1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22 /* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 2012, 2017 by Delphix. All rights reserved. 25 * Copyright (c) 2014 Integros [integros.com] 26 * Copyright 2015 Joyent, Inc. 27 * Copyright 2017 Nexenta Systems, Inc. 28 */ 29 30 /* Portions Copyright 2007 Jeremy Teo */ 31 /* Portions Copyright 2010 Robert Milkowski */ 32 33 #include <sys/types.h> 34 #include <sys/param.h> 35 #include <sys/time.h> 36 #include <sys/systm.h> 37 #include <sys/sysmacros.h> 38 #include <sys/resource.h> 39 #include <sys/vfs.h> 40 #include <sys/vfs_opreg.h> 41 #include <sys/vnode.h> 42 #include <sys/file.h> 43 #include <sys/stat.h> 44 #include <sys/kmem.h> 45 #include <sys/taskq.h> 46 #include <sys/uio.h> 47 #include <sys/vmsystm.h> 48 #include <sys/atomic.h> 49 #include <sys/vm.h> 50 #include <vm/seg_vn.h> 51 #include <vm/pvn.h> 52 #include <vm/as.h> 53 #include <vm/kpm.h> 54 #include <vm/seg_kpm.h> 55 #include <sys/mman.h> 56 #include <sys/pathname.h> 57 #include <sys/cmn_err.h> 58 #include <sys/errno.h> 59 #include <sys/unistd.h> 60 #include <sys/zfs_dir.h> 61 #include <sys/zfs_acl.h> 62 #include <sys/zfs_ioctl.h> 63 #include <sys/fs/zfs.h> 64 #include <sys/dmu.h> 65 #include <sys/dmu_objset.h> 66 #include <sys/spa.h> 67 #include <sys/txg.h> 68 #include <sys/dbuf.h> 69 #include <sys/zap.h> 70 #include <sys/sa.h> 71 #include <sys/dirent.h> 72 #include <sys/policy.h> 73 #include <sys/sunddi.h> 74 #include <sys/filio.h> 75 #include <sys/sid.h> 76 #include "fs/fs_subr.h" 77 #include <sys/zfs_ctldir.h> 78 #include <sys/zfs_fuid.h> 79 #include <sys/zfs_sa.h> 80 #include <sys/dnlc.h> 81 #include <sys/zfs_rlock.h> 82 #include <sys/extdirent.h> 83 #include <sys/kidmap.h> 84 #include <sys/cred.h> 85 #include <sys/attr.h> 86 #include <sys/zil.h> 87 88 /* 89 * Programming rules. 90 * 91 * Each vnode op performs some logical unit of work. To do this, the ZPL must 92 * properly lock its in-core state, create a DMU transaction, do the work, 93 * record this work in the intent log (ZIL), commit the DMU transaction, 94 * and wait for the intent log to commit if it is a synchronous operation. 95 * Moreover, the vnode ops must work in both normal and log replay context. 96 * The ordering of events is important to avoid deadlocks and references 97 * to freed memory. The example below illustrates the following Big Rules: 98 * 99 * (1) A check must be made in each zfs thread for a mounted file system. 100 * This is done avoiding races using ZFS_ENTER(zfsvfs). 101 * A ZFS_EXIT(zfsvfs) is needed before all returns. Any znodes 102 * must be checked with ZFS_VERIFY_ZP(zp). Both of these macros 103 * can return EIO from the calling function. 104 * 105 * (2) VN_RELE() should always be the last thing except for zil_commit() 106 * (if necessary) and ZFS_EXIT(). This is for 3 reasons: 107 * First, if it's the last reference, the vnode/znode 108 * can be freed, so the zp may point to freed memory. Second, the last 109 * reference will call zfs_zinactive(), which may induce a lot of work -- 110 * pushing cached pages (which acquires range locks) and syncing out 111 * cached atime changes. Third, zfs_zinactive() may require a new tx, 112 * which could deadlock the system if you were already holding one. 113 * If you must call VN_RELE() within a tx then use VN_RELE_ASYNC(). 114 * 115 * (3) All range locks must be grabbed before calling dmu_tx_assign(), 116 * as they can span dmu_tx_assign() calls. 117 * 118 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to 119 * dmu_tx_assign(). This is critical because we don't want to block 120 * while holding locks. 121 * 122 * If no ZPL locks are held (aside from ZFS_ENTER()), use TXG_WAIT. This 123 * reduces lock contention and CPU usage when we must wait (note that if 124 * throughput is constrained by the storage, nearly every transaction 125 * must wait). 126 * 127 * Note, in particular, that if a lock is sometimes acquired before 128 * the tx assigns, and sometimes after (e.g. z_lock), then failing 129 * to use a non-blocking assign can deadlock the system. The scenario: 130 * 131 * Thread A has grabbed a lock before calling dmu_tx_assign(). 132 * Thread B is in an already-assigned tx, and blocks for this lock. 133 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open() 134 * forever, because the previous txg can't quiesce until B's tx commits. 135 * 136 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT, 137 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent 138 * calls to dmu_tx_assign(), pass TXG_NOTHROTTLE in addition to TXG_NOWAIT, 139 * to indicate that this operation has already called dmu_tx_wait(). 140 * This will ensure that we don't retry forever, waiting a short bit 141 * each time. 142 * 143 * (5) If the operation succeeded, generate the intent log entry for it 144 * before dropping locks. This ensures that the ordering of events 145 * in the intent log matches the order in which they actually occurred. 146 * During ZIL replay the zfs_log_* functions will update the sequence 147 * number to indicate the zil transaction has replayed. 148 * 149 * (6) At the end of each vnode op, the DMU tx must always commit, 150 * regardless of whether there were any errors. 151 * 152 * (7) After dropping all locks, invoke zil_commit(zilog, foid) 153 * to ensure that synchronous semantics are provided when necessary. 154 * 155 * In general, this is how things should be ordered in each vnode op: 156 * 157 * ZFS_ENTER(zfsvfs); // exit if unmounted 158 * top: 159 * zfs_dirent_lock(&dl, ...) // lock directory entry (may VN_HOLD()) 160 * rw_enter(...); // grab any other locks you need 161 * tx = dmu_tx_create(...); // get DMU tx 162 * dmu_tx_hold_*(); // hold each object you might modify 163 * error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT); 164 * if (error) { 165 * rw_exit(...); // drop locks 166 * zfs_dirent_unlock(dl); // unlock directory entry 167 * VN_RELE(...); // release held vnodes 168 * if (error == ERESTART) { 169 * waited = B_TRUE; 170 * dmu_tx_wait(tx); 171 * dmu_tx_abort(tx); 172 * goto top; 173 * } 174 * dmu_tx_abort(tx); // abort DMU tx 175 * ZFS_EXIT(zfsvfs); // finished in zfs 176 * return (error); // really out of space 177 * } 178 * error = do_real_work(); // do whatever this VOP does 179 * if (error == 0) 180 * zfs_log_*(...); // on success, make ZIL entry 181 * dmu_tx_commit(tx); // commit DMU tx -- error or not 182 * rw_exit(...); // drop locks 183 * zfs_dirent_unlock(dl); // unlock directory entry 184 * VN_RELE(...); // release held vnodes 185 * zil_commit(zilog, foid); // synchronous when necessary 186 * ZFS_EXIT(zfsvfs); // finished in zfs 187 * return (error); // done, report error 188 */ 189 190 /* ARGSUSED */ 191 static int 192 zfs_open(vnode_t **vpp, int flag, cred_t *cr, caller_context_t *ct) 193 { 194 znode_t *zp = VTOZ(*vpp); 195 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 196 197 ZFS_ENTER(zfsvfs); 198 ZFS_VERIFY_ZP(zp); 199 200 if ((flag & FWRITE) && (zp->z_pflags & ZFS_APPENDONLY) && 201 ((flag & FAPPEND) == 0)) { 202 ZFS_EXIT(zfsvfs); 203 return (SET_ERROR(EPERM)); 204 } 205 206 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 207 ZTOV(zp)->v_type == VREG && 208 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) { 209 if (fs_vscan(*vpp, cr, 0) != 0) { 210 ZFS_EXIT(zfsvfs); 211 return (SET_ERROR(EACCES)); 212 } 213 } 214 215 /* Keep a count of the synchronous opens in the znode */ 216 if (flag & (FSYNC | FDSYNC)) 217 atomic_inc_32(&zp->z_sync_cnt); 218 219 ZFS_EXIT(zfsvfs); 220 return (0); 221 } 222 223 /* ARGSUSED */ 224 static int 225 zfs_close(vnode_t *vp, int flag, int count, offset_t offset, cred_t *cr, 226 caller_context_t *ct) 227 { 228 znode_t *zp = VTOZ(vp); 229 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 230 231 /* 232 * Clean up any locks held by this process on the vp. 233 */ 234 cleanlocks(vp, ddi_get_pid(), 0); 235 cleanshares(vp, ddi_get_pid()); 236 237 ZFS_ENTER(zfsvfs); 238 ZFS_VERIFY_ZP(zp); 239 240 /* Decrement the synchronous opens in the znode */ 241 if ((flag & (FSYNC | FDSYNC)) && (count == 1)) 242 atomic_dec_32(&zp->z_sync_cnt); 243 244 if (!zfs_has_ctldir(zp) && zp->z_zfsvfs->z_vscan && 245 ZTOV(zp)->v_type == VREG && 246 !(zp->z_pflags & ZFS_AV_QUARANTINED) && zp->z_size > 0) 247 VERIFY(fs_vscan(vp, cr, 1) == 0); 248 249 ZFS_EXIT(zfsvfs); 250 return (0); 251 } 252 253 /* 254 * Lseek support for finding holes (cmd == _FIO_SEEK_HOLE) and 255 * data (cmd == _FIO_SEEK_DATA). "off" is an in/out parameter. 256 */ 257 static int 258 zfs_holey(vnode_t *vp, int cmd, offset_t *off) 259 { 260 znode_t *zp = VTOZ(vp); 261 uint64_t noff = (uint64_t)*off; /* new offset */ 262 uint64_t file_sz; 263 int error; 264 boolean_t hole; 265 266 file_sz = zp->z_size; 267 if (noff >= file_sz) { 268 return (SET_ERROR(ENXIO)); 269 } 270 271 if (cmd == _FIO_SEEK_HOLE) 272 hole = B_TRUE; 273 else 274 hole = B_FALSE; 275 276 error = dmu_offset_next(zp->z_zfsvfs->z_os, zp->z_id, hole, &noff); 277 278 if (error == ESRCH) 279 return (SET_ERROR(ENXIO)); 280 281 /* 282 * We could find a hole that begins after the logical end-of-file, 283 * because dmu_offset_next() only works on whole blocks. If the 284 * EOF falls mid-block, then indicate that the "virtual hole" 285 * at the end of the file begins at the logical EOF, rather than 286 * at the end of the last block. 287 */ 288 if (noff > file_sz) { 289 ASSERT(hole); 290 noff = file_sz; 291 } 292 293 if (noff < *off) 294 return (error); 295 *off = noff; 296 return (error); 297 } 298 299 /* ARGSUSED */ 300 static int 301 zfs_ioctl(vnode_t *vp, int com, intptr_t data, int flag, cred_t *cred, 302 int *rvalp, caller_context_t *ct) 303 { 304 offset_t off; 305 offset_t ndata; 306 dmu_object_info_t doi; 307 int error; 308 zfsvfs_t *zfsvfs; 309 znode_t *zp; 310 311 switch (com) { 312 case _FIOFFS: 313 { 314 return (zfs_sync(vp->v_vfsp, 0, cred)); 315 316 /* 317 * The following two ioctls are used by bfu. Faking out, 318 * necessary to avoid bfu errors. 319 */ 320 } 321 case _FIOGDIO: 322 case _FIOSDIO: 323 { 324 return (0); 325 } 326 327 case _FIO_SEEK_DATA: 328 case _FIO_SEEK_HOLE: 329 { 330 if (ddi_copyin((void *)data, &off, sizeof (off), flag)) 331 return (SET_ERROR(EFAULT)); 332 333 zp = VTOZ(vp); 334 zfsvfs = zp->z_zfsvfs; 335 ZFS_ENTER(zfsvfs); 336 ZFS_VERIFY_ZP(zp); 337 338 /* offset parameter is in/out */ 339 error = zfs_holey(vp, com, &off); 340 ZFS_EXIT(zfsvfs); 341 if (error) 342 return (error); 343 if (ddi_copyout(&off, (void *)data, sizeof (off), flag)) 344 return (SET_ERROR(EFAULT)); 345 return (0); 346 } 347 case _FIO_COUNT_FILLED: 348 { 349 /* 350 * _FIO_COUNT_FILLED adds a new ioctl command which 351 * exposes the number of filled blocks in a 352 * ZFS object. 353 */ 354 zp = VTOZ(vp); 355 zfsvfs = zp->z_zfsvfs; 356 ZFS_ENTER(zfsvfs); 357 ZFS_VERIFY_ZP(zp); 358 359 /* 360 * Wait for all dirty blocks for this object 361 * to get synced out to disk, and the DMU info 362 * updated. 363 */ 364 error = dmu_object_wait_synced(zfsvfs->z_os, zp->z_id); 365 if (error) { 366 ZFS_EXIT(zfsvfs); 367 return (error); 368 } 369 370 /* 371 * Retrieve fill count from DMU object. 372 */ 373 error = dmu_object_info(zfsvfs->z_os, zp->z_id, &doi); 374 if (error) { 375 ZFS_EXIT(zfsvfs); 376 return (error); 377 } 378 379 ndata = doi.doi_fill_count; 380 381 ZFS_EXIT(zfsvfs); 382 if (ddi_copyout(&ndata, (void *)data, sizeof (ndata), flag)) 383 return (SET_ERROR(EFAULT)); 384 return (0); 385 } 386 } 387 return (SET_ERROR(ENOTTY)); 388 } 389 390 /* 391 * Utility functions to map and unmap a single physical page. These 392 * are used to manage the mappable copies of ZFS file data, and therefore 393 * do not update ref/mod bits. 394 */ 395 caddr_t 396 zfs_map_page(page_t *pp, enum seg_rw rw) 397 { 398 if (kpm_enable) 399 return (hat_kpm_mapin(pp, 0)); 400 ASSERT(rw == S_READ || rw == S_WRITE); 401 return (ppmapin(pp, PROT_READ | ((rw == S_WRITE) ? PROT_WRITE : 0), 402 (caddr_t)-1)); 403 } 404 405 void 406 zfs_unmap_page(page_t *pp, caddr_t addr) 407 { 408 if (kpm_enable) { 409 hat_kpm_mapout(pp, 0, addr); 410 } else { 411 ppmapout(addr); 412 } 413 } 414 415 /* 416 * When a file is memory mapped, we must keep the IO data synchronized 417 * between the DMU cache and the memory mapped pages. What this means: 418 * 419 * On Write: If we find a memory mapped page, we write to *both* 420 * the page and the dmu buffer. 421 */ 422 static void 423 update_pages(vnode_t *vp, int64_t start, int len, objset_t *os, uint64_t oid) 424 { 425 int64_t off; 426 427 off = start & PAGEOFFSET; 428 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 429 page_t *pp; 430 uint64_t nbytes = MIN(PAGESIZE - off, len); 431 432 if (pp = page_lookup(vp, start, SE_SHARED)) { 433 caddr_t va; 434 435 va = zfs_map_page(pp, S_WRITE); 436 (void) dmu_read(os, oid, start+off, nbytes, va+off, 437 DMU_READ_PREFETCH); 438 zfs_unmap_page(pp, va); 439 page_unlock(pp); 440 } 441 len -= nbytes; 442 off = 0; 443 } 444 } 445 446 /* 447 * When a file is memory mapped, we must keep the IO data synchronized 448 * between the DMU cache and the memory mapped pages. What this means: 449 * 450 * On Read: We "read" preferentially from memory mapped pages, 451 * else we default from the dmu buffer. 452 * 453 * NOTE: We will always "break up" the IO into PAGESIZE uiomoves when 454 * the file is memory mapped. 455 */ 456 static int 457 mappedread(vnode_t *vp, int nbytes, uio_t *uio) 458 { 459 znode_t *zp = VTOZ(vp); 460 int64_t start, off; 461 int len = nbytes; 462 int error = 0; 463 464 start = uio->uio_loffset; 465 off = start & PAGEOFFSET; 466 for (start &= PAGEMASK; len > 0; start += PAGESIZE) { 467 page_t *pp; 468 uint64_t bytes = MIN(PAGESIZE - off, len); 469 470 if (pp = page_lookup(vp, start, SE_SHARED)) { 471 caddr_t va; 472 473 va = zfs_map_page(pp, S_READ); 474 error = uiomove(va + off, bytes, UIO_READ, uio); 475 zfs_unmap_page(pp, va); 476 page_unlock(pp); 477 } else { 478 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl), 479 uio, bytes); 480 } 481 len -= bytes; 482 off = 0; 483 if (error) 484 break; 485 } 486 return (error); 487 } 488 489 offset_t zfs_read_chunk_size = 1024 * 1024; /* Tunable */ 490 491 /* 492 * Read bytes from specified file into supplied buffer. 493 * 494 * IN: vp - vnode of file to be read from. 495 * uio - structure supplying read location, range info, 496 * and return buffer. 497 * ioflag - SYNC flags; used to provide FRSYNC semantics. 498 * cr - credentials of caller. 499 * ct - caller context 500 * 501 * OUT: uio - updated offset and range, buffer filled. 502 * 503 * RETURN: 0 on success, error code on failure. 504 * 505 * Side Effects: 506 * vp - atime updated if byte count > 0 507 */ 508 /* ARGSUSED */ 509 static int 510 zfs_read(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 511 { 512 znode_t *zp = VTOZ(vp); 513 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 514 ssize_t n, nbytes; 515 int error = 0; 516 rl_t *rl; 517 xuio_t *xuio = NULL; 518 519 ZFS_ENTER(zfsvfs); 520 ZFS_VERIFY_ZP(zp); 521 522 if (zp->z_pflags & ZFS_AV_QUARANTINED) { 523 ZFS_EXIT(zfsvfs); 524 return (SET_ERROR(EACCES)); 525 } 526 527 /* 528 * Validate file offset 529 */ 530 if (uio->uio_loffset < (offset_t)0) { 531 ZFS_EXIT(zfsvfs); 532 return (SET_ERROR(EINVAL)); 533 } 534 535 /* 536 * Fasttrack empty reads 537 */ 538 if (uio->uio_resid == 0) { 539 ZFS_EXIT(zfsvfs); 540 return (0); 541 } 542 543 /* 544 * Check for mandatory locks 545 */ 546 if (MANDMODE(zp->z_mode)) { 547 if (error = chklock(vp, FREAD, 548 uio->uio_loffset, uio->uio_resid, uio->uio_fmode, ct)) { 549 ZFS_EXIT(zfsvfs); 550 return (error); 551 } 552 } 553 554 /* 555 * If we're in FRSYNC mode, sync out this znode before reading it. 556 */ 557 if (ioflag & FRSYNC || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 558 zil_commit(zfsvfs->z_log, zp->z_id); 559 560 /* 561 * Lock the range against changes. 562 */ 563 rl = zfs_range_lock(zp, uio->uio_loffset, uio->uio_resid, RL_READER); 564 565 /* 566 * If we are reading past end-of-file we can skip 567 * to the end; but we might still need to set atime. 568 */ 569 if (uio->uio_loffset >= zp->z_size) { 570 error = 0; 571 goto out; 572 } 573 574 ASSERT(uio->uio_loffset < zp->z_size); 575 n = MIN(uio->uio_resid, zp->z_size - uio->uio_loffset); 576 577 if ((uio->uio_extflg == UIO_XUIO) && 578 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) { 579 int nblk; 580 int blksz = zp->z_blksz; 581 uint64_t offset = uio->uio_loffset; 582 583 xuio = (xuio_t *)uio; 584 if ((ISP2(blksz))) { 585 nblk = (P2ROUNDUP(offset + n, blksz) - P2ALIGN(offset, 586 blksz)) / blksz; 587 } else { 588 ASSERT(offset + n <= blksz); 589 nblk = 1; 590 } 591 (void) dmu_xuio_init(xuio, nblk); 592 593 if (vn_has_cached_data(vp)) { 594 /* 595 * For simplicity, we always allocate a full buffer 596 * even if we only expect to read a portion of a block. 597 */ 598 while (--nblk >= 0) { 599 (void) dmu_xuio_add(xuio, 600 dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 601 blksz), 0, blksz); 602 } 603 } 604 } 605 606 while (n > 0) { 607 nbytes = MIN(n, zfs_read_chunk_size - 608 P2PHASE(uio->uio_loffset, zfs_read_chunk_size)); 609 610 if (vn_has_cached_data(vp)) { 611 error = mappedread(vp, nbytes, uio); 612 } else { 613 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl), 614 uio, nbytes); 615 } 616 if (error) { 617 /* convert checksum errors into IO errors */ 618 if (error == ECKSUM) 619 error = SET_ERROR(EIO); 620 break; 621 } 622 623 n -= nbytes; 624 } 625 out: 626 zfs_range_unlock(rl); 627 628 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 629 ZFS_EXIT(zfsvfs); 630 return (error); 631 } 632 633 /* 634 * Write the bytes to a file. 635 * 636 * IN: vp - vnode of file to be written to. 637 * uio - structure supplying write location, range info, 638 * and data buffer. 639 * ioflag - FAPPEND, FSYNC, and/or FDSYNC. FAPPEND is 640 * set if in append mode. 641 * cr - credentials of caller. 642 * ct - caller context (NFS/CIFS fem monitor only) 643 * 644 * OUT: uio - updated offset and range. 645 * 646 * RETURN: 0 on success, error code on failure. 