518 * Attempt to lock directory; fail if entry already exists. 2519 */ 2520 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, ZNEW)) { 2521 ZFS_EXIT(zfsvfs); 2522 return (error); 2523 } 2524 2525 tx = dmu_tx_create(zfsvfs->z_os); 2526 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 2527 dmu_tx_hold_bonus(tx, dzp->z_id); 2528 dmu_tx_hold_zap(tx, dzp->z_id, 1); 2529 if (dzp->z_phys->zp_flags & ZFS_INHERIT_ACE) 2530 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, SPA_MAXBLOCKSIZE); 2531 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2532 if (error) { 2533 dmu_tx_abort(tx); 2534 zfs_dirent_unlock(dl); 2535 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2536 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 2537 goto top; 2538 } 2539 ZFS_EXIT(zfsvfs); 2540 return (error); 2541 } 2542 2543 dmu_buf_will_dirty(dzp->z_dbuf, tx); 2544 2545 /* 2546 * Create a new object for the symlink. 2547 * Put the link content into bonus buffer if it will fit; 2548 * otherwise, store it just like any other file data. 2549 */ 2550 zoid = 0; 2551 if (sizeof (znode_phys_t) + len <= dmu_bonus_max()) { 2552 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, len); 2553 if (len != 0) 2554 bcopy(link, zp->z_phys + 1, len); 2555 } else { 2556 dmu_buf_t *dbp; 2557 zfs_mknode(dzp, vap, &zoid, tx, cr, 0, &zp, 0); 2558 2559 rw_enter(&zp->z_grow_lock, RW_WRITER); 2560 error = zfs_grow_blocksize(zp, len, tx); 2561 rw_exit(&zp->z_grow_lock); 2562 if (error) 2563 goto out; 2564 2565 dbp = dmu_buf_hold(zfsvfs->z_os, zoid, 0); 2566 dmu_buf_will_dirty(dbp, tx); 2567 2568 ASSERT3U(len, <=, dbp->db_size); 2569 bcopy(link, dbp->db_data, len); 2570 dmu_buf_rele(dbp); 2571 } 2572 zp->z_phys->zp_size = len; 2573 2574 /* 2575 * Insert the new object into the directory. 2576 */ 2577 (void) zfs_link_create(dl, zp, tx, ZNEW); 2578 out: 2579 if (error == 0) 2580 seq = zfs_log_symlink(zilog, tx, TX_SYMLINK, 2581 dzp, zp, name, link); 2582 2583 dmu_tx_commit(tx); 2584 2585 zfs_dirent_unlock(dl); 2586 2587 VN_RELE(ZTOV(zp)); 2588 2589 zil_commit(zilog, seq, 0); 2590 2591 ZFS_EXIT(zfsvfs); 2592 return (error); 2593 } 2594 2595 /* 2596 * Return, in the buffer contained in the provided uio structure, 2597 * the symbolic path referred to by vp. 2598 * 2599 * IN: vp - vnode of symbolic link. 2600 * uoip - structure to contain the link path. 2601 * cr - credentials of caller. 2602 * 2603 * OUT: uio - structure to contain the link path. 2604 * 2605 * RETURN: 0 if success 2606 * error code if failure 2607 * 2608 * Timestamps: 2609 * vp - atime updated 2610 */ 2611 /* ARGSUSED */ 2612 static int 2613 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr) 2614 { 2615 znode_t *zp = VTOZ(vp); 2616 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2617 size_t bufsz; 2618 int error; 2619 2620 ZFS_ENTER(zfsvfs); 2621 2622 bufsz = (size_t)zp->z_phys->zp_size; 2623 if (bufsz + sizeof (znode_phys_t) <= zp->z_dbuf->db_size) { 2624 error = uiomove(zp->z_phys + 1, 2625 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 2626 } else { 2627 dmu_buf_t *dbp = dmu_buf_hold(zfsvfs->z_os, zp->z_id, 0); 2628 if ((error = dmu_buf_read_canfail(dbp)) != 0) { 2629 dmu_buf_rele(dbp); 2630 ZFS_EXIT(zfsvfs); 2631 return (error); 2632 } 2633 error = uiomove(dbp->db_data, 2634 MIN((size_t)bufsz, uio->uio_resid), UIO_READ, uio); 2635 dmu_buf_rele(dbp); 2636 } 2637 2638 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2639 ZFS_EXIT(zfsvfs); 2640 return (error); 2641 } 2642 2643 /* 2644 * Insert a new entry into directory tdvp referencing svp. 2645 * 2646 * IN: tdvp - Directory to contain new entry. 2647 * svp - vnode of new entry. 2648 * name - name of new entry. 2649 * cr - credentials of caller. 