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/sysmacros.h> 32#include <sys/vfs.h> 33#include <sys/vnode.h> 34#include <sys/file.h> 35#include <sys/kmem.h> 36#include <sys/uio.h> 37#include <sys/pathname.h> 38#include <sys/cmn_err.h> 39#include <sys/errno.h> 40#include <sys/stat.h> 41#include <sys/sunddi.h> 42#include <sys/random.h> 43#include <sys/policy.h> 44#include <sys/zfs_dir.h> 45#include <sys/zfs_acl.h> 46#include <sys/zfs_vnops.h> 47#include <sys/fs/zfs.h> 48#include <sys/zap.h> 49#include <sys/dmu.h> 50#include <sys/atomic.h> 51#include <sys/zfs_ctldir.h> 52#include <sys/zfs_fuid.h> 53#include <sys/sa.h> 54#include <sys/zfs_sa.h> 55#include <sys/dmu_objset.h> 56#include <sys/dsl_dir.h> 57 58/* 59 * zfs_match_find() is used by zfs_dirent_lock() to perform zap lookups 60 * of names after deciding which is the appropriate lookup interface. 61 */ 62static int 63zfs_match_find(zfsvfs_t *zfsvfs, znode_t *dzp, const char *name, 64 matchtype_t mt, boolean_t update, int *deflags, pathname_t *rpnp, 65 uint64_t *zoid) 66{ 67 boolean_t conflict = B_FALSE; 68 int error; 69 70 if (zfsvfs->z_norm) { 71 size_t bufsz = 0; 72 char *buf = NULL; 73 74 if (rpnp) { 75 buf = rpnp->pn_buf; 76 bufsz = rpnp->pn_bufsize; 77 } 78 79 /* 80 * In the non-mixed case we only expect there would ever 81 * be one match, but we need to use the normalizing lookup. 82 */ 83 error = zap_lookup_norm(zfsvfs->z_os, dzp->z_id, name, 8, 1, 84 zoid, mt, buf, bufsz, &conflict); 85 } else { 86 error = zap_lookup(zfsvfs->z_os, dzp->z_id, name, 8, 1, zoid); 87 } 88 89 /* 90 * Allow multiple entries provided the first entry is 91 * the object id. Non-zpl consumers may safely make 92 * use of the additional space. 93 * 94 * XXX: This should be a feature flag for compatibility 95 */ 96 if (error == EOVERFLOW) 97 error = 0; 98 99 if (zfsvfs->z_norm && !error && deflags) 100 *deflags = conflict ? ED_CASE_CONFLICT : 0; 101 102 *zoid = ZFS_DIRENT_OBJ(*zoid); 103 104 return (error); 105} 106 107/* 108 * Lock a directory entry. A dirlock on <dzp, name> protects that name 109 * in dzp's directory zap object. As long as you hold a dirlock, you can 110 * assume two things: (1) dzp cannot be reaped, and (2) no other thread 111 * can change the zap entry for (i.e. link or unlink) this name. 112 * 113 * Input arguments: 114 * dzp - znode for directory 115 * name - name of entry to lock 116 * flag - ZNEW: if the entry already exists, fail with EEXIST. 117 * ZEXISTS: if the entry does not exist, fail with ENOENT. 118 * ZSHARED: allow concurrent access with other ZSHARED callers. 119 * ZXATTR: we want dzp's xattr directory 120 * ZCILOOK: On a mixed sensitivity file system, 121 * this lookup should be case-insensitive. 122 * ZCIEXACT: On a purely case-insensitive file system, 123 * this lookup should be case-sensitive. 124 * ZRENAMING: we are locking for renaming, force narrow locks 125 * ZHAVELOCK: Don't grab the z_name_lock for this call. The 126 * current thread already holds it. 127 * 128 * Output arguments: 129 * zpp - pointer to the znode for the entry (NULL if there isn't one) 130 * dlpp - pointer to the dirlock for this entry (NULL on error) 131 * direntflags - (case-insensitive lookup only) 132 * flags if multiple case-sensitive matches exist in directory 133 * realpnp - (case-insensitive lookup only) 134 * actual name matched within the directory 135 * 136 * Return value: 0 on success or errno on failure. 137 * 138 * NOTE: Always checks for, and rejects, '.' and '..'. 139 * NOTE: For case-insensitive file systems we take wide locks (see below), 140 * but return znode pointers to a single match. 141 */ 142int 143zfs_dirent_lock(zfs_dirlock_t **dlpp, znode_t *dzp, char *name, 144 znode_t **zpp, int flag, int *direntflags, pathname_t *realpnp) 145{ 146 zfsvfs_t *zfsvfs = ZTOZSB(dzp); 147 zfs_dirlock_t *dl; 148 boolean_t update; 149 matchtype_t mt = 0; 150 uint64_t zoid; 151 int error = 0; 152 int cmpflags; 153 154 *zpp = NULL; 155 *dlpp = NULL; 156 157 /* 158 * Verify that we are not trying to lock '.', '..', or '.zfs' 159 */ 160 if ((name[0] == '.' && 161 (name[1] == '\0' || (name[1] == '.' && name[2] == '\0'))) || 162 (zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0)) 163 return (SET_ERROR(EEXIST)); 164 165 /* 166 * Case sensitivity and normalization preferences are set when 167 * the file system is created. These are stored in the 168 * zfsvfs->z_case and zfsvfs->z_norm fields. These choices 169 * affect what vnodes can be cached in the DNLC, how we 170 * perform zap lookups, and the "width" of our dirlocks. 