647 * 648 * Timestamps: 649 * vp - ctime|mtime updated if byte count > 0 650 */ 651 652 /* ARGSUSED */ 653 static int 654 zfs_write(vnode_t *vp, uio_t *uio, int ioflag, cred_t *cr, caller_context_t *ct) 655 { 656 znode_t *zp = VTOZ(vp); 657 rlim64_t limit = uio->uio_llimit; 658 ssize_t start_resid = uio->uio_resid; 659 ssize_t tx_bytes; 660 uint64_t end_size; 661 dmu_tx_t *tx; 662 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 663 zilog_t *zilog; 664 offset_t woff; 665 ssize_t n, nbytes; 666 rl_t *rl; 667 int max_blksz = zfsvfs->z_max_blksz; 668 int error = 0; 669 arc_buf_t *abuf; 670 iovec_t *aiov = NULL; 671 xuio_t *xuio = NULL; 672 int i_iov = 0; 673 int iovcnt = uio->uio_iovcnt; 674 iovec_t *iovp = uio->uio_iov; 675 int write_eof; 676 int count = 0; 677 sa_bulk_attr_t bulk[4]; 678 uint64_t mtime[2], ctime[2]; 679 680 /* 681 * Fasttrack empty write 682 */ 683 n = start_resid; 684 if (n == 0) 685 return (0); 686 687 if (limit == RLIM64_INFINITY || limit > MAXOFFSET_T) 688 limit = MAXOFFSET_T; 689 690 ZFS_ENTER(zfsvfs); 691 ZFS_VERIFY_ZP(zp); 692 693 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 694 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 695 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 696 &zp->z_size, 8); 697 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 698 &zp->z_pflags, 8); 699 700 /* 701 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our 702 * callers might not be able to detect properly that we are read-only, 703 * so check it explicitly here. 704 */ 705 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 706 ZFS_EXIT(zfsvfs); 707 return (SET_ERROR(EROFS)); 708 } 709 710 /* 711 * If immutable or not appending then return EPERM. 712 * Intentionally allow ZFS_READONLY through here. 713 * See zfs_zaccess_common() 714 */ 715 if ((zp->z_pflags & ZFS_IMMUTABLE) || 716 ((zp->z_pflags & ZFS_APPENDONLY) && !(ioflag & FAPPEND) && 717 (uio->uio_loffset < zp->z_size))) { 718 ZFS_EXIT(zfsvfs); 719 return (SET_ERROR(EPERM)); 720 } 721 722 zilog = zfsvfs->z_log; 723 724 /* 725 * Validate file offset 726 */ 727 woff = ioflag & FAPPEND ? zp->z_size : uio->uio_loffset; 728 if (woff < 0) { 729 ZFS_EXIT(zfsvfs); 730 return (SET_ERROR(EINVAL)); 731 } 732 733 /* 734 * Check for mandatory locks before calling zfs_range_lock() 735 * in order to prevent a deadlock with locks set via fcntl(). 736 */ 737 if (MANDMODE((mode_t)zp->z_mode) && 738 (error = chklock(vp, FWRITE, woff, n, uio->uio_fmode, ct)) != 0) { 739 ZFS_EXIT(zfsvfs); 740 return (error); 741 } 742 743 /* 744 * Pre-fault the pages to ensure slow (eg NFS) pages 745 * don't hold up txg. 746 * Skip this if uio contains loaned arc_buf. 747 */ 748 if ((uio->uio_extflg == UIO_XUIO) && 749 (((xuio_t *)uio)->xu_type == UIOTYPE_ZEROCOPY)) 750 xuio = (xuio_t *)uio; 751 else 752 uio_prefaultpages(MIN(n, max_blksz), uio); 753 754 /* 755 * If in append mode, set the io offset pointer to eof. 756 */ 757 if (ioflag & FAPPEND) { 758 /* 759 * Obtain an appending range lock to guarantee file append 760 * semantics. We reset the write offset once we have the lock. 761 */ 762 rl = zfs_range_lock(zp, 0, n, RL_APPEND); 763 woff = rl->r_off; 764 if (rl->r_len == UINT64_MAX) { 765 /* 766 * We overlocked the file because this write will cause 767 * the file block size to increase. 768 * Note that zp_size cannot change with this lock held. 769 */ 770 woff = zp->z_size; 771 } 772 uio->uio_loffset = woff; 773 } else { 774 /* 775 * Note that if the file block size will change as a result of 776 * this write, then this range lock will lock the entire file 777 * so that we can re-write the block safely. 778 */ 779 rl = zfs_range_lock(zp, woff, n, RL_WRITER); 780 } 781 782 if (woff >= limit) { 783 zfs_range_unlock(rl); 784 ZFS_EXIT(zfsvfs); 785 return (SET_ERROR(EFBIG)); 786 } 787 788 if ((woff + n) > limit || woff > (limit - n)) 789 n = limit - woff; 790 791 /* Will this write extend the file length? */ 792 write_eof = (woff + n > zp->z_size); 793 794 end_size = MAX(zp->z_size, woff + n); 795 796 /* 797 * Write the file in reasonable size chunks. Each chunk is written 798 * in a separate transaction; this keeps the intent log records small 799 * and allows us to do more fine-grained space accounting. 800 */ 801 while (n > 0) { 802 abuf = NULL; 803 woff = uio->uio_loffset; 804 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 805 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 806 if (abuf != NULL) 807 dmu_return_arcbuf(abuf); 808 error = SET_ERROR(EDQUOT); 809 break; 810 } 811 812 if (xuio && abuf == NULL) { 813 ASSERT(i_iov < iovcnt); 814 aiov = &iovp[i_iov]; 815 abuf = dmu_xuio_arcbuf(xuio, i_iov); 816 dmu_xuio_clear(xuio, i_iov); 817 DTRACE_PROBE3(zfs_cp_write, int, i_iov, 818 iovec_t *, aiov, arc_buf_t *, abuf); 819 ASSERT((aiov->iov_base == abuf->b_data) || 820 ((char *)aiov->iov_base - (char *)abuf->b_data + 821 aiov->iov_len == arc_buf_size(abuf))); 822 i_iov++; 823 } else if (abuf == NULL && n >= max_blksz && 824 woff >= zp->z_size && 825 P2PHASE(woff, max_blksz) == 0 && 826 zp->z_blksz == max_blksz) { 827 /* 828 * This write covers a full block. "Borrow" a buffer 829 * from the dmu so that we can fill it before we enter 830 * a transaction. This avoids the possibility of 831 * holding up the transaction if the data copy hangs 832 * up on a pagefault (e.g., from an NFS server mapping). 833 */ 834 size_t cbytes; 835 836 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 837 max_blksz); 838 ASSERT(abuf != NULL); 839 ASSERT(arc_buf_size(abuf) == max_blksz); 840 if (error = uiocopy(abuf->b_data, max_blksz, 841 UIO_WRITE, uio, &cbytes)) { 842 dmu_return_arcbuf(abuf); 843 break; 844 } 845 ASSERT(cbytes == max_blksz); 846 } 847 848 /* 849 * Start a transaction. 850 */ 851 tx = dmu_tx_create(zfsvfs->z_os); 852 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 853 dmu_tx_hold_write(tx, zp->z_id, woff, MIN(n, max_blksz)); 854 zfs_sa_upgrade_txholds(tx, zp); 855 error = dmu_tx_assign(tx, TXG_WAIT); 856 if (error) { 857 dmu_tx_abort(tx); 858 if (abuf != NULL) 859 dmu_return_arcbuf(abuf); 860 break; 861 } 862 863 /* 864 * If zfs_range_lock() over-locked we grow the blocksize 865 * and then reduce the lock range. This will only happen 866 * on the first iteration since zfs_range_reduce() will 867 * shrink down r_len to the appropriate size. 868 */ 869 if (rl->r_len == UINT64_MAX) { 870 uint64_t new_blksz; 871 872 if (zp->z_blksz > max_blksz) { 873 /* 874 * File's blocksize is already larger than the 875 * "recordsize" property. Only let it grow to 876 * the next power of 2. 877 */ 878 ASSERT(!ISP2(zp->z_blksz)); 879 new_blksz = MIN(end_size, 880 1 << highbit64(zp->z_blksz)); 881 } else { 882 new_blksz = MIN(end_size, max_blksz); 883 } 884 zfs_grow_blocksize(zp, new_blksz, tx); 885 zfs_range_reduce(rl, woff, n); 886 } 887 888 /* 889 * XXX - should we really limit each write to z_max_blksz? 890 * Perhaps we should use SPA_MAXBLOCKSIZE chunks? 891 */ 892 nbytes = MIN(n, max_blksz - P2PHASE(woff, max_blksz)); 893 894 if (abuf == NULL) { 895 tx_bytes = uio->uio_resid; 896 error = dmu_write_uio_dbuf(sa_get_db(zp->z_sa_hdl), 897 uio, nbytes, tx); 898 tx_bytes -= uio->uio_resid; 899 } else { 900 tx_bytes = nbytes; 901 ASSERT(xuio == NULL || tx_bytes == aiov->iov_len); 902 /* 903 * If this is not a full block write, but we are 904 * extending the file past EOF and this data starts 905 * block-aligned, use assign_arcbuf(). Otherwise, 906 * write via dmu_write(). 907 */ 908 if (tx_bytes < max_blksz && (!write_eof || 909 aiov->iov_base != abuf->b_data)) { 910 ASSERT(xuio); 911 dmu_write(zfsvfs->z_os, zp->z_id, woff, 912 aiov->iov_len, aiov->iov_base, tx); 913 dmu_return_arcbuf(abuf); 914 xuio_stat_wbuf_copied(); 915 } else { 916 ASSERT(xuio || tx_bytes == max_blksz); 917 dmu_assign_arcbuf(sa_get_db(zp->z_sa_hdl), 918 woff, abuf, tx); 919 } 920 ASSERT(tx_bytes <= uio->uio_resid); 921 uioskip(uio, tx_bytes); 922 } 923 if (tx_bytes && vn_has_cached_data(vp)) { 924 update_pages(vp, woff, 925 tx_bytes, zfsvfs->z_os, zp->z_id); 926 } 927 928 /* 929 * If we made no progress, we're done. If we made even 930 * partial progress, update the znode and ZIL accordingly. 931 */ 932 if (tx_bytes == 0) { 933 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 934 (void *)&zp->z_size, sizeof (uint64_t), tx); 935 dmu_tx_commit(tx); 936 ASSERT(error != 0); 937 break; 938 } 939 940 /* 941 * Clear Set-UID/Set-GID bits on successful write if not 942 * privileged and at least one of the excute bits is set. 943 * 944 * It would be nice to to this after all writes have 945 * been done, but that would still expose the ISUID/ISGID 946 * to another app after the partial write is committed. 947 * 948 * Note: we don't call zfs_fuid_map_id() here because 949 * user 0 is not an ephemeral uid. 950 */ 951 mutex_enter(&zp->z_acl_lock); 952 if ((zp->z_mode & (S_IXUSR | (S_IXUSR >> 3) | 953 (S_IXUSR >> 6))) != 0 && 954 (zp->z_mode & (S_ISUID | S_ISGID)) != 0 && 955 secpolicy_vnode_setid_retain(cr, 956 (zp->z_mode & S_ISUID) != 0 && zp->z_uid == 0) != 0) { 957 uint64_t newmode; 958 zp->z_mode &= ~(S_ISUID | S_ISGID); 959 newmode = zp->z_mode; 960 (void) sa_update(zp->z_sa_hdl, SA_ZPL_MODE(zfsvfs), 961 (void *)&newmode, sizeof (uint64_t), tx); 962 } 963 mutex_exit(&zp->z_acl_lock); 964 965 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 966 B_TRUE); 967 968 /* 969 * Update the file size (zp_size) if it has changed; 970 * account for possible concurrent updates. 971 */ 972 while ((end_size = zp->z_size) < uio->uio_loffset) { 973 (void) atomic_cas_64(&zp->z_size, end_size, 974 uio->uio_loffset); 975 ASSERT(error == 0); 976 } 977 /* 978 * If we are replaying and eof is non zero then force 979 * the file size to the specified eof. Note, there's no 980 * concurrency during replay. 981 */ 982 if (zfsvfs->z_replay && zfsvfs->z_replay_eof != 0) 983 zp->z_size = zfsvfs->z_replay_eof; 984 985 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 986 987 zfs_log_write(zilog, tx, TX_WRITE, zp, woff, tx_bytes, ioflag); 988 dmu_tx_commit(tx); 989 990 if (error != 0) 991 break; 992 ASSERT(tx_bytes == nbytes); 993 n -= nbytes; 994 995 if (!xuio && n > 0) 996 uio_prefaultpages(MIN(n, max_blksz), uio); 997 } 998 999 zfs_range_unlock(rl); 1000 1001 /* 1002 * If we're in replay mode, or we made no progress, return error. 1003 * Otherwise, it's at least a partial write, so it's successful. 1004 */ 1005 if (zfsvfs->z_replay || uio->uio_resid == start_resid) { 1006 ZFS_EXIT(zfsvfs); 1007 return (error); 1008 } 1009 1010 if (ioflag & (FSYNC | FDSYNC) || 1011 zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1012 zil_commit(zilog, zp->z_id); 1013 1014 ZFS_EXIT(zfsvfs); 1015 return (0); 1016 } 1017 1018 /* ARGSUSED */ 1019 void 1020 zfs_get_done(zgd_t *zgd, int error) 1021 { 1022 znode_t *zp = zgd->zgd_private; 1023 objset_t *os = zp->z_zfsvfs->z_os; 1024 1025 if (zgd->zgd_db) 1026 dmu_buf_rele(zgd->zgd_db, zgd); 1027 1028 zfs_range_unlock(zgd->zgd_rl); 1029 1030 /* 1031 * Release the vnode asynchronously as we currently have the 1032 * txg stopped from syncing. 1033 */ 1034 VN_RELE_ASYNC(ZTOV(zp), dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 1035 1036 kmem_free(zgd, sizeof (zgd_t)); 1037 } 1038 1039 #ifdef DEBUG 1040 static int zil_fault_io = 0; 1041 #endif 1042 1043 /* 1044 * Get data to generate a TX_WRITE intent log record. 1045 */ 1046 int 1047 zfs_get_data(void *arg, lr_write_t *lr, char *buf, struct lwb *lwb, zio_t *zio) 1048 { 1049 zfsvfs_t *zfsvfs = arg; 1050 objset_t *os = zfsvfs->z_os; 1051 znode_t *zp; 1052 uint64_t object = lr->lr_foid; 1053 uint64_t offset = lr->lr_offset; 1054 uint64_t size = lr->lr_length; 1055 dmu_buf_t *db; 1056 zgd_t *zgd; 1057 int error = 0; 1058 1059 ASSERT3P(lwb, !=, NULL); 1060 ASSERT3P(zio, !=, NULL); 1061 ASSERT3U(size, !=, 0); 1062 1063 /* 1064 * Nothing to do if the file has been removed 1065 */ 1066 if (zfs_zget(zfsvfs, object, &zp) != 0) 1067 return (SET_ERROR(ENOENT)); 1068 if (zp->z_unlinked) { 1069 /* 1070 * Release the vnode asynchronously as we currently have the 1071 * txg stopped from syncing. 1072 */ 1073 VN_RELE_ASYNC(ZTOV(zp), 1074 dsl_pool_vnrele_taskq(dmu_objset_pool(os))); 1075 return (SET_ERROR(ENOENT)); 1076 } 1077 1078 zgd = (zgd_t *)kmem_zalloc(sizeof (zgd_t), KM_SLEEP); 1079 zgd->zgd_lwb = lwb; 1080 zgd->zgd_private = zp; 1081 1082 /* 1083 * Write records come in two flavors: immediate and indirect. 1084 * For small writes it's cheaper to store the data with the 1085 * log record (immediate); for large writes it's cheaper to 1086 * sync the data and get a pointer to it (indirect) so that 1087 * we don't have to write the data twice. 1088 */ 1089 if (buf != NULL) { /* immediate write */ 1090 zgd->zgd_rl = zfs_range_lock(zp, offset, size, RL_READER); 1091 /* test for truncation needs to be done while range locked */ 1092 if (offset >= zp->z_size) { 1093 error = SET_ERROR(ENOENT); 1094 } else { 1095 error = dmu_read(os, object, offset, size, buf, 1096 DMU_READ_NO_PREFETCH); 1097 } 1098 ASSERT(error == 0 || error == ENOENT); 1099 } else { /* indirect write */ 1100 /* 1101 * Have to lock the whole block to ensure when it's 1102 * written out and its checksum is being calculated 1103 * that no one can change the data. We need to re-check 1104 * blocksize after we get the lock in case it's changed! 1105 */ 1106 for (;;) { 1107 uint64_t blkoff; 1108 size = zp->z_blksz; 1109 blkoff = ISP2(size) ? P2PHASE(offset, size) : offset; 1110 offset -= blkoff; 1111 zgd->zgd_rl = zfs_range_lock(zp, offset, size, 1112 RL_READER); 1113 if (zp->z_blksz == size) 1114 break; 1115 offset += blkoff; 1116 zfs_range_unlock(zgd->zgd_rl); 1117 } 1118 /* test for truncation needs to be done while range locked */ 1119 if (lr->lr_offset >= zp->z_size) 1120 error = SET_ERROR(ENOENT); 1121 #ifdef DEBUG 1122 if (zil_fault_io) { 1123 error = SET_ERROR(EIO); 1124 zil_fault_io = 0; 1125 } 1126 #endif 1127 if (error == 0) 1128 error = dmu_buf_hold(os, object, offset, zgd, &db, 1129 DMU_READ_NO_PREFETCH); 1130 1131 if (error == 0) { 1132 blkptr_t *bp = &lr->lr_blkptr; 1133 1134 zgd->zgd_db = db; 1135 zgd->zgd_bp = bp; 1136 1137 ASSERT(db->db_offset == offset); 1138 ASSERT(db->db_size == size); 1139 1140 error = dmu_sync(zio, lr->lr_common.lrc_txg, 1141 zfs_get_done, zgd); 1142 ASSERT(error || lr->lr_length <= size); 1143 1144 /* 1145 * On success, we need to wait for the write I/O 1146 * initiated by dmu_sync() to complete before we can 1147 * release this dbuf. We will finish everything up 1148 * in the zfs_get_done() callback. 1149 */ 1150 if (error == 0) 1151 return (0); 1152 1153 if (error == EALREADY) { 1154 lr->lr_common.lrc_txtype = TX_WRITE2; 1155 /* 1156 * TX_WRITE2 relies on the data previously 1157 * written by the TX_WRITE that caused 1158 * EALREADY. We zero out the BP because 1159 * it is the old, currently-on-disk BP. 1160 */ 1161 zgd->zgd_bp = NULL; 1162 BP_ZERO(bp); 1163 error = 0; 1164 } 1165 } 1166 } 1167 1168 zfs_get_done(zgd, error); 1169 1170 return (error); 1171 } 1172 1173 /*ARGSUSED*/ 1174 static int 1175 zfs_access(vnode_t *vp, int mode, int flag, cred_t *cr, 1176 caller_context_t *ct) 1177 { 1178 znode_t *zp = VTOZ(vp); 1179 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 1180 int error; 1181 1182 ZFS_ENTER(zfsvfs); 1183 ZFS_VERIFY_ZP(zp); 1184 1185 if (flag & V_ACE_MASK) 1186 error = zfs_zaccess(zp, mode, flag, B_FALSE, cr); 1187 else 1188 error = zfs_zaccess_rwx(zp, mode, flag, cr); 1189 1190 ZFS_EXIT(zfsvfs); 1191 return (error); 1192 } 1193 1194 /* 1195 * If vnode is for a device return a specfs vnode instead. 1196 */ 1197 static int 1198 specvp_check(vnode_t **vpp, cred_t *cr) 1199 { 1200 int error = 0; 1201 1202 if (IS_DEVVP(*vpp)) { 1203 struct vnode *svp; 1204 1205 svp = specvp(*vpp, (*vpp)->v_rdev, (*vpp)->v_type, cr); 1206 VN_RELE(*vpp); 1207 if (svp == NULL) 1208 error = SET_ERROR(ENOSYS); 1209 *vpp = svp; 1210 } 1211 return (error); 1212 } 1213 1214 1215 /* 1216 * Lookup an entry in a directory, or an extended attribute directory. 1217 * If it exists, return a held vnode reference for it. 1218 * 1219 * IN: dvp - vnode of directory to search. 1220 * nm - name of entry to lookup. 1221 * pnp - full pathname to lookup [UNUSED]. 1222 * flags - LOOKUP_XATTR set if looking for an attribute. 1223 * rdir - root directory vnode [UNUSED]. 1224 * cr - credentials of caller. 1225 * ct - caller context 1226 * direntflags - directory lookup flags 1227 * realpnp - returned pathname. 1228 * 1229 * OUT: vpp - vnode of located entry, NULL if not found. 1230 * 1231 * RETURN: 0 on success, error code on failure. 1232 * 1233 * Timestamps: 1234 * NA 1235 */ 1236 /* ARGSUSED */ 1237 static int 1238 zfs_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, struct pathname *pnp, 1239 int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct, 1240 int *direntflags, pathname_t *realpnp) 1241 { 1242 znode_t *zdp = VTOZ(dvp); 1243 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 1244 int error = 0; 1245 1246 /* 1247 * Fast path lookup, however we must skip DNLC lookup 1248 * for case folding or normalizing lookups because the 1249 * DNLC code only stores the passed in name. This means 1250 * creating 'a' and removing 'A' on a case insensitive 1251 * file system would work, but DNLC still thinks 'a' 1252 * exists and won't let you create it again on the next 1253 * pass through fast path. 1254 */ 1255 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) { 1256 1257 if (dvp->v_type != VDIR) { 1258 return (SET_ERROR(ENOTDIR)); 1259 } else if (zdp->z_sa_hdl == NULL) { 1260 return (SET_ERROR(EIO)); 1261 } 1262 1263 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) { 1264 error = zfs_fastaccesschk_execute(zdp, cr); 1265 if (!error) { 1266 *vpp = dvp; 1267 VN_HOLD(*vpp); 1268 return (0); 1269 } 1270 return (error); 1271 } else if (!zdp->z_zfsvfs->z_norm && 1272 (zdp->z_zfsvfs->z_case == ZFS_CASE_SENSITIVE)) { 1273 1274 vnode_t *tvp = dnlc_lookup(dvp, nm); 1275 1276 if (tvp) { 1277 error = zfs_fastaccesschk_execute(zdp, cr); 1278 if (error) { 1279 VN_RELE(tvp); 1280 return (error); 1281 } 1282 if (tvp == DNLC_NO_VNODE) { 1283 VN_RELE(tvp); 1284 return (SET_ERROR(ENOENT)); 1285 } else { 1286 *vpp = tvp; 1287 return (specvp_check(vpp, cr)); 1288 } 1289 } 1290 } 1291 } 1292 1293 DTRACE_PROBE2(zfs__fastpath__lookup__miss, vnode_t *, dvp, char *, nm); 1294 1295 ZFS_ENTER(zfsvfs); 1296 ZFS_VERIFY_ZP(zdp); 1297 1298 *vpp = NULL; 1299 1300 if (flags & LOOKUP_XATTR) { 1301 /* 1302 * If the xattr property is off, refuse the lookup request. 1303 */ 1304 if (!(zfsvfs->z_vfs->vfs_flag & VFS_XATTR)) { 1305 ZFS_EXIT(zfsvfs); 1306 return (SET_ERROR(EINVAL)); 1307 } 1308 1309 /* 1310 * We don't allow recursive attributes.. 1311 * Maybe someday we will. 1312 */ 1313 if (zdp->z_pflags & ZFS_XATTR) { 1314 ZFS_EXIT(zfsvfs); 1315 return (SET_ERROR(EINVAL)); 1316 } 1317 1318 if (error = zfs_get_xattrdir(VTOZ(dvp), vpp, cr, flags)) { 1319 ZFS_EXIT(zfsvfs); 1320 return (error); 1321 } 1322 1323 /* 1324 * Do we have permission to get into attribute directory? 1325 */ 1326 1327 if (error = zfs_zaccess(VTOZ(*vpp), ACE_EXECUTE, 0, 1328 B_FALSE, cr)) { 1329 VN_RELE(*vpp); 1330 *vpp = NULL; 1331 } 1332 1333 ZFS_EXIT(zfsvfs); 1334 return (error); 1335 } 1336 1337 if (dvp->v_type != VDIR) { 1338 ZFS_EXIT(zfsvfs); 1339 return (SET_ERROR(ENOTDIR)); 1340 } 1341 1342 /* 1343 * Check accessibility of directory. 