1/* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 22/* 23 * Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved. 24 * Copyright (c) 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/extdirent.h> 37#include <sys/file.h> 38#include <sys/kmem.h> 39#include <sys/uio.h> 40#include <sys/cmn_err.h> 41#include <sys/errno.h> 42#include <sys/stat.h> 43#include <sys/unistd.h> 44#include <sys/sunddi.h> 45#include <sys/random.h> 46#include <sys/policy.h> 47#include <sys/condvar.h> 48#include <sys/callb.h> 49#include <sys/smp.h> 50#include <sys/zfs_dir.h> 51#include <sys/zfs_acl.h> 52#include <sys/fs/zfs.h> 53#include <sys/zap.h> 54#include <sys/dmu.h> 55#include <sys/atomic.h> 56#include <sys/zfs_ctldir.h> 57#include <sys/zfs_fuid.h> 58#include <sys/sa.h> 59#include <sys/zfs_sa.h> 60#include <sys/dmu_objset.h> 61#include <sys/dsl_dir.h> 62 63#include <sys/ccompat.h> 64 65/* 66 * zfs_match_find() is used by zfs_dirent_lookup() to perform zap lookups 67 * of names after deciding which is the appropriate lookup interface. 68 */ 69static int 70zfs_match_find(zfsvfs_t *zfsvfs, znode_t *dzp, const char *name, 71 matchtype_t mt, uint64_t *zoid) 72{ 73 int error; 74 75 if (zfsvfs->z_norm) { 76 77 /* 78 * In the non-mixed case we only expect there would ever 79 * be one match, but we need to use the normalizing lookup. 80 */ 81 error = zap_lookup_norm(zfsvfs->z_os, dzp->z_id, name, 8, 1, 82 zoid, mt, NULL, 0, NULL); 83 } else { 84 error = zap_lookup(zfsvfs->z_os, dzp->z_id, name, 8, 1, zoid); 85 } 86 *zoid = ZFS_DIRENT_OBJ(*zoid); 87 88 return (error); 89} 90 91/* 92 * Look up a directory entry under a locked vnode. 93 * dvp being locked gives us a guarantee that there are no concurrent 94 * modification of the directory and, thus, if a node can be found in 95 * the directory, then it must not be unlinked. 96 * 97 * Input arguments: 98 * dzp - znode for directory 99 * name - name of entry to lock 100 * flag - ZNEW: if the entry already exists, fail with EEXIST. 101 * ZEXISTS: if the entry does not exist, fail with ENOENT. 102 * ZXATTR: we want dzp's xattr directory 103 * 104 * Output arguments: 105 * zpp - pointer to the znode for the entry (NULL if there isn't one) 106 * 107 * Return value: 0 on success or errno on failure. 108 * 109 * NOTE: Always checks for, and rejects, '.' and '..'. 110 */ 111int 112zfs_dirent_lookup(znode_t *dzp, const char *name, znode_t **zpp, int flag) 113{ 114 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 115 znode_t *zp; 116 matchtype_t mt = 0; 117 uint64_t zoid; 118 int error = 0; 119 120 if (zfsvfs->z_replay == B_FALSE) 121 ASSERT_VOP_LOCKED(ZTOV(dzp), __func__); 122 123 *zpp = NULL; 124 125 /* 126 * Verify that we are not trying to lock '.', '..', or '.zfs' 127 */ 128 if (name[0] == '.' && 129 (((name[1] == '\0') || (name[1] == '.' && name[2] == '\0')) || 130 (zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0))) 131 return (SET_ERROR(EEXIST)); 132 133 /* 134 * Case sensitivity and normalization preferences are set when 135 * the file system is created. These are stored in the 136 * zfsvfs->z_case and zfsvfs->z_norm fields. These choices 137 * affect how we perform zap lookups. 138 * 139 * When matching we may need to normalize & change case according to 140 * FS settings. 141 * 142 * Note that a normalized match is necessary for a case insensitive 143 * filesystem when the lookup request is not exact because normalization 144 * can fold case independent of normalizing code point sequences. 