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) 2013, 2016 by Delphix. All rights reserved. 25 * Copyright 2017 Nexenta Systems, Inc. 26 */ 27 28 #include <sys/types.h> 29 #include <sys/param.h> 30 #include <sys/time.h> 31 #include <sys/systm.h> 32 #include <sys/sysmacros.h> 33 #include <sys/resource.h> 34 #include <sys/vfs.h> 35 #include <sys/vnode.h> 36 #include <sys/file.h> 37 #include <sys/kmem.h> 38 #include <sys/uio.h> 39 #include <sys/cmn_err.h> 40 #include <sys/errno.h> 41 #include <sys/stat.h> 42 #include <sys/unistd.h> 43 #include <sys/sunddi.h> 44 #include <sys/random.h> 45 #include <sys/policy.h> 46 #include <sys/condvar.h> 47 #include <sys/callb.h> 48 #include <sys/smp.h> 49 #include <sys/zfs_dir.h> 50 #include <sys/zfs_acl.h> 51 #include <sys/fs/zfs.h> 52 #include <sys/zap.h> 53 #include <sys/dmu.h> 54 #include <sys/atomic.h> 55 #include <sys/zfs_ctldir.h> 56 #include <sys/zfs_fuid.h> 57 #include <sys/sa.h> 58 #include <sys/zfs_sa.h> 59 #include <sys/dmu_objset.h> 60 #include <sys/dsl_dir.h> 61 62 #include <sys/ccompat.h> 63 64 /* 65 * zfs_match_find() is used by zfs_dirent_lookup() to perform zap lookups 66 * of names after deciding which is the appropriate lookup interface. 67 */ 68 static int 69 zfs_match_find(zfsvfs_t *zfsvfs, znode_t *dzp, const char *name, 70 matchtype_t mt, uint64_t *zoid) 71 { 72 int error; 73 74 if (zfsvfs->z_norm) { 75 76 /* 77 * In the non-mixed case we only expect there would ever 78 * be one match, but we need to use the normalizing lookup. 79 */ 80 error = zap_lookup_norm(zfsvfs->z_os, dzp->z_id, name, 8, 1, 81 zoid, mt, NULL, 0, NULL); 82 } else { 83 error = zap_lookup(zfsvfs->z_os, dzp->z_id, name, 8, 1, zoid); 84 } 85 *zoid = ZFS_DIRENT_OBJ(*zoid); 86 87 return (error); 88 } 89 90 /* 91 * Look up a directory entry under a locked vnode. 92 * dvp being locked gives us a guarantee that there are no concurrent 93 * modification of the directory and, thus, if a node can be found in 94 * the directory, then it must not be unlinked. 95 * 96 * Input arguments: 97 * dzp - znode for directory 98 * name - name of entry to lock 99 * flag - ZNEW: if the entry already exists, fail with EEXIST. 100 * ZEXISTS: if the entry does not exist, fail with ENOENT. 101 * ZXATTR: we want dzp's xattr directory 102 * 103 * Output arguments: 104 * zpp - pointer to the znode for the entry (NULL if there isn't one) 105 * 106 * Return value: 0 on success or errno on failure. 107 * 108 * NOTE: Always checks for, and rejects, '.' and '..'. 109 */ 110 int 111 zfs_dirent_lookup(znode_t *dzp, const char *name, znode_t **zpp, int flag) 112 { 113 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 114 znode_t *zp; 115 matchtype_t mt = 0; 116 uint64_t zoid; 117 int error = 0; 118 119 if (zfsvfs->z_replay == B_FALSE) 120 ASSERT_VOP_LOCKED(ZTOV(dzp), __func__); 121 122 *zpp = NULL; 123 124 /* 125 * Verify that we are not trying to lock '.', '..', or '.zfs' 126 */ 127 if (name[0] == '.' && 128 (((name[1] == '\0') || (name[1] == '.' && name[2] == '\0')) || 129 (zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0))) 130 return (SET_ERROR(EEXIST)); 131 132 /* 133 * Case sensitivity and normalization preferences are set when 134 * the file system is created. These are stored in the 135 * zfsvfs->z_case and zfsvfs->z_norm fields. These choices 136 * affect how we perform zap lookups. 137 * 138 * When matching we may need to normalize & change case according to 139 * FS settings. 