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 https://opensource.org/licenses/CDDL-1.0. 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, 2018 by Delphix. All rights reserved. 25 * Copyright (c) 2015 by Chunwei Chen. All rights reserved. 26 * Copyright 2017 Nexenta Systems, Inc. 27 */ 28 29 /* Portions Copyright 2007 Jeremy Teo */ 30 /* Portions Copyright 2010 Robert Milkowski */ 31 32 33 #include <sys/types.h> 34 #include <sys/param.h> 35 #include <sys/time.h> 36 #include <sys/sysmacros.h> 37 #include <sys/vfs.h> 38 #include <sys/file.h> 39 #include <sys/stat.h> 40 #include <sys/kmem.h> 41 #include <sys/taskq.h> 42 #include <sys/uio.h> 43 #include <sys/vmsystm.h> 44 #include <sys/atomic.h> 45 #include <sys/pathname.h> 46 #include <sys/cmn_err.h> 47 #include <sys/errno.h> 48 #include <sys/zfs_dir.h> 49 #include <sys/zfs_acl.h> 50 #include <sys/zfs_ioctl.h> 51 #include <sys/fs/zfs.h> 52 #include <sys/dmu.h> 53 #include <sys/dmu_objset.h> 54 #include <sys/spa.h> 55 #include <sys/txg.h> 56 #include <sys/dbuf.h> 57 #include <sys/zap.h> 58 #include <sys/sa.h> 59 #include <sys/policy.h> 60 #include <sys/sunddi.h> 61 #include <sys/sid.h> 62 #include <sys/zfs_ctldir.h> 63 #include <sys/zfs_fuid.h> 64 #include <sys/zfs_quota.h> 65 #include <sys/zfs_sa.h> 66 #include <sys/zfs_vnops.h> 67 #include <sys/zfs_rlock.h> 68 #include <sys/cred.h> 69 #include <sys/zpl.h> 70 #include <sys/zil.h> 71 #include <sys/sa_impl.h> 72 73 /* 74 * Programming rules. 75 * 76 * Each vnode op performs some logical unit of work. To do this, the ZPL must 77 * properly lock its in-core state, create a DMU transaction, do the work, 78 * record this work in the intent log (ZIL), commit the DMU transaction, 79 * and wait for the intent log to commit if it is a synchronous operation. 80 * Moreover, the vnode ops must work in both normal and log replay context. 81 * The ordering of events is important to avoid deadlocks and references 82 * to freed memory. The example below illustrates the following Big Rules: 83 * 84 * (1) A check must be made in each zfs thread for a mounted file system. 85 * This is done avoiding races using zfs_enter(zfsvfs). 86 * A zfs_exit(zfsvfs) is needed before all returns. Any znodes 87 * must be checked with zfs_verify_zp(zp). Both of these macros 88 * can return EIO from the calling function. 89 * 90 * (2) zrele() should always be the last thing except for zil_commit() (if 91 * necessary) and zfs_exit(). This is for 3 reasons: First, if it's the 92 * last reference, the vnode/znode can be freed, so the zp may point to 93 * freed memory. Second, the last reference will call zfs_zinactive(), 94 * which may induce a lot of work -- pushing cached pages (which acquires 95 * range locks) and syncing out cached atime changes. Third, 96 * zfs_zinactive() may require a new tx, which could deadlock the system 97 * if you were already holding one. This deadlock occurs because the tx 98 * currently being operated on prevents a txg from syncing, which 99 * prevents the new tx from progressing, resulting in a deadlock. If you 100 * must call zrele() within a tx, use zfs_zrele_async(). Note that iput() 101 * is a synonym for zrele(). 102 * 103 * (3) All range locks must be grabbed before calling dmu_tx_assign(), 104 * as they can span dmu_tx_assign() calls. 105 * 106 * (4) If ZPL locks are held, pass TXG_NOWAIT as the second argument to 107 * dmu_tx_assign(). This is critical because we don't want to block 108 * while holding locks. 109 * 110 * If no ZPL locks are held (aside from zfs_enter()), use TXG_WAIT. This 111 * reduces lock contention and CPU usage when we must wait (note that if 112 * throughput is constrained by the storage, nearly every transaction 113 * must wait). 114 * 115 * Note, in particular, that if a lock is sometimes acquired before 116 * the tx assigns, and sometimes after (e.g. z_lock), then failing 117 * to use a non-blocking assign can deadlock the system. The scenario: 118 * 119 * Thread A has grabbed a lock before calling dmu_tx_assign(). 120 * Thread B is in an already-assigned tx, and blocks for this lock. 121 * Thread A calls dmu_tx_assign(TXG_WAIT) and blocks in txg_wait_open() 122 * forever, because the previous txg can't quiesce until B's tx commits. 123 * 124 * If dmu_tx_assign() returns ERESTART and zfsvfs->z_assign is TXG_NOWAIT, 125 * then drop all locks, call dmu_tx_wait(), and try again. On subsequent 126 * calls to dmu_tx_assign(), pass TXG_NOTHROTTLE in addition to TXG_NOWAIT, 127 * to indicate that this operation has already called dmu_tx_wait(). 128 * This will ensure that we don't retry forever, waiting a short bit 129 * each time. 130 * 131 * (5) If the operation succeeded, generate the intent log entry for it 132 * before dropping locks. This ensures that the ordering of events 133 * in the intent log matches the order in which they actually occurred. 134 * During ZIL replay the zfs_log_* functions will update the sequence 135 * number to indicate the zil transaction has replayed. 136 * 137 * (6) At the end of each vnode op, the DMU tx must always commit, 138 * regardless of whether there were any errors. 139 * 140 * (7) After dropping all locks, invoke zil_commit(zilog, foid) 141 * to ensure that synchronous semantics are provided when necessary. 142 * 143 * In general, this is how things should be ordered in each vnode op: 144 * 145 * zfs_enter(zfsvfs); // exit if unmounted 146 * top: 147 * zfs_dirent_lock(&dl, ...) // lock directory entry (may igrab()) 148 * rw_enter(...); // grab any other locks you need 149 * tx = dmu_tx_create(...); // get DMU tx 150 * dmu_tx_hold_*(); // hold each object you might modify 151 * error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT); 152 * if (error) { 153 * rw_exit(...); // drop locks 154 * zfs_dirent_unlock(dl); // unlock directory entry 155 * zrele(...); // release held znodes 156 * if (error == ERESTART) { 157 * waited = B_TRUE; 158 * dmu_tx_wait(tx); 159 * dmu_tx_abort(tx); 160 * goto top; 161 * } 162 * dmu_tx_abort(tx); // abort DMU tx 163 * zfs_exit(zfsvfs); // finished in zfs 164 * return (error); // really out of space 165 * } 166 * error = do_real_work(); // do whatever this VOP does 167 * if (error == 0) 168 * zfs_log_*(...); // on success, make ZIL entry 169 * dmu_tx_commit(tx); // commit DMU tx -- error or not 170 * rw_exit(...); // drop locks 171 * zfs_dirent_unlock(dl); // unlock directory entry 172 * zrele(...); // release held znodes 173 * zil_commit(zilog, foid); // synchronous when necessary 174 * zfs_exit(zfsvfs); // finished in zfs 175 * return (error); // done, report error 176 */ 177 int 178 zfs_open(struct inode *ip, int mode, int flag, cred_t *cr) 179 { 180 (void) cr; 181 znode_t *zp = ITOZ(ip); 182 zfsvfs_t *zfsvfs = ITOZSB(ip); 183 int error; 184 185 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 186 return (error); 187 188 /* Honor ZFS_APPENDONLY file attribute */ 189 if ((mode & FMODE_WRITE) && (zp->z_pflags & ZFS_APPENDONLY) && 190 ((flag & O_APPEND) == 0)) { 191 zfs_exit(zfsvfs, FTAG); 192 return (SET_ERROR(EPERM)); 193 } 194 195 /* Keep a count of the synchronous opens in the znode */ 196 if (flag & O_SYNC) 197 atomic_inc_32(&zp->z_sync_cnt); 198 199 zfs_exit(zfsvfs, FTAG); 200 return (0); 201 } 202 203 int 204 zfs_close(struct inode *ip, int flag, cred_t *cr) 205 { 206 (void) cr; 207 znode_t *zp = ITOZ(ip); 208 zfsvfs_t *zfsvfs = ITOZSB(ip); 209 int error; 210 211 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 212 return (error); 213 214 /* Decrement the synchronous opens in the znode */ 215 if (flag & O_SYNC) 216 atomic_dec_32(&zp->z_sync_cnt); 217 218 zfs_exit(zfsvfs, FTAG); 219 return (0); 220 } 221 222 #if defined(_KERNEL) 223 224 static int zfs_fillpage(struct inode *ip, struct page *pp); 225 226 /* 227 * When a file is memory mapped, we must keep the IO data synchronized 228 * between the DMU cache and the memory mapped pages. Update all mapped 229 * pages with the contents of the coresponding dmu buffer. 230 */ 231 void 232 update_pages(znode_t *zp, int64_t start, int len, objset_t *os) 233 { 234 struct address_space *mp = ZTOI(zp)->i_mapping; 235 int64_t off = start & (PAGE_SIZE - 1); 236 237 for (start &= PAGE_MASK; len > 0; start += PAGE_SIZE) { 238 uint64_t nbytes = MIN(PAGE_SIZE - off, len); 239 240 struct page *pp = find_lock_page(mp, start >> PAGE_SHIFT); 241 if (pp) { 242 if (mapping_writably_mapped(mp)) 243 flush_dcache_page(pp); 244 245 void *pb = kmap(pp); 246 int error = dmu_read(os, zp->z_id, start + off, 247 nbytes, pb + off, DMU_READ_PREFETCH); 248 kunmap(pp); 249 250 if (error) { 251 SetPageError(pp); 252 ClearPageUptodate(pp); 253 } else { 254 ClearPageError(pp); 255 SetPageUptodate(pp); 256 257 if (mapping_writably_mapped(mp)) 258 flush_dcache_page(pp); 259 260 mark_page_accessed(pp); 261 } 262 263 unlock_page(pp); 264 put_page(pp); 265 } 266 267 len -= nbytes; 268 off = 0; 269 } 270 } 271 272 /* 273 * When a file is memory mapped, we must keep the I/O data synchronized 274 * between the DMU cache and the memory mapped pages. Preferentially read 275 * from memory mapped pages, otherwise fallback to reading through the dmu. 276 */ 277 int 278 mappedread(znode_t *zp, int nbytes, zfs_uio_t *uio) 279 { 280 struct inode *ip = ZTOI(zp); 281 struct address_space *mp = ip->i_mapping; 282 int64_t start = uio->uio_loffset; 283 int64_t off = start & (PAGE_SIZE - 1); 284 int len = nbytes; 285 int error = 0; 286 287 for (start &= PAGE_MASK; len > 0; start += PAGE_SIZE) { 288 uint64_t bytes = MIN(PAGE_SIZE - off, len); 289 290 struct page *pp = find_lock_page(mp, start >> PAGE_SHIFT); 291 if (pp) { 292 /* 293 * If filemap_fault() retries there exists a window 294 * where the page will be unlocked and not up to date. 295 * In this case we must try and fill the page. 296 */ 297 if (unlikely(!PageUptodate(pp))) { 298 error = zfs_fillpage(ip, pp); 299 if (error) { 300 unlock_page(pp); 301 put_page(pp); 302 return (error); 303 } 304 } 305 306 ASSERT(PageUptodate(pp) || PageDirty(pp)); 307 308 unlock_page(pp); 309 310 void *pb = kmap(pp); 311 error = zfs_uiomove(pb + off, bytes, UIO_READ, uio); 312 kunmap(pp); 313 314 if (mapping_writably_mapped(mp)) 315 flush_dcache_page(pp); 316 317 mark_page_accessed(pp); 318 put_page(pp); 319 } else { 320 error = dmu_read_uio_dbuf(sa_get_db(zp->z_sa_hdl), 321 uio, bytes); 322 } 323 324 len -= bytes; 325 off = 0; 326 327 if (error) 328 break; 329 } 330 331 return (error); 332 } 333 #endif /* _KERNEL */ 334 335 static unsigned long zfs_delete_blocks = DMU_MAX_DELETEBLKCNT; 336 337 /* 338 * Write the bytes to a file. 339 * 340 * IN: zp - znode of file to be written to 341 * data - bytes to write 342 * len - number of bytes to write 343 * pos - offset to start writing at 344 * 345 * OUT: resid - remaining bytes to write 346 * 347 * RETURN: 0 if success 348 * positive error code if failure. EIO is returned 349 * for a short write when residp isn't provided. 350 * 351 * Timestamps: 352 * zp - ctime|mtime updated if byte count > 0 353 */ 354 int 355 zfs_write_simple(znode_t *zp, const void *data, size_t len, 356 loff_t pos, size_t *residp) 357 { 358 fstrans_cookie_t cookie; 359 int error; 360 361 struct iovec iov; 362 iov.iov_base = (void *)data; 363 iov.iov_len = len; 364 365 zfs_uio_t uio; 366 zfs_uio_iovec_init(&uio, &iov, 1, pos, UIO_SYSSPACE, len, 0); 367 368 cookie = spl_fstrans_mark(); 369 error = zfs_write(zp, &uio, 0, kcred); 370 spl_fstrans_unmark(cookie); 371 372 if (error == 0) { 373 if (residp != NULL) 374 *residp = zfs_uio_resid(&uio); 375 else if (zfs_uio_resid(&uio) != 0) 376 error = SET_ERROR(EIO); 377 } 378 379 return (error); 380 } 381 382 static void 383 zfs_rele_async_task(void *arg) 384 { 385 iput(arg); 386 } 387 388 void 389 zfs_zrele_async(znode_t *zp) 390 { 391 struct inode *ip = ZTOI(zp); 392 objset_t *os = ITOZSB(ip)->z_os; 393 394 ASSERT(atomic_read(&ip->i_count) > 0); 395 ASSERT(os != NULL); 396 397 /* 398 * If decrementing the count would put us at 0, we can't do it inline 399 * here, because that would be synchronous. Instead, dispatch an iput 400 * to run later. 401 * 402 * For more information on the dangers of a synchronous iput, see the 403 * header comment of this file. 404 */ 405 if (!atomic_add_unless(&ip->i_count, -1, 1)) { 406 VERIFY(taskq_dispatch(dsl_pool_zrele_taskq(dmu_objset_pool(os)), 407 zfs_rele_async_task, ip, TQ_SLEEP) != TASKQID_INVALID); 408 } 409 } 410 411 412 /* 413 * Lookup an entry in a directory, or an extended attribute directory. 414 * If it exists, return a held inode reference for it. 415 * 416 * IN: zdp - znode of directory to search. 417 * nm - name of entry to lookup. 418 * flags - LOOKUP_XATTR set if looking for an attribute. 419 * cr - credentials of caller. 420 * direntflags - directory lookup flags 421 * realpnp - returned pathname. 422 * 423 * OUT: zpp - znode of located entry, NULL if not found. 424 * 425 * RETURN: 0 on success, error code on failure. 426 * 427 * Timestamps: 428 * NA 429 */ 430 int 431 zfs_lookup(znode_t *zdp, char *nm, znode_t **zpp, int flags, cred_t *cr, 432 int *direntflags, pathname_t *realpnp) 433 { 434 zfsvfs_t *zfsvfs = ZTOZSB(zdp); 435 int error = 0; 436 437 /* 438 * Fast path lookup, however we must skip DNLC lookup 439 * for case folding or normalizing lookups because the 440 * DNLC code only stores the passed in name. This means 441 * creating 'a' and removing 'A' on a case insensitive 442 * file system would work, but DNLC still thinks 'a' 443 * exists and won't let you create it again on the next 444 * pass through fast path. 445 */ 446 if (!(flags & (LOOKUP_XATTR | FIGNORECASE))) { 447 448 if (!S_ISDIR(ZTOI(zdp)->i_mode)) { 449 return (SET_ERROR(ENOTDIR)); 450 } else if (zdp->z_sa_hdl == NULL) { 451 return (SET_ERROR(EIO)); 452 } 453 454 if (nm[0] == 0 || (nm[0] == '.' && nm[1] == '\0')) { 455 error = zfs_fastaccesschk_execute(zdp, cr); 456 if (!error) { 457 *zpp = zdp; 458 zhold(*zpp); 459 return (0); 460 } 461 return (error); 462 } 463 } 464 465 if ((error = zfs_enter_verify_zp(zfsvfs, zdp, FTAG)) != 0) 466 return (error); 467 468 *zpp = NULL; 469 470 if (flags & LOOKUP_XATTR) { 471 /* 472 * We don't allow recursive attributes.. 473 * Maybe someday we will. 474 */ 475 if (zdp->z_pflags & ZFS_XATTR) { 476 zfs_exit(zfsvfs, FTAG); 477 return (SET_ERROR(EINVAL)); 478 } 479 480 if ((error = zfs_get_xattrdir(zdp, zpp, cr, flags))) { 481 zfs_exit(zfsvfs, FTAG); 482 return (error); 483 } 484 485 /* 486 * Do we have permission to get into attribute directory? 487 */ 488 489 if ((error = zfs_zaccess(*zpp, ACE_EXECUTE, 0, 490 B_TRUE, cr, zfs_init_idmap))) { 491 zrele(*zpp); 492 *zpp = NULL; 493 } 494 495 zfs_exit(zfsvfs, FTAG); 496 return (error); 497 } 498 499 if (!S_ISDIR(ZTOI(zdp)->i_mode)) { 500 zfs_exit(zfsvfs, FTAG); 501 return (SET_ERROR(ENOTDIR)); 502 } 503 504 /* 505 * Check accessibility of directory. 