zfs_dir.c revision 208131
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 * Copyright 2008 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 26#include <sys/types.h> 27#include <sys/param.h> 28#include <sys/time.h> 29#include <sys/systm.h> 30#include <sys/sysmacros.h> 31#include <sys/resource.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/cmn_err.h> 38#include <sys/errno.h> 39#include <sys/stat.h> 40#include <sys/unistd.h> 41#include <sys/sunddi.h> 42#include <sys/random.h> 43#include <sys/policy.h> 44#include <sys/kcondvar.h> 45#include <sys/callb.h> 46#include <sys/smp.h> 47#include <sys/zfs_dir.h> 48#include <sys/zfs_acl.h> 49#include <sys/fs/zfs.h> 50#include <sys/zap.h> 51#include <sys/dmu.h> 52#include <sys/atomic.h> 53#include <sys/zfs_ctldir.h> 54#include <sys/zfs_fuid.h> 55#include <sys/dnlc.h> 56#include <sys/extdirent.h> 57 58/* 59 * zfs_match_find() is used by zfs_dirent_lock() to peform 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, char *name, boolean_t exact, 64 boolean_t update, int *deflags, pathname_t *rpnp, uint64_t *zoid) 65{ 66 int error; 67 68 if (zfsvfs->z_norm) { 69 matchtype_t mt = MT_FIRST; 70 boolean_t conflict = B_FALSE; 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 if (exact) 79 mt = MT_EXACT; 80 /* 81 * In the non-mixed case we only expect there would ever 82 * be one match, but we need to use the normalizing lookup. 83 */ 84 error = zap_lookup_norm(zfsvfs->z_os, dzp->z_id, name, 8, 1, 85 zoid, mt, buf, bufsz, &conflict); 86 if (!error && deflags) 87 *deflags = conflict ? ED_CASE_CONFLICT : 0; 88 } else { 89 error = zap_lookup(zfsvfs->z_os, dzp->z_id, name, 8, 1, zoid); 90 } 91 *zoid = ZFS_DIRENT_OBJ(*zoid); 92 93 if (error == ENOENT && update) 94 dnlc_update(ZTOV(dzp), name, DNLC_NO_VNODE); 95 96 return (error); 97} 98 99/* 100 * Lock a directory entry. A dirlock on <dzp, name> protects that name 101 * in dzp's directory zap object. As long as you hold a dirlock, you can 102 * assume two things: (1) dzp cannot be reaped, and (2) no other thread 103 * can change the zap entry for (i.e. link or unlink) this name. 104 * 105 * Input arguments: 106 * dzp - znode for directory 107 * name - name of entry to lock 108 * flag - ZNEW: if the entry already exists, fail with EEXIST. 109 * ZEXISTS: if the entry does not exist, fail with ENOENT. 110 * ZSHARED: allow concurrent access with other ZSHARED callers. 111 * ZXATTR: we want dzp's xattr directory 112 * ZCILOOK: On a mixed sensitivity file system, 113 * this lookup should be case-insensitive. 114 * ZCIEXACT: On a purely case-insensitive file system, 115 * this lookup should be case-sensitive. 116 * ZRENAMING: we are locking for renaming, force narrow locks 117 * ZHAVELOCK: Don't grab the z_name_lock for this call. The 118 * current thread already holds it. 119 * 120 * Output arguments: 121 * zpp - pointer to the znode for the entry (NULL if there isn't one) 122 * dlpp - pointer to the dirlock for this entry (NULL on error) 123 * direntflags - (case-insensitive lookup only) 124 * flags if multiple case-sensitive matches exist in directory 125 * realpnp - (case-insensitive lookup only) 126 * actual name matched within the directory 127 * 128 * Return value: 0 on success or errno on failure. 129 * 130 * NOTE: Always checks for, and rejects, '.' and '..'. 131 * NOTE: For case-insensitive file systems we take wide locks (see below), 132 * but return znode pointers to a single match. 