140 * 141 * Note that a normalized match is necessary for a case insensitive 142 * filesystem when the lookup request is not exact because normalization 143 * can fold case independent of normalizing code point sequences. 144 * 145 * See the table above zfs_dropname(). 146 */ 147 if (zfsvfs->z_norm != 0) { 148 mt = MT_NORMALIZE; 149 150 /* 151 * Determine if the match needs to honor the case specified in 152 * lookup, and if so keep track of that so that during 153 * normalization we don't fold case. 154 */ 155 if (zfsvfs->z_case == ZFS_CASE_MIXED) { 156 mt |= MT_MATCH_CASE; 157 } 158 } 159 160 /* 161 * Only look in or update the DNLC if we are looking for the 162 * name on a file system that does not require normalization 163 * or case folding. We can also look there if we happen to be 164 * on a non-normalizing, mixed sensitivity file system IF we 165 * are looking for the exact name. 166 * 167 * NB: we do not need to worry about this flag for ZFS_CASE_SENSITIVE 168 * because in that case MT_EXACT and MT_FIRST should produce exactly 169 * the same result. 170 */ 171 172 if (dzp->z_unlinked && !(flag & ZXATTR)) 173 return (ENOENT); 174 if (flag & ZXATTR) { 175 error = sa_lookup(dzp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &zoid, 176 sizeof (zoid)); 177 if (error == 0) 178 error = (zoid == 0 ? ENOENT : 0); 179 } else { 180 error = zfs_match_find(zfsvfs, dzp, name, mt, &zoid); 181 } 182 if (error) { 183 if (error != ENOENT || (flag & ZEXISTS)) { 184 return (error); 185 } 186 } else { 187 if (flag & ZNEW) { 188 return (SET_ERROR(EEXIST)); 189 } 190 error = zfs_zget(zfsvfs, zoid, &zp); 191 if (error) 192 return (error); 193 ASSERT(!zp->z_unlinked); 194 *zpp = zp; 195 } 196 197 return (0); 198 } 199 200 static int 201 zfs_dd_lookup(znode_t *dzp, znode_t **zpp) 202 { 203 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 204 znode_t *zp; 205 uint64_t parent; 206 int error; 207 208 #ifdef ZFS_DEBUG 209 if (zfsvfs->z_replay == B_FALSE) 210 ASSERT_VOP_LOCKED(ZTOV(dzp), __func__); 211 #endif 212 if (dzp->z_unlinked) 213 return (ENOENT); 214 215 if ((error = sa_lookup(dzp->z_sa_hdl, 216 SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0) 217 return (error); 218 219 error = zfs_zget(zfsvfs, parent, &zp); 220 if (error == 0) 221 *zpp = zp; 222 return (error); 223 } 224 225 int 226 zfs_dirlook(znode_t *dzp, const char *name, znode_t **zpp) 227 { 228 zfsvfs_t *zfsvfs __unused = dzp->z_zfsvfs; 229 znode_t *zp = NULL; 230 int error = 0; 231 232 #ifdef ZFS_DEBUG 233 if (zfsvfs->z_replay == B_FALSE) 234 ASSERT_VOP_LOCKED(ZTOV(dzp), __func__); 235 #endif 236 if (dzp->z_unlinked) 237 return (SET_ERROR(ENOENT)); 238 239 if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) { 240 *zpp = dzp; 241 } else if (name[0] == '.' && name[1] == '.' && name[2] == 0) { 242 error = zfs_dd_lookup(dzp, &zp); 243 if (error == 0) 244 *zpp = zp; 245 } else { 246 error = zfs_dirent_lookup(dzp, name, &zp, ZEXISTS); 247 if (error == 0) { 248 dzp->z_zn_prefetch = B_TRUE; /* enable prefetching */ 249 *zpp = zp; 250 } 251 } 252 return (error); 253 } 254 255 /* 256 * unlinked Set (formerly known as the "delete queue") Error Handling 257 * 258 * When dealing with the unlinked set, we dmu_tx_hold_zap(), but we 259 * don't specify the name of the entry that we will be manipulating. We 260 * also fib and say that we won't be adding any new entries to the 261 * unlinked set, even though we might (this is to lower the minimum file 262 * size that can be deleted in a full filesystem). So on the small 263 * chance that the nlink list is using a fat zap (ie. has more than 264 * 2000 entries), we *may* not pre-read a block that's needed. 