506 */ 507 508 if ((error = zfs_zaccess(zdp, ACE_EXECUTE, 0, B_FALSE, cr, 509 zfs_init_idmap))) { 510 zfs_exit(zfsvfs, FTAG); 511 return (error); 512 } 513 514 if (zfsvfs->z_utf8 && u8_validate(nm, strlen(nm), 515 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 516 zfs_exit(zfsvfs, FTAG); 517 return (SET_ERROR(EILSEQ)); 518 } 519 520 error = zfs_dirlook(zdp, nm, zpp, flags, direntflags, realpnp); 521 if ((error == 0) && (*zpp)) 522 zfs_znode_update_vfs(*zpp); 523 524 zfs_exit(zfsvfs, FTAG); 525 return (error); 526 } 527 528 /* 529 * Attempt to create a new entry in a directory. If the entry 530 * already exists, truncate the file if permissible, else return 531 * an error. Return the ip of the created or trunc'd file. 532 * 533 * IN: dzp - znode of directory to put new file entry in. 534 * name - name of new file entry. 535 * vap - attributes of new file. 536 * excl - flag indicating exclusive or non-exclusive mode. 537 * mode - mode to open file with. 538 * cr - credentials of caller. 539 * flag - file flag. 540 * vsecp - ACL to be set 541 * mnt_ns - user namespace of the mount 542 * 543 * OUT: zpp - znode of created or trunc'd entry. 544 * 545 * RETURN: 0 on success, error code on failure. 546 * 547 * Timestamps: 548 * dzp - ctime|mtime updated if new entry created 549 * zp - ctime|mtime always, atime if new 550 */ 551 int 552 zfs_create(znode_t *dzp, char *name, vattr_t *vap, int excl, 553 int mode, znode_t **zpp, cred_t *cr, int flag, vsecattr_t *vsecp, 554 zidmap_t *mnt_ns) 555 { 556 znode_t *zp; 557 zfsvfs_t *zfsvfs = ZTOZSB(dzp); 558 zilog_t *zilog; 559 objset_t *os; 560 zfs_dirlock_t *dl; 561 dmu_tx_t *tx; 562 int error; 563 uid_t uid; 564 gid_t gid; 565 zfs_acl_ids_t acl_ids; 566 boolean_t fuid_dirtied; 567 boolean_t have_acl = B_FALSE; 568 boolean_t waited = B_FALSE; 569 boolean_t skip_acl = (flag & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 570 571 /* 572 * If we have an ephemeral id, ACL, or XVATTR then 573 * make sure file system is at proper version 574 */ 575 576 gid = crgetgid(cr); 577 uid = crgetuid(cr); 578 579 if (zfsvfs->z_use_fuids == B_FALSE && 580 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 581 return (SET_ERROR(EINVAL)); 582 583 if (name == NULL) 584 return (SET_ERROR(EINVAL)); 585 586 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) 587 return (error); 588 os = zfsvfs->z_os; 589 zilog = zfsvfs->z_log; 590 591 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 592 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 593 zfs_exit(zfsvfs, FTAG); 594 return (SET_ERROR(EILSEQ)); 595 } 596 597 if (vap->va_mask & ATTR_XVATTR) { 598 if ((error = secpolicy_xvattr((xvattr_t *)vap, 599 crgetuid(cr), cr, vap->va_mode)) != 0) { 600 zfs_exit(zfsvfs, FTAG); 601 return (error); 602 } 603 } 604 605 top: 606 *zpp = NULL; 607 if (*name == '\0') { 608 /* 609 * Null component name refers to the directory itself. 610 */ 611 zhold(dzp); 612 zp = dzp; 613 dl = NULL; 614 error = 0; 615 } else { 616 /* possible igrab(zp) */ 617 int zflg = 0; 618 619 if (flag & FIGNORECASE) 620 zflg |= ZCILOOK; 621 622 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 623 NULL, NULL); 624 if (error) { 625 if (have_acl) 626 zfs_acl_ids_free(&acl_ids); 627 if (strcmp(name, "..") == 0) 628 error = SET_ERROR(EISDIR); 629 zfs_exit(zfsvfs, FTAG); 630 return (error); 631 } 632 } 633 634 if (zp == NULL) { 635 uint64_t txtype; 636 uint64_t projid = ZFS_DEFAULT_PROJID; 637 638 /* 639 * Create a new file object and update the directory 640 * to reference it. 641 */ 642 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, skip_acl, cr, 643 mnt_ns))) { 644 if (have_acl) 645 zfs_acl_ids_free(&acl_ids); 646 goto out; 647 } 648 649 /* 650 * We only support the creation of regular files in 651 * extended attribute directories. 652 */ 653 654 if ((dzp->z_pflags & ZFS_XATTR) && !S_ISREG(vap->va_mode)) { 655 if (have_acl) 656 zfs_acl_ids_free(&acl_ids); 657 error = SET_ERROR(EINVAL); 658 goto out; 659 } 660 661 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap, 662 cr, vsecp, &acl_ids, mnt_ns)) != 0) 663 goto out; 664 have_acl = B_TRUE; 665 666 if (S_ISREG(vap->va_mode) || S_ISDIR(vap->va_mode)) 667 projid = zfs_inherit_projid(dzp); 668 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, projid)) { 669 zfs_acl_ids_free(&acl_ids); 670 error = SET_ERROR(EDQUOT); 671 goto out; 672 } 673 674 tx = dmu_tx_create(os); 675 676 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 677 ZFS_SA_BASE_ATTR_SIZE); 678 679 fuid_dirtied = zfsvfs->z_fuid_dirty; 680 if (fuid_dirtied) 681 zfs_fuid_txhold(zfsvfs, tx); 682 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 683 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 684 if (!zfsvfs->z_use_sa && 685 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 686 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 687 0, acl_ids.z_aclp->z_acl_bytes); 688 } 689 690 error = dmu_tx_assign(tx, 691 (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT); 692 if (error) { 693 zfs_dirent_unlock(dl); 694 if (error == ERESTART) { 695 waited = B_TRUE; 696 dmu_tx_wait(tx); 697 dmu_tx_abort(tx); 698 goto top; 699 } 700 zfs_acl_ids_free(&acl_ids); 701 dmu_tx_abort(tx); 702 zfs_exit(zfsvfs, FTAG); 703 return (error); 704 } 705 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 706 707 error = zfs_link_create(dl, zp, tx, ZNEW); 708 if (error != 0) { 709 /* 710 * Since, we failed to add the directory entry for it, 711 * delete the newly created dnode. 712 */ 713 zfs_znode_delete(zp, tx); 714 remove_inode_hash(ZTOI(zp)); 715 zfs_acl_ids_free(&acl_ids); 716 dmu_tx_commit(tx); 717 goto out; 718 } 719 720 if (fuid_dirtied) 721 zfs_fuid_sync(zfsvfs, tx); 722 723 txtype = zfs_log_create_txtype(Z_FILE, vsecp, vap); 724 if (flag & FIGNORECASE) 725 txtype |= TX_CI; 726 zfs_log_create(zilog, tx, txtype, dzp, zp, name, 727 vsecp, acl_ids.z_fuidp, vap); 728 zfs_acl_ids_free(&acl_ids); 729 dmu_tx_commit(tx); 730 } else { 731 int aflags = (flag & O_APPEND) ? V_APPEND : 0; 732 733 if (have_acl) 734 zfs_acl_ids_free(&acl_ids); 735 736 /* 737 * A directory entry already exists for this name. 738 */ 739 /* 740 * Can't truncate an existing file if in exclusive mode. 741 */ 742 if (excl) { 743 error = SET_ERROR(EEXIST); 744 goto out; 745 } 746 /* 747 * Can't open a directory for writing. 748 */ 749 if (S_ISDIR(ZTOI(zp)->i_mode)) { 750 error = SET_ERROR(EISDIR); 751 goto out; 752 } 753 /* 754 * Verify requested access to file. 755 */ 756 if (mode && (error = zfs_zaccess_rwx(zp, mode, aflags, cr, 757 mnt_ns))) { 758 goto out; 759 } 760 761 mutex_enter(&dzp->z_lock); 762 dzp->z_seq++; 763 mutex_exit(&dzp->z_lock); 764 765 /* 766 * Truncate regular files if requested. 767 */ 768 if (S_ISREG(ZTOI(zp)->i_mode) && 769 (vap->va_mask & ATTR_SIZE) && (vap->va_size == 0)) { 770 /* we can't hold any locks when calling zfs_freesp() */ 771 if (dl) { 772 zfs_dirent_unlock(dl); 773 dl = NULL; 774 } 775 error = zfs_freesp(zp, 0, 0, mode, TRUE); 776 } 777 } 778 out: 779 780 if (dl) 781 zfs_dirent_unlock(dl); 782 783 if (error) { 784 if (zp) 785 zrele(zp); 786 } else { 787 zfs_znode_update_vfs(dzp); 788 zfs_znode_update_vfs(zp); 789 *zpp = zp; 790 } 791 792 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 793 zil_commit(zilog, 0); 794 795 zfs_exit(zfsvfs, FTAG); 796 return (error); 797 } 798 799 int 800 zfs_tmpfile(struct inode *dip, vattr_t *vap, int excl, 801 int mode, struct inode **ipp, cred_t *cr, int flag, vsecattr_t *vsecp, 802 zidmap_t *mnt_ns) 803 { 804 (void) excl, (void) mode, (void) flag; 805 znode_t *zp = NULL, *dzp = ITOZ(dip); 806 zfsvfs_t *zfsvfs = ITOZSB(dip); 807 objset_t *os; 808 dmu_tx_t *tx; 809 int error; 810 uid_t uid; 811 gid_t gid; 812 zfs_acl_ids_t acl_ids; 813 uint64_t projid = ZFS_DEFAULT_PROJID; 814 boolean_t fuid_dirtied; 815 boolean_t have_acl = B_FALSE; 816 boolean_t waited = B_FALSE; 817 818 /* 819 * If we have an ephemeral id, ACL, or XVATTR then 820 * make sure file system is at proper version 821 */ 822 823 gid = crgetgid(cr); 824 uid = crgetuid(cr); 825 826 if (zfsvfs->z_use_fuids == B_FALSE && 827 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 828 return (SET_ERROR(EINVAL)); 829 830 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) 831 return (error); 832 os = zfsvfs->z_os; 833 834 if (vap->va_mask & ATTR_XVATTR) { 835 if ((error = secpolicy_xvattr((xvattr_t *)vap, 836 crgetuid(cr), cr, vap->va_mode)) != 0) { 837 zfs_exit(zfsvfs, FTAG); 838 return (error); 839 } 840 } 841 842 top: 843 *ipp = NULL; 844 845 /* 846 * Create a new file object and update the directory 847 * to reference it. 848 */ 849 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr, mnt_ns))) { 850 if (have_acl) 851 zfs_acl_ids_free(&acl_ids); 852 goto out; 853 } 854 855 if (!have_acl && (error = zfs_acl_ids_create(dzp, 0, vap, 856 cr, vsecp, &acl_ids, mnt_ns)) != 0) 857 goto out; 858 have_acl = B_TRUE; 859 860 if (S_ISREG(vap->va_mode) || S_ISDIR(vap->va_mode)) 861 projid = zfs_inherit_projid(dzp); 862 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, projid)) { 863 zfs_acl_ids_free(&acl_ids); 864 error = SET_ERROR(EDQUOT); 865 goto out; 866 } 867 868 tx = dmu_tx_create(os); 869 870 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 871 ZFS_SA_BASE_ATTR_SIZE); 872 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 873 874 fuid_dirtied = zfsvfs->z_fuid_dirty; 875 if (fuid_dirtied) 876 zfs_fuid_txhold(zfsvfs, tx); 877 if (!zfsvfs->z_use_sa && 878 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 879 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 880 0, acl_ids.z_aclp->z_acl_bytes); 881 } 882 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT); 883 if (error) { 884 if (error == ERESTART) { 885 waited = B_TRUE; 886 dmu_tx_wait(tx); 887 dmu_tx_abort(tx); 888 goto top; 889 } 890 zfs_acl_ids_free(&acl_ids); 891 dmu_tx_abort(tx); 892 zfs_exit(zfsvfs, FTAG); 893 return (error); 894 } 895 zfs_mknode(dzp, vap, tx, cr, IS_TMPFILE, &zp, &acl_ids); 896 897 if (fuid_dirtied) 898 zfs_fuid_sync(zfsvfs, tx); 899 900 /* Add to unlinked set */ 901 zp->z_unlinked = B_TRUE; 902 zfs_unlinked_add(zp, tx); 903 zfs_acl_ids_free(&acl_ids); 904 dmu_tx_commit(tx); 905 out: 906 907 if (error) { 908 if (zp) 909 zrele(zp); 910 } else { 911 zfs_znode_update_vfs(dzp); 912 zfs_znode_update_vfs(zp); 913 *ipp = ZTOI(zp); 914 } 915 916 zfs_exit(zfsvfs, FTAG); 917 return (error); 918 } 919 920 /* 921 * Remove an entry from a directory. 922 * 923 * IN: dzp - znode of directory to remove entry from. 924 * name - name of entry to remove. 925 * cr - credentials of caller. 926 * flags - case flags. 927 * 928 * RETURN: 0 if success 929 * error code if failure 930 * 931 * Timestamps: 932 * dzp - ctime|mtime 933 * ip - ctime (if nlink > 0) 934 */ 935 936 static uint64_t null_xattr = 0; 937 938 int 939 zfs_remove(znode_t *dzp, char *name, cred_t *cr, int flags) 940 { 941 znode_t *zp; 942 znode_t *xzp; 943 zfsvfs_t *zfsvfs = ZTOZSB(dzp); 944 zilog_t *zilog; 945 uint64_t acl_obj, xattr_obj; 946 uint64_t xattr_obj_unlinked = 0; 947 uint64_t obj = 0; 948 uint64_t links; 949 zfs_dirlock_t *dl; 950 dmu_tx_t *tx; 951 boolean_t may_delete_now, delete_now = FALSE; 952 boolean_t unlinked, toobig = FALSE; 953 uint64_t txtype; 954 pathname_t *realnmp = NULL; 955 pathname_t realnm; 956 int error; 957 int zflg = ZEXISTS; 958 boolean_t waited = B_FALSE; 959 960 if (name == NULL) 961 return (SET_ERROR(EINVAL)); 962 963 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) 964 return (error); 965 zilog = zfsvfs->z_log; 966 967 if (flags & FIGNORECASE) { 968 zflg |= ZCILOOK; 969 pn_alloc(&realnm); 970 realnmp = &realnm; 971 } 972 973 top: 974 xattr_obj = 0; 975 xzp = NULL; 976 /* 977 * Attempt to lock directory; fail if entry doesn't exist. 978 */ 979 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 980 NULL, realnmp))) { 981 if (realnmp) 982 pn_free(realnmp); 983 zfs_exit(zfsvfs, FTAG); 984 return (error); 985 } 986 987 if ((error = zfs_zaccess_delete(dzp, zp, cr, zfs_init_idmap))) { 988 goto out; 989 } 990 991 /* 992 * Need to use rmdir for removing directories. 993 */ 994 if (S_ISDIR(ZTOI(zp)->i_mode)) { 995 error = SET_ERROR(EPERM); 996 goto out; 997 } 998 999 mutex_enter(&zp->z_lock); 1000 may_delete_now = atomic_read(&ZTOI(zp)->i_count) == 1 && 1001 !zn_has_cached_data(zp, 0, LLONG_MAX); 1002 mutex_exit(&zp->z_lock); 1003 1004 /* 1005 * We may delete the znode now, or we may put it in the unlinked set; 1006 * it depends on whether we're the last link, and on whether there are 1007 * other holds on the inode. So we dmu_tx_hold() the right things to 1008 * allow for either case. 1009 */ 1010 obj = zp->z_id; 1011 tx = dmu_tx_create(zfsvfs->z_os); 1012 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1013 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1014 zfs_sa_upgrade_txholds(tx, zp); 1015 zfs_sa_upgrade_txholds(tx, dzp); 1016 if (may_delete_now) { 1017 toobig = zp->z_size > zp->z_blksz * zfs_delete_blocks; 1018 /* if the file is too big, only hold_free a token amount */ 1019 dmu_tx_hold_free(tx, zp->z_id, 0, 1020 (toobig ? DMU_MAX_ACCESS : DMU_OBJECT_END)); 1021 } 1022 1023 /* are there any extended attributes? */ 1024 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 1025 &xattr_obj, sizeof (xattr_obj)); 1026 if (error == 0 && xattr_obj) { 1027 error = zfs_zget(zfsvfs, xattr_obj, &xzp); 1028 ASSERT0(error); 1029 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 1030 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE); 1031 } 1032 1033 mutex_enter(&zp->z_lock); 1034 if ((acl_obj = zfs_external_acl(zp)) != 0 && may_delete_now) 1035 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 1036 mutex_exit(&zp->z_lock); 1037 1038 /* charge as an update -- would be nice not to charge at all */ 1039 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1040 1041 /* 1042 * Mark this transaction as typically resulting in a net free of space 1043 */ 1044 dmu_tx_mark_netfree(tx); 1045 1046 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT); 1047 if (error) { 1048 zfs_dirent_unlock(dl); 1049 if (error == ERESTART) { 1050 waited = B_TRUE; 1051 dmu_tx_wait(tx); 1052 dmu_tx_abort(tx); 1053 zrele(zp); 1054 if (xzp) 1055 zrele(xzp); 1056 goto top; 1057 } 1058 if (realnmp) 1059 pn_free(realnmp); 1060 dmu_tx_abort(tx); 1061 zrele(zp); 1062 if (xzp) 1063 zrele(xzp); 1064 zfs_exit(zfsvfs, FTAG); 1065 return (error); 1066 } 1067 1068 /* 1069 * Remove the directory entry. 1070 */ 1071 error = zfs_link_destroy(dl, zp, tx, zflg, &unlinked); 1072 1073 if (error) { 1074 dmu_tx_commit(tx); 1075 goto out; 1076 } 1077 1078 if (unlinked) { 1079 /* 1080 * Hold z_lock so that we can make sure that the ACL obj 1081 * hasn't changed. Could have been deleted due to 1082 * zfs_sa_upgrade(). 