145 * 146 * See the table above zfs_dropname(). 147 */ 148 if (zfsvfs->z_norm != 0) { 149 mt = MT_NORMALIZE; 150 151 /* 152 * Determine if the match needs to honor the case specified in 153 * lookup, and if so keep track of that so that during 154 * normalization we don't fold case. 155 */ 156 if (zfsvfs->z_case == ZFS_CASE_MIXED) { 157 mt |= MT_MATCH_CASE; 158 } 159 } 160 161 /* 162 * Only look in or update the DNLC if we are looking for the 163 * name on a file system that does not require normalization 164 * or case folding. We can also look there if we happen to be 165 * on a non-normalizing, mixed sensitivity file system IF we 166 * are looking for the exact name. 167 * 168 * NB: we do not need to worry about this flag for ZFS_CASE_SENSITIVE 169 * because in that case MT_EXACT and MT_FIRST should produce exactly 170 * the same result. 171 */ 172 173 if (dzp->z_unlinked && !(flag & ZXATTR)) 174 return (ENOENT); 175 if (flag & ZXATTR) { 176 error = sa_lookup(dzp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &zoid, 177 sizeof (zoid)); 178 if (error == 0) 179 error = (zoid == 0 ? ENOENT : 0); 180 } else { 181 error = zfs_match_find(zfsvfs, dzp, name, mt, &zoid); 182 } 183 if (error) { 184 if (error != ENOENT || (flag & ZEXISTS)) { 185 return (error); 186 } 187 } else { 188 if (flag & ZNEW) { 189 return (SET_ERROR(EEXIST)); 190 } 191 error = zfs_zget(zfsvfs, zoid, &zp); 192 if (error) 193 return (error); 194 ASSERT(!zp->z_unlinked); 195 *zpp = zp; 196 } 197 198 return (0); 199} 200 201static int 202zfs_dd_lookup(znode_t *dzp, znode_t **zpp) 203{ 204 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 205 znode_t *zp; 206 uint64_t parent; 207 int error; 208 209#ifdef ZFS_DEBUG 210 if (zfsvfs->z_replay == B_FALSE) 211 ASSERT_VOP_LOCKED(ZTOV(dzp), __func__); 212#endif 213 if (dzp->z_unlinked) 214 return (ENOENT); 215 216 if ((error = sa_lookup(dzp->z_sa_hdl, 217 SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0) 218 return (error); 219 220 error = zfs_zget(zfsvfs, parent, &zp); 221 if (error == 0) 222 *zpp = zp; 223 return (error); 224} 225 226int 227zfs_dirlook(znode_t *dzp, const char *name, znode_t **zpp) 228{ 229 zfsvfs_t *zfsvfs __unused = dzp->z_zfsvfs; 230 znode_t *zp = NULL; 231 int error = 0; 232 233#ifdef ZFS_DEBUG 234 if (zfsvfs->z_replay == B_FALSE) 235 ASSERT_VOP_LOCKED(ZTOV(dzp), __func__); 236#endif 237 if (dzp->z_unlinked) 238 return (SET_ERROR(ENOENT)); 239 240 if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) { 241 *zpp = dzp; 242 } else if (name[0] == '.' && name[1] == '.' && name[2] == 0) { 243 error = zfs_dd_lookup(dzp, &zp); 244 if (error == 0) 245 *zpp = zp; 246 } else { 247 error = zfs_dirent_lookup(dzp, name, &zp, ZEXISTS); 248 if (error == 0) { 249 dzp->z_zn_prefetch = B_TRUE; /* enable prefetching */ 250 *zpp = zp; 251 } 252 } 253 return (error); 254} 255 256/* 257 * unlinked Set (formerly known as the "delete queue") Error Handling 258 * 259 * When dealing with the unlinked set, we dmu_tx_hold_zap(), but we 260 * don't specify the name of the entry that we will be manipulating. We 261 * also fib and say that we won't be adding any new entries to the 262 * unlinked set, even though we might (this is to lower the minimum file 263 * size that can be deleted in a full filesystem). So on the small 264 * chance that the nlink list is using a fat zap (ie. has more than 265 * 2000 entries), we *may* not pre-read a block that's needed. 