265 * Therefore it is remotely possible for some of the assertions 266 * regarding the unlinked set below to fail due to i/o error. On a 267 * nondebug system, this will result in the space being leaked. 268 */ 269 void 270 zfs_unlinked_add(znode_t *zp, dmu_tx_t *tx) 271 { 272 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 273 274 ASSERT(zp->z_unlinked); 275 ASSERT3U(zp->z_links, ==, 0); 276 277 VERIFY0(zap_add_int(zfsvfs->z_os, zfsvfs->z_unlinkedobj, zp->z_id, tx)); 278 279 dataset_kstats_update_nunlinks_kstat(&zfsvfs->z_kstat, 1); 280 } 281 282 /* 283 * Clean up any znodes that had no links when we either crashed or 284 * (force) umounted the file system. 285 */ 286 void 287 zfs_unlinked_drain(zfsvfs_t *zfsvfs) 288 { 289 zap_cursor_t zc; 290 zap_attribute_t zap; 291 dmu_object_info_t doi; 292 znode_t *zp; 293 dmu_tx_t *tx; 294 int error; 295 296 /* 297 * Iterate over the contents of the unlinked set. 298 */ 299 for (zap_cursor_init(&zc, zfsvfs->z_os, zfsvfs->z_unlinkedobj); 300 zap_cursor_retrieve(&zc, &zap) == 0; 301 zap_cursor_advance(&zc)) { 302 303 /* 304 * See what kind of object we have in list 305 */ 306 307 error = dmu_object_info(zfsvfs->z_os, 308 zap.za_first_integer, &doi); 309 if (error != 0) 310 continue; 311 312 ASSERT((doi.doi_type == DMU_OT_PLAIN_FILE_CONTENTS) || 313 (doi.doi_type == DMU_OT_DIRECTORY_CONTENTS)); 314 /* 315 * We need to re-mark these list entries for deletion, 316 * so we pull them back into core and set zp->z_unlinked. 317 */ 318 error = zfs_zget(zfsvfs, zap.za_first_integer, &zp); 319 320 /* 321 * We may pick up znodes that are already marked for deletion. 322 * This could happen during the purge of an extended attribute 323 * directory. All we need to do is skip over them, since they 324 * are already in the system marked z_unlinked. 325 */ 326 if (error != 0) 327 continue; 328 329 vn_lock(ZTOV(zp), LK_EXCLUSIVE | LK_RETRY); 330 331 /* 332 * Due to changes in zfs_rmnode we need to make sure the 333 * link count is set to zero here. 334 */ 335 if (zp->z_links != 0) { 336 tx = dmu_tx_create(zfsvfs->z_os); 337 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 338 error = dmu_tx_assign(tx, TXG_WAIT); 339 if (error != 0) { 340 dmu_tx_abort(tx); 341 vput(ZTOV(zp)); 342 continue; 343 } 344 zp->z_links = 0; 345 VERIFY0(sa_update(zp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs), 346 &zp->z_links, sizeof (zp->z_links), tx)); 347 dmu_tx_commit(tx); 348 } 349 350 zp->z_unlinked = B_TRUE; 351 vput(ZTOV(zp)); 352 } 353 zap_cursor_fini(&zc); 354 } 355 356 /* 357 * Delete the entire contents of a directory. Return a count 358 * of the number of entries that could not be deleted. If we encounter 359 * an error, return a count of at least one so that the directory stays 360 * in the unlinked set. 361 * 362 * NOTE: this function assumes that the directory is inactive, 363 * so there is no need to lock its entries before deletion. 364 * Also, it assumes the directory contents is *only* regular 365 * files. 