171 * 172 * A normal dirlock locks a single name. Note that with 173 * normalization a name can be composed multiple ways, but 174 * when normalized, these names all compare equal. A wide 175 * dirlock locks multiple names. We need these when the file 176 * system is supporting mixed-mode access. It is sometimes 177 * necessary to lock all case permutations of file name at 178 * once so that simultaneous case-insensitive/case-sensitive 179 * behaves as rationally as possible. 180 */ 181 182 /* 183 * When matching we may need to normalize & change case according to 184 * FS settings. 185 * 186 * Note that a normalized match is necessary for a case insensitive 187 * filesystem when the lookup request is not exact because normalization 188 * can fold case independent of normalizing code point sequences. 189 * 190 * See the table above zfs_dropname(). 191 */ 192 if (zfsvfs->z_norm != 0) { 193 mt = MT_NORMALIZE; 194 195 /* 196 * Determine if the match needs to honor the case specified in 197 * lookup, and if so keep track of that so that during 198 * normalization we don't fold case. 199 */ 200 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE && 201 (flag & ZCIEXACT)) || 202 (zfsvfs->z_case == ZFS_CASE_MIXED && !(flag & ZCILOOK))) { 203 mt |= MT_MATCH_CASE; 204 } 205 } 206 207 /* 208 * Only look in or update the DNLC if we are looking for the 209 * name on a file system that does not require normalization 210 * or case folding. We can also look there if we happen to be 211 * on a non-normalizing, mixed sensitivity file system IF we 212 * are looking for the exact name. 213 * 214 * Maybe can add TO-UPPERed version of name to dnlc in ci-only 215 * case for performance improvement? 216 */ 217 update = !zfsvfs->z_norm || 218 (zfsvfs->z_case == ZFS_CASE_MIXED && 219 !(zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER) && !(flag & ZCILOOK)); 220 221 /* 222 * ZRENAMING indicates we are in a situation where we should 223 * take narrow locks regardless of the file system's 224 * preferences for normalizing and case folding. This will 225 * prevent us deadlocking trying to grab the same wide lock 226 * twice if the two names happen to be case-insensitive 227 * matches. 228 */ 229 if (flag & ZRENAMING) 230 cmpflags = 0; 231 else 232 cmpflags = zfsvfs->z_norm; 233 234 /* 235 * Wait until there are no locks on this name. 236 * 237 * Don't grab the lock if it is already held. However, cannot 238 * have both ZSHARED and ZHAVELOCK together. 239 */ 240 ASSERT(!(flag & ZSHARED) || !(flag & ZHAVELOCK)); 241 if (!(flag & ZHAVELOCK)) 242 rw_enter(&dzp->z_name_lock, RW_READER); 243 244 mutex_enter(&dzp->z_lock); 245 for (;;) { 246 if (dzp->z_unlinked && !(flag & ZXATTR)) { 247 mutex_exit(&dzp->z_lock); 248 if (!(flag & ZHAVELOCK)) 249 rw_exit(&dzp->z_name_lock); 250 return (SET_ERROR(ENOENT)); 251 } 252 for (dl = dzp->z_dirlocks; dl != NULL; dl = dl->dl_next) { 253 if ((u8_strcmp(name, dl->dl_name, 0, cmpflags, 254 U8_UNICODE_LATEST, &error) == 0) || error != 0) 255 break; 256 } 257 if (error != 0) { 258 mutex_exit(&dzp->z_lock); 259 if (!(flag & ZHAVELOCK)) 260 rw_exit(&dzp->z_name_lock); 261 return (SET_ERROR(ENOENT)); 262 } 263 if (dl == NULL) { 264 /* 265 * Allocate a new dirlock and add it to the list. 266 */ 267 dl = kmem_alloc(sizeof (zfs_dirlock_t), KM_SLEEP); 268 cv_init(&dl->dl_cv, NULL, CV_DEFAULT, NULL); 269 dl->dl_name = name; 270 dl->dl_sharecnt = 0; 271 dl->dl_namelock = 0; 272 dl->dl_namesize = 0; 273 dl->dl_dzp = dzp; 274 dl->dl_next = dzp->z_dirlocks; 275 dzp->z_dirlocks = dl; 276 break; 277 } 278 if ((flag & ZSHARED) && dl->dl_sharecnt != 0) 279 break; 280 cv_wait(&dl->dl_cv, &dzp->z_lock); 281 } 282 283 /* 284 * If the z_name_lock was NOT held for this dirlock record it. 285 */ 286 if (flag & ZHAVELOCK) 287 dl->dl_namelock = 1; 288 289 if ((flag & ZSHARED) && ++dl->dl_sharecnt > 1 && dl->dl_namesize == 0) { 290 /* 291 * We're the second shared reference to dl. Make a copy of 292 * dl_name in case the first thread goes away before we do. 293 * Note that we initialize the new name before storing its 294 * pointer into dl_name, because the first thread may load 295 * dl->dl_name at any time. It'll either see the old value, 296 * which belongs to it, or the new shared copy; either is OK. 297 */ 298 dl->dl_namesize = strlen(dl->dl_name) + 1; 299 name = kmem_alloc(dl->dl_namesize, KM_SLEEP); 300 bcopy(dl->dl_name, name, dl->dl_namesize); 301 dl->dl_name = name; 302 } 303 304 mutex_exit(&dzp->z_lock); 305 306 /* 307 * We have a dirlock on the name. (Note that it is the dirlock, 308 * not the dzp's z_lock, that protects the name in the zap object.) 