266 * Therefore it is remotely possible for some of the assertions 267 * regarding the unlinked set below to fail due to i/o error. On a 268 * nondebug system, this will result in the space being leaked. 269 */ 270void 271zfs_unlinked_add(znode_t *zp, dmu_tx_t *tx) 272{ 273 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 274 275 ASSERT(zp->z_unlinked); 276 ASSERT(zp->z_links == 0); 277 278 VERIFY3U(0, ==, 279 zap_add_int(zfsvfs->z_os, zfsvfs->z_unlinkedobj, zp->z_id, tx)); 280 281 dataset_kstats_update_nunlinks_kstat(&zfsvfs->z_kstat, 1); 282} 283 284/* 285 * Clean up any znodes that had no links when we either crashed or 286 * (force) umounted the file system. 287 */ 288void 289zfs_unlinked_drain(zfsvfs_t *zfsvfs) 290{ 291 zap_cursor_t zc; 292 zap_attribute_t zap; 293 dmu_object_info_t doi; 294 znode_t *zp; 295 dmu_tx_t *tx; 296 int error; 297 298 /* 299 * Iterate over the contents of the unlinked set. 300 */ 301 for (zap_cursor_init(&zc, zfsvfs->z_os, zfsvfs->z_unlinkedobj); 302 zap_cursor_retrieve(&zc, &zap) == 0; 303 zap_cursor_advance(&zc)) { 304 305 /* 306 * See what kind of object we have in list 307 */ 308 309 error = dmu_object_info(zfsvfs->z_os, 310 zap.za_first_integer, &doi); 311 if (error != 0) 312 continue; 313 314 ASSERT((doi.doi_type == DMU_OT_PLAIN_FILE_CONTENTS) || 315 (doi.doi_type == DMU_OT_DIRECTORY_CONTENTS)); 316 /* 317 * We need to re-mark these list entries for deletion, 318 * so we pull them back into core and set zp->z_unlinked. 319 */ 320 error = zfs_zget(zfsvfs, zap.za_first_integer, &zp); 321 322 /* 323 * We may pick up znodes that are already marked for deletion. 324 * This could happen during the purge of an extended attribute 325 * directory. All we need to do is skip over them, since they 326 * are already in the system marked z_unlinked. 327 */ 328 if (error != 0) 329 continue; 330 331 vn_lock(ZTOV(zp), LK_EXCLUSIVE | LK_RETRY); 332 333 /* 334 * Due to changes in zfs_rmnode we need to make sure the 335 * link count is set to zero here. 336 */ 337 if (zp->z_links != 0) { 338 tx = dmu_tx_create(zfsvfs->z_os); 339 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_FALSE); 340 error = dmu_tx_assign(tx, TXG_WAIT); 341 if (error != 0) { 342 dmu_tx_abort(tx); 343 vput(ZTOV(zp)); 344 continue; 345 } 346 zp->z_links = 0; 347 VERIFY0(sa_update(zp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs), 348 &zp->z_links, sizeof (zp->z_links), tx)); 349 dmu_tx_commit(tx); 350 } 351 352 zp->z_unlinked = B_TRUE; 353 vput(ZTOV(zp)); 354 } 355 zap_cursor_fini(&zc); 356} 357 358/* 359 * Delete the entire contents of a directory. Return a count 360 * of the number of entries that could not be deleted. If we encounter 361 * an error, return a count of at least one so that the directory stays 362 * in the unlinked set. 363 * 364 * NOTE: this function assumes that the directory is inactive, 365 * so there is no need to lock its entries before deletion. 366 * Also, it assumes the directory contents is *only* regular 367 * files. 