1344 */ 1345 1346 if (error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr)) { 1347 ZFS_EXIT(zfsvfs); 1348 return (error); 1349 } 1350 1351 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm), 1352 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1353 ZFS_EXIT(zfsvfs); 1354 return (SET_ERROR(EILSEQ)); 1355 } 1356 1357 error = zfs_dirlook(zdp, nm, vpp, flags, direntflags, realpnp); 1358 if (error == 0) 1359 error = specvp_check(vpp, cr); 1360 1361 ZFS_EXIT(zfsvfs); 1362 return (error); 1363 } 1364 1365 /* 1366 * Attempt to create a new entry in a directory. If the entry 1367 * already exists, truncate the file if permissible, else return 1368 * an error. Return the vp of the created or trunc'd file. 1369 * 1370 * IN: dvp - vnode of directory to put new file entry in. 1371 * name - name of new file entry. 1372 * vap - attributes of new file. 1373 * excl - flag indicating exclusive or non-exclusive mode. 1374 * mode - mode to open file with. 1375 * cr - credentials of caller. 1376 * flag - large file flag [UNUSED]. 1377 * ct - caller context 1378 * vsecp - ACL to be set 1379 * 1380 * OUT: vpp - vnode of created or trunc'd entry. 1381 * 1382 * RETURN: 0 on success, error code on failure. 1383 * 1384 * Timestamps: 1385 * dvp - ctime|mtime updated if new entry created 1386 * vp - ctime|mtime always, atime if new 1387 */ 1388 1389 /* ARGSUSED */ 1390 static int 1391 zfs_create(vnode_t *dvp, char *name, vattr_t *vap, vcexcl_t excl, 1392 int mode, vnode_t **vpp, cred_t *cr, int flag, caller_context_t *ct, 1393 vsecattr_t *vsecp) 1394 { 1395 znode_t *zp, *dzp = VTOZ(dvp); 1396 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1397 zilog_t *zilog; 1398 objset_t *os; 1399 zfs_dirlock_t *dl; 1400 dmu_tx_t *tx; 1401 int error; 1402 ksid_t *ksid; 1403 uid_t uid; 1404 gid_t gid = crgetgid(cr); 1405 zfs_acl_ids_t acl_ids; 1406 boolean_t fuid_dirtied; 1407 boolean_t have_acl = B_FALSE; 1408 boolean_t waited = B_FALSE; 1409 1410 /* 1411 * If we have an ephemeral id, ACL, or XVATTR then 1412 * make sure file system is at proper version 1413 */ 1414 1415 ksid = crgetsid(cr, KSID_OWNER); 1416 if (ksid) 1417 uid = ksid_getid(ksid); 1418 else 1419 uid = crgetuid(cr); 1420 1421 if (zfsvfs->z_use_fuids == B_FALSE && 1422 (vsecp || (vap->va_mask & AT_XVATTR) || 1423 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 1424 return (SET_ERROR(EINVAL)); 1425 1426 ZFS_ENTER(zfsvfs); 1427 ZFS_VERIFY_ZP(dzp); 1428 os = zfsvfs->z_os; 1429 zilog = zfsvfs->z_log; 1430 1431 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 1432 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1433 ZFS_EXIT(zfsvfs); 1434 return (SET_ERROR(EILSEQ)); 1435 } 1436 1437 if (vap->va_mask & AT_XVATTR) { 1438 if ((error = secpolicy_xvattr((xvattr_t *)vap, 1439 crgetuid(cr), cr, vap->va_type)) != 0) { 1440 ZFS_EXIT(zfsvfs); 1441 return (error); 1442 } 1443 } 1444 top: 1445 *vpp = NULL; 1446 1447 if ((vap->va_mode & VSVTX) && secpolicy_vnode_stky_modify(cr)) 1448 vap->va_mode &= ~VSVTX; 1449 1450 if (*name == '\0') { 1451 /* 1452 * Null component name refers to the directory itself. 1453 */ 1454 VN_HOLD(dvp); 1455 zp = dzp; 1456 dl = NULL; 1457 error = 0; 1458 } else { 1459 /* possible VN_HOLD(zp) */ 1460 int zflg = 0; 1461 1462 if (flag & FIGNORECASE) 1463 zflg |= ZCILOOK; 1464 1465 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1466 NULL, NULL); 1467 if (error) { 1468 if (have_acl) 1469 zfs_acl_ids_free(&acl_ids); 1470 if (strcmp(name, "..") == 0) 1471 error = SET_ERROR(EISDIR); 1472 ZFS_EXIT(zfsvfs); 1473 return (error); 1474 } 1475 } 1476 1477 if (zp == NULL) { 1478 uint64_t txtype; 1479 1480 /* 1481 * Create a new file object and update the directory 1482 * to reference it. 1483 */ 1484 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 1485 if (have_acl) 1486 zfs_acl_ids_free(&acl_ids); 1487 goto out; 1488 } 1489 1490 /* 1491 * We only support the creation of regular files in 1492 * extended attribute directories. 1493 */ 1494 1495 if ((dzp->z_pflags & ZFS_XATTR) && 1496 (vap->va_type != VREG)) { 1497 if (have_acl) 1498 zfs_acl_ids_free(&acl_ids); 1499 error = SET_ERROR(EINVAL); 1500 goto out; 1501 } 1502 1503 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap, 1504 cr, vsecp, &acl_ids)) != 0) 1505 goto out; 1506 have_acl = B_TRUE; 1507 1508 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 1509 zfs_acl_ids_free(&acl_ids); 1510 error = SET_ERROR(EDQUOT); 1511 goto out; 1512 } 1513 1514 tx = dmu_tx_create(os); 1515 1516 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 1517 ZFS_SA_BASE_ATTR_SIZE); 1518 1519 fuid_dirtied = zfsvfs->z_fuid_dirty; 1520 if (fuid_dirtied) 1521 zfs_fuid_txhold(zfsvfs, tx); 1522 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 1523 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 1524 if (!zfsvfs->z_use_sa && 1525 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 1526 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 1527 0, acl_ids.z_aclp->z_acl_bytes); 1528 } 1529 error = dmu_tx_assign(tx, 1530 (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT); 1531 if (error) { 1532 zfs_dirent_unlock(dl); 1533 if (error == ERESTART) { 1534 waited = B_TRUE; 1535 dmu_tx_wait(tx); 1536 dmu_tx_abort(tx); 1537 goto top; 1538 } 1539 zfs_acl_ids_free(&acl_ids); 1540 dmu_tx_abort(tx); 1541 ZFS_EXIT(zfsvfs); 1542 return (error); 1543 } 1544 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 1545 1546 if (fuid_dirtied) 1547 zfs_fuid_sync(zfsvfs, tx); 1548 1549 (void) zfs_link_create(dl, zp, tx, ZNEW); 1550 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap); 1551 if (flag & FIGNORECASE) 1552 txtype |= TX_CI; 1553 zfs_log_create(zilog, tx, txtype, dzp, zp, name, 1554 vsecp, acl_ids.z_fuidp, vap); 1555 zfs_acl_ids_free(&acl_ids); 1556 dmu_tx_commit(tx); 1557 } else { 1558 int aflags = (flag & FAPPEND) ? V_APPEND : 0; 1559 1560 if (have_acl) 1561 zfs_acl_ids_free(&acl_ids); 1562 have_acl = B_FALSE; 1563 1564 /* 1565 * A directory entry already exists for this name. 1566 */ 1567 /* 1568 * Can't truncate an existing file if in exclusive mode. 1569 */ 1570 if (excl == EXCL) { 1571 error = SET_ERROR(EEXIST); 1572 goto out; 1573 } 1574 /* 1575 * Can't open a directory for writing. 1576 */ 1577 if ((ZTOV(zp)->v_type == VDIR) && (mode & S_IWRITE)) { 1578 error = SET_ERROR(EISDIR); 1579 goto out; 1580 } 1581 /* 1582 * Verify requested access to file. 1583 */ 1584 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr))) { 1585 goto out; 1586 } 1587 1588 mutex_enter(&dzp->z_lock); 1589 dzp->z_seq++; 1590 mutex_exit(&dzp->z_lock); 1591 1592 /* 1593 * Truncate regular files if requested. 1594 */ 1595 if ((ZTOV(zp)->v_type == VREG) && 1596 (vap->va_mask & AT_SIZE) && (vap->va_size == 0)) { 1597 /* we can't hold any locks when calling zfs_freesp() */ 1598 zfs_dirent_unlock(dl); 1599 dl = NULL; 1600 error = zfs_freesp(zp, 0, 0, mode, TRUE); 1601 if (error == 0) { 1602 vnevent_create(ZTOV(zp), ct); 1603 } 1604 } 1605 } 1606 out: 1607 1608 if (dl) 1609 zfs_dirent_unlock(dl); 1610 1611 if (error) { 1612 if (zp) 1613 VN_RELE(ZTOV(zp)); 1614 } else { 1615 *vpp = ZTOV(zp); 1616 error = specvp_check(vpp, cr); 1617 } 1618 1619 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1620 zil_commit(zilog, 0); 1621 1622 ZFS_EXIT(zfsvfs); 1623 return (error); 1624 } 1625 1626 /* 1627 * Remove an entry from a directory. 1628 * 1629 * IN: dvp - vnode of directory to remove entry from. 1630 * name - name of entry to remove. 1631 * cr - credentials of caller. 1632 * ct - caller context 1633 * flags - case flags 1634 * 1635 * RETURN: 0 on success, error code on failure. 1636 * 1637 * Timestamps: 1638 * dvp - ctime|mtime 1639 * vp - ctime (if nlink > 0) 1640 */ 1641 1642 uint64_t null_xattr = 0; 1643 1644 /*ARGSUSED*/ 1645 static int 1646 zfs_remove(vnode_t *dvp, char *name, cred_t *cr, caller_context_t *ct, 1647 int flags) 1648 { 1649 znode_t *zp, *dzp = VTOZ(dvp); 1650 znode_t *xzp; 1651 vnode_t *vp; 1652 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1653 zilog_t *zilog; 1654 uint64_t acl_obj, xattr_obj; 1655 uint64_t xattr_obj_unlinked = 0; 1656 uint64_t obj = 0; 1657 zfs_dirlock_t *dl; 1658 dmu_tx_t *tx; 1659 boolean_t may_delete_now, delete_now = FALSE; 1660 boolean_t unlinked, toobig = FALSE; 1661 uint64_t txtype; 1662 pathname_t *realnmp = NULL; 1663 pathname_t realnm; 1664 int error; 1665 int zflg = ZEXISTS; 1666 boolean_t waited = B_FALSE; 1667 1668 ZFS_ENTER(zfsvfs); 1669 ZFS_VERIFY_ZP(dzp); 1670 zilog = zfsvfs->z_log; 1671 1672 if (flags & FIGNORECASE) { 1673 zflg |= ZCILOOK; 1674 pn_alloc(&realnm); 1675 realnmp = &realnm; 1676 } 1677 1678 top: 1679 xattr_obj = 0; 1680 xzp = NULL; 1681 /* 1682 * Attempt to lock directory; fail if entry doesn't exist. 1683 */ 1684 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1685 NULL, realnmp)) { 1686 if (realnmp) 1687 pn_free(realnmp); 1688 ZFS_EXIT(zfsvfs); 1689 return (error); 1690 } 1691 1692 vp = ZTOV(zp); 1693 1694 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 1695 goto out; 1696 } 1697 1698 /* 1699 * Need to use rmdir for removing directories. 1700 */ 1701 if (vp->v_type == VDIR) { 1702 error = SET_ERROR(EPERM); 1703 goto out; 1704 } 1705 1706 vnevent_remove(vp, dvp, name, ct); 1707 1708 if (realnmp) 1709 dnlc_remove(dvp, realnmp->pn_buf); 1710 else 1711 dnlc_remove(dvp, name); 1712 1713 mutex_enter(&vp->v_lock); 1714 may_delete_now = vp->v_count == 1 && !vn_has_cached_data(vp); 1715 mutex_exit(&vp->v_lock); 1716 1717 /* 1718 * We may delete the znode now, or we may put it in the unlinked set; 1719 * it depends on whether we're the last link, and on whether there are 1720 * other holds on the vnode. So we dmu_tx_hold() the right things to 1721 * allow for either case. 1722 */ 1723 obj = zp->z_id; 1724 tx = dmu_tx_create(zfsvfs->z_os); 1725 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1726 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1727 zfs_sa_upgrade_txholds(tx, zp); 1728 zfs_sa_upgrade_txholds(tx, dzp); 1729 if (may_delete_now) { 1730 toobig = 1731 zp->z_size > zp->z_blksz * DMU_MAX_DELETEBLKCNT; 1732 /* if the file is too big, only hold_free a token amount */ 1733 dmu_tx_hold_free(tx, zp->z_id, 0, 1734 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END)); 1735 } 1736 1737 /* are there any extended attributes? */ 1738 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 1739 &xattr_obj, sizeof (xattr_obj)); 1740 if (error == 0 && xattr_obj) { 1741 error = zfs_zget(zfsvfs, xattr_obj, &xzp); 1742 ASSERT0(error); 1743 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 1744 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE); 1745 } 1746 1747 mutex_enter(&zp->z_lock); 1748 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now) 1749 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 1750 mutex_exit(&zp->z_lock); 1751 1752 /* charge as an update -- would be nice not to charge at all */ 1753 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1754 1755 /* 1756 * Mark this transaction as typically resulting in a net free of space 1757 */ 1758 dmu_tx_mark_netfree(tx); 1759 1760 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT); 1761 if (error) { 1762 zfs_dirent_unlock(dl); 1763 VN_RELE(vp); 1764 if (xzp) 1765 VN_RELE(ZTOV(xzp)); 1766 if (error == ERESTART) { 1767 waited = B_TRUE; 1768 dmu_tx_wait(tx); 1769 dmu_tx_abort(tx); 1770 goto top; 1771 } 1772 if (realnmp) 1773 pn_free(realnmp); 1774 dmu_tx_abort(tx); 1775 ZFS_EXIT(zfsvfs); 1776 return (error); 1777 } 1778 1779 /* 1780 * Remove the directory entry. 1781 */ 1782 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked); 1783 1784 if (error) { 1785 dmu_tx_commit(tx); 1786 goto out; 1787 } 1788 1789 if (unlinked) { 1790 /* 1791 * Hold z_lock so that we can make sure that the ACL obj 1792 * hasn't changed. Could have been deleted due to 1793 * zfs_sa_upgrade(). 1794 */ 1795 mutex_enter(&zp->z_lock); 1796 mutex_enter(&vp->v_lock); 1797 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 1798 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked)); 1799 delete_now = may_delete_now && !toobig && 1800 vp->v_count == 1 && !vn_has_cached_data(vp) && 1801 xattr_obj == xattr_obj_unlinked && zfs_external_acl(zp) == 1802 acl_obj; 1803 mutex_exit(&vp->v_lock); 1804 } 1805 1806 if (delete_now) { 1807 if (xattr_obj_unlinked) { 1808 ASSERT3U(xzp->z_links, ==, 2); 1809 mutex_enter(&xzp->z_lock); 1810 xzp->z_unlinked = 1; 1811 xzp->z_links = 0; 1812 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs), 1813 &xzp->z_links, sizeof (xzp->z_links), tx); 1814 ASSERT3U(error, ==, 0); 1815 mutex_exit(&xzp->z_lock); 1816 zfs_unlinked_add(xzp, tx); 1817 1818 if (zp->z_is_sa) 1819 error = sa_remove(zp->z_sa_hdl, 1820 SA_ZPL_XATTR(zfsvfs), tx); 1821 else 1822 error = sa_update(zp->z_sa_hdl, 1823 SA_ZPL_XATTR(zfsvfs), &null_xattr, 1824 sizeof (uint64_t), tx); 1825 ASSERT0(error); 1826 } 1827 mutex_enter(&vp->v_lock); 1828 VN_RELE_LOCKED(vp); 1829 ASSERT0(vp->v_count); 1830 mutex_exit(&vp->v_lock); 1831 mutex_exit(&zp->z_lock); 1832 zfs_znode_delete(zp, tx); 1833 } else if (unlinked) { 1834 mutex_exit(&zp->z_lock); 1835 zfs_unlinked_add(zp, tx); 1836 } 1837 1838 txtype = TX_REMOVE; 1839 if (flags & FIGNORECASE) 1840 txtype |= TX_CI; 1841 zfs_log_remove(zilog, tx, txtype, dzp, name, obj); 1842 1843 dmu_tx_commit(tx); 1844 out: 1845 if (realnmp) 1846 pn_free(realnmp); 1847 1848 zfs_dirent_unlock(dl); 1849 1850 if (!delete_now) 1851 VN_RELE(vp); 1852 if (xzp) 1853 VN_RELE(ZTOV(xzp)); 1854 1855 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1856 zil_commit(zilog, 0); 1857 1858 ZFS_EXIT(zfsvfs); 1859 return (error); 1860 } 1861 1862 /* 1863 * Create a new directory and insert it into dvp using the name 1864 * provided. Return a pointer to the inserted directory. 1865 * 1866 * IN: dvp - vnode of directory to add subdir to. 1867 * dirname - name of new directory. 1868 * vap - attributes of new directory. 1869 * cr - credentials of caller. 1870 * ct - caller context 1871 * flags - case flags 1872 * vsecp - ACL to be set 1873 * 1874 * OUT: vpp - vnode of created directory. 1875 * 1876 * RETURN: 0 on success, error code on failure. 1877 * 1878 * Timestamps: 1879 * dvp - ctime|mtime updated 1880 * vp - ctime|mtime|atime updated 1881 */ 1882 /*ARGSUSED*/ 1883 static int 1884 zfs_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t **vpp, cred_t *cr, 1885 caller_context_t *ct, int flags, vsecattr_t *vsecp) 1886 { 1887 znode_t *zp, *dzp = VTOZ(dvp); 1888 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 1889 zilog_t *zilog; 1890 zfs_dirlock_t *dl; 1891 uint64_t txtype; 1892 dmu_tx_t *tx; 1893 int error; 1894 int zf = ZNEW; 1895 ksid_t *ksid; 1896 uid_t uid; 1897 gid_t gid = crgetgid(cr); 1898 zfs_acl_ids_t acl_ids; 1899 boolean_t fuid_dirtied; 1900 boolean_t waited = B_FALSE; 1901 1902 ASSERT(vap->va_type == VDIR); 1903 1904 /* 1905 * If we have an ephemeral id, ACL, or XVATTR then 1906 * make sure file system is at proper version 1907 */ 1908 1909 ksid = crgetsid(cr, KSID_OWNER); 1910 if (ksid) 1911 uid = ksid_getid(ksid); 1912 else 1913 uid = crgetuid(cr); 1914 if (zfsvfs->z_use_fuids == B_FALSE && 1915 (vsecp || (vap->va_mask & AT_XVATTR) || 1916 IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 1917 return (SET_ERROR(EINVAL)); 1918 1919 ZFS_ENTER(zfsvfs); 1920 ZFS_VERIFY_ZP(dzp); 1921 zilog = zfsvfs->z_log; 1922 1923 if (dzp->z_pflags & ZFS_XATTR) { 1924 ZFS_EXIT(zfsvfs); 1925 return (SET_ERROR(EINVAL)); 1926 } 1927 1928 if (zfsvfs->z_utf8 && u8_validate(dirname, 1929 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1930 ZFS_EXIT(zfsvfs); 1931 return (SET_ERROR(EILSEQ)); 1932 } 1933 if (flags & FIGNORECASE) 1934 zf |= ZCILOOK; 1935 1936 if (vap->va_mask & AT_XVATTR) { 1937 if ((error = secpolicy_xvattr((xvattr_t *)vap, 1938 crgetuid(cr), cr, vap->va_type)) != 0) { 1939 ZFS_EXIT(zfsvfs); 1940 return (error); 1941 } 1942 } 1943 1944 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, 1945 vsecp, &acl_ids)) != 0) { 1946 ZFS_EXIT(zfsvfs); 1947 return (error); 1948 } 1949 /* 1950 * First make sure the new directory doesn't exist. 1951 * 1952 * Existence is checked first to make sure we don't return 1953 * EACCES instead of EEXIST which can cause some applications 1954 * to fail. 1955 */ 1956 top: 1957 *vpp = NULL; 1958 1959 if (error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf, 1960 NULL, NULL)) { 1961 zfs_acl_ids_free(&acl_ids); 1962 ZFS_EXIT(zfsvfs); 1963 return (error); 1964 } 1965 1966 if (error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr)) { 1967 zfs_acl_ids_free(&acl_ids); 1968 zfs_dirent_unlock(dl); 1969 ZFS_EXIT(zfsvfs); 1970 return (error); 1971 } 1972 1973 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 1974 zfs_acl_ids_free(&acl_ids); 1975 zfs_dirent_unlock(dl); 1976 ZFS_EXIT(zfsvfs); 1977 return (SET_ERROR(EDQUOT)); 1978 } 1979 1980 /* 1981 * Add a new entry to the directory. 1982 */ 1983 tx = dmu_tx_create(zfsvfs->z_os); 1984 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 1985 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 1986 fuid_dirtied = zfsvfs->z_fuid_dirty; 1987 if (fuid_dirtied) 1988 zfs_fuid_txhold(zfsvfs, tx); 1989 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 1990 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 1991 acl_ids.z_aclp->z_acl_bytes); 1992 } 1993 1994 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 1995 ZFS_SA_BASE_ATTR_SIZE); 1996 1997 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT); 1998 if (error) { 1999 zfs_dirent_unlock(dl); 2000 if (error == ERESTART) { 2001 waited = B_TRUE; 2002 dmu_tx_wait(tx); 2003 dmu_tx_abort(tx); 2004 goto top; 2005 } 2006 zfs_acl_ids_free(&acl_ids); 2007 dmu_tx_abort(tx); 2008 ZFS_EXIT(zfsvfs); 2009 return (error); 2010 } 2011 2012 /* 2013 * Create new node. 2014 */ 2015 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 2016 2017 if (fuid_dirtied) 2018 zfs_fuid_sync(zfsvfs, tx); 2019 2020 /* 2021 * Now put new name in parent dir. 2022 */ 2023 (void) zfs_link_create(dl, zp, tx, ZNEW); 2024 2025 *vpp = ZTOV(zp); 2026 2027 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap); 2028 if (flags & FIGNORECASE) 2029 txtype |= TX_CI; 2030 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp, 2031 acl_ids.z_fuidp, vap); 2032 2033 zfs_acl_ids_free(&acl_ids); 2034 2035 dmu_tx_commit(tx); 2036 2037 zfs_dirent_unlock(dl); 2038 2039 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2040 zil_commit(zilog, 0); 2041 2042 ZFS_EXIT(zfsvfs); 2043 return (0); 2044 } 2045 2046 /* 2047 * Remove a directory subdir entry. If the current working 2048 * directory is the same as the subdir to be removed, the 2049 * remove will fail. 2050 * 2051 * IN: dvp - vnode of directory to remove from. 2052 * name - name of directory to be removed. 2053 * cwd - vnode of current working directory. 2054 * cr - credentials of caller. 2055 * ct - caller context 2056 * flags - case flags 2057 * 2058 * RETURN: 0 on success, error code on failure. 2059 * 2060 * Timestamps: 2061 * dvp - ctime|mtime updated 2062 */ 2063 /*ARGSUSED*/ 2064 static int 2065 zfs_rmdir(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr, 2066 caller_context_t *ct, int flags) 2067 { 2068 znode_t *dzp = VTOZ(dvp); 2069 znode_t *zp; 2070 vnode_t *vp; 2071 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 2072 zilog_t *zilog; 2073 zfs_dirlock_t *dl; 2074 dmu_tx_t *tx; 2075 int error; 2076 int zflg = ZEXISTS; 2077 boolean_t waited = B_FALSE; 2078 2079 ZFS_ENTER(zfsvfs); 2080 ZFS_VERIFY_ZP(dzp); 2081 zilog = zfsvfs->z_log; 2082 2083 if (flags & FIGNORECASE) 2084 zflg |= ZCILOOK; 2085 top: 2086 zp = NULL; 2087 2088 /* 2089 * Attempt to lock directory; fail if entry doesn't exist. 