2650 * 2651 * RETURN: 0 if success 2652 * error code if failure 2653 * 2654 * Timestamps: 2655 * tdvp - ctime|mtime updated 2656 * svp - ctime updated 2657 */ 2658 /* ARGSUSED */ 2659 static int 2660 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr) 2661 { 2662 znode_t *dzp = VTOZ(tdvp); 2663 znode_t *tzp, *szp; 2664 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2665 zilog_t *zilog = zfsvfs->z_log; 2666 uint64_t seq = 0; 2667 zfs_dirlock_t *dl; 2668 dmu_tx_t *tx; 2669 vnode_t *realvp; 2670 int error; 2671 2672 ASSERT(tdvp->v_type == VDIR); 2673 2674 ZFS_ENTER(zfsvfs); 2675 2676 if (VOP_REALVP(svp, &realvp) == 0) 2677 svp = realvp; 2678 2679 if (svp->v_vfsp != tdvp->v_vfsp) { 2680 ZFS_EXIT(zfsvfs); 2681 return (EXDEV); 2682 } 2683 2684 szp = VTOZ(svp); 2685 top: 2686 /* 2687 * We do not support links between attributes and non-attributes 2688 * because of the potential security risk of creating links 2689 * into "normal" file space in order to circumvent restrictions 2690 * imposed in attribute space. 2691 */ 2692 if ((szp->z_phys->zp_flags & ZFS_XATTR) != 2693 (dzp->z_phys->zp_flags & ZFS_XATTR)) { 2694 ZFS_EXIT(zfsvfs); 2695 return (EINVAL); 2696 } 2697 2698 /* 2699 * POSIX dictates that we return EPERM here. 2700 * Better choices include ENOTSUP or EISDIR. 2701 */ 2702 if (svp->v_type == VDIR) { 2703 ZFS_EXIT(zfsvfs); 2704 return (EPERM); 2705 } 2706 2707 if ((uid_t)szp->z_phys->zp_uid != crgetuid(cr) && 2708 secpolicy_basic_link(cr) != 0) { 2709 ZFS_EXIT(zfsvfs); 2710 return (EPERM); 2711 } 2712 2713 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, cr)) { 2714 ZFS_EXIT(zfsvfs); 2715 return (error); 2716 } 2717 2718 /* 2719 * Attempt to lock directory; fail if entry already exists. 2720 */ 2721 if (error = zfs_dirent_lock(&dl, dzp, name, &tzp, ZNEW)) { 2722 ZFS_EXIT(zfsvfs); 2723 return (error); 2724 } 2725 2726 tx = dmu_tx_create(zfsvfs->z_os); 2727 dmu_tx_hold_bonus(tx, szp->z_id); 2728 dmu_tx_hold_zap(tx, dzp->z_id, 1); 2729 error = dmu_tx_assign(tx, zfsvfs->z_assign); 2730 if (error) { 2731 dmu_tx_abort(tx); 2732 zfs_dirent_unlock(dl); 2733 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2734 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 2735 goto top; 2736 } 2737 ZFS_EXIT(zfsvfs); 2738 return (error); 2739 } 2740 2741 error = zfs_link_create(dl, szp, tx, 0); 2742 2743 if (error == 0) 2744 seq = zfs_log_link(zilog, tx, TX_LINK, dzp, szp, name); 2745 2746 dmu_tx_commit(tx); 2747 2748 zfs_dirent_unlock(dl); 2749 2750 zil_commit(zilog, seq, 0); 2751 2752 ZFS_EXIT(zfsvfs); 2753 return (error); 2754 } 2755 2756 /* 2757 * zfs_null_putapage() is used when the file system has been force 2758 * unmounted. It just drops the pages. 2759 */ 2760 /* ARGSUSED */ 2761 static int 2762 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 2763 size_t *lenp, int flags, cred_t *cr) 2764 { 2765 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR); 2766 return (0); 2767 } 2768 2769 /* ARGSUSED */ 2770 static int 2771 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 2772 size_t *lenp, int flags, cred_t *cr) 2773 { 2774 znode_t *zp = VTOZ(vp); 2775 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2776 zilog_t *zilog = zfsvfs->z_log; 2777 uint64_t seq = 0; 2778 dmu_tx_t *tx; 2779 u_offset_t off; 2780 ssize_t len; 2781 caddr_t va; 2782 int err; 2783 2784 top: 2785 rw_enter(&zp->z_grow_lock, RW_READER); 2786 2787 off = pp->p_offset; 2788 len = MIN(PAGESIZE, zp->z_phys->zp_size - off); 2789 2790 tx = dmu_tx_create(zfsvfs->z_os); 2791 dmu_tx_hold_write(tx, zp->z_id, off, len); 2792 dmu_tx_hold_bonus(tx, zp->z_id); 2793 err = dmu_tx_assign(tx, zfsvfs->z_assign); 2794 if (err != 0) { 2795 dmu_tx_abort(tx); 2796 rw_exit(&zp->z_grow_lock); 2797 if (err == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 2798 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 2799 goto top; 2800 } 2801 goto out; 2802 } 2803 2804 va = ppmapin(pp, PROT_READ | PROT_WRITE, (caddr_t)-1); 2805 2806 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx); 2807 2808 ppmapout(va); 2809 2810 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 2811 seq = zfs_log_write(zilog, tx, TX_WRITE, zp, off, len, 0, NULL); 2812 dmu_tx_commit(tx); 2813 2814 rw_exit(&zp->z_grow_lock); 2815 2816 pvn_write_done(pp, B_WRITE | flags); 2817 if (offp) 2818 *offp = off; 2819 if (lenp) 2820 *lenp = len; 2821 2822 zil_commit(zilog, seq, (flags & B_ASYNC) ? 