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, kcred->user_ns))) { 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 kcred->user_ns))) { 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 zuserns_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 zuserns_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, kcred->user_ns))) { 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, zuserns_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, kcred->user_ns))) { 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(struct user_namespace *user_ns, struct inode *ip, 1656 struct kstat *sp) 1657 { 1658 znode_t *zp = ITOZ(ip); 1659 zfsvfs_t *zfsvfs = ITOZSB(ip); 1660 uint32_t blksize; 1661 u_longlong_t nblocks; 1662 int error; 1663 1664 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 1665 return (error); 1666 1667 mutex_enter(&zp->z_lock); 1668 1669 zpl_generic_fillattr(user_ns, ip, sp); 1670 /* 1671 * +1 link count for root inode with visible '.zfs' directory. 1672 */ 1673 if ((zp->z_id == zfsvfs->z_root) && zfs_show_ctldir(zp)) 1674 if (sp->nlink < ZFS_LINK_MAX) 1675 sp->nlink++; 1676 1677 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks); 1678 sp->blksize = blksize; 1679 sp->blocks = nblocks; 1680 1681 if (unlikely(zp->z_blksz == 0)) { 1682 /* 1683 * Block size hasn't been set; suggest maximal I/O transfers. 1684 */ 1685 sp->blksize = zfsvfs->z_max_blksz; 1686 } 1687 1688 mutex_exit(&zp->z_lock); 1689 1690 /* 1691 * Required to prevent NFS client from detecting different inode 1692 * numbers of snapshot root dentry before and after snapshot mount. 1693 */ 1694 if (zfsvfs->z_issnap) { 1695 if (ip->i_sb->s_root->d_inode == ip) 1696 sp->ino = ZFSCTL_INO_SNAPDIRS - 1697 dmu_objset_id(zfsvfs->z_os); 1698 } 1699 1700 zfs_exit(zfsvfs, FTAG); 1701 1702 return (0); 1703 } 1704 1705 /* 1706 * For the operation of changing file's user/group/project, we need to 1707 * handle not only the main object that is assigned to the file directly, 1708 * but also the ones that are used by the file via hidden xattr directory. 1709 * 1710 * Because the xattr directory may contains many EA entries, as to it may 1711 * be impossible to change all of them via the transaction of changing the 1712 * main object's user/group/project attributes. Then we have to change them 1713 * via other multiple independent transactions one by one. It may be not good 1714 * solution, but we have no better idea yet. 1715 */ 1716 static int 1717 zfs_setattr_dir(znode_t *dzp) 1718 { 1719 struct inode *dxip = ZTOI(dzp); 1720 struct inode *xip = NULL; 1721 zfsvfs_t *zfsvfs = ZTOZSB(dzp); 1722 objset_t *os = zfsvfs->z_os; 1723 zap_cursor_t zc; 1724 zap_attribute_t zap; 1725 zfs_dirlock_t *dl; 1726 znode_t *zp = NULL; 1727 dmu_tx_t *tx = NULL; 1728 uint64_t uid, gid; 1729 sa_bulk_attr_t bulk[4]; 1730 int count; 1731 int err; 1732 1733 zap_cursor_init(&zc, os, dzp->z_id); 1734 while ((err = zap_cursor_retrieve(&zc, &zap)) == 0) { 1735 count = 0; 1736 if (zap.za_integer_length != 8 || zap.za_num_integers != 1) { 1737 err = ENXIO; 1738 break; 1739 } 1740 1741 err = zfs_dirent_lock(&dl, dzp, (char *)zap.za_name, &zp, 1742 ZEXISTS, NULL, NULL); 1743 if (err == ENOENT) 1744 goto next; 1745 if (err) 1746 break; 1747 1748 xip = ZTOI(zp); 1749 if (KUID_TO_SUID(xip->i_uid) == KUID_TO_SUID(dxip->i_uid) && 1750 KGID_TO_SGID(xip->i_gid) == KGID_TO_SGID(dxip->i_gid) && 1751 zp->z_projid == dzp->z_projid) 1752 goto next; 1753 1754 tx = dmu_tx_create(os); 1755 if (!(zp->z_pflags & ZFS_PROJID)) 1756 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 1757 else 1758 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 1759 1760 err = dmu_tx_assign(tx, TXG_WAIT); 1761 if (err) 1762 break; 1763 1764 mutex_enter(&dzp->z_lock); 1765 1766 if (KUID_TO_SUID(xip->i_uid) != KUID_TO_SUID(dxip->i_uid)) { 1767 xip->i_uid = dxip->i_uid; 1768 uid = zfs_uid_read(dxip); 1769 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 1770 &uid, sizeof (uid)); 1771 } 1772 1773 if (KGID_TO_SGID(xip->i_gid) != KGID_TO_SGID(dxip->i_gid)) { 1774 xip->i_gid = dxip->i_gid; 1775 gid = zfs_gid_read(dxip); 1776 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), NULL, 1777 &gid, sizeof (gid)); 1778 } 1779 1780 if (zp->z_projid != dzp->z_projid) { 1781 if (!(zp->z_pflags & ZFS_PROJID)) { 1782 zp->z_pflags |= ZFS_PROJID; 1783 SA_ADD_BULK_ATTR(bulk, count, 1784 SA_ZPL_FLAGS(zfsvfs), NULL, &zp->z_pflags, 1785 sizeof (zp->z_pflags)); 1786 } 1787 1788 zp->z_projid = dzp->z_projid; 1789 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PROJID(zfsvfs), 1790 NULL, &zp->z_projid, sizeof (zp->z_projid)); 1791 } 1792 1793 mutex_exit(&dzp->z_lock); 1794 1795 if (likely(count > 0)) { 1796 err = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 1797 dmu_tx_commit(tx); 1798 } else { 1799 dmu_tx_abort(tx); 1800 } 1801 tx = NULL; 1802 if (err != 0 && err != ENOENT) 1803 break; 1804 1805 next: 1806 if (zp) { 1807 zrele(zp); 1808 zp = NULL; 1809 zfs_dirent_unlock(dl); 1810 } 1811 zap_cursor_advance(&zc); 1812 } 1813 1814 if (tx) 1815 dmu_tx_abort(tx); 1816 if (zp) { 1817 zrele(zp); 1818 zfs_dirent_unlock(dl); 1819 } 1820 zap_cursor_fini(&zc); 1821 1822 return (err == ENOENT ? 0 : err); 1823 } 1824 1825 /* 1826 * Set the file attributes to the values contained in the 1827 * vattr structure. 1828 * 1829 * IN: zp - znode of file to be modified. 1830 * vap - new attribute values. 1831 * If ATTR_XVATTR set, then optional attrs are being set 1832 * flags - ATTR_UTIME set if non-default time values provided. 1833 * - ATTR_NOACLCHECK (CIFS context only). 1834 * cr - credentials of caller. 1835 * mnt_ns - user namespace of the mount 1836 * 1837 * RETURN: 0 if success 1838 * error code if failure 1839 * 1840 * Timestamps: 1841 * ip - ctime updated, mtime updated if size changed. 1842 */ 1843 int 1844 zfs_setattr(znode_t *zp, vattr_t *vap, int flags, cred_t *cr, zuserns_t *mnt_ns) 1845 { 1846 struct inode *ip; 1847 zfsvfs_t *zfsvfs = ZTOZSB(zp); 1848 objset_t *os = zfsvfs->z_os; 1849 zilog_t *zilog; 1850 dmu_tx_t *tx; 1851 vattr_t oldva; 1852 xvattr_t *tmpxvattr; 1853 uint_t mask = vap->va_mask; 1854 uint_t saved_mask = 0; 1855 int trim_mask = 0; 1856 uint64_t new_mode; 1857 uint64_t new_kuid = 0, new_kgid = 0, new_uid, new_gid; 1858 uint64_t xattr_obj; 1859 uint64_t mtime[2], ctime[2], atime[2]; 1860 uint64_t projid = ZFS_INVALID_PROJID; 1861 znode_t *attrzp; 1862 int need_policy = FALSE; 1863 int err, err2 = 0; 1864 zfs_fuid_info_t *fuidp = NULL; 1865 xvattr_t *xvap = (xvattr_t *)vap; /* vap may be an xvattr_t * */ 1866 xoptattr_t *xoap; 1867 zfs_acl_t *aclp; 1868 boolean_t skipaclchk = (flags & ATTR_NOACLCHECK) ? B_TRUE : B_FALSE; 1869 boolean_t fuid_dirtied = B_FALSE; 1870 boolean_t handle_eadir = B_FALSE; 1871 sa_bulk_attr_t *bulk, *xattr_bulk; 1872 int count = 0, xattr_count = 0, bulks = 8; 1873 1874 if (mask == 0) 1875 return (0); 1876 1877 if ((err = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 1878 return (err); 1879 ip = ZTOI(zp); 1880 1881 /* 1882 * If this is a xvattr_t, then get a pointer to the structure of 1883 * optional attributes. If this is NULL, then we have a vattr_t. 1884 */ 1885 xoap = xva_getxoptattr(xvap); 1886 if (xoap != NULL && (mask & ATTR_XVATTR)) { 1887 if (XVA_ISSET_REQ(xvap, XAT_PROJID)) { 1888 if (!dmu_objset_projectquota_enabled(os) || 1889 (!S_ISREG(ip->i_mode) && !S_ISDIR(ip->i_mode))) { 1890 zfs_exit(zfsvfs, FTAG); 1891 return (SET_ERROR(ENOTSUP)); 1892 } 1893 1894 projid = xoap->xoa_projid; 1895 if (unlikely(projid == ZFS_INVALID_PROJID)) { 1896 zfs_exit(zfsvfs, FTAG); 1897 return (SET_ERROR(EINVAL)); 1898 } 1899 1900 if (projid == zp->z_projid && zp->z_pflags & ZFS_PROJID) 1901 projid = ZFS_INVALID_PROJID; 1902 else 1903 need_policy = TRUE; 1904 } 1905 1906 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT) && 1907 (xoap->xoa_projinherit != 1908 ((zp->z_pflags & ZFS_PROJINHERIT) != 0)) && 1909 (!dmu_objset_projectquota_enabled(os) || 1910 (!S_ISREG(ip->i_mode) && !S_ISDIR(ip->i_mode)))) { 1911 zfs_exit(zfsvfs, FTAG); 1912 return (SET_ERROR(ENOTSUP)); 1913 } 1914 } 1915 1916 zilog = zfsvfs->z_log; 1917 1918 /* 1919 * Make sure that if we have ephemeral uid/gid or xvattr specified 1920 * that file system is at proper version level 1921 */ 1922 1923 if (zfsvfs->z_use_fuids == B_FALSE && 1924 (((mask & ATTR_UID) && IS_EPHEMERAL(vap->va_uid)) || 1925 ((mask & ATTR_GID) && IS_EPHEMERAL(vap->va_gid)) || 1926 (mask & ATTR_XVATTR))) { 1927 zfs_exit(zfsvfs, FTAG); 1928 return (SET_ERROR(EINVAL)); 1929 } 1930 1931 if (mask & ATTR_SIZE && S_ISDIR(ip->i_mode)) { 1932 zfs_exit(zfsvfs, FTAG); 1933 return (SET_ERROR(EISDIR)); 1934 } 1935 1936 if (mask & ATTR_SIZE && !S_ISREG(ip->i_mode) && !S_ISFIFO(ip->i_mode)) { 1937 zfs_exit(zfsvfs, FTAG); 1938 return (SET_ERROR(EINVAL)); 1939 } 1940 1941 tmpxvattr = kmem_alloc(sizeof (xvattr_t), KM_SLEEP); 1942 xva_init(tmpxvattr); 1943 1944 bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * bulks, KM_SLEEP); 1945 xattr_bulk = kmem_alloc(sizeof (sa_bulk_attr_t) * bulks, KM_SLEEP); 1946 1947 /* 1948 * Immutable files can only alter immutable bit and atime 1949 */ 1950 if ((zp->z_pflags & ZFS_IMMUTABLE) && 1951 ((mask & (ATTR_SIZE|ATTR_UID|ATTR_GID|ATTR_MTIME|ATTR_MODE)) || 1952 ((mask & ATTR_XVATTR) && XVA_ISSET_REQ(xvap, XAT_CREATETIME)))) { 1953 err = SET_ERROR(EPERM); 1954 goto out3; 1955 } 1956 1957 if ((mask & ATTR_SIZE) && (zp->z_pflags & ZFS_READONLY)) { 1958 err = SET_ERROR(EPERM); 1959 goto out3; 1960 } 1961 1962 /* 1963 * Verify timestamps doesn't overflow 32 bits. 1964 * ZFS can handle large timestamps, but 32bit syscalls can't 1965 * handle times greater than 2039. This check should be removed 1966 * once large timestamps are fully supported. 