2090 */ 2091 if (error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 2092 NULL, NULL)) { 2093 ZFS_EXIT(zfsvfs); 2094 return (error); 2095 } 2096 2097 vp = ZTOV(zp); 2098 2099 if (error = zfs_zaccess_delete(dzp, zp, cr)) { 2100 goto out; 2101 } 2102 2103 if (vp->v_type != VDIR) { 2104 error = SET_ERROR(ENOTDIR); 2105 goto out; 2106 } 2107 2108 if (vp == cwd) { 2109 error = SET_ERROR(EINVAL); 2110 goto out; 2111 } 2112 2113 vnevent_rmdir(vp, dvp, name, ct); 2114 2115 /* 2116 * Grab a lock on the directory to make sure that noone is 2117 * trying to add (or lookup) entries while we are removing it. 2118 */ 2119 rw_enter(&zp->z_name_lock, RW_WRITER); 2120 2121 /* 2122 * Grab a lock on the parent pointer to make sure we play well 2123 * with the treewalk and directory rename code. 2124 */ 2125 rw_enter(&zp->z_parent_lock, RW_WRITER); 2126 2127 tx = dmu_tx_create(zfsvfs->z_os); 2128 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 2129 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 2130 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 2131 zfs_sa_upgrade_txholds(tx, zp); 2132 zfs_sa_upgrade_txholds(tx, dzp); 2133 dmu_tx_mark_netfree(tx); 2134 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT); 2135 if (error) { 2136 rw_exit(&zp->z_parent_lock); 2137 rw_exit(&zp->z_name_lock); 2138 zfs_dirent_unlock(dl); 2139 VN_RELE(vp); 2140 if (error == ERESTART) { 2141 waited = B_TRUE; 2142 dmu_tx_wait(tx); 2143 dmu_tx_abort(tx); 2144 goto top; 2145 } 2146 dmu_tx_abort(tx); 2147 ZFS_EXIT(zfsvfs); 2148 return (error); 2149 } 2150 2151 error = zfs_link_destroy(dl, zp, tx, zflg, NULL); 2152 2153 if (error == 0) { 2154 uint64_t txtype = TX_RMDIR; 2155 if (flags & FIGNORECASE) 2156 txtype |= TX_CI; 2157 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT); 2158 } 2159 2160 dmu_tx_commit(tx); 2161 2162 rw_exit(&zp->z_parent_lock); 2163 rw_exit(&zp->z_name_lock); 2164 out: 2165 zfs_dirent_unlock(dl); 2166 2167 VN_RELE(vp); 2168 2169 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 2170 zil_commit(zilog, 0); 2171 2172 ZFS_EXIT(zfsvfs); 2173 return (error); 2174 } 2175 2176 /* 2177 * Read as many directory entries as will fit into the provided 2178 * buffer from the given directory cursor position (specified in 2179 * the uio structure). 2180 * 2181 * IN: vp - vnode of directory to read. 2182 * uio - structure supplying read location, range info, 2183 * and return buffer. 2184 * cr - credentials of caller. 2185 * ct - caller context 2186 * flags - case flags 2187 * 2188 * OUT: uio - updated offset and range, buffer filled. 2189 * eofp - set to true if end-of-file detected. 2190 * 2191 * RETURN: 0 on success, error code on failure. 2192 * 2193 * Timestamps: 2194 * vp - atime updated 2195 * 2196 * Note that the low 4 bits of the cookie returned by zap is always zero. 2197 * This allows us to use the low range for "special" directory entries: 2198 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 2199 * we use the offset 2 for the '.zfs' directory. 2200 */ 2201 /* ARGSUSED */ 2202 static int 2203 zfs_readdir(vnode_t *vp, uio_t *uio, cred_t *cr, int *eofp, 2204 caller_context_t *ct, int flags) 2205 { 2206 znode_t *zp = VTOZ(vp); 2207 iovec_t *iovp; 2208 edirent_t *eodp; 2209 dirent64_t *odp; 2210 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2211 objset_t *os; 2212 caddr_t outbuf; 2213 size_t bufsize; 2214 zap_cursor_t zc; 2215 zap_attribute_t zap; 2216 uint_t bytes_wanted; 2217 uint64_t offset; /* must be unsigned; checks for < 1 */ 2218 uint64_t parent; 2219 int local_eof; 2220 int outcount; 2221 int error; 2222 uint8_t prefetch; 2223 boolean_t check_sysattrs; 2224 2225 ZFS_ENTER(zfsvfs); 2226 ZFS_VERIFY_ZP(zp); 2227 2228 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 2229 &parent, sizeof (parent))) != 0) { 2230 ZFS_EXIT(zfsvfs); 2231 return (error); 2232 } 2233 2234 /* 2235 * If we are not given an eof variable, 2236 * use a local one. 2237 */ 2238 if (eofp == NULL) 2239 eofp = &local_eof; 2240 2241 /* 2242 * Check for valid iov_len. 2243 */ 2244 if (uio->uio_iov->iov_len <= 0) { 2245 ZFS_EXIT(zfsvfs); 2246 return (SET_ERROR(EINVAL)); 2247 } 2248 2249 /* 2250 * Quit if directory has been removed (posix) 2251 */ 2252 if ((*eofp = zp->z_unlinked) != 0) { 2253 ZFS_EXIT(zfsvfs); 2254 return (0); 2255 } 2256 2257 error = 0; 2258 os = zfsvfs->z_os; 2259 offset = uio->uio_loffset; 2260 prefetch = zp->z_zn_prefetch; 2261 2262 /* 2263 * Initialize the iterator cursor. 2264 */ 2265 if (offset <= 3) { 2266 /* 2267 * Start iteration from the beginning of the directory. 2268 */ 2269 zap_cursor_init(&zc, os, zp->z_id); 2270 } else { 2271 /* 2272 * The offset is a serialized cursor. 2273 */ 2274 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 2275 } 2276 2277 /* 2278 * Get space to change directory entries into fs independent format. 2279 */ 2280 iovp = uio->uio_iov; 2281 bytes_wanted = iovp->iov_len; 2282 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) { 2283 bufsize = bytes_wanted; 2284 outbuf = kmem_alloc(bufsize, KM_SLEEP); 2285 odp = (struct dirent64 *)outbuf; 2286 } else { 2287 bufsize = bytes_wanted; 2288 outbuf = NULL; 2289 odp = (struct dirent64 *)iovp->iov_base; 2290 } 2291 eodp = (struct edirent *)odp; 2292 2293 /* 2294 * If this VFS supports the system attribute view interface; and 2295 * we're looking at an extended attribute directory; and we care 2296 * about normalization conflicts on this vfs; then we must check 2297 * for normalization conflicts with the sysattr name space. 2298 */ 2299 check_sysattrs = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 2300 (vp->v_flag & V_XATTRDIR) && zfsvfs->z_norm && 2301 (flags & V_RDDIR_ENTFLAGS); 2302 2303 /* 2304 * Transform to file-system independent format 2305 */ 2306 outcount = 0; 2307 while (outcount < bytes_wanted) { 2308 ino64_t objnum; 2309 ushort_t reclen; 2310 off64_t *next = NULL; 2311 2312 /* 2313 * Special case `.', `..', and `.zfs'. 2314 */ 2315 if (offset == 0) { 2316 (void) strcpy(zap.za_name, "."); 2317 zap.za_normalization_conflict = 0; 2318 objnum = zp->z_id; 2319 } else if (offset == 1) { 2320 (void) strcpy(zap.za_name, ".."); 2321 zap.za_normalization_conflict = 0; 2322 objnum = parent; 2323 } else if (offset == 2 && zfs_show_ctldir(zp)) { 2324 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 2325 zap.za_normalization_conflict = 0; 2326 objnum = ZFSCTL_INO_ROOT; 2327 } else { 2328 /* 2329 * Grab next entry. 2330 */ 2331 if (error = zap_cursor_retrieve(&zc, &zap)) { 2332 if ((*eofp = (error == ENOENT)) != 0) 2333 break; 2334 else 2335 goto update; 2336 } 2337 2338 if (zap.za_integer_length != 8 || 2339 zap.za_num_integers != 1) { 2340 cmn_err(CE_WARN, "zap_readdir: bad directory " 2341 "entry, obj = %lld, offset = %lld\n", 2342 (u_longlong_t)zp->z_id, 2343 (u_longlong_t)offset); 2344 error = SET_ERROR(ENXIO); 2345 goto update; 2346 } 2347 2348 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); 2349 /* 2350 * MacOS X can extract the object type here such as: 2351 * uint8_t type = ZFS_DIRENT_TYPE(zap.za_first_integer); 2352 */ 2353 2354 if (check_sysattrs && !zap.za_normalization_conflict) { 2355 zap.za_normalization_conflict = 2356 xattr_sysattr_casechk(zap.za_name); 2357 } 2358 } 2359 2360 if (flags & V_RDDIR_ACCFILTER) { 2361 /* 2362 * If we have no access at all, don't include 2363 * this entry in the returned information 2364 */ 2365 znode_t *ezp; 2366 if (zfs_zget(zp->z_zfsvfs, objnum, &ezp) != 0) 2367 goto skip_entry; 2368 if (!zfs_has_access(ezp, cr)) { 2369 VN_RELE(ZTOV(ezp)); 2370 goto skip_entry; 2371 } 2372 VN_RELE(ZTOV(ezp)); 2373 } 2374 2375 if (flags & V_RDDIR_ENTFLAGS) 2376 reclen = EDIRENT_RECLEN(strlen(zap.za_name)); 2377 else 2378 reclen = DIRENT64_RECLEN(strlen(zap.za_name)); 2379 2380 /* 2381 * Will this entry fit in the buffer? 2382 */ 2383 if (outcount + reclen > bufsize) { 2384 /* 2385 * Did we manage to fit anything in the buffer? 2386 */ 2387 if (!outcount) { 2388 error = SET_ERROR(EINVAL); 2389 goto update; 2390 } 2391 break; 2392 } 2393 if (flags & V_RDDIR_ENTFLAGS) { 2394 /* 2395 * Add extended flag entry: 2396 */ 2397 eodp->ed_ino = objnum; 2398 eodp->ed_reclen = reclen; 2399 /* NOTE: ed_off is the offset for the *next* entry */ 2400 next = &(eodp->ed_off); 2401 eodp->ed_eflags = zap.za_normalization_conflict ? 2402 ED_CASE_CONFLICT : 0; 2403 (void) strncpy(eodp->ed_name, zap.za_name, 2404 EDIRENT_NAMELEN(reclen)); 2405 eodp = (edirent_t *)((intptr_t)eodp + reclen); 2406 } else { 2407 /* 2408 * Add normal entry: 2409 */ 2410 odp->d_ino = objnum; 2411 odp->d_reclen = reclen; 2412 /* NOTE: d_off is the offset for the *next* entry */ 2413 next = &(odp->d_off); 2414 (void) strncpy(odp->d_name, zap.za_name, 2415 DIRENT64_NAMELEN(reclen)); 2416 odp = (dirent64_t *)((intptr_t)odp + reclen); 2417 } 2418 outcount += reclen; 2419 2420 ASSERT(outcount <= bufsize); 2421 2422 /* Prefetch znode */ 2423 if (prefetch) 2424 dmu_prefetch(os, objnum, 0, 0, 0, 2425 ZIO_PRIORITY_SYNC_READ); 2426 2427 skip_entry: 2428 /* 2429 * Move to the next entry, fill in the previous offset. 2430 */ 2431 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 2432 zap_cursor_advance(&zc); 2433 offset = zap_cursor_serialize(&zc); 2434 } else { 2435 offset += 1; 2436 } 2437 if (next) 2438 *next = offset; 2439 } 2440 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 2441 2442 if (uio->uio_segflg == UIO_SYSSPACE && uio->uio_iovcnt == 1) { 2443 iovp->iov_base += outcount; 2444 iovp->iov_len -= outcount; 2445 uio->uio_resid -= outcount; 2446 } else if (error = uiomove(outbuf, (long)outcount, UIO_READ, uio)) { 2447 /* 2448 * Reset the pointer. 2449 */ 2450 offset = uio->uio_loffset; 2451 } 2452 2453 update: 2454 zap_cursor_fini(&zc); 2455 if (uio->uio_segflg != UIO_SYSSPACE || uio->uio_iovcnt != 1) 2456 kmem_free(outbuf, bufsize); 2457 2458 if (error == ENOENT) 2459 error = 0; 2460 2461 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 2462 2463 uio->uio_loffset = offset; 2464 ZFS_EXIT(zfsvfs); 2465 return (error); 2466 } 2467 2468 ulong_t zfs_fsync_sync_cnt = 4; 2469 2470 static int 2471 zfs_fsync(vnode_t *vp, int syncflag, cred_t *cr, caller_context_t *ct) 2472 { 2473 znode_t *zp = VTOZ(vp); 2474 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2475 2476 /* 2477 * Regardless of whether this is required for standards conformance, 2478 * this is the logical behavior when fsync() is called on a file with 2479 * dirty pages. We use B_ASYNC since the ZIL transactions are already 2480 * going to be pushed out as part of the zil_commit(). 2481 */ 2482 if (vn_has_cached_data(vp) && !(syncflag & FNODSYNC) && 2483 (vp->v_type == VREG) && !(IS_SWAPVP(vp))) 2484 (void) VOP_PUTPAGE(vp, (offset_t)0, (size_t)0, B_ASYNC, cr, ct); 2485 2486 (void) tsd_set(zfs_fsyncer_key, (void *)zfs_fsync_sync_cnt); 2487 2488 if (zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) { 2489 ZFS_ENTER(zfsvfs); 2490 ZFS_VERIFY_ZP(zp); 2491 zil_commit(zfsvfs->z_log, zp->z_id); 2492 ZFS_EXIT(zfsvfs); 2493 } 2494 return (0); 2495 } 2496 2497 2498 /* 2499 * Get the requested file attributes and place them in the provided 2500 * vattr structure. 2501 * 2502 * IN: vp - vnode of file. 2503 * vap - va_mask identifies requested attributes. 2504 * If AT_XVATTR set, then optional attrs are requested 2505 * flags - ATTR_NOACLCHECK (CIFS server context) 2506 * cr - credentials of caller. 2507 * ct - caller context 2508 * 2509 * OUT: vap - attribute values. 2510 * 2511 * RETURN: 0 (always succeeds). 2512 */ 2513 /* ARGSUSED */ 2514 static int 2515 zfs_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2516 caller_context_t *ct) 2517 { 2518 znode_t *zp = VTOZ(vp); 2519 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2520 int error = 0; 2521 uint64_t links; 2522 uint64_t mtime[2], ctime[2]; 2523 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2524 xoptattr_t *xoap = NULL; 2525 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2526 sa_bulk_attr_t bulk[2]; 2527 int count = 0; 2528 2529 ZFS_ENTER(zfsvfs); 2530 ZFS_VERIFY_ZP(zp); 2531 2532 zfs_fuid_map_ids(zp, cr, &vap->va_uid, &vap->va_gid); 2533 2534 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 2535 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 2536 2537 if ((error = sa_bulk_lookup(zp->z_sa_hdl, bulk, count)) != 0) { 2538 ZFS_EXIT(zfsvfs); 2539 return (error); 2540 } 2541 2542 /* 2543 * If ACL is trivial don't bother looking for ACE_READ_ATTRIBUTES. 2544 * Also, if we are the owner don't bother, since owner should 2545 * always be allowed to read basic attributes of file. 2546 */ 2547 if (!(zp->z_pflags & ZFS_ACL_TRIVIAL) && 2548 (vap->va_uid != crgetuid(cr))) { 2549 if (error = zfs_zaccess(zp, ACE_READ_ATTRIBUTES, 0, 2550 skipaclchk, cr)) { 2551 ZFS_EXIT(zfsvfs); 2552 return (error); 2553 } 2554 } 2555 2556 /* 2557 * Return all attributes. It's cheaper to provide the answer 2558 * than to determine whether we were asked the question. 2559 */ 2560 2561 mutex_enter(&zp->z_lock); 2562 vap->va_type = vp->v_type; 2563 vap->va_mode = zp->z_mode & MODEMASK; 2564 vap->va_fsid = zp->z_zfsvfs->z_vfs->vfs_dev; 2565 vap->va_nodeid = zp->z_id; 2566 if ((vp->v_flag & VROOT) && zfs_show_ctldir(zp)) 2567 links = zp->z_links + 1; 2568 else 2569 links = zp->z_links; 2570 vap->va_nlink = MIN(links, UINT32_MAX); /* nlink_t limit! */ 2571 vap->va_size = zp->z_size; 2572 vap->va_rdev = vp->v_rdev; 2573 vap->va_seq = zp->z_seq; 2574 2575 /* 2576 * Add in any requested optional attributes and the create time. 2577 * Also set the corresponding bits in the returned attribute bitmap. 2578 */ 2579 if ((xoap = xva_getxoptattr(xvap)) != NULL && zfsvfs->z_use_fuids) { 2580 if (XVA_ISSET_REQ(xvap, XAT_ARCHIVE)) { 2581 xoap->xoa_archive = 2582 ((zp->z_pflags & ZFS_ARCHIVE) != 0); 2583 XVA_SET_RTN(xvap, XAT_ARCHIVE); 2584 } 2585 2586 if (XVA_ISSET_REQ(xvap, XAT_READONLY)) { 2587 xoap->xoa_readonly = 2588 ((zp->z_pflags & ZFS_READONLY) != 0); 2589 XVA_SET_RTN(xvap, XAT_READONLY); 2590 } 2591 2592 if (XVA_ISSET_REQ(xvap, XAT_SYSTEM)) { 2593 xoap->xoa_system = 2594 ((zp->z_pflags & ZFS_SYSTEM) != 0); 2595 XVA_SET_RTN(xvap, XAT_SYSTEM); 2596 } 2597 2598 if (XVA_ISSET_REQ(xvap, XAT_HIDDEN)) { 2599 xoap->xoa_hidden = 2600 ((zp->z_pflags & ZFS_HIDDEN) != 0); 2601 XVA_SET_RTN(xvap, XAT_HIDDEN); 2602 } 2603 2604 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2605 xoap->xoa_nounlink = 2606 ((zp->z_pflags & ZFS_NOUNLINK) != 0); 2607 XVA_SET_RTN(xvap, XAT_NOUNLINK); 2608 } 2609 2610 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2611 xoap->xoa_immutable = 2612 ((zp->z_pflags & ZFS_IMMUTABLE) != 0); 2613 XVA_SET_RTN(xvap, XAT_IMMUTABLE); 2614 } 2615 2616 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2617 xoap->xoa_appendonly = 2618 ((zp->z_pflags & ZFS_APPENDONLY) != 0); 2619 XVA_SET_RTN(xvap, XAT_APPENDONLY); 2620 } 2621 2622 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2623 xoap->xoa_nodump = 2624 ((zp->z_pflags & ZFS_NODUMP) != 0); 2625 XVA_SET_RTN(xvap, XAT_NODUMP); 2626 } 2627 2628 if (XVA_ISSET_REQ(xvap, XAT_OPAQUE)) { 2629 xoap->xoa_opaque = 2630 ((zp->z_pflags & ZFS_OPAQUE) != 0); 2631 XVA_SET_RTN(xvap, XAT_OPAQUE); 2632 } 2633 2634 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2635 xoap->xoa_av_quarantined = 2636 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0); 2637 XVA_SET_RTN(xvap, XAT_AV_QUARANTINED); 2638 } 2639 2640 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2641 xoap->xoa_av_modified = 2642 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0); 2643 XVA_SET_RTN(xvap, XAT_AV_MODIFIED); 2644 } 2645 2646 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) && 2647 vp->v_type == VREG) { 2648 zfs_sa_get_scanstamp(zp, xvap); 2649 } 2650 2651 if (XVA_ISSET_REQ(xvap, XAT_CREATETIME)) { 2652 uint64_t times[2]; 2653 2654 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs), 2655 times, sizeof (times)); 2656 ZFS_TIME_DECODE(&xoap->xoa_createtime, times); 2657 XVA_SET_RTN(xvap, XAT_CREATETIME); 2658 } 2659 2660 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 2661 xoap->xoa_reparse = ((zp->z_pflags & ZFS_REPARSE) != 0); 2662 XVA_SET_RTN(xvap, XAT_REPARSE); 2663 } 2664 if (XVA_ISSET_REQ(xvap, XAT_GEN)) { 2665 xoap->xoa_generation = zp->z_gen; 2666 XVA_SET_RTN(xvap, XAT_GEN); 2667 } 2668 2669 if (XVA_ISSET_REQ(xvap, XAT_OFFLINE)) { 2670 xoap->xoa_offline = 2671 ((zp->z_pflags & ZFS_OFFLINE) != 0); 2672 XVA_SET_RTN(xvap, XAT_OFFLINE); 2673 } 2674 2675 if (XVA_ISSET_REQ(xvap, XAT_SPARSE)) { 2676 xoap->xoa_sparse = 2677 ((zp->z_pflags & ZFS_SPARSE) != 0); 2678 XVA_SET_RTN(xvap, XAT_SPARSE); 2679 } 2680 } 2681 2682 ZFS_TIME_DECODE(&vap->va_atime, zp->z_atime); 2683 ZFS_TIME_DECODE(&vap->va_mtime, mtime); 2684 ZFS_TIME_DECODE(&vap->va_ctime, ctime); 2685 2686 mutex_exit(&zp->z_lock); 2687 2688 sa_object_size(zp->z_sa_hdl, &vap->va_blksize, &vap->va_nblocks); 2689 2690 if (zp->z_blksz == 0) { 2691 /* 2692 * Block size hasn't been set; suggest maximal I/O transfers. 2693 */ 2694 vap->va_blksize = zfsvfs->z_max_blksz; 2695 } 2696 2697 ZFS_EXIT(zfsvfs); 2698 return (0); 2699 } 2700 2701 /* 2702 * Set the file attributes to the values contained in the 2703 * vattr structure. 2704 * 2705 * IN: vp - vnode of file to be modified. 2706 * vap - new attribute values. 2707 * If AT_XVATTR set, then optional attrs are being set 2708 * flags - ATTR_UTIME set if non-default time values provided. 2709 * - ATTR_NOACLCHECK (CIFS context only). 2710 * cr - credentials of caller. 2711 * ct - caller context 2712 * 2713 * RETURN: 0 on success, error code on failure. 2714 * 2715 * Timestamps: 2716 * vp - ctime updated, mtime updated if size changed. 