133 */ 134int 135zfs_dirent_lock(zfs_dirlock_t **dlpp, znode_t *dzp, char *name, znode_t **zpp, 136 int flag, int *direntflags, pathname_t *realpnp) 137{ 138 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 139 zfs_dirlock_t *dl; 140 boolean_t update; 141 boolean_t exact; 142 uint64_t zoid; 143 vnode_t *vp = NULL; 144 int error = 0; 145 int cmpflags; 146 147 *zpp = NULL; 148 *dlpp = NULL; 149 150 /* 151 * Verify that we are not trying to lock '.', '..', or '.zfs' 152 */ 153 if (name[0] == '.' && 154 (name[1] == '\0' || (name[1] == '.' && name[2] == '\0')) || 155 zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0) 156 return (EEXIST); 157 158 /* 159 * Case sensitivity and normalization preferences are set when 160 * the file system is created. These are stored in the 161 * zfsvfs->z_case and zfsvfs->z_norm fields. These choices 162 * affect what vnodes can be cached in the DNLC, how we 163 * perform zap lookups, and the "width" of our dirlocks. 164 * 165 * A normal dirlock locks a single name. Note that with 166 * normalization a name can be composed multiple ways, but 167 * when normalized, these names all compare equal. A wide 168 * dirlock locks multiple names. We need these when the file 169 * system is supporting mixed-mode access. It is sometimes 170 * necessary to lock all case permutations of file name at 171 * once so that simultaneous case-insensitive/case-sensitive 172 * behaves as rationally as possible. 173 */ 174 175 /* 176 * Decide if exact matches should be requested when performing 177 * a zap lookup on file systems supporting case-insensitive 178 * access. 179 */ 180 exact = 181 ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE) && (flag & ZCIEXACT)) || 182 ((zfsvfs->z_case == ZFS_CASE_MIXED) && !(flag & ZCILOOK)); 183 184 /* 185 * Only look in or update the DNLC if we are looking for the 186 * name on a file system that does not require normalization 187 * or case folding. We can also look there if we happen to be 188 * on a non-normalizing, mixed sensitivity file system IF we 189 * are looking for the exact name. 190 * 191 * Maybe can add TO-UPPERed version of name to dnlc in ci-only 192 * case for performance improvement? 193 */ 194 update = !zfsvfs->z_norm || 195 ((zfsvfs->z_case == ZFS_CASE_MIXED) && 196 !(zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER) && !(flag & ZCILOOK)); 197 198 /* 199 * ZRENAMING indicates we are in a situation where we should 200 * take narrow locks regardless of the file system's 201 * preferences for normalizing and case folding. This will 202 * prevent us deadlocking trying to grab the same wide lock 203 * twice if the two names happen to be case-insensitive 204 * matches. 205 */ 206 if (flag & ZRENAMING) 207 cmpflags = 0; 208 else 209 cmpflags = zfsvfs->z_norm; 210 211 /* 212 * Wait until there are no locks on this name. 213 * 214 * Don't grab the the lock if it is already held. However, cannot 215 * have both ZSHARED and ZHAVELOCK together. 216 */ 217 ASSERT(!(flag & ZSHARED) || !(flag & ZHAVELOCK)); 218 if (!(flag & ZHAVELOCK)) 219 rw_enter(&dzp->z_name_lock, RW_READER); 220 221 mutex_enter(&dzp->z_lock); 222 for (;;) { 223 if (dzp->z_unlinked) { 224 mutex_exit(&dzp->z_lock); 225 if (!(flag & ZHAVELOCK)) 226 rw_exit(&dzp->z_name_lock); 227 return (ENOENT); 228 } 229 for (dl = dzp->z_dirlocks; dl != NULL; dl = dl->dl_next) { 230 if ((u8_strcmp(name, dl->dl_name, 0, cmpflags, 231 U8_UNICODE_LATEST, &error) == 0) || error != 0) 232 break; 233 } 234 if (error != 0) { 235 mutex_exit(&dzp->z_lock); 236 if (!