1967 */ 1968 if (mask & (ATTR_ATIME | ATTR_MTIME)) { 1969 if (((mask & ATTR_ATIME) && 1970 TIMESPEC_OVERFLOW(&vap->va_atime)) || 1971 ((mask & ATTR_MTIME) && 1972 TIMESPEC_OVERFLOW(&vap->va_mtime))) { 1973 err = SET_ERROR(EOVERFLOW); 1974 goto out3; 1975 } 1976 } 1977 1978 top: 1979 attrzp = NULL; 1980 aclp = NULL; 1981 1982 /* Can this be moved to before the top label? */ 1983 if (zfs_is_readonly(zfsvfs)) { 1984 err = SET_ERROR(EROFS); 1985 goto out3; 1986 } 1987 1988 /* 1989 * First validate permissions 1990 */ 1991 1992 if (mask & ATTR_SIZE) { 1993 err = zfs_zaccess(zp, ACE_WRITE_DATA, 0, skipaclchk, cr, 1994 mnt_ns); 1995 if (err) 1996 goto out3; 1997 1998 /* 1999 * XXX - Note, we are not providing any open 2000 * mode flags here (like FNDELAY), so we may 2001 * block if there are locks present... this 2002 * should be addressed in openat(). 2003 */ 2004 /* XXX - would it be OK to generate a log record here? */ 2005 err = zfs_freesp(zp, vap->va_size, 0, 0, FALSE); 2006 if (err) 2007 goto out3; 2008 } 2009 2010 if (mask & (ATTR_ATIME|ATTR_MTIME) || 2011 ((mask & ATTR_XVATTR) && (XVA_ISSET_REQ(xvap, XAT_HIDDEN) || 2012 XVA_ISSET_REQ(xvap, XAT_READONLY) || 2013 XVA_ISSET_REQ(xvap, XAT_ARCHIVE) || 2014 XVA_ISSET_REQ(xvap, XAT_OFFLINE) || 2015 XVA_ISSET_REQ(xvap, XAT_SPARSE) || 2016 XVA_ISSET_REQ(xvap, XAT_CREATETIME) || 2017 XVA_ISSET_REQ(xvap, XAT_SYSTEM)))) { 2018 need_policy = zfs_zaccess(zp, ACE_WRITE_ATTRIBUTES, 0, 2019 skipaclchk, cr, mnt_ns); 2020 } 2021 2022 if (mask & (ATTR_UID|ATTR_GID)) { 2023 int idmask = (mask & (ATTR_UID|ATTR_GID)); 2024 int take_owner; 2025 int take_group; 2026 uid_t uid; 2027 gid_t gid; 2028 2029 /* 2030 * NOTE: even if a new mode is being set, 2031 * we may clear S_ISUID/S_ISGID bits. 2032 */ 2033 2034 if (!(mask & ATTR_MODE)) 2035 vap->va_mode = zp->z_mode; 2036 2037 /* 2038 * Take ownership or chgrp to group we are a member of 2039 */ 2040 2041 uid = zfs_uid_to_vfsuid((struct user_namespace *)mnt_ns, 2042 zfs_i_user_ns(ip), vap->va_uid); 2043 gid = zfs_gid_to_vfsgid((struct user_namespace *)mnt_ns, 2044 zfs_i_user_ns(ip), vap->va_gid); 2045 take_owner = (mask & ATTR_UID) && (uid == crgetuid(cr)); 2046 take_group = (mask & ATTR_GID) && 2047 zfs_groupmember(zfsvfs, gid, cr); 2048 2049 /* 2050 * If both ATTR_UID and ATTR_GID are set then take_owner and 2051 * take_group must both be set in order to allow taking 2052 * ownership. 2053 * 2054 * Otherwise, send the check through secpolicy_vnode_setattr() 2055 * 2056 */ 2057 2058 if (((idmask == (ATTR_UID|ATTR_GID)) && 2059 take_owner && take_group) || 2060 ((idmask == ATTR_UID) && take_owner) || 2061 ((idmask == ATTR_GID) && take_group)) { 2062 if (zfs_zaccess(zp, ACE_WRITE_OWNER, 0, 2063 skipaclchk, cr, mnt_ns) == 0) { 2064 /* 2065 * Remove setuid/setgid for non-privileged users 2066 */ 2067 (void) secpolicy_setid_clear(vap, cr); 2068 trim_mask = (mask & (ATTR_UID|ATTR_GID)); 2069 } else { 2070 need_policy = TRUE; 2071 } 2072 } else { 2073 need_policy = TRUE; 2074 } 2075 } 2076 2077 mutex_enter(&zp->z_lock); 2078 oldva.va_mode = zp->z_mode; 2079 zfs_fuid_map_ids(zp, cr, &oldva.va_uid, &oldva.va_gid); 2080 if (mask & ATTR_XVATTR) { 2081 /* 2082 * Update xvattr mask to include only those attributes 2083 * that are actually changing. 2084 * 2085 * the bits will be restored prior to actually setting 2086 * the attributes so the caller thinks they were set. 2087 */ 2088 if (XVA_ISSET_REQ(xvap, XAT_APPENDONLY)) { 2089 if (xoap->xoa_appendonly != 2090 ((zp->z_pflags & ZFS_APPENDONLY) != 0)) { 2091 need_policy = TRUE; 2092 } else { 2093 XVA_CLR_REQ(xvap, XAT_APPENDONLY); 2094 XVA_SET_REQ(tmpxvattr, XAT_APPENDONLY); 2095 } 2096 } 2097 2098 if (XVA_ISSET_REQ(xvap, XAT_PROJINHERIT)) { 2099 if (xoap->xoa_projinherit != 2100 ((zp->z_pflags & ZFS_PROJINHERIT) != 0)) { 2101 need_policy = TRUE; 2102 } else { 2103 XVA_CLR_REQ(xvap, XAT_PROJINHERIT); 2104 XVA_SET_REQ(tmpxvattr, XAT_PROJINHERIT); 2105 } 2106 } 2107 2108 if (XVA_ISSET_REQ(xvap, XAT_NOUNLINK)) { 2109 if (xoap->xoa_nounlink != 2110 ((zp->z_pflags & ZFS_NOUNLINK) != 0)) { 2111 need_policy = TRUE; 2112 } else { 2113 XVA_CLR_REQ(xvap, XAT_NOUNLINK); 2114 XVA_SET_REQ(tmpxvattr, XAT_NOUNLINK); 2115 } 2116 } 2117 2118 if (XVA_ISSET_REQ(xvap, XAT_IMMUTABLE)) { 2119 if (xoap->xoa_immutable != 2120 ((zp->z_pflags & ZFS_IMMUTABLE) != 0)) { 2121 need_policy = TRUE; 2122 } else { 2123 XVA_CLR_REQ(xvap, XAT_IMMUTABLE); 2124 XVA_SET_REQ(tmpxvattr, XAT_IMMUTABLE); 2125 } 2126 } 2127 2128 if (XVA_ISSET_REQ(xvap, XAT_NODUMP)) { 2129 if (xoap->xoa_nodump != 2130 ((zp->z_pflags & ZFS_NODUMP) != 0)) { 2131 need_policy = TRUE; 2132 } else { 2133 XVA_CLR_REQ(xvap, XAT_NODUMP); 2134 XVA_SET_REQ(tmpxvattr, XAT_NODUMP); 2135 } 2136 } 2137 2138 if (XVA_ISSET_REQ(xvap, XAT_AV_MODIFIED)) { 2139 if (xoap->xoa_av_modified != 2140 ((zp->z_pflags & ZFS_AV_MODIFIED) != 0)) { 2141 need_policy = TRUE; 2142 } else { 2143 XVA_CLR_REQ(xvap, XAT_AV_MODIFIED); 2144 XVA_SET_REQ(tmpxvattr, XAT_AV_MODIFIED); 2145 } 2146 } 2147 2148 if (XVA_ISSET_REQ(xvap, XAT_AV_QUARANTINED)) { 2149 if ((!S_ISREG(ip->i_mode) && 2150 xoap->xoa_av_quarantined) || 2151 xoap->xoa_av_quarantined != 2152 ((zp->z_pflags & ZFS_AV_QUARANTINED) != 0)) { 2153 need_policy = TRUE; 2154 } else { 2155 XVA_CLR_REQ(xvap, XAT_AV_QUARANTINED); 2156 XVA_SET_REQ(tmpxvattr, XAT_AV_QUARANTINED); 2157 } 2158 } 2159 2160 if (XVA_ISSET_REQ(xvap, XAT_REPARSE)) { 2161 mutex_exit(&zp->z_lock); 2162 err = SET_ERROR(EPERM); 2163 goto out3; 2164 } 2165 2166 if (need_policy == FALSE && 2167 (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP) || 2168 XVA_ISSET_REQ(xvap, XAT_OPAQUE))) { 2169 need_policy = TRUE; 2170 } 2171 } 2172 2173 mutex_exit(&zp->z_lock); 2174 2175 if (mask & ATTR_MODE) { 2176 if (zfs_zaccess(zp, ACE_WRITE_ACL, 0, skipaclchk, cr, 2177 mnt_ns) == 0) { 2178 err = secpolicy_setid_setsticky_clear(ip, vap, 2179 &oldva, cr, mnt_ns, zfs_i_user_ns(ip)); 2180 if (err) 2181 goto out3; 2182 trim_mask |= ATTR_MODE; 2183 } else { 2184 need_policy = TRUE; 2185 } 2186 } 2187 2188 if (need_policy) { 2189 /* 2190 * If trim_mask is set then take ownership 2191 * has been granted or write_acl is present and user 2192 * has the ability to modify mode. In that case remove 2193 * UID|GID and or MODE from mask so that 2194 * secpolicy_vnode_setattr() doesn't revoke it. 2195 */ 2196 2197 if (trim_mask) { 2198 saved_mask = vap->va_mask; 2199 vap->va_mask &= ~trim_mask; 2200 } 2201 err = secpolicy_vnode_setattr(cr, ip, vap, &oldva, flags, 2202 zfs_zaccess_unix, zp); 2203 if (err) 2204 goto out3; 2205 2206 if (trim_mask) 2207 vap->va_mask |= saved_mask; 2208 } 2209 2210 /* 2211 * secpolicy_vnode_setattr, or take ownership may have 2212 * changed va_mask 2213 */ 2214 mask = vap->va_mask; 2215 2216 if ((mask & (ATTR_UID | ATTR_GID)) || projid != ZFS_INVALID_PROJID) { 2217 handle_eadir = B_TRUE; 2218 err = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 2219 &xattr_obj, sizeof (xattr_obj)); 2220 2221 if (err == 0 && xattr_obj) { 2222 err = zfs_zget(ZTOZSB(zp), xattr_obj, &attrzp); 2223 if (err) 2224 goto out2; 2225 } 2226 if (mask & ATTR_UID) { 2227 new_kuid = zfs_fuid_create(zfsvfs, 2228 (uint64_t)vap->va_uid, cr, ZFS_OWNER, &fuidp); 2229 if (new_kuid != KUID_TO_SUID(ZTOI(zp)->i_uid) && 2230 zfs_id_overquota(zfsvfs, DMU_USERUSED_OBJECT, 2231 new_kuid)) { 2232 if (attrzp) 2233 zrele(attrzp); 2234 err = SET_ERROR(EDQUOT); 2235 goto out2; 2236 } 2237 } 2238 2239 if (mask & ATTR_GID) { 2240 new_kgid = zfs_fuid_create(zfsvfs, 2241 (uint64_t)vap->va_gid, cr, ZFS_GROUP, &fuidp); 2242 if (new_kgid != KGID_TO_SGID(ZTOI(zp)->i_gid) && 2243 zfs_id_overquota(zfsvfs, DMU_GROUPUSED_OBJECT, 2244 new_kgid)) { 2245 if (attrzp) 2246 zrele(attrzp); 2247 err = SET_ERROR(EDQUOT); 2248 goto out2; 2249 } 2250 } 2251 2252 if (projid != ZFS_INVALID_PROJID && 2253 zfs_id_overquota(zfsvfs, DMU_PROJECTUSED_OBJECT, projid)) { 2254 if (attrzp) 2255 zrele(attrzp); 2256 err = EDQUOT; 2257 goto out2; 2258 } 2259 } 2260 tx = dmu_tx_create(os); 2261 2262 if (mask & ATTR_MODE) { 2263 uint64_t pmode = zp->z_mode; 2264 uint64_t acl_obj; 2265 new_mode = (pmode & S_IFMT) | (vap->va_mode & ~S_IFMT); 2266 2267 if (ZTOZSB(zp)->z_acl_mode == ZFS_ACL_RESTRICTED && 2268 !(zp->z_pflags & ZFS_ACL_TRIVIAL)) { 2269 err = EPERM; 2270 goto out; 2271 } 2272 2273 if ((err = zfs_acl_chmod_setattr(zp, &aclp, new_mode))) 2274 goto out; 2275 2276 mutex_enter(&zp->z_lock); 2277 if (!zp->z_is_sa && ((acl_obj = zfs_external_acl(zp)) != 0)) { 2278 /* 2279 * Are we upgrading ACL from old V0 format 2280 * to V1 format? 2281 */ 2282 if (zfsvfs->z_version >= ZPL_VERSION_FUID && 2283 zfs_znode_acl_version(zp) == 2284 ZFS_ACL_VERSION_INITIAL) { 2285 dmu_tx_hold_free(tx, acl_obj, 0, 2286 DMU_OBJECT_END); 2287 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2288 0, aclp->z_acl_bytes); 2289 } else { 2290 dmu_tx_hold_write(tx, acl_obj, 0, 2291 aclp->z_acl_bytes); 2292 } 2293 } else if (!zp->z_is_sa && aclp->z_acl_bytes > ZFS_ACE_SPACE) { 2294 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2295 0, aclp->z_acl_bytes); 2296 } 2297 mutex_exit(&zp->z_lock); 2298 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 2299 } else { 2300 if (((mask & ATTR_XVATTR) && 2301 XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) || 2302 (projid != ZFS_INVALID_PROJID && 2303 !