366 */ 367 static int 368 zfs_purgedir(znode_t *dzp) 369 { 370 zap_cursor_t zc; 371 zap_attribute_t zap; 372 znode_t *xzp; 373 dmu_tx_t *tx; 374 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 375 int skipped = 0; 376 int error; 377 378 for (zap_cursor_init(&zc, zfsvfs->z_os, dzp->z_id); 379 (error = zap_cursor_retrieve(&zc, &zap)) == 0; 380 zap_cursor_advance(&zc)) { 381 error = zfs_zget(zfsvfs, 382 ZFS_DIRENT_OBJ(zap.za_first_integer), &xzp); 383 if (error) { 384 skipped += 1; 385 continue; 386 } 387 388 vn_lock(ZTOV(xzp), LK_EXCLUSIVE | LK_RETRY); 389 ASSERT((ZTOV(xzp)->v_type == VREG) || 390 (ZTOV(xzp)->v_type == VLNK)); 391 392 tx = dmu_tx_create(zfsvfs->z_os); 393 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 394 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, zap.za_name); 395 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE); 396 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 397 /* Is this really needed ? */ 398 zfs_sa_upgrade_txholds(tx, xzp); 399 dmu_tx_mark_netfree(tx); 400 error = dmu_tx_assign(tx, TXG_WAIT); 401 if (error) { 402 dmu_tx_abort(tx); 403 vput(ZTOV(xzp)); 404 skipped += 1; 405 continue; 406 } 407 408 error = zfs_link_destroy(dzp, zap.za_name, xzp, tx, 0, NULL); 409 if (error) 410 skipped += 1; 411 dmu_tx_commit(tx); 412 413 vput(ZTOV(xzp)); 414 } 415 zap_cursor_fini(&zc); 416 if (error != ENOENT) 417 skipped += 1; 418 return (skipped); 419 } 420 421 extern taskq_t *zfsvfs_taskq; 422 423 void 424 zfs_rmnode(znode_t *zp) 425 { 426 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 427 objset_t *os = zfsvfs->z_os; 428 dmu_tx_t *tx; 429 uint64_t z_id = zp->z_id; 430 uint64_t acl_obj; 431 uint64_t xattr_obj; 432 uint64_t count; 433 int error; 434 435 ASSERT3U(zp->z_links, ==, 0); 436 if (zfsvfs->z_replay == B_FALSE) 437 ASSERT_VOP_ELOCKED(ZTOV(zp), __func__); 438 439 /* 440 * If this is an attribute directory, purge its contents. 441 */ 442 if (ZTOV(zp) != NULL && ZTOV(zp)->v_type == VDIR && 443 (zp->z_pflags & ZFS_XATTR)) { 444 if (zfs_purgedir(zp) != 0) { 445 /* 446 * Not enough space to delete some xattrs. 447 * Leave it in the unlinked set. 448 */ 449 ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id); 450 zfs_znode_dmu_fini(zp); 451 zfs_znode_free(zp); 452 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 453 return; 454 } 455 } else { 456 /* 457 * Free up all the data in the file. We don't do this for 458 * XATTR directories because we need truncate and remove to be 459 * in the same tx, like in zfs_znode_delete(). Otherwise, if 460 * we crash here we'll end up with an inconsistent truncated 461 * zap object in the delete queue. Note a truncated file is 462 * harmless since it only contains user data. 463 */ 464 error = dmu_free_long_range(os, zp->z_id, 0, DMU_OBJECT_END); 465 if (error) { 466 /* 467 * Not enough space or we were interrupted by unmount. 468 * Leave the file in the unlinked set. 469 */ 470 ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id); 471 zfs_znode_dmu_fini(zp); 472 zfs_znode_free(zp); 473 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 474 return; 475 } 476 } 477 478 /* 479 * If the file has extended attributes, we're going to unlink 480 * the xattr dir. 481 */ 482 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 483 &xattr_obj, sizeof (xattr_obj)); 484 if (error) 485 xattr_obj = 0; 486 487 acl_obj = zfs_external_acl(zp); 488 489 /* 490 * Set up the final transaction. 491 */ 492 tx = dmu_tx_create(os); 493 dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END); 494 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 495 if (xattr_obj) 496 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, TRUE, NULL); 497 if (acl_obj) 498 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 499 500 zfs_sa_upgrade_txholds(tx, zp); 501 error = dmu_tx_assign(tx, TXG_WAIT); 502 if (error) { 503 /* 504 * Not enough space to delete the file. Leave it in the 505 * unlinked set, leaking it until the fs is remounted (at 506 * which point we'll call zfs_unlinked_drain() to process it). 