(flag & ZHAVELOCK)) 237 rw_exit(&dzp->z_name_lock); 238 return (ENOENT); 239 } 240 if (dl == NULL) { 241 /* 242 * Allocate a new dirlock and add it to the list. 243 */ 244 dl = kmem_alloc(sizeof (zfs_dirlock_t), KM_SLEEP); 245 cv_init(&dl->dl_cv, NULL, CV_DEFAULT, NULL); 246 dl->dl_name = name; 247 dl->dl_sharecnt = 0; 248 dl->dl_namelock = 0; 249 dl->dl_namesize = 0; 250 dl->dl_dzp = dzp; 251 dl->dl_next = dzp->z_dirlocks; 252 dzp->z_dirlocks = dl; 253 break; 254 } 255 if ((flag & ZSHARED) && dl->dl_sharecnt != 0) 256 break; 257 cv_wait(&dl->dl_cv, &dzp->z_lock); 258 } 259 260 /* 261 * If the z_name_lock was NOT held for this dirlock record it. 262 */ 263 if (flag & ZHAVELOCK) 264 dl->dl_namelock = 1; 265 266 if ((flag & ZSHARED) && ++dl->dl_sharecnt > 1 && dl->dl_namesize == 0) { 267 /* 268 * We're the second shared reference to dl. Make a copy of 269 * dl_name in case the first thread goes away before we do. 270 * Note that we initialize the new name before storing its 271 * pointer into dl_name, because the first thread may load 272 * dl->dl_name at any time. He'll either see the old value, 273 * which is his, or the new shared copy; either is OK. 274 */ 275 dl->dl_namesize = strlen(dl->dl_name) + 1; 276 name = kmem_alloc(dl->dl_namesize, KM_SLEEP); 277 bcopy(dl->dl_name, name, dl->dl_namesize); 278 dl->dl_name = name; 279 } 280 281 mutex_exit(&dzp->z_lock); 282 283 /* 284 * We have a dirlock on the name. (Note that it is the dirlock, 285 * not the dzp's z_lock, that protects the name in the zap object.) 286 * See if there's an object by this name; if so, put a hold on it. 287 */ 288 if (flag & ZXATTR) { 289 zoid = dzp->z_phys->zp_xattr; 290 error = (zoid == 0 ? ENOENT : 0); 291 } else { 292 if (update) 293 vp = dnlc_lookup(ZTOV(dzp), name); 294 if (vp == DNLC_NO_VNODE) { 295 VN_RELE(vp); 296 error = ENOENT; 297 } else if (vp) { 298 if (flag & ZNEW) { 299 zfs_dirent_unlock(dl); 300 VN_RELE(vp); 301 return (EEXIST); 302 } 303 *dlpp = dl; 304 *zpp = VTOZ(vp); 305 return (0); 306 } else { 307 error = zfs_match_find(zfsvfs, dzp, name, exact, 308 update, direntflags, realpnp, &zoid); 309 } 310 } 311 if (error) { 312 if (error != ENOENT || (flag & ZEXISTS)) { 313 zfs_dirent_unlock(dl); 314 return (error); 315 } 316 } else { 317 if (flag & ZNEW) { 318 zfs_dirent_unlock(dl); 319 return (EEXIST); 320 } 321 error = zfs_zget(zfsvfs, zoid, zpp); 322 if (error) { 323 zfs_dirent_unlock(dl); 324 return (error); 325 } 326 if (!(flag & ZXATTR) && update) 327 dnlc_update(ZTOV(dzp), name, ZTOV(*zpp)); 328 } 329 330 *dlpp = dl; 331 332 return (0); 333} 334 335/* 336 * Unlock this directory entry and wake anyone who was waiting for it. 337 */ 338void 339zfs_dirent_unlock(zfs_dirlock_t *dl) 340{ 341 znode_t *dzp = dl->dl_dzp; 342 zfs_dirlock_t **prev_dl, *cur_dl; 343 344 mutex_enter(&dzp->z_lock); 345 346 if (!dl->dl_namelock) 347 rw_exit(&dzp->z_name_lock); 348 349 if (dl->dl_sharecnt > 1) { 350 dl->dl_sharecnt--; 351 mutex_exit(&dzp->z_lock); 352 return; 353 } 354 prev_dl = &dzp->z_dirlocks; 355 while ((cur_dl = *prev_dl) != dl) 356 prev_dl = &cur_dl->dl_next; 357 *prev_dl = dl->dl_next; 358 cv_broadcast(&dl->dl_cv); 359 mutex_exit(&dzp->z_lock); 360 361 if (dl->dl_namesize != 0) 362 kmem_free(dl->dl_name, dl->dl_namesize); 363 cv_destroy(&dl->dl_cv); 364 kmem_free(dl, sizeof (*dl)); 365} 366 367/* 368 * Look up an entry in a directory. 369 * 370 * NOTE: '.' and '..' are handled as special cases because 371 * no directory entries are actually stored for them. If this is 372 * the root of a filesystem, then '.zfs' is also treated as a 373 * special pseudo-directory. 