368 */ 369static int 370zfs_purgedir(znode_t *dzp) 371{ 372 zap_cursor_t zc; 373 zap_attribute_t zap; 374 znode_t *xzp; 375 dmu_tx_t *tx; 376 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 377 int skipped = 0; 378 int error; 379 380 for (zap_cursor_init(&zc, zfsvfs->z_os, dzp->z_id); 381 (error = zap_cursor_retrieve(&zc, &zap)) == 0; 382 zap_cursor_advance(&zc)) { 383 error = zfs_zget(zfsvfs, 384 ZFS_DIRENT_OBJ(zap.za_first_integer), &xzp); 385 if (error) { 386 skipped += 1; 387 continue; 388 } 389 390 vn_lock(ZTOV(xzp), LK_EXCLUSIVE | LK_RETRY); 391 ASSERT((ZTOV(xzp)->v_type == VREG) || 392 (ZTOV(xzp)->v_type == VLNK)); 393 394 tx = dmu_tx_create(zfsvfs->z_os); 395 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 396 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, zap.za_name); 397 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE); 398 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 399 /* Is this really needed ? */ 400 zfs_sa_upgrade_txholds(tx, xzp); 401 dmu_tx_mark_netfree(tx); 402 error = dmu_tx_assign(tx, TXG_WAIT); 403 if (error) { 404 dmu_tx_abort(tx); 405 vput(ZTOV(xzp)); 406 skipped += 1; 407 continue; 408 } 409 410 error = zfs_link_destroy(dzp, zap.za_name, xzp, tx, 0, NULL); 411 if (error) 412 skipped += 1; 413 dmu_tx_commit(tx); 414 415 vput(ZTOV(xzp)); 416 } 417 zap_cursor_fini(&zc); 418 if (error != ENOENT) 419 skipped += 1; 420 return (skipped); 421} 422 423extern taskq_t *zfsvfs_taskq; 424 425void 426zfs_rmnode(znode_t *zp) 427{ 428 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 429 objset_t *os = zfsvfs->z_os; 430 dmu_tx_t *tx; 431 uint64_t acl_obj; 432 uint64_t xattr_obj; 433 uint64_t count; 434 int error; 435 436 ASSERT(zp->z_links == 0); 437 if (zfsvfs->z_replay == B_FALSE) 438 ASSERT_VOP_ELOCKED(ZTOV(zp), __func__); 439 440 /* 441 * If this is an attribute directory, purge its contents. 442 */ 443 if (ZTOV(zp) != NULL && ZTOV(zp)->v_type == VDIR && 444 (zp->z_pflags & ZFS_XATTR)) { 445 if (zfs_purgedir(zp) != 0) { 446 /* 447 * Not enough space to delete some xattrs. 448 * Leave it in the unlinked set. 449 */ 450 zfs_znode_dmu_fini(zp); 451 zfs_znode_free(zp); 452 return; 453 } 454 } else { 455 /* 456 * Free up all the data in the file. We don't do this for 457 * XATTR directories because we need truncate and remove to be 458 * in the same tx, like in zfs_znode_delete(). Otherwise, if 459 * we crash here we'll end up with an inconsistent truncated 460 * zap object in the delete queue. Note a truncated file is 461 * harmless since it only contains user data. 462 */ 463 error = dmu_free_long_range(os, zp->z_id, 0, DMU_OBJECT_END); 464 if (error) { 465 /* 466 * Not enough space or we were interrupted by unmount. 467 * Leave the file in the unlinked set. 468 */ 469 zfs_znode_dmu_fini(zp); 470 zfs_znode_free(zp); 471 return; 472 } 473 } 474 475 /* 476 * If the file has extended attributes, we're going to unlink 477 * the xattr dir. 478 */ 479 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 480 &xattr_obj, sizeof (xattr_obj)); 481 if (error) 482 xattr_obj = 0; 483 484 acl_obj = zfs_external_acl(zp); 485 486 /* 487 * Set up the final transaction. 488 */ 489 tx = dmu_tx_create(os); 490 dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END); 491 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 492 if (xattr_obj) 493 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, TRUE, NULL); 494 if (acl_obj) 495 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 496 497 zfs_sa_upgrade_txholds(tx, zp); 498 error = dmu_tx_assign(tx, TXG_WAIT); 499 if (error) { 500 /* 501 * Not enough space to delete the file. Leave it in the 502 * unlinked set, leaking it until the fs is remounted (at 503 * which point we'll call zfs_unlinked_drain() to process it). 