1083 */ 1084 mutex_enter(&zp->z_lock); 1085 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 1086 &xattr_obj_unlinked, sizeof (xattr_obj_unlinked)); 1087 delete_now = may_delete_now && !toobig && 1088 atomic_read(&ZTOI(zp)->i_count) == 1 && 1089 !zn_has_cached_data(zp, 0, LLONG_MAX) && 1090 xattr_obj == xattr_obj_unlinked && 1091 zfs_external_acl(zp) == acl_obj; 1092 VERIFY_IMPLY(xattr_obj_unlinked, xzp); 1093 } 1094 1095 if (delete_now) { 1096 if (xattr_obj_unlinked) { 1097 ASSERT3U(ZTOI(xzp)->i_nlink, ==, 2); 1098 mutex_enter(&xzp->z_lock); 1099 xzp->z_unlinked = B_TRUE; 1100 clear_nlink(ZTOI(xzp)); 1101 links = 0; 1102 error = sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs), 1103 &links, sizeof (links), tx); 1104 ASSERT3U(error, ==, 0); 1105 mutex_exit(&xzp->z_lock); 1106 zfs_unlinked_add(xzp, tx); 1107 1108 if (zp->z_is_sa) 1109 error = sa_remove(zp->z_sa_hdl, 1110 SA_ZPL_XATTR(zfsvfs), tx); 1111 else 1112 error = sa_update(zp->z_sa_hdl, 1113 SA_ZPL_XATTR(zfsvfs), &null_xattr, 1114 sizeof (uint64_t), tx); 1115 ASSERT0(error); 1116 } 1117 /* 1118 * Add to the unlinked set because a new reference could be 1119 * taken concurrently resulting in a deferred destruction. 1120 */ 1121 zfs_unlinked_add(zp, tx); 1122 mutex_exit(&zp->z_lock); 1123 } else if (unlinked) { 1124 mutex_exit(&zp->z_lock); 1125 zfs_unlinked_add(zp, tx); 1126 } 1127 1128 txtype = TX_REMOVE; 1129 if (flags & FIGNORECASE) 1130 txtype |= TX_CI; 1131 zfs_log_remove(zilog, tx, txtype, dzp, name, obj, unlinked); 1132 1133 dmu_tx_commit(tx); 1134 out: 1135 if (realnmp) 1136 pn_free(realnmp); 1137 1138 zfs_dirent_unlock(dl); 1139 zfs_znode_update_vfs(dzp); 1140 zfs_znode_update_vfs(zp); 1141 1142 if (delete_now) 1143 zrele(zp); 1144 else 1145 zfs_zrele_async(zp); 1146 1147 if (xzp) { 1148 zfs_znode_update_vfs(xzp); 1149 zfs_zrele_async(xzp); 1150 } 1151 1152 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1153 zil_commit(zilog, 0); 1154 1155 zfs_exit(zfsvfs, FTAG); 1156 return (error); 1157 } 1158 1159 /* 1160 * Create a new directory and insert it into dzp using the name 1161 * provided. Return a pointer to the inserted directory. 1162 * 1163 * IN: dzp - znode of directory to add subdir to. 1164 * dirname - name of new directory. 1165 * vap - attributes of new directory. 1166 * cr - credentials of caller. 1167 * flags - case flags. 1168 * vsecp - ACL to be set 1169 * mnt_ns - user namespace of the mount 1170 * 1171 * OUT: zpp - znode of created directory. 1172 * 1173 * RETURN: 0 if success 1174 * error code if failure 1175 * 1176 * Timestamps: 1177 * dzp - ctime|mtime updated 1178 * zpp - ctime|mtime|atime updated 1179 */ 1180 int 1181 zfs_mkdir(znode_t *dzp, char *dirname, vattr_t *vap, znode_t **zpp, 1182 cred_t *cr, int flags, vsecattr_t *vsecp, zidmap_t *mnt_ns) 1183 { 1184 znode_t *zp; 1185 zfsvfs_t *zfsvfs = ZTOZSB(dzp); 1186 zilog_t *zilog; 1187 zfs_dirlock_t *dl; 1188 uint64_t txtype; 1189 dmu_tx_t *tx; 1190 int error; 1191 int zf = ZNEW; 1192 uid_t uid; 1193 gid_t gid = crgetgid(cr); 1194 zfs_acl_ids_t acl_ids; 1195 boolean_t fuid_dirtied; 1196 boolean_t waited = B_FALSE; 1197 1198 ASSERT(S_ISDIR(vap->va_mode)); 1199 1200 /* 1201 * If we have an ephemeral id, ACL, or XVATTR then 1202 * make sure file system is at proper version 1203 */ 1204 1205 uid = crgetuid(cr); 1206 if (zfsvfs->z_use_fuids == B_FALSE && 1207 (vsecp || IS_EPHEMERAL(uid) || IS_EPHEMERAL(gid))) 1208 return (SET_ERROR(EINVAL)); 1209 1210 if (dirname == NULL) 1211 return (SET_ERROR(EINVAL)); 1212 1213 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) 1214 return (error); 1215 zilog = zfsvfs->z_log; 1216 1217 if (dzp->z_pflags & ZFS_XATTR) { 1218 zfs_exit(zfsvfs, FTAG); 1219 return (SET_ERROR(EINVAL)); 1220 } 1221 1222 if (zfsvfs->z_utf8 && u8_validate(dirname, 1223 strlen(dirname), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 1224 zfs_exit(zfsvfs, FTAG); 1225 return (SET_ERROR(EILSEQ)); 1226 } 1227 if (flags & FIGNORECASE) 1228 zf |= ZCILOOK; 1229 1230 if (vap->va_mask & ATTR_XVATTR) { 1231 if ((error = secpolicy_xvattr((xvattr_t *)vap, 1232 crgetuid(cr), cr, vap->va_mode)) != 0) { 1233 zfs_exit(zfsvfs, FTAG); 1234 return (error); 1235 } 1236 } 1237 1238 if ((error = zfs_acl_ids_create(dzp, 0, vap, cr, 1239 vsecp, &acl_ids, mnt_ns)) != 0) { 1240 zfs_exit(zfsvfs, FTAG); 1241 return (error); 1242 } 1243 /* 1244 * First make sure the new directory doesn't exist. 1245 * 1246 * Existence is checked first to make sure we don't return 1247 * EACCES instead of EEXIST which can cause some applications 1248 * to fail. 1249 */ 1250 top: 1251 *zpp = NULL; 1252 1253 if ((error = zfs_dirent_lock(&dl, dzp, dirname, &zp, zf, 1254 NULL, NULL))) { 1255 zfs_acl_ids_free(&acl_ids); 1256 zfs_exit(zfsvfs, FTAG); 1257 return (error); 1258 } 1259 1260 if ((error = zfs_zaccess(dzp, ACE_ADD_SUBDIRECTORY, 0, B_FALSE, cr, 1261 mnt_ns))) { 1262 zfs_acl_ids_free(&acl_ids); 1263 zfs_dirent_unlock(dl); 1264 zfs_exit(zfsvfs, FTAG); 1265 return (error); 1266 } 1267 1268 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, zfs_inherit_projid(dzp))) { 1269 zfs_acl_ids_free(&acl_ids); 1270 zfs_dirent_unlock(dl); 1271 zfs_exit(zfsvfs, FTAG); 1272 return (SET_ERROR(EDQUOT)); 1273 } 1274 1275 /* 1276 * Add a new entry to the directory. 1277 */ 1278 tx = dmu_tx_create(zfsvfs->z_os); 1279 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, dirname); 1280 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 1281 fuid_dirtied = zfsvfs->z_fuid_dirty; 1282 if (fuid_dirtied) 1283 zfs_fuid_txhold(zfsvfs, tx); 1284 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 1285 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 1286 acl_ids.z_aclp->z_acl_bytes); 1287 } 1288 1289 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 1290 ZFS_SA_BASE_ATTR_SIZE); 1291 1292 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT); 1293 if (error) { 1294 zfs_dirent_unlock(dl); 1295 if (error == ERESTART) { 1296 waited = B_TRUE; 1297 dmu_tx_wait(tx); 1298 dmu_tx_abort(tx); 1299 goto top; 1300 } 1301 zfs_acl_ids_free(&acl_ids); 1302 dmu_tx_abort(tx); 1303 zfs_exit(zfsvfs, FTAG); 1304 return (error); 1305 } 1306 1307 /* 1308 * Create new node. 1309 */ 1310 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 1311 1312 /* 1313 * Now put new name in parent dir. 1314 */ 1315 error = zfs_link_create(dl, zp, tx, ZNEW); 1316 if (error != 0) { 1317 zfs_znode_delete(zp, tx); 1318 remove_inode_hash(ZTOI(zp)); 1319 goto out; 1320 } 1321 1322 if (fuid_dirtied) 1323 zfs_fuid_sync(zfsvfs, tx); 1324 1325 *zpp = zp; 1326 1327 txtype = zfs_log_create_txtype(Z_DIR, vsecp, vap); 1328 if (flags & FIGNORECASE) 1329 txtype |= TX_CI; 1330 zfs_log_create(zilog, tx, txtype, dzp, zp, dirname, vsecp, 1331 acl_ids.z_fuidp, vap); 1332 1333 out: 1334 zfs_acl_ids_free(&acl_ids); 1335 1336 dmu_tx_commit(tx); 1337 1338 zfs_dirent_unlock(dl); 1339 1340 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1341 zil_commit(zilog, 0); 1342 1343 if (error != 0) { 1344 zrele(zp); 1345 } else { 1346 zfs_znode_update_vfs(dzp); 1347 zfs_znode_update_vfs(zp); 1348 } 1349 zfs_exit(zfsvfs, FTAG); 1350 return (error); 1351 } 1352 1353 /* 1354 * Remove a directory subdir entry. If the current working 1355 * directory is the same as the subdir to be removed, the 1356 * remove will fail. 1357 * 1358 * IN: dzp - znode of directory to remove from. 1359 * name - name of directory to be removed. 1360 * cwd - inode of current working directory. 1361 * cr - credentials of caller. 1362 * flags - case flags 1363 * 1364 * RETURN: 0 on success, error code on failure. 1365 * 1366 * Timestamps: 1367 * dzp - ctime|mtime updated 1368 */ 1369 int 1370 zfs_rmdir(znode_t *dzp, char *name, znode_t *cwd, cred_t *cr, 1371 int flags) 1372 { 1373 znode_t *zp; 1374 zfsvfs_t *zfsvfs = ZTOZSB(dzp); 1375 zilog_t *zilog; 1376 zfs_dirlock_t *dl; 1377 dmu_tx_t *tx; 1378 int error; 1379 int zflg = ZEXISTS; 1380 boolean_t waited = B_FALSE; 1381 1382 if (name == NULL) 1383 return (SET_ERROR(EINVAL)); 1384 1385 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) 1386 return (error); 1387 zilog = zfsvfs->z_log; 1388 1389 if (flags & FIGNORECASE) 1390 zflg |= ZCILOOK; 1391 top: 1392 zp = NULL; 1393 1394 /* 1395 * Attempt to lock directory; fail if entry doesn't exist. 1396 */ 1397 if ((error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, 1398 NULL, NULL))) { 1399 zfs_exit(zfsvfs, FTAG); 1400 return (error); 1401 } 1402 1403 if ((error = zfs_zaccess_delete(dzp, zp, cr, zfs_init_idmap))) { 1404 goto out; 1405 } 1406 1407 if (!S_ISDIR(ZTOI(zp)->i_mode)) { 1408 error = SET_ERROR(ENOTDIR); 1409 goto out; 1410 } 1411 1412 if (zp == cwd) { 1413 error = SET_ERROR(EINVAL); 1414 goto out; 1415 } 1416 1417 /* 1418 * Grab a lock on the directory to make sure that no one is 1419 * trying to add (or lookup) entries while we are removing it. 1420 */ 1421 rw_enter(&zp->z_name_lock, RW_WRITER); 1422 1423 /* 1424 * Grab a lock on the parent pointer to make sure we play well 1425 * with the treewalk and directory rename code. 1426 */ 1427 rw_enter(&zp->z_parent_lock, RW_WRITER); 1428 1429 tx = dmu_tx_create(zfsvfs->z_os); 1430 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, name); 1431 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1432 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 1433 zfs_sa_upgrade_txholds(tx, zp); 1434 zfs_sa_upgrade_txholds(tx, dzp); 1435 dmu_tx_mark_netfree(tx); 1436 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT); 1437 if (error) { 1438 rw_exit(&zp->z_parent_lock); 1439 rw_exit(&zp->z_name_lock); 1440 zfs_dirent_unlock(dl); 1441 if (error == ERESTART) { 1442 waited = B_TRUE; 1443 dmu_tx_wait(tx); 1444 dmu_tx_abort(tx); 1445 zrele(zp); 1446 goto top; 1447 } 1448 dmu_tx_abort(tx); 1449 zrele(zp); 1450 zfs_exit(zfsvfs, FTAG); 1451 return (error); 1452 } 1453 1454 error = zfs_link_destroy(dl, zp, tx, zflg, NULL); 1455 1456 if (error == 0) { 1457 uint64_t txtype = TX_RMDIR; 1458 if (flags & FIGNORECASE) 1459 txtype |= TX_CI; 1460 zfs_log_remove(zilog, tx, txtype, dzp, name, ZFS_NO_OBJECT, 1461 B_FALSE); 1462 } 1463 1464 dmu_tx_commit(tx); 1465 1466 rw_exit(&zp->z_parent_lock); 1467 rw_exit(&zp->z_name_lock); 1468 out: 1469 zfs_dirent_unlock(dl); 1470 1471 zfs_znode_update_vfs(dzp); 1472 zfs_znode_update_vfs(zp); 1473 zrele(zp); 1474 1475 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 1476 zil_commit(zilog, 0); 1477 1478 zfs_exit(zfsvfs, FTAG); 1479 return (error); 1480 } 1481 1482 /* 1483 * Read directory entries from the given directory cursor position and emit 1484 * name and position for each entry. 1485 * 1486 * IN: ip - inode of directory to read. 1487 * ctx - directory entry context. 1488 * cr - credentials of caller. 1489 * 1490 * RETURN: 0 if success 1491 * error code if failure 1492 * 1493 * Timestamps: 1494 * ip - atime updated 1495 * 1496 * Note that the low 4 bits of the cookie returned by zap is always zero. 1497 * This allows us to use the low range for "special" directory entries: 1498 * We use 0 for '.', and 1 for '..'. If this is the root of the filesystem, 1499 * we use the offset 2 for the '.zfs' directory. 1500 */ 1501 int 1502 zfs_readdir(struct inode *ip, zpl_dir_context_t *ctx, cred_t *cr) 1503 { 1504 (void) cr; 1505 znode_t *zp = ITOZ(ip); 1506 zfsvfs_t *zfsvfs = ITOZSB(ip); 1507 objset_t *os; 1508 zap_cursor_t zc; 1509 zap_attribute_t zap; 1510 int error; 1511 uint8_t prefetch; 1512 uint8_t type; 1513 int done = 0; 1514 uint64_t parent; 1515 uint64_t offset; /* must be unsigned; checks for < 1 */ 1516 1517 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 1518 return (error); 1519 1520 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 1521 &parent, sizeof (parent))) != 0) 1522 goto out; 1523 1524 /* 1525 * Quit if directory has been removed (posix) 1526 */ 1527 if (zp->z_unlinked) 1528 goto out; 1529 1530 error = 0; 1531 os = zfsvfs->z_os; 1532 offset = ctx->pos; 1533 prefetch = zp->z_zn_prefetch; 1534 1535 /* 1536 * Initialize the iterator cursor. 1537 */ 1538 if (offset <= 3) { 1539 /* 1540 * Start iteration from the beginning of the directory. 1541 */ 1542 zap_cursor_init(&zc, os, zp->z_id); 1543 } else { 1544 /* 1545 * The offset is a serialized cursor. 1546 */ 1547 zap_cursor_init_serialized(&zc, os, zp->z_id, offset); 1548 } 1549 1550 /* 1551 * Transform to file-system independent format 1552 */ 1553 while (!done) { 1554 uint64_t objnum; 1555 /* 1556 * Special case `.', `..', and `.zfs'. 1557 */ 1558 if (offset == 0) { 1559 (void) strcpy(zap.za_name, "."); 1560 zap.za_normalization_conflict = 0; 1561 objnum = zp->z_id; 1562 type = DT_DIR; 1563 } else if (offset == 1) { 1564 (void) strcpy(zap.za_name, ".."); 1565 zap.za_normalization_conflict = 0; 1566 objnum = parent; 1567 type = DT_DIR; 1568 } else if (offset == 2 && zfs_show_ctldir(zp)) { 1569 (void) strcpy(zap.za_name, ZFS_CTLDIR_NAME); 1570 zap.za_normalization_conflict = 0; 1571 objnum = ZFSCTL_INO_ROOT; 1572 type = DT_DIR; 1573 } else { 1574 /* 1575 * Grab next entry. 1576 */ 1577 if ((error = zap_cursor_retrieve(&zc, &zap))) { 1578 if (error == ENOENT) 1579 break; 1580 else 1581 goto update; 1582 } 1583 1584 /* 1585 * Allow multiple entries provided the first entry is 1586 * the object id. Non-zpl consumers may safely make 1587 * use of the additional space. 1588 * 1589 * XXX: This should be a feature flag for compatibility 1590 */ 1591 if (zap.za_integer_length != 8 || 1592 zap.za_num_integers == 0) { 1593 cmn_err(CE_WARN, "zap_readdir: bad directory " 1594 "entry, obj = %lld, offset = %lld, " 1595 "length = %d, num = %lld\n", 1596 (u_longlong_t)zp->z_id, 1597 (u_longlong_t)offset, 1598 zap.za_integer_length, 1599 (u_longlong_t)zap.za_num_integers); 1600 error = SET_ERROR(ENXIO); 1601 goto update; 1602 } 1603 1604 objnum = ZFS_DIRENT_OBJ(zap.za_first_integer); 1605 type = ZFS_DIRENT_TYPE(zap.za_first_integer); 1606 } 1607 1608 done = !zpl_dir_emit(ctx, zap.za_name, strlen(zap.za_name), 1609 objnum, type); 1610 if (done) 1611 break; 1612 1613 /* Prefetch znode */ 1614 if (prefetch) { 1615 dmu_prefetch(os, objnum, 0, 0, 0, 1616 ZIO_PRIORITY_SYNC_READ); 1617 } 1618 1619 /* 1620 * Move to the next entry, fill in the previous offset. 1621 */ 1622 if (offset > 2 || (offset == 2 && !zfs_show_ctldir(zp))) { 1623 zap_cursor_advance(&zc); 1624 offset = zap_cursor_serialize(&zc); 1625 } else { 1626 offset += 1; 1627 } 1628 ctx->pos = offset; 1629 } 1630 zp->z_zn_prefetch = B_FALSE; /* a lookup will re-enable pre-fetching */ 1631 1632 update: 1633 zap_cursor_fini(&zc); 1634 if (error == ENOENT) 1635 error = 0; 1636 out: 1637 zfs_exit(zfsvfs, FTAG); 1638 1639 return (error); 1640 } 1641 1642 /* 1643 * Get the basic file attributes and place them in the provided kstat 1644 * structure. The inode is assumed to be the authoritative source 1645 * for most of the attributes. However, the znode currently has the 1646 * authoritative atime, blksize, and block count. 1647 * 1648 * IN: ip - inode of file. 1649 * 1650 * OUT: sp - kstat values. 