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 */ 68static int 69zfs_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 */ 110int 111zfs_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 200static int 201zfs_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 225int 226zfs_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 */ 269void 270zfs_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 */ 286void 287zfs_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 */ 367static int 368zfs_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 421extern taskq_t *zfsvfs_taskq; 422 423void 424zfs_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 zfs_znode_free(zp); 547 548 dmu_tx_commit(tx); 549 550 if (xattr_obj) { 551 /* 552 * We're using the FreeBSD taskqueue API here instead of 553 * the Solaris taskq API since the FreeBSD API allows for a 554 * task to be enqueued multiple times but executed once. 555 */ 556 taskqueue_enqueue(zfsvfs_taskq->tq_queue, 557 &zfsvfs->z_unlinked_drain_task); 558 } 559} 560 561static uint64_t 562zfs_dirent(znode_t *zp, uint64_t mode) 563{ 564 uint64_t de = zp->z_id; 565 566 if (zp->z_zfsvfs->z_version >= ZPL_VERSION_DIRENT_TYPE) 567 de |= IFTODT(mode) << 60; 568 return (de); 569} 570 571/* 572 * Link zp into dzp. Can only fail if zp has been unlinked. 573 */ 574int 575zfs_link_create(znode_t *dzp, const char *name, znode_t *zp, dmu_tx_t *tx, 576 int flag) 577{ 578 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 579 vnode_t *vp = ZTOV(zp); 580 uint64_t value; 581 int zp_is_dir = (vp->v_type == VDIR); 582 sa_bulk_attr_t bulk[5]; 583 uint64_t mtime[2], ctime[2]; 584 int count = 0; 585 int error; 586 587 if (zfsvfs->z_replay == B_FALSE) { 588 ASSERT_VOP_ELOCKED(ZTOV(dzp), __func__); 589 ASSERT_VOP_ELOCKED(ZTOV(zp), __func__); 590 } 591 if (zp_is_dir) { 592 if (dzp->z_links >= ZFS_LINK_MAX) 593 return (SET_ERROR(EMLINK)); 594 } 595 if (!(flag & ZRENAMING)) { 596 if (zp->z_unlinked) { /* no new links to unlinked zp */ 597 ASSERT(!(flag & (ZNEW | ZEXISTS))); 598 return (SET_ERROR(ENOENT)); 599 } 600 if (zp->z_links >= ZFS_LINK_MAX - zp_is_dir) { 601 return (SET_ERROR(EMLINK)); 602 } 603 zp->z_links++; 604 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 605 &zp->z_links, sizeof (zp->z_links)); 606 607 } else { 608 ASSERT(!zp->z_unlinked); 609 } 610 value = zfs_dirent(zp, zp->z_mode); 611 error = zap_add(zp->z_zfsvfs->z_os, dzp->z_id, name, 612 8, 1, &value, tx); 613 614 /* 615 * zap_add could fail to add the entry if it exceeds the capacity of the 616 * leaf-block and zap_leaf_split() failed to help. 617 * The caller of this routine is responsible for failing the transaction 618 * which will rollback the SA updates done above. 619 */ 620 if (error != 0) { 621 if (!(flag & ZRENAMING) && !(flag & ZNEW)) 622 zp->z_links--; 623 return (error); 624 } 625 626 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, 627 &dzp->z_id, sizeof (dzp->z_id)); 628 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 629 &zp->z_pflags, sizeof (zp->z_pflags)); 630 631 if (!