(zp->z_pflags & ZFS_PROJID))) 2304 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 2305 else 2306 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 2307 } 2308 2309 if (attrzp) { 2310 dmu_tx_hold_sa(tx, attrzp->z_sa_hdl, B_FALSE); 2311 } 2312 2313 fuid_dirtied = zfsvfs->z_fuid_dirty; 2314 if (fuid_dirtied) 2315 zfs_fuid_txhold(zfsvfs, tx); 2316 2317 zfs_sa_upgrade_txholds(tx, zp); 2318 2319 err = dmu_tx_assign(tx, TXG_WAIT); 2320 if (err) 2321 goto out; 2322 2323 count = 0; 2324 /* 2325 * Set each attribute requested. 2326 * We group settings according to the locks they need to acquire. 2327 * 2328 * Note: you cannot set ctime directly, although it will be 2329 * updated as a side-effect of calling this function. 2330 */ 2331 2332 if (projid != ZFS_INVALID_PROJID && !(zp->z_pflags & ZFS_PROJID)) { 2333 /* 2334 * For the existed object that is upgraded from old system, 2335 * its on-disk layout has no slot for the project ID attribute. 2336 * But quota accounting logic needs to access related slots by 2337 * offset directly. So we need to adjust old objects' layout 2338 * to make the project ID to some unified and fixed offset. 2339 */ 2340 if (attrzp) 2341 err = sa_add_projid(attrzp->z_sa_hdl, tx, projid); 2342 if (err == 0) 2343 err = sa_add_projid(zp->z_sa_hdl, tx, projid); 2344 2345 if (unlikely(err == EEXIST)) 2346 err = 0; 2347 else if (err != 0) 2348 goto out; 2349 else 2350 projid = ZFS_INVALID_PROJID; 2351 } 2352 2353 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE)) 2354 mutex_enter(&zp->z_acl_lock); 2355 mutex_enter(&zp->z_lock); 2356 2357 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 2358 &zp->z_pflags, sizeof (zp->z_pflags)); 2359 2360 if (attrzp) { 2361 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE)) 2362 mutex_enter(&attrzp->z_acl_lock); 2363 mutex_enter(&attrzp->z_lock); 2364 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 2365 SA_ZPL_FLAGS(zfsvfs), NULL, &attrzp->z_pflags, 2366 sizeof (attrzp->z_pflags)); 2367 if (projid != ZFS_INVALID_PROJID) { 2368 attrzp->z_projid = projid; 2369 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 2370 SA_ZPL_PROJID(zfsvfs), NULL, &attrzp->z_projid, 2371 sizeof (attrzp->z_projid)); 2372 } 2373 } 2374 2375 if (mask & (ATTR_UID|ATTR_GID)) { 2376 2377 if (mask & ATTR_UID) { 2378 ZTOI(zp)->i_uid = SUID_TO_KUID(new_kuid); 2379 new_uid = zfs_uid_read(ZTOI(zp)); 2380 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_UID(zfsvfs), NULL, 2381 &new_uid, sizeof (new_uid)); 2382 if (attrzp) { 2383 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 2384 SA_ZPL_UID(zfsvfs), NULL, &new_uid, 2385 sizeof (new_uid)); 2386 ZTOI(attrzp)->i_uid = SUID_TO_KUID(new_uid); 2387 } 2388 } 2389 2390 if (mask & ATTR_GID) { 2391 ZTOI(zp)->i_gid = SGID_TO_KGID(new_kgid); 2392 new_gid = zfs_gid_read(ZTOI(zp)); 2393 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_GID(zfsvfs), 2394 NULL, &new_gid, sizeof (new_gid)); 2395 if (attrzp) { 2396 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 2397 SA_ZPL_GID(zfsvfs), NULL, &new_gid, 2398 sizeof (new_gid)); 2399 ZTOI(attrzp)->i_gid = SGID_TO_KGID(new_kgid); 2400 } 2401 } 2402 if (!(mask & ATTR_MODE)) { 2403 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), 2404 NULL, &new_mode, sizeof (new_mode)); 2405 new_mode = zp->z_mode; 2406 } 2407 err = zfs_acl_chown_setattr(zp); 2408 ASSERT(err == 0); 2409 if (attrzp) { 2410 err = zfs_acl_chown_setattr(attrzp); 2411 ASSERT(err == 0); 2412 } 2413 } 2414 2415 if (mask & ATTR_MODE) { 2416 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MODE(zfsvfs), NULL, 2417 &new_mode, sizeof (new_mode)); 2418 zp->z_mode = ZTOI(zp)->i_mode = new_mode; 2419 ASSERT3P(aclp, !=, NULL); 2420 err = zfs_aclset_common(zp, aclp, cr, tx); 2421 ASSERT0(err); 2422 if (zp->z_acl_cached) 2423 zfs_acl_free(zp->z_acl_cached); 2424 zp->z_acl_cached = aclp; 2425 aclp = NULL; 2426 } 2427 2428 if ((mask & ATTR_ATIME) || zp->z_atime_dirty) { 2429 zp->z_atime_dirty = B_FALSE; 2430 ZFS_TIME_ENCODE(&ip->i_atime, atime); 2431 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_ATIME(zfsvfs), NULL, 2432 &atime, sizeof (atime)); 2433 } 2434 2435 if (mask & (ATTR_MTIME | ATTR_SIZE)) { 2436 ZFS_TIME_ENCODE(&vap->va_mtime, mtime); 2437 ZTOI(zp)->i_mtime = zpl_inode_timestamp_truncate( 2438 vap->va_mtime, ZTOI(zp)); 2439 2440 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 2441 mtime, sizeof (mtime)); 2442 } 2443 2444 if (mask & (ATTR_CTIME | ATTR_SIZE)) { 2445 ZFS_TIME_ENCODE(&vap->va_ctime, ctime); 2446 ZTOI(zp)->i_ctime = zpl_inode_timestamp_truncate(vap->va_ctime, 2447 ZTOI(zp)); 2448 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 2449 ctime, sizeof (ctime)); 2450 } 2451 2452 if (projid != ZFS_INVALID_PROJID) { 2453 zp->z_projid = projid; 2454 SA_ADD_BULK_ATTR(bulk, count, 2455 SA_ZPL_PROJID(zfsvfs), NULL, &zp->z_projid, 2456 sizeof (zp->z_projid)); 2457 } 2458 2459 if (attrzp && mask) { 2460 SA_ADD_BULK_ATTR(xattr_bulk, xattr_count, 2461 SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 2462 sizeof (ctime)); 2463 } 2464 2465 /* 2466 * Do this after setting timestamps to prevent timestamp 2467 * update from toggling bit 2468 */ 2469 2470 if (xoap && (mask & ATTR_XVATTR)) { 2471 2472 /* 2473 * restore trimmed off masks 2474 * so that return masks can be set for caller. 2475 */ 2476 2477 if (XVA_ISSET_REQ(tmpxvattr, XAT_APPENDONLY)) { 2478 XVA_SET_REQ(xvap, XAT_APPENDONLY); 2479 } 2480 if (XVA_ISSET_REQ(tmpxvattr, XAT_NOUNLINK)) { 2481 XVA_SET_REQ(xvap, XAT_NOUNLINK); 2482 } 2483 if (XVA_ISSET_REQ(tmpxvattr, XAT_IMMUTABLE)) { 2484 XVA_SET_REQ(xvap, XAT_IMMUTABLE); 2485 } 2486 if (XVA_ISSET_REQ(tmpxvattr, XAT_NODUMP)) { 2487 XVA_SET_REQ(xvap, XAT_NODUMP); 2488 } 2489 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_MODIFIED)) { 2490 XVA_SET_REQ(xvap, XAT_AV_MODIFIED); 2491 } 2492 if (XVA_ISSET_REQ(tmpxvattr, XAT_AV_QUARANTINED)) { 2493 XVA_SET_REQ(xvap, XAT_AV_QUARANTINED); 2494 } 2495 if (XVA_ISSET_REQ(tmpxvattr, XAT_PROJINHERIT)) { 2496 XVA_SET_REQ(xvap, XAT_PROJINHERIT); 2497 } 2498 2499 if (XVA_ISSET_REQ(xvap, XAT_AV_SCANSTAMP)) 2500 ASSERT(S_ISREG(ip->i_mode)); 2501 2502 zfs_xvattr_set(zp, xvap, tx); 2503 } 2504 2505 if (fuid_dirtied) 2506 zfs_fuid_sync(zfsvfs, tx); 2507 2508 if (mask != 0) 2509 zfs_log_setattr(zilog, tx, TX_SETATTR, zp, vap, mask, fuidp); 2510 2511 mutex_exit(&zp->z_lock); 2512 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE)) 2513 mutex_exit(&zp->z_acl_lock); 2514 2515 if (attrzp) { 2516 if (mask & (ATTR_UID|ATTR_GID|ATTR_MODE)) 2517 mutex_exit(&attrzp->z_acl_lock); 2518 mutex_exit(&attrzp->z_lock); 2519 } 2520 out: 2521 if (err == 0 && xattr_count > 0) { 2522 err2 = sa_bulk_update(attrzp->z_sa_hdl, xattr_bulk, 2523 xattr_count, tx); 2524 ASSERT(err2 == 0); 2525 } 2526 2527 if (aclp) 2528 zfs_acl_free(aclp); 2529 2530 if (fuidp) { 2531 zfs_fuid_info_free(fuidp); 2532 fuidp = NULL; 2533 } 2534 2535 if (err) { 2536 dmu_tx_abort(tx); 2537 if (attrzp) 2538 zrele(attrzp); 2539 if (err == ERESTART) 2540 goto top; 2541 } else { 2542 if (count > 0) 2543 err2 = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 2544 dmu_tx_commit(tx); 2545 if (attrzp) { 2546 if (err2 == 0 && handle_eadir) 2547 err = zfs_setattr_dir(attrzp); 2548 zrele(attrzp); 2549 } 2550 zfs_znode_update_vfs(zp); 2551 } 2552 2553 out2: 2554 if (os->os_sync == ZFS_SYNC_ALWAYS) 2555 zil_commit(zilog, 0); 2556 2557 out3: 2558 kmem_free(xattr_bulk, sizeof (sa_bulk_attr_t) * bulks); 2559 kmem_free(bulk, sizeof (sa_bulk_attr_t) * bulks); 2560 kmem_free(tmpxvattr, sizeof (xvattr_t)); 2561 zfs_exit(zfsvfs, FTAG); 2562 return (err); 2563 } 2564 2565 typedef struct zfs_zlock { 2566 krwlock_t *zl_rwlock; /* lock we acquired */ 2567 znode_t *zl_znode; /* znode we held */ 2568 struct zfs_zlock *zl_next; /* next in list */ 2569 } zfs_zlock_t; 2570 2571 /* 2572 * Drop locks and release vnodes that were held by zfs_rename_lock(). 2573 */ 2574 static void 2575 zfs_rename_unlock(zfs_zlock_t **zlpp) 2576 { 2577 zfs_zlock_t *zl; 2578 2579 while ((zl = *zlpp) != NULL) { 2580 if (zl->zl_znode != NULL) 2581 zfs_zrele_async(zl->zl_znode); 2582 rw_exit(zl->zl_rwlock); 2583 *zlpp = zl->zl_next; 2584 kmem_free(zl, sizeof (*zl)); 2585 } 2586 } 2587 2588 /* 2589 * Search back through the directory tree, using the ".." entries. 2590 * Lock each directory in the chain to prevent concurrent renames. 2591 * Fail any attempt to move a directory into one of its own descendants. 2592 * XXX - z_parent_lock can overlap with map or grow locks 2593 */ 2594 static int 2595 zfs_rename_lock(znode_t *szp, znode_t *tdzp, znode_t *sdzp, zfs_zlock_t **zlpp) 2596 { 2597 zfs_zlock_t *zl; 2598 znode_t *zp = tdzp; 2599 uint64_t rootid = ZTOZSB(zp)->z_root; 2600 uint64_t oidp = zp->z_id; 2601 krwlock_t *rwlp = &szp->z_parent_lock; 2602 krw_t rw = RW_WRITER; 2603 2604 /* 2605 * First pass write-locks szp and compares to zp->z_id. 2606 * Later passes read-lock zp and compare to zp->z_parent. 