309 * See if there's an object by this name; if so, put a hold on it. 310 */ 311 if (flag & ZXATTR) { 312 error = sa_lookup(dzp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &zoid, 313 sizeof (zoid)); 314 if (error == 0) 315 error = (zoid == 0 ? SET_ERROR(ENOENT) : 0); 316 } else { 317 error = zfs_match_find(zfsvfs, dzp, name, mt, 318 update, direntflags, realpnp, &zoid); 319 } 320 if (error) { 321 if (error != ENOENT || (flag & ZEXISTS)) { 322 zfs_dirent_unlock(dl); 323 return (error); 324 } 325 } else { 326 if (flag & ZNEW) { 327 zfs_dirent_unlock(dl); 328 return (SET_ERROR(EEXIST)); 329 } 330 error = zfs_zget(zfsvfs, zoid, zpp); 331 if (error) { 332 zfs_dirent_unlock(dl); 333 return (error); 334 } 335 } 336 337 *dlpp = dl; 338 339 return (0); 340} 341 342/* 343 * Unlock this directory entry and wake anyone who was waiting for it. 344 */ 345void 346zfs_dirent_unlock(zfs_dirlock_t *dl) 347{ 348 znode_t *dzp = dl->dl_dzp; 349 zfs_dirlock_t **prev_dl, *cur_dl; 350 351 mutex_enter(&dzp->z_lock); 352 353 if (!dl->dl_namelock) 354 rw_exit(&dzp->z_name_lock); 355 356 if (dl->dl_sharecnt > 1) { 357 dl->dl_sharecnt--; 358 mutex_exit(&dzp->z_lock); 359 return; 360 } 361 prev_dl = &dzp->z_dirlocks; 362 while ((cur_dl = *prev_dl) != dl) 363 prev_dl = &cur_dl->dl_next; 364 *prev_dl = dl->dl_next; 365 cv_broadcast(&dl->dl_cv); 366 mutex_exit(&dzp->z_lock); 367 368 if (dl->dl_namesize != 0) 369 kmem_free(dl->dl_name, dl->dl_namesize); 370 cv_destroy(&dl->dl_cv); 371 kmem_free(dl, sizeof (*dl)); 372} 373 374/* 375 * Look up an entry in a directory. 376 * 377 * NOTE: '.' and '..' are handled as special cases because 378 * no directory entries are actually stored for them. If this is 379 * the root of a filesystem, then '.zfs' is also treated as a 380 * special pseudo-directory. 381 */ 382int 383zfs_dirlook(znode_t *dzp, char *name, znode_t **zpp, int flags, 384 int *deflg, pathname_t *rpnp) 385{ 386 zfs_dirlock_t *dl; 387 znode_t *zp; 388 struct inode *ip; 389 int error = 0; 390 uint64_t parent; 391 392 if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) { 393 *zpp = dzp; 394 zhold(*zpp); 395 } else if (name[0] == '.' && name[1] == '.' && name[2] == 0) { 396 zfsvfs_t *zfsvfs = ZTOZSB(dzp); 397 398 /* 399 * If we are a snapshot mounted under .zfs, return 400 * the inode pointer for the snapshot directory. 401 */ 402 if ((error = sa_lookup(dzp->z_sa_hdl, 403 SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0) 404 return (error); 405 406 if (parent == dzp->z_id && zfsvfs->z_parent != zfsvfs) { 407 error = zfsctl_root_lookup(zfsvfs->z_parent->z_ctldir, 408 "snapshot", &ip, 0, kcred, NULL, NULL); 409 *zpp = ITOZ(ip); 410 return (error); 411 } 412 rw_enter(&dzp->z_parent_lock, RW_READER); 413 error = zfs_zget(zfsvfs, parent, &zp); 414 if (error == 0) 415 *zpp = zp; 416 rw_exit(&dzp->z_parent_lock); 417 } else if (zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0) { 418 ip = zfsctl_root(dzp); 419 *zpp = ITOZ(ip); 420 } else { 421 int zf; 422 423 zf = ZEXISTS | ZSHARED; 424 if (flags & FIGNORECASE) 425 zf |= ZCILOOK; 426 427 error = zfs_dirent_lock(&dl, dzp, name, &zp, zf, deflg, rpnp); 428 if (error == 0) { 429 *zpp = zp; 430 zfs_dirent_unlock(dl); 431 dzp->z_zn_prefetch = B_TRUE; /* enable prefetching */ 432 } 433 rpnp = NULL; 434 } 435 436 if ((flags & FIGNORECASE) && rpnp && !error) 437 (void) strlcpy(rpnp->pn_buf, name, rpnp->pn_bufsize); 438 439 return (error); 440} 441 442/* 443 * unlinked Set (formerly known as the "delete queue") Error Handling 444 * 445 * When dealing with the unlinked set, we dmu_tx_hold_zap(), but we 446 * don't specify the name of the entry that we will be manipulating. We 447 * also fib and say that we won't be adding any new entries to the 448 * unlinked set, even though we might (this is to lower the minimum file 449 * size that can be deleted in a full filesystem). So on the small 450 * chance that the nlink list is using a fat zap (ie. has more than 451 * 2000 entries), we *may* not pre-read a block that's needed. 452 * Therefore it is remotely possible for some of the assertions 453 * regarding the unlinked set below to fail due to i/o error. On a 454 * nondebug system, this will result in the space being leaked. 