504 */ 505 dmu_tx_abort(tx); 506 zfs_znode_dmu_fini(zp); 507 zfs_znode_free(zp); 508 return; 509 } 510 511 /* 512 * FreeBSD's implementation of zfs_zget requires a vnode to back it. 513 * This means that we could end up calling into getnewvnode while 514 * calling zfs_rmnode as a result of a prior call to getnewvnode 515 * trying to clear vnodes out of the cache. If this repeats we can 516 * recurse enough that we overflow our stack. To avoid this, we 517 * avoid calling zfs_zget on the xattr znode and instead simply add 518 * it to the unlinked set and schedule a call to zfs_unlinked_drain. 519 */ 520 if (xattr_obj) { 521 /* Add extended attribute directory to the unlinked set. */ 522 VERIFY3U(0, ==, 523 zap_add_int(os, zfsvfs->z_unlinkedobj, xattr_obj, tx)); 524 } 525 526 mutex_enter(&os->os_dsl_dataset->ds_dir->dd_activity_lock); 527 528 /* Remove this znode from the unlinked set */ 529 VERIFY3U(0, ==, 530 zap_remove_int(os, zfsvfs->z_unlinkedobj, zp->z_id, tx)); 531 532 if (zap_count(os, zfsvfs->z_unlinkedobj, &count) == 0 && count == 0) { 533 cv_broadcast(&os->os_dsl_dataset->ds_dir->dd_activity_cv); 534 } 535 536 mutex_exit(&os->os_dsl_dataset->ds_dir->dd_activity_lock); 537 538 dataset_kstats_update_nunlinked_kstat(&zfsvfs->z_kstat, 1); 539 540 zfs_znode_delete(zp, tx); 541 542 dmu_tx_commit(tx); 543 544 if (xattr_obj) { 545 /* 546 * We're using the FreeBSD taskqueue API here instead of 547 * the Solaris taskq API since the FreeBSD API allows for a 548 * task to be enqueued multiple times but executed once. 549 */ 550 taskqueue_enqueue(zfsvfs_taskq->tq_queue, 551 &zfsvfs->z_unlinked_drain_task); 552 } 553} 554 555static uint64_t 556zfs_dirent(znode_t *zp, uint64_t mode) 557{ 558 uint64_t de = zp->z_id; 559 560 if (zp->z_zfsvfs->z_version >= ZPL_VERSION_DIRENT_TYPE) 561 de |= IFTODT(mode) << 60; 562 return (de); 563} 564 565/* 566 * Link zp into dzp. Can only fail if zp has been unlinked. 567 */ 568int 569zfs_link_create(znode_t *dzp, const char *name, znode_t *zp, dmu_tx_t *tx, 570 int flag) 571{ 572 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 573 vnode_t *vp = ZTOV(zp); 574 uint64_t value; 575 int zp_is_dir = (vp->v_type == VDIR); 576 sa_bulk_attr_t bulk[5]; 577 uint64_t mtime[2], ctime[2]; 578 int count = 0; 579 int error; 580 581 if (zfsvfs->z_replay == B_FALSE) { 582 ASSERT_VOP_ELOCKED(ZTOV(dzp), __func__); 583 ASSERT_VOP_ELOCKED(ZTOV(zp), __func__); 584 } 585 if (zp_is_dir) { 586 if (dzp->z_links >= ZFS_LINK_MAX) 587 return (SET_ERROR(EMLINK)); 588 } 589 if (!(flag & ZRENAMING)) { 590 if (zp->z_unlinked) { /* no new links to unlinked zp */ 591 ASSERT(!(flag & (ZNEW | ZEXISTS))); 592 return (SET_ERROR(ENOENT)); 593 } 594 if (zp->z_links >= ZFS_LINK_MAX - zp_is_dir) { 595 return (SET_ERROR(EMLINK)); 596 } 597 zp->z_links++; 598 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 599 &zp->z_links, sizeof (zp->z_links)); 600 601 } else { 602 ASSERT(zp->z_unlinked == 0); 603 } 604 value = zfs_dirent(zp, zp->z_mode); 605 error = zap_add(zp->z_zfsvfs->z_os, dzp->z_id, name, 606 8, 1, &value, tx); 607 608 /* 609 * zap_add could fail to add the entry if it exceeds the capacity of the 610 * leaf-block and zap_leaf_split() failed to help. 611 * The caller of this routine is responsible for failing the transaction 612 * which will rollback the SA updates done above. 613 */ 614 if (error != 0) { 615 if (!(flag & ZRENAMING) && !(flag & ZNEW)) 616 zp->z_links--; 617 return (error); 618 } 619 620 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, 621 &dzp->z_id, sizeof (dzp->z_id)); 622 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 623 &zp->z_pflags, sizeof (zp->z_pflags)); 624 625 if (!