2717 */ 2718 /* ARGSUSED */ 2719 static int 2720 zfs_setattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr, 2721 caller_context_t *ct) 2722 { 2723 znode_t *zp = VTOZ(vp); 2724 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 2725 zilog_t *zilog; 2726 dmu_tx_t *tx; 2727 vattr_t oldva; 2728 xvattr_t tmpxvattr; 2729 uint_t mask = vap->va_mask; 2730 uint_t saved_mask = 0; 2731 int trim_mask = 0; 2732 uint64_t new_mode; 2733 uint64_t new_uid, new_gid; 2734 uint64_t xattr_obj; 2735 uint64_t mtime[2], ctime[2]; 2736 znode_t *attrzp; 2737 int need_policy = FALSE; 2738 int err, err2; 2739 zfs_fuid_info_t *fuidp = NULL; 2740 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 2741 xoptattr_t *xoap; 2742 zfs_acl_t *aclp; 2743 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 2744 boolean_t fuid_dirtied = B_FALSE; 2745 sa_bulk_attr_t bulk[7], xattr_bulk[7]; 2746 int count = 0, xattr_count = 0; 2747 2748 if (mask == 0) 2749 return (0); 2750 2751 if (mask & AT_NOSET) 2752 return (SET_ERROR(EINVAL)); 2753 2754 ZFS_ENTER(zfsvfs); 2755 ZFS_VERIFY_ZP(zp); 2756 2757 zilog = zfsvfs->z_log; 2758 2759 /* 2760 * Make sure that if we have ephemeral uid/gid or xvattr specified 2761 * that file system is at proper version level 2762 */ 2763 2764 if (zfsvfs->z_use_fuids == B_FALSE && 2765 (((mask & AT_UID) && IS_EPHEMERAL(vap->va_uid)) || 2766 ((mask & AT_GID) && IS_EPHEMERAL(vap->va_gid)) || 2767 (mask & AT_XVATTR))) { 2768 ZFS_EXIT(zfsvfs); 2769 return (SET_ERROR(EINVAL)); 2770 } 2771 2772 if (mask & AT_SIZE && vp->v_type == VDIR) { 2773 ZFS_EXIT(zfsvfs); 2774 return (SET_ERROR(EISDIR)); 2775 } 2776 2777 if (mask & AT_SIZE && vp->v_type != VREG && vp->v_type != VFIFO) { 2778 ZFS_EXIT(zfsvfs); 2779 return (SET_ERROR(EINVAL)); 2780 } 2781 2782 /* 2783 * If this is an xvattr_t, then get a pointer to the structure of 2784 * optional attributes. If this is NULL, then we have a vattr_t. 2785 */ 2786 xoap = xva_getxoptattr(xvap); 2787 2788 xva_init(&tmpxvattr); 2789 2790 /* 2791 * Immutable files can only alter immutable bit and atime 2792 */ 2793 if ((zp->z_pflags & ZFS_IMMUTABLE) && 2794 ((mask & (AT_SIZE|AT_UID|AT_GID|AT_MTIME|AT_MODE)) || 2795 ((mask & AT_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) { 2796 ZFS_EXIT(zfsvfs); 2797 return (SET_ERROR(EPERM)); 2798 } 2799 2800 /* 2801 * Note: ZFS_READONLY is handled in zfs_zaccess_common. 2802 */ 2803 2804 /* 2805 * Verify timestamps doesn't overflow 32 bits. 2806 * ZFS can handle large timestamps, but 32bit syscalls can't 2807 * handle times greater than 2039. This check should be removed 2808 * once large timestamps are fully supported. 2809 */ 2810 if (mask & (AT_ATIME | AT_MTIME)) { 2811 if (((mask & AT_ATIME) && TIMESPEC_OVERFLOW(&vap->va_atime)) || 2812 ((mask & AT_MTIME) && TIMESPEC_OVERFLOW(&vap->va_mtime))) { 2813 ZFS_EXIT(zfsvfs); 2814 return (SET_ERROR(EOVERFLOW)); 2815 } 2816 } 2817 2818 top: 2819 attrzp = NULL; 2820 aclp = NULL; 2821 2822 /* Can this be moved to before the top label? */ 2823 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 2824 ZFS_EXIT(zfsvfs); 2825 return (SET_ERROR(EROFS)); 2826 } 2827 2828 /* 2829 * First validate permissions 2830 */ 2831 2832 if (mask & AT_SIZE) { 2833 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr); 2834 if (err) { 2835 ZFS_EXIT(zfsvfs); 2836 return (err); 2837 } 2838 /* 2839 * XXX - Note, we are not providing any open 2840 * mode flags here (like FNDELAY), so we may 2841 * block if there are locks present... this 2842 * should be addressed in openat(). 2843 */ 2844 /* XXX - would it be OK to generate a log record here? */ 2845 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 2846 if (err) { 2847 ZFS_EXIT(zfsvfs); 2848 return (err); 2849 } 2850 2851 if (vap->va_size == 0) 2852 vnevent_truncate(ZTOV(zp), ct); 2853 } 2854 2855 if (mask & (AT_ATIME|AT_MTIME) || 2856 ((mask & AT_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) || 2857 XVA_ISSET_REQ(xvap, XAT_READONLY) || 2858 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) || 2859 XVA_ISSET_REQ(xvap, XAT_OFFLINE) || 2860 XVA_ISSET_REQ(xvap, XAT_SPARSE) || 2861 XVA_ISSET_REQ(xvap, XAT_CREATETIME) || 2862 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) { 2863 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0, 2864 skipaclchk, cr); 2865 } 2866 2867 if (mask & (AT_UID|AT_GID)) { 2868 int idmask = (mask & (AT_UID|AT_GID)); 2869 int take_owner; 2870 int take_group; 2871 2872 /* 2873 * NOTE: even if a new mode is being set, 2874 * we may clear S_ISUID/S_ISGID bits. 2875 */ 2876 2877 if (!(mask & AT_MODE)) 2878 vap->va_mode = zp->z_mode; 2879 2880 /* 2881 * Take ownership or chgrp to group we are a member of 2882 */ 2883 2884 take_owner = (mask & AT_UID) && (vap->va_uid == crgetuid(cr)); 2885 take_group = (mask & AT_GID) && 2886 zfs_groupmember(zfsvfs, vap->va_gid, cr); 2887 2888 /* 2889 * If both AT_UID and AT_GID are set then take_owner and 2890 * take_group must both be set in order to allow taking 2891 * ownership. 2892 * 2893 * Otherwise, send the check through secpolicy_vnode_setattr() 2894 * 2895 */ 2896 2897 if (((idmask == (AT_UID|AT_GID)) && take_owner && take_group) || 2898 ((idmask == AT_UID) && take_owner) || 2899 ((idmask == AT_GID) && take_group)) { 2900 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0, 2901 skipaclchk, cr) == 0) { 2902 /* 2903 * Remove setuid/setgid for non-privileged users 2904 */ 2905 secpolicy_setid_clear(vap, cr); 2906 trim_mask = (mask & (AT_UID|AT_GID)); 2907 } else { 2908 need_policy = TRUE; 2909 } 2910 } else { 2911 need_policy = TRUE; 2912 } 2913 } 2914 2915 mutex_enter(&zp->z_lock); 2916 oldva.va_mode = zp->z_mode; 2917 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid); 2918 if (mask & AT_XVATTR) { 2919 /* 2920 * Update xvattr mask to include only those attributes 2921 * that are actually changing. 2922 * 2923 * the bits will be restored prior to actually setting 2924 * the attributes so the caller thinks they were set. 2925 */ 2926 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2927 if (xoap->xoa_appendonly != 2928 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) { 2929 need_policy = TRUE; 2930 } else { 2931 XVA_CLR_REQ(xvap, XAT_APPENDONLY); 2932 XVA_SET_REQ(&tmpxvattr, XAT_APPENDONLY); 2933 } 2934 } 2935 2936 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2937 if (xoap->xoa_nounlink != 2938 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) { 2939 need_policy = TRUE; 2940 } else { 2941 XVA_CLR_REQ(xvap, XAT_NOUNLINK); 2942 XVA_SET_REQ(&tmpxvattr, XAT_NOUNLINK); 2943 } 2944 } 2945 2946 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2947 if (xoap->xoa_immutable != 2948 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) { 2949 need_policy = TRUE; 2950 } else { 2951 XVA_CLR_REQ(xvap, XAT_IMMUTABLE); 2952 XVA_SET_REQ(&tmpxvattr, XAT_IMMUTABLE); 2953 } 2954 } 2955 2956 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2957 if (xoap->xoa_nodump != 2958 ((zp->z_pflags & ZFS_NODUMP) != 0)) { 2959 need_policy = TRUE; 2960 } else { 2961 XVA_CLR_REQ(xvap, XAT_NODUMP); 2962 XVA_SET_REQ(&tmpxvattr, XAT_NODUMP); 2963 } 2964 } 2965 2966 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2967 if (xoap->xoa_av_modified != 2968 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) { 2969 need_policy = TRUE; 2970 } else { 2971 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED); 2972 XVA_SET_REQ(&tmpxvattr, XAT_AV_MODIFIED); 2973 } 2974 } 2975 2976 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2977 if ((vp->v_type != VREG && 2978 xoap->xoa_av_quarantined) || 2979 xoap->xoa_av_quarantined != 2980 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) { 2981 need_policy = TRUE; 2982 } else { 2983 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED); 2984 XVA_SET_REQ(&tmpxvattr, XAT_AV_QUARANTINED); 2985 } 2986 } 2987 2988 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 2989 mutex_exit(&zp->z_lock); 2990 ZFS_EXIT(zfsvfs); 2991 return (SET_ERROR(EPERM)); 2992 } 2993 2994 if (need_policy == FALSE && 2995 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) || 2996 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) { 2997 need_policy = TRUE; 2998 } 2999 } 3000 3001 mutex_exit(&zp->z_lock); 3002 3003 if (mask & AT_MODE) { 3004 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr) == 0) { 3005 err = secpolicy_setid_setsticky_clear(vp, vap, 3006 &oldva, cr); 3007 if (err) { 3008 ZFS_EXIT(zfsvfs); 3009 return (err); 3010 } 3011 trim_mask |= AT_MODE; 3012 } else { 3013 need_policy = TRUE; 3014 } 3015 } 3016 3017 if (need_policy) { 3018 /* 3019 * If trim_mask is set then take ownership 3020 * has been granted or write_acl is present and user 3021 * has the ability to modify mode. In that case remove 3022 * UID|GID and or MODE from mask so that 3023 * secpolicy_vnode_setattr() doesn't revoke it. 3024 */ 3025 3026 if (trim_mask) { 3027 saved_mask = vap->va_mask; 3028 vap->va_mask &= ~trim_mask; 3029 } 3030 err = secpolicy_vnode_setattr(cr, vp, vap, &oldva, flags, 3031 (int (*)(void *, int, cred_t *))zfs_zaccess_unix, zp); 3032 if (err) { 3033 ZFS_EXIT(zfsvfs); 3034 return (err); 3035 } 3036 3037 if (trim_mask) 3038 vap->va_mask |= saved_mask; 3039 } 3040 3041 /* 3042 * secpolicy_vnode_setattr, or take ownership may have 3043 * changed va_mask 3044 */ 3045 mask = vap->va_mask; 3046 3047 if ((mask & (AT_UID | AT_GID))) { 3048 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 3049 &xattr_obj, sizeof (xattr_obj)); 3050 3051 if (err == 0 && xattr_obj) { 3052 err = zfs_zget(zp->z_zfsvfs, xattr_obj, &attrzp); 3053 if (err) 3054 goto out2; 3055 } 3056 if (mask & AT_UID) { 3057 new_uid = zfs_fuid_create(zfsvfs, 3058 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp); 3059 if (new_uid != zp->z_uid && 3060 zfs_fuid_overquota(zfsvfs, B_FALSE, new_uid)) { 3061 if (attrzp) 3062 VN_RELE(ZTOV(attrzp)); 3063 err = SET_ERROR(EDQUOT); 3064 goto out2; 3065 } 3066 } 3067 3068 if (mask & AT_GID) { 3069 new_gid = zfs_fuid_create(zfsvfs, (uint64_t)vap->va_gid, 3070 cr, ZFS_GROUP, &fuidp); 3071 if (new_gid != zp->z_gid && 3072 zfs_fuid_overquota(zfsvfs, B_TRUE, new_gid)) { 3073 if (attrzp) 3074 VN_RELE(ZTOV(attrzp)); 3075 err = SET_ERROR(EDQUOT); 3076 goto out2; 3077 } 3078 } 3079 } 3080 tx = dmu_tx_create(zfsvfs->z_os); 3081 3082 if (mask & AT_MODE) { 3083 uint64_t pmode = zp->z_mode; 3084 uint64_t acl_obj; 3085 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 3086 3087 if (zp->z_zfsvfs->z_acl_mode == ZFS_ACL_RESTRICTED && 3088 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) { 3089 err = SET_ERROR(EPERM); 3090 goto out; 3091 } 3092 3093 if (err = zfs_acl_chmod_setattr(zp, &aclp, new_mode)) 3094 goto out; 3095 3096 mutex_enter(&zp->z_lock); 3097 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) { 3098 /* 3099 * Are we upgrading ACL from old V0 format 3100 * to V1 format? 3101 */ 3102 if (zfsvfs->z_version >= ZPL_VERSION_FUID && 3103 zfs_znode_acl_version(zp) == 3104 ZFS_ACL_VERSION_INITIAL) { 3105 dmu_tx_hold_free(tx, acl_obj, 0, 3106 DMU_OBJECT_END); 3107 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 3108 0, aclp->z_acl_bytes); 3109 } else { 3110 dmu_tx_hold_write(tx, acl_obj, 0, 3111 aclp->z_acl_bytes); 3112 } 3113 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) { 3114 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 3115 0, aclp->z_acl_bytes); 3116 } 3117 mutex_exit(&zp->z_lock); 3118 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 3119 } else { 3120 if ((mask & AT_XVATTR) && 3121 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 3122 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 3123 else 3124 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 3125 } 3126 3127 if (attrzp) { 3128 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE); 3129 } 3130 3131 fuid_dirtied = zfsvfs->z_fuid_dirty; 3132 if (fuid_dirtied) 3133 zfs_fuid_txhold(zfsvfs, tx); 3134 3135 zfs_sa_upgrade_txholds(tx, zp); 3136 3137 err = dmu_tx_assign(tx, TXG_WAIT); 3138 if (err) 3139 goto out; 3140 3141 count = 0; 3142 /* 3143 * Set each attribute requested. 3144 * We group settings according to the locks they need to acquire. 3145 * 3146 * Note: you cannot set ctime directly, although it will be 3147 * updated as a side-effect of calling this function. 3148 */ 3149 3150 3151 if (mask & (AT_UID|AT_GID|AT_MODE)) 3152 mutex_enter(&zp->z_acl_lock); 3153 mutex_enter(&zp->z_lock); 3154 3155 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 3156 &zp->z_pflags, sizeof (zp->z_pflags)); 3157 3158 if (attrzp) { 3159 if (mask & (AT_UID|AT_GID|AT_MODE)) 3160 mutex_enter(&attrzp->z_acl_lock); 3161 mutex_enter(&attrzp->z_lock); 3162 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3163 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags, 3164 sizeof (attrzp->z_pflags)); 3165 } 3166 3167 if (mask & (AT_UID|AT_GID)) { 3168 3169 if (mask & AT_UID) { 3170 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 3171 &new_uid, sizeof (new_uid)); 3172 zp->z_uid = new_uid; 3173 if (attrzp) { 3174 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3175 SA_ZPL_UID(zfsvfs), NULL, &new_uid, 3176 sizeof (new_uid)); 3177 attrzp->z_uid = new_uid; 3178 } 3179 } 3180 3181 if (mask & AT_GID) { 3182 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), 3183 NULL, &new_gid, sizeof (new_gid)); 3184 zp->z_gid = new_gid; 3185 if (attrzp) { 3186 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3187 SA_ZPL_GID(zfsvfs), NULL, &new_gid, 3188 sizeof (new_gid)); 3189 attrzp->z_gid = new_gid; 3190 } 3191 } 3192 if (!(mask & AT_MODE)) { 3193 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), 3194 NULL, &new_mode, sizeof (new_mode)); 3195 new_mode = zp->z_mode; 3196 } 3197 err = zfs_acl_chown_setattr(zp); 3198 ASSERT(err == 0); 3199 if (attrzp) { 3200 err = zfs_acl_chown_setattr(attrzp); 3201 ASSERT(err == 0); 3202 } 3203 } 3204 3205 if (mask & AT_MODE) { 3206 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 3207 &new_mode, sizeof (new_mode)); 3208 zp->z_mode = new_mode; 3209 ASSERT3U((uintptr_t)aclp, !=, NULL); 3210 err = zfs_aclset_common(zp, aclp, cr, tx); 3211 ASSERT0(err); 3212 if (zp->z_acl_cached) 3213 zfs_acl_free(zp->z_acl_cached); 3214 zp->z_acl_cached = aclp; 3215 aclp = NULL; 3216 } 3217 3218 3219 if (mask & AT_ATIME) { 3220 ZFS_TIME_ENCODE(&vap->va_atime, zp->z_atime); 3221 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 3222 &zp->z_atime, sizeof (zp->z_atime)); 3223 } 3224 3225 if (mask & AT_MTIME) { 3226 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 3227 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 3228 mtime, sizeof (mtime)); 3229 } 3230 3231 /* XXX - shouldn't this be done *before* the ATIME/MTIME checks? */ 3232 if (mask & AT_SIZE && !(mask & AT_MTIME)) { 3233 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), 3234 NULL, mtime, sizeof (mtime)); 3235 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 3236 &ctime, sizeof (ctime)); 3237 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 3238 B_TRUE); 3239 } else if (mask != 0) { 3240 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 3241 &ctime, sizeof (ctime)); 3242 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime, 3243 B_TRUE); 3244 if (attrzp) { 3245 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 3246 SA_ZPL_CTIME(zfsvfs), NULL, 3247 &ctime, sizeof (ctime)); 3248 zfs_tstamp_update_setup(attrzp, STATE_CHANGED, 3249 mtime, ctime, B_TRUE); 3250 } 3251 } 3252 /* 3253 * Do this after setting timestamps to prevent timestamp 3254 * update from toggling bit 3255 */ 3256 3257 if (xoap && (mask & AT_XVATTR)) { 3258 3259 /* 3260 * restore trimmed off masks 3261 * so that return masks can be set for caller. 3262 */ 3263 3264 if (XVA_ISSET_REQ(&tmpxvattr, XAT_APPENDONLY)) { 3265 XVA_SET_REQ(xvap, XAT_APPENDONLY); 3266 } 3267 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NOUNLINK)) { 3268 XVA_SET_REQ(xvap, XAT_NOUNLINK); 3269 } 3270 if (XVA_ISSET_REQ(&tmpxvattr, XAT_IMMUTABLE)) { 3271 XVA_SET_REQ(xvap, XAT_IMMUTABLE); 3272 } 3273 if (XVA_ISSET_REQ(&tmpxvattr, XAT_NODUMP)) { 3274 XVA_SET_REQ(xvap, XAT_NODUMP); 3275 } 3276 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_MODIFIED)) { 3277 XVA_SET_REQ(xvap, XAT_AV_MODIFIED); 3278 } 3279 if (XVA_ISSET_REQ(&tmpxvattr, XAT_AV_QUARANTINED)) { 3280 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED); 3281 } 3282 3283 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 3284 ASSERT(vp->v_type == VREG); 3285 3286 zfs_xvattr_set(zp, xvap, tx); 3287 } 3288 3289 if (fuid_dirtied) 3290 zfs_fuid_sync(zfsvfs, tx); 3291 3292 if (mask != 0) 3293 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp); 3294 3295 mutex_exit(&zp->z_lock); 3296 if (mask & (AT_UID|AT_GID|AT_MODE)) 3297 mutex_exit(&zp->z_acl_lock); 3298 3299 if (attrzp) { 3300 if (mask & (AT_UID|AT_GID|AT_MODE)) 3301 mutex_exit(&attrzp->z_acl_lock); 3302 mutex_exit(&attrzp->z_lock); 3303 } 3304 out: 3305 if (err == 0 && attrzp) { 3306 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk, 3307 xattr_count, tx); 3308 ASSERT(err2 == 0); 3309 } 3310 3311 if (attrzp) 3312 VN_RELE(ZTOV(attrzp)); 3313 3314 if (aclp) 3315 zfs_acl_free(aclp); 3316 3317 if (fuidp) { 3318 zfs_fuid_info_free(fuidp); 3319 fuidp = NULL; 3320 } 3321 3322 if (err) { 3323 dmu_tx_abort(tx); 3324 if (err == ERESTART) 3325 goto top; 3326 } else { 3327 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 3328 dmu_tx_commit(tx); 3329 } 3330 3331 out2: 3332 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3333 zil_commit(zilog, 0); 3334 3335 ZFS_EXIT(zfsvfs); 3336 return (err); 3337 } 3338 3339 typedef struct zfs_zlock { 3340 krwlock_t *zl_rwlock; /* lock we acquired */ 3341 znode_t *zl_znode; /* znode we held */ 3342 struct zfs_zlock *zl_next; /* next in list */ 3343 } zfs_zlock_t; 3344 3345 /* 3346 * Drop locks and release vnodes that were held by zfs_rename_lock(). 3347 */ 3348 static void 3349 zfs_rename_unlock(zfs_zlock_t **zlpp) 3350 { 3351 zfs_zlock_t *zl; 3352 3353 while ((zl = *zlpp) != NULL) { 3354 if (zl->zl_znode != NULL) 3355 VN_RELE(ZTOV(zl->zl_znode)); 3356 rw_exit(zl->zl_rwlock); 3357 *zlpp = zl->zl_next; 3358 kmem_free(zl, sizeof (*zl)); 3359 } 3360 } 3361 3362 /* 3363 * Search back through the directory tree, using the ".." entries. 3364 * Lock each directory in the chain to prevent concurrent renames. 3365 * Fail any attempt to move a directory into one of its own descendants. 3366 * XXX - z_parent_lock can overlap with map or grow locks 3367 */ 3368 static int 3369 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 3370 { 3371 zfs_zlock_t *zl; 3372 znode_t *zp = tdzp; 3373 uint64_t rootid = zp->z_zfsvfs->z_root; 3374 uint64_t oidp = zp->z_id; 3375 krwlock_t *rwlp = &szp->z_parent_lock; 3376 krw_t rw = RW_WRITER; 3377 3378 /* 3379 * First pass write-locks szp and compares to zp->z_id. 3380 * Later passes read-lock zp and compare to zp->z_parent. 3381 */ 3382 do { 3383 if (!rw_tryenter(rwlp, rw)) { 3384 /* 3385 * Another thread is renaming in this path. 3386 * Note that if we are a WRITER, we don't have any 3387 * parent_locks held yet. 3388 */ 3389 if (rw == RW_READER && zp->z_id > szp->z_id) { 3390 /* 3391 * Drop our locks and restart 3392 */ 3393 zfs_rename_unlock(&zl); 3394 *zlpp = NULL; 3395 zp = tdzp; 3396 oidp = zp->z_id; 3397 rwlp = &szp->z_parent_lock; 3398 rw = RW_WRITER; 3399 continue; 3400 } else { 3401 /* 3402 * Wait for other thread to drop its locks 3403 */ 3404 rw_enter(rwlp, rw); 3405 } 3406 } 3407 3408 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 3409 zl->zl_rwlock = rwlp; 3410 zl->zl_znode = NULL; 3411 zl->zl_next = *zlpp; 3412 *zlpp = zl; 3413 3414 if (oidp == szp->z_id) /* We're a descendant of szp */ 3415 return (SET_ERROR(EINVAL)); 3416 3417 if (oidp == rootid) /* We've hit the top */ 3418 return (0); 3419 3420 if (rw == RW_READER) { /* i.e. not the first pass */ 3421 int error = zfs_zget(zp->z_zfsvfs, oidp, &zp); 3422 if (error) 3423 return (error); 3424 zl->zl_znode = zp; 3425 } 3426 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zp->z_zfsvfs), 3427 &oidp, sizeof (oidp)); 3428 rwlp = &zp->z_parent_lock; 3429 rw = RW_READER; 3430 3431 } while (zp->z_id != sdzp->z_id); 3432 3433 return (0); 3434 } 3435 3436 /* 3437 * Move an entry from the provided source directory to the target 3438 * directory. Change the entry name as indicated. 