2607 */ 2608 do { 2609 if (!rw_tryenter(rwlp, rw)) { 2610 /* 2611 * Another thread is renaming in this path. 2612 * Note that if we are a WRITER, we don't have any 2613 * parent_locks held yet. 2614 */ 2615 if (rw == RW_READER && zp->z_id > szp->z_id) { 2616 /* 2617 * Drop our locks and restart 2618 */ 2619 zfs_rename_unlock(&zl); 2620 *zlpp = NULL; 2621 zp = tdzp; 2622 oidp = zp->z_id; 2623 rwlp = &szp->z_parent_lock; 2624 rw = RW_WRITER; 2625 continue; 2626 } else { 2627 /* 2628 * Wait for other thread to drop its locks 2629 */ 2630 rw_enter(rwlp, rw); 2631 } 2632 } 2633 2634 zl = kmem_alloc(sizeof (*zl), KM_SLEEP); 2635 zl->zl_rwlock = rwlp; 2636 zl->zl_znode = NULL; 2637 zl->zl_next = *zlpp; 2638 *zlpp = zl; 2639 2640 if (oidp == szp->z_id) /* We're a descendant of szp */ 2641 return (SET_ERROR(EINVAL)); 2642 2643 if (oidp == rootid) /* We've hit the top */ 2644 return (0); 2645 2646 if (rw == RW_READER) { /* i.e. not the first pass */ 2647 int error = zfs_zget(ZTOZSB(zp), oidp, &zp); 2648 if (error) 2649 return (error); 2650 zl->zl_znode = zp; 2651 } 2652 (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_PARENT(ZTOZSB(zp)), 2653 &oidp, sizeof (oidp)); 2654 rwlp = &zp->z_parent_lock; 2655 rw = RW_READER; 2656 2657 } while (zp->z_id != sdzp->z_id); 2658 2659 return (0); 2660 } 2661 2662 /* 2663 * Move an entry from the provided source directory to the target 2664 * directory. Change the entry name as indicated. 2665 * 2666 * IN: sdzp - Source directory containing the "old entry". 2667 * snm - Old entry name. 2668 * tdzp - Target directory to contain the "new entry". 2669 * tnm - New entry name. 2670 * cr - credentials of caller. 2671 * flags - case flags 2672 * rflags - RENAME_* flags 2673 * wa_vap - attributes for RENAME_WHITEOUT (must be a char 0:0). 2674 * mnt_ns - user namespace of the mount 2675 * 2676 * RETURN: 0 on success, error code on failure. 2677 * 2678 * Timestamps: 2679 * sdzp,tdzp - ctime|mtime updated 2680 */ 2681 int 2682 zfs_rename(znode_t *sdzp, char *snm, znode_t *tdzp, char *tnm, 2683 cred_t *cr, int flags, uint64_t rflags, vattr_t *wo_vap, zuserns_t *mnt_ns) 2684 { 2685 znode_t *szp, *tzp; 2686 zfsvfs_t *zfsvfs = ZTOZSB(sdzp); 2687 zilog_t *zilog; 2688 zfs_dirlock_t *sdl, *tdl; 2689 dmu_tx_t *tx; 2690 zfs_zlock_t *zl; 2691 int cmp, serr, terr; 2692 int error = 0; 2693 int zflg = 0; 2694 boolean_t waited = B_FALSE; 2695 /* Needed for whiteout inode creation. */ 2696 boolean_t fuid_dirtied; 2697 zfs_acl_ids_t acl_ids; 2698 boolean_t have_acl = B_FALSE; 2699 znode_t *wzp = NULL; 2700 2701 2702 if (snm == NULL || tnm == NULL) 2703 return (SET_ERROR(EINVAL)); 2704 2705 if (rflags & ~(RENAME_NOREPLACE | RENAME_EXCHANGE | RENAME_WHITEOUT)) 2706 return (SET_ERROR(EINVAL)); 2707 2708 /* Already checked by Linux VFS, but just to make sure. */ 2709 if (rflags & RENAME_EXCHANGE && 2710 (rflags & (RENAME_NOREPLACE | RENAME_WHITEOUT))) 2711 return (SET_ERROR(EINVAL)); 2712 2713 /* 2714 * Make sure we only get wo_vap iff. RENAME_WHITEOUT and that it's the 2715 * right kind of vattr_t for the whiteout file. These are set 2716 * internally by ZFS so should never be incorrect. 2717 */ 2718 VERIFY_EQUIV(rflags & RENAME_WHITEOUT, wo_vap != NULL); 2719 VERIFY_IMPLY(wo_vap, wo_vap->va_mode == S_IFCHR); 2720 VERIFY_IMPLY(wo_vap, wo_vap->va_rdev == makedevice(0, 0)); 2721 2722 if ((error = zfs_enter_verify_zp(zfsvfs, sdzp, FTAG)) != 0) 2723 return (error); 2724 zilog = zfsvfs->z_log; 2725 2726 if ((error = zfs_verify_zp(tdzp)) != 0) { 2727 zfs_exit(zfsvfs, FTAG); 2728 return (error); 2729 } 2730 2731 /* 2732 * We check i_sb because snapshots and the ctldir must have different 2733 * super blocks. 2734 */ 2735 if (ZTOI(tdzp)->i_sb != ZTOI(sdzp)->i_sb || 2736 zfsctl_is_node(ZTOI(tdzp))) { 2737 zfs_exit(zfsvfs, FTAG); 2738 return (SET_ERROR(EXDEV)); 2739 } 2740 2741 if (zfsvfs->z_utf8 && u8_validate(tnm, 2742 strlen(tnm), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 2743 zfs_exit(zfsvfs, FTAG); 2744 return (SET_ERROR(EILSEQ)); 2745 } 2746 2747 if (flags & FIGNORECASE) 2748 zflg |= ZCILOOK; 2749 2750 top: 2751 szp = NULL; 2752 tzp = NULL; 2753 zl = NULL; 2754 2755 /* 2756 * This is to prevent the creation of links into attribute space 2757 * by renaming a linked file into/outof an attribute directory. 2758 * See the comment in zfs_link() for why this is considered bad. 2759 */ 2760 if ((tdzp->z_pflags & ZFS_XATTR) != (sdzp->z_pflags & ZFS_XATTR)) { 2761 zfs_exit(zfsvfs, FTAG); 2762 return (SET_ERROR(EINVAL)); 2763 } 2764 2765 /* 2766 * Lock source and target directory entries. To prevent deadlock, 2767 * a lock ordering must be defined. We lock the directory with 2768 * the smallest object id first, or if it's a tie, the one with 2769 * the lexically first name. 2770 */ 2771 if (sdzp->z_id < tdzp->z_id) { 2772 cmp = -1; 2773 } else if (sdzp->z_id > tdzp->z_id) { 2774 cmp = 1; 2775 } else { 2776 /* 2777 * First compare the two name arguments without 2778 * considering any case folding. 2779 */ 2780 int nofold = (zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER); 2781 2782 cmp = u8_strcmp(snm, tnm, 0, nofold, U8_UNICODE_LATEST, &error); 2783 ASSERT(error == 0 || !zfsvfs->z_utf8); 2784 if (cmp == 0) { 2785 /* 2786 * POSIX: "If the old argument and the new argument 2787 * both refer to links to the same existing file, 2788 * the rename() function shall return successfully 2789 * and perform no other action." 2790 */ 2791 zfs_exit(zfsvfs, FTAG); 2792 return (0); 2793 } 2794 /* 2795 * If the file system is case-folding, then we may 2796 * have some more checking to do. A case-folding file 2797 * system is either supporting mixed case sensitivity 2798 * access or is completely case-insensitive. Note 2799 * that the file system is always case preserving. 2800 * 2801 * In mixed sensitivity mode case sensitive behavior 2802 * is the default. FIGNORECASE must be used to 2803 * explicitly request case insensitive behavior. 2804 * 2805 * If the source and target names provided differ only 2806 * by case (e.g., a request to rename 'tim' to 'Tim'), 2807 * we will treat this as a special case in the 2808 * case-insensitive mode: as long as the source name 2809 * is an exact match, we will allow this to proceed as 2810 * a name-change request. 2811 */ 2812 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE || 2813 (zfsvfs->z_case == ZFS_CASE_MIXED && 2814 flags & FIGNORECASE)) && 2815 u8_strcmp(snm, tnm, 0, zfsvfs->z_norm, U8_UNICODE_LATEST, 2816 &error) == 0) { 2817 /* 2818 * case preserving rename request, require exact 2819 * name matches 2820 */ 2821 zflg |= ZCIEXACT; 2822 zflg &= ~ZCILOOK; 2823 } 2824 } 2825 2826 /* 2827 * If the source and destination directories are the same, we should 2828 * grab the z_name_lock of that directory only once. 2829 */ 2830 if (sdzp == tdzp) { 2831 zflg |= ZHAVELOCK; 2832 rw_enter(&sdzp->z_name_lock, RW_READER); 2833 } 2834 2835 if (cmp < 0) { 2836 serr = zfs_dirent_lock(&sdl, sdzp, snm, &szp, 2837 ZEXISTS | zflg, NULL, NULL); 2838 terr = zfs_dirent_lock(&tdl, 2839 tdzp, tnm, &tzp, ZRENAMING | zflg, NULL, NULL); 2840 } else { 2841 terr = zfs_dirent_lock(&tdl, 2842 tdzp, tnm, &tzp, zflg, NULL, NULL); 2843 serr = zfs_dirent_lock(&sdl, 2844 sdzp, snm, &szp, ZEXISTS | ZRENAMING | zflg, 2845 NULL, NULL); 2846 } 2847 2848 if (serr) { 2849 /* 2850 * Source entry invalid or not there. 2851 */ 2852 if (!terr) { 2853 zfs_dirent_unlock(tdl); 2854 if (tzp) 2855 zrele(tzp); 2856 } 2857 2858 if (sdzp == tdzp) 2859 rw_exit(&sdzp->z_name_lock); 2860 2861 if (strcmp(snm, "..") == 0) 2862 serr = EINVAL; 2863 zfs_exit(zfsvfs, FTAG); 2864 return (serr); 2865 } 2866 if (terr) { 2867 zfs_dirent_unlock(sdl); 2868 zrele(szp); 2869 2870 if (sdzp == tdzp) 2871 rw_exit(&sdzp->z_name_lock); 2872 2873 if (strcmp(tnm, "..") == 0) 2874 terr = EINVAL; 2875 zfs_exit(zfsvfs, FTAG); 2876 return (terr); 2877 } 2878 2879 /* 2880 * If we are using project inheritance, means if the directory has 2881 * ZFS_PROJINHERIT set, then its descendant directories will inherit 2882 * not only the project ID, but also the ZFS_PROJINHERIT flag. Under 2883 * such case, we only allow renames into our tree when the project 2884 * IDs are the same. 2885 */ 2886 if (tdzp->z_pflags & ZFS_PROJINHERIT && 2887 tdzp->z_projid != szp->z_projid) { 2888 error = SET_ERROR(EXDEV); 2889 goto out; 2890 } 2891 2892 /* 2893 * Must have write access at the source to remove the old entry 2894 * and write access at the target to create the new entry. 2895 * Note that if target and source are the same, this can be 2896 * done in a single check. 2897 */ 2898 if ((error = zfs_zaccess_rename(sdzp, szp, tdzp, tzp, cr, mnt_ns))) 2899 goto out; 2900 2901 if (S_ISDIR(ZTOI(szp)->i_mode)) { 2902 /* 2903 * Check to make sure rename is valid. 2904 * Can't do a move like this: /usr/a/b to /usr/a/b/c/d 2905 */ 2906 if ((error = zfs_rename_lock(szp, tdzp, sdzp, &zl))) 2907 goto out; 2908 } 2909 2910 /* 2911 * Does target exist? 2912 */ 2913 if (tzp) { 2914 if (rflags & RENAME_NOREPLACE) { 2915 error = SET_ERROR(EEXIST); 2916 goto out; 2917 } 2918 /* 2919 * Source and target must be the same type (unless exchanging). 2920 */ 2921 if (!(rflags & RENAME_EXCHANGE)) { 2922 boolean_t s_is_dir = S_ISDIR(ZTOI(szp)->i_mode) != 0; 2923 boolean_t t_is_dir = S_ISDIR(ZTOI(tzp)->i_mode) != 0; 2924 2925 if (s_is_dir != t_is_dir) { 2926 error = SET_ERROR(s_is_dir ? ENOTDIR : EISDIR); 2927 goto out; 2928 } 2929 } 2930 /* 2931 * POSIX dictates that when the source and target 2932 * entries refer to the same file object, rename 2933 * must do nothing and exit without error. 