(flag & ZNEW)) { 626 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 627 ctime, sizeof (ctime)); 628 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, 629 ctime); 630 } 631 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 632 ASSERT0(error); 633 634 dzp->z_size++; 635 dzp->z_links += zp_is_dir; 636 count = 0; 637 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 638 &dzp->z_size, sizeof (dzp->z_size)); 639 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 640 &dzp->z_links, sizeof (dzp->z_links)); 641 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 642 mtime, sizeof (mtime)); 643 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 644 ctime, sizeof (ctime)); 645 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 646 &dzp->z_pflags, sizeof (dzp->z_pflags)); 647 zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime); 648 error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx); 649 ASSERT0(error); 650 return (0); 651} 652 653/* 654 * The match type in the code for this function should conform to: 655 * 656 * ------------------------------------------------------------------------ 657 * fs type | z_norm | lookup type | match type 658 * ---------|-------------|-------------|---------------------------------- 659 * CS !norm | 0 | 0 | 0 (exact) 660 * CS norm | formX | 0 | MT_NORMALIZE 661 * CI !norm | upper | !ZCIEXACT | MT_NORMALIZE 662 * CI !norm | upper | ZCIEXACT | MT_NORMALIZE | MT_MATCH_CASE 663 * CI norm | upper|formX | !ZCIEXACT | MT_NORMALIZE 664 * CI norm | upper|formX | ZCIEXACT | MT_NORMALIZE | MT_MATCH_CASE 665 * CM !norm | upper | !ZCILOOK | MT_NORMALIZE | MT_MATCH_CASE 666 * CM !norm | upper | ZCILOOK | MT_NORMALIZE 667 * CM norm | upper|formX | !ZCILOOK | MT_NORMALIZE | MT_MATCH_CASE 668 * CM norm | upper|formX | ZCILOOK | MT_NORMALIZE 669 * 670 * Abbreviations: 671 * CS = Case Sensitive, CI = Case Insensitive, CM = Case Mixed 672 * upper = case folding set by fs type on creation (U8_TEXTPREP_TOUPPER) 673 * formX = unicode normalization form set on fs creation 674 */ 675static int 676zfs_dropname(znode_t *dzp, const char *name, znode_t *zp, dmu_tx_t *tx, 677 int flag) 678{ 679 int error; 680 681 if (zp->z_zfsvfs->z_norm) { 682 matchtype_t mt = MT_NORMALIZE; 683 684 if (zp->z_zfsvfs->z_case == ZFS_CASE_MIXED) { 685 mt |= MT_MATCH_CASE; 686 } 687 688 error = zap_remove_norm(zp->z_zfsvfs->z_os, dzp->z_id, 689 name, mt, tx); 690 } else { 691 error = zap_remove(zp->z_zfsvfs->z_os, dzp->z_id, name, tx); 692 } 693 694 return (error); 695} 696 697/* 698 * Unlink zp from dzp, and mark zp for deletion if this was the last link. 699 * Can fail if zp is a mount point (EBUSY) or a non-empty directory (EEXIST). 700 * If 'unlinkedp' is NULL, we put unlinked znodes on the unlinked list. 701 * If it's non-NULL, we use it to indicate whether the znode needs deletion, 702 * and it's the caller's job to do it. 703 */ 704int 705zfs_link_destroy(znode_t *dzp, const char *name, znode_t *zp, dmu_tx_t *tx, 706 int flag, boolean_t *unlinkedp) 707{ 708 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 709 vnode_t *vp = ZTOV(zp); 710 int zp_is_dir = (vp->v_type == VDIR); 711 boolean_t unlinked = B_FALSE; 712 sa_bulk_attr_t bulk[5]; 713 uint64_t mtime[2], ctime[2]; 714 int count = 0; 715 int error; 716 717 if (zfsvfs->z_replay == B_FALSE) { 718 ASSERT_VOP_ELOCKED(ZTOV(dzp), __func__); 719 ASSERT_VOP_ELOCKED(ZTOV(zp), __func__); 720 } 721 if (!