1651 * 1652 * RETURN: 0 (always succeeds) 1653 */ 1654 int 1655 zfs_getattr_fast(zidmap_t *user_ns, struct inode *ip, struct kstat *sp) 1656 { 1657 znode_t *zp = ITOZ(ip); 1658 zfsvfs_t *zfsvfs = ITOZSB(ip); 1659 uint32_t blksize; 1660 u_longlong_t nblocks; 1661 int error; 1662 1663 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 1664 return (error); 1665 1666 mutex_enter(&zp->z_lock); 1667 1668 zpl_generic_fillattr(user_ns, ip, sp); 1669 /* 1670 * +1 link count for root inode with visible '.zfs' directory. 1671 */ 1672 if ((zp->z_id == zfsvfs->z_root) && zfs_show_ctldir(zp)) 1673 if (sp->nlink < ZFS_LINK_MAX) 1674 sp->nlink++; 1675 1676 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks); 1677 sp->blksize = blksize; 1678 sp->blocks = nblocks; 1679 1680 if (unlikely(zp->z_blksz == 0)) { 1681 /* 1682 * Block size hasn't been set; suggest maximal I/O transfers. 1683 */ 1684 sp->blksize = zfsvfs->z_max_blksz; 1685 } 1686 1687 mutex_exit(&zp->z_lock); 1688 1689 /* 1690 * Required to prevent NFS client from detecting different inode 1691 * numbers of snapshot root dentry before and after snapshot mount. 1692 */ 1693 if (zfsvfs->z_issnap) { 1694 if (ip->i_sb->s_root->d_inode == ip) 1695 sp->ino = ZFSCTL_INO_SNAPDIRS - 1696 dmu_objset_id(zfsvfs->z_os); 1697 } 1698 1699 zfs_exit(zfsvfs, FTAG); 1700 1701 return (0); 1702 } 1703 1704 /* 1705 * For the operation of changing file's user/group/project, we need to 1706 * handle not only the main object that is assigned to the file directly, 1707 * but also the ones that are used by the file via hidden xattr directory. 1708 * 1709 * Because the xattr directory may contains many EA entries, as to it may 1710 * be impossible to change all of them via the transaction of changing the 1711 * main object's user/group/project attributes. Then we have to change them 1712 * via other multiple independent transactions one by one. It may be not good 1713 * solution, but we have no better idea yet. 1714 */ 1715 static int 1716 zfs_setattr_dir(znode_t *dzp) 1717 { 1718 struct inode *dxip = ZTOI(dzp); 1719 struct inode *xip = NULL; 1720 zfsvfs_t *zfsvfs = ZTOZSB(dzp); 1721 objset_t *os = zfsvfs->z_os; 1722 zap_cursor_t zc; 1723 zap_attribute_t zap; 1724 zfs_dirlock_t *dl; 1725 znode_t *zp = NULL; 1726 dmu_tx_t *tx = NULL; 1727 uint64_t uid, gid; 1728 sa_bulk_attr_t bulk[4]; 1729 int count; 1730 int err; 1731 1732 zap_cursor_init(&zc, os, dzp->z_id); 1733 while ((err = zap_cursor_retrieve(&zc, &zap)) == 0) { 1734 count = 0; 1735 if (zap.za_integer_length != 8 || zap.za_num_integers != 1) { 1736 err = ENXIO; 1737 break; 1738 } 1739 1740 err = zfs_dirent_lock(&dl, dzp, (char *)zap.za_name, &zp, 1741 ZEXISTS, NULL, NULL); 1742 if (err == ENOENT) 1743 goto next; 1744 if (err) 1745 break; 1746 1747 xip = ZTOI(zp); 1748 if (KUID_TO_SUID(xip->i_uid) == KUID_TO_SUID(dxip->i_uid) && 1749 KGID_TO_SGID(xip->i_gid) == KGID_TO_SGID(dxip->i_gid) && 1750 zp->z_projid == dzp->z_projid) 1751 goto next; 1752 1753 tx = dmu_tx_create(os); 1754 if (!(zp->z_pflags & ZFS_PROJID)) 1755 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 1756 else 1757 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1758 1759 err = dmu_tx_assign(tx, TXG_WAIT); 1760 if (err) 1761 break; 1762 1763 mutex_enter(&dzp->z_lock); 1764 1765 if (KUID_TO_SUID(xip->i_uid) != KUID_TO_SUID(dxip->i_uid)) { 1766 xip->i_uid = dxip->i_uid; 1767 uid = zfs_uid_read(dxip); 1768 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 1769 &uid, sizeof (uid)); 1770 } 1771 1772 if (KGID_TO_SGID(xip->i_gid) != KGID_TO_SGID(dxip->i_gid)) { 1773 xip->i_gid = dxip->i_gid; 1774 gid = zfs_gid_read(dxip); 1775 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, 1776 &gid, sizeof (gid)); 1777 } 1778 1779 if (zp->z_projid != dzp->z_projid) { 1780 if (!(zp->z_pflags & ZFS_PROJID)) { 1781 zp->z_pflags |= ZFS_PROJID; 1782 SA_ADD_BULK_ATTR(bulk, count, 1783 SA_ZPL_FLAGS(zfsvfs), NULL, &zp->z_pflags, 1784 sizeof (zp->z_pflags)); 1785 } 1786 1787 zp->z_projid = dzp->z_projid; 1788 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PROJID(zfsvfs), 1789 NULL, &zp->z_projid, sizeof (zp->z_projid)); 1790 } 1791 1792 mutex_exit(&dzp->z_lock); 1793 1794 if (likely(count > 0)) { 1795 err = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1796 dmu_tx_commit(tx); 1797 } else { 1798 dmu_tx_abort(tx); 1799 } 1800 tx = NULL; 1801 if (err != 0 && err != ENOENT) 1802 break; 1803 1804 next: 1805 if (zp) { 1806 zrele(zp); 1807 zp = NULL; 1808 zfs_dirent_unlock(dl); 1809 } 1810 zap_cursor_advance(&zc); 1811 } 1812 1813 if (tx) 1814 dmu_tx_abort(tx); 1815 if (zp) { 1816 zrele(zp); 1817 zfs_dirent_unlock(dl); 1818 } 1819 zap_cursor_fini(&zc); 1820 1821 return (err == ENOENT ? 0 : err); 1822 } 1823 1824 /* 1825 * Set the file attributes to the values contained in the 1826 * vattr structure. 1827 * 1828 * IN: zp - znode of file to be modified. 1829 * vap - new attribute values. 1830 * If ATTR_XVATTR set, then optional attrs are being set 1831 * flags - ATTR_UTIME set if non-default time values provided. 1832 * - ATTR_NOACLCHECK (CIFS context only). 1833 * cr - credentials of caller. 1834 * mnt_ns - user namespace of the mount 1835 * 1836 * RETURN: 0 if success 1837 * error code if failure 1838 * 1839 * Timestamps: 1840 * ip - ctime updated, mtime updated if size changed. 1841 */ 1842 int 1843 zfs_setattr(znode_t *zp, vattr_t *vap, int flags, cred_t *cr, zidmap_t *mnt_ns) 1844 { 1845 struct inode *ip; 1846 zfsvfs_t *zfsvfs = ZTOZSB(zp); 1847 objset_t *os = zfsvfs->z_os; 1848 zilog_t *zilog; 1849 dmu_tx_t *tx; 1850 vattr_t oldva; 1851 xvattr_t *tmpxvattr; 1852 uint_t mask = vap->va_mask; 1853 uint_t saved_mask = 0; 1854 int trim_mask = 0; 1855 uint64_t new_mode; 1856 uint64_t new_kuid = 0, new_kgid = 0, new_uid, new_gid; 1857 uint64_t xattr_obj; 1858 uint64_t mtime[2], ctime[2], atime[2]; 1859 uint64_t projid = ZFS_INVALID_PROJID; 1860 znode_t *attrzp; 1861 int need_policy = FALSE; 1862 int err, err2 = 0; 1863 zfs_fuid_info_t *fuidp = NULL; 1864 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 1865 xoptattr_t *xoap; 1866 zfs_acl_t *aclp; 1867 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 1868 boolean_t fuid_dirtied = B_FALSE; 1869 boolean_t handle_eadir = B_FALSE; 1870 sa_bulk_attr_t *bulk, *xattr_bulk; 1871 int count = 0, xattr_count = 0, bulks = 8; 1872 1873 if (mask == 0) 1874 return (0); 1875 1876 if ((err = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 1877 return (err); 1878 ip = ZTOI(zp); 1879 1880 /* 1881 * If this is a xvattr_t, then get a pointer to the structure of 1882 * optional attributes. If this is NULL, then we have a vattr_t. 1883 */ 1884 xoap = xva_getxoptattr(xvap); 1885 if (xoap != NULL && (mask & ATTR_XVATTR)) { 1886 if (XVA_ISSET_REQ(xvap, XAT_PROJID)) { 1887 if (!dmu_objset_projectquota_enabled(os) || 1888 (!S_ISREG(ip->i_mode) && !S_ISDIR(ip->i_mode))) { 1889 zfs_exit(zfsvfs, FTAG); 1890 return (SET_ERROR(ENOTSUP)); 1891 } 1892 1893 projid = xoap->xoa_projid; 1894 if (unlikely(projid == ZFS_INVALID_PROJID)) { 1895 zfs_exit(zfsvfs, FTAG); 1896 return (SET_ERROR(EINVAL)); 1897 } 1898 1899 if (projid == zp->z_projid && zp->z_pflags & ZFS_PROJID) 1900 projid = ZFS_INVALID_PROJID; 1901 else 1902 need_policy = TRUE; 1903 } 1904 1905 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT) && 1906 (xoap->xoa_projinherit != 1907 ((zp->z_pflags & ZFS_PROJINHERIT) != 0)) && 1908 (!dmu_objset_projectquota_enabled(os) || 1909 (!S_ISREG(ip->i_mode) && !S_ISDIR(ip->i_mode)))) { 1910 zfs_exit(zfsvfs, FTAG); 1911 return (SET_ERROR(ENOTSUP)); 1912 } 1913 } 1914 1915 zilog = zfsvfs->z_log; 1916 1917 /* 1918 * Make sure that if we have ephemeral uid/gid or xvattr specified 1919 * that file system is at proper version level 1920 */ 1921 1922 if (zfsvfs->z_use_fuids == B_FALSE && 1923 (((mask & ATTR_UID) && IS_EPHEMERAL(vap->va_uid)) || 1924 ((mask & ATTR_GID) && IS_EPHEMERAL(vap->va_gid)) || 1925 (mask & ATTR_XVATTR))) { 1926 zfs_exit(zfsvfs, FTAG); 1927 return (SET_ERROR(EINVAL)); 1928 } 1929 1930 if (mask & ATTR_SIZE && S_ISDIR(ip->i_mode)) { 1931 zfs_exit(zfsvfs, FTAG); 1932 return (SET_ERROR(EISDIR)); 1933 } 1934 1935 if (mask & ATTR_SIZE && !S_ISREG(ip->i_mode) && !S_ISFIFO(ip->i_mode)) { 1936 zfs_exit(zfsvfs, FTAG); 1937 return (SET_ERROR(EINVAL)); 1938 } 1939 1940 tmpxvattr = kmem_alloc(sizeof (xvattr_t), KM_SLEEP); 1941 xva_init(tmpxvattr); 1942 1943 bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * bulks, KM_SLEEP); 1944 xattr_bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * bulks, KM_SLEEP); 1945 1946 /* 1947 * Immutable files can only alter immutable bit and atime 1948 */ 1949 if ((zp->z_pflags & ZFS_IMMUTABLE) && 1950 ((mask & (ATTR_SIZE|ATTR_UID|ATTR_GID|ATTR_MTIME|ATTR_MODE)) || 1951 ((mask & ATTR_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) { 1952 err = SET_ERROR(EPERM); 1953 goto out3; 1954 } 1955 1956 if ((mask & ATTR_SIZE) && (zp->z_pflags & ZFS_READONLY)) { 1957 err = SET_ERROR(EPERM); 1958 goto out3; 1959 } 1960 1961 /* 1962 * Verify timestamps doesn't overflow 32 bits. 1963 * ZFS can handle large timestamps, but 32bit syscalls can't 1964 * handle times greater than 2039. This check should be removed 1965 * once large timestamps are fully supported. 1966 */ 1967 if (mask & (ATTR_ATIME | ATTR_MTIME)) { 1968 if (((mask & ATTR_ATIME) && 1969 TIMESPEC_OVERFLOW(&vap->va_atime)) || 1970 ((mask & ATTR_MTIME) && 1971 TIMESPEC_OVERFLOW(&vap->va_mtime))) { 1972 err = SET_ERROR(EOVERFLOW); 1973 goto out3; 1974 } 1975 } 1976 1977 top: 1978 attrzp = NULL; 1979 aclp = NULL; 1980 1981 /* Can this be moved to before the top label? */ 1982 if (zfs_is_readonly(zfsvfs)) { 1983 err = SET_ERROR(EROFS); 1984 goto out3; 1985 } 1986 1987 /* 1988 * First validate permissions 1989 */ 1990 1991 if (mask & ATTR_SIZE) { 1992 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr, 1993 mnt_ns); 1994 if (err) 1995 goto out3; 1996 1997 /* 1998 * XXX - Note, we are not providing any open 1999 * mode flags here (like FNDELAY), so we may 2000 * block if there are locks present... this 2001 * should be addressed in openat(). 2002 */ 2003 /* XXX - would it be OK to generate a log record here? */ 2004 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 2005 if (err) 2006 goto out3; 2007 } 2008 2009 if (mask & (ATTR_ATIME|ATTR_MTIME) || 2010 ((mask & ATTR_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) || 2011 XVA_ISSET_REQ(xvap, XAT_READONLY) || 2012 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) || 2013 XVA_ISSET_REQ(xvap, XAT_OFFLINE) || 2014 XVA_ISSET_REQ(xvap, XAT_SPARSE) || 2015 XVA_ISSET_REQ(xvap, XAT_CREATETIME) || 2016 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) { 2017 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0, 2018 skipaclchk, cr, mnt_ns); 2019 } 2020 2021 if (mask & (ATTR_UID|ATTR_GID)) { 2022 int idmask = (mask & (ATTR_UID|ATTR_GID)); 2023 int take_owner; 2024 int take_group; 2025 uid_t uid; 2026 gid_t gid; 2027 2028 /* 2029 * NOTE: even if a new mode is being set, 2030 * we may clear S_ISUID/S_ISGID bits. 2031 */ 2032 2033 if (!(mask & ATTR_MODE)) 2034 vap->va_mode = zp->z_mode; 2035 2036 /* 2037 * Take ownership or chgrp to group we are a member of 2038 */ 2039 2040 uid = zfs_uid_to_vfsuid(mnt_ns, zfs_i_user_ns(ip), 2041 vap->va_uid); 2042 gid = zfs_gid_to_vfsgid(mnt_ns, zfs_i_user_ns(ip), 2043 vap->va_gid); 2044 take_owner = (mask & ATTR_UID) && (uid == crgetuid(cr)); 2045 take_group = (mask & ATTR_GID) && 2046 zfs_groupmember(zfsvfs, gid, cr); 2047 2048 /* 2049 * If both ATTR_UID and ATTR_GID are set then take_owner and 2050 * take_group must both be set in order to allow taking 2051 * ownership. 2052 * 2053 * Otherwise, send the check through secpolicy_vnode_setattr() 2054 * 2055 */ 2056 2057 if (((idmask == (ATTR_UID|ATTR_GID)) && 2058 take_owner && take_group) || 2059 ((idmask == ATTR_UID) && take_owner) || 2060 ((idmask == ATTR_GID) && take_group)) { 2061 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0, 2062 skipaclchk, cr, mnt_ns) == 0) { 2063 /* 2064 * Remove setuid/setgid for non-privileged users 2065 */ 2066 (void) secpolicy_setid_clear(vap, cr); 2067 trim_mask = (mask & (ATTR_UID|ATTR_GID)); 2068 } else { 2069 need_policy = TRUE; 2070 } 2071 } else { 2072 need_policy = TRUE; 2073 } 2074 } 2075 2076 mutex_enter(&zp->z_lock); 2077 oldva.va_mode = zp->z_mode; 2078 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid); 2079 if (mask & ATTR_XVATTR) { 2080 /* 2081 * Update xvattr mask to include only those attributes 2082 * that are actually changing. 2083 * 2084 * the bits will be restored prior to actually setting 2085 * the attributes so the caller thinks they were set. 2086 */ 2087 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2088 if (xoap->xoa_appendonly != 2089 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) { 2090 need_policy = TRUE; 2091 } else { 2092 XVA_CLR_REQ(xvap, XAT_APPENDONLY); 2093 XVA_SET_REQ(tmpxvattr, XAT_APPENDONLY); 2094 } 2095 } 2096 2097 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) { 2098 if (xoap->xoa_projinherit != 2099 ((zp->z_pflags & ZFS_PROJINHERIT) != 0)) { 2100 need_policy = TRUE; 2101 } else { 2102 XVA_CLR_REQ(xvap, XAT_PROJINHERIT); 2103 XVA_SET_REQ(tmpxvattr, XAT_PROJINHERIT); 2104 } 2105 } 2106 2107 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2108 if (xoap->xoa_nounlink != 2109 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) { 2110 need_policy = TRUE; 2111 } else { 2112 XVA_CLR_REQ(xvap, XAT_NOUNLINK); 2113 XVA_SET_REQ(tmpxvattr, XAT_NOUNLINK); 2114 } 2115 } 2116 2117 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2118 if (xoap->xoa_immutable != 2119 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) { 2120 need_policy = TRUE; 2121 } else { 2122 XVA_CLR_REQ(xvap, XAT_IMMUTABLE); 2123 XVA_SET_REQ(tmpxvattr, XAT_IMMUTABLE); 2124 } 2125 } 2126 2127 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2128 if (xoap->xoa_nodump != 2129 ((zp->z_pflags & ZFS_NODUMP) != 0)) { 2130 need_policy = TRUE; 2131 } else { 2132 XVA_CLR_REQ(xvap, XAT_NODUMP); 2133 XVA_SET_REQ(tmpxvattr, XAT_NODUMP); 2134 } 2135 } 2136 2137 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2138 if (xoap->xoa_av_modified != 2139 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) { 2140 need_policy = TRUE; 2141 } else { 2142 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED); 2143 XVA_SET_REQ(tmpxvattr, XAT_AV_MODIFIED); 2144 } 2145 } 2146 2147 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2148 if ((!S_ISREG(ip->i_mode) && 2149 xoap->xoa_av_quarantined) || 2150 xoap->xoa_av_quarantined != 2151 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) { 2152 need_policy = TRUE; 2153 } else { 2154 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED); 2155 XVA_SET_REQ(tmpxvattr, XAT_AV_QUARANTINED); 2156 } 2157 } 2158 2159 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 2160 mutex_exit(&zp->z_lock); 2161 err = SET_ERROR(EPERM); 2162 goto out3; 2163 } 2164 2165 if (need_policy == FALSE && 2166 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) || 2167 