2934 */ 2935 if (szp->z_id == tzp->z_id) { 2936 error = 0; 2937 goto out; 2938 } 2939 } else if (rflags & RENAME_EXCHANGE) { 2940 /* Target must exist for RENAME_EXCHANGE. */ 2941 error = SET_ERROR(ENOENT); 2942 goto out; 2943 } 2944 2945 /* Set up inode creation for RENAME_WHITEOUT. */ 2946 if (rflags & RENAME_WHITEOUT) { 2947 /* 2948 * Whiteout files are not regular files or directories, so to 2949 * match zfs_create() we do not inherit the project id. 2950 */ 2951 uint64_t wo_projid = ZFS_DEFAULT_PROJID; 2952 2953 error = zfs_zaccess(sdzp, ACE_ADD_FILE, 0, B_FALSE, cr, mnt_ns); 2954 if (error) 2955 goto out; 2956 2957 if (!have_acl) { 2958 error = zfs_acl_ids_create(sdzp, 0, wo_vap, cr, NULL, 2959 &acl_ids, mnt_ns); 2960 if (error) 2961 goto out; 2962 have_acl = B_TRUE; 2963 } 2964 2965 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, wo_projid)) { 2966 error = SET_ERROR(EDQUOT); 2967 goto out; 2968 } 2969 } 2970 2971 tx = dmu_tx_create(zfsvfs->z_os); 2972 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 2973 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE); 2974 dmu_tx_hold_zap(tx, sdzp->z_id, 2975 (rflags & RENAME_EXCHANGE) ? TRUE : FALSE, snm); 2976 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, tnm); 2977 if (sdzp != tdzp) { 2978 dmu_tx_hold_sa(tx, tdzp->z_sa_hdl, B_FALSE); 2979 zfs_sa_upgrade_txholds(tx, tdzp); 2980 } 2981 if (tzp) { 2982 dmu_tx_hold_sa(tx, tzp->z_sa_hdl, B_FALSE); 2983 zfs_sa_upgrade_txholds(tx, tzp); 2984 } 2985 if (rflags & RENAME_WHITEOUT) { 2986 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 2987 ZFS_SA_BASE_ATTR_SIZE); 2988 2989 dmu_tx_hold_zap(tx, sdzp->z_id, TRUE, snm); 2990 dmu_tx_hold_sa(tx, sdzp->z_sa_hdl, B_FALSE); 2991 if (!zfsvfs->z_use_sa && 2992 acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 2993 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 2994 0, acl_ids.z_aclp->z_acl_bytes); 2995 } 2996 } 2997 fuid_dirtied = zfsvfs->z_fuid_dirty; 2998 if (fuid_dirtied) 2999 zfs_fuid_txhold(zfsvfs, tx); 3000 zfs_sa_upgrade_txholds(tx, szp); 3001 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 3002 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT); 3003 if (error) { 3004 if (zl != NULL) 3005 zfs_rename_unlock(&zl); 3006 zfs_dirent_unlock(sdl); 3007 zfs_dirent_unlock(tdl); 3008 3009 if (sdzp == tdzp) 3010 rw_exit(&sdzp->z_name_lock); 3011 3012 if (error == ERESTART) { 3013 waited = B_TRUE; 3014 dmu_tx_wait(tx); 3015 dmu_tx_abort(tx); 3016 zrele(szp); 3017 if (tzp) 3018 zrele(tzp); 3019 goto top; 3020 } 3021 dmu_tx_abort(tx); 3022 zrele(szp); 3023 if (tzp) 3024 zrele(tzp); 3025 zfs_exit(zfsvfs, FTAG); 3026 return (error); 3027 } 3028 3029 /* 3030 * Unlink the source. 3031 */ 3032 szp->z_pflags |= ZFS_AV_MODIFIED; 3033 if (tdzp->z_pflags & ZFS_PROJINHERIT) 3034 szp->z_pflags |= ZFS_PROJINHERIT; 3035 3036 error = sa_update(szp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs), 3037 (void *)&szp->z_pflags, sizeof (uint64_t), tx); 3038 VERIFY0(error); 3039 3040 error = zfs_link_destroy(sdl, szp, tx, ZRENAMING, NULL); 3041 if (error) 3042 goto commit; 3043 3044 /* 3045 * Unlink the target. 3046 */ 3047 if (tzp) { 3048 int tzflg = zflg; 3049 3050 if (rflags & RENAME_EXCHANGE) { 3051 /* This inode will be re-linked soon. */ 3052 tzflg |= ZRENAMING; 3053 3054 tzp->z_pflags |= ZFS_AV_MODIFIED; 3055 if (sdzp->z_pflags & ZFS_PROJINHERIT) 3056 tzp->z_pflags |= ZFS_PROJINHERIT; 3057 3058 error = sa_update(tzp->z_sa_hdl, SA_ZPL_FLAGS(zfsvfs), 3059 (void *)&tzp->z_pflags, sizeof (uint64_t), tx); 3060 ASSERT0(error); 3061 } 3062 error = zfs_link_destroy(tdl, tzp, tx, tzflg, NULL); 3063 if (error) 3064 goto commit_link_szp; 3065 } 3066 3067 /* 3068 * Create the new target links: 3069 * * We always link the target. 3070 * * RENAME_EXCHANGE: Link the old target to the source. 3071 * * RENAME_WHITEOUT: Create a whiteout inode in-place of the source. 3072 */ 3073 error = zfs_link_create(tdl, szp, tx, ZRENAMING); 3074 if (error) { 3075 /* 3076 * If we have removed the existing target, a subsequent call to 3077 * zfs_link_create() to add back the same entry, but with a new 3078 * dnode (szp), should not fail. 3079 */ 3080 ASSERT3P(tzp, ==, NULL); 3081 goto commit_link_tzp; 3082 } 3083 3084 switch (rflags & (RENAME_EXCHANGE | RENAME_WHITEOUT)) { 3085 case RENAME_EXCHANGE: 3086 error = zfs_link_create(sdl, tzp, tx, ZRENAMING); 3087 /* 3088 * The same argument as zfs_link_create() failing for 3089 * szp applies here, since the source directory must 3090 * have had an entry we are replacing. 3091 */ 3092 ASSERT0(error); 3093 if (error) 3094 goto commit_unlink_td_szp; 3095 break; 3096 case RENAME_WHITEOUT: 3097 zfs_mknode(sdzp, wo_vap, tx, cr, 0, &wzp, &acl_ids); 3098 error = zfs_link_create(sdl, wzp, tx, ZNEW); 3099 if (error) { 3100 zfs_znode_delete(wzp, tx); 3101 remove_inode_hash(ZTOI(wzp)); 3102 goto commit_unlink_td_szp; 3103 } 3104 break; 3105 } 3106 3107 if (fuid_dirtied) 3108 zfs_fuid_sync(zfsvfs, tx); 3109 3110 switch (rflags & (RENAME_EXCHANGE | RENAME_WHITEOUT)) { 3111 case RENAME_EXCHANGE: 3112 zfs_log_rename_exchange(zilog, tx, 3113 (flags & FIGNORECASE ? TX_CI : 0), sdzp, sdl->dl_name, 3114 tdzp, tdl->dl_name, szp); 3115 break; 3116 case RENAME_WHITEOUT: 3117 zfs_log_rename_whiteout(zilog, tx, 3118 (flags & FIGNORECASE ? TX_CI : 0), sdzp, sdl->dl_name, 3119 tdzp, tdl->dl_name, szp, wzp); 3120 break; 3121 default: 3122 ASSERT0(rflags & ~RENAME_NOREPLACE); 3123 zfs_log_rename(zilog, tx, (flags & FIGNORECASE ? TX_CI : 0), 3124 sdzp, sdl->dl_name, tdzp, tdl->dl_name, szp); 3125 break; 3126 } 3127 3128 commit: 3129 dmu_tx_commit(tx); 3130 out: 3131 if (have_acl) 3132 zfs_acl_ids_free(&acl_ids); 3133 3134 zfs_znode_update_vfs(sdzp); 3135 if (sdzp == tdzp) 3136 rw_exit(&sdzp->z_name_lock); 3137 3138 if (sdzp != tdzp) 3139 zfs_znode_update_vfs(tdzp); 3140 3141 zfs_znode_update_vfs(szp); 3142 zrele(szp); 3143 if (wzp) { 3144 zfs_znode_update_vfs(wzp); 3145 zrele(wzp); 3146 } 3147 if (tzp) { 3148 zfs_znode_update_vfs(tzp); 3149 zrele(tzp); 3150 } 3151 3152 if (zl != NULL) 3153 zfs_rename_unlock(&zl); 3154 3155 zfs_dirent_unlock(sdl); 3156 zfs_dirent_unlock(tdl); 3157 3158 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3159 zil_commit(zilog, 0); 3160 3161 zfs_exit(zfsvfs, FTAG); 3162 return (error); 3163 3164 /* 3165 * Clean-up path for broken link state. 3166 * 3167 * At this point we are in a (very) bad state, so we need to do our 3168 * best to correct the state. In particular, all of the nlinks are 3169 * wrong because we were destroying and creating links with ZRENAMING. 3170 * 3171 * In some form, all of these operations have to resolve the state: 3172 * 3173 * * link_destroy() *must* succeed. Fortunately, this is very likely 3174 * since we only just created it. 3175 * 3176 * * link_create()s are allowed to fail (though they shouldn't because 3177 * we only just unlinked them and are putting the entries back 3178 * during clean-up). But if they fail, we can just forcefully drop 3179 * the nlink value to (at the very least) avoid broken nlink values 3180 * -- though in the case of non-empty directories we will have to 3181 * panic (otherwise we'd have a leaked directory with a broken ..). 3182 */ 3183 commit_unlink_td_szp: 3184 VERIFY0(zfs_link_destroy(tdl, szp, tx, ZRENAMING, NULL)); 3185 commit_link_tzp: 3186 if (tzp) { 3187 if (zfs_link_create(tdl, tzp, tx, ZRENAMING)) 3188 VERIFY0(zfs_drop_nlink(tzp, tx, NULL)); 3189 } 3190 commit_link_szp: 3191 if (zfs_link_create(sdl, szp, tx, ZRENAMING)) 3192 VERIFY0(zfs_drop_nlink(szp, tx, NULL)); 3193 goto commit; 3194 } 3195 3196 /* 3197 * Insert the indicated symbolic reference entry into the directory. 3198 * 3199 * IN: dzp - Directory to contain new symbolic link. 3200 * name - Name of directory entry in dip. 3201 * vap - Attributes of new entry. 3202 * link - Name for new symlink entry. 3203 * cr - credentials of caller. 3204 * flags - case flags 3205 * mnt_ns - user namespace of the mount 3206 * 3207 * OUT: zpp - Znode for new symbolic link. 3208 * 3209 * RETURN: 0 on success, error code on failure. 3210 * 3211 * Timestamps: 3212 * dip - ctime|mtime updated 3213 */ 3214 int 3215 zfs_symlink(znode_t *dzp, char *name, vattr_t *vap, char *link, 3216 znode_t **zpp, cred_t *cr, int flags, zuserns_t *mnt_ns) 3217 { 3218 znode_t *zp; 3219 zfs_dirlock_t *dl; 3220 dmu_tx_t *tx; 3221 zfsvfs_t *zfsvfs = ZTOZSB(dzp); 3222 zilog_t *zilog; 3223 uint64_t len = strlen(link); 3224 int error; 3225 int zflg = ZNEW; 3226 zfs_acl_ids_t acl_ids; 3227 boolean_t fuid_dirtied; 3228 uint64_t txtype = TX_SYMLINK; 3229 boolean_t waited = B_FALSE; 3230 3231 ASSERT(S_ISLNK(vap->va_mode)); 3232 3233 if (name == NULL) 3234 return (SET_ERROR(EINVAL)); 3235 3236 if ((error = zfs_enter_verify_zp(zfsvfs, dzp, FTAG)) != 0) 3237 return (error); 3238 zilog = zfsvfs->z_log; 3239 3240 if (zfsvfs->z_utf8 && u8_validate(name, strlen(name), 3241 NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3242 zfs_exit(zfsvfs, FTAG); 3243 return (SET_ERROR(EILSEQ)); 3244 } 3245 if (flags & FIGNORECASE) 3246 zflg |= ZCILOOK; 3247 3248 if (len > MAXPATHLEN) { 3249 zfs_exit(zfsvfs, FTAG); 3250 return (SET_ERROR(ENAMETOOLONG)); 3251 } 3252 3253 if ((error = zfs_acl_ids_create(dzp, 0, 3254 vap, cr, NULL, &acl_ids, mnt_ns)) != 0) { 3255 zfs_exit(zfsvfs, FTAG); 3256 return (error); 3257 } 3258 top: 3259 *zpp = NULL; 3260 3261 /* 3262 * Attempt to lock directory; fail if entry already exists. 