(flag & ZRENAMING)) { 722 723 if (zp_is_dir && !zfs_dirempty(zp)) 724 return (SET_ERROR(ENOTEMPTY)); 725 726 /* 727 * If we get here, we are going to try to remove the object. 728 * First try removing the name from the directory; if that 729 * fails, return the error. 730 */ 731 error = zfs_dropname(dzp, name, zp, tx, flag); 732 if (error != 0) { 733 return (error); 734 } 735 736 if (zp->z_links <= zp_is_dir) { 737 zfs_panic_recover("zfs: link count on vnode %p is %u, " 738 "should be at least %u", zp->z_vnode, 739 (int)zp->z_links, 740 zp_is_dir + 1); 741 zp->z_links = zp_is_dir + 1; 742 } 743 if (--zp->z_links == zp_is_dir) { 744 zp->z_unlinked = B_TRUE; 745 zp->z_links = 0; 746 unlinked = B_TRUE; 747 } else { 748 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), 749 NULL, &ctime, sizeof (ctime)); 750 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 751 NULL, &zp->z_pflags, sizeof (zp->z_pflags)); 752 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, 753 ctime); 754 } 755 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), 756 NULL, &zp->z_links, sizeof (zp->z_links)); 757 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 758 count = 0; 759 ASSERT0(error); 760 } else { 761 ASSERT(zp->z_unlinked == 0); 762 error = zfs_dropname(dzp, name, zp, tx, flag); 763 if (error != 0) 764 return (error); 765 } 766 767 dzp->z_size--; /* one dirent removed */ 768 dzp->z_links -= zp_is_dir; /* ".." link from zp */ 769 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), 770 NULL, &dzp->z_links, sizeof (dzp->z_links)); 771 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), 772 NULL, &dzp->z_size, sizeof (dzp->z_size)); 773 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), 774 NULL, ctime, sizeof (ctime)); 775 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), 776 NULL, mtime, sizeof (mtime)); 777 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 778 NULL, &dzp->z_pflags, sizeof (dzp->z_pflags)); 779 zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime); 780 error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx); 781 ASSERT0(error); 782 783 if (unlinkedp != NULL) 784 *unlinkedp = unlinked; 785 else if (unlinked) 786 zfs_unlinked_add(zp, tx); 787 788 return (0); 789} 790 791/* 792 * Indicate whether the directory is empty. 793 */ 794boolean_t 795zfs_dirempty(znode_t *dzp) 796{ 797 return (dzp->z_size == 2); 798} 799 800int 801zfs_make_xattrdir(znode_t *zp, vattr_t *vap, znode_t **xvpp, cred_t *cr) 802{ 803 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 804 znode_t *xzp; 805 dmu_tx_t *tx; 806 int error; 807 zfs_acl_ids_t acl_ids; 808 boolean_t fuid_dirtied; 809 uint64_t parent __unused; 810 811 *xvpp = NULL; 812 813 if ((error = zfs_acl_ids_create(zp, IS_XATTR, vap, cr, NULL, 814 &acl_ids)) != 0) 815 return (error); 816 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, 0)) { 817 zfs_acl_ids_free(&acl_ids); 818 return (SET_ERROR(EDQUOT)); 819 } 820 821 getnewvnode_reserve_(); 822 823 tx = dmu_tx_create(zfsvfs->z_os); 824 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 825 ZFS_SA_BASE_ATTR_SIZE); 826 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 827 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 828 fuid_dirtied = zfsvfs->z_fuid_dirty; 829 if (fuid_dirtied) 830 zfs_fuid_txhold(zfsvfs, tx); 831 error = dmu_tx_assign(tx, TXG_WAIT); 832 if (error) { 833 zfs_acl_ids_free(&acl_ids); 834 dmu_tx_abort(tx); 835 getnewvnode_drop_reserve(); 836 return (error); 837 } 838 zfs_mknode(zp, vap, tx, cr, IS_XATTR, &xzp, &acl_ids); 