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) { 2168 need_policy = TRUE; 2169 } 2170 } 2171 2172 mutex_exit(&zp->z_lock); 2173 2174 if (mask & ATTR_MODE) { 2175 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr, 2176 mnt_ns) == 0) { 2177 err = secpolicy_setid_setsticky_clear(ip, vap, 2178 &oldva, cr, mnt_ns, zfs_i_user_ns(ip)); 2179 if (err) 2180 goto out3; 2181 trim_mask |= ATTR_MODE; 2182 } else { 2183 need_policy = TRUE; 2184 } 2185 } 2186 2187 if (need_policy) { 2188 /* 2189 * If trim_mask is set then take ownership 2190 * has been granted or write_acl is present and user 2191 * has the ability to modify mode. In that case remove 2192 * UID|GID and or MODE from mask so that 2193 * secpolicy_vnode_setattr() doesn't revoke it. 2194 */ 2195 2196 if (trim_mask) { 2197 saved_mask = vap->va_mask; 2198 vap->va_mask &= ~trim_mask; 2199 } 2200 err = secpolicy_vnode_setattr(cr, ip, vap, &oldva, flags, 2201 zfs_zaccess_unix, zp); 2202 if (err) 2203 goto out3; 2204 2205 if (trim_mask) 2206 vap->va_mask |= saved_mask; 2207 } 2208 2209 /* 2210 * secpolicy_vnode_setattr, or take ownership may have 2211 * changed va_mask 2212 */ 2213 mask = vap->va_mask; 2214 2215 if ((mask & (ATTR_UID | ATTR_GID)) || projid != ZFS_INVALID_PROJID) { 2216 handle_eadir = B_TRUE; 2217 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 2218 &xattr_obj, sizeof (xattr_obj)); 2219 2220 if (err == 0 && xattr_obj) { 2221 err = zfs_zget(ZTOZSB(zp), xattr_obj, &attrzp); 2222 if (err) 2223 goto out2; 2224 } 2225 if (mask & ATTR_UID) { 2226 new_kuid = zfs_fuid_create(zfsvfs, 2227 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp); 2228 if (new_kuid != KUID_TO_SUID(ZTOI(zp)->i_uid) && 2229 zfs_id_overquota(zfsvfs, DMU_USERUSED_OBJECT, 2230 new_kuid)) { 2231 if (attrzp) 2232 zrele(attrzp); 2233 err = SET_ERROR(EDQUOT); 2234 goto out2; 2235 } 2236 } 2237 2238 if (mask & ATTR_GID) { 2239 new_kgid = zfs_fuid_create(zfsvfs, 2240 (uint64_t)vap->va_gid, cr, ZFS_GROUP, &fuidp); 2241 if (new_kgid != KGID_TO_SGID(ZTOI(zp)->i_gid) && 2242 zfs_id_overquota(zfsvfs, DMU_GROUPUSED_OBJECT, 2243 new_kgid)) { 2244 if (attrzp) 2245 zrele(attrzp); 2246 err = SET_ERROR(EDQUOT); 2247 goto out2; 2248 } 2249 } 2250 2251 if (projid != ZFS_INVALID_PROJID && 2252 zfs_id_overquota(zfsvfs, DMU_PROJECTUSED_OBJECT, projid)) { 2253 if (attrzp) 2254 zrele(attrzp); 2255 err = EDQUOT; 2256 goto out2; 2257 } 2258 } 2259 tx = dmu_tx_create(os); 2260 2261 if (mask & ATTR_MODE) { 2262 uint64_t pmode = zp->z_mode; 2263 uint64_t acl_obj; 2264 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 2265 2266 if (ZTOZSB(zp)->z_acl_mode == ZFS_ACL_RESTRICTED && 2267 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) { 2268 err = EPERM; 2269 goto out; 2270 } 2271 2272 if ((err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))) 2273 goto out; 2274 2275 mutex_enter(&zp->z_lock); 2276 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) { 2277 /* 2278 * Are we upgrading ACL from old V0 format 2279 * to V1 format? 2280 */ 2281 if (zfsvfs->z_version >= ZPL_VERSION_FUID && 2282 zfs_znode_acl_version(zp) == 2283 ZFS_ACL_VERSION_INITIAL) { 2284 dmu_tx_hold_free(tx, acl_obj, 0, 2285 DMU_OBJECT_END); 2286 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2287 0, aclp->z_acl_bytes); 2288 } else { 2289 dmu_tx_hold_write(tx, acl_obj, 0, 2290 aclp->z_acl_bytes); 2291 } 2292 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) { 2293 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2294 0, aclp->z_acl_bytes); 2295 } 2296 mutex_exit(&zp->z_lock); 2297 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 2298 } else { 2299 if (((mask & ATTR_XVATTR) && 2300 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) || 2301 (projid != ZFS_INVALID_PROJID && 2302 !(zp->z_pflags & ZFS_PROJID))) 2303 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 2304 else 2305 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 2306 } 2307 2308 if (attrzp) { 2309 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE); 2310 } 2311 2312 fuid_dirtied = zfsvfs->z_fuid_dirty; 2313 if (fuid_dirtied) 2314 zfs_fuid_txhold(zfsvfs, tx); 2315 2316 zfs_sa_upgrade_txholds(tx, zp); 2317 2318 err = dmu_tx_assign(tx, TXG_WAIT); 2319 if (err) 2320 goto out; 2321 2322 count = 0; 2323 /* 2324 * Set each attribute requested. 2325 * We group settings according to the locks they need to acquire. 2326 * 2327 * Note: you cannot set ctime directly, although it will be 2328 * updated as a side-effect of calling this function. 2329 */ 2330 2331 if (projid != ZFS_INVALID_PROJID && !(zp->z_pflags & ZFS_PROJID)) { 2332 /* 2333 * For the existed object that is upgraded from old system, 2334 * its on-disk layout has no slot for the project ID attribute. 2335 * But quota accounting logic needs to access related slots by 2336 * offset directly. So we need to adjust old objects' layout 2337 * to make the project ID to some unified and fixed offset. 2338 */ 2339 if (attrzp) 2340 err = sa_add_projid(attrzp->z_sa_hdl, tx, projid); 2341 if (err == 0) 2342 err = sa_add_projid(zp->z_sa_hdl, tx, projid); 2343 2344 if (unlikely(err == EEXIST)) 2345 err = 0; 2346 else if (err != 0) 2347 goto out; 2348 else 2349 projid = ZFS_INVALID_PROJID; 2350 } 2351 2352 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE)) 2353 mutex_enter(&zp->z_acl_lock); 2354 mutex_enter(&zp->z_lock); 2355 2356 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 2357 &zp->z_pflags, sizeof (zp->z_pflags)); 2358 2359 if (attrzp) { 2360 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE)) 2361 mutex_enter(&attrzp->z_acl_lock); 2362 mutex_enter(&attrzp->z_lock); 2363 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 2364 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags, 2365 sizeof (attrzp->z_pflags)); 2366 if (projid != ZFS_INVALID_PROJID) { 2367 attrzp->z_projid = projid; 2368 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 2369 SA_ZPL_PROJID(zfsvfs), NULL, &attrzp->z_projid, 2370 sizeof (attrzp->z_projid)); 2371 } 2372 } 2373 2374 if (mask & (ATTR_UID|ATTR_GID)) { 2375 2376 if (mask & ATTR_UID) { 2377 ZTOI(zp)->i_uid = SUID_TO_KUID(new_kuid); 2378 new_uid = zfs_uid_read(ZTOI(zp)); 2379 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 2380 &new_uid, sizeof (new_uid)); 2381 if (attrzp) { 2382 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 2383 SA_ZPL_UID(zfsvfs), NULL, &new_uid, 2384 sizeof (new_uid)); 2385 ZTOI(attrzp)->i_uid = SUID_TO_KUID(new_uid); 2386 } 2387 } 2388 2389 if (mask & ATTR_GID) { 2390 ZTOI(zp)->i_gid = SGID_TO_KGID(new_kgid); 2391 new_gid = zfs_gid_read(ZTOI(zp)); 2392 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), 2393 NULL, &new_gid, sizeof (new_gid)); 2394 if (attrzp) { 2395 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 2396 SA_ZPL_GID(zfsvfs), NULL, &new_gid, 2397 sizeof (new_gid)); 2398 ZTOI(attrzp)->i_gid = SGID_TO_KGID(new_kgid); 2399 } 2400 } 2401 if (!(mask & ATTR_MODE)) { 2402 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), 2403 NULL, &new_mode, sizeof (new_mode)); 2404 new_mode = zp->z_mode; 2405 } 2406 err = zfs_acl_chown_setattr(zp); 2407 ASSERT(err == 0); 2408 if (attrzp) { 2409 err = zfs_acl_chown_setattr(attrzp); 2410 ASSERT(err == 0); 2411 } 2412 } 2413 2414 if (mask & ATTR_MODE) { 2415 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 2416 &new_mode, sizeof (new_mode)); 2417 zp->z_mode = ZTOI(zp)->i_mode = new_mode; 2418 ASSERT3P(aclp, !=, NULL); 2419 err = zfs_aclset_common(zp, aclp, cr, tx); 2420 ASSERT0(err); 2421 if (zp->z_acl_cached) 2422 zfs_acl_free(zp->z_acl_cached); 2423 zp->z_acl_cached = aclp; 2424 aclp = NULL; 2425 } 2426 2427 if ((mask & ATTR_ATIME) || zp->z_atime_dirty) { 2428 zp->z_atime_dirty = B_FALSE; 2429 ZFS_TIME_ENCODE(&ip->i_atime, atime); 2430 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 2431 &atime, sizeof (atime)); 2432 } 2433 2434 if (mask & (ATTR_MTIME | ATTR_SIZE)) { 2435 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 2436 ZTOI(zp)->i_mtime = zpl_inode_timestamp_truncate( 2437 vap->va_mtime, ZTOI(zp)); 2438 2439 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 2440 mtime, sizeof (mtime)); 2441 } 2442 2443 if (mask & (ATTR_CTIME | ATTR_SIZE)) { 2444 ZFS_TIME_ENCODE(&vap->va_ctime, ctime); 2445 ZTOI(zp)->i_ctime = zpl_inode_timestamp_truncate(vap->va_ctime, 2446 ZTOI(zp)); 2447 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 2448 ctime, sizeof (ctime)); 2449 } 2450 2451 if (projid != ZFS_INVALID_PROJID) { 2452 zp->z_projid = projid; 2453 SA_ADD_BULK_ATTR(bulk, count, 2454 SA_ZPL_PROJID(zfsvfs), NULL, &zp->z_projid, 2455 sizeof (zp->z_projid)); 2456 } 2457 2458 if (attrzp && mask) { 2459 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 2460 SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 2461 sizeof (ctime)); 2462 } 2463 2464 /* 2465 * Do this after setting timestamps to prevent timestamp 2466 * update from toggling bit 2467 */ 2468 2469 if (xoap && (mask & ATTR_XVATTR)) { 2470 2471 /* 2472 * restore trimmed off masks 2473 * so that return masks can be set for caller. 2474 */ 2475 2476 if (XVA_ISSET_REQ(tmpxvattr, XAT_APPENDONLY)) { 2477 XVA_SET_REQ(xvap, XAT_APPENDONLY); 2478 } 2479 if (XVA_ISSET_REQ(tmpxvattr, XAT_NOUNLINK)) { 2480 XVA_SET_REQ(xvap, XAT_NOUNLINK); 2481 } 2482 if (XVA_ISSET_REQ(tmpxvattr, XAT_IMMUTABLE)) { 2483 XVA_SET_REQ(xvap, XAT_IMMUTABLE); 2484 } 2485 if (XVA_ISSET_REQ(tmpxvattr, XAT_NODUMP)) { 2486 XVA_SET_REQ(xvap, XAT_NODUMP); 2487 } 2488 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_MODIFIED)) { 2489 XVA_SET_REQ(xvap, XAT_AV_MODIFIED); 2490 } 2491 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_QUARANTINED)) { 2492 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED); 2493 } 2494 if (XVA_ISSET_REQ(tmpxvattr, XAT_PROJINHERIT)) { 2495 XVA_SET_REQ(xvap, XAT_PROJINHERIT); 2496 } 2497 2498 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 2499 ASSERT(S_ISREG(ip->i_mode)); 2500 2501 zfs_xvattr_set(zp, xvap, tx); 2502 } 2503 2504 if (fuid_dirtied) 2505 zfs_fuid_sync(zfsvfs, tx); 2506 2507 if (mask != 0) 2508 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp); 2509 2510 mutex_exit(&zp->z_lock); 2511 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE)) 2512 mutex_exit(&zp->z_acl_lock); 2513 2514 if (attrzp) { 2515 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE)) 2516 mutex_exit(&attrzp->z_acl_lock); 2517 mutex_exit(&attrzp->z_lock); 2518 } 2519 out: 2520 if (err == 0 && xattr_count > 0) { 2521 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk, 2522 xattr_count, tx); 2523 ASSERT(err2 == 0); 2524 } 2525 2526 if (aclp) 2527 zfs_acl_free(aclp); 2528 2529 if (fuidp) { 2530 zfs_fuid_info_free(fuidp); 2531 fuidp = NULL; 2532 } 2533 2534 if (err) { 2535 dmu_tx_abort(tx); 2536 if (attrzp) 2537 zrele(attrzp); 2538 if (err == ERESTART) 2539 goto top; 2540 } else { 2541 if (count > 0) 2542 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 2543 dmu_tx_commit(tx); 2544 if (attrzp) { 2545 if (err2 == 0 && handle_eadir) 2546 err = zfs_setattr_dir(attrzp); 2547 zrele(attrzp); 2548 } 2549 zfs_znode_update_vfs(zp); 2550 } 2551 2552 out2: 2553 if (os->os_sync == ZFS_SYNC_ALWAYS) 2554 zil_commit(zilog, 0); 2555 2556 out3: 2557 kmem_free(xattr_bulk, sizeof (sa_bulk_attr_t) * bulks); 2558 kmem_free(bulk, sizeof (sa_bulk_attr_t) * bulks); 2559 kmem_free(tmpxvattr, sizeof (xvattr_t)); 2560 zfs_exit(zfsvfs, FTAG); 2561 return (err); 2562 } 2563 2564 typedef struct zfs_zlock { 2565 krwlock_t *zl_rwlock; /* lock we acquired */ 2566 znode_t *zl_znode; /* znode we held */ 2567 struct zfs_zlock *zl_next; /* next in list */ 2568 } zfs_zlock_t; 2569 2570 /* 2571 * Drop locks and release vnodes that were held by zfs_rename_lock(). 2572 */ 2573 static void 2574 zfs_rename_unlock(zfs_zlock_t **zlpp) 2575 { 2576 zfs_zlock_t *zl; 2577 2578 while ((zl = *zlpp) != NULL) { 2579 if (zl->zl_znode != NULL) 2580 zfs_zrele_async(zl->zl_znode); 2581 rw_exit(zl->zl_rwlock); 2582 *zlpp = zl->zl_next; 2583 kmem_free(zl, sizeof (*zl)); 2584 } 2585 } 2586 2587 /* 2588 * Search back through the directory tree, using the ".." entries. 2589 * Lock each directory in the chain to prevent concurrent renames. 2590 * Fail any attempt to move a directory into one of its own descendants. 2591 * XXX - z_parent_lock can overlap with map or grow locks 2592 */ 2593 static int 2594 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 2595 { 2596 zfs_zlock_t *zl; 2597 znode_t *zp = tdzp; 2598 uint64_t rootid = ZTOZSB(zp)->z_root; 2599 uint64_t oidp = zp->z_id; 2600 krwlock_t *rwlp = &szp->z_parent_lock; 2601 krw_t rw = RW_WRITER; 2602 2603 /* 2604 * First pass write-locks szp and compares to zp->z_id. 2605 * Later passes read-lock zp and compare to zp->z_parent. 2606 */ 2607 do { 2608 if (!rw_tryenter(rwlp, rw)) { 2609 /* 2610 * Another thread is renaming in this path. 2611 * Note that if we are a WRITER, we don't have any 2612 * parent_locks held yet. 2613 */ 2614 if (rw == RW_READER && zp->z_id > szp->z_id) { 2615 /* 2616 * Drop our locks and restart 2617 */ 2618 zfs_rename_unlock(&zl); 2619 *zlpp = NULL; 2620 zp = tdzp; 2621 oidp = zp->z_id; 2622 rwlp = &szp->z_parent_lock; 2623 rw = RW_WRITER; 2624 continue; 2625 } else { 2626 /* 2627 * Wait for other thread to drop its locks 2628 */ 2629 rw_enter(rwlp, rw); 2630 } 2631 } 2632 2633 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 2634 zl->zl_rwlock = rwlp; 2635 zl->zl_znode = NULL; 2636 zl->zl_next = *zlpp; 2637 *zlpp = zl; 2638 2639 if (oidp == szp->z_id) /* We're a descendant of szp */ 2640 return (SET_ERROR(EINVAL)); 2641 2642 if (oidp == rootid) /* We've hit the top */ 2643 return (0); 2644 2645 if (rw == RW_READER) { /* i.e. not the first pass */ 2646 int error = zfs_zget(ZTOZSB(zp), oidp, &zp); 2647 if (error) 2648 return (error); 2649 zl->zl_znode = zp; 2650 } 2651 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(ZTOZSB(zp)), 2652 &oidp, sizeof (oidp)); 2653 rwlp = &zp->z_parent_lock; 2654 rw = RW_READER; 2655 2656 } while (zp->z_id != sdzp->z_id); 2657 2658 return (0); 2659 } 2660 2661 /* 2662 * Move an entry from the provided source directory to the target 2663 * directory. Change the entry name as indicated. 2664 * 2665 * IN: sdzp - Source directory containing the "old entry". 2666 * snm - Old entry name. 2667 * tdzp - Target directory to contain the "new entry". 2668 * tnm - New entry name. 2669 * cr - credentials of caller. 2670 * flags - case flags 2671 * rflags - RENAME_* flags 2672 * wa_vap - attributes for RENAME_WHITEOUT (must be a char 0:0). 2673 * mnt_ns - user namespace of the mount 2674 * 2675 * RETURN: 0 on success, error code on failure. 