374 */ 375int 376zfs_dirlook(znode_t *dzp, char *name, vnode_t **vpp, int flags, 377 int *deflg, pathname_t *rpnp) 378{ 379 zfs_dirlock_t *dl; 380 znode_t *zp; 381 int error = 0; 382 383 if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) { 384 *vpp = ZTOV(dzp); 385 VN_HOLD(*vpp); 386 } else if (name[0] == '.' && name[1] == '.' && name[2] == 0) { 387 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 388 /* 389 * If we are a snapshot mounted under .zfs, return 390 * the vp for the snapshot directory. 391 */ 392 if (dzp->z_phys->zp_parent == dzp->z_id && 393 zfsvfs->z_parent != zfsvfs) { 394 error = zfsctl_root_lookup(zfsvfs->z_parent->z_ctldir, 395 "snapshot", vpp, NULL, 0, NULL, kcred, 396 NULL, NULL, NULL); 397 return (error); 398 } 399 rw_enter(&dzp->z_parent_lock, RW_READER); 400 error = zfs_zget(zfsvfs, dzp->z_phys->zp_parent, &zp); 401 if (error == 0) 402 *vpp = ZTOV(zp); 403 rw_exit(&dzp->z_parent_lock); 404 } else if (zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0) { 405 *vpp = zfsctl_root(dzp); 406 } else { 407 int zf; 408 409 zf = ZEXISTS | ZSHARED; 410 if (flags & FIGNORECASE) 411 zf |= ZCILOOK; 412 413 error = zfs_dirent_lock(&dl, dzp, name, &zp, zf, deflg, rpnp); 414 if (error == 0) { 415 *vpp = ZTOV(zp); 416 zfs_dirent_unlock(dl); 417 dzp->z_zn_prefetch = B_TRUE; /* enable prefetching */ 418 } 419 rpnp = NULL; 420 } 421 422 if ((flags & FIGNORECASE) && rpnp && !error) 423 (void) strlcpy(rpnp->pn_buf, name, rpnp->pn_bufsize); 424 425 return (error); 426} 427 428/* 429 * unlinked Set (formerly known as the "delete queue") Error Handling 430 * 431 * When dealing with the unlinked set, we dmu_tx_hold_zap(), but we 432 * don't specify the name of the entry that we will be manipulating. We 433 * also fib and say that we won't be adding any new entries to the 434 * unlinked set, even though we might (this is to lower the minimum file 435 * size that can be deleted in a full filesystem). So on the small 436 * chance that the nlink list is using a fat zap (ie. has more than 437 * 2000 entries), we *may* not pre-read a block that's needed. 438 * Therefore it is remotely possible for some of the assertions 439 * regarding the unlinked set below to fail due to i/o error. On a 440 * nondebug system, this will result in the space being leaked. 441 */ 442void 443zfs_unlinked_add(znode_t *zp, dmu_tx_t *tx) 444{ 445 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 446 447 ASSERT(zp->z_unlinked); 448 ASSERT3U(zp->z_phys->zp_links, ==, 0); 449 450 VERIFY3U(0, ==, 451 zap_add_int(zfsvfs->z_os, zfsvfs->z_unlinkedobj, zp->z_id, tx)); 452} 453 454/* 455 * Clean up any znodes that had no links when we either crashed or 456 * (force) umounted the file system. 457 */ 458void 459zfs_unlinked_drain(zfsvfs_t *zfsvfs) 460{ 461 zap_cursor_t zc; 462 zap_attribute_t zap; 463 dmu_object_info_t doi; 464 znode_t *zp; 465 int error; 466 467 /* 468 * Interate over the contents of the unlinked set. 469 */ 470 for (zap_cursor_init(&zc, zfsvfs->z_os, zfsvfs->z_unlinkedobj); 471 zap_cursor_retrieve(&zc, &zap) == 0; 472 zap_cursor_advance(&zc)) { 473 474 /* 475 * See what kind of object we have in list 476 */ 477 478 error = dmu_object_info(zfsvfs->z_os, 479 zap.za_first_integer, &doi); 480 if (error != 0) 481 continue; 482 483 ASSERT((doi.doi_type == DMU_OT_PLAIN_FILE_CONTENTS) || 484 (doi.doi_type == DMU_OT_DIRECTORY_CONTENTS)); 485 /* 486 * We need to re-mark these list entries for deletion, 487 * so we pull them back into core and set zp->z_unlinked. 488 */ 489 error = zfs_zget(zfsvfs, zap.za_first_integer, &zp); 490 491 /* 492 * We may pick up znodes that are already marked for deletion. 493 * This could happen during the purge of an extended attribute 494 * directory. All we need to do is skip over them, since they 495 * are already in the system marked z_unlinked. 