507 */ 508 dmu_tx_abort(tx); 509 ZFS_OBJ_HOLD_ENTER(zfsvfs, z_id); 510 zfs_znode_dmu_fini(zp); 511 zfs_znode_free(zp); 512 ZFS_OBJ_HOLD_EXIT(zfsvfs, z_id); 513 return; 514 } 515 516 /* 517 * FreeBSD's implementation of zfs_zget requires a vnode to back it. 518 * This means that we could end up calling into getnewvnode while 519 * calling zfs_rmnode as a result of a prior call to getnewvnode 520 * trying to clear vnodes out of the cache. If this repeats we can 521 * recurse enough that we overflow our stack. To avoid this, we 522 * avoid calling zfs_zget on the xattr znode and instead simply add 523 * it to the unlinked set and schedule a call to zfs_unlinked_drain. 524 */ 525 if (xattr_obj) { 526 /* Add extended attribute directory to the unlinked set. */ 527 VERIFY3U(0, ==, 528 zap_add_int(os, zfsvfs->z_unlinkedobj, xattr_obj, tx)); 529 } 530 531 mutex_enter(&os->os_dsl_dataset->ds_dir->dd_activity_lock); 532 533 /* Remove this znode from the unlinked set */ 534 VERIFY3U(0, ==, 535 zap_remove_int(os, zfsvfs->z_unlinkedobj, zp->z_id, tx)); 536 537 if (zap_count(os, zfsvfs->z_unlinkedobj, &count) == 0 && count == 0) { 538 cv_broadcast(&os->os_dsl_dataset->ds_dir->dd_activity_cv); 539 } 540 541 mutex_exit(&os->os_dsl_dataset->ds_dir->dd_activity_lock); 542 543 dataset_kstats_update_nunlinked_kstat(&zfsvfs->z_kstat, 1); 544 545 zfs_znode_delete(zp, tx); 546 547 dmu_tx_commit(tx); 548 549 if (xattr_obj) { 550 /* 551 * We're using the FreeBSD taskqueue API here instead of 552 * the Solaris taskq API since the FreeBSD API allows for a 553 * task to be enqueued multiple times but executed once. 554 */ 555 taskqueue_enqueue(zfsvfs_taskq->tq_queue, 556 &zfsvfs->z_unlinked_drain_task); 557 } 558 } 559 560 static uint64_t 561 zfs_dirent(znode_t *zp, uint64_t mode) 562 { 563 uint64_t de = zp->z_id; 564 565 if (zp->z_zfsvfs->z_version >= ZPL_VERSION_DIRENT_TYPE) 566 de |= IFTODT(mode) << 60; 567 return (de); 568 } 569 570 /* 571 * Link zp into dzp. Can only fail if zp has been unlinked. 572 */ 573 int 574 zfs_link_create(znode_t *dzp, const char *name, znode_t *zp, dmu_tx_t *tx, 575 int flag) 576 { 577 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 578 vnode_t *vp = ZTOV(zp); 579 uint64_t value; 580 int zp_is_dir = (vp->v_type == VDIR); 581 sa_bulk_attr_t bulk[5]; 582 uint64_t mtime[2], ctime[2]; 583 int count = 0; 584 int error; 585 586 if (zfsvfs->z_replay == B_FALSE) { 587 ASSERT_VOP_ELOCKED(ZTOV(dzp), __func__); 588 ASSERT_VOP_ELOCKED(ZTOV(zp), __func__); 589 } 590 if (zp_is_dir) { 591 if (dzp->z_links >= ZFS_LINK_MAX) 592 return (SET_ERROR(EMLINK)); 593 } 594 if (!(flag & ZRENAMING)) { 595 if (zp->z_unlinked) { /* no new links to unlinked zp */ 596 ASSERT(!(flag & (ZNEW | ZEXISTS))); 597 return (SET_ERROR(ENOENT)); 598 } 599 if (zp->z_links >= ZFS_LINK_MAX - zp_is_dir) { 600 return (SET_ERROR(EMLINK)); 601 } 602 zp->z_links++; 603 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 604 &zp->z_links, sizeof (zp->z_links)); 605 606 } else { 607 ASSERT(!zp->z_unlinked); 608 } 609 value = zfs_dirent(zp, zp->z_mode); 610 error = zap_add(zp->z_zfsvfs->z_os, dzp->z_id, name, 611 8, 1, &value, tx); 612 613 /* 614 * zap_add could fail to add the entry if it exceeds the capacity of the 615 * leaf-block and zap_leaf_split() failed to help. 616 * The caller of this routine is responsible for failing the transaction 617 * which will rollback the SA updates done above. 618 */ 619 if (error != 0) { 620 if (!(flag & ZRENAMING) && !