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