496 */ 497 if (error != 0) 498 continue; 499 500 zp->z_unlinked = B_TRUE; 501 VN_RELE(ZTOV(zp)); 502 } 503 zap_cursor_fini(&zc); 504} 505 506/* 507 * Delete the entire contents of a directory. Return a count 508 * of the number of entries that could not be deleted. If we encounter 509 * an error, return a count of at least one so that the directory stays 510 * in the unlinked set. 511 * 512 * NOTE: this function assumes that the directory is inactive, 513 * so there is no need to lock its entries before deletion. 514 * Also, it assumes the directory contents is *only* regular 515 * files. 516 */ 517static int 518zfs_purgedir(znode_t *dzp) 519{ 520 zap_cursor_t zc; 521 zap_attribute_t zap; 522 znode_t *xzp; 523 dmu_tx_t *tx; 524 zfsvfs_t *zfsvfs = dzp->z_zfsvfs; 525 zfs_dirlock_t dl; 526 int skipped = 0; 527 int error; 528 529 for (zap_cursor_init(&zc, zfsvfs->z_os, dzp->z_id); 530 (error = zap_cursor_retrieve(&zc, &zap)) == 0; 531 zap_cursor_advance(&zc)) { 532 error = zfs_zget(zfsvfs, 533 ZFS_DIRENT_OBJ(zap.za_first_integer), &xzp); 534 if (error) { 535 skipped += 1; 536 continue; 537 } 538 539 ASSERT((ZTOV(xzp)->v_type == VREG) || 540 (ZTOV(xzp)->v_type == VLNK)); 541 542 tx = dmu_tx_create(zfsvfs->z_os); 543 dmu_tx_hold_bonus(tx, dzp->z_id); 544 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, zap.za_name); 545 dmu_tx_hold_bonus(tx, xzp->z_id); 546 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 547 error = dmu_tx_assign(tx, TXG_WAIT); 548 if (error) { 549 dmu_tx_abort(tx); 550 VN_RELE(ZTOV(xzp)); 551 skipped += 1; 552 continue; 553 } 554 bzero(&dl, sizeof (dl)); 555 dl.dl_dzp = dzp; 556 dl.dl_name = zap.za_name; 557 558 error = zfs_link_destroy(&dl, xzp, tx, 0, NULL); 559 if (error) 560 skipped += 1; 561 dmu_tx_commit(tx); 562 563 VN_RELE(ZTOV(xzp)); 564 } 565 zap_cursor_fini(&zc); 566 if (error != ENOENT) 567 skipped += 1; 568 return (skipped); 569} 570 571void 572zfs_rmnode(znode_t *zp) 573{ 574 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 575 objset_t *os = zfsvfs->z_os; 576 znode_t *xzp = NULL; 577 dmu_tx_t *tx; 578 uint64_t acl_obj; 579 int error; 580 581 ASSERT(zp->z_phys->zp_links == 0); 582 583 /* 584 * If this is a ZIL replay then leave the object in the unlinked set. 585 * Otherwise we can get a deadlock, because the delete can be 586 * quite large and span multiple tx's and txgs, but each replay 587 * creates a tx to atomically run the replay function and mark the 588 * replay record as complete. We deadlock trying to start a tx in 589 * a new txg to further the deletion but can't because the replay 590 * tx hasn't finished. 591 * 592 * We actually delete the object if we get a failure to create an 593 * object in zil_replay_log_record(), or after calling zil_replay(). 594 */ 595 if (zfsvfs->z_assign >= TXG_INITIAL) { 596 zfs_znode_dmu_fini(zp); 597 zfs_znode_free(zp); 598 return; 599 } 600 601 /* 602 * If this is an attribute directory, purge its contents. 603 */ 604 if (ZTOV(zp) != NULL && ZTOV(zp)->v_type == VDIR && 605 (zp->z_phys->zp_flags & ZFS_XATTR)) { 606 if (zfs_purgedir(zp) != 0) { 607 /* 608 * Not enough space to delete some xattrs. 609 * Leave it in the unlinked set. 610 */ 611 zfs_znode_dmu_fini(zp); 612 zfs_znode_free(zp); 613 return; 614 } 615 } 616 617 /* 618 * Free up all the data in the file. 619 */ 620 error = dmu_free_long_range(os, zp->z_id, 0, DMU_OBJECT_END); 621 if (error) { 622 /* 623 * Not enough space. Leave the file in the unlinked set. 624 */ 625 zfs_znode_dmu_fini(zp); 626 zfs_znode_free(zp); 627 return; 628 } 629 630 /* 631 * If the file has extended attributes, we're going to unlink 632 * the xattr dir. 633 */ 634 if (zp->z_phys->zp_xattr) { 635 error = zfs_zget(zfsvfs, zp->z_phys->zp_xattr, &xzp); 636 ASSERT(error == 0); 637 } 638 639 acl_obj = zp->z_phys->zp_acl.z_acl_extern_obj; 640 641 /* 642 * Set up the final transaction. 643 */ 644 tx = dmu_tx_create(os); 645 dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END); 646 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL); 647 if (xzp) { 648 dmu_tx_hold_bonus(tx, xzp->z_id); 649 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, TRUE, NULL); 650 } 651 if (acl_obj) 652 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END); 653 error = dmu_tx_assign(tx, TXG_WAIT); 654 if (error) { 655 /* 656 * Not enough space to delete the file. Leave it in the 657 * unlinked set, leaking it until the fs is remounted (at 658 * which point we'll call zfs_unlinked_drain() to process it). 659 */ 660 dmu_tx_abort(tx); 661 zfs_znode_dmu_fini(zp); 662 zfs_znode_free(zp); 663 goto out; 664 } 665 666 if (xzp) { 667 dmu_buf_will_dirty(xzp->z_dbuf, tx); 668 mutex_enter(&xzp->z_lock); 669 xzp->z_unlinked = B_TRUE; /* mark xzp for deletion */ 670 xzp->z_phys->zp_links = 0; /* no more links to it */ 671 mutex_exit(&xzp->z_lock); 672 zfs_unlinked_add(xzp, tx); 673 } 674 675 /* Remove this znode from the unlinked set */ 676 VERIFY3U(0, ==, 677 zap_remove_int(zfsvfs->z_os, zfsvfs->z_unlinkedobj, zp->z_id, tx)); 678 679 zfs_znode_delete(zp, tx); 680 681 dmu_tx_commit(tx); 682out: 683 if (xzp) 684 VN_RELE(ZTOV(xzp)); 685} 686 687static uint64_t 688zfs_dirent(znode_t *zp) 689{ 690 uint64_t de = zp->z_id; 691 if (zp->z_zfsvfs->z_version >= ZPL_VERSION_DIRENT_TYPE) 692 de |= IFTODT((zp)->z_phys->zp_mode) << 60; 693 return (de); 694} 695 696/* 697 * Link zp into dl. Can only fail if zp has been unlinked. 698 */ 699int 700zfs_link_create(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag) 701{ 702 znode_t *dzp = dl->dl_dzp; 703 vnode_t *vp = ZTOV(zp); 704 uint64_t value; 705 int zp_is_dir = (vp->v_type == VDIR); 706 int error; 707 708 dmu_buf_will_dirty(zp->z_dbuf, tx); 709 mutex_enter(&zp->z_lock); 710 711 if (!(flag & ZRENAMING)) { 712 if (zp->z_unlinked) { /* no new links to unlinked zp */ 713 ASSERT(!(flag & (ZNEW | ZEXISTS))); 714 mutex_exit(&zp->z_lock); 715 return (ENOENT); 716 } 717 zp->z_phys->zp_links++; 718 } 719 zp->z_phys->zp_parent = dzp->z_id; /* dzp is now zp's parent */ 720 721 if (!(flag & ZNEW)) 722 zfs_time_stamper_locked(zp, STATE_CHANGED, tx); 723 mutex_exit(&zp->z_lock); 724 725 dmu_buf_will_dirty(dzp->z_dbuf, tx); 726 mutex_enter(&dzp->z_lock); 727 dzp->z_phys->zp_size++; /* one dirent added */ 728 dzp->z_phys->zp_links += zp_is_dir; /* ".." link from zp */ 729 zfs_time_stamper_locked(dzp, CONTENT_MODIFIED, tx); 730 mutex_exit(&dzp->z_lock); 731 732 value = zfs_dirent(zp); 733 error = zap_add(zp->z_zfsvfs->z_os, dzp->z_id, dl->dl_name, 734 8, 1, &value, tx); 735 ASSERT(error == 0); 736 737 dnlc_update(ZTOV(dzp), dl->dl_name, vp); 738 739 return (0); 740} 741 742/* 743 * Unlink zp from dl, and mark zp for deletion if this was the last link. 744 * Can fail if zp is a mount point (EBUSY) or a non-empty directory (EEXIST). 745 * If 'unlinkedp' is NULL, we put unlinked znodes on the unlinked list. 746 * If it's non-NULL, we use it to indicate whether the znode needs deletion, 747 * and it's the caller's job to do it. 748 */ 749int 750zfs_link_destroy(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag, 751 boolean_t *unlinkedp) 752{ 753 znode_t *dzp = dl->dl_dzp; 754 vnode_t *vp = ZTOV(zp); 755 int zp_is_dir = (vp->v_type == VDIR); 756 boolean_t unlinked = B_FALSE; 757 int error; 758 759 dnlc_remove(ZTOV(dzp), dl->dl_name); 760 761 if (!(flag & ZRENAMING)) { 762 dmu_buf_will_dirty(zp->z_dbuf, tx); 763 764 if (vn_vfswlock(vp)) /* prevent new mounts on zp */ 765 return (EBUSY); 766 767 if (vn_ismntpt(vp)) { /* don't remove mount point */ 768 vn_vfsunlock(vp); 769 return (EBUSY); 770 } 771 772 mutex_enter(&zp->z_lock); 773 if (zp_is_dir && !zfs_dirempty(zp)) { /* dir not empty */ 774 mutex_exit(&zp->z_lock); 775 vn_vfsunlock(vp); 776 return (ENOTEMPTY); 777 } 778 if (zp->z_phys->zp_links <= zp_is_dir) { 779 zfs_panic_recover("zfs: link count on vnode %p is %u, " 780 "should be at least %u", zp->z_vnode, 781 (int)zp->z_phys->zp_links, 782 zp_is_dir + 1); 783 zp->z_phys->zp_links = zp_is_dir + 1; 784 } 785 if (--zp->z_phys->zp_links == zp_is_dir) { 786 zp->z_unlinked = B_TRUE; 787 zp->z_phys->zp_links = 0; 788 unlinked = B_TRUE; 789 } else { 790 zfs_time_stamper_locked(zp, STATE_CHANGED, tx); 791 } 792 mutex_exit(&zp->z_lock); 793 vn_vfsunlock(vp); 794 } 795 796 dmu_buf_will_dirty(dzp->z_dbuf, tx); 797 mutex_enter(&dzp->z_lock); 798 dzp->z_phys->zp_size--; /* one dirent removed */ 799 dzp->z_phys->zp_links -= zp_is_dir; /* ".." link from zp */ 800 zfs_time_stamper_locked(dzp, CONTENT_MODIFIED, tx); 801 mutex_exit(&dzp->z_lock); 802 803 if (zp->z_zfsvfs->z_norm) { 804 if (((zp->z_zfsvfs->z_case == ZFS_CASE_INSENSITIVE) && 805 (flag & ZCIEXACT)) || 806 ((zp->z_zfsvfs->z_case == ZFS_CASE_MIXED) && 807 !(flag & ZCILOOK))) 808 error = zap_remove_norm(zp->z_zfsvfs->z_os, 809 dzp->z_id, dl->dl_name, MT_EXACT, tx); 810 else 811 error = zap_remove_norm(zp->z_zfsvfs->z_os, 812 dzp->z_id, dl->dl_name, MT_FIRST, tx); 813 } else { 814 error = zap_remove(zp->z_zfsvfs->z_os, 815 dzp->z_id, dl->dl_name, tx); 816 } 817 ASSERT(error == 0); 818 819 if (unlinkedp != NULL) 820 *unlinkedp = unlinked; 821 else if (unlinked) 822 zfs_unlinked_add(zp, tx); 823 824 return (0); 825} 826 827/* 828 * Indicate whether the directory is empty. Works with or without z_lock 829 * held, but can only be consider a hint in the latter case. Returns true 830 * if only "." and ".." remain and there's no work in progress. 831 */ 832boolean_t 833zfs_dirempty(znode_t *dzp) 834{ 835 return (dzp->z_phys->zp_size == 2 && dzp->z_dirlocks == 0); 836} 837 838int 839zfs_make_xattrdir(znode_t *zp, vattr_t *vap, vnode_t **xvpp, cred_t *cr) 840{ 841 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 842 znode_t *xzp; 843 dmu_tx_t *tx; 844 int error; 845 zfs_fuid_info_t *fuidp = NULL; 846 847 *xvpp = NULL; 848 849 /* 850 * In FreeBSD, access checking for creating an EA is being done 851 * in zfs_setextattr(), 852 */ 853#ifndef __FreeBSD__ 854 if (error = zfs_zaccess(zp, ACE_WRITE_NAMED_ATTRS, 0, B_FALSE, cr)) 855 return (error); 856#endif 857 858 tx = dmu_tx_create(zfsvfs->z_os); 859 dmu_tx_hold_bonus(tx, zp->z_id); 860 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL); 861 if (IS_EPHEMERAL(crgetuid(cr)) || IS_EPHEMERAL(crgetgid(cr))) { 862 if (zfsvfs->z_fuid_obj == 0) { 863 dmu_tx_hold_bonus(tx, DMU_NEW_OBJECT); 864 dmu_tx_hold_write(tx, DMU_NEW_OBJECT, 0, 865 FUID_SIZE_ESTIMATE(zfsvfs)); 866 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, FALSE, NULL); 867 } else { 868 dmu_tx_hold_bonus(tx, zfsvfs->z_fuid_obj); 869 dmu_tx_hold_write(tx, zfsvfs->z_fuid_obj, 0, 870 FUID_SIZE_ESTIMATE(zfsvfs)); 871 } 872 } 873 error = dmu_tx_assign(tx, zfsvfs->z_assign); 874 if (error) { 875 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) 876 dmu_tx_wait(tx); 877 dmu_tx_abort(tx); 878 return (error); 879 } 880 zfs_mknode(zp, vap, tx, cr, IS_XATTR, &xzp, 0, NULL, &fuidp); 881 ASSERT(xzp->z_phys->zp_parent == zp->z_id); 882 dmu_buf_will_dirty(zp->z_dbuf, tx); 883 zp->z_phys->zp_xattr = xzp->z_id; 884 885 (void) zfs_log_create(zfsvfs->z_log, tx, TX_MKXATTR, zp, 886 xzp, "", NULL, fuidp, vap); 887 if (fuidp) 888 zfs_fuid_info_free(fuidp); 889 dmu_tx_commit(tx); 890 891 *xvpp = ZTOV(xzp); 892 893 return (0); 894} 895 896/* 897 * Return a znode for the extended attribute directory for zp. 898 * ** If the directory does not already exist, it is created ** 899 * 900 * IN: zp - znode to obtain attribute directory from 901 * cr - credentials of caller 902 * flags - flags from the VOP_LOOKUP call 903 * 904 * OUT: xzpp - pointer to extended attribute znode 905 * 906 * RETURN: 0 on success 907 * error number on failure 908 */ 909int 910zfs_get_xattrdir(znode_t *zp, vnode_t **xvpp, cred_t *cr, int flags) 911{ 912 zfsvfs_t *zfsvfs = zp->z_zfsvfs; 913 znode_t *xzp; 914 zfs_dirlock_t *dl; 915 vattr_t va; 916 int error; 917top: 918 error = zfs_dirent_lock(&dl, zp, "", &xzp, ZXATTR, NULL, NULL); 919 if (error) 920 return (error); 921 922 if (xzp != NULL) { 923 *xvpp = ZTOV(xzp); 924 zfs_dirent_unlock(dl); 925 return (0); 926 } 927 928 ASSERT(zp->z_phys->zp_xattr == 0); 929 930 if (!(flags & CREATE_XATTR_DIR)) { 931 zfs_dirent_unlock(dl); 932#ifdef __FreeBSD__ 933 return (ENOATTR); 934#else 935 return (ENOENT); 936#endif 937 } 938 939 if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) { 940 zfs_dirent_unlock(dl); 941 return (EROFS); 942 } 943 944 /* 945 * The ability to 'create' files in an attribute 946 * directory comes from the write_xattr permission on the base file. 947 * 948 * The ability to 'search' an attribute directory requires 949 * read_xattr permission on the base file. 950 * 951 * Once in a directory the ability to read/write attributes 952 * is controlled by the permissions on the attribute file. 953 */ 954 va.va_mask = AT_TYPE | AT_MODE | AT_UID | AT_GID; 955 va.va_type = VDIR; 956 va.va_mode = S_IFDIR | S_ISVTX | 0777; 957 zfs_fuid_map_ids(zp, cr, &va.va_uid, &va.va_gid); 958 959 error = zfs_make_xattrdir(zp, &va, xvpp, cr); 960 zfs_dirent_unlock(dl); 961 962 if (error == ERESTART && zfsvfs->z_assign == TXG_NOWAIT) { 963 /* NB: we already did dmu_tx_wait() if necessary */ 964 goto top; 965 } 966 if (error == 0) 967 VOP_UNLOCK(*xvpp, 0); 968 969 return (error); 970} 971 972/* 973 * Decide whether it is okay to remove within a sticky directory. 974 * 975 * In sticky directories, write access is not sufficient; 976 * you can remove entries from a directory only if: 977 * 978 * you own the directory, 979 * you own the entry, 980 * the entry is a plain file and you have write access, 981 * or you are privileged (checked in secpolicy...). 982 * 983 * The function returns 0 if remove access is granted. 984 */ 985int 986zfs_sticky_remove_access(znode_t *zdp, znode_t *zp, cred_t *cr) 987{ 988 uid_t uid; 989 uid_t downer; 990 uid_t fowner; 991 zfsvfs_t *zfsvfs = zdp->z_zfsvfs; 992 993 if (zdp->z_zfsvfs->z_assign >= TXG_INITIAL) /* ZIL replay */ 994 return (0); 995 996 if ((zdp->z_phys->zp_mode & S_ISVTX) == 0) 997 return (0); 998 999 downer = zfs_fuid_map_id(zfsvfs, zdp->z_phys->zp_uid, cr, ZFS_OWNER); 1000 fowner = zfs_fuid_map_id(zfsvfs, zp->z_phys->zp_uid, cr, ZFS_OWNER); 1001 1002 if ((uid = crgetuid(cr)) == downer || uid == fowner || 1003 (ZTOV(zp)->v_type == VREG && 1004 zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr) == 0)) 1005 return (0); 1006 else 1007 return (secpolicy_vnode_remove(ZTOV(zp), cr)); 1008} 1009