455 */ 456void 457zfs_unlinked_add(znode_t *zp, dmu_tx_t *tx) 458{ 459 zfsvfs_t *zfsvfs = ZTOZSB(zp); 460 461 ASSERT(zp->z_unlinked); 462 ASSERT(ZTOI(zp)->i_nlink == 0); 463 464 VERIFY3U(0, ==, 465 zap_add_int(zfsvfs->z_os, zfsvfs->z_unlinkedobj, zp->z_id, tx)); 466 467 dataset_kstats_update_nunlinks_kstat(&zfsvfs->z_kstat, 1); 468} 469 470/* 471 * Clean up any znodes that had no links when we either crashed or 472 * (force) umounted the file system. 473 */ 474static void 475zfs_unlinked_drain_task(void *arg) 476{ 477 zfsvfs_t *zfsvfs = arg; 478 zap_cursor_t zc; 479 zap_attribute_t zap; 480 dmu_object_info_t doi; 481 znode_t *zp; 482 int error; 483 484 ASSERT3B(zfsvfs->z_draining, ==, B_TRUE); 485 486 /* 487 * Iterate over the contents of the unlinked set. 488 */ 489 for (zap_cursor_init(&zc, zfsvfs->z_os, zfsvfs->z_unlinkedobj); 490 zap_cursor_retrieve(&zc, &zap) == 0 && !zfsvfs->z_drain_cancel; 491 zap_cursor_advance(&zc)) { 492 493 /* 494 * See what kind of object we have in list 495 */ 496 497 error = dmu_object_info(zfsvfs->z_os, 498 zap.za_first_integer, &doi); 499 if (error != 0) 500 continue; 501 502 ASSERT((doi.doi_type == DMU_OT_PLAIN_FILE_CONTENTS) || 503 (doi.doi_type == DMU_OT_DIRECTORY_CONTENTS)); 504 /* 505 * We need to re-mark these list entries for deletion, 506 * so we pull them back into core and set zp->z_unlinked. 507 */ 508 error = zfs_zget(zfsvfs, zap.za_first_integer, &zp); 509 510 /* 511 * We may pick up znodes that are already marked for deletion. 512 * This could happen during the purge of an extended attribute 513 * directory. All we need to do is skip over them, since they 514 * are already in the system marked z_unlinked. 515 */ 516 if (error != 0) 517 continue; 518 519 zp->z_unlinked = B_TRUE; 520 521 /* 522 * zrele() decrements the znode's ref count and may cause 523 * it to be synchronously freed. We interrupt freeing 524 * of this znode by checking the return value of 525 * dmu_objset_zfs_unmounting() in dmu_free_long_range() 526 * when an unmount is requested. 527 */ 528 zrele(zp); 529 ASSERT3B(zfsvfs->z_unmounted, ==, B_FALSE); 530 } 531 zap_cursor_fini(&zc); 532 533 zfsvfs->z_draining = B_FALSE; 534 zfsvfs->z_drain_task = TASKQID_INVALID; 535} 536 537/* 538 * Sets z_draining then tries to dispatch async unlinked drain. 539 * If that fails executes synchronous unlinked drain. 540 */ 541void 542zfs_unlinked_drain(zfsvfs_t *zfsvfs) 543{ 544 ASSERT3B(zfsvfs->z_unmounted, ==, B_FALSE); 545 ASSERT3B(zfsvfs->z_draining, ==, B_FALSE); 546 547 zfsvfs->z_draining = B_TRUE; 548 zfsvfs->z_drain_cancel = B_FALSE; 549 550 zfsvfs->z_drain_task = taskq_dispatch( 551 dsl_pool_unlinked_drain_taskq(dmu_objset_pool(zfsvfs->z_os)), 552 zfs_unlinked_drain_task, zfsvfs, TQ_SLEEP); 553 if (zfsvfs->z_drain_task == TASKQID_INVALID) { 554 zfs_dbgmsg("async zfs_unlinked_drain dispatch failed"); 555 zfs_unlinked_drain_task(zfsvfs); 556 } 557} 558 559/* 560 * Wait for the unlinked drain taskq task to stop. This will interrupt the 561 * unlinked set processing if it is in progress. 562 */ 563void 564zfs_unlinked_drain_stop_wait(zfsvfs_t *zfsvfs) 565{ 566 ASSERT3B(zfsvfs->z_unmounted, ==, B_FALSE); 567 568 if (zfsvfs->z_draining) { 569 zfsvfs->z_drain_cancel = B_TRUE; 570 taskq_cancel_id(dsl_pool_unlinked_drain_taskq( 571 dmu_objset_pool(zfsvfs->z_os)), zfsvfs->z_drain_task); 572 zfsvfs->z_drain_task = TASKQID_INVALID; 573 zfsvfs->z_draining = B_FALSE; 574 } 575} 576 577/* 578 * Delete the entire contents of a directory. Return a count 579 * of the number of entries that could not be deleted. If we encounter 580 * an error, return a count of at least one so that the directory stays 581 * in the unlinked set. 582 * 583 * NOTE: this function assumes that the directory is inactive, 584 * so there is no need to lock its entries before deletion. 585 * Also, it assumes the directory contents is *only* regular 586 * files. 587 */ 588static int 589zfs_purgedir(znode_t *dzp) 590{ 591 zap_cursor_t zc; 592 zap_attribute_t zap; 593 znode_t *xzp; 594 dmu_tx_t *tx; 595 zfsvfs_t *zfsvfs = ZTOZSB(dzp); 596 zfs_dirlock_t dl; 597 int skipped = 0; 598 int error; 599 600 for (zap_cursor_init(&zc, zfsvfs->z_os, dzp->z_id); 601 (error = zap_cursor_retrieve(&zc, &zap)) == 0; 602 zap_cursor_advance(&zc)) { 603 error = zfs_zget(zfsvfs, 604 ZFS_DIRENT_OBJ(zap.za_first_integer), &xzp); 605 if (error) { 606 skipped += 1; 607 continue; 608 } 609 610 ASSERT(S_ISREG(ZTOI(xzp)->i_mode) || 611 S_ISLNK(ZTOI(xzp)->i_mode)); 612 613 tx = dmu_tx_create(zfsvfs->z_os); 614 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE); 615 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, zap.za_name); 616 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE); 617 