3439 * 3440 * IN: sdvp - Source directory containing the "old entry". 3441 * snm - Old entry name. 3442 * tdvp - Target directory to contain the "new entry". 3443 * tnm - New entry name. 3444 * cr - credentials of caller. 3445 * ct - caller context 3446 * flags - case flags 3447 * 3448 * RETURN: 0 on success, error code on failure. 3449 * 3450 * Timestamps: 3451 * sdvp,tdvp - ctime|mtime updated 3452 */ 3453 /*ARGSUSED*/ 3454 static int 3455 zfs_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm, cred_t *cr, 3456 caller_context_t *ct, int flags) 3457 { 3458 znode_t *tdzp, *szp, *tzp; 3459 znode_t *sdzp = VTOZ(sdvp); 3460 zfsvfs_t *zfsvfs = sdzp->z_zfsvfs; 3461 zilog_t *zilog; 3462 vnode_t *realvp; 3463 zfs_dirlock_t *sdl, *tdl; 3464 dmu_tx_t *tx; 3465 zfs_zlock_t *zl; 3466 int cmp, serr, terr; 3467 int error = 0, rm_err = 0; 3468 int zflg = 0; 3469 boolean_t waited = B_FALSE; 3470 3471 ZFS_ENTER(zfsvfs); 3472 ZFS_VERIFY_ZP(sdzp); 3473 zilog = zfsvfs->z_log; 3474 3475 /* 3476 * Make sure we have the real vp for the target directory. 3477 */ 3478 if (VOP_REALVP(tdvp, &realvp, ct) == 0) 3479 tdvp = realvp; 3480 3481 tdzp = VTOZ(tdvp); 3482 ZFS_VERIFY_ZP(tdzp); 3483 3484 /* 3485 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the 3486 * ctldir appear to have the same v_vfsp. 3487 */ 3488 if (tdzp->z_zfsvfs != zfsvfs || zfsctl_is_node(tdvp)) { 3489 ZFS_EXIT(zfsvfs); 3490 return (SET_ERROR(EXDEV)); 3491 } 3492 3493 if (zfsvfs->z_utf8 && u8_validate(tnm, 3494 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3495 ZFS_EXIT(zfsvfs); 3496 return (SET_ERROR(EILSEQ)); 3497 } 3498 3499 if (flags & FIGNORECASE) 3500 zflg |= ZCILOOK; 3501 3502 top: 3503 szp = NULL; 3504 tzp = NULL; 3505 zl = NULL; 3506 3507 /* 3508 * This is to prevent the creation of links into attribute space 3509 * by renaming a linked file into/outof an attribute directory. 3510 * See the comment in zfs_link() for why this is considered bad. 3511 */ 3512 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) { 3513 ZFS_EXIT(zfsvfs); 3514 return (SET_ERROR(EINVAL)); 3515 } 3516 3517 /* 3518 * Lock source and target directory entries. To prevent deadlock, 3519 * a lock ordering must be defined. We lock the directory with 3520 * the smallest object id first, or if it's a tie, the one with 3521 * the lexically first name. 3522 */ 3523 if (sdzp->z_id < tdzp->z_id) { 3524 cmp = -1; 3525 } else if (sdzp->z_id > tdzp->z_id) { 3526 cmp = 1; 3527 } else { 3528 /* 3529 * First compare the two name arguments without 3530 * considering any case folding. 3531 */ 3532 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER); 3533 3534 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error); 3535 ASSERT(error == 0 || !zfsvfs->z_utf8); 3536 if (cmp == 0) { 3537 /* 3538 * POSIX: "If the old argument and the new argument 3539 * both refer to links to the same existing file, 3540 * the rename() function shall return successfully 3541 * and perform no other action." 3542 */ 3543 ZFS_EXIT(zfsvfs); 3544 return (0); 3545 } 3546 /* 3547 * If the file system is case-folding, then we may 3548 * have some more checking to do. A case-folding file 3549 * system is either supporting mixed case sensitivity 3550 * access or is completely case-insensitive. Note 3551 * that the file system is always case preserving. 3552 * 3553 * In mixed sensitivity mode case sensitive behavior 3554 * is the default. FIGNORECASE must be used to 3555 * explicitly request case insensitive behavior. 3556 * 3557 * If the source and target names provided differ only 3558 * by case (e.g., a request to rename 'tim' to 'Tim'), 3559 * we will treat this as a special case in the 3560 * case-insensitive mode: as long as the source name 3561 * is an exact match, we will allow this to proceed as 3562 * a name-change request. 3563 */ 3564 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 3565 (zfsvfs->z_case == ZFS_CASE_MIXED && 3566 flags & FIGNORECASE)) && 3567 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST, 3568 &error) == 0) { 3569 /* 3570 * case preserving rename request, require exact 3571 * name matches 3572 */ 3573 zflg |= ZCIEXACT; 3574 zflg &= ~ZCILOOK; 3575 } 3576 } 3577 3578 /* 3579 * If the source and destination directories are the same, we should 3580 * grab the z_name_lock of that directory only once. 3581 */ 3582 if (sdzp == tdzp) { 3583 zflg |= ZHAVELOCK; 3584 rw_enter(&sdzp->z_name_lock, RW_READER); 3585 } 3586 3587 if (cmp < 0) { 3588 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, 3589 ZEXISTS | zflg, NULL, NULL); 3590 terr = zfs_dirent_lock(&tdl, 3591 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL); 3592 } else { 3593 terr = zfs_dirent_lock(&tdl, 3594 tdzp, tnm, &tzp, zflg, NULL, NULL); 3595 serr = zfs_dirent_lock(&sdl, 3596 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg, 3597 NULL, NULL); 3598 } 3599 3600 if (serr) { 3601 /* 3602 * Source entry invalid or not there. 3603 */ 3604 if (!terr) { 3605 zfs_dirent_unlock(tdl); 3606 if (tzp) 3607 VN_RELE(ZTOV(tzp)); 3608 } 3609 3610 if (sdzp == tdzp) 3611 rw_exit(&sdzp->z_name_lock); 3612 3613 if (strcmp(snm, "..") == 0) 3614 serr = SET_ERROR(EINVAL); 3615 ZFS_EXIT(zfsvfs); 3616 return (serr); 3617 } 3618 if (terr) { 3619 zfs_dirent_unlock(sdl); 3620 VN_RELE(ZTOV(szp)); 3621 3622 if (sdzp == tdzp) 3623 rw_exit(&sdzp->z_name_lock); 3624 3625 if (strcmp(tnm, "..") == 0) 3626 terr = SET_ERROR(EINVAL); 3627 ZFS_EXIT(zfsvfs); 3628 return (terr); 3629 } 3630 3631 /* 3632 * Must have write access at the source to remove the old entry 3633 * and write access at the target to create the new entry. 3634 * Note that if target and source are the same, this can be 3635 * done in a single check. 3636 */ 3637 3638 if (error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr)) 3639 goto out; 3640 3641 if (ZTOV(szp)->v_type == VDIR) { 3642 /* 3643 * Check to make sure rename is valid. 3644 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 3645 */ 3646 if (error = zfs_rename_lock(szp, tdzp, sdzp, &zl)) 3647 goto out; 3648 } 3649 3650 /* 3651 * Does target exist? 3652 */ 3653 if (tzp) { 3654 /* 3655 * Source and target must be the same type. 3656 */ 3657 if (ZTOV(szp)->v_type == VDIR) { 3658 if (ZTOV(tzp)->v_type != VDIR) { 3659 error = SET_ERROR(ENOTDIR); 3660 goto out; 3661 } 3662 } else { 3663 if (ZTOV(tzp)->v_type == VDIR) { 3664 error = SET_ERROR(EISDIR); 3665 goto out; 3666 } 3667 } 3668 /* 3669 * POSIX dictates that when the source and target 3670 * entries refer to the same file object, rename 3671 * must do nothing and exit without error. 3672 */ 3673 if (szp->z_id == tzp->z_id) { 3674 error = 0; 3675 goto out; 3676 } 3677 } 3678 3679 vnevent_pre_rename_src(ZTOV(szp), sdvp, snm, ct); 3680 if (tzp) 3681 vnevent_pre_rename_dest(ZTOV(tzp), tdvp, tnm, ct); 3682 3683 /* 3684 * notify the target directory if it is not the same 3685 * as source directory. 3686 */ 3687 if (tdvp != sdvp) { 3688 vnevent_pre_rename_dest_dir(tdvp, ZTOV(szp), tnm, ct); 3689 } 3690 3691 tx = dmu_tx_create(zfsvfs->z_os); 3692 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 3693 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE); 3694 dmu_tx_hold_zap(tx, sdzp->z_id, FALSE, snm); 3695 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 3696 if (sdzp != tdzp) { 3697 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE); 3698 zfs_sa_upgrade_txholds(tx, tdzp); 3699 } 3700 if (tzp) { 3701 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE); 3702 zfs_sa_upgrade_txholds(tx, tzp); 3703 } 3704 3705 zfs_sa_upgrade_txholds(tx, szp); 3706 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 3707 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT); 3708 if (error) { 3709 if (zl != NULL) 3710 zfs_rename_unlock(&zl); 3711 zfs_dirent_unlock(sdl); 3712 zfs_dirent_unlock(tdl); 3713 3714 if (sdzp == tdzp) 3715 rw_exit(&sdzp->z_name_lock); 3716 3717 VN_RELE(ZTOV(szp)); 3718 if (tzp) 3719 VN_RELE(ZTOV(tzp)); 3720 if (error == ERESTART) { 3721 waited = B_TRUE; 3722 dmu_tx_wait(tx); 3723 dmu_tx_abort(tx); 3724 goto top; 3725 } 3726 dmu_tx_abort(tx); 3727 ZFS_EXIT(zfsvfs); 3728 return (error); 3729 } 3730 3731 if (tzp) /* Attempt to remove the existing target */ 3732 error = rm_err = zfs_link_destroy(tdl, tzp, tx, zflg, NULL); 3733 3734 if (error == 0) { 3735 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 3736 if (error == 0) { 3737 szp->z_pflags |= ZFS_AV_MODIFIED; 3738 3739 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs), 3740 (void *)&szp->z_pflags, sizeof (uint64_t), tx); 3741 ASSERT0(error); 3742 3743 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 3744 if (error == 0) { 3745 zfs_log_rename(zilog, tx, TX_RENAME | 3746 (flags & FIGNORECASE ? TX_CI : 0), sdzp, 3747 sdl->dl_name, tdzp, tdl->dl_name, szp); 3748 3749 /* 3750 * Update path information for the target vnode 3751 */ 3752 vn_renamepath(tdvp, ZTOV(szp), tnm, 3753 strlen(tnm)); 3754 } else { 3755 /* 3756 * At this point, we have successfully created 3757 * the target name, but have failed to remove 3758 * the source name. Since the create was done 3759 * with the ZRENAMING flag, there are 3760 * complications; for one, the link count is 3761 * wrong. The easiest way to deal with this 3762 * is to remove the newly created target, and 3763 * return the original error. This must 3764 * succeed; fortunately, it is very unlikely to 3765 * fail, since we just created it. 3766 */ 3767 VERIFY3U(zfs_link_destroy(tdl, szp, tx, 3768 ZRENAMING, NULL), ==, 0); 3769 } 3770 } 3771 } 3772 3773 dmu_tx_commit(tx); 3774 3775 if (tzp && rm_err == 0) 3776 vnevent_rename_dest(ZTOV(tzp), tdvp, tnm, ct); 3777 3778 if (error == 0) { 3779 vnevent_rename_src(ZTOV(szp), sdvp, snm, ct); 3780 /* notify the target dir if it is not the same as source dir */ 3781 if (tdvp != sdvp) 3782 vnevent_rename_dest_dir(tdvp, ct); 3783 } 3784 out: 3785 if (zl != NULL) 3786 zfs_rename_unlock(&zl); 3787 3788 zfs_dirent_unlock(sdl); 3789 zfs_dirent_unlock(tdl); 3790 3791 if (sdzp == tdzp) 3792 rw_exit(&sdzp->z_name_lock); 3793 3794 3795 VN_RELE(ZTOV(szp)); 3796 if (tzp) 3797 VN_RELE(ZTOV(tzp)); 3798 3799 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3800 zil_commit(zilog, 0); 3801 3802 ZFS_EXIT(zfsvfs); 3803 return (error); 3804 } 3805 3806 /* 3807 * Insert the indicated symbolic reference entry into the directory. 3808 * 3809 * IN: dvp - Directory to contain new symbolic link. 3810 * link - Name for new symlink entry. 3811 * vap - Attributes of new entry. 3812 * cr - credentials of caller. 3813 * ct - caller context 3814 * flags - case flags 3815 * 3816 * RETURN: 0 on success, error code on failure. 3817 * 3818 * Timestamps: 3819 * dvp - ctime|mtime updated 3820 */ 3821 /*ARGSUSED*/ 3822 static int 3823 zfs_symlink(vnode_t *dvp, char *name, vattr_t *vap, char *link, cred_t *cr, 3824 caller_context_t *ct, int flags) 3825 { 3826 znode_t *zp, *dzp = VTOZ(dvp); 3827 zfs_dirlock_t *dl; 3828 dmu_tx_t *tx; 3829 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 3830 zilog_t *zilog; 3831 uint64_t len = strlen(link); 3832 int error; 3833 int zflg = ZNEW; 3834 zfs_acl_ids_t acl_ids; 3835 boolean_t fuid_dirtied; 3836 uint64_t txtype = TX_SYMLINK; 3837 boolean_t waited = B_FALSE; 3838 3839 ASSERT(vap->va_type == VLNK); 3840 3841 ZFS_ENTER(zfsvfs); 3842 ZFS_VERIFY_ZP(dzp); 3843 zilog = zfsvfs->z_log; 3844 3845 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 3846 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3847 ZFS_EXIT(zfsvfs); 3848 return (SET_ERROR(EILSEQ)); 3849 } 3850 if (flags & FIGNORECASE) 3851 zflg |= ZCILOOK; 3852 3853 if (len > MAXPATHLEN) { 3854 ZFS_EXIT(zfsvfs); 3855 return (SET_ERROR(ENAMETOOLONG)); 3856 } 3857 3858 if ((error = zfs_acl_ids_create(dzp, 0, 3859 vap, cr, NULL, &acl_ids)) != 0) { 3860 ZFS_EXIT(zfsvfs); 3861 return (error); 3862 } 3863 top: 3864 /* 3865 * Attempt to lock directory; fail if entry already exists. 3866 */ 3867 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL); 3868 if (error) { 3869 zfs_acl_ids_free(&acl_ids); 3870 ZFS_EXIT(zfsvfs); 3871 return (error); 3872 } 3873 3874 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 3875 zfs_acl_ids_free(&acl_ids); 3876 zfs_dirent_unlock(dl); 3877 ZFS_EXIT(zfsvfs); 3878 return (error); 3879 } 3880 3881 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) { 3882 zfs_acl_ids_free(&acl_ids); 3883 zfs_dirent_unlock(dl); 3884 ZFS_EXIT(zfsvfs); 3885 return (SET_ERROR(EDQUOT)); 3886 } 3887 tx = dmu_tx_create(zfsvfs->z_os); 3888 fuid_dirtied = zfsvfs->z_fuid_dirty; 3889 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 3890 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 3891 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 3892 ZFS_SA_BASE_ATTR_SIZE + len); 3893 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 3894 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 3895 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 3896 acl_ids.z_aclp->z_acl_bytes); 3897 } 3898 if (fuid_dirtied) 3899 zfs_fuid_txhold(zfsvfs, tx); 3900 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT); 3901 if (error) { 3902 zfs_dirent_unlock(dl); 3903 if (error == ERESTART) { 3904 waited = B_TRUE; 3905 dmu_tx_wait(tx); 3906 dmu_tx_abort(tx); 3907 goto top; 3908 } 3909 zfs_acl_ids_free(&acl_ids); 3910 dmu_tx_abort(tx); 3911 ZFS_EXIT(zfsvfs); 3912 return (error); 3913 } 3914 3915 /* 3916 * Create a new object for the symlink. 3917 * for version 4 ZPL datsets the symlink will be an SA attribute 3918 */ 3919 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 3920 3921 if (fuid_dirtied) 3922 zfs_fuid_sync(zfsvfs, tx); 3923 3924 mutex_enter(&zp->z_lock); 3925 if (zp->z_is_sa) 3926 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs), 3927 link, len, tx); 3928 else 3929 zfs_sa_symlink(zp, link, len, tx); 3930 mutex_exit(&zp->z_lock); 3931 3932 zp->z_size = len; 3933 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 3934 &zp->z_size, sizeof (zp->z_size), tx); 3935 /* 3936 * Insert the new object into the directory. 3937 */ 3938 (void) zfs_link_create(dl, zp, tx, ZNEW); 3939 3940 if (flags & FIGNORECASE) 3941 txtype |= TX_CI; 3942 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link); 3943 3944 zfs_acl_ids_free(&acl_ids); 3945 3946 dmu_tx_commit(tx); 3947 3948 zfs_dirent_unlock(dl); 3949 3950 VN_RELE(ZTOV(zp)); 3951 3952 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3953 zil_commit(zilog, 0); 3954 3955 ZFS_EXIT(zfsvfs); 3956 return (error); 3957 } 3958 3959 /* 3960 * Return, in the buffer contained in the provided uio structure, 3961 * the symbolic path referred to by vp. 3962 * 3963 * IN: vp - vnode of symbolic link. 3964 * uio - structure to contain the link path. 3965 * cr - credentials of caller. 3966 * ct - caller context 3967 * 3968 * OUT: uio - structure containing the link path. 3969 * 3970 * RETURN: 0 on success, error code on failure. 3971 * 3972 * Timestamps: 3973 * vp - atime updated 3974 */ 3975 /* ARGSUSED */ 3976 static int 3977 zfs_readlink(vnode_t *vp, uio_t *uio, cred_t *cr, caller_context_t *ct) 3978 { 3979 znode_t *zp = VTOZ(vp); 3980 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 3981 int error; 3982 3983 ZFS_ENTER(zfsvfs); 3984 ZFS_VERIFY_ZP(zp); 3985 3986 mutex_enter(&zp->z_lock); 3987 if (zp->z_is_sa) 3988 error = sa_lookup_uio(zp->z_sa_hdl, 3989 SA_ZPL_SYMLINK(zfsvfs), uio); 3990 else 3991 error = zfs_sa_readlink(zp, uio); 3992 mutex_exit(&zp->z_lock); 3993 3994 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 3995 3996 ZFS_EXIT(zfsvfs); 3997 return (error); 3998 } 3999 4000 /* 4001 * Insert a new entry into directory tdvp referencing svp. 4002 * 4003 * IN: tdvp - Directory to contain new entry. 4004 * svp - vnode of new entry. 4005 * name - name of new entry. 4006 * cr - credentials of caller. 4007 * ct - caller context 4008 * 4009 * RETURN: 0 on success, error code on failure. 4010 * 4011 * Timestamps: 4012 * tdvp - ctime|mtime updated 4013 * svp - ctime updated 4014 */ 4015 /* ARGSUSED */ 4016 static int 4017 zfs_link(vnode_t *tdvp, vnode_t *svp, char *name, cred_t *cr, 4018 caller_context_t *ct, int flags) 4019 { 4020 znode_t *dzp = VTOZ(tdvp); 4021 znode_t *tzp, *szp; 4022 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 4023 zilog_t *zilog; 4024 zfs_dirlock_t *dl; 4025 dmu_tx_t *tx; 4026 vnode_t *realvp; 4027 int error; 4028 int zf = ZNEW; 4029 uint64_t parent; 4030 uid_t owner; 4031 boolean_t waited = B_FALSE; 4032 4033 ASSERT(tdvp->v_type == VDIR); 4034 4035 ZFS_ENTER(zfsvfs); 4036 ZFS_VERIFY_ZP(dzp); 4037 zilog = zfsvfs->z_log; 4038 4039 if (VOP_REALVP(svp, &realvp, ct) == 0) 4040 svp = realvp; 4041 4042 /* 4043 * POSIX dictates that we return EPERM here. 4044 * Better choices include ENOTSUP or EISDIR. 4045 */ 4046 if (svp->v_type == VDIR) { 4047 ZFS_EXIT(zfsvfs); 4048 return (SET_ERROR(EPERM)); 4049 } 4050 4051 szp = VTOZ(svp); 4052 ZFS_VERIFY_ZP(szp); 4053 4054 /* 4055 * We check z_zfsvfs rather than v_vfsp here, because snapshots and the 4056 * ctldir appear to have the same v_vfsp. 4057 */ 4058 if (szp->z_zfsvfs != zfsvfs || zfsctl_is_node(svp)) { 4059 ZFS_EXIT(zfsvfs); 4060 return (SET_ERROR(EXDEV)); 4061 } 4062 4063 /* Prevent links to .zfs/shares files */ 4064 4065 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 4066 &parent, sizeof (uint64_t))) != 0) { 4067 ZFS_EXIT(zfsvfs); 4068 return (error); 4069 } 4070 if (parent == zfsvfs->z_shares_dir) { 4071 ZFS_EXIT(zfsvfs); 4072 return (SET_ERROR(EPERM)); 4073 } 4074 4075 if (zfsvfs->z_utf8 && u8_validate(name, 4076 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 4077 ZFS_EXIT(zfsvfs); 4078 return (SET_ERROR(EILSEQ)); 4079 } 4080 if (flags & FIGNORECASE) 4081 zf |= ZCILOOK; 4082 4083 /* 4084 * We do not support links between attributes and non-attributes 4085 * because of the potential security risk of creating links 4086 * into "normal" file space in order to circumvent restrictions 4087 * imposed in attribute space. 4088 */ 4089 if ((szp->z_pflags & ZFS_XATTR) != (dzp->z_pflags & ZFS_XATTR)) { 4090 ZFS_EXIT(zfsvfs); 4091 return (SET_ERROR(EINVAL)); 4092 } 4093 4094 4095 owner = zfs_fuid_map_id(zfsvfs, szp->z_uid, cr, ZFS_OWNER); 4096 if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) { 4097 ZFS_EXIT(zfsvfs); 4098 return (SET_ERROR(EPERM)); 4099 } 4100 4101 if (error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr)) { 4102 ZFS_EXIT(zfsvfs); 4103 return (error); 4104 } 4105 4106 top: 4107 /* 4108 * Attempt to lock directory; fail if entry already exists. 