(flag & ZNEW)) 621 zp->z_links--; 622 return (error); 623 } 624 625 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, 626 &dzp->z_id, sizeof (dzp->z_id)); 627 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 628 &zp->z_pflags, sizeof (zp->z_pflags)); 629 630 if (!(flag & ZNEW)) { 631 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 632 ctime, sizeof (ctime)); 633 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, 634 ctime); 635 } 636 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 637 ASSERT0(error); 638 639 dzp->z_size++; 640 dzp->z_links += zp_is_dir; 641 count = 0; 642 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 643 &dzp->z_size, sizeof (dzp->z_size)); 644 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 645 &dzp->z_links, sizeof (dzp->z_links)); 646 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 647 mtime, sizeof (mtime)); 648 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 649 ctime, sizeof (ctime)); 650 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 651 &dzp->z_pflags, sizeof (dzp->z_pflags)); 652 zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime); 653 error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx); 654 ASSERT0(error); 655 return (0); 656 } 657 658 /* 659 * The match type in the code for this function should conform to: 660 * 661 * ------------------------------------------------------------------------ 662 * fs type | z_norm | lookup type | match type 663 * ---------|-------------|-------------|---------------------------------- 664 * CS !norm | 0 | 0 | 0 (exact) 665 * CS norm | formX | 0 | MT_NORMALIZE 666 * CI !norm | upper | !ZCIEXACT | MT_NORMALIZE 667 * CI !norm | upper | ZCIEXACT | MT_NORMALIZE | MT_MATCH_CASE 668 * CI norm | upper|formX | !ZCIEXACT | MT_NORMALIZE 669 * CI norm | upper|formX | ZCIEXACT | MT_NORMALIZE | MT_MATCH_CASE 670 * CM !norm | upper | !ZCILOOK | MT_NORMALIZE | MT_MATCH_CASE 671 * CM !norm | upper | ZCILOOK | MT_NORMALIZE 672 * CM norm | upper|formX | !ZCILOOK | MT_NORMALIZE | MT_MATCH_CASE 673 * CM norm | upper|formX | ZCILOOK | MT_NORMALIZE 674 * 675 * Abbreviations: 676 * CS = Case Sensitive, CI = Case Insensitive, CM = Case Mixed 677 * upper = case folding set by fs type on creation (U8_TEXTPREP_TOUPPER) 678 * formX = unicode normalization form set on fs creation 679 */ 680 static int 681 zfs_dropname(znode_t *dzp, const char *name, znode_t *zp, dmu_tx_t *tx, 682 int flag) 683 { 684 int error; 685 686 if (zp->z_zfsvfs->z_norm) { 687 matchtype_t mt = MT_NORMALIZE; 688 689 if (zp->z_zfsvfs->z_case == ZFS_CASE_MIXED) { 690 mt |= MT_MATCH_CASE; 691 } 692 693 error = zap_remove_norm(zp->z_zfsvfs->z_os, dzp->z_id, 694 name, mt, tx); 695 } else { 696 error = zap_remove(zp->z_zfsvfs->z_os, dzp->z_id, name, tx); 697 } 698 699 return (error); 700 } 701 702 /* 703 * Unlink zp from dzp, and mark zp for deletion if this was the last link. 704 * Can fail if zp is a mount point (EBUSY) or a non-empty directory (EEXIST). 705 * If 'unlinkedp' is NULL, we put unlinked znodes on the unlinked list. 706 * If it's non-NULL, we use it to indicate whether the znode needs deletion, 707 * and it's the caller's job to do it. 708 */ 709 int 710 zfs_link_destroy(znode_t *dzp, const char *name, znode_t *zp, dmu_tx_t *tx, 711 int flag, boolean_t *unlinkedp) 712 { 713 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 714 vnode_t *vp = ZTOV(zp); 715 int zp_is_dir = (vp->v_type == VDIR); 716 boolean_t unlinked = B_FALSE; 717 sa_bulk_attr_t bulk[5]; 718 uint64_t mtime[2], ctime[2]; 719 int count = 0; 720 int error; 721 722 if (zfsvfs->z_replay == B_FALSE) { 723 ASSERT_VOP_ELOCKED(ZTOV(dzp), __func__); 724 ASSERT_VOP_ELOCKED(ZTOV(zp), __func__); 725 } 726 if (!