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 (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
  23  */
  24 
  25 #include <sys/types.h>
  26 #include <sys/param.h>
  27 #include <sys/time.h>
  28 #include <sys/systm.h>
  29 #include <sys/sysmacros.h>
  30 #include <sys/resource.h>
  31 #include <sys/vfs.h>
  32 #include <sys/vnode.h>
  33 #include <sys/file.h>
  34 #include <sys/mode.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/unistd.h>
  42 #include <sys/sunddi.h>
  43 #include <sys/random.h>
  44 #include <sys/policy.h>
  45 #include <sys/zfs_dir.h>
  46 #include <sys/zfs_acl.h>
  47 #include <sys/fs/zfs.h>
  48 #include "fs/fs_subr.h"
  49 #include <sys/zap.h>
  50 #include <sys/dmu.h>
  51 #include <sys/atomic.h>
  52 #include <sys/zfs_ctldir.h>
  53 #include <sys/zfs_fuid.h>
  54 #include <sys/sa.h>
  55 #include <sys/zfs_sa.h>
  56 #include <sys/dnlc.h>
  57 #include <sys/extdirent.h>
  58 
  59 /*
  60  * zfs_match_find() is used by zfs_dirent_lock() to peform zap lookups
  61  * of names after deciding which is the appropriate lookup interface.
  62  */
  63 static int
  64 zfs_match_find(zfsvfs_t *zfsvfs, znode_t *dzp, char *name, boolean_t exact,
  65     boolean_t update, int *deflags, pathname_