0 : FDSYNC); 2823 out: 2824 return (err); 2825 } 2826 2827 /* 2828 * Copy the portion of the file indicated from pages into the file. 2829 * The pages are stored in a page list attached to the files vnode. 2830 * 2831 * IN: vp - vnode of file to push page data to. 2832 * off - position in file to put data. 2833 * len - amount of data to write. 2834 * flags - flags to control the operation. 2835 * cr - credentials of caller. 2836 * 2837 * RETURN: 0 if success 2838 * error code if failure 2839 * 2840 * Timestamps: 2841 * vp - ctime|mtime updated 2842 */ 2843 static int 2844 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr) 2845 { 2846 znode_t *zp = VTOZ(vp); 2847 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2848 page_t *pp; 2849 size_t io_len; 2850 u_offset_t io_off; 2851 int error = 0; 2852 2853 ZFS_ENTER(zfsvfs); 2854 2855 ASSERT(zp->z_dbuf_held && zp->z_phys); 2856 2857 if (len == 0) { 2858 /* 2859 * Search the entire vp list for pages >= off. 2860 */ 2861 error = pvn_vplist_dirty(vp, (u_offset_t)off, zfs_putapage, 2862 flags, cr); 2863 ZFS_EXIT(zfsvfs); 2864 return (error); 2865 } 2866 2867 if (off > zp->z_phys->zp_size) { 2868 /* past end of file */ 2869 ZFS_EXIT(zfsvfs); 2870 return (0); 2871 } 2872 2873 len = MIN(len, zp->z_phys->zp_size - off); 2874 2875 io_off = off; 2876 while (io_off < off + len) { 2877 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { 2878 pp = page_lookup(vp, io_off, 2879 (flags & (B_INVAL | B_FREE)) ? 2880 SE_EXCL : SE_SHARED); 2881 } else { 2882 pp = page_lookup_nowait(vp, io_off, 2883 (flags & B_FREE) ? SE_EXCL : SE_SHARED); 2884 } 2885 2886 if (pp != NULL && pvn_getdirty(pp, flags)) { 2887 int err; 2888 2889 /* 2890 * Found a dirty page to push 2891 */ 2892 if (err = 2893 zfs_putapage(vp, pp, &io_off, &io_len, flags, cr)) 2894 error = err; 2895 } else { 2896 io_len = PAGESIZE; 2897 } 2898 io_off += io_len; 2899 } 2900 ZFS_EXIT(zfsvfs); 2901 return (error); 2902 } 2903 2904 void 2905 zfs_inactive(vnode_t *vp, cred_t *cr) 2906 { 2907 znode_t *zp = VTOZ(vp); 2908 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2909 int error; 2910 2911 rw_enter(&zfsvfs->z_um_lock, RW_READER); 2912 if (zfsvfs->z_unmounted2) { 2913 ASSERT(zp->z_dbuf_held == 0); 2914 2915 if (vn_has_cached_data(vp)) { 2916 (void) pvn_vplist_dirty(vp, 0, zfs_null_putapage, 2917 B_INVAL, cr); 2918 } 2919 2920 vp->v_count = 0; /* count arrives as 1 */ 2921 zfs_znode_free(zp); 2922 rw_exit(&zfsvfs->z_um_lock); 2923 VFS_RELE(zfsvfs->z_vfs); 2924 return; 2925 } 2926 2927 /* 2928 * Attempt to push any data in the page cache. If this fails 2929 * we will get kicked out later in zfs_zinactive(). 2930 */ 2931 if (vn_has_cached_data(vp)) { 2932 (void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC, 2933 cr); 2934 } 2935 2936 if (zp->z_atime_dirty && zp->z_reap == 0) { 2937 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 2938 2939 dmu_tx_hold_bonus(tx, zp->z_id); 2940 error = dmu_tx_assign(tx, TXG_WAIT); 2941 if (error) { 2942 dmu_tx_abort(tx); 2943 } else { 2944 dmu_buf_will_dirty(zp->z_dbuf, tx); 2945 mutex_enter(&zp->z_lock); 2946 zp->z_atime_dirty = 0; 2947 mutex_exit(&zp->z_lock); 2948 dmu_tx_commit(tx); 2949 } 2950 } 2951 2952 zfs_zinactive(zp); 2953 rw_exit(&zfsvfs->z_um_lock); 2954 } 2955 2956 /* 2957 * Bounds-check the seek operation. 2958 * 2959 * IN: vp - vnode seeking within 2960 * ooff - old file offset 2961 * noffp - pointer to new file offset 2962 * 2963 * RETURN: 0 if success 2964 * EINVAL if new offset invalid 2965 */ 2966 /* ARGSUSED */ 2967 static int 2968 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp) 2969 { 2970 if (vp->v_type == VDIR) 2971 return (0); 2972 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0); 2973 } 2974 2975 /* 2976 * Pre-filter the generic locking function to trap attempts to place 2977 * a mandatory lock on a memory mapped file. 2978 */ 2979 static int 2980 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset, 2981 flk_callback_t *flk_cbp, cred_t *cr) 2982 { 2983 znode_t *zp = VTOZ(vp); 2984 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2985 uint_t cnt = 1; 2986 int error; 2987 2988 ZFS_ENTER(zfsvfs); 2989 2990 /* 2991 * If file is being mapped, disallow frlock. We set the mapcnt to 2992 * -1 here to signal that we are in the process of setting a lock. 2993 * This prevents a race with zfs_map(). 