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 618 /* Is this really needed ? */ 619 zfs_sa_upgrade_txholds(tx, xzp); 620 dmu_tx_mark_netfree(tx); 621 error = dmu_tx_assign(tx, TXG_WAIT); 622 if (error) { 623 dmu_tx_abort(tx); 624 zfs_zrele_async(xzp); 625 skipped += 1; 626 continue; 627 } 628 bzero(&dl, sizeof (dl)); 629 dl.dl_dzp = dzp; 630 dl.dl_name = zap.za_name; 631 632 error = zfs_link_destroy(&dl, xzp, tx, 0, NULL); 633 if (error) 634 skipped += 1; 635 dmu_tx_commit(tx); 636 637 zfs_zrele_async(xzp); 638 } 639 zap_cursor_fini(&zc); 640 if (error != ENOENT) 641 skipped += 1; 642 return (skipped); 643} 644 645void 646zfs_rmnode(znode_t *zp) 647{ 648 zfsvfs_t *zfsvfs = ZTOZSB(zp); 649 objset_t *os = zfsvfs->z_os; 650 znode_t *xzp = NULL; 651 dmu_tx_t *tx; 652 uint64_t acl_obj; 653 uint64_t xattr_obj; 654 uint64_t links; 655 int error; 656 657 ASSERT(ZTOI(zp)->i_nlink == 0); 658 ASSERT(atomic_read(&ZTOI(zp)->i_count) == 0); 659 660 /* 661 * If this is an attribute directory, purge its contents. 662 */ 663 if (S_ISDIR(ZTOI(zp)->i_mode) && (zp->z_pflags & ZFS_XATTR)) { 664 if (zfs_purgedir(zp) != 0) { 665 /* 666 * Not enough space to delete some xattrs. 667 * Leave it in the unlinked set. 668 */ 669 zfs_znode_dmu_fini(zp); 670 671 return; 672 } 673 } 674 675 /* 676 * Free up all the data in the file. We don't do this for directories 677 * because we need truncate and remove to be in the same tx, like in 678 * zfs_znode_delete(). Otherwise, if we crash here we'll end up with 679 * an inconsistent truncated zap object in the delete queue. Note a 680 * truncated file is harmless since it only contains user data. 681 */ 682 if (S_ISREG(ZTOI(zp)->i_mode)) { 683 error = dmu_free_long_range(os, zp->z_id, 0, DMU_OBJECT_END); 684 if (error) { 685 /* 686 * Not enough space or we were interrupted by unmount. 687 * Leave the file in the unlinked set. 688 */ 689 zfs_znode_dmu_fini(zp); 690 return; 691 } 692 } 693 694 /* 695 * If the file has extended attributes, we're going to unlink 696 * the xattr dir. 697 */ 698 error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), 699 &xattr_obj, sizeof (xattr_obj)); 700 if (error == 0 && xattr_obj) { 701 error = zfs_zget(zfsvfs, xattr_obj, &xzp); 702 ASSERT(error == 0); 703 } 704 705 acl_obj = zfs_external_acl(zp); 706 707 /* 708 * Set up the final transaction. 709 */ 710 tx = dmu_tx_create(os); 711 dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END); 712 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 713 if (xzp) { 714 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, TRUE, NULL); 715 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE); 716 } 717 if (acl_obj) 718 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 719 720 zfs_sa_upgrade_txholds(tx, zp); 721 error = dmu_tx_assign(tx, TXG_WAIT); 722 if (error) { 723 /* 724 * Not enough space to delete the file. Leave it in the 725 * unlinked set, leaking it until the fs is remounted (at 726 * which point we'll call zfs_unlinked_drain() to process it). 727 */ 728 dmu_tx_abort(tx); 729 zfs_znode_dmu_fini(zp); 730 goto out; 731 } 732 733 if (xzp) { 734 ASSERT(error == 0); 735 mutex_enter(&xzp->z_lock); 736 xzp->z_unlinked = B_TRUE; /* mark xzp for deletion */ 737 clear_nlink(ZTOI(xzp)); /* no more links to it */ 738 links = 0; 739 VERIFY(0 == sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs), 740 &links, sizeof (links), tx)); 741 mutex_exit(&xzp->z_lock); 742 zfs_unlinked_add(xzp, tx); 743 } 744 745 mutex_enter(&os->os_dsl_dataset->ds_dir->dd_activity_lock); 746 747 /* 748 * Remove this znode from the unlinked set. If a has rollback has 749 * occurred while a file is open and unlinked. Then when the file 750 * is closed post rollback it will not exist in the rolled back 751 * version of the unlinked object. 752 */ 753 error = zap_remove_int(zfsvfs->z_os, zfsvfs->z_unlinkedobj, 754 zp->z_id, tx); 755 VERIFY(error == 0 || error == ENOENT); 756 757 uint64_t count; 758 if (zap_count(os, zfsvfs->z_unlinkedobj, &count) == 0 && count == 0) { 759 cv_broadcast(&os->os_dsl_dataset->ds_dir->dd_activity_cv); 760 } 761 762 mutex_exit(&os->os_dsl_dataset->ds_dir->dd_activity_lock); 763 764 dataset_kstats_update_nunlinked_kstat(&zfsvfs->z_kstat, 1); 765 766 zfs_znode_delete(zp, tx); 767 768 dmu_tx_commit(tx); 769out: 770 if (xzp) 771 zfs_zrele_async(xzp); 772} 773 774static uint64_t 775zfs_dirent(znode_t *zp, uint64_t mode) 776{ 777 uint64_t de = zp->z_id; 778 779 if (ZTOZSB(zp)->z_version >= ZPL_VERSION_DIRENT_TYPE) 780 de |= IFTODT(mode) << 60; 781 return (de); 782} 783 784/* 785 * Link zp into dl. Can fail in the following cases : 786 * - if zp has been unlinked. 