(flag & ZRENAMING)) { 727 728 if (zp_is_dir && !zfs_dirempty(zp)) 729 return (SET_ERROR(ENOTEMPTY)); 730 731 /* 732 * If we get here, we are going to try to remove the object. 733 * First try removing the name from the directory; if that 734 * fails, return the error. 735 */ 736 error = zfs_dropname(dzp, name, zp, tx, flag); 737 if (error != 0) { 738 return (error); 739 } 740 741 if (zp->z_links <= zp_is_dir) { 742 zfs_panic_recover("zfs: link count on vnode %p is %u, " 743 "should be at least %u", zp->z_vnode, 744 (int)zp->z_links, 745 zp_is_dir + 1); 746 zp->z_links = zp_is_dir + 1; 747 } 748 if (--zp->z_links == zp_is_dir) { 749 zp->z_unlinked = B_TRUE; 750 zp->z_links = 0; 751 unlinked = B_TRUE; 752 } else { 753 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), 754 NULL, &ctime, sizeof (ctime)); 755 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 756 NULL, &zp->z_pflags, sizeof (zp->z_pflags)); 757 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, 758 ctime); 759 } 760 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), 761 NULL, &zp->z_links, sizeof (zp->z_links)); 762 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 763 count = 0; 764 ASSERT0(error); 765 } else { 766 ASSERT(!zp->z_unlinked); 767 error = zfs_dropname(dzp, name, zp, tx, flag); 768 if (error != 0) 769 return (error); 770 } 771 772 dzp->z_size--; /* one dirent removed */ 773 dzp->z_links -= zp_is_dir; /* ".." link from zp */ 774 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), 775 NULL, &dzp->z_links, sizeof (dzp->z_links)); 776 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), 777 NULL, &dzp->z_size, sizeof (dzp->z_size)); 778 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), 779 NULL, ctime, sizeof (ctime)); 780 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), 781 NULL, mtime, sizeof (mtime)); 782 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 783 NULL, &dzp->z_pflags, sizeof (dzp->z_pflags)); 784 zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime); 785 error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx); 786 ASSERT0(error); 787 788 if (unlinkedp != NULL) 789 *unlinkedp = unlinked; 790 else if (unlinked) 791 zfs_unlinked_add(zp, tx); 792 793 return (0); 794 } 795 796 /* 797 * Indicate whether the directory is empty. 798 */ 799 boolean_t 800 zfs_dirempty(znode_t *dzp) 801 { 802 return (dzp->z_size == 2); 803 } 804 805 int 806 zfs_make_xattrdir(znode_t *zp, vattr_t *vap, znode_t **xvpp, cred_t *cr) 807 { 808 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 809 znode_t *xzp; 810 dmu_tx_t *tx; 811 int error; 812 zfs_acl_ids_t acl_ids; 813 boolean_t fuid_dirtied; 814 uint64_t parent __maybe_unused; 815 816 *xvpp = NULL; 817 818 if ((error = zfs_acl_ids_create(zp, IS_XATTR, vap, cr, NULL, 819 &acl_ids, NULL)) != 0) 820 return (error); 821 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, 0)) { 822 zfs_acl_ids_free(&acl_ids); 823 return (SET_ERROR(EDQUOT)); 824 } 825 826 getnewvnode_reserve_(); 827 828 tx = dmu_tx_create(zfsvfs->z_os); 829 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 830 ZFS_SA_BASE_ATTR_SIZE); 831 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 832 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 833 fuid_dirtied = zfsvfs->z_fuid_dirty; 834 if (fuid_dirtied) 835 zfs_fuid_txhold(zfsvfs, tx); 836 error = dmu_tx_assign(tx, TXG_WAIT); 837 if (error) { 838 zfs_acl_ids_free(&acl_ids); 839 dmu_tx_abort(tx); 840 getnewvnode_drop_reserve(); 