2994 * XXX - well, sort of; since zfs_map() does not change z_mapcnt, 2995 * we could be in the middle of zfs_map() and still call fs_frlock(). 2996 * Also, we are doing no checking in zfs_addmap() (where z_mapcnt 2997 * *is* manipulated). 2998 */ 2999 if (MANDMODE((mode_t)zp->z_phys->zp_mode) && 3000 (int)(cnt = atomic_cas_32(&zp->z_mapcnt, 0, -1)) > 0) { 3001 ZFS_EXIT(zfsvfs); 3002 return (EAGAIN); 3003 } 3004 error = fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr); 3005 ASSERT((cnt != 0) || ((int)atomic_cas_32(&zp->z_mapcnt, -1, 0) == -1)); 3006 ZFS_EXIT(zfsvfs); 3007 return (error); 3008 } 3009 3010 /* 3011 * If we can't find a page in the cache, we will create a new page 3012 * and fill it with file data. For efficiency, we may try to fill 3013 * multiple pages as once (klustering). 3014 */ 3015 static int 3016 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, 3017 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw) 3018 { 3019 znode_t *zp = VTOZ(vp); 3020 page_t *pp, *cur_pp; 3021 objset_t *os = zp->z_zfsvfs->z_os; 3022 caddr_t va; 3023 u_offset_t io_off, total; 3024 uint64_t oid = zp->z_id; 3025 size_t io_len; 3026 int err; 3027 3028 /* 3029 * If we are only asking for a single page don't bother klustering. 3030 */ 3031 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE || 3032 off > zp->z_phys->zp_size) { 3033 io_off = off; 3034 io_len = PAGESIZE; 3035 pp = page_create_va(vp, io_off, io_len, PG_WAIT, seg, addr); 3036 } else { 3037 /* 3038 * Try to fill a kluster of pages (a blocks worth). 3039 */ 3040 size_t klen; 3041 u_offset_t koff; 3042 3043 if (!ISP2(zp->z_blksz)) { 3044 /* Only one block in the file. */ 3045 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 3046 koff = 0; 3047 } else { 3048 klen = plsz; 3049 koff = P2ALIGN(off, (u_offset_t)klen); 3050 } 3051 if (klen > zp->z_phys->zp_size) 3052 klen = P2ROUNDUP(zp->z_phys->zp_size, 3053 (uint64_t)PAGESIZE); 3054 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 3055 &io_len, koff, klen, 0); 3056 } 3057 if (pp == NULL) { 3058 /* 3059 * Some other thread entered the page before us. 3060 * Return to zfs_getpage to retry the lookup. 3061 */ 3062 *pl = NULL; 3063 return (0); 3064 } 3065 3066 /* 3067 * Fill the pages in the kluster. 3068 */ 3069 cur_pp = pp; 3070 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) { 3071 ASSERT(io_off == cur_pp->p_offset); 3072 va = ppmapin(cur_pp, PROT_READ | PROT_WRITE, (caddr_t)-1); 3073 err = dmu_read_canfail(os, oid, io_off, PAGESIZE, va); 3074 ppmapout(va); 3075 if (err) { 3076 /* On error, toss the entire kluster */ 3077 pvn_read_done(pp, B_ERROR); 3078 return (err); 3079 } 3080 cur_pp = cur_pp->p_next; 3081 } 3082 out: 3083 /* 3084 * Fill in the page list array from the kluster. If 3085 * there are too many pages in the kluster, return 3086 * as many pages as possible starting from the desired 3087 * offset `off'. 3088 * NOTE: the page list will always be null terminated. 3089 */ 3090 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 3091 3092 return (0); 3093 } 3094 3095 /* 3096 * Return pointers to the pages for the file region [off, off + len] 3097 * in the pl array. If plsz is greater than len, this function may 3098 * also return page pointers from before or after the specified 3099 * region (i.e. some region [off', off' + plsz]). These additional 3100 * pages are only returned if they are already in the cache, or were 3101 * created as part of a klustered read. 3102 * 3103 * IN: vp - vnode of file to get data from. 3104 * off - position in file to get data from. 3105 * len - amount of data to retrieve. 3106 * plsz - length of provided page list. 3107 * seg - segment to obtain pages for. 3108 * addr - virtual address of fault. 3109 * rw - mode of created pages. 3110 * cr - credentials of caller. 