787 * - if the number of entries with the same hash (aka. colliding entries) 788 * exceed the capacity of a leaf-block of fatzap and splitting of the 789 * leaf-block does not help. 790 */ 791int 792zfs_link_create(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag) 793{ 794 znode_t *dzp = dl->dl_dzp; 795 zfsvfs_t *zfsvfs = ZTOZSB(zp); 796 uint64_t value; 797 int zp_is_dir = S_ISDIR(ZTOI(zp)->i_mode); 798 sa_bulk_attr_t bulk[5]; 799 uint64_t mtime[2], ctime[2]; 800 uint64_t links; 801 int count = 0; 802 int error; 803 804 mutex_enter(&zp->z_lock); 805 806 if (!(flag & ZRENAMING)) { 807 if (zp->z_unlinked) { /* no new links to unlinked zp */ 808 ASSERT(!(flag & (ZNEW | ZEXISTS))); 809 mutex_exit(&zp->z_lock); 810 return (SET_ERROR(ENOENT)); 811 } 812 if (!(flag & ZNEW)) { 813 /* 814 * ZNEW nodes come from zfs_mknode() where the link 815 * count has already been initialised 816 */ 817 inc_nlink(ZTOI(zp)); 818 links = ZTOI(zp)->i_nlink; 819 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), 820 NULL, &links, sizeof (links)); 821 } 822 } 823 824 value = zfs_dirent(zp, zp->z_mode); 825 error = zap_add(ZTOZSB(zp)->z_os, dzp->z_id, dl->dl_name, 8, 1, 826 &value, tx); 827 828 /* 829 * zap_add could fail to add the entry if it exceeds the capacity of the 830 * leaf-block and zap_leaf_split() failed to help. 831 * The caller of this routine is responsible for failing the transaction 832 * which will rollback the SA updates done above. 833 */ 834 if (error != 0) { 835 if (!(flag & ZRENAMING) && !(flag & ZNEW)) 836 drop_nlink(ZTOI(zp)); 837 mutex_exit(&zp->z_lock); 838 return (error); 839 } 840 841 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL, 842 &dzp->z_id, sizeof (dzp->z_id)); 843 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 844 &zp->z_pflags, sizeof (zp->z_pflags)); 845 846 if (!(flag & ZNEW)) { 847 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 848 ctime, sizeof (ctime)); 849 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, 850 ctime); 851 } 852 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 853 ASSERT(error == 0); 854 855 mutex_exit(&zp->z_lock); 856 857 mutex_enter(&dzp->z_lock); 858 dzp->z_size++; 859 if (zp_is_dir) 860 inc_nlink(ZTOI(dzp)); 861 links = ZTOI(dzp)->i_nlink; 862 count = 0; 863 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL, 864 &dzp->z_size, sizeof (dzp->z_size)); 865 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL, 866 &links, sizeof (links)); 867 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL, 868 mtime, sizeof (mtime)); 869 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL, 870 ctime, sizeof (ctime)); 871 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL, 872 &dzp->z_pflags, sizeof (dzp->z_pflags)); 873 zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime); 874 error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx); 875 ASSERT(error == 0); 876 mutex_exit(&dzp->z_lock); 877 878 return (0); 879} 880 881/* 882 * The match type in the code for this function should conform to: 883 * 884 * ------------------------------------------------------------------------ 885 * fs type | z_norm | lookup type | match type 886 * ---------|-------------|-------------|---------------------------------- 887 * CS !norm | 0 | 0 | 0 (exact) 888 * CS norm | formX | 0 | MT_NORMALIZE 889 * CI !norm | upper | !ZCIEXACT | MT_NORMALIZE 890 * CI !norm | upper | ZCIEXACT | MT_NORMALIZE | MT_MATCH_CASE 891 * CI norm | upper|formX | !ZCIEXACT | MT_NORMALIZE 892 * CI norm | upper|formX | ZCIEXACT | MT_NORMALIZE | MT_MATCH_CASE 893 * CM !norm | upper | !ZCILOOK | MT_NORMALIZE | MT_MATCH_CASE 894 * CM !norm | upper | ZCILOOK | MT_NORMALIZE 895 * CM norm | upper|formX | !ZCILOOK | MT_NORMALIZE | MT_MATCH_CASE 896 * CM norm | upper|formX | ZCILOOK | MT_NORMALIZE 897 * 898 * Abbreviations: 899 * CS = Case Sensitive, CI = Case Insensitive, CM = Case Mixed 900 * upper = case folding set by fs type on creation (U8_TEXTPREP_TOUPPER) 901 * formX = unicode normalization form set on fs creation 902 */ 903static int 904zfs_dropname(zfs_dirlock_t *dl, znode_t *zp, znode_t *dzp, dmu_tx_t *tx, 905 int flag) 906{ 907 int error; 908 909 if (ZTOZSB(zp)->z_norm) { 910 matchtype_t mt = MT_NORMALIZE; 911 912 if ((ZTOZSB(zp)->z_case == ZFS_CASE_INSENSITIVE && 913 (flag & ZCIEXACT)) || 914 (ZTOZSB(zp)->z_case == ZFS_CASE_MIXED && 915 !