841 return (error); 842 } 843 zfs_mknode(zp, vap, tx, cr, IS_XATTR, &xzp, &acl_ids); 844 845 if (fuid_dirtied) 846 zfs_fuid_sync(zfsvfs, tx); 847 848 ASSERT0(sa_lookup(xzp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), &parent, 849 sizeof (parent))); 850 ASSERT3U(parent, ==, zp->z_id); 851 852 VERIFY0(sa_update(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &xzp->z_id, 853 sizeof (xzp->z_id), tx)); 854 855 zfs_log_create(zfsvfs->z_log, tx, TX_MKXATTR, zp, xzp, "", NULL, 856 acl_ids.z_fuidp, vap); 857 858 zfs_acl_ids_free(&acl_ids); 859 dmu_tx_commit(tx); 860 861 getnewvnode_drop_reserve(); 862 863 *xvpp = xzp; 864 865 return (0); 866 } 867 868 /* 869 * Return a znode for the extended attribute directory for zp. 870 * ** If the directory does not already exist, it is created ** 871 * 872 * IN: zp - znode to obtain attribute directory from 873 * cr - credentials of caller 874 * flags - flags from the VOP_LOOKUP call 875 * 876 * OUT: xzpp - pointer to extended attribute znode 877 * 878 * RETURN: 0 on success 879 * error number on failure 880 */ 881 int 882 zfs_get_xattrdir(znode_t *zp, znode_t **xzpp, cred_t *cr, int flags) 883 { 884 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 885 znode_t *xzp; 886 vattr_t va; 887 int error; 888 top: 889 error = zfs_dirent_lookup(zp, "", &xzp, ZXATTR); 890 if (error) 891 return (error); 892 893 if (xzp != NULL) { 894 *xzpp = xzp; 895 return (0); 896 } 897 898 899 if (!(flags & CREATE_XATTR_DIR)) 900 return (SET_ERROR(ENOATTR)); 901 902 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 903 return (SET_ERROR(EROFS)); 904 } 905 906 /* 907 * The ability to 'create' files in an attribute 908 * directory comes from the write_xattr permission on the base file. 909 * 910 * The ability to 'search' an attribute directory requires 911 * read_xattr permission on the base file. 912 * 913 * Once in a directory the ability to read/write attributes 914 * is controlled by the permissions on the attribute file. 915 */ 916 va.va_mask = AT_MODE | AT_UID | AT_GID; 917 va.va_type = VDIR; 918 va.va_mode = S_IFDIR | S_ISVTX | 0777; 919 zfs_fuid_map_ids(zp, cr, &va.va_uid, &va.va_gid); 920 921 error = zfs_make_xattrdir(zp, &va, xzpp, cr); 922 923 if (error == ERESTART) { 924 /* NB: we already did dmu_tx_wait() if necessary */ 925 goto top; 926 } 927 if (error == 0) 928 VOP_UNLOCK1(ZTOV(*xzpp)); 929 930 return (error); 931 } 932 933 /* 934 * Decide whether it is okay to remove within a sticky directory. 935 * 936 * In sticky directories, write access is not sufficient; 937 * you can remove entries from a directory only if: 938 * 939 * you own the directory, 940 * you own the entry, 941 * the entry is a plain file and you have write access, 942 * or you are privileged (checked in secpolicy...). 943 * 944 * The function returns 0 if remove access is granted. 945 */ 946 int 947 zfs_sticky_remove_access(znode_t *zdp, znode_t *zp, cred_t *cr) 948 { 949 uid_t uid; 950 uid_t downer; 951 uid_t fowner; 952 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 953 954 if (zdp->z_zfsvfs->z_replay) 955 return (0); 956 957 if ((zdp->z_mode & S_ISVTX) == 0) 958 return (0); 959 960 downer = zfs_fuid_map_id(zfsvfs, zdp->z_uid, cr, ZFS_OWNER); 961 fowner = zfs_fuid_map_id(zfsvfs, zp->z_uid, cr, ZFS_OWNER); 962 963 if ((uid = crgetuid(cr)) == downer || uid == fowner || 964 (ZTOV(zp)->v_type == VREG && 965 zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr, NULL) == 0)) 966 return (0); 967 else 968 return (secpolicy_vnode_remove(ZTOV(zp), cr)); 969 } 970