3111 * 3112 * OUT: protp - protection mode of created pages. 3113 * pl - list of pages created. 3114 * 3115 * RETURN: 0 if success 3116 * error code if failure 3117 * 3118 * Timestamps: 3119 * vp - atime updated 3120 */ 3121 /* ARGSUSED */ 3122 static int 3123 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 3124 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 3125 enum seg_rw rw, cred_t *cr) 3126 { 3127 znode_t *zp = VTOZ(vp); 3128 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3129 page_t *pp, **pl0 = pl; 3130 int cnt = 0, need_unlock = 0, err = 0; 3131 3132 ZFS_ENTER(zfsvfs); 3133 3134 if (protp) 3135 *protp = PROT_ALL; 3136 3137 ASSERT(zp->z_dbuf_held && zp->z_phys); 3138 3139 /* no faultahead (for now) */ 3140 if (pl == NULL) { 3141 ZFS_EXIT(zfsvfs); 3142 return (0); 3143 } 3144 3145 /* can't fault past EOF */ 3146 if (off >= zp->z_phys->zp_size) { 3147 ZFS_EXIT(zfsvfs); 3148 return (EFAULT); 3149 } 3150 3151 /* 3152 * Make sure nobody restructures the file (changes block size) 3153 * in the middle of the getpage. 3154 */ 3155 rw_enter(&zp->z_grow_lock, RW_READER); 3156 3157 /* 3158 * If we already own the lock, then we must be page faulting 3159 * in the middle of a write to this file (i.e., we are writing 3160 * to this file using data from a mapped region of the file). 3161 */ 3162 if (!rw_owner(&zp->z_map_lock)) { 3163 rw_enter(&zp->z_map_lock, RW_WRITER); 3164 need_unlock = TRUE; 3165 } 3166 3167 /* 3168 * Loop through the requested range [off, off + len] looking 3169 * for pages. If we don't find a page, we will need to create 3170 * a new page and fill it with data from the file. 3171 */ 3172 while (len > 0) { 3173 if (plsz < PAGESIZE) 3174 break; 3175 if (pp = page_lookup(vp, off, SE_SHARED)) { 3176 *pl++ = pp; 3177 off += PAGESIZE; 3178 addr += PAGESIZE; 3179 len -= PAGESIZE; 3180 plsz -= PAGESIZE; 3181 } else { 3182 err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw); 3183 /* 3184 * klustering may have changed our region 3185 * to be block aligned. 3186 */ 3187 if (((pp = *pl) != 0) && (off != pp->p_offset)) { 3188 int delta = off - pp->p_offset; 3189 len += delta; 3190 off -= delta; 3191 addr -= delta; 3192 } 3193 while (*pl) { 3194 pl++; 3195 cnt++; 3196 off += PAGESIZE; 3197 addr += PAGESIZE; 3198 plsz -= PAGESIZE; 3199 if (len > PAGESIZE) 3200 len -= PAGESIZE; 3201 else 3202 len = 0; 3203 } 3204 } 3205 if (err) 3206 goto out; 3207 } 3208 3209 /* 3210 * Fill out the page array with any pages already in the cache. 3211 */ 3212 while (plsz > 0) { 3213 pp = page_lookup_nowait(vp, off, SE_SHARED); 3214 if (pp == NULL) 3215 break; 3216 *pl++ = pp; 3217 off += PAGESIZE; 3218 plsz -= PAGESIZE; 3219 } 3220 3221 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 3222 out: 3223 if (err) { 3224 /* 3225 * Release any pages we have locked. 3226 */ 3227 while (pl > pl0) 3228 page_unlock(*--pl); 3229 } 3230 *pl = NULL; 3231 3232 if (need_unlock) 3233 rw_exit(&zp->z_map_lock); 3234 rw_exit(&zp->z_grow_lock); 3235 3236 ZFS_EXIT(zfsvfs); 3237 return (err); 3238 } 3239 3240 static int 3241 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 3242 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr) 3243 { 3244 znode_t *zp = VTOZ(vp); 3245 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3246 segvn_crargs_t vn_a; 3247 int error; 3248 3249 ZFS_ENTER(zfsvfs); 3250 3251 if (vp->v_flag & VNOMAP) { 3252 ZFS_EXIT(zfsvfs); 3253 return (ENOSYS); 3254 } 3255 3256 if (off < 0 || len > MAXOFFSET_T - off) { 3257 ZFS_EXIT(zfsvfs); 3258 return (ENXIO); 3259 } 3260 3261 if (vp->v_type != VREG) { 3262 ZFS_EXIT(zfsvfs); 3263 return (ENODEV); 3264 } 3265 3266 /* 3267 * If file is locked, disallow mapping. 3268 * XXX - since we don't modify z_mapcnt here, there is nothing 3269 * to stop a file lock being placed immediately after we complete 3270 * this check. 