3263 */ 3264 error = zfs_dirent_lock(&dl, dzp, name, &zp, zflg, NULL, NULL); 3265 if (error) { 3266 zfs_acl_ids_free(&acl_ids); 3267 zfs_exit(zfsvfs, FTAG); 3268 return (error); 3269 } 3270 3271 if ((error = zfs_zaccess(dzp, ACE_ADD_FILE, 0, B_FALSE, cr, mnt_ns))) { 3272 zfs_acl_ids_free(&acl_ids); 3273 zfs_dirent_unlock(dl); 3274 zfs_exit(zfsvfs, FTAG); 3275 return (error); 3276 } 3277 3278 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, ZFS_DEFAULT_PROJID)) { 3279 zfs_acl_ids_free(&acl_ids); 3280 zfs_dirent_unlock(dl); 3281 zfs_exit(zfsvfs, FTAG); 3282 return (SET_ERROR(EDQUOT)); 3283 } 3284 tx = dmu_tx_create(zfsvfs->z_os); 3285 fuid_dirtied = zfsvfs->z_fuid_dirty; 3286 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, MAX(1, len)); 3287 dmu_tx_hold_zap(tx, dzp->z_id, TRUE, name); 3288 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 3289 ZFS_SA_BASE_ATTR_SIZE + len); 3290 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 3291 if (!zfsvfs->z_use_sa && acl_ids.z_aclp->z_acl_bytes > ZFS_ACE_SPACE) { 3292 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 3293 acl_ids.z_aclp->z_acl_bytes); 3294 } 3295 if (fuid_dirtied) 3296 zfs_fuid_txhold(zfsvfs, tx); 3297 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT); 3298 if (error) { 3299 zfs_dirent_unlock(dl); 3300 if (error == ERESTART) { 3301 waited = B_TRUE; 3302 dmu_tx_wait(tx); 3303 dmu_tx_abort(tx); 3304 goto top; 3305 } 3306 zfs_acl_ids_free(&acl_ids); 3307 dmu_tx_abort(tx); 3308 zfs_exit(zfsvfs, FTAG); 3309 return (error); 3310 } 3311 3312 /* 3313 * Create a new object for the symlink. 3314 * for version 4 ZPL datasets the symlink will be an SA attribute 3315 */ 3316 zfs_mknode(dzp, vap, tx, cr, 0, &zp, &acl_ids); 3317 3318 if (fuid_dirtied) 3319 zfs_fuid_sync(zfsvfs, tx); 3320 3321 mutex_enter(&zp->z_lock); 3322 if (zp->z_is_sa) 3323 error = sa_update(zp->z_sa_hdl, SA_ZPL_SYMLINK(zfsvfs), 3324 link, len, tx); 3325 else 3326 zfs_sa_symlink(zp, link, len, tx); 3327 mutex_exit(&zp->z_lock); 3328 3329 zp->z_size = len; 3330 (void) sa_update(zp->z_sa_hdl, SA_ZPL_SIZE(zfsvfs), 3331 &zp->z_size, sizeof (zp->z_size), tx); 3332 /* 3333 * Insert the new object into the directory. 3334 */ 3335 error = zfs_link_create(dl, zp, tx, ZNEW); 3336 if (error != 0) { 3337 zfs_znode_delete(zp, tx); 3338 remove_inode_hash(ZTOI(zp)); 3339 } else { 3340 if (flags & FIGNORECASE) 3341 txtype |= TX_CI; 3342 zfs_log_symlink(zilog, tx, txtype, dzp, zp, name, link); 3343 3344 zfs_znode_update_vfs(dzp); 3345 zfs_znode_update_vfs(zp); 3346 } 3347 3348 zfs_acl_ids_free(&acl_ids); 3349 3350 dmu_tx_commit(tx); 3351 3352 zfs_dirent_unlock(dl); 3353 3354 if (error == 0) { 3355 *zpp = zp; 3356 3357 if (zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3358 zil_commit(zilog, 0); 3359 } else { 3360 zrele(zp); 3361 } 3362 3363 zfs_exit(zfsvfs, FTAG); 3364 return (error); 3365 } 3366 3367 /* 3368 * Return, in the buffer contained in the provided uio structure, 3369 * the symbolic path referred to by ip. 3370 * 3371 * IN: ip - inode of symbolic link 3372 * uio - structure to contain the link path. 3373 * cr - credentials of caller. 3374 * 3375 * RETURN: 0 if success 3376 * error code if failure 3377 * 3378 * Timestamps: 3379 * ip - atime updated 3380 */ 3381 int 3382 zfs_readlink(struct inode *ip, zfs_uio_t *uio, cred_t *cr) 3383 { 3384 (void) cr; 3385 znode_t *zp = ITOZ(ip); 3386 zfsvfs_t *zfsvfs = ITOZSB(ip); 3387 int error; 3388 3389 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 3390 return (error); 3391 3392 mutex_enter(&zp->z_lock); 3393 if (zp->z_is_sa) 3394 error = sa_lookup_uio(zp->z_sa_hdl, 3395 SA_ZPL_SYMLINK(zfsvfs), uio); 3396 else 3397 error = zfs_sa_readlink(zp, uio); 3398 mutex_exit(&zp->z_lock); 3399 3400 zfs_exit(zfsvfs, FTAG); 3401 return (error); 3402 } 3403 3404 /* 3405 * Insert a new entry into directory tdzp referencing szp. 3406 * 3407 * IN: tdzp - Directory to contain new entry. 3408 * szp - znode of new entry. 3409 * name - name of new entry. 3410 * cr - credentials of caller. 3411 * flags - case flags. 3412 * 3413 * RETURN: 0 if success 3414 * error code if failure 3415 * 3416 * Timestamps: 3417 * tdzp - ctime|mtime updated 3418 * szp - ctime updated 3419 */ 3420 int 3421 zfs_link(znode_t *tdzp, znode_t *szp, char *name, cred_t *cr, 3422 int flags) 3423 { 3424 struct inode *sip = ZTOI(szp); 3425 znode_t *tzp; 3426 zfsvfs_t *zfsvfs = ZTOZSB(tdzp); 3427 zilog_t *zilog; 3428 zfs_dirlock_t *dl; 3429 dmu_tx_t *tx; 3430 int error; 3431 int zf = ZNEW; 3432 uint64_t parent; 3433 uid_t owner; 3434 boolean_t waited = B_FALSE; 3435 boolean_t is_tmpfile = 0; 3436 uint64_t txg; 3437 #ifdef HAVE_TMPFILE 3438 is_tmpfile = (sip->i_nlink == 0 && (sip->i_state & I_LINKABLE)); 3439 #endif 3440 ASSERT(S_ISDIR(ZTOI(tdzp)->i_mode)); 3441 3442 if (name == NULL) 3443 return (SET_ERROR(EINVAL)); 3444 3445 if ((error = zfs_enter_verify_zp(zfsvfs, tdzp, FTAG)) != 0) 3446 return (error); 3447 zilog = zfsvfs->z_log; 3448 3449 /* 3450 * POSIX dictates that we return EPERM here. 3451 * Better choices include ENOTSUP or EISDIR. 3452 */ 3453 if (S_ISDIR(sip->i_mode)) { 3454 zfs_exit(zfsvfs, FTAG); 3455 return (SET_ERROR(EPERM)); 3456 } 3457 3458 if ((error = zfs_verify_zp(szp)) != 0) { 3459 zfs_exit(zfsvfs, FTAG); 3460 return (error); 3461 } 3462 3463 /* 3464 * If we are using project inheritance, means if the directory has 3465 * ZFS_PROJINHERIT set, then its descendant directories will inherit 3466 * not only the project ID, but also the ZFS_PROJINHERIT flag. Under 3467 * such case, we only allow hard link creation in our tree when the 3468 * project IDs are the same. 3469 */ 3470 if (tdzp->z_pflags & ZFS_PROJINHERIT && 3471 tdzp->z_projid != szp->z_projid) { 3472 zfs_exit(zfsvfs, FTAG); 3473 return (SET_ERROR(EXDEV)); 3474 } 3475 3476 /* 3477 * We check i_sb because snapshots and the ctldir must have different 3478 * super blocks. 3479 */ 3480 if (sip->i_sb != ZTOI(tdzp)->i_sb || zfsctl_is_node(sip)) { 3481 zfs_exit(zfsvfs, FTAG); 3482 return (SET_ERROR(EXDEV)); 3483 } 3484 3485 /* Prevent links to .zfs/shares files */ 3486 3487 if ((error = sa_lookup(szp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 3488 &parent, sizeof (uint64_t))) != 0) { 3489 zfs_exit(zfsvfs, FTAG); 3490 return (error); 3491 } 3492 if (parent == zfsvfs->z_shares_dir) { 3493 zfs_exit(zfsvfs, FTAG); 3494 return (SET_ERROR(EPERM)); 3495 } 3496 3497 if (zfsvfs->z_utf8 && u8_validate(name, 3498 strlen(name), NULL, U8_VALIDATE_ENTIRE, &error) < 0) { 3499 zfs_exit(zfsvfs, FTAG); 3500 return (SET_ERROR(EILSEQ)); 3501 } 3502 if (flags & FIGNORECASE) 3503 zf |= ZCILOOK; 3504 3505 /* 3506 * We do not support links between attributes and non-attributes 3507 * because of the potential security risk of creating links 3508 * into "normal" file space in order to circumvent restrictions 3509 * imposed in attribute space. 3510 */ 3511 if ((szp->z_pflags & ZFS_XATTR) != (tdzp->z_pflags & ZFS_XATTR)) { 3512 zfs_exit(zfsvfs, FTAG); 3513 return (SET_ERROR(EINVAL)); 3514 } 3515 3516 owner = zfs_fuid_map_id(zfsvfs, KUID_TO_SUID(sip->i_uid), 3517 cr, ZFS_OWNER); 3518 if (owner != crgetuid(cr) && secpolicy_basic_link(cr) != 0) { 3519 zfs_exit(zfsvfs, FTAG); 3520 return (SET_ERROR(EPERM)); 3521 } 3522 3523 if ((error = zfs_zaccess(tdzp, ACE_ADD_FILE, 0, B_FALSE, cr, 3524 kcred->user_ns))) { 3525 zfs_exit(zfsvfs, FTAG); 3526 return (error); 3527 } 3528 3529 top: 3530 /* 3531 * Attempt to lock directory; fail if entry already exists. 3532 */ 3533 error = zfs_dirent_lock(&dl, tdzp, name, &tzp, zf, NULL, NULL); 3534 if (error) { 3535 zfs_exit(zfsvfs, FTAG); 3536 return (error); 3537 } 3538 3539 tx = dmu_tx_create(zfsvfs->z_os); 3540 dmu_tx_hold_sa(tx, szp->z_sa_hdl, B_FALSE); 3541 dmu_tx_hold_zap(tx, tdzp->z_id, TRUE, name); 3542 if (is_tmpfile) 3543 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 3544 3545 zfs_sa_upgrade_txholds(tx, szp); 3546 zfs_sa_upgrade_txholds(tx, tdzp); 3547 error = dmu_tx_assign(tx, (waited ? TXG_NOTHROTTLE : 0) | TXG_NOWAIT); 3548 if (error) { 3549 zfs_dirent_unlock(dl); 3550 if (error == ERESTART) { 3551 waited = B_TRUE; 3552 dmu_tx_wait(tx); 3553 dmu_tx_abort(tx); 3554 goto top; 3555 } 3556 dmu_tx_abort(tx); 3557 zfs_exit(zfsvfs, FTAG); 3558 return (error); 3559 } 3560 /* unmark z_unlinked so zfs_link_create will not reject */ 3561 if (is_tmpfile) 3562 szp->z_unlinked = B_FALSE; 3563 error = zfs_link_create(dl, szp, tx, 0); 3564 3565 if (error == 0) { 3566 uint64_t txtype = TX_LINK; 3567 /* 3568 * tmpfile is created to be in z_unlinkedobj, so remove it. 3569 * Also, we don't log in ZIL, because all previous file 3570 * operation on the tmpfile are ignored by ZIL. Instead we 3571 * always wait for txg to sync to make sure all previous 3572 * operation are sync safe. 3573 */ 3574 if (is_tmpfile) { 3575 VERIFY(zap_remove_int(zfsvfs->z_os, 3576 zfsvfs->z_unlinkedobj, szp->z_id, tx) == 0); 3577 } else { 3578 if (flags & FIGNORECASE) 3579 txtype |= TX_CI; 3580 zfs_log_link(zilog, tx, txtype, tdzp, szp, name); 3581 } 3582 } else if (is_tmpfile) { 3583 /* restore z_unlinked since when linking failed */ 3584 szp->z_unlinked = B_TRUE; 3585 } 3586 txg = dmu_tx_get_txg(tx); 3587 dmu_tx_commit(tx); 3588 3589 zfs_dirent_unlock(dl); 3590 3591 if (!is_tmpfile && zfsvfs->z_os->os_sync == ZFS_SYNC_ALWAYS) 3592 zil_commit(zilog, 0); 3593 3594 if (is_tmpfile && zfsvfs->z_os->os_sync != ZFS_SYNC_DISABLED) 3595 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), txg); 3596 3597 zfs_znode_update_vfs(tdzp); 3598 zfs_znode_update_vfs(szp); 3599 zfs_exit(zfsvfs, FTAG); 3600 return (error); 3601 } 3602 3603 static void 3604 zfs_putpage_sync_commit_cb(void *arg) 3605 { 3606 struct page *pp = arg; 3607 3608 ClearPageError(pp); 3609 end_page_writeback(pp); 3610 } 3611 3612 static void 3613 zfs_putpage_async_commit_cb(void *arg) 3614 { 3615 struct page *pp = arg; 3616 znode_t *zp = ITOZ(pp->mapping->host); 3617 3618 ClearPageError(pp); 3619 end_page_writeback(pp); 3620 atomic_dec_32(&zp->z_async_writes_cnt); 3621 } 3622 3623 /* 3624 * Push a page out to disk, once the page is on stable storage the 3625 * registered commit callback will be run as notification of completion. 