2676 * 2677 * Timestamps: 2678 * sdzp,tdzp - ctime|mtime updated 2679 */ 2680 int 2681 zfs_rename(znode_t *sdzp, char *snm, znode_t *tdzp, char *tnm, 2682 cred_t *cr, int flags, uint64_t rflags, vattr_t *wo_vap, zidmap_t *mnt_ns) 2683 { 2684 znode_t *szp, *tzp; 2685 zfsvfs_t *zfsvfs = ZTOZSB(sdzp); 2686 zilog_t *zilog; 2687 zfs_dirlock_t *sdl, *tdl; 2688 dmu_tx_t *tx; 2689 zfs_zlock_t *zl; 2690 int cmp, serr, terr; 2691 int error = 0; 2692 int zflg = 0; 2693 boolean_t waited = B_FALSE; 2694 /* Needed for whiteout inode creation. */ 2695 boolean_t fuid_dirtied; 2696 zfs_acl_ids_t acl_ids; 2697 boolean_t have_acl = B_FALSE; 2698 znode_t *wzp = NULL; 2699 2700 2701 if (snm == NULL || tnm == NULL) 2702 return (SET_ERROR(EINVAL)); 2703 2704 if (rflags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 2705 return (SET_ERROR(EINVAL)); 2706 2707 /* Already checked by Linux VFS, but just to make sure. */ 2708 if (rflags & RENAME_EXCHANGE && 2709 (rflags & (RENAME_NOREPLACE | RENAME_WHITEOUT))) 2710 return (SET_ERROR(EINVAL)); 2711 2712 /* 2713 * Make sure we only get wo_vap iff. RENAME_WHITEOUT and that it's the 2714 * right kind of vattr_t for the whiteout file. These are set 2715 * internally by ZFS so should never be incorrect. 2716 */ 2717 VERIFY_EQUIV(rflags & RENAME_WHITEOUT, wo_vap != NULL); 2718 VERIFY_IMPLY(wo_vap, wo_vap->va_mode == S_IFCHR); 2719 VERIFY_IMPLY(wo_vap, wo_vap->va_rdev == makedevice(0, 0)); 2720 2721 if ((error = zfs_enter_verify_zp(zfsvfs, sdzp, FTAG)) != 0) 2722 return (error); 2723 zilog = zfsvfs->z_log; 2724 2725 if ((error = zfs_verify_zp(tdzp)) != 0) { 2726 zfs_exit(zfsvfs, FTAG); 2727 return (error); 2728 } 2729 2730 /* 2731 * We check i_sb because snapshots and the ctldir must have different 2732 * super blocks. 2733 */ 2734 if (ZTOI(tdzp)->i_sb != ZTOI(sdzp)->i_sb || 2735 zfsctl_is_node(ZTOI(tdzp))) { 2736 zfs_exit(zfsvfs, FTAG); 2737 return (SET_ERROR(EXDEV)); 2738 } 2739 2740 if (zfsvfs->z_utf8 && u8_validate(tnm, 2741 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 2742 zfs_exit(zfsvfs, FTAG); 2743 return (SET_ERROR(EILSEQ)); 2744 } 2745 2746 if (flags & FIGNORECASE) 2747 zflg |= ZCILOOK; 2748 2749 top: 2750 szp = NULL; 2751 tzp = NULL; 2752 zl = NULL; 2753 2754 /* 2755 * This is to prevent the creation of links into attribute space 2756 * by renaming a linked file into/outof an attribute directory. 2757 * See the comment in zfs_link() for why this is considered bad. 2758 */ 2759 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) { 2760 zfs_exit(zfsvfs, FTAG); 2761 return (SET_ERROR(EINVAL)); 2762 } 2763 2764 /* 2765 * Lock source and target directory entries. To prevent deadlock, 2766 * a lock ordering must be defined. We lock the directory with 2767 * the smallest object id first, or if it's a tie, the one with 2768 * the lexically first name. 2769 */ 2770 if (sdzp->z_id < tdzp->z_id) { 2771 cmp = -1; 2772 } else if (sdzp->z_id > tdzp->z_id) { 2773 cmp = 1; 2774 } else { 2775 /* 2776 * First compare the two name arguments without 2777 * considering any case folding. 2778 */ 2779 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER); 2780 2781 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error); 2782 ASSERT(error == 0 || !zfsvfs->z_utf8); 2783 if (cmp == 0) { 2784 /* 2785 * POSIX: "If the old argument and the new argument 2786 * both refer to links to the same existing file, 2787 * the rename() function shall return successfully 2788 * and perform no other action." 2789 */ 2790 zfs_exit(zfsvfs, FTAG); 2791 return (0); 2792 } 2793 /* 2794 * If the file system is case-folding, then we may 2795 * have some more checking to do. A case-folding file 2796 * system is either supporting mixed case sensitivity 2797 * access or is completely case-insensitive. Note 2798 * that the file system is always case preserving. 2799 * 2800 * In mixed sensitivity mode case sensitive behavior 2801 * is the default. FIGNORECASE must be used to 2802 * explicitly request case insensitive behavior. 2803 * 2804 * If the source and target names provided differ only 2805 * by case (e.g., a request to rename 'tim' to 'Tim'), 2806 * we will treat this as a special case in the 2807 * case-insensitive mode: as long as the source name 2808 * is an exact match, we will allow this to proceed as 2809 * a name-change request. 2810 */ 2811 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 2812 (zfsvfs->z_case == ZFS_CASE_MIXED && 2813 flags & FIGNORECASE)) && 2814 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST, 2815 &error) == 0) { 2816 /* 2817 * case preserving rename request, require exact 2818 * name matches 2819 */ 2820 zflg |= ZCIEXACT; 2821 zflg &= ~ZCILOOK; 2822 } 2823 } 2824 2825 /* 2826 * If the source and destination directories are the same, we should 2827 * grab the z_name_lock of that directory only once. 2828 */ 2829 if (sdzp == tdzp) { 2830 zflg |= ZHAVELOCK; 2831 rw_enter(&sdzp->z_name_lock, RW_READER); 2832 } 2833 2834 if (cmp < 0) { 2835 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, 2836 ZEXISTS | zflg, NULL, NULL); 2837 terr = zfs_dirent_lock(&tdl, 2838 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL); 2839 } else { 2840 terr = zfs_dirent_lock(&tdl, 2841 tdzp, tnm, &tzp, zflg, NULL, NULL); 2842 serr = zfs_dirent_lock(&sdl, 2843 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg, 2844 NULL, NULL); 2845 } 2846 2847 if (serr) { 2848 /* 2849 * Source entry invalid or not there. 2850 */ 2851 if (!terr) { 2852 zfs_dirent_unlock(tdl); 2853 if (tzp) 2854 zrele(tzp); 2855 } 2856 2857 if (sdzp == tdzp) 2858 rw_exit(&sdzp->z_name_lock); 2859 2860 if (strcmp(snm, "..") == 0) 2861 serr = EINVAL; 2862 zfs_exit(zfsvfs, FTAG); 2863 return (serr); 2864 } 2865 if (terr) { 2866 zfs_dirent_unlock(sdl); 2867 zrele(szp); 2868 2869 if (sdzp == tdzp) 2870 rw_exit(&sdzp->z_name_lock); 2871 2872 if (strcmp(tnm, "..") == 0) 2873 terr = EINVAL; 2874 zfs_exit(zfsvfs, FTAG); 2875 return (terr); 2876 } 2877 2878 /* 2879 * If we are using project inheritance, means if the directory has 2880 * ZFS_PROJINHERIT set, then its descendant directories will inherit 2881 * not only the project ID, but also the ZFS_PROJINHERIT flag. Under 2882 * such case, we only allow renames into our tree when the project 2883 * IDs are the same. 2884 */ 2885 if (tdzp->z_pflags & ZFS_PROJINHERIT && 2886 tdzp->z_projid != szp->z_projid) { 2887 error = SET_ERROR(EXDEV); 2888 goto out; 2889 } 2890 2891 /* 2892 * Must have write access at the source to remove the old entry 2893 * and write access at the target to create the new entry. 2894 * Note that if target and source are the same, this can be 2895 * done in a single check. 2896 */ 2897 if ((error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr, mnt_ns))) 2898 goto out; 2899 2900 if (S_ISDIR(ZTOI(szp)->i_mode)) { 2901 /* 2902 * Check to make sure rename is valid. 2903 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 2904 */ 2905 if ((error = zfs_rename_lock(szp, tdzp, sdzp, &zl))) 2906 goto out; 2907 } 2908 2909 /* 2910 * Does target exist? 2911 */ 2912 if (tzp) { 2913 if (rflags & RENAME_NOREPLACE) { 2914 error = SET_ERROR(EEXIST); 2915 goto out; 2916 } 2917 /* 2918 * Source and target must be the same type (unless exchanging). 2919 */ 2920 if (!(rflags & RENAME_EXCHANGE)) { 2921 boolean_t s_is_dir = S_ISDIR(ZTOI(szp)->i_mode) != 0; 2922 boolean_t t_is_dir = S_ISDIR(ZTOI(tzp)->i_mode) != 0; 2923 2924 if (s_is_dir != t_is_dir) { 2925 error = SET_ERROR(s_is_dir ? ENOTDIR : EISDIR); 2926 goto out; 2927 } 2928 } 2929 /* 2930 * POSIX dictates that when the source and target 2931 * entries refer to the same file object, rename 2932 * must do nothing and exit without error. 2933 */ 2934 if (szp->z_id == tzp->z_id) { 2935 error = 0; 2936 goto out; 2937 } 2938 } else if (rflags & RENAME_EXCHANGE) { 2939 /* Target must exist for RENAME_EXCHANGE. */ 2940 error = SET_ERROR(ENOENT); 2941 goto out; 2942 } 2943 2944 /* Set up inode creation for RENAME_WHITEOUT. */ 2945 if (rflags & RENAME_WHITEOUT) { 2946 /* 2947 * Whiteout files are not regular files or directories, so to 2948 * match zfs_create() we do not inherit the project id. 2949 */ 2950 uint64_t wo_projid = ZFS_DEFAULT_PROJID; 2951 2952 error = zfs_zaccess(sdzp, ACE_ADD_FILE, 0, B_FALSE, cr, mnt_ns); 2953 if (error) 2954 goto out; 2955 2956 if (!have_acl) { 2957 error = zfs_acl_ids_create(sdzp, 0, wo_vap, cr, NULL, 2958 &acl_ids, mnt_ns); 2959 if (error) 2960 goto out; 2961 have_acl = B_TRUE; 2962 } 2963 2964 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, wo_projid)) { 2965 error = SET_ERROR(EDQUOT); 2966 goto out; 2967 } 2968 } 2969 2970 tx = dmu_tx_create(zfsvfs->z_os); 2971 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 2972 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE); 2973 dmu_tx_hold_zap(tx, sdzp->z_id, 2974 (rflags & RENAME_EXCHANGE) ? TRUE : FALSE, snm); 2975 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 2976 if (sdzp != tdzp) { 2977 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE); 2978 zfs_sa_upgrade_txholds(tx, tdzp); 2979 } 2980 if (tzp) { 2981 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE); 2982 zfs_sa_upgrade_txholds(tx, tzp); 2983 } 2984 if (rflags & RENAME_WHITEOUT) { 2985 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 2986 ZFS_SA_BASE_ATTR_SIZE); 2987 2988 dmu_tx_hold_zap(tx, sdzp->z_id, TRUE, snm); 2989 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE); 2990 if (!zfsvfs->z_use_sa && 2991 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 2992 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2993 0, acl_ids.z_aclp->z_acl_bytes); 2994 } 2995 } 2996 fuid_dirtied = zfsvfs->z_fuid_dirty; 2997 if (fuid_dirtied) 2998 zfs_fuid_txhold(zfsvfs, tx); 2999 zfs_sa_upgrade_txholds(tx, szp); 3000 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 3001 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT); 3002 if (error) { 3003 if (zl != NULL) 3004 zfs_rename_unlock(&zl); 3005 zfs_dirent_unlock(sdl); 3006 zfs_dirent_unlock(tdl); 3007 3008 if (sdzp == tdzp) 3009 rw_exit(&sdzp->z_name_lock); 3010 3011 if (error == ERESTART) { 3012 waited = B_TRUE; 3013 dmu_tx_wait(tx); 3014 dmu_tx_abort(tx); 3015 zrele(szp); 3016 if (tzp) 3017 zrele(tzp); 3018 goto top; 3019 } 3020 dmu_tx_abort(tx); 3021 zrele(szp); 3022 if (tzp) 3023 zrele(tzp); 3024 zfs_exit(zfsvfs, FTAG); 3025 return (error); 3026 } 3027 3028 /* 3029 * Unlink the source. 3030 */ 3031 szp->z_pflags |= ZFS_AV_MODIFIED; 3032 if (tdzp->z_pflags & ZFS_PROJINHERIT) 3033 szp->z_pflags |= ZFS_PROJINHERIT; 3034 3035 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs), 3036 (void *)&szp->z_pflags, sizeof (uint64_t), tx); 3037 VERIFY0(error); 3038 3039 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 3040 if (error) 3041 goto commit; 3042 3043 /* 3044 * Unlink the target. 3045 */ 3046 if (tzp) { 3047 int tzflg = zflg; 3048 3049 if (rflags & RENAME_EXCHANGE) { 3050 /* This inode will be re-linked soon. */ 3051 tzflg |= ZRENAMING; 3052 3053 tzp->z_pflags |= ZFS_AV_MODIFIED; 3054 if (sdzp->z_pflags & ZFS_PROJINHERIT) 3055 tzp->z_pflags |= ZFS_PROJINHERIT; 3056 3057 error = sa_update(tzp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs), 3058 (void *)&tzp->z_pflags, sizeof (uint64_t), tx); 3059 ASSERT0(error); 3060 } 3061 error = zfs_link_destroy(tdl, tzp, tx, tzflg, NULL); 3062 if (error) 3063 goto commit_link_szp; 3064 } 3065 3066 /* 3067 * Create the new target links: 3068 * * We always link the target. 3069 * * RENAME_EXCHANGE: Link the old target to the source. 3070 * * RENAME_WHITEOUT: Create a whiteout inode in-place of the source. 3071 */ 3072 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 3073 if (error) { 3074 /* 3075 * If we have removed the existing target, a subsequent call to 3076 * zfs_link_create() to add back the same entry, but with a new 3077 * dnode (szp), should not fail. 3078 */ 3079 ASSERT3P(tzp, ==, NULL); 3080 goto commit_link_tzp; 3081 } 3082 3083 switch (rflags & (RENAME_EXCHANGE | RENAME_WHITEOUT)) { 3084 case RENAME_EXCHANGE: 3085 error = zfs_link_create(sdl, tzp, tx, ZRENAMING); 3086 /* 3087 * The same argument as zfs_link_create() failing for 3088 * szp applies here, since the source directory must 3089 * have had an entry we are replacing. 3090 */ 3091 ASSERT0(error); 3092 if (error) 3093 goto commit_unlink_td_szp; 3094 break; 3095 case RENAME_WHITEOUT: 3096 zfs_mknode(sdzp, wo_vap, tx, cr, 0, &wzp, &acl_ids); 3097 error = zfs_link_create(sdl, wzp, tx, ZNEW); 3098 if (error) { 3099 zfs_znode_delete(wzp, tx); 3100 remove_inode_hash(ZTOI(wzp)); 3101 goto commit_unlink_td_szp; 3102 } 3103 break; 3104 } 3105 3106 if (fuid_dirtied) 3107 zfs_fuid_sync(zfsvfs, tx); 3108 3109 switch (rflags & (RENAME_EXCHANGE | RENAME_WHITEOUT)) { 3110 case RENAME_EXCHANGE: 3111 zfs_log_rename_exchange(zilog, tx, 3112 (flags & FIGNORECASE ? TX_CI : 0), sdzp, sdl->dl_name, 3113 tdzp, tdl->dl_name, szp); 3114 break; 3115 case RENAME_WHITEOUT: 3116 zfs_log_rename_whiteout(zilog, tx, 3117 (flags & FIGNORECASE ? TX_CI : 0), sdzp, sdl->dl_name, 3118 tdzp, tdl->dl_name, szp, wzp); 3119 break; 3120 default: 3121 ASSERT0(rflags & ~RENAME_NOREPLACE); 3122 zfs_log_rename(zilog, tx, (flags & FIGNORECASE ? TX_CI : 0), 3123 sdzp, sdl->dl_name, tdzp, tdl->dl_name, szp); 3124 break; 3125 } 3126 3127 commit: 3128 dmu_tx_commit(tx); 3129 out: 3130 if (have_acl) 3131 zfs_acl_ids_free(&acl_ids); 3132 3133 zfs_znode_update_vfs(sdzp); 3134 if (sdzp == tdzp) 3135 rw_exit(&sdzp->z_name_lock); 3136 3137 if (sdzp != tdzp) 3138 zfs_znode_update_vfs(tdzp); 3139 3140 zfs_znode_update_vfs(szp); 3141 zrele(szp); 3142 if (wzp) { 3143 zfs_znode_update_vfs(wzp); 3144 zrele(wzp); 3145 } 3146 if (tzp) { 3147 zfs_znode_update_vfs(tzp); 3148 zrele(tzp); 3149 } 3150 3151 if (zl != NULL) 3152 zfs_rename_unlock(&zl); 3153 3154 zfs_dirent_unlock(sdl); 3155 zfs_dirent_unlock(tdl); 3156 3157 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3158 zil_commit(zilog, 0); 3159 3160 zfs_exit(zfsvfs, FTAG); 3161 return (error); 3162 3163 /* 3164 * Clean-up path for broken link state. 3165 * 3166 * At this point we are in a (very) bad state, so we need to do our 3167 * best to correct the state. In particular, all of the nlinks are 3168 * wrong because we were destroying and creating links with ZRENAMING. 3169 * 3170 * In some form, all of these operations have to resolve the state: 3171 * 3172 * * link_destroy() *must* succeed. Fortunately, this is very likely 3173 * since we only just created it. 3174 * 3175 * * link_create()s are allowed to fail (though they shouldn't because 3176 * we only just unlinked them and are putting the entries back 3177 * during clean-up). But if they fail, we can just forcefully drop 3178 * the nlink value to (at the very least) avoid broken nlink values 3179 * -- though in the case of non-empty directories we will have to 3180 * panic (otherwise we'd have a leaked directory with a broken ..). 3181 */ 3182 commit_unlink_td_szp: 3183 VERIFY0(zfs_link_destroy(tdl, szp, tx, ZRENAMING, NULL)); 3184 commit_link_tzp: 3185 if (tzp) { 3186 if (zfs_link_create(tdl, tzp, tx, ZRENAMING)) 3187 VERIFY0(zfs_drop_nlink(tzp, tx, NULL)); 3188 } 3189 commit_link_szp: 3190 if (zfs_link_create(sdl, szp, tx, ZRENAMING)) 3191 VERIFY0(zfs_drop_nlink(szp, tx, NULL)); 3192 goto commit; 3193 } 3194 3195 /* 3196 * Insert the indicated symbolic reference entry into the directory. 