4109 */ 4110 error = zfs_dirent_lock(&dl, dzp, name, &tzp, zf, NULL, NULL); 4111 if (error) { 4112 ZFS_EXIT(zfsvfs); 4113 return (error); 4114 } 4115 4116 tx = dmu_tx_create(zfsvfs->z_os); 4117 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 4118 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 4119 zfs_sa_upgrade_txholds(tx, szp); 4120 zfs_sa_upgrade_txholds(tx, dzp); 4121 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT); 4122 if (error) { 4123 zfs_dirent_unlock(dl); 4124 if (error == ERESTART) { 4125 waited = B_TRUE; 4126 dmu_tx_wait(tx); 4127 dmu_tx_abort(tx); 4128 goto top; 4129 } 4130 dmu_tx_abort(tx); 4131 ZFS_EXIT(zfsvfs); 4132 return (error); 4133 } 4134 4135 error = zfs_link_create(dl, szp, tx, 0); 4136 4137 if (error == 0) { 4138 uint64_t txtype = TX_LINK; 4139 if (flags & FIGNORECASE) 4140 txtype |= TX_CI; 4141 zfs_log_link(zilog, tx, txtype, dzp, szp, name); 4142 } 4143 4144 dmu_tx_commit(tx); 4145 4146 zfs_dirent_unlock(dl); 4147 4148 if (error == 0) { 4149 vnevent_link(svp, ct); 4150 } 4151 4152 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4153 zil_commit(zilog, 0); 4154 4155 ZFS_EXIT(zfsvfs); 4156 return (error); 4157 } 4158 4159 /* 4160 * zfs_null_putapage() is used when the file system has been force 4161 * unmounted. It just drops the pages. 4162 */ 4163 /* ARGSUSED */ 4164 static int 4165 zfs_null_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 4166 size_t *lenp, int flags, cred_t *cr) 4167 { 4168 pvn_write_done(pp, B_INVAL|B_FORCE|B_ERROR); 4169 return (0); 4170 } 4171 4172 /* 4173 * Push a page out to disk, klustering if possible. 4174 * 4175 * IN: vp - file to push page to. 4176 * pp - page to push. 4177 * flags - additional flags. 4178 * cr - credentials of caller. 4179 * 4180 * OUT: offp - start of range pushed. 4181 * lenp - len of range pushed. 4182 * 4183 * RETURN: 0 on success, error code on failure. 4184 * 4185 * NOTE: callers must have locked the page to be pushed. On 4186 * exit, the page (and all other pages in the kluster) must be 4187 * unlocked. 4188 */ 4189 /* ARGSUSED */ 4190 static int 4191 zfs_putapage(vnode_t *vp, page_t *pp, u_offset_t *offp, 4192 size_t *lenp, int flags, cred_t *cr) 4193 { 4194 znode_t *zp = VTOZ(vp); 4195 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4196 dmu_tx_t *tx; 4197 u_offset_t off, koff; 4198 size_t len, klen; 4199 int err; 4200 4201 off = pp->p_offset; 4202 len = PAGESIZE; 4203 /* 4204 * If our blocksize is bigger than the page size, try to kluster 4205 * multiple pages so that we write a full block (thus avoiding 4206 * a read-modify-write). 4207 */ 4208 if (off < zp->z_size && zp->z_blksz > PAGESIZE) { 4209 klen = P2ROUNDUP((ulong_t)zp->z_blksz, PAGESIZE); 4210 koff = ISP2(klen) ? P2ALIGN(off, (u_offset_t)klen) : 0; 4211 ASSERT(koff <= zp->z_size); 4212 if (koff + klen > zp->z_size) 4213 klen = P2ROUNDUP(zp->z_size - koff, (uint64_t)PAGESIZE); 4214 pp = pvn_write_kluster(vp, pp, &off, &len, koff, klen, flags); 4215 } 4216 ASSERT3U(btop(len), ==, btopr(len)); 4217 4218 /* 4219 * Can't push pages past end-of-file. 4220 */ 4221 if (off >= zp->z_size) { 4222 /* ignore all pages */ 4223 err = 0; 4224 goto out; 4225 } else if (off + len > zp->z_size) { 4226 int npages = btopr(zp->z_size - off); 4227 page_t *trunc; 4228 4229 page_list_break(&pp, &trunc, npages); 4230 /* ignore pages past end of file */ 4231 if (trunc) 4232 pvn_write_done(trunc, flags); 4233 len = zp->z_size - off; 4234 } 4235 4236 if (zfs_owner_overquota(zfsvfs, zp, B_FALSE) || 4237 zfs_owner_overquota(zfsvfs, zp, B_TRUE)) { 4238 err = SET_ERROR(EDQUOT); 4239 goto out; 4240 } 4241 tx = dmu_tx_create(zfsvfs->z_os); 4242 dmu_tx_hold_write(tx, zp->z_id, off, len); 4243 4244 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4245 zfs_sa_upgrade_txholds(tx, zp); 4246 err = dmu_tx_assign(tx, TXG_WAIT); 4247 if (err != 0) { 4248 dmu_tx_abort(tx); 4249 goto out; 4250 } 4251 4252 if (zp->z_blksz <= PAGESIZE) { 4253 caddr_t va = zfs_map_page(pp, S_READ); 4254 ASSERT3U(len, <=, PAGESIZE); 4255 dmu_write(zfsvfs->z_os, zp->z_id, off, len, va, tx); 4256 zfs_unmap_page(pp, va); 4257 } else { 4258 err = dmu_write_pages(zfsvfs->z_os, zp->z_id, off, len, pp, tx); 4259 } 4260 4261 if (err == 0) { 4262 uint64_t mtime[2], ctime[2]; 4263 sa_bulk_attr_t bulk[3]; 4264 int count = 0; 4265 4266 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 4267 &mtime, 16); 4268 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 4269 &ctime, 16); 4270 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 4271 &zp->z_pflags, 8); 4272 zfs_tstamp_update_setup(zp, CONTENT_MODIFIED, mtime, ctime, 4273 B_TRUE); 4274 err = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 4275 ASSERT0(err); 4276 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, off, len, 0); 4277 } 4278 dmu_tx_commit(tx); 4279 4280 out: 4281 pvn_write_done(pp, (err ? B_ERROR : 0) | flags); 4282 if (offp) 4283 *offp = off; 4284 if (lenp) 4285 *lenp = len; 4286 4287 return (err); 4288 } 4289 4290 /* 4291 * Copy the portion of the file indicated from pages into the file. 4292 * The pages are stored in a page list attached to the files vnode. 4293 * 4294 * IN: vp - vnode of file to push page data to. 4295 * off - position in file to put data. 4296 * len - amount of data to write. 4297 * flags - flags to control the operation. 4298 * cr - credentials of caller. 4299 * ct - caller context. 4300 * 4301 * RETURN: 0 on success, error code on failure. 4302 * 4303 * Timestamps: 4304 * vp - ctime|mtime updated 4305 */ 4306 /*ARGSUSED*/ 4307 static int 4308 zfs_putpage(vnode_t *vp, offset_t off, size_t len, int flags, cred_t *cr, 4309 caller_context_t *ct) 4310 { 4311 znode_t *zp = VTOZ(vp); 4312 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4313 page_t *pp; 4314 size_t io_len; 4315 u_offset_t io_off; 4316 uint_t blksz; 4317 rl_t *rl; 4318 int error = 0; 4319 4320 ZFS_ENTER(zfsvfs); 4321 ZFS_VERIFY_ZP(zp); 4322 4323 /* 4324 * There's nothing to do if no data is cached. 4325 */ 4326 if (!vn_has_cached_data(vp)) { 4327 ZFS_EXIT(zfsvfs); 4328 return (0); 4329 } 4330 4331 /* 4332 * Align this request to the file block size in case we kluster. 4333 * XXX - this can result in pretty aggresive locking, which can 4334 * impact simultanious read/write access. One option might be 4335 * to break up long requests (len == 0) into block-by-block 4336 * operations to get narrower locking. 4337 */ 4338 blksz = zp->z_blksz; 4339 if (ISP2(blksz)) 4340 io_off = P2ALIGN_TYPED(off, blksz, u_offset_t); 4341 else 4342 io_off = 0; 4343 if (len > 0 && ISP2(blksz)) 4344 io_len = P2ROUNDUP_TYPED(len + (off - io_off), blksz, size_t); 4345 else 4346 io_len = 0; 4347 4348 if (io_len == 0) { 4349 /* 4350 * Search the entire vp list for pages >= io_off. 4351 */ 4352 rl = zfs_range_lock(zp, io_off, UINT64_MAX, RL_WRITER); 4353 error = pvn_vplist_dirty(vp, io_off, zfs_putapage, flags, cr); 4354 goto out; 4355 } 4356 rl = zfs_range_lock(zp, io_off, io_len, RL_WRITER); 4357 4358 if (off > zp->z_size) { 4359 /* past end of file */ 4360 zfs_range_unlock(rl); 4361 ZFS_EXIT(zfsvfs); 4362 return (0); 4363 } 4364 4365 len = MIN(io_len, P2ROUNDUP(zp->z_size, PAGESIZE) - io_off); 4366 4367 for (off = io_off; io_off < off + len; io_off += io_len) { 4368 if ((flags & B_INVAL) || ((flags & B_ASYNC) == 0)) { 4369 pp = page_lookup(vp, io_off, 4370 (flags & (B_INVAL | B_FREE)) ? SE_EXCL : SE_SHARED); 4371 } else { 4372 pp = page_lookup_nowait(vp, io_off, 4373 (flags & B_FREE) ? SE_EXCL : SE_SHARED); 4374 } 4375 4376 if (pp != NULL && pvn_getdirty(pp, flags)) { 4377 int err; 4378 4379 /* 4380 * Found a dirty page to push 4381 */ 4382 err = zfs_putapage(vp, pp, &io_off, &io_len, flags, cr); 4383 if (err) 4384 error = err; 4385 } else { 4386 io_len = PAGESIZE; 4387 } 4388 } 4389 out: 4390 zfs_range_unlock(rl); 4391 if ((flags & B_ASYNC) == 0 || zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 4392 zil_commit(zfsvfs->z_log, zp->z_id); 4393 ZFS_EXIT(zfsvfs); 4394 return (error); 4395 } 4396 4397 /*ARGSUSED*/ 4398 void 4399 zfs_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct) 4400 { 4401 znode_t *zp = VTOZ(vp); 4402 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4403 int error; 4404 4405 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 4406 if (zp->z_sa_hdl == NULL) { 4407 /* 4408 * The fs has been unmounted, or we did a 4409 * suspend/resume and this file no longer exists. 4410 */ 4411 if (vn_has_cached_data(vp)) { 4412 (void) pvn_vplist_dirty(vp, 0, zfs_null_putapage, 4413 B_INVAL, cr); 4414 } 4415 4416 mutex_enter(&zp->z_lock); 4417 mutex_enter(&vp->v_lock); 4418 ASSERT(vp->v_count == 1); 4419 VN_RELE_LOCKED(vp); 4420 mutex_exit(&vp->v_lock); 4421 mutex_exit(&zp->z_lock); 4422 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4423 zfs_znode_free(zp); 4424 return; 4425 } 4426 4427 /* 4428 * Attempt to push any data in the page cache. If this fails 4429 * we will get kicked out later in zfs_zinactive(). 4430 */ 4431 if (vn_has_cached_data(vp)) { 4432 (void) pvn_vplist_dirty(vp, 0, zfs_putapage, B_INVAL|B_ASYNC, 4433 cr); 4434 } 4435 4436 if (zp->z_atime_dirty && zp->z_unlinked == 0) { 4437 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 4438 4439 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 4440 zfs_sa_upgrade_txholds(tx, zp); 4441 error = dmu_tx_assign(tx, TXG_WAIT); 4442 if (error) { 4443 dmu_tx_abort(tx); 4444 } else { 4445 mutex_enter(&zp->z_lock); 4446 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs), 4447 (void *)&zp->z_atime, sizeof (zp->z_atime), tx); 4448 zp->z_atime_dirty = 0; 4449 mutex_exit(&zp->z_lock); 4450 dmu_tx_commit(tx); 4451 } 4452 } 4453 4454 zfs_zinactive(zp); 4455 rw_exit(&zfsvfs->z_teardown_inactive_lock); 4456 } 4457 4458 /* 4459 * Bounds-check the seek operation. 4460 * 4461 * IN: vp - vnode seeking within 4462 * ooff - old file offset 4463 * noffp - pointer to new file offset 4464 * ct - caller context 4465 * 4466 * RETURN: 0 on success, EINVAL if new offset invalid. 4467 */ 4468 /* ARGSUSED */ 4469 static int 4470 zfs_seek(vnode_t *vp, offset_t ooff, offset_t *noffp, 4471 caller_context_t *ct) 4472 { 4473 if (vp->v_type == VDIR) 4474 return (0); 4475 return ((*noffp < 0 || *noffp > MAXOFFSET_T) ? EINVAL : 0); 4476 } 4477 4478 /* 4479 * Pre-filter the generic locking function to trap attempts to place 4480 * a mandatory lock on a memory mapped file. 4481 */ 4482 static int 4483 zfs_frlock(vnode_t *vp, int cmd, flock64_t *bfp, int flag, offset_t offset, 4484 flk_callback_t *flk_cbp, cred_t *cr, caller_context_t *ct) 4485 { 4486 znode_t *zp = VTOZ(vp); 4487 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4488 4489 ZFS_ENTER(zfsvfs); 4490 ZFS_VERIFY_ZP(zp); 4491 4492 /* 4493 * We are following the UFS semantics with respect to mapcnt 4494 * here: If we see that the file is mapped already, then we will 4495 * return an error, but we don't worry about races between this 4496 * function and zfs_map(). 4497 */ 4498 if (zp->z_mapcnt > 0 && MANDMODE(zp->z_mode)) { 4499 ZFS_EXIT(zfsvfs); 4500 return (SET_ERROR(EAGAIN)); 4501 } 4502 ZFS_EXIT(zfsvfs); 4503 return (fs_frlock(vp, cmd, bfp, flag, offset, flk_cbp, cr, ct)); 4504 } 4505 4506 /* 4507 * If we can't find a page in the cache, we will create a new page 4508 * and fill it with file data. For efficiency, we may try to fill 4509 * multiple pages at once (klustering) to fill up the supplied page 4510 * list. Note that the pages to be filled are held with an exclusive 4511 * lock to prevent access by other threads while they are being filled. 4512 */ 4513 static int 4514 zfs_fillpage(vnode_t *vp, u_offset_t off, struct seg *seg, 4515 caddr_t addr, page_t *pl[], size_t plsz, enum seg_rw rw) 4516 { 4517 znode_t *zp = VTOZ(vp); 4518 page_t *pp, *cur_pp; 4519 objset_t *os = zp->z_zfsvfs->z_os; 4520 u_offset_t io_off, total; 4521 size_t io_len; 4522 int err; 4523 4524 if (plsz == PAGESIZE || zp->z_blksz <= PAGESIZE) { 4525 /* 4526 * We only have a single page, don't bother klustering 4527 */ 4528 io_off = off; 4529 io_len = PAGESIZE; 4530 pp = page_create_va(vp, io_off, io_len, 4531 PG_EXCL | PG_WAIT, seg, addr); 4532 } else { 4533 /* 4534 * Try to find enough pages to fill the page list 4535 */ 4536 pp = pvn_read_kluster(vp, off, seg, addr, &io_off, 4537 &io_len, off, plsz, 0); 4538 } 4539 if (pp == NULL) { 4540 /* 4541 * The page already exists, nothing to do here. 4542 */ 4543 *pl = NULL; 4544 return (0); 4545 } 4546 4547 /* 4548 * Fill the pages in the kluster. 4549 */ 4550 cur_pp = pp; 4551 for (total = io_off + io_len; io_off < total; io_off += PAGESIZE) { 4552 caddr_t va; 4553 4554 ASSERT3U(io_off, ==, cur_pp->p_offset); 4555 va = zfs_map_page(cur_pp, S_WRITE); 4556 err = dmu_read(os, zp->z_id, io_off, PAGESIZE, va, 4557 DMU_READ_PREFETCH); 4558 zfs_unmap_page(cur_pp, va); 4559 if (err) { 4560 /* On error, toss the entire kluster */ 4561 pvn_read_done(pp, B_ERROR); 4562 /* convert checksum errors into IO errors */ 4563 if (err == ECKSUM) 4564 err = SET_ERROR(EIO); 4565 return (err); 4566 } 4567 cur_pp = cur_pp->p_next; 4568 } 4569 4570 /* 4571 * Fill in the page list array from the kluster starting 4572 * from the desired offset `off'. 4573 * NOTE: the page list will always be null terminated. 4574 */ 4575 pvn_plist_init(pp, pl, plsz, off, io_len, rw); 4576 ASSERT(pl == NULL || (*pl)->p_offset == off); 4577 4578 return (0); 4579 } 4580 4581 /* 4582 * Return pointers to the pages for the file region [off, off + len] 4583 * in the pl array. If plsz is greater than len, this function may 4584 * also return page pointers from after the specified region 4585 * (i.e. the region [off, off + plsz]). These additional pages are 4586 * only returned if they are already in the cache, or were created as 4587 * part of a klustered read. 4588 * 4589 * IN: vp - vnode of file to get data from. 4590 * off - position in file to get data from. 4591 * len - amount of data to retrieve. 4592 * plsz - length of provided page list. 4593 * seg - segment to obtain pages for. 4594 * addr - virtual address of fault. 4595 * rw - mode of created pages. 4596 * cr - credentials of caller. 4597 * ct - caller context. 4598 * 4599 * OUT: protp - protection mode of created pages. 4600 * pl - list of pages created. 4601 * 4602 * RETURN: 0 on success, error code on failure. 4603 * 4604 * Timestamps: 4605 * vp - atime updated 4606 */ 4607 /* ARGSUSED */ 4608 static int 4609 zfs_getpage(vnode_t *vp, offset_t off, size_t len, uint_t *protp, 4610 page_t *pl[], size_t plsz, struct seg *seg, caddr_t addr, 4611 enum seg_rw rw, cred_t *cr, caller_context_t *ct) 4612 { 4613 znode_t *zp = VTOZ(vp); 4614 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4615 page_t **pl0 = pl; 4616 int err = 0; 4617 4618 /* we do our own caching, faultahead is unnecessary */ 4619 if (pl == NULL) 4620 return (0); 4621 else if (len > plsz) 4622 len = plsz; 4623 else 4624 len = P2ROUNDUP(len, PAGESIZE); 4625 ASSERT(plsz >= len); 4626 4627 ZFS_ENTER(zfsvfs); 4628 ZFS_VERIFY_ZP(zp); 4629 4630 if (protp) 4631 *protp = PROT_ALL; 4632 4633 /* 4634 * Loop through the requested range [off, off + len) looking 4635 * for pages. If we don't find a page, we will need to create 4636 * a new page and fill it with data from the file. 4637 */ 4638 while (len > 0) { 4639 if (*pl = page_lookup(vp, off, SE_SHARED)) 4640 *(pl+1) = NULL; 4641 else if (err = zfs_fillpage(vp, off, seg, addr, pl, plsz, rw)) 4642 goto out; 4643 while (*pl) { 4644 ASSERT3U((*pl)->p_offset, ==, off); 4645 off += PAGESIZE; 4646 addr += PAGESIZE; 4647 if (len > 0) { 4648 ASSERT3U(len, >=, PAGESIZE); 4649 len -= PAGESIZE; 4650 } 4651 ASSERT3U(plsz, >=, PAGESIZE); 4652 plsz -= PAGESIZE; 4653 pl++; 4654 } 4655 } 4656 4657 /* 4658 * Fill out the page array with any pages already in the cache. 4659 */ 4660 while (plsz > 0 && 4661 (*pl++ = page_lookup_nowait(vp, off, SE_SHARED))) { 4662 off += PAGESIZE; 4663 plsz -= PAGESIZE; 4664 } 4665 out: 4666 if (err) { 4667 /* 4668 * Release any pages we have previously locked. 4669 */ 4670 while (pl > pl0) 4671 page_unlock(*--pl); 4672 } else { 4673 ZFS_ACCESSTIME_STAMP(zfsvfs, zp); 4674 } 4675 4676 *pl = NULL; 4677 4678 ZFS_EXIT(zfsvfs); 4679 return (err); 4680 } 4681 4682 /* 4683 * Request a memory map for a section of a file. This code interacts 4684 * with common code and the VM system as follows: 4685 * 4686 * - common code calls mmap(), which ends up in smmap_common() 4687 * - this calls VOP_MAP(), which takes you into (say) zfs 4688 * - zfs_map() calls as_map(), passing segvn_create() as the callback 4689 * - segvn_create() creates the new segment and calls VOP_ADDMAP() 4690 * - zfs_addmap() updates z_mapcnt 4691 */ 4692 /*ARGSUSED*/ 4693 static int 4694 zfs_map(vnode_t *vp, offset_t off, struct as *as, caddr_t *addrp, 4695 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4696 caller_context_t *ct) 4697 { 4698 znode_t *zp = VTOZ(vp); 4699 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4700 segvn_crargs_t vn_a; 4701 int error; 4702 4703 ZFS_ENTER(zfsvfs); 4704 ZFS_VERIFY_ZP(zp); 4705 4706 /* 4707 * Note: ZFS_READONLY is handled in zfs_zaccess_common. 4708 */ 4709 4710 if ((prot & PROT_WRITE) && (zp->z_pflags & 4711 (ZFS_IMMUTABLE | ZFS_APPENDONLY))) { 4712 ZFS_EXIT(zfsvfs); 4713 return (SET_ERROR(EPERM)); 4714 } 4715 4716 if ((prot & (PROT_READ | PROT_EXEC)) && 4717 (zp->z_pflags & ZFS_AV_QUARANTINED)) { 4718 ZFS_EXIT(zfsvfs); 4719 return (SET_ERROR(EACCES)); 4720 } 4721 4722 if (vp->v_flag & VNOMAP) { 4723 ZFS_EXIT(zfsvfs); 4724 return (SET_ERROR(ENOSYS)); 4725 } 4726 4727 if (off < 0 || len > MAXOFFSET_T - off) { 4728 ZFS_EXIT(zfsvfs); 4729 return (SET_ERROR(ENXIO)); 4730 } 4731 4732 if (vp->v_type != VREG) { 4733 ZFS_EXIT(zfsvfs); 4734 return (SET_ERROR(ENODEV)); 4735 } 4736 4737 /* 4738 * If file is locked, disallow mapping. 