839 840 if (fuid_dirtied) 841 zfs_fuid_sync(zfsvfs, tx); 842 843#ifdef ZFS_DEBUG 844 error = sa_lookup(xzp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 845 &parent, sizeof (parent)); 846 ASSERT(error == 0 && parent == zp->z_id); 847#endif 848 849 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &xzp->z_id, 850 sizeof (xzp->z_id), tx)); 851 852 zfs_log_create(zfsvfs->z_log, tx, TX_MKXATTR, zp, xzp, "", NULL, 853 acl_ids.z_fuidp, vap); 854 855 zfs_acl_ids_free(&acl_ids); 856 dmu_tx_commit(tx); 857 858 getnewvnode_drop_reserve(); 859 860 *xvpp = xzp; 861 862 return (0); 863} 864 865/* 866 * Return a znode for the extended attribute directory for zp. 867 * ** If the directory does not already exist, it is created ** 868 * 869 * IN: zp - znode to obtain attribute directory from 870 * cr - credentials of caller 871 * flags - flags from the VOP_LOOKUP call 872 * 873 * OUT: xzpp - pointer to extended attribute znode 874 * 875 * RETURN: 0 on success 876 * error number on failure 877 */ 878int 879zfs_get_xattrdir(znode_t *zp, znode_t **xzpp, cred_t *cr, int flags) 880{ 881 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 882 znode_t *xzp; 883 vattr_t va; 884 int error; 885top: 886 error = zfs_dirent_lookup(zp, "", &xzp, ZXATTR); 887 if (error) 888 return (error); 889 890 if (xzp != NULL) { 891 *xzpp = xzp; 892 return (0); 893 } 894 895 896 if (!(flags & CREATE_XATTR_DIR)) 897 return (SET_ERROR(ENOATTR)); 898 899 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 900 return (SET_ERROR(EROFS)); 901 } 902 903 /* 904 * The ability to 'create' files in an attribute 905 * directory comes from the write_xattr permission on the base file. 906 * 907 * The ability to 'search' an attribute directory requires 908 * read_xattr permission on the base file. 909 * 910 * Once in a directory the ability to read/write attributes 911 * is controlled by the permissions on the attribute file. 912 */ 913 va.va_mask = AT_MODE | AT_UID | AT_GID; 914 va.va_type = VDIR; 915 va.va_mode = S_IFDIR | S_ISVTX | 0777; 916 zfs_fuid_map_ids(zp, cr, &va.va_uid, &va.va_gid); 917 918 error = zfs_make_xattrdir(zp, &va, xzpp, cr); 919 920 if (error == ERESTART) { 921 /* NB: we already did dmu_tx_wait() if necessary */ 922 goto top; 923 } 924 if (error == 0) 925 VOP_UNLOCK1(ZTOV(*xzpp)); 926 927 return (error); 928} 929 930/* 931 * Decide whether it is okay to remove within a sticky directory. 932 * 933 * In sticky directories, write access is not sufficient; 934 * you can remove entries from a directory only if: 935 * 936 * you own the directory, 937 * you own the entry, 938 * the entry is a plain file and you have write access, 939 * or you are privileged (checked in secpolicy...). 940 * 941 * The function returns 0 if remove access is granted. 942 */ 943int 944zfs_sticky_remove_access(znode_t *zdp, znode_t *zp, cred_t *cr) 945{ 946 uid_t uid; 947 uid_t downer; 948 uid_t fowner; 949 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 950 951 if (zdp->z_zfsvfs->z_replay) 952 return (0); 953 954 if ((zdp->z_mode & S_ISVTX) == 0) 955 return (0); 956 957 downer = zfs_fuid_map_id(zfsvfs, zdp->z_uid, cr, ZFS_OWNER); 958 fowner = zfs_fuid_map_id(zfsvfs, zp->z_uid, cr, ZFS_OWNER); 959 960 if ((uid = crgetuid(cr)) == downer || uid == fowner || 961 (ZTOV(zp)->v_type == VREG && 962 zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr) == 0)) 963 return (0); 964 else 965 return (secpolicy_vnode_remove(ZTOV(zp), cr)); 966} 967