3197 * 3198 * IN: dzp - Directory to contain new symbolic link. 3199 * name - Name of directory entry in dip. 3200 * vap - Attributes of new entry. 3201 * link - Name for new symlink entry. 3202 * cr - credentials of caller. 3203 * flags - case flags 3204 * mnt_ns - user namespace of the mount 3205 * 3206 * OUT: zpp - Znode for new symbolic link. 3207 * 3208 * RETURN: 0 on success, error code on failure. 3209 * 3210 * Timestamps: 3211 * dip - ctime|mtime updated 3212 */ 3213 int 3214 zfs_symlink(znode_t *dzp, char *name, vattr_t *vap, char *link, 3215 znode_t **zpp, cred_t *cr, int flags, zidmap_t *mnt_ns) 3216 { 3217 znode_t *zp; 3218 zfs_dirlock_t *dl; 3219 dmu_tx_t *tx; 3220 zfsvfs_t *zfsvfs = ZTOZSB(dzp); 3221 zilog_t *zilog; 3222 uint64_t len = strlen(link); 3223 int error; 3224 int zflg = ZNEW; 3225 zfs_acl_ids_t acl_ids; 3226 boolean_t fuid_dirtied; 3227 uint64_t txtype = TX_SYMLINK; 3228 boolean_t waited = B_FALSE; 3229 3230 ASSERT(S_ISLNK(vap->va_mode)); 3231 3232 if (name == NULL) 3233 return (SET_ERROR(EINVAL)); 3234 3235 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) 3236 return (error); 3237 zilog = zfsvfs->z_log; 3238 3239 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 3240 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3241 zfs_exit(zfsvfs, FTAG); 3242 return (SET_ERROR(EILSEQ)); 3243 } 3244 if (flags & FIGNORECASE) 3245 zflg |= ZCILOOK; 3246 3247 if (len > MAXPATHLEN) { 3248 zfs_exit(zfsvfs, FTAG); 3249 return (SET_ERROR(ENAMETOOLONG)); 3250 } 3251 3252 if ((error = zfs_acl_ids_create(dzp, 0, 3253 vap, cr, NULL, &acl_ids, mnt_ns)) != 0) { 3254 zfs_exit(zfsvfs, FTAG); 3255 return (error); 3256 } 3257 top: 3258 *zpp = NULL; 3259 3260 /* 3261 * Attempt to lock directory; fail if entry already exists. 3262 */ 3263 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL); 3264 if (error) { 3265 zfs_acl_ids_free(&acl_ids); 3266 zfs_exit(zfsvfs, FTAG); 3267 return (error); 3268 } 3269 3270 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr, mnt_ns))) { 3271 zfs_acl_ids_free(&acl_ids); 3272 zfs_dirent_unlock(dl); 3273 zfs_exit(zfsvfs, FTAG); 3274 return (error); 3275 } 3276 3277 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, ZFS_DEFAULT_PROJID)) { 3278 zfs_acl_ids_free(&acl_ids); 3279 zfs_dirent_unlock(dl); 3280 zfs_exit(zfsvfs, FTAG); 3281 return (SET_ERROR(EDQUOT)); 3282 } 3283 tx = dmu_tx_create(zfsvfs->z_os); 3284 fuid_dirtied = zfsvfs->z_fuid_dirty; 3285 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 3286 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 3287 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 3288 ZFS_SA_BASE_ATTR_SIZE + len); 3289 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 3290 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 3291 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 3292 acl_ids.z_aclp->z_acl_bytes); 3293 } 3294 if (fuid_dirtied) 3295 zfs_fuid_txhold(zfsvfs, tx); 3296 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT); 3297 if (error) { 3298 zfs_dirent_unlock(dl); 3299 if (error == ERESTART) { 3300 waited = B_TRUE; 3301 dmu_tx_wait(tx); 3302 dmu_tx_abort(tx); 3303 goto top; 3304 } 3305 zfs_acl_ids_free(&acl_ids); 3306 dmu_tx_abort(tx); 3307 zfs_exit(zfsvfs, FTAG); 3308 return (error); 3309 } 3310 3311 /* 3312 * Create a new object for the symlink. 3313 * for version 4 ZPL datasets the symlink will be an SA attribute 3314 */ 3315 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 3316 3317 if (fuid_dirtied) 3318 zfs_fuid_sync(zfsvfs, tx); 3319 3320 mutex_enter(&zp->z_lock); 3321 if (zp->z_is_sa) 3322 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs), 3323 link, len, tx); 3324 else 3325 zfs_sa_symlink(zp, link, len, tx); 3326 mutex_exit(&zp->z_lock); 3327 3328 zp->z_size = len; 3329 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 3330 &zp->z_size, sizeof (zp->z_size), tx); 3331 /* 3332 * Insert the new object into the directory. 3333 */ 3334 error = zfs_link_create(dl, zp, tx, ZNEW); 3335 if (error != 0) { 3336 zfs_znode_delete(zp, tx); 3337 remove_inode_hash(ZTOI(zp)); 3338 } else { 3339 if (flags & FIGNORECASE) 3340 txtype |= TX_CI; 3341 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link); 3342 3343 zfs_znode_update_vfs(dzp); 3344 zfs_znode_update_vfs(zp); 3345 } 3346 3347 zfs_acl_ids_free(&acl_ids); 3348 3349 dmu_tx_commit(tx); 3350 3351 zfs_dirent_unlock(dl); 3352 3353 if (error == 0) { 3354 *zpp = zp; 3355 3356 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3357 zil_commit(zilog, 0); 3358 } else { 3359 zrele(zp); 3360 } 3361 3362 zfs_exit(zfsvfs, FTAG); 3363 return (error); 3364 } 3365 3366 /* 3367 * Return, in the buffer contained in the provided uio structure, 3368 * the symbolic path referred to by ip. 3369 * 3370 * IN: ip - inode of symbolic link 3371 * uio - structure to contain the link path. 3372 * cr - credentials of caller. 3373 * 3374 * RETURN: 0 if success 3375 * error code if failure 3376 * 3377 * Timestamps: 3378 * ip - atime updated 3379 */ 3380 int 3381 zfs_readlink(struct inode *ip, zfs_uio_t *uio, cred_t *cr) 3382 { 3383 (void) cr; 3384 znode_t *zp = ITOZ(ip); 3385 zfsvfs_t *zfsvfs = ITOZSB(ip); 3386 int error; 3387 3388 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 3389 return (error); 3390 3391 mutex_enter(&zp->z_lock); 3392 if (zp->z_is_sa) 3393 error = sa_lookup_uio(zp->z_sa_hdl, 3394 SA_ZPL_SYMLINK(zfsvfs), uio); 3395 else 3396 error = zfs_sa_readlink(zp, uio); 3397 mutex_exit(&zp->z_lock); 3398 3399 zfs_exit(zfsvfs, FTAG); 3400 return (error); 3401 } 3402 3403 /* 3404 * Insert a new entry into directory tdzp referencing szp. 3405 * 3406 * IN: tdzp - Directory to contain new entry. 3407 * szp - znode of new entry. 3408 * name - name of new entry. 3409 * cr - credentials of caller. 3410 * flags - case flags. 3411 * 3412 * RETURN: 0 if success 3413 * error code if failure 3414 * 3415 * Timestamps: 3416 * tdzp - ctime|mtime updated 3417 * szp - ctime updated 3418 */ 3419 int 3420 zfs_link(znode_t *tdzp, znode_t *szp, char *name, cred_t *cr, 3421 int flags) 3422 { 3423 struct inode *sip = ZTOI(szp); 3424 znode_t *tzp; 3425 zfsvfs_t *zfsvfs = ZTOZSB(tdzp); 3426 zilog_t *zilog; 3427 zfs_dirlock_t *dl; 3428 dmu_tx_t *tx; 3429 int error; 3430 int zf = ZNEW; 3431 uint64_t parent; 3432 uid_t owner; 3433 boolean_t waited = B_FALSE; 3434 boolean_t is_tmpfile = 0; 3435 uint64_t txg; 3436 #ifdef HAVE_TMPFILE 3437 is_tmpfile = (sip->i_nlink == 0 && (sip->i_state & I_LINKABLE)); 3438 #endif 3439 ASSERT(S_ISDIR(ZTOI(tdzp)->i_mode)); 3440 3441 if (name == NULL) 3442 return (SET_ERROR(EINVAL)); 3443 3444 if ((error = zfs_enter_verify_zp(zfsvfs, tdzp, FTAG)) != 0) 3445 return (error); 3446 zilog = zfsvfs->z_log; 3447 3448 /* 3449 * POSIX dictates that we return EPERM here. 3450 * Better choices include ENOTSUP or EISDIR. 3451 */ 3452 if (S_ISDIR(sip->i_mode)) { 3453 zfs_exit(zfsvfs, FTAG); 3454 return (SET_ERROR(EPERM)); 3455 } 3456 3457 if ((error = zfs_verify_zp(szp)) != 0) { 3458 zfs_exit(zfsvfs, FTAG); 3459 return (error); 3460 } 3461 3462 /* 3463 * If we are using project inheritance, means if the directory has 3464 * ZFS_PROJINHERIT set, then its descendant directories will inherit 3465 * not only the project ID, but also the ZFS_PROJINHERIT flag. Under 3466 * such case, we only allow hard link creation in our tree when the 3467 * project IDs are the same. 3468 */ 3469 if (tdzp->z_pflags & ZFS_PROJINHERIT && 3470 tdzp->z_projid != szp->z_projid) { 3471 zfs_exit(zfsvfs, FTAG); 3472 return (SET_ERROR(EXDEV)); 3473 } 3474 3475 /* 3476 * We check i_sb because snapshots and the ctldir must have different 3477 * super blocks. 3478 */ 3479 if (sip->i_sb != ZTOI(tdzp)->i_sb || zfsctl_is_node(sip)) { 3480 zfs_exit(zfsvfs, FTAG); 3481 return (SET_ERROR(EXDEV)); 3482 } 3483 3484 /* Prevent links to .zfs/shares files */ 3485 3486 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 3487 &parent, sizeof (uint64_t))) != 0) { 3488 zfs_exit(zfsvfs, FTAG); 3489 return (error); 3490 } 3491 if (parent == zfsvfs->z_shares_dir) { 3492 zfs_exit(zfsvfs, FTAG); 3493 return (SET_ERROR(EPERM)); 3494 } 3495 3496 if (zfsvfs->z_utf8 && u8_validate(name, 3497 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3498 zfs_exit(zfsvfs, FTAG); 3499 return (SET_ERROR(EILSEQ)); 3500 } 3501 if (flags & FIGNORECASE) 3502 zf |= ZCILOOK; 3503 3504 /* 3505 * We do not support links between attributes and non-attributes 3506 * because of the potential security risk of creating links 3507 * into "normal" file space in order to circumvent restrictions 3508 * imposed in attribute space. 3509 */ 3510 if ((szp->z_pflags & ZFS_XATTR) != (tdzp->z_pflags & ZFS_XATTR)) { 3511 zfs_exit(zfsvfs, FTAG); 3512 return (SET_ERROR(EINVAL)); 3513 } 3514 3515 owner = zfs_fuid_map_id(zfsvfs, KUID_TO_SUID(sip->i_uid), 3516 cr, ZFS_OWNER); 3517 if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) { 3518 zfs_exit(zfsvfs, FTAG); 3519 return (SET_ERROR(EPERM)); 3520 } 3521 3522 if ((error = zfs_zaccess(tdzp, ACE_ADD_FILE, 0, B_FALSE, cr, 3523 zfs_init_idmap))) { 3524 zfs_exit(zfsvfs, FTAG); 3525 return (error); 3526 } 3527 3528 top: 3529 /* 3530 * Attempt to lock directory; fail if entry already exists. 3531 */ 3532 error = zfs_dirent_lock(&dl, tdzp, name, &tzp, zf, NULL, NULL); 3533 if (error) { 3534 zfs_exit(zfsvfs, FTAG); 3535 return (error); 3536 } 3537 3538 tx = dmu_tx_create(zfsvfs->z_os); 3539 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 3540 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, name); 3541 if (is_tmpfile) 3542 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 3543 3544 zfs_sa_upgrade_txholds(tx, szp); 3545 zfs_sa_upgrade_txholds(tx, tdzp); 3546 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT); 3547 if (error) { 3548 zfs_dirent_unlock(dl); 3549 if (error == ERESTART) { 3550 waited = B_TRUE; 3551 dmu_tx_wait(tx); 3552 dmu_tx_abort(tx); 3553 goto top; 3554 } 3555 dmu_tx_abort(tx); 3556 zfs_exit(zfsvfs, FTAG); 3557 return (error); 3558 } 3559 /* unmark z_unlinked so zfs_link_create will not reject */ 3560 if (is_tmpfile) 3561 szp->z_unlinked = B_FALSE; 3562 error = zfs_link_create(dl, szp, tx, 0); 3563 3564 if (error == 0) { 3565 uint64_t txtype = TX_LINK; 3566 /* 3567 * tmpfile is created to be in z_unlinkedobj, so remove it. 3568 * Also, we don't log in ZIL, because all previous file 3569 * operation on the tmpfile are ignored by ZIL. Instead we 3570 * always wait for txg to sync to make sure all previous 3571 * operation are sync safe. 3572 */ 3573 if (is_tmpfile) { 3574 VERIFY(zap_remove_int(zfsvfs->z_os, 3575 zfsvfs->z_unlinkedobj, szp->z_id, tx) == 0); 3576 } else { 3577 if (flags & FIGNORECASE) 3578 txtype |= TX_CI; 3579 zfs_log_link(zilog, tx, txtype, tdzp, szp, name); 3580 } 3581 } else if (is_tmpfile) { 3582 /* restore z_unlinked since when linking failed */ 3583 szp->z_unlinked = B_TRUE; 3584 } 3585 txg = dmu_tx_get_txg(tx); 3586 dmu_tx_commit(tx); 3587 3588 zfs_dirent_unlock(dl); 3589 3590 if (!is_tmpfile && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3591 zil_commit(zilog, 0); 3592 3593 if (is_tmpfile && zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) 3594 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), txg); 3595 3596 zfs_znode_update_vfs(tdzp); 3597 zfs_znode_update_vfs(szp); 3598 zfs_exit(zfsvfs, FTAG); 3599 return (error); 3600 } 3601 3602 static void 3603 zfs_putpage_sync_commit_cb(void *arg) 3604 { 3605 struct page *pp = arg; 3606 3607 ClearPageError(pp); 3608 end_page_writeback(pp); 3609 } 3610 3611 static void 3612 zfs_putpage_async_commit_cb(void *arg) 3613 { 3614 struct page *pp = arg; 3615 znode_t *zp = ITOZ(pp->mapping->host); 3616 3617 ClearPageError(pp); 3618 end_page_writeback(pp); 3619 atomic_dec_32(&zp->z_async_writes_cnt); 3620 } 3621 3622 /* 3623 * Push a page out to disk, once the page is on stable storage the 3624 * registered commit callback will be run as notification of completion. 3625 * 3626 * IN: ip - page mapped for inode. 3627 * pp - page to push (page is locked) 3628 * wbc - writeback control data 3629 * for_sync - does the caller intend to wait synchronously for the 3630 * page writeback to complete? 3631 * 3632 * RETURN: 0 if success 3633 * error code if failure 3634 * 3635 * Timestamps: 3636 * ip - ctime|mtime updated 3637 */ 3638 int 3639 zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc, 3640 boolean_t for_sync) 3641 { 3642 znode_t *zp = ITOZ(ip); 3643 zfsvfs_t *zfsvfs = ITOZSB(ip); 3644 loff_t offset; 3645 loff_t pgoff; 3646 unsigned int pglen; 3647 dmu_tx_t *tx; 3648 caddr_t va; 3649 int err = 0; 3650 uint64_t mtime[2], ctime[2]; 3651 sa_bulk_attr_t bulk[3]; 3652 int cnt = 0; 3653 struct address_space *mapping; 3654 3655 if ((err = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 3656 return (err); 3657 3658 ASSERT(PageLocked(pp)); 3659 3660 pgoff = page_offset(pp); /* Page byte-offset in file */ 3661 offset = i_size_read(ip); /* File length in bytes */ 3662 pglen = MIN(PAGE_SIZE, /* Page length in bytes */ 3663 P2ROUNDUP(offset, PAGE_SIZE)-pgoff); 3664 3665 /* Page is beyond end of file */ 3666 if (pgoff >= offset) { 3667 unlock_page(pp); 3668 zfs_exit(zfsvfs, FTAG); 3669 return (0); 3670 } 3671 3672 /* Truncate page length to end of file */ 3673 if (pgoff + pglen > offset) 3674 pglen = offset - pgoff; 3675 3676 #if 0 3677 /* 3678 * FIXME: Allow mmap writes past its quota. The correct fix 3679 * is to register a page_mkwrite() handler to count the page 3680 * against its quota when it is about to be dirtied. 3681 */ 3682 if (zfs_id_overblockquota(zfsvfs, DMU_USERUSED_OBJECT, 3683 KUID_TO_SUID(ip->i_uid)) || 3684 zfs_id_overblockquota(zfsvfs, DMU_GROUPUSED_OBJECT, 3685 KGID_TO_SGID(ip->i_gid)) || 3686 (zp->z_projid != ZFS_DEFAULT_PROJID && 3687 zfs_id_overblockquota(zfsvfs, DMU_PROJECTUSED_OBJECT, 3688 zp->z_projid))) { 3689 err = EDQUOT; 3690 } 3691 #endif 3692 3693 /* 3694 * The ordering here is critical and must adhere to the following 3695 * rules in order to avoid deadlocking in either zfs_read() or 3696 * zfs_free_range() due to a lock inversion. 3697 * 3698 * 1) The page must be unlocked prior to acquiring the range lock. 3699 * This is critical because zfs_read() calls find_lock_page() 3700 * which may block on the page lock while holding the range lock. 3701 * 3702 * 2) Before setting or clearing write back on a page the range lock 3703 * must be held in order to prevent a lock inversion with the 3704 * zfs_free_range() function. 3705 * 3706 * This presents a problem because upon entering this function the 3707 * page lock is already held. To safely acquire the range lock the 3708 * page lock must be dropped. This creates a window where another 3709 * process could truncate, invalidate, dirty, or write out the page. 3710 * 3711 * Therefore, after successfully reacquiring the range and page locks 3712 * the current page state is checked. In the common case everything 3713 * will be as is expected and it can be written out. However, if 3714 * the page state has changed it must be handled accordingly. 