(flag & ZCILOOK))) { 916 mt |= MT_MATCH_CASE; 917 } 918 919 error = zap_remove_norm(ZTOZSB(zp)->z_os, dzp->z_id, 920 dl->dl_name, mt, tx); 921 } else { 922 error = zap_remove(ZTOZSB(zp)->z_os, dzp->z_id, dl->dl_name, 923 tx); 924 } 925 926 return (error); 927} 928 929/* 930 * Unlink zp from dl, and mark zp for deletion if this was the last link. Can 931 * fail if zp is a mount point (EBUSY) or a non-empty directory (ENOTEMPTY). 932 * If 'unlinkedp' is NULL, we put unlinked znodes on the unlinked list. 933 * If it's non-NULL, we use it to indicate whether the znode needs deletion, 934 * and it's the caller's job to do it. 935 */ 936int 937zfs_link_destroy(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag, 938 boolean_t *unlinkedp) 939{ 940 znode_t *dzp = dl->dl_dzp; 941 zfsvfs_t *zfsvfs = ZTOZSB(dzp); 942 int zp_is_dir = S_ISDIR(ZTOI(zp)->i_mode); 943 boolean_t unlinked = B_FALSE; 944 sa_bulk_attr_t bulk[5]; 945 uint64_t mtime[2], ctime[2]; 946 uint64_t links; 947 int count = 0; 948 int error; 949 950 if (!(flag & ZRENAMING)) { 951 mutex_enter(&zp->z_lock); 952 953 if (zp_is_dir && !zfs_dirempty(zp)) { 954 mutex_exit(&zp->z_lock); 955 return (SET_ERROR(ENOTEMPTY)); 956 } 957 958 /* 959 * If we get here, we are going to try to remove the object. 960 * First try removing the name from the directory; if that 961 * fails, return the error. 962 */ 963 error = zfs_dropname(dl, zp, dzp, tx, flag); 964 if (error != 0) { 965 mutex_exit(&zp->z_lock); 966 return (error); 967 } 968 969 if (ZTOI(zp)->i_nlink <= zp_is_dir) { 970 zfs_panic_recover("zfs: link count on %lu is %u, " 971 "should be at least %u", zp->z_id, 972 (int)ZTOI(zp)->i_nlink, zp_is_dir + 1); 973 set_nlink(ZTOI(zp), zp_is_dir + 1); 974 } 975 drop_nlink(ZTOI(zp)); 976 if (ZTOI(zp)->i_nlink == zp_is_dir) { 977 zp->z_unlinked = B_TRUE; 978 clear_nlink(ZTOI(zp)); 979 unlinked = B_TRUE; 980 } else { 981 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), 982 NULL, &ctime, sizeof (ctime)); 983 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 984 NULL, &zp->z_pflags, sizeof (zp->z_pflags)); 985 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, 986 ctime); 987 } 988 links = ZTOI(zp)->i_nlink; 989 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), 990 NULL, &links, sizeof (links)); 991 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx); 992 count = 0; 993 ASSERT(error == 0); 994 mutex_exit(&zp->z_lock); 995 } else { 996 error = zfs_dropname(dl, zp, dzp, tx, flag); 997 if (error != 0) 998 return (error); 999 } 1000 1001 mutex_enter(&dzp->z_lock); 1002 dzp->z_size--; /* one dirent removed */ 1003 if (zp_is_dir) 1004 drop_nlink(ZTOI(dzp)); /* ".." link from zp */ 1005 links = ZTOI(dzp)->i_nlink; 1006 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), 1007 NULL, &links, sizeof (links)); 1008 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), 1009 NULL, &dzp->z_size, sizeof (dzp->z_size)); 1010 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), 1011 NULL, ctime, sizeof (ctime)); 1012 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), 1013 NULL, mtime, sizeof (mtime)); 1014 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), 1015 NULL, &dzp->z_pflags, sizeof (dzp->z_pflags)); 1016 zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime); 1017 error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx); 1018 ASSERT(error == 0); 1019 mutex_exit(&dzp->z_lock); 1020 1021 if (unlinkedp != NULL) 1022 *unlinkedp = unlinked; 1023 else if (unlinked) 1024 zfs_unlinked_add(zp, tx); 1025 1026 return (0); 1027} 1028 1029/* 1030 * Indicate whether the directory is empty. Works with or without z_lock 1031 * held, but can only be consider a hint in the latter case. Returns true 1032 * if only "." and ".." remain and there's no work in progress. 1033 * 1034 * The internal ZAP size, rather than zp->z_size, needs to be checked since 1035 * some consumers (Lustre) do not strictly maintain an accurate SA_ZPL_SIZE. 1036 */ 1037boolean_t 1038zfs_dirempty(znode_t *dzp) 1039{ 1040 zfsvfs_t *zfsvfs = ZTOZSB(dzp); 1041 uint64_t count; 1042 int error; 1043 1044 if (dzp->z_dirlocks != NULL) 1045 return (B_FALSE); 1046 1047 error = zap_count(zfsvfs->z_os, dzp->z_id, &count); 1048 if (error != 0 || count != 0) 1049 return (B_FALSE); 1050 1051 return (B_TRUE); 1052} 1053 1054int 1055zfs_make_xattrdir(znode_t *zp, vattr_t *vap, znode_t **xzpp, cred_t *cr) 1056{ 1057 zfsvfs_t *zfsvfs = ZTOZSB(zp); 1058 znode_t *xzp; 1059 dmu_tx_t *tx; 1060 int error; 1061 zfs_acl_ids_t acl_ids; 1062 boolean_t fuid_dirtied; 1063#ifdef ZFS_DEBUG 1064 uint64_t parent; 1065#endif 