3271 */ 3272 if (MANDMODE((mode_t)zp->z_phys->zp_mode)) { 3273 if (vn_has_flocks(vp) || zp->z_mapcnt == -1) { 3274 ZFS_EXIT(zfsvfs); 3275 return (EAGAIN); 3276 } 3277 } 3278 3279 as_rangelock(as); 3280 if ((flags & MAP_FIXED) == 0) { 3281 map_addr(addrp, len, off, 1, flags); 3282 if (*addrp == NULL) { 3283 as_rangeunlock(as); 3284 ZFS_EXIT(zfsvfs); 3285 return (ENOMEM); 3286 } 3287 } else { 3288 /* 3289 * User specified address - blow away any previous mappings 3290 */ 3291 (void) as_unmap(as, *addrp, len); 3292 } 3293 3294 vn_a.vp = vp; 3295 vn_a.offset = (u_offset_t)off; 3296 vn_a.type = flags & MAP_TYPE; 3297 vn_a.prot = prot; 3298 vn_a.maxprot = maxprot; 3299 vn_a.cred = cr; 3300 vn_a.amp = NULL; 3301 vn_a.flags = flags & ~MAP_TYPE; 3302 3303 error = as_map(as, *addrp, len, segvn_create, &vn_a); 3304 3305 as_rangeunlock(as); 3306 ZFS_EXIT(zfsvfs); 3307 return (error); 3308 } 3309 3310 /* ARGSUSED */ 3311 static int 3312 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 3313 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr) 3314 { 3315 /* 3316 * XXX - shouldn't we be checking for file locks here? 3317 */ 3318 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, 0); 3319 atomic_add_32(&VTOZ(vp)->z_mapcnt, btopr(len)); 3320 return (0); 3321 } 3322 3323 /* ARGSUSED */ 3324 static int 3325 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 3326 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr) 3327 { 3328 atomic_add_32(&VTOZ(vp)->z_mapcnt, -btopr(len)); 3329 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, 0); 3330 return (0); 3331 } 3332 3333 /* 3334 * Free or allocate space in a file. Currently, this function only 3335 * supports the `F_FREESP' command. However, this command is somewhat 3336 * misnamed, as its functionality includes the ability to allocate as 3337 * well as free space. 3338 * 3339 * IN: vp - vnode of file to free data in. 3340 * cmd - action to take (only F_FREESP supported). 3341 * bfp - section of file to free/alloc. 3342 * flag - current file open mode flags. 3343 * offset - current file offset. 3344 * cr - credentials of caller [UNUSED]. 3345 * 3346 * RETURN: 0 if success 3347 * error code if failure 3348 * 3349 * Timestamps: 3350 * vp - ctime|mtime updated 3351 * 3352 * NOTE: This function is limited in that it will only permit space to 3353 * be freed at the end of a file. In essence, this function simply 3354 * allows one to set the file size. 3355 */ 3356 /* ARGSUSED */ 3357 static int 3358 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag, 3359 offset_t offset, cred_t *cr, caller_context_t *ct) 3360 { 3361 dmu_tx_t *tx; 3362 znode_t *zp = VTOZ(vp); 3363 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3364 zilog_t *zilog = zfsvfs->z_log; 3365 uint64_t seq = 0; 3366 uint64_t off, len; 3367 int error; 3368 3369 ZFS_ENTER(zfsvfs); 3370 3371 top: 3372 if (cmd != F_FREESP) { 3373 ZFS_EXIT(zfsvfs); 3374 return (EINVAL); 3375 } 3376 3377 if (error = convoff(vp, bfp, 0, offset)) { 3378 ZFS_EXIT(zfsvfs); 3379 return (error); 3380 } 3381 3382 if (bfp->l_len < 0) { 3383 ZFS_EXIT(zfsvfs); 3384 return (EINVAL); 3385 } 3386 3387 off = bfp->l_start; 3388 len = bfp->l_len; 3389 tx = dmu_tx_create(zfsvfs->z_os); 3390 /* 3391 * Grab the grow_lock to serialize this change with 3392 * respect to other file size changes. 3393 */ 3394 dmu_tx_hold_bonus(tx, zp->z_id); 3395 rw_enter(&zp->z_grow_lock, RW_WRITER); 3396 if (off + len > zp->z_blksz && zp->z_blksz < zfsvfs->z_max_blksz && 3397 off >= zp->z_phys->zp_size) { 3398 /* 3399 * We are increasing the length of the file, 3400 * and this may mean a block size increase. 3401 */ 3402 dmu_tx_hold_write(tx, zp->z_id, 0, 3403 MIN(off + len, zfsvfs->z_max_blksz)); 3404 } else if (off < zp->z_phys->zp_size) { 3405 /* 3406 * If len == 0, we are truncating the file. 3407 */ 3408 dmu_tx_hold_free(tx, zp->z_id, off, len ? len : DMU_OBJECT_END); 3409 } 3410 3411 error = dmu_tx_assign(tx, zfsvfs->z_assign); 3412 if (error) { 3413 dmu_tx_abort(tx); 3414 rw_exit(&zp->z_grow_lock); 3415 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 3416 txg_wait_open(dmu_objset_pool(zfsvfs->z_os), 0); 3417 goto top; 3418 } 3419 ZFS_EXIT(zfsvfs); 3420 return (error); 3421 } 3422 3423 error = zfs_freesp(zp, off, len, flag, tx, cr); 3424 3425 if (error == 0) { 3426 zfs_time_stamper(zp, CONTENT_MODIFIED, tx); 3427 seq = zfs_log_truncate(zilog, tx, TX_TRUNCATE, zp, off, len); 3428 } 3429 3430 rw_exit(&zp->z_grow_lock); 3431 3432 dmu_tx_commit(tx); 3433 3434 zil_commit(zilog, seq, 0); 3435 3436 ZFS_EXIT(zfsvfs); 3437 return (error); 3438 } 3439 3440 static int 3441 zfs_fid(vnode_t *vp, fid_t *fidp) 3442 { 3443 znode_t *zp = VTOZ(vp); 3444 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3445 uint32_t gen = (uint32_t)zp->z_phys->zp_gen; 3446 uint64_t object = zp->z_id; 3447 zfid_short_t *zfid; 3448 int size, i; 3449 3450 ZFS_ENTER(zfsvfs); 3451 3452 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 3453 if (fidp->fid_len < size) { 3454 fidp->fid_len = size; 3455 return (ENOSPC); 3456 } 3457 3458 zfid = (zfid_short_t *)fidp; 3459 3460 zfid->zf_len = size; 3461 3462 for (i = 0; i < sizeof (zfid->zf_object); i++) 3463 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 3464 3465 /* Must have a non-zero generation number to distinguish from .zfs */ 3466 if (gen == 0) 3467 gen = 1; 3468 for (i = 0; i < sizeof (zfid->zf_gen); i++) 3469 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 3470 3471 if (size == LONG_FID_LEN) { 3472 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 3473 zfid_long_t *zlfid; 3474 3475 zlfid = (zfid_long_t *)fidp; 3476 3477 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 3478 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 3479 3480 /* XXX - this should be the generation number for the objset */ 3481 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 3482 zlfid->zf_setgen[i] = 0; 3483 } 3484 3485 ZFS_EXIT(zfsvfs); 3486 return (0); 3487 } 3488 3489 static int 3490 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr) 3491 { 3492 znode_t *zp, *xzp; 3493 zfsvfs_t *zfsvfs; 3494 zfs_dirlock_t *dl; 3495 int error; 3496 3497 switch (cmd) { 3498 case _PC_LINK_MAX: 3499 *valp = ULONG_MAX; 3500 return (0); 3501 3502 case _PC_FILESIZEBITS: 3503 *valp = 64; 3504 return (0); 3505 3506 case _PC_XATTR_EXISTS: 3507 zp = VTOZ(vp); 3508 zfsvfs = zp->z_zfsvfs; 3509 ZFS_ENTER(zfsvfs); 3510 *valp = 0; 3511 error = zfs_dirent_lock(&dl, zp, "", &xzp, 3512 ZXATTR | ZEXISTS | ZSHARED); 3513 if (error == 0) { 3514 zfs_dirent_unlock(dl); 3515 if (!zfs_dirempty(xzp)) 3516 *valp = 1; 3517 VN_RELE(ZTOV(xzp)); 3518 } else if (error == ENOENT) { 3519 /* 3520 * If there aren't extended attributes, it's the 3521 * same as having zero of them. 3522 */ 3523 error = 0; 3524 } 3525 ZFS_EXIT(zfsvfs); 3526 return (error); 3527 3528 case _PC_ACL_ENABLED: 3529 *valp = _ACL_ACE_ENABLED; 3530 return (0); 3531 3532 case _PC_MIN_HOLE_SIZE: 3533 *valp = (ulong_t)SPA_MINBLOCKSIZE; 3534 return (0); 3535 3536 default: 3537 return (fs_pathconf(vp, cmd, valp, cr)); 3538 } 3539 } 3540 3541 /*ARGSUSED*/ 3542 static int 3543 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr) 3544 { 3545 znode_t *zp = VTOZ(vp); 3546 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3547 int error; 3548 3549 ZFS_ENTER(zfsvfs); 3550 error = zfs_getacl(zp, vsecp, cr); 3551 ZFS_EXIT(zfsvfs); 3552 3553 return (error); 3554 } 3555 3556 /*ARGSUSED*/ 3557 static int 3558 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr) 3559 { 3560 znode_t *zp = VTOZ(vp); 3561 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3562 int error; 3563 3564 ZFS_ENTER(zfsvfs); 3565 error = zfs_setacl(zp, vsecp, cr); 3566 ZFS_EXIT(zfsvfs); 3567 return (error); 3568 } 3569 3570 /* 3571 * Predeclare these here so that the compiler assumes that 3572 * this is an "old style" function declaration that does 3573 * not include arguments => we won't get type mismatch errors 3574 * in the initializations that follow. 