3715 */ 3716 mapping = pp->mapping; 3717 redirty_page_for_writepage(wbc, pp); 3718 unlock_page(pp); 3719 3720 zfs_locked_range_t *lr = zfs_rangelock_enter(&zp->z_rangelock, 3721 pgoff, pglen, RL_WRITER); 3722 lock_page(pp); 3723 3724 /* Page mapping changed or it was no longer dirty, we're done */ 3725 if (unlikely((mapping != pp->mapping) || !PageDirty(pp))) { 3726 unlock_page(pp); 3727 zfs_rangelock_exit(lr); 3728 zfs_exit(zfsvfs, FTAG); 3729 return (0); 3730 } 3731 3732 /* Another process started write block if required */ 3733 if (PageWriteback(pp)) { 3734 unlock_page(pp); 3735 zfs_rangelock_exit(lr); 3736 3737 if (wbc->sync_mode != WB_SYNC_NONE) { 3738 /* 3739 * Speed up any non-sync page writebacks since 3740 * they may take several seconds to complete. 3741 * Refer to the comment in zpl_fsync() (when 3742 * HAVE_FSYNC_RANGE is defined) for details. 3743 */ 3744 if (atomic_load_32(&zp->z_async_writes_cnt) > 0) { 3745 zil_commit(zfsvfs->z_log, zp->z_id); 3746 } 3747 3748 if (PageWriteback(pp)) 3749 #ifdef HAVE_PAGEMAP_FOLIO_WAIT_BIT 3750 folio_wait_bit(page_folio(pp), PG_writeback); 3751 #else 3752 wait_on_page_bit(pp, PG_writeback); 3753 #endif 3754 } 3755 3756 zfs_exit(zfsvfs, FTAG); 3757 return (0); 3758 } 3759 3760 /* Clear the dirty flag the required locks are held */ 3761 if (!clear_page_dirty_for_io(pp)) { 3762 unlock_page(pp); 3763 zfs_rangelock_exit(lr); 3764 zfs_exit(zfsvfs, FTAG); 3765 return (0); 3766 } 3767 3768 /* 3769 * Counterpart for redirty_page_for_writepage() above. This page 3770 * was in fact not skipped and should not be counted as if it were. 3771 */ 3772 wbc->pages_skipped--; 3773 if (!for_sync) 3774 atomic_inc_32(&zp->z_async_writes_cnt); 3775 set_page_writeback(pp); 3776 unlock_page(pp); 3777 3778 tx = dmu_tx_create(zfsvfs->z_os); 3779 dmu_tx_hold_write(tx, zp->z_id, pgoff, pglen); 3780 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 3781 zfs_sa_upgrade_txholds(tx, zp); 3782 3783 err = dmu_tx_assign(tx, TXG_NOWAIT); 3784 if (err != 0) { 3785 if (err == ERESTART) 3786 dmu_tx_wait(tx); 3787 3788 dmu_tx_abort(tx); 3789 #ifdef HAVE_VFS_FILEMAP_DIRTY_FOLIO 3790 filemap_dirty_folio(page_mapping(pp), page_folio(pp)); 3791 #else 3792 __set_page_dirty_nobuffers(pp); 3793 #endif 3794 ClearPageError(pp); 3795 end_page_writeback(pp); 3796 if (!for_sync) 3797 atomic_dec_32(&zp->z_async_writes_cnt); 3798 zfs_rangelock_exit(lr); 3799 zfs_exit(zfsvfs, FTAG); 3800 return (err); 3801 } 3802 3803 va = kmap(pp); 3804 ASSERT3U(pglen, <=, PAGE_SIZE); 3805 dmu_write(zfsvfs->z_os, zp->z_id, pgoff, pglen, va, tx); 3806 kunmap(pp); 3807 3808 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 3809 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 3810 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(zfsvfs), NULL, 3811 &zp->z_pflags, 8); 3812 3813 /* Preserve the mtime and ctime provided by the inode */ 3814 ZFS_TIME_ENCODE(&ip->i_mtime, mtime); 3815 ZFS_TIME_ENCODE(&ip->i_ctime, ctime); 3816 zp->z_atime_dirty = B_FALSE; 3817 zp->z_seq++; 3818 3819 err = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx); 3820 3821 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, pgoff, pglen, 0, 3822 for_sync ? zfs_putpage_sync_commit_cb : 3823 zfs_putpage_async_commit_cb, pp); 3824 3825 dmu_tx_commit(tx); 3826 3827 zfs_rangelock_exit(lr); 3828 3829 if (wbc->sync_mode != WB_SYNC_NONE) { 3830 /* 3831 * Note that this is rarely called under writepages(), because 3832 * writepages() normally handles the entire commit for 3833 * performance reasons. 3834 */ 3835 zil_commit(zfsvfs->z_log, zp->z_id); 3836 } else if (!for_sync && atomic_load_32(&zp->z_sync_writes_cnt) > 0) { 3837 /* 3838 * If the caller does not intend to wait synchronously 3839 * for this page writeback to complete and there are active 3840 * synchronous calls on this file, do a commit so that 3841 * the latter don't accidentally end up waiting for 3842 * our writeback to complete. Refer to the comment in 3843 * zpl_fsync() (when HAVE_FSYNC_RANGE is defined) for details. 3844 */ 3845 zil_commit(zfsvfs->z_log, zp->z_id); 3846 } 3847 3848 dataset_kstats_update_write_kstats(&zfsvfs->z_kstat, pglen); 3849 3850 zfs_exit(zfsvfs, FTAG); 3851 return (err); 3852 } 3853 3854 /* 3855 * Update the system attributes when the inode has been dirtied. For the 3856 * moment we only update the mode, atime, mtime, and ctime. 3857 */ 3858 int 3859 zfs_dirty_inode(struct inode *ip, int flags) 3860 { 3861 znode_t *zp = ITOZ(ip); 3862 zfsvfs_t *zfsvfs = ITOZSB(ip); 3863 dmu_tx_t *tx; 3864 uint64_t mode, atime[2], mtime[2], ctime[2]; 3865 sa_bulk_attr_t bulk[4]; 3866 int error = 0; 3867 int cnt = 0; 3868 3869 if (zfs_is_readonly(zfsvfs) || dmu_objset_is_snapshot(zfsvfs->z_os)) 3870 return (0); 3871 3872 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 3873 return (error); 3874 3875 #ifdef I_DIRTY_TIME 3876 /* 3877 * This is the lazytime semantic introduced in Linux 4.0 3878 * This flag will only be called from update_time when lazytime is set. 3879 * (Note, I_DIRTY_SYNC will also set if not lazytime) 3880 * Fortunately mtime and ctime are managed within ZFS itself, so we 3881 * only need to dirty atime. 3882 */ 3883 if (flags == I_DIRTY_TIME) { 3884 zp->z_atime_dirty = B_TRUE; 3885 goto out; 3886 } 3887 #endif 3888 3889 tx = dmu_tx_create(zfsvfs->z_os); 3890 3891 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 3892 zfs_sa_upgrade_txholds(tx, zp); 3893 3894 error = dmu_tx_assign(tx, TXG_WAIT); 3895 if (error) { 3896 dmu_tx_abort(tx); 3897 goto out; 3898 } 3899 3900 mutex_enter(&zp->z_lock); 3901 zp->z_atime_dirty = B_FALSE; 3902 3903 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8); 3904 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(zfsvfs), NULL, &atime, 16); 3905 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 3906 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 3907 3908 /* Preserve the mode, mtime and ctime provided by the inode */ 3909 ZFS_TIME_ENCODE(&ip->i_atime, atime); 3910 ZFS_TIME_ENCODE(&ip->i_mtime, mtime); 3911 ZFS_TIME_ENCODE(&ip->i_ctime, ctime); 3912 mode = ip->i_mode; 3913 3914 zp->z_mode = mode; 3915 3916 error = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx); 3917 mutex_exit(&zp->z_lock); 3918 3919 dmu_tx_commit(tx); 3920 out: 3921 zfs_exit(zfsvfs, FTAG); 3922 return (error); 3923 } 3924 3925 void 3926 zfs_inactive(struct inode *ip) 3927 { 3928 znode_t *zp = ITOZ(ip); 3929 zfsvfs_t *zfsvfs = ITOZSB(ip); 3930 uint64_t atime[2]; 3931 int error; 3932 int need_unlock = 0; 3933 3934 /* Only read lock if we haven't already write locked, e.g. rollback */ 3935 if (!RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)) { 3936 need_unlock = 1; 3937 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 3938 } 3939 if (zp->z_sa_hdl == NULL) { 3940 if (need_unlock) 3941 rw_exit(&zfsvfs->z_teardown_inactive_lock); 3942 return; 3943 } 3944 3945 if (zp->z_atime_dirty && zp->z_unlinked == B_FALSE) { 3946 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 3947 3948 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 3949 zfs_sa_upgrade_txholds(tx, zp); 3950 error = dmu_tx_assign(tx, TXG_WAIT); 3951 if (error) { 3952 dmu_tx_abort(tx); 3953 } else { 3954 ZFS_TIME_ENCODE(&ip->i_atime, atime); 3955 mutex_enter(&zp->z_lock); 3956 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs), 3957 (void *)&atime, sizeof (atime), tx); 3958 zp->z_atime_dirty = B_FALSE; 3959 mutex_exit(&zp->z_lock); 3960 dmu_tx_commit(tx); 3961 } 3962 } 3963 3964 zfs_zinactive(zp); 3965 if (need_unlock) 3966 rw_exit(&zfsvfs->z_teardown_inactive_lock); 3967 } 3968 3969 /* 3970 * Fill pages with data from the disk. 3971 */ 3972 static int 3973 zfs_fillpage(struct inode *ip, struct page *pp) 3974 { 3975 zfsvfs_t *zfsvfs = ITOZSB(ip); 3976 loff_t i_size = i_size_read(ip); 3977 u_offset_t io_off = page_offset(pp); 3978 size_t io_len = PAGE_SIZE; 3979 3980 ASSERT3U(io_off, <, i_size); 3981 3982 if (io_off + io_len > i_size) 3983 io_len = i_size - io_off; 3984 3985 void *va = kmap(pp); 3986 int error = dmu_read(zfsvfs->z_os, ITOZ(ip)->z_id, io_off, 3987 io_len, va, DMU_READ_PREFETCH); 3988 if (io_len != PAGE_SIZE) 3989 memset((char *)va + io_len, 0, PAGE_SIZE - io_len); 3990 kunmap(pp); 3991 3992 if (error) { 3993 /* convert checksum errors into IO errors */ 3994 if (error == ECKSUM) 3995 error = SET_ERROR(EIO); 3996 3997 SetPageError(pp); 3998 ClearPageUptodate(pp); 3999 } else { 4000 ClearPageError(pp); 4001 SetPageUptodate(pp); 4002 } 4003 4004 return (error); 4005 } 4006 4007 /* 4008 * Uses zfs_fillpage to read data from the file and fill the page. 4009 * 4010 * IN: ip - inode of file to get data from. 4011 * pp - page to read 4012 * 4013 * RETURN: 0 on success, error code on failure. 4014 * 4015 * Timestamps: 4016 * vp - atime updated 4017 */ 4018 int 4019 zfs_getpage(struct inode *ip, struct page *pp) 4020 { 4021 zfsvfs_t *zfsvfs = ITOZSB(ip); 4022 znode_t *zp = ITOZ(ip); 4023 int error; 4024 4025 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 4026 return (error); 4027 4028 error = zfs_fillpage(ip, pp); 4029 if (error == 0) 4030 dataset_kstats_update_read_kstats(&zfsvfs->z_kstat, PAGE_SIZE); 4031 4032 zfs_exit(zfsvfs, FTAG); 4033 4034 return (error); 4035 } 4036 4037 /* 4038 * Check ZFS specific permissions to memory map a section of a file. 4039 * 4040 * IN: ip - inode of the file to mmap 4041 * off - file offset 4042 * addrp - start address in memory region 4043 * len - length of memory region 4044 * vm_flags- address flags 4045 * 4046 * RETURN: 0 if success 4047 * error code if failure 4048 */ 4049 int 4050 zfs_map(struct inode *ip, offset_t off, caddr_t *addrp, size_t len, 4051 unsigned long vm_flags) 4052 { 4053 (void) addrp; 4054 znode_t *zp = ITOZ(ip); 4055 zfsvfs_t *zfsvfs = ITOZSB(ip); 4056 int error; 4057 4058 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 4059 return (error); 4060 4061 if ((vm_flags & VM_WRITE) && (zp->z_pflags & 4062 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) { 4063 zfs_exit(zfsvfs, FTAG); 4064 return (SET_ERROR(EPERM)); 4065 } 4066 4067 if ((vm_flags & (VM_READ | VM_EXEC)) && 4068 (zp->z_pflags & ZFS_AV_QUARANTINED)) { 4069 zfs_exit(zfsvfs, FTAG); 4070 return (SET_ERROR(EACCES)); 4071 } 4072 4073 if (off < 0 || len > MAXOFFSET_T - off) { 4074 zfs_exit(zfsvfs, FTAG); 4075 return (SET_ERROR(ENXIO)); 4076 } 4077 4078 zfs_exit(zfsvfs, FTAG); 4079 return (0); 4080 } 4081 4082 /* 4083 * Free or allocate space in a file. Currently, this function only 4084 * supports the `F_FREESP' command. However, this command is somewhat 4085 * misnamed, as its functionality includes the ability to allocate as 4086 * well as free space. 4087 * 4088 * IN: zp - znode of file to free data in. 4089 * cmd - action to take (only F_FREESP supported). 4090 * bfp - section of file to free/alloc. 4091 * flag - current file open mode flags. 4092 * offset - current file offset. 4093 * cr - credentials of caller. 4094 * 4095 * RETURN: 0 on success, error code on failure. 4096 * 4097 * Timestamps: 4098 * zp - ctime|mtime updated 4099 */ 4100 int 4101 zfs_space(znode_t *zp, int cmd, flock64_t *bfp, int flag, 4102 offset_t offset, cred_t *cr) 4103 { 4104 (void) offset; 4105 zfsvfs_t *zfsvfs = ZTOZSB(zp); 4106 uint64_t off, len; 4107 int error; 4108 4109 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 4110 return (error); 4111 4112 if (cmd != F_FREESP) { 4113 zfs_exit(zfsvfs, FTAG); 4114 return (SET_ERROR(EINVAL)); 4115 } 4116 4117 /* 4118 * Callers might not be able to detect properly that we are read-only, 4119 * so check it explicitly here. 4120 */ 4121 if (zfs_is_readonly(zfsvfs)) { 4122 zfs_exit(zfsvfs, FTAG); 4123 return (SET_ERROR(EROFS)); 4124 } 4125 4126 if (bfp->l_len < 0) { 4127 zfs_exit(zfsvfs, FTAG); 4128 return (SET_ERROR(EINVAL)); 4129 } 4130 4131 /* 4132 * Permissions aren't checked on Solaris because on this OS 4133 * zfs_space() can only be called with an opened file handle. 4134 * On Linux we can get here through truncate_range() which 4135 * operates directly on inodes, so we need to check access rights. 4136 */ 4137 if ((error = zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr, 4138 zfs_init_idmap))) { 4139 zfs_exit(zfsvfs, FTAG); 4140 return (error); 4141 } 4142 4143 off = bfp->l_start; 4144 len = bfp->l_len; /* 0 means from off to end of file */ 4145 4146 error = zfs_freesp(zp, off, len, flag, TRUE); 4147 4148 zfs_exit(zfsvfs, FTAG); 4149 return (error); 4150 } 4151 4152 int 4153 zfs_fid(struct inode *ip, fid_t *fidp) 4154 { 4155 znode_t *zp = ITOZ(ip); 4156 zfsvfs_t *zfsvfs = ITOZSB(ip); 4157 uint32_t gen; 4158 uint64_t gen64; 4159 uint64_t object = zp->z_id; 4160 zfid_short_t *zfid; 4161 int size, i, error; 4162 4163 if ((error = zfs_enter(zfsvfs, FTAG)) != 0) 4164 return (error); 4165 4166 if (fidp->fid_len < SHORT_FID_LEN) { 4167 fidp->fid_len = SHORT_FID_LEN; 4168 zfs_exit(zfsvfs, FTAG); 4169 return (SET_ERROR(ENOSPC)); 4170 } 4171 4172 if ((error = zfs_verify_zp(zp)) != 0) { 4173 zfs_exit(zfsvfs, FTAG); 4174 return (error); 4175 } 4176 4177 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), 4178 &gen64, sizeof (uint64_t))) != 0) { 4179 zfs_exit(zfsvfs, FTAG); 4180 return (error); 4181 } 4182 4183 gen = (uint32_t)gen64; 4184 4185 size = SHORT_FID_LEN; 4186 4187 zfid = (zfid_short_t *)fidp; 4188 4189 zfid->zf_len = size; 4190 4191 for (i = 0; i < sizeof (zfid->zf_object); i++) 4192 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 4193 4194 /* Must have a non-zero generation number to distinguish from .zfs */ 4195 if (gen == 0) 4196 gen = 1; 4197 for (i = 0; i < sizeof (zfid->zf_gen); i++) 4198 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 4199 4200 zfs_exit(zfsvfs, FTAG); 4201 return (0); 4202 } 4203 4204 #if defined(_KERNEL) 4205 EXPORT_SYMBOL(zfs_open); 4206 EXPORT_SYMBOL(zfs_close); 4207 EXPORT_SYMBOL(zfs_lookup); 4208 EXPORT_SYMBOL(zfs_create); 4209 EXPORT_SYMBOL(zfs_tmpfile); 4210 EXPORT_SYMBOL(zfs_remove); 4211 EXPORT_SYMBOL(zfs_mkdir); 4212 EXPORT_SYMBOL(zfs_rmdir); 4213 EXPORT_SYMBOL(zfs_readdir); 4214 EXPORT_SYMBOL(zfs_getattr_fast); 4215 EXPORT_SYMBOL(zfs_setattr); 4216 EXPORT_SYMBOL(zfs_rename); 4217 EXPORT_SYMBOL(zfs_symlink); 4218 EXPORT_SYMBOL(zfs_readlink); 4219 EXPORT_SYMBOL(zfs_link); 4220 EXPORT_SYMBOL(zfs_inactive); 4221 EXPORT_SYMBOL(zfs_space); 4222 EXPORT_SYMBOL(zfs_fid); 4223 EXPORT_SYMBOL(zfs_getpage); 4224 EXPORT_SYMBOL(zfs_putpage); 4225 EXPORT_SYMBOL(zfs_dirty_inode); 4226 EXPORT_SYMBOL(zfs_map); 4227 4228 /* CSTYLED */ 4229 module_param(zfs_delete_blocks, ulong, 0644); 4230 MODULE_PARM_DESC(zfs_delete_blocks, "Delete files larger than N blocks async"); 4231 4232 #endif 4233