(flag & ZNEW)) { 632 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 633 ctime, sizeof (ctime)); 634 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, 635 ctime); 636 } 637 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 638 ASSERT0(error); 639 640 dzp->z_size++; 641 dzp->z_links += zp_is_dir; 642 count = 0; 643 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 644 &dzp->z_size, sizeof (dzp->z_size)); 645 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 646 &dzp->z_links, sizeof (dzp->z_links)); 647 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 648 mtime, sizeof (mtime)); 649 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 650 ctime, sizeof (ctime)); 651 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 652 &dzp->z_pflags, sizeof (dzp->z_pflags)); 653 zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime); 654 error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx); 655 ASSERT0(error); 656 return (0); 657} 658 659/* 660 * The match type in the code for this function should conform to: 661 * 662 * ------------------------------------------------------------------------ 663 * fs type | z_norm | lookup type | match type 664 * ---------|-------------|-------------|---------------------------------- 665 * CS !norm | 0 | 0 | 0 (exact) 666 * CS norm | formX | 0 | MT_NORMALIZE 667 * CI !norm | upper | !ZCIEXACT | MT_NORMALIZE 668 * CI !norm | upper | ZCIEXACT | MT_NORMALIZE | MT_MATCH_CASE 669 * CI norm | upper|formX | !ZCIEXACT | MT_NORMALIZE 670 * CI norm | upper|formX | ZCIEXACT | MT_NORMALIZE | MT_MATCH_CASE 671 * CM !norm | upper | !ZCILOOK | MT_NORMALIZE | MT_MATCH_CASE 672 * CM !norm | upper | ZCILOOK | MT_NORMALIZE 673 * CM norm | upper|formX | !ZCILOOK | MT_NORMALIZE | MT_MATCH_CASE 674 * CM norm | upper|formX | ZCILOOK | MT_NORMALIZE 675 * 676 * Abbreviations: 677 * CS = Case Sensitive, CI = Case Insensitive, CM = Case Mixed 678 * upper = case folding set by fs type on creation (U8_TEXTPREP_TOUPPER) 679 * formX = unicode normalization form set on fs creation 680 */ 681static int 682zfs_dropname(znode_t *dzp, const char *name, znode_t *zp, dmu_tx_t *tx, 683 int flag) 684{ 685 int error; 686 687 if (zp->z_zfsvfs->z_norm) { 688 matchtype_t mt = MT_NORMALIZE; 689 690 if (zp->z_zfsvfs->z_case == ZFS_CASE_MIXED) { 691 mt |= MT_MATCH_CASE; 692 } 693 694 error = zap_remove_norm(zp->z_zfsvfs->z_os, dzp->z_id, 695 name, mt, tx); 696 } else { 697 error = zap_remove(zp->z_zfsvfs->z_os, dzp->z_id, name, tx); 698 } 699 700 return (error); 701} 702 703/* 704 * Unlink zp from dzp, and mark zp for deletion if this was the last link. 705 * Can fail if zp is a mount point (EBUSY) or a non-empty directory (EEXIST). 706 * If 'unlinkedp' is NULL, we put unlinked znodes on the unlinked list. 707 * If it's non-NULL, we use it to indicate whether the znode needs deletion, 708 * and it's the caller's job to do it. 709 */ 710int 711zfs_link_destroy(znode_t *dzp, const char *name, znode_t *zp, dmu_tx_t *tx, 712 int flag, boolean_t *unlinkedp) 713{ 714 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 715 vnode_t *vp = ZTOV(zp); 716 int zp_is_dir = (vp->v_type == VDIR); 717 boolean_t unlinked = B_FALSE; 718 sa_bulk_attr_t bulk[5]; 719 uint64_t mtime[2], ctime[2]; 720 int count = 0; 721 int error; 722 723 if (zfsvfs->z_replay == B_FALSE) { 724 ASSERT_VOP_ELOCKED(ZTOV(dzp), __func__); 725 ASSERT_VOP_ELOCKED(ZTOV(zp), __func__); 726 } 727 if (!(flag & ZRENAMING)) { 728 729 if (zp_is_dir && !zfs_dirempty(zp)) 730 return (SET_ERROR(ENOTEMPTY)); 731 732 /* 733 * If we get here, we are going to try to remove the object. 734 * First try removing the name from the directory; if that 735 * fails, return the error. 