1066 1067 *xzpp = NULL; 1068 1069 if ((error = zfs_zaccess(zp, ACE_WRITE_NAMED_ATTRS, 0, B_FALSE, cr))) 1070 return (error); 1071 1072 if ((error = zfs_acl_ids_create(zp, IS_XATTR, vap, cr, NULL, 1073 &acl_ids)) != 0) 1074 return (error); 1075 if (zfs_acl_ids_overquota(zfsvfs, &acl_ids, zp->z_projid)) { 1076 zfs_acl_ids_free(&acl_ids); 1077 return (SET_ERROR(EDQUOT)); 1078 } 1079 1080 tx = dmu_tx_create(zfsvfs->z_os); 1081 dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes + 1082 ZFS_SA_BASE_ATTR_SIZE); 1083 dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE); 1084 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 1085 fuid_dirtied = zfsvfs->z_fuid_dirty; 1086 if (fuid_dirtied) 1087 zfs_fuid_txhold(zfsvfs, tx); 1088 error = dmu_tx_assign(tx, TXG_WAIT); 1089 if (error) { 1090 zfs_acl_ids_free(&acl_ids); 1091 dmu_tx_abort(tx); 1092 return (error); 1093 } 1094 zfs_mknode(zp, vap, tx, cr, IS_XATTR, &xzp, &acl_ids); 1095 1096 if (fuid_dirtied) 1097 zfs_fuid_sync(zfsvfs, tx); 1098 1099#ifdef ZFS_DEBUG 1100 error = sa_lookup(xzp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs), 1101 &parent, sizeof (parent)); 1102 ASSERT(error == 0 && parent == zp->z_id); 1103#endif 1104 1105 VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &xzp->z_id, 1106 sizeof (xzp->z_id), tx)); 1107 1108 if (!zp->z_unlinked) 1109 zfs_log_create(zfsvfs->z_log, tx, TX_MKXATTR, zp, xzp, "", NULL, 1110 acl_ids.z_fuidp, vap); 1111 1112 zfs_acl_ids_free(&acl_ids); 1113 dmu_tx_commit(tx); 1114 1115 *xzpp = xzp; 1116 1117 return (0); 1118} 1119 1120/* 1121 * Return a znode for the extended attribute directory for zp. 1122 * ** If the directory does not already exist, it is created ** 1123 * 1124 * IN: zp - znode to obtain attribute directory from 1125 * cr - credentials of caller 1126 * flags - flags from the VOP_LOOKUP call 1127 * 1128 * OUT: xipp - pointer to extended attribute znode 1129 * 1130 * RETURN: 0 on success 1131 * error number on failure 1132 */ 1133int 1134zfs_get_xattrdir(znode_t *zp, znode_t **xzpp, cred_t *cr, int flags) 1135{ 1136 zfsvfs_t *zfsvfs = ZTOZSB(zp); 1137 znode_t *xzp; 1138 zfs_dirlock_t *dl; 1139 vattr_t va; 1140 int error; 1141top: 1142 error = zfs_dirent_lock(&dl, zp, "", &xzp, ZXATTR, NULL, NULL); 1143 if (error) 1144 return (error); 1145 1146 if (xzp != NULL) { 1147 *xzpp = xzp; 1148 zfs_dirent_unlock(dl); 1149 return (0); 1150 } 1151 1152 if (!(flags & CREATE_XATTR_DIR)) { 1153 zfs_dirent_unlock(dl); 1154 return (SET_ERROR(ENOENT)); 1155 } 1156 1157 if (zfs_is_readonly(zfsvfs)) { 1158 zfs_dirent_unlock(dl); 1159 return (SET_ERROR(EROFS)); 1160 } 1161 1162 /* 1163 * The ability to 'create' files in an attribute 1164 * directory comes from the write_xattr permission on the base file. 1165 * 1166 * The ability to 'search' an attribute directory requires 1167 * read_xattr permission on the base file. 1168 * 1169 * Once in a directory the ability to read/write attributes 1170 * is controlled by the permissions on the attribute file. 1171 */ 1172 va.va_mask = ATTR_MODE | ATTR_UID | ATTR_GID; 1173 va.va_mode = S_IFDIR | S_ISVTX | 0777; 1174 zfs_fuid_map_ids(zp, cr, &va.va_uid, &va.va_gid); 1175 1176 va.va_dentry = NULL; 1177 error = zfs_make_xattrdir(zp, &va, xzpp, cr); 1178 zfs_dirent_unlock(dl); 1179 1180 if (error == ERESTART) { 1181 /* NB: we already did dmu_tx_wait() if necessary */ 1182 goto top; 1183 } 1184 1185 return (error); 1186} 1187 1188/* 1189 * Decide whether it is okay to remove within a sticky directory. 1190 * 1191 * In sticky directories, write access is not sufficient; 1192 * you can remove entries from a directory only if: 1193 * 1194 * you own the directory, 1195 * you own the entry, 1196 * you have write access to the entry, 1197 * or you are privileged (checked in secpolicy...). 1198 * 1199 * The function returns 0 if remove access is granted. 1200 */ 1201int 1202zfs_sticky_remove_access(znode_t *zdp, znode_t *zp, cred_t *cr) 1203{ 1204 uid_t uid; 1205 uid_t downer; 1206 uid_t fowner; 1207 zfsvfs_t *zfsvfs = ZTOZSB(zdp); 1208 1209 if (zfsvfs->z_replay) 1210 return (0); 1211 1212 if ((zdp->z_mode & S_ISVTX) == 0) 1213 return (0); 1214 1215 downer = zfs_fuid_map_id(zfsvfs, KUID_TO_SUID(ZTOI(zdp)->i_uid), 1216 cr, ZFS_OWNER); 1217 fowner = zfs_fuid_map_id(zfsvfs, KUID_TO_SUID(ZTOI(zp)->i_uid), 1218 cr, ZFS_OWNER); 1219 1220 if ((uid = crgetuid(cr)) == downer || uid == fowner || 1221 zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr) == 0) 1222 return (0); 1223 else 1224 return (secpolicy_vnode_remove(cr)); 1225} 1226