3575 */ 3576 static int zfs_inval(); 3577 static int zfs_isdir(); 3578 3579 static int 3580 zfs_inval() 3581 { 3582 return (EINVAL); 3583 } 3584 3585 static int 3586 zfs_isdir() 3587 { 3588 return (EISDIR); 3589 } 3590 /* 3591 * Directory vnode operations template 3592 */ 3593 vnodeops_t *zfs_dvnodeops; 3594 const fs_operation_def_t zfs_dvnodeops_template[] = { 3595 VOPNAME_OPEN, zfs_open, 3596 VOPNAME_CLOSE, zfs_close, 3597 VOPNAME_READ, zfs_isdir, 3598 VOPNAME_WRITE, zfs_isdir, 3599 VOPNAME_IOCTL, zfs_ioctl, 3600 VOPNAME_GETATTR, zfs_getattr, 3601 VOPNAME_SETATTR, zfs_setattr, 3602 VOPNAME_ACCESS, zfs_access, 3603 VOPNAME_LOOKUP, zfs_lookup, 3604 VOPNAME_CREATE, zfs_create, 3605 VOPNAME_REMOVE, zfs_remove, 3606 VOPNAME_LINK, zfs_link, 3607 VOPNAME_RENAME, zfs_rename, 3608 VOPNAME_MKDIR, zfs_mkdir, 3609 VOPNAME_RMDIR, zfs_rmdir, 3610 VOPNAME_READDIR, zfs_readdir, 3611 VOPNAME_SYMLINK, zfs_symlink, 3612 VOPNAME_FSYNC, zfs_fsync, 3613 VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive, 3614 VOPNAME_FID, zfs_fid, 3615 VOPNAME_SEEK, zfs_seek, 3616 VOPNAME_PATHCONF, zfs_pathconf, 3617 VOPNAME_GETSECATTR, zfs_getsecattr, 3618 VOPNAME_SETSECATTR, zfs_setsecattr, 3619 NULL, NULL 3620 }; 3621 3622 /* 3623 * Regular file vnode operations template 3624 */ 3625 vnodeops_t *zfs_fvnodeops; 3626 const fs_operation_def_t zfs_fvnodeops_template[] = { 3627 VOPNAME_OPEN, zfs_open, 3628 VOPNAME_CLOSE, zfs_close, 3629 VOPNAME_READ, zfs_read, 3630 VOPNAME_WRITE, zfs_write, 3631 VOPNAME_IOCTL, zfs_ioctl, 3632 VOPNAME_GETATTR, zfs_getattr, 3633 VOPNAME_SETATTR, zfs_setattr, 3634 VOPNAME_ACCESS, zfs_access, 3635 VOPNAME_LOOKUP, zfs_lookup, 3636 VOPNAME_RENAME, zfs_rename, 3637 VOPNAME_FSYNC, zfs_fsync, 3638 VOPNAME_INACTIVE, (fs_generic_func_p)zfs_inactive, 3639 VOPNAME_FID, zfs_fid, 3640 VOPNAME_SEEK, zfs_seek, 3641 VOPNAME_FRLOCK, zfs_frlock, 3642 VOPNAME_SPACE, zfs_space, 3643 VOPNAME_GETPAGE, zfs_getpage, 3644 VOPNAME_PUTPAGE, zfs_putpage, 3645 VOPNAME_MAP, (fs_generic_func_p) zfs_map, 3646 VOPNAME_ADDMAP, (fs_generic_func_p) zfs_addmap, 3647 VOPNAME_DELMAP, zfs_delmap, 3648 VOPNAME_PATHCONF, zfs_pathconf, 3649 VOPNAME_GETSECATTR, zfs_getsecattr, 3650 VOPNAME_SETSECATTR, zfs_setsecattr, 3651 VOPNAME_VNEVENT, fs_vnevent_support, 3652 NULL, NULL 3653 }; 3654 3655 /* 3656 * Symbolic link vnode operations template 3657 */ 3658 vnodeops_t *zfs_symvnodeops; 3659 const fs_operation_def_t zfs_symvnodeops_template[] = { 3660 VOPNAME_GETATTR, zfs_getattr, 3661 VOPNAME_SETATTR, zfs_setattr, 3662 VOPNAME_ACCESS, zfs_access, 3663 VOPNAME_RENAME, zfs_rename, 3664 VOPNAME_READLINK, zfs_readlink, 3665 VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive, 3666 VOPNAME_FID, zfs_fid, 3667 VOPNAME_PATHCONF, zfs_pathconf, 3668 VOPNAME_VNEVENT, fs_vnevent_support, 3669 NULL, NULL 3670 }; 3671 3672 /* 3673 * Extended attribute directory vnode operations template 3674 * This template is identical to the directory vnodes 3675 * operation template except for restricted operations: 3676 * VOP_MKDIR() 3677 * VOP_SYMLINK() 3678 * Note that there are other restrictions embedded in: 3679 * zfs_create() - restrict type to VREG 3680 * zfs_link() - no links into/out of attribute space 3681 * zfs_rename() - no moves into/out of attribute space 3682 */ 3683 vnodeops_t *zfs_xdvnodeops; 3684 const fs_operation_def_t zfs_xdvnodeops_template[] = { 3685 VOPNAME_OPEN, zfs_open, 3686 VOPNAME_CLOSE, zfs_close, 3687 VOPNAME_IOCTL, zfs_ioctl, 3688 VOPNAME_GETATTR, zfs_getattr, 3689 VOPNAME_SETATTR, zfs_setattr, 3690 VOPNAME_ACCESS, zfs_access, 3691 VOPNAME_LOOKUP, zfs_lookup, 3692 VOPNAME_CREATE, zfs_create, 3693 VOPNAME_REMOVE, zfs_remove, 3694 VOPNAME_LINK, zfs_link, 3695 VOPNAME_RENAME, zfs_rename, 3696 VOPNAME_MKDIR, zfs_inval, 3697 VOPNAME_RMDIR, zfs_rmdir, 3698 VOPNAME_READDIR, zfs_readdir, 3699 VOPNAME_SYMLINK, zfs_inval, 3700 VOPNAME_FSYNC, zfs_fsync, 3701 VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive, 3702 VOPNAME_FID, zfs_fid, 3703 VOPNAME_SEEK, zfs_seek, 3704 VOPNAME_PATHCONF, zfs_pathconf, 3705 VOPNAME_GETSECATTR, zfs_getsecattr, 3706 VOPNAME_SETSECATTR, zfs_setsecattr, 3707 VOPNAME_VNEVENT, fs_vnevent_support, 3708 NULL, NULL 3709 }; 3710 3711 /* 3712 * Error vnode operations template 3713 */ 3714 vnodeops_t *zfs_evnodeops; 3715 const fs_operation_def_t zfs_evnodeops_template[] = { 3716 VOPNAME_INACTIVE, (fs_generic_func_p) zfs_inactive, 3717 VOPNAME_PATHCONF, zfs_pathconf, 3718 NULL, NULL 3719 }; 3720