4739 */ 4740 if (MANDMODE(zp->z_mode) && vn_has_flocks(vp)) { 4741 ZFS_EXIT(zfsvfs); 4742 return (SET_ERROR(EAGAIN)); 4743 } 4744 4745 as_rangelock(as); 4746 error = choose_addr(as, addrp, len, off, ADDR_VACALIGN, flags); 4747 if (error != 0) { 4748 as_rangeunlock(as); 4749 ZFS_EXIT(zfsvfs); 4750 return (error); 4751 } 4752 4753 vn_a.vp = vp; 4754 vn_a.offset = (u_offset_t)off; 4755 vn_a.type = flags & MAP_TYPE; 4756 vn_a.prot = prot; 4757 vn_a.maxprot = maxprot; 4758 vn_a.cred = cr; 4759 vn_a.amp = NULL; 4760 vn_a.flags = flags & ~MAP_TYPE; 4761 vn_a.szc = 0; 4762 vn_a.lgrp_mem_policy_flags = 0; 4763 4764 error = as_map(as, *addrp, len, segvn_create, &vn_a); 4765 4766 as_rangeunlock(as); 4767 ZFS_EXIT(zfsvfs); 4768 return (error); 4769 } 4770 4771 /* ARGSUSED */ 4772 static int 4773 zfs_addmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 4774 size_t len, uchar_t prot, uchar_t maxprot, uint_t flags, cred_t *cr, 4775 caller_context_t *ct) 4776 { 4777 uint64_t pages = btopr(len); 4778 4779 atomic_add_64(&VTOZ(vp)->z_mapcnt, pages); 4780 return (0); 4781 } 4782 4783 /* 4784 * The reason we push dirty pages as part of zfs_delmap() is so that we get a 4785 * more accurate mtime for the associated file. Since we don't have a way of 4786 * detecting when the data was actually modified, we have to resort to 4787 * heuristics. If an explicit msync() is done, then we mark the mtime when the 4788 * last page is pushed. The problem occurs when the msync() call is omitted, 4789 * which by far the most common case: 4790 * 4791 * open() 4792 * mmap() 4793 * <modify memory> 4794 * munmap() 4795 * close() 4796 * <time lapse> 4797 * putpage() via fsflush 4798 * 4799 * If we wait until fsflush to come along, we can have a modification time that 4800 * is some arbitrary point in the future. In order to prevent this in the 4801 * common case, we flush pages whenever a (MAP_SHARED, PROT_WRITE) mapping is 4802 * torn down. 4803 */ 4804 /* ARGSUSED */ 4805 static int 4806 zfs_delmap(vnode_t *vp, offset_t off, struct as *as, caddr_t addr, 4807 size_t len, uint_t prot, uint_t maxprot, uint_t flags, cred_t *cr, 4808 caller_context_t *ct) 4809 { 4810 uint64_t pages = btopr(len); 4811 4812 ASSERT3U(VTOZ(vp)->z_mapcnt, >=, pages); 4813 atomic_add_64(&VTOZ(vp)->z_mapcnt, -pages); 4814 4815 if ((flags & MAP_SHARED) && (prot & PROT_WRITE) && 4816 vn_has_cached_data(vp)) 4817 (void) VOP_PUTPAGE(vp, off, len, B_ASYNC, cr, ct); 4818 4819 return (0); 4820 } 4821 4822 /* 4823 * Free or allocate space in a file. Currently, this function only 4824 * supports the `F_FREESP' command. However, this command is somewhat 4825 * misnamed, as its functionality includes the ability to allocate as 4826 * well as free space. 4827 * 4828 * IN: vp - vnode of file to free data in. 4829 * cmd - action to take (only F_FREESP supported). 4830 * bfp - section of file to free/alloc. 4831 * flag - current file open mode flags. 4832 * offset - current file offset. 4833 * cr - credentials of caller [UNUSED]. 4834 * ct - caller context. 4835 * 4836 * RETURN: 0 on success, error code on failure. 4837 * 4838 * Timestamps: 4839 * vp - ctime|mtime updated 4840 */ 4841 /* ARGSUSED */ 4842 static int 4843 zfs_space(vnode_t *vp, int cmd, flock64_t *bfp, int flag, 4844 offset_t offset, cred_t *cr, caller_context_t *ct) 4845 { 4846 znode_t *zp = VTOZ(vp); 4847 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4848 uint64_t off, len; 4849 int error; 4850 4851 ZFS_ENTER(zfsvfs); 4852 ZFS_VERIFY_ZP(zp); 4853 4854 if (cmd != F_FREESP) { 4855 ZFS_EXIT(zfsvfs); 4856 return (SET_ERROR(EINVAL)); 4857 } 4858 4859 /* 4860 * In a case vp->v_vfsp != zp->z_zfsvfs->z_vfs (e.g. snapshots) our 4861 * callers might not be able to detect properly that we are read-only, 4862 * so check it explicitly here. 4863 */ 4864 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 4865 ZFS_EXIT(zfsvfs); 4866 return (SET_ERROR(EROFS)); 4867 } 4868 4869 if (error = convoff(vp, bfp, 0, offset)) { 4870 ZFS_EXIT(zfsvfs); 4871 return (error); 4872 } 4873 4874 if (bfp->l_len < 0) { 4875 ZFS_EXIT(zfsvfs); 4876 return (SET_ERROR(EINVAL)); 4877 } 4878 4879 off = bfp->l_start; 4880 len = bfp->l_len; /* 0 means from off to end of file */ 4881 4882 error = zfs_freesp(zp, off, len, flag, TRUE); 4883 4884 if (error == 0 && off == 0 && len == 0) 4885 vnevent_truncate(ZTOV(zp), ct); 4886 4887 ZFS_EXIT(zfsvfs); 4888 return (error); 4889 } 4890 4891 /*ARGSUSED*/ 4892 static int 4893 zfs_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct) 4894 { 4895 znode_t *zp = VTOZ(vp); 4896 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 4897 uint32_t gen; 4898 uint64_t gen64; 4899 uint64_t object = zp->z_id; 4900 zfid_short_t *zfid; 4901 int size, i, error; 4902 4903 ZFS_ENTER(zfsvfs); 4904 ZFS_VERIFY_ZP(zp); 4905 4906 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), 4907 &gen64, sizeof (uint64_t))) != 0) { 4908 ZFS_EXIT(zfsvfs); 4909 return (error); 4910 } 4911 4912 gen = (uint32_t)gen64; 4913 4914 size = (zfsvfs->z_parent != zfsvfs) ? LONG_FID_LEN : SHORT_FID_LEN; 4915 if (fidp->fid_len < size) { 4916 fidp->fid_len = size; 4917 ZFS_EXIT(zfsvfs); 4918 return (SET_ERROR(ENOSPC)); 4919 } 4920 4921 zfid = (zfid_short_t *)fidp; 4922 4923 zfid->zf_len = size; 4924 4925 for (i = 0; i < sizeof (zfid->zf_object); i++) 4926 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 4927 4928 /* Must have a non-zero generation number to distinguish from .zfs */ 4929 if (gen == 0) 4930 gen = 1; 4931 for (i = 0; i < sizeof (zfid->zf_gen); i++) 4932 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 4933 4934 if (size == LONG_FID_LEN) { 4935 uint64_t objsetid = dmu_objset_id(zfsvfs->z_os); 4936 zfid_long_t *zlfid; 4937 4938 zlfid = (zfid_long_t *)fidp; 4939 4940 for (i = 0; i < sizeof (zlfid->zf_setid); i++) 4941 zlfid->zf_setid[i] = (uint8_t)(objsetid >> (8 * i)); 4942 4943 /* XXX - this should be the generation number for the objset */ 4944 for (i = 0; i < sizeof (zlfid->zf_setgen); i++) 4945 zlfid->zf_setgen[i] = 0; 4946 } 4947 4948 ZFS_EXIT(zfsvfs); 4949 return (0); 4950 } 4951 4952 static int 4953 zfs_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr, 4954 caller_context_t *ct) 4955 { 4956 znode_t *zp, *xzp; 4957 zfsvfs_t *zfsvfs; 4958 zfs_dirlock_t *dl; 4959 int error; 4960 4961 switch (cmd) { 4962 case _PC_LINK_MAX: 4963 *valp = ULONG_MAX; 4964 return (0); 4965 4966 case _PC_FILESIZEBITS: 4967 *valp = 64; 4968 return (0); 4969 4970 case _PC_XATTR_EXISTS: 4971 zp = VTOZ(vp); 4972 zfsvfs = zp->z_zfsvfs; 4973 ZFS_ENTER(zfsvfs); 4974 ZFS_VERIFY_ZP(zp); 4975 *valp = 0; 4976 error = zfs_dirent_lock(&dl, zp, "", &xzp, 4977 ZXATTR | ZEXISTS | ZSHARED, NULL, NULL); 4978 if (error == 0) { 4979 zfs_dirent_unlock(dl); 4980 if (!zfs_dirempty(xzp)) 4981 *valp = 1; 4982 VN_RELE(ZTOV(xzp)); 4983 } else if (error == ENOENT) { 4984 /* 4985 * If there aren't extended attributes, it's the 4986 * same as having zero of them. 4987 */ 4988 error = 0; 4989 } 4990 ZFS_EXIT(zfsvfs); 4991 return (error); 4992 4993 case _PC_SATTR_ENABLED: 4994 case _PC_SATTR_EXISTS: 4995 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_SYSATTR_VIEWS) && 4996 (vp->v_type == VREG || vp->v_type == VDIR); 4997 return (0); 4998 4999 case _PC_ACCESS_FILTERING: 5000 *valp = vfs_has_feature(vp->v_vfsp, VFSFT_ACCESS_FILTER) && 5001 vp->v_type == VDIR; 5002 return (0); 5003 5004 case _PC_ACL_ENABLED: 5005 *valp = _ACL_ACE_ENABLED; 5006 return (0); 5007 5008 case _PC_MIN_HOLE_SIZE: 5009 *valp = (ulong_t)SPA_MINBLOCKSIZE; 5010 return (0); 5011 5012 case _PC_TIMESTAMP_RESOLUTION: 5013 /* nanosecond timestamp resolution */ 5014 *valp = 1L; 5015 return (0); 5016 5017 default: 5018 return (fs_pathconf(vp, cmd, valp, cr, ct)); 5019 } 5020 } 5021 5022 /*ARGSUSED*/ 5023 static int 5024 zfs_getsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 5025 caller_context_t *ct) 5026 { 5027 znode_t *zp = VTOZ(vp); 5028 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5029 int error; 5030 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 5031 5032 ZFS_ENTER(zfsvfs); 5033 ZFS_VERIFY_ZP(zp); 5034 error = zfs_getacl(zp, vsecp, skipaclchk, cr); 5035 ZFS_EXIT(zfsvfs); 5036 5037 return (error); 5038 } 5039 5040 /*ARGSUSED*/ 5041 static int 5042 zfs_setsecattr(vnode_t *vp, vsecattr_t *vsecp, int flag, cred_t *cr, 5043 caller_context_t *ct) 5044 { 5045 znode_t *zp = VTOZ(vp); 5046 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5047 int error; 5048 boolean_t skipaclchk = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 5049 zilog_t *zilog = zfsvfs->z_log; 5050 5051 ZFS_ENTER(zfsvfs); 5052 ZFS_VERIFY_ZP(zp); 5053 5054 error = zfs_setacl(zp, vsecp, skipaclchk, cr); 5055 5056 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 5057 zil_commit(zilog, 0); 5058 5059 ZFS_EXIT(zfsvfs); 5060 return (error); 5061 } 5062 5063 /* 5064 * The smallest read we may consider to loan out an arcbuf. 5065 * This must be a power of 2. 5066 */ 5067 int zcr_blksz_min = (1 << 10); /* 1K */ 5068 /* 5069 * If set to less than the file block size, allow loaning out of an 5070 * arcbuf for a partial block read. This must be a power of 2. 5071 */ 5072 int zcr_blksz_max = (1 << 17); /* 128K */ 5073 5074 /*ARGSUSED*/ 5075 static int 5076 zfs_reqzcbuf(vnode_t *vp, enum uio_rw ioflag, xuio_t *xuio, cred_t *cr, 5077 caller_context_t *ct) 5078 { 5079 znode_t *zp = VTOZ(vp); 5080 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 5081 int max_blksz = zfsvfs->z_max_blksz; 5082 uio_t *uio = &xuio->xu_uio; 5083 ssize_t size = uio->uio_resid; 5084 offset_t offset = uio->uio_loffset; 5085 int blksz; 5086 int fullblk, i; 5087 arc_buf_t *abuf; 5088 ssize_t maxsize; 5089 int preamble, postamble; 5090 5091 if (xuio->xu_type != UIOTYPE_ZEROCOPY) 5092 return (SET_ERROR(EINVAL)); 5093 5094 ZFS_ENTER(zfsvfs); 5095 ZFS_VERIFY_ZP(zp); 5096 switch (ioflag) { 5097 case UIO_WRITE: 5098 /* 5099 * Loan out an arc_buf for write if write size is bigger than 5100 * max_blksz, and the file's block size is also max_blksz. 5101 */ 5102 blksz = max_blksz; 5103 if (size < blksz || zp->z_blksz != blksz) { 5104 ZFS_EXIT(zfsvfs); 5105 return (SET_ERROR(EINVAL)); 5106 } 5107 /* 5108 * Caller requests buffers for write before knowing where the 5109 * write offset might be (e.g. NFS TCP write). 5110 */ 5111 if (offset == -1) { 5112 preamble = 0; 5113 } else { 5114 preamble = P2PHASE(offset, blksz); 5115 if (preamble) { 5116 preamble = blksz - preamble; 5117 size -= preamble; 5118 } 5119 } 5120 5121 postamble = P2PHASE(size, blksz); 5122 size -= postamble; 5123 5124 fullblk = size / blksz; 5125 (void) dmu_xuio_init(xuio, 5126 (preamble != 0) + fullblk + (postamble != 0)); 5127 DTRACE_PROBE3(zfs_reqzcbuf_align, int, preamble, 5128 int, postamble, int, 5129 (preamble != 0) + fullblk + (postamble != 0)); 5130 5131 /* 5132 * Have to fix iov base/len for partial buffers. They 5133 * currently represent full arc_buf's. 5134 */ 5135 if (preamble) { 5136 /* data begins in the middle of the arc_buf */ 5137 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5138 blksz); 5139 ASSERT(abuf); 5140 (void) dmu_xuio_add(xuio, abuf, 5141 blksz - preamble, preamble); 5142 } 5143 5144 for (i = 0; i < fullblk; i++) { 5145 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5146 blksz); 5147 ASSERT(abuf); 5148 (void) dmu_xuio_add(xuio, abuf, 0, blksz); 5149 } 5150 5151 if (postamble) { 5152 /* data ends in the middle of the arc_buf */ 5153 abuf = dmu_request_arcbuf(sa_get_db(zp->z_sa_hdl), 5154 blksz); 5155 ASSERT(abuf); 5156 (void) dmu_xuio_add(xuio, abuf, 0, postamble); 5157 } 5158 break; 5159 case UIO_READ: 5160 /* 5161 * Loan out an arc_buf for read if the read size is larger than 5162 * the current file block size. Block alignment is not 5163 * considered. Partial arc_buf will be loaned out for read. 5164 */ 5165 blksz = zp->z_blksz; 5166 if (blksz < zcr_blksz_min) 5167 blksz = zcr_blksz_min; 5168 if (blksz > zcr_blksz_max) 5169 blksz = zcr_blksz_max; 5170 /* avoid potential complexity of dealing with it */ 5171 if (blksz > max_blksz) { 5172 ZFS_EXIT(zfsvfs); 5173 return (SET_ERROR(EINVAL)); 5174 } 5175 5176 maxsize = zp->z_size - uio->uio_loffset; 5177 if (size > maxsize) 5178 size = maxsize; 5179 5180 if (size < blksz || vn_has_cached_data(vp)) { 5181 ZFS_EXIT(zfsvfs); 5182 return (SET_ERROR(EINVAL)); 5183 } 5184 break; 5185 default: 5186 ZFS_EXIT(zfsvfs); 5187 return (SET_ERROR(EINVAL)); 5188 } 5189 5190 uio->uio_extflg = UIO_XUIO; 5191 XUIO_XUZC_RW(xuio) = ioflag; 5192 ZFS_EXIT(zfsvfs); 5193 return (0); 5194 } 5195 5196 /*ARGSUSED*/ 5197 static int 5198 zfs_retzcbuf(vnode_t *vp, xuio_t *xuio, cred_t *cr, caller_context_t *ct) 5199 { 5200 int i; 5201 arc_buf_t *abuf; 5202 int ioflag = XUIO_XUZC_RW(xuio); 5203 5204 ASSERT(xuio->xu_type == UIOTYPE_ZEROCOPY); 5205 5206 i = dmu_xuio_cnt(xuio); 5207 while (i-- > 0) { 5208 abuf = dmu_xuio_arcbuf(xuio, i); 5209 /* 5210 * if abuf == NULL, it must be a write buffer 5211 * that has been returned in zfs_write(). 5212 */ 5213 if (abuf) 5214 dmu_return_arcbuf(abuf); 5215 ASSERT(abuf || ioflag == UIO_WRITE); 5216 } 5217 5218 dmu_xuio_fini(xuio); 5219 return (0); 5220 } 5221 5222 /* 5223 * Predeclare these here so that the compiler assumes that 5224 * this is an "old style" function declaration that does 5225 * not include arguments => we won't get type mismatch errors 5226 * in the initializations that follow. 5227 */ 5228 static int zfs_inval(); 5229 static int zfs_isdir(); 5230 5231 static int 5232 zfs_inval() 5233 { 5234 return (SET_ERROR(EINVAL)); 5235 } 5236 5237 static int 5238 zfs_isdir() 5239 { 5240 return (SET_ERROR(EISDIR)); 5241 } 5242 /* 5243 * Directory vnode operations template 5244 */ 5245 vnodeops_t *zfs_dvnodeops; 5246 const fs_operation_def_t zfs_dvnodeops_template[] = { 5247 VOPNAME_OPEN, { .vop_open = zfs_open }, 5248 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5249 VOPNAME_READ, { .error = zfs_isdir }, 5250 VOPNAME_WRITE, { .error = zfs_isdir }, 5251 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5252 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5253 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5254 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5255 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5256 VOPNAME_CREATE, { .vop_create = zfs_create }, 5257 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 5258 VOPNAME_LINK, { .vop_link = zfs_link }, 5259 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5260 VOPNAME_MKDIR, { .vop_mkdir = zfs_mkdir }, 5261 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 5262 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 5263 VOPNAME_SYMLINK, { .vop_symlink = zfs_symlink }, 5264 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5265 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5266 VOPNAME_FID, { .vop_fid = zfs_fid }, 5267 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5268 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5269 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5270 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5271 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5272 NULL, NULL 5273 }; 5274 5275 /* 5276 * Regular file vnode operations template 5277 */ 5278 vnodeops_t *zfs_fvnodeops; 5279 const fs_operation_def_t zfs_fvnodeops_template[] = { 5280 VOPNAME_OPEN, { .vop_open = zfs_open }, 5281 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5282 VOPNAME_READ, { .vop_read = zfs_read }, 5283 VOPNAME_WRITE, { .vop_write = zfs_write }, 5284 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5285 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5286 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5287 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5288 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5289 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5290 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5291 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5292 VOPNAME_FID, { .vop_fid = zfs_fid }, 5293 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5294 VOPNAME_FRLOCK, { .vop_frlock = zfs_frlock }, 5295 VOPNAME_SPACE, { .vop_space = zfs_space }, 5296 VOPNAME_GETPAGE, { .vop_getpage = zfs_getpage }, 5297 VOPNAME_PUTPAGE, { .vop_putpage = zfs_putpage }, 5298 VOPNAME_MAP, { .vop_map = zfs_map }, 5299 VOPNAME_ADDMAP, { .vop_addmap = zfs_addmap }, 5300 VOPNAME_DELMAP, { .vop_delmap = zfs_delmap }, 5301 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5302 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5303 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5304 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5305 VOPNAME_REQZCBUF, { .vop_reqzcbuf = zfs_reqzcbuf }, 5306 VOPNAME_RETZCBUF, { .vop_retzcbuf = zfs_retzcbuf }, 5307 NULL, NULL 5308 }; 5309 5310 /* 5311 * Symbolic link vnode operations template 5312 */ 5313 vnodeops_t *zfs_symvnodeops; 5314 const fs_operation_def_t zfs_symvnodeops_template[] = { 5315 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5316 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5317 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5318 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5319 VOPNAME_READLINK, { .vop_readlink = zfs_readlink }, 5320 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5321 VOPNAME_FID, { .vop_fid = zfs_fid }, 5322 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5323 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5324 NULL, NULL 5325 }; 5326 5327 /* 5328 * special share hidden files vnode operations template 5329 */ 5330 vnodeops_t *zfs_sharevnodeops; 5331 const fs_operation_def_t zfs_sharevnodeops_template[] = { 5332 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5333 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5334 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5335 VOPNAME_FID, { .vop_fid = zfs_fid }, 5336 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5337 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5338 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5339 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5340 NULL, NULL 5341 }; 5342 5343 /* 5344 * Extended attribute directory vnode operations template 5345 * 5346 * This template is identical to the directory vnodes 5347 * operation template except for restricted operations: 5348 * VOP_MKDIR() 5349 * VOP_SYMLINK() 5350 * 5351 * Note that there are other restrictions embedded in: 5352 * zfs_create() - restrict type to VREG 5353 * zfs_link() - no links into/out of attribute space 5354 * zfs_rename() - no moves into/out of attribute space 5355 */ 5356 vnodeops_t *zfs_xdvnodeops; 5357 const fs_operation_def_t zfs_xdvnodeops_template[] = { 5358 VOPNAME_OPEN, { .vop_open = zfs_open }, 5359 VOPNAME_CLOSE, { .vop_close = zfs_close }, 5360 VOPNAME_IOCTL, { .vop_ioctl = zfs_ioctl }, 5361 VOPNAME_GETATTR, { .vop_getattr = zfs_getattr }, 5362 VOPNAME_SETATTR, { .vop_setattr = zfs_setattr }, 5363 VOPNAME_ACCESS, { .vop_access = zfs_access }, 5364 VOPNAME_LOOKUP, { .vop_lookup = zfs_lookup }, 5365 VOPNAME_CREATE, { .vop_create = zfs_create }, 5366 VOPNAME_REMOVE, { .vop_remove = zfs_remove }, 5367 VOPNAME_LINK, { .vop_link = zfs_link }, 5368 VOPNAME_RENAME, { .vop_rename = zfs_rename }, 5369 VOPNAME_MKDIR, { .error = zfs_inval }, 5370 VOPNAME_RMDIR, { .vop_rmdir = zfs_rmdir }, 5371 VOPNAME_READDIR, { .vop_readdir = zfs_readdir }, 5372 VOPNAME_SYMLINK, { .error = zfs_inval }, 5373 VOPNAME_FSYNC, { .vop_fsync = zfs_fsync }, 5374 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5375 VOPNAME_FID, { .vop_fid = zfs_fid }, 5376 VOPNAME_SEEK, { .vop_seek = zfs_seek }, 5377 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5378 VOPNAME_GETSECATTR, { .vop_getsecattr = zfs_getsecattr }, 5379 VOPNAME_SETSECATTR, { .vop_setsecattr = zfs_setsecattr }, 5380 VOPNAME_VNEVENT, { .vop_vnevent = fs_vnevent_support }, 5381 NULL, NULL 5382 }; 5383 5384 /* 5385 * Error vnode operations template 5386 */ 5387 vnodeops_t *zfs_evnodeops; 5388 const fs_operation_def_t zfs_evnodeops_template[] = { 5389 VOPNAME_INACTIVE, { .vop_inactive = zfs_inactive }, 5390 VOPNAME_PATHCONF, { .vop_pathconf = zfs_pathconf }, 5391 NULL, NULL 5392 }; 5393