3626 * 3627 * IN: ip - page mapped for inode. 3628 * pp - page to push (page is locked) 3629 * wbc - writeback control data 3630 * for_sync - does the caller intend to wait synchronously for the 3631 * page writeback to complete? 3632 * 3633 * RETURN: 0 if success 3634 * error code if failure 3635 * 3636 * Timestamps: 3637 * ip - ctime|mtime updated 3638 */ 3639 int 3640 zfs_putpage(struct inode *ip, struct page *pp, struct writeback_control *wbc, 3641 boolean_t for_sync) 3642 { 3643 znode_t *zp = ITOZ(ip); 3644 zfsvfs_t *zfsvfs = ITOZSB(ip); 3645 loff_t offset; 3646 loff_t pgoff; 3647 unsigned int pglen; 3648 dmu_tx_t *tx; 3649 caddr_t va; 3650 int err = 0; 3651 uint64_t mtime[2], ctime[2]; 3652 sa_bulk_attr_t bulk[3]; 3653 int cnt = 0; 3654 struct address_space *mapping; 3655 3656 if ((err = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 3657 return (err); 3658 3659 ASSERT(PageLocked(pp)); 3660 3661 pgoff = page_offset(pp); /* Page byte-offset in file */ 3662 offset = i_size_read(ip); /* File length in bytes */ 3663 pglen = MIN(PAGE_SIZE, /* Page length in bytes */ 3664 P2ROUNDUP(offset, PAGE_SIZE)-pgoff); 3665 3666 /* Page is beyond end of file */ 3667 if (pgoff >= offset) { 3668 unlock_page(pp); 3669 zfs_exit(zfsvfs, FTAG); 3670 return (0); 3671 } 3672 3673 /* Truncate page length to end of file */ 3674 if (pgoff + pglen > offset) 3675 pglen = offset - pgoff; 3676 3677 #if 0 3678 /* 3679 * FIXME: Allow mmap writes past its quota. The correct fix 3680 * is to register a page_mkwrite() handler to count the page 3681 * against its quota when it is about to be dirtied. 3682 */ 3683 if (zfs_id_overblockquota(zfsvfs, DMU_USERUSED_OBJECT, 3684 KUID_TO_SUID(ip->i_uid)) || 3685 zfs_id_overblockquota(zfsvfs, DMU_GROUPUSED_OBJECT, 3686 KGID_TO_SGID(ip->i_gid)) || 3687 (zp->z_projid != ZFS_DEFAULT_PROJID && 3688 zfs_id_overblockquota(zfsvfs, DMU_PROJECTUSED_OBJECT, 3689 zp->z_projid))) { 3690 err = EDQUOT; 3691 } 3692 #endif 3693 3694 /* 3695 * The ordering here is critical and must adhere to the following 3696 * rules in order to avoid deadlocking in either zfs_read() or 3697 * zfs_free_range() due to a lock inversion. 3698 * 3699 * 1) The page must be unlocked prior to acquiring the range lock. 3700 * This is critical because zfs_read() calls find_lock_page() 3701 * which may block on the page lock while holding the range lock. 3702 * 3703 * 2) Before setting or clearing write back on a page the range lock 3704 * must be held in order to prevent a lock inversion with the 3705 * zfs_free_range() function. 3706 * 3707 * This presents a problem because upon entering this function the 3708 * page lock is already held. To safely acquire the range lock the 3709 * page lock must be dropped. This creates a window where another 3710 * process could truncate, invalidate, dirty, or write out the page. 3711 * 3712 * Therefore, after successfully reacquiring the range and page locks 3713 * the current page state is checked. In the common case everything 3714 * will be as is expected and it can be written out. However, if 3715 * the page state has changed it must be handled accordingly. 3716 */ 3717 mapping = pp->mapping; 3718 redirty_page_for_writepage(wbc, pp); 3719 unlock_page(pp); 3720 3721 zfs_locked_range_t *lr = zfs_rangelock_enter(&zp->z_rangelock, 3722 pgoff, pglen, RL_WRITER); 3723 lock_page(pp); 3724 3725 /* Page mapping changed or it was no longer dirty, we're done */ 3726 if (unlikely((mapping != pp->mapping) || !PageDirty(pp))) { 3727 unlock_page(pp); 3728 zfs_rangelock_exit(lr); 3729 zfs_exit(zfsvfs, FTAG); 3730 return (0); 3731 } 3732 3733 /* Another process started write block if required */ 3734 if (PageWriteback(pp)) { 3735 unlock_page(pp); 3736 zfs_rangelock_exit(lr); 3737 3738 if (wbc->sync_mode != WB_SYNC_NONE) { 3739 /* 3740 * Speed up any non-sync page writebacks since 3741 * they may take several seconds to complete. 3742 * Refer to the comment in zpl_fsync() (when 3743 * HAVE_FSYNC_RANGE is defined) for details. 3744 */ 3745 if (atomic_load_32(&zp->z_async_writes_cnt) > 0) { 3746 zil_commit(zfsvfs->z_log, zp->z_id); 3747 } 3748 3749 if (PageWriteback(pp)) 3750 #ifdef HAVE_PAGEMAP_FOLIO_WAIT_BIT 3751 folio_wait_bit(page_folio(pp), PG_writeback); 3752 #else 3753 wait_on_page_bit(pp, PG_writeback); 3754 #endif 3755 } 3756 3757 zfs_exit(zfsvfs, FTAG); 3758 return (0); 3759 } 3760 3761 /* Clear the dirty flag the required locks are held */ 3762 if (!clear_page_dirty_for_io(pp)) { 3763 unlock_page(pp); 3764 zfs_rangelock_exit(lr); 3765 zfs_exit(zfsvfs, FTAG); 3766 return (0); 3767 } 3768 3769 /* 3770 * Counterpart for redirty_page_for_writepage() above. This page 3771 * was in fact not skipped and should not be counted as if it were. 3772 */ 3773 wbc->pages_skipped--; 3774 if (!for_sync) 3775 atomic_inc_32(&zp->z_async_writes_cnt); 3776 set_page_writeback(pp); 3777 unlock_page(pp); 3778 3779 tx = dmu_tx_create(zfsvfs->z_os); 3780 dmu_tx_hold_write(tx, zp->z_id, pgoff, pglen); 3781 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 3782 zfs_sa_upgrade_txholds(tx, zp); 3783 3784 err = dmu_tx_assign(tx, TXG_NOWAIT); 3785 if (err != 0) { 3786 if (err == ERESTART) 3787 dmu_tx_wait(tx); 3788 3789 dmu_tx_abort(tx); 3790 #ifdef HAVE_VFS_FILEMAP_DIRTY_FOLIO 3791 filemap_dirty_folio(page_mapping(pp), page_folio(pp)); 3792 #else 3793 __set_page_dirty_nobuffers(pp); 3794 #endif 3795 ClearPageError(pp); 3796 end_page_writeback(pp); 3797 if (!for_sync) 3798 atomic_dec_32(&zp->z_async_writes_cnt); 3799 zfs_rangelock_exit(lr); 3800 zfs_exit(zfsvfs, FTAG); 3801 return (err); 3802 } 3803 3804 va = kmap(pp); 3805 ASSERT3U(pglen, <=, PAGE_SIZE); 3806 dmu_write(zfsvfs->z_os, zp->z_id, pgoff, pglen, va, tx); 3807 kunmap(pp); 3808 3809 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 3810 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 3811 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_FLAGS(zfsvfs), NULL, 3812 &zp->z_pflags, 8); 3813 3814 /* Preserve the mtime and ctime provided by the inode */ 3815 ZFS_TIME_ENCODE(&ip->i_mtime, mtime); 3816 ZFS_TIME_ENCODE(&ip->i_ctime, ctime); 3817 zp->z_atime_dirty = B_FALSE; 3818 zp->z_seq++; 3819 3820 err = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx); 3821 3822 zfs_log_write(zfsvfs->z_log, tx, TX_WRITE, zp, pgoff, pglen, 0, 3823 for_sync ? zfs_putpage_sync_commit_cb : 3824 zfs_putpage_async_commit_cb, pp); 3825 3826 dmu_tx_commit(tx); 3827 3828 zfs_rangelock_exit(lr); 3829 3830 if (wbc->sync_mode != WB_SYNC_NONE) { 3831 /* 3832 * Note that this is rarely called under writepages(), because 3833 * writepages() normally handles the entire commit for 3834 * performance reasons. 3835 */ 3836 zil_commit(zfsvfs->z_log, zp->z_id); 3837 } else if (!for_sync && atomic_load_32(&zp->z_sync_writes_cnt) > 0) { 3838 /* 3839 * If the caller does not intend to wait synchronously 3840 * for this page writeback to complete and there are active 3841 * synchronous calls on this file, do a commit so that 3842 * the latter don't accidentally end up waiting for 3843 * our writeback to complete. Refer to the comment in 3844 * zpl_fsync() (when HAVE_FSYNC_RANGE is defined) for details. 3845 */ 3846 zil_commit(zfsvfs->z_log, zp->z_id); 3847 } 3848 3849 dataset_kstats_update_write_kstats(&zfsvfs->z_kstat, pglen); 3850 3851 zfs_exit(zfsvfs, FTAG); 3852 return (err); 3853 } 3854 3855 /* 3856 * Update the system attributes when the inode has been dirtied. For the 3857 * moment we only update the mode, atime, mtime, and ctime. 3858 */ 3859 int 3860 zfs_dirty_inode(struct inode *ip, int flags) 3861 { 3862 znode_t *zp = ITOZ(ip); 3863 zfsvfs_t *zfsvfs = ITOZSB(ip); 3864 dmu_tx_t *tx; 3865 uint64_t mode, atime[2], mtime[2], ctime[2]; 3866 sa_bulk_attr_t bulk[4]; 3867 int error = 0; 3868 int cnt = 0; 3869 3870 if (zfs_is_readonly(zfsvfs) || dmu_objset_is_snapshot(zfsvfs->z_os)) 3871 return (0); 3872 3873 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 3874 return (error); 3875 3876 #ifdef I_DIRTY_TIME 3877 /* 3878 * This is the lazytime semantic introduced in Linux 4.0 3879 * This flag will only be called from update_time when lazytime is set. 3880 * (Note, I_DIRTY_SYNC will also set if not lazytime) 3881 * Fortunately mtime and ctime are managed within ZFS itself, so we 3882 * only need to dirty atime. 