736 */ 737 error = zfs_dropname(dzp, name, zp, tx, flag); 738 if (error != 0) { 739 return (error); 740 } 741 742 if (zp->z_links <= zp_is_dir) { 743 zfs_panic_recover("zfs: link count on vnode %p is %u, " 744 "should be at least %u", zp->z_vnode, 745 (int)zp->z_links, 746 zp_is_dir + 1); 747 zp->z_links = zp_is_dir + 1; 748 } 749 if (--zp->z_links == zp_is_dir) { 750 zp->z_unlinked = B_TRUE; 751 zp->z_links = 0; 752 unlinked = B_TRUE; 753 } else { 754 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), 755 NULL, &ctime, sizeof (ctime)); 756 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 757 NULL, &zp->z_pflags, sizeof (zp->z_pflags)); 758 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, 759 ctime); 760 } 761 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), 762 NULL, &zp->z_links, sizeof (zp->z_links)); 763 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 764 count = 0; 765 ASSERT0(error); 766 } else { 767 ASSERT(!zp->z_unlinked); 768 error = zfs_dropname(dzp, name, zp, tx, flag); 769 if (error != 0) 770 return (error); 771 } 772 773 dzp->z_size--; /* one dirent removed */ 774 dzp->z_links -= zp_is_dir; /* ".." link from zp */ 775 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), 776 NULL, &dzp->z_links, sizeof (dzp->z_links)); 777 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), 778 NULL, &dzp->z_size, sizeof (dzp->z_size)); 779 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), 780 NULL, ctime, sizeof (ctime)); 781 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), 782 NULL, mtime, sizeof (mtime)); 783 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 784 NULL, &dzp->z_pflags, sizeof (dzp->z_pflags)); 785 zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime); 786 error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx); 787 ASSERT0(error); 788 789 if (unlinkedp != NULL) 790 *unlinkedp = unlinked; 791 else if (unlinked) 792 zfs_unlinked_add(zp, tx); 793 794 return (0); 795} 796 797/* 798 * Indicate whether the directory is empty. 799 */ 800boolean_t 801zfs_dirempty(znode_t *dzp) 802{ 803 return (dzp->z_size == 2); 804} 805 806int 807zfs_make_xattrdir(znode_t *zp, vattr_t *vap, znode_t **xvpp, cred_t *cr) 808{ 809 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 810 znode_t *xzp; 811 dmu_tx_t *tx; 812 int error; 813 zfs_acl_ids_t acl_ids; 814 boolean_t fuid_dirtied; 815 uint64_t parent __maybe_unused; 816 817 *xvpp = NULL; 818 819 if ((error = zfs_acl_ids_create(zp, IS_XATTR, vap, cr, NULL, 820 &acl_ids, NULL)) != 0) 821 return (error); 822 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, 0)) { 823 zfs_acl_ids_free(&acl_ids); 824 return (SET_ERROR(EDQUOT)); 825 } 826 827 getnewvnode_reserve_(); 828 829 tx = dmu_tx_create(zfsvfs->z_os); 830 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 831 ZFS_SA_BASE_ATTR_SIZE); 832 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 833 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 834 fuid_dirtied = zfsvfs->z_fuid_dirty; 835 if (fuid_dirtied) 836 zfs_fuid_txhold(zfsvfs, tx); 837 error = dmu_tx_assign(tx, TXG_WAIT); 838 if (error) { 839 zfs_acl_ids_free(&acl_ids); 840 dmu_tx_abort(tx); 841 getnewvnode_drop_reserve(); 842 return (error); 843 } 844 zfs_mknode(zp, vap, tx, cr, IS_XATTR, &xzp, &acl_ids); 845 846 if (fuid_dirtied) 847 zfs_fuid_sync(zfsvfs, tx); 848 849 ASSERT0(sa_lookup(xzp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), &parent, 850 sizeof (parent))); 851 ASSERT3U(parent, ==, zp->z_id); 852 853 VERIFY0(sa_update(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &xzp->z_id, 854 sizeof (xzp->z_id), tx)); 855 856 zfs_log_create(zfsvfs->z_log, tx, TX_MKXATTR, zp, xzp, "", NULL, 857 acl_ids.z_fuidp, vap); 858 859 zfs_acl_ids_free(&acl_ids); 860 dmu_tx_commit(tx); 861 862 getnewvnode_drop_reserve(); 863 864 *xvpp = xzp; 865 866 return (0); 867} 868 869/* 870 * Return a znode for the extended attribute directory for zp. 871 * ** If the directory does not already exist, it is created ** 872 * 873 * IN: zp - znode to obtain attribute directory from 874 * cr - credentials of caller 875 * flags - flags from the VOP_LOOKUP call 876 * 877 * OUT: xzpp - pointer to extended attribute znode 878 * 879 * RETURN: 0 on success 880 * error number on failure 881 */ 882int 883zfs_get_xattrdir(znode_t *zp, znode_t **xzpp, cred_t *cr, int flags) 884{ 885 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 886 znode_t *xzp; 887 vattr_t va; 888 int error; 889top: 890 error = zfs_dirent_lookup(zp, "", &xzp, ZXATTR); 891 if (error) 892 return (error); 893 894 if (xzp != NULL) { 895 *xzpp = xzp; 896 return (0); 897 } 898 899 900 if (!(flags & CREATE_XATTR_DIR)) 901 return (SET_ERROR(ENOATTR)); 902 903 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 904 return (SET_ERROR(EROFS)); 905 } 906 907 /* 908 * The ability to 'create' files in an attribute 909 * directory comes from the write_xattr permission on the base file. 910 * 911 * The ability to 'search' an attribute directory requires 912 * read_xattr permission on the base file. 913 * 914 * Once in a directory the ability to read/write attributes 915 * is controlled by the permissions on the attribute file. 916 */ 917 va.va_mask = AT_MODE | AT_UID | AT_GID; 918 va.va_type = VDIR; 919 va.va_mode = S_IFDIR | S_ISVTX | 0777; 920 zfs_fuid_map_ids(zp, cr, &va.va_uid, &va.va_gid); 921 922 error = zfs_make_xattrdir(zp, &va, xzpp, cr); 923 924 if (error == ERESTART) { 925 /* NB: we already did dmu_tx_wait() if necessary */ 926 goto top; 927 } 928 if (error == 0) 929 VOP_UNLOCK1(ZTOV(*xzpp)); 930 931 return (error); 932} 933 934/* 935 * Decide whether it is okay to remove within a sticky directory. 936 * 937 * In sticky directories, write access is not sufficient; 938 * you can remove entries from a directory only if: 939 * 940 * you own the directory, 941 * you own the entry, 942 * the entry is a plain file and you have write access, 943 * or you are privileged (checked in secpolicy...). 944 * 945 * The function returns 0 if remove access is granted. 946 */ 947int 948zfs_sticky_remove_access(znode_t *zdp, znode_t *zp, cred_t *cr) 949{ 950 uid_t uid; 951 uid_t downer; 952 uid_t fowner; 953 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 954 955 if (zdp->z_zfsvfs->z_replay) 956 return (0); 957 958 if ((zdp->z_mode & S_ISVTX) == 0) 959 return (0); 960 961 downer = zfs_fuid_map_id(zfsvfs, zdp->z_uid, cr, ZFS_OWNER); 962 fowner = zfs_fuid_map_id(zfsvfs, zp->z_uid, cr, ZFS_OWNER); 963 964 if ((uid = crgetuid(cr)) == downer || uid == fowner || 965 (ZTOV(zp)->v_type == VREG && 966 zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr, NULL) == 0)) 967 return (0); 968 else 969 return (secpolicy_vnode_remove(ZTOV(zp), cr)); 970} 971