3883 */ 3884 if (flags == I_DIRTY_TIME) { 3885 zp->z_atime_dirty = B_TRUE; 3886 goto out; 3887 } 3888 #endif 3889 3890 tx = dmu_tx_create(zfsvfs->z_os); 3891 3892 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 3893 zfs_sa_upgrade_txholds(tx, zp); 3894 3895 error = dmu_tx_assign(tx, TXG_WAIT); 3896 if (error) { 3897 dmu_tx_abort(tx); 3898 goto out; 3899 } 3900 3901 mutex_enter(&zp->z_lock); 3902 zp->z_atime_dirty = B_FALSE; 3903 3904 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MODE(zfsvfs), NULL, &mode, 8); 3905 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_ATIME(zfsvfs), NULL, &atime, 16); 3906 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_MTIME(zfsvfs), NULL, &mtime, 16); 3907 SA_ADD_BULK_ATTR(bulk, cnt, SA_ZPL_CTIME(zfsvfs), NULL, &ctime, 16); 3908 3909 /* Preserve the mode, mtime and ctime provided by the inode */ 3910 ZFS_TIME_ENCODE(&ip->i_atime, atime); 3911 ZFS_TIME_ENCODE(&ip->i_mtime, mtime); 3912 ZFS_TIME_ENCODE(&ip->i_ctime, ctime); 3913 mode = ip->i_mode; 3914 3915 zp->z_mode = mode; 3916 3917 error = sa_bulk_update(zp->z_sa_hdl, bulk, cnt, tx); 3918 mutex_exit(&zp->z_lock); 3919 3920 dmu_tx_commit(tx); 3921 out: 3922 zfs_exit(zfsvfs, FTAG); 3923 return (error); 3924 } 3925 3926 void 3927 zfs_inactive(struct inode *ip) 3928 { 3929 znode_t *zp = ITOZ(ip); 3930 zfsvfs_t *zfsvfs = ITOZSB(ip); 3931 uint64_t atime[2]; 3932 int error; 3933 int need_unlock = 0; 3934 3935 /* Only read lock if we haven't already write locked, e.g. rollback */ 3936 if (!RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock)) { 3937 need_unlock = 1; 3938 rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_READER); 3939 } 3940 if (zp->z_sa_hdl == NULL) { 3941 if (need_unlock) 3942 rw_exit(&zfsvfs->z_teardown_inactive_lock); 3943 return; 3944 } 3945 3946 if (zp->z_atime_dirty && zp->z_unlinked == B_FALSE) { 3947 dmu_tx_t *tx = dmu_tx_create(zfsvfs->z_os); 3948 3949 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 3950 zfs_sa_upgrade_txholds(tx, zp); 3951 error = dmu_tx_assign(tx, TXG_WAIT); 3952 if (error) { 3953 dmu_tx_abort(tx); 3954 } else { 3955 ZFS_TIME_ENCODE(&ip->i_atime, atime); 3956 mutex_enter(&zp->z_lock); 3957 (void) sa_update(zp->z_sa_hdl, SA_ZPL_ATIME(zfsvfs), 3958 (void *)&atime, sizeof (atime), tx); 3959 zp->z_atime_dirty = B_FALSE; 3960 mutex_exit(&zp->z_lock); 3961 dmu_tx_commit(tx); 3962 } 3963 } 3964 3965 zfs_zinactive(zp); 3966 if (need_unlock) 3967 rw_exit(&zfsvfs->z_teardown_inactive_lock); 3968 } 3969 3970 /* 3971 * Fill pages with data from the disk. 3972 */ 3973 static int 3974 zfs_fillpage(struct inode *ip, struct page *pp) 3975 { 3976 zfsvfs_t *zfsvfs = ITOZSB(ip); 3977 loff_t i_size = i_size_read(ip); 3978 u_offset_t io_off = page_offset(pp); 3979 size_t io_len = PAGE_SIZE; 3980 3981 ASSERT3U(io_off, <, i_size); 3982 3983 if (io_off + io_len > i_size) 3984 io_len = i_size - io_off; 3985 3986 void *va = kmap(pp); 3987 int error = dmu_read(zfsvfs->z_os, ITOZ(ip)->z_id, io_off, 3988 io_len, va, DMU_READ_PREFETCH); 3989 if (io_len != PAGE_SIZE) 3990 memset((char *)va + io_len, 0, PAGE_SIZE - io_len); 3991 kunmap(pp); 3992 3993 if (error) { 3994 /* convert checksum errors into IO errors */ 3995 if (error == ECKSUM) 3996 error = SET_ERROR(EIO); 3997 3998 SetPageError(pp); 3999 ClearPageUptodate(pp); 4000 } else { 4001 ClearPageError(pp); 4002 SetPageUptodate(pp); 4003 } 4004 4005 return (error); 4006 } 4007 4008 /* 4009 * Uses zfs_fillpage to read data from the file and fill the page. 4010 * 4011 * IN: ip - inode of file to get data from. 4012 * pp - page to read 4013 * 4014 * RETURN: 0 on success, error code on failure. 4015 * 4016 * Timestamps: 4017 * vp - atime updated 4018 */ 4019 int 4020 zfs_getpage(struct inode *ip, struct page *pp) 4021 { 4022 zfsvfs_t *zfsvfs = ITOZSB(ip); 4023 znode_t *zp = ITOZ(ip); 4024 int error; 4025 4026 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 4027 return (error); 4028 4029 error = zfs_fillpage(ip, pp); 4030 if (error == 0) 4031 dataset_kstats_update_read_kstats(&zfsvfs->z_kstat, PAGE_SIZE); 4032 4033 zfs_exit(zfsvfs, FTAG); 4034 4035 return (error); 4036 } 4037 4038 /* 4039 * Check ZFS specific permissions to memory map a section of a file. 4040 * 4041 * IN: ip - inode of the file to mmap 4042 * off - file offset 4043 * addrp - start address in memory region 4044 * len - length of memory region 4045 * vm_flags- address flags 4046 * 4047 * RETURN: 0 if success 4048 * error code if failure 4049 */ 4050 int 4051 zfs_map(struct inode *ip, offset_t off, caddr_t *addrp, size_t len, 4052 unsigned long vm_flags) 4053 { 4054 (void) addrp; 4055 znode_t *zp = ITOZ(ip); 4056 zfsvfs_t *zfsvfs = ITOZSB(ip); 4057 int error; 4058 4059 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 4060 return (error); 4061 4062 if ((vm_flags & VM_WRITE) && (zp->z_pflags & 4063 (ZFS_IMMUTABLE | ZFS_READONLY | ZFS_APPENDONLY))) { 4064 zfs_exit(zfsvfs, FTAG); 4065 return (SET_ERROR(EPERM)); 4066 } 4067 4068 if ((vm_flags & (VM_READ | VM_EXEC)) && 4069 (zp->z_pflags & ZFS_AV_QUARANTINED)) { 4070 zfs_exit(zfsvfs, FTAG); 4071 return (SET_ERROR(EACCES)); 4072 } 4073 4074 if (off < 0 || len > MAXOFFSET_T - off) { 4075 zfs_exit(zfsvfs, FTAG); 4076 return (SET_ERROR(ENXIO)); 4077 } 4078 4079 zfs_exit(zfsvfs, FTAG); 4080 return (0); 4081 } 4082 4083 /* 4084 * Free or allocate space in a file. Currently, this function only 4085 * supports the `F_FREESP' command. However, this command is somewhat 4086 * misnamed, as its functionality includes the ability to allocate as 4087 * well as free space. 4088 * 4089 * IN: zp - znode of file to free data in. 4090 * cmd - action to take (only F_FREESP supported). 4091 * bfp - section of file to free/alloc. 4092 * flag - current file open mode flags. 4093 * offset - current file offset. 4094 * cr - credentials of caller. 4095 * 4096 * RETURN: 0 on success, error code on failure. 4097 * 4098 * Timestamps: 4099 * zp - ctime|mtime updated 4100 */ 4101 int 4102 zfs_space(znode_t *zp, int cmd, flock64_t *bfp, int flag, 4103 offset_t offset, cred_t *cr) 4104 { 4105 (void) offset; 4106 zfsvfs_t *zfsvfs = ZTOZSB(zp); 4107 uint64_t off, len; 4108 int error; 4109 4110 if ((error = zfs_enter_verify_zp(zfsvfs, zp, FTAG)) != 0) 4111 return (error); 4112 4113 if (cmd != F_FREESP) { 4114 zfs_exit(zfsvfs, FTAG); 4115 return (SET_ERROR(EINVAL)); 4116 } 4117 4118 /* 4119 * Callers might not be able to detect properly that we are read-only, 4120 * so check it explicitly here. 4121 */ 4122 if (zfs_is_readonly(zfsvfs)) { 4123 zfs_exit(zfsvfs, FTAG); 4124 return (SET_ERROR(EROFS)); 4125 } 4126 4127 if (bfp->l_len < 0) { 4128 zfs_exit(zfsvfs, FTAG); 4129 return (SET_ERROR(EINVAL)); 4130 } 4131 4132 /* 4133 * Permissions aren't checked on Solaris because on this OS 4134 * zfs_space() can only be called with an opened file handle. 4135 * On Linux we can get here through truncate_range() which 4136 * operates directly on inodes, so we need to check access rights. 4137 */ 4138 if ((error = zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr, 4139 kcred->user_ns))) { 4140 zfs_exit(zfsvfs, FTAG); 4141 return (error); 4142 } 4143 4144 off = bfp->l_start; 4145 len = bfp->l_len; /* 0 means from off to end of file */ 4146 4147 error = zfs_freesp(zp, off, len, flag, TRUE); 4148 4149 zfs_exit(zfsvfs, FTAG); 4150 return (error); 4151 } 4152 4153 int 4154 zfs_fid(struct inode *ip, fid_t *fidp) 4155 { 4156 znode_t *zp = ITOZ(ip); 4157 zfsvfs_t *zfsvfs = ITOZSB(ip); 4158 uint32_t gen; 4159 uint64_t gen64; 4160 uint64_t object = zp->z_id; 4161 zfid_short_t *zfid; 4162 int size, i, error; 4163 4164 if ((error = zfs_enter(zfsvfs, FTAG)) != 0) 4165 return (error); 4166 4167 if (fidp->fid_len < SHORT_FID_LEN) { 4168 fidp->fid_len = SHORT_FID_LEN; 4169 zfs_exit(zfsvfs, FTAG); 4170 return (SET_ERROR(ENOSPC)); 4171 } 4172 4173 if ((error = zfs_verify_zp(zp)) != 0) { 4174 zfs_exit(zfsvfs, FTAG); 4175 return (error); 4176 } 4177 4178 if ((error = sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), 4179 &gen64, sizeof (uint64_t))) != 0) { 4180 zfs_exit(zfsvfs, FTAG); 4181 return (error); 4182 } 4183 4184 gen = (uint32_t)gen64; 4185 4186 size = SHORT_FID_LEN; 4187 4188 zfid = (zfid_short_t *)fidp; 4189 4190 zfid->zf_len = size; 4191 4192 for (i = 0; i < sizeof (zfid->zf_object); i++) 4193 zfid->zf_object[i] = (uint8_t)(object >> (8 * i)); 4194 4195 /* Must have a non-zero generation number to distinguish from .zfs */ 4196 if (gen == 0) 4197 gen = 1; 4198 for (i = 0; i < sizeof (zfid->zf_gen); i++) 4199 zfid->zf_gen[i] = (uint8_t)(gen >> (8 * i)); 4200 4201 zfs_exit(zfsvfs, FTAG); 4202 return (0); 4203 } 4204 4205 #if defined(_KERNEL) 4206 EXPORT_SYMBOL(zfs_open); 4207 EXPORT_SYMBOL(zfs_close); 4208 EXPORT_SYMBOL(zfs_lookup); 4209 EXPORT_SYMBOL(zfs_create); 4210 EXPORT_SYMBOL(zfs_tmpfile); 4211 EXPORT_SYMBOL(zfs_remove); 4212 EXPORT_SYMBOL(zfs_mkdir); 4213 EXPORT_SYMBOL(zfs_rmdir); 4214 EXPORT_SYMBOL(zfs_readdir); 4215 EXPORT_SYMBOL(zfs_getattr_fast); 4216 EXPORT_SYMBOL(zfs_setattr); 4217 EXPORT_SYMBOL(zfs_rename); 4218 EXPORT_SYMBOL(zfs_symlink); 4219 EXPORT_SYMBOL(zfs_readlink); 4220 EXPORT_SYMBOL(zfs_link); 4221 EXPORT_SYMBOL(zfs_inactive); 4222 EXPORT_SYMBOL(zfs_space); 4223 EXPORT_SYMBOL(zfs_fid); 4224 EXPORT_SYMBOL(zfs_getpage); 4225 EXPORT_SYMBOL(zfs_putpage); 4226 EXPORT_SYMBOL(zfs_dirty_inode); 4227 EXPORT_SYMBOL(zfs_map); 4228 4229 /* CSTYLED */ 4230 module_param(zfs_delete_blocks, ulong, 0644); 4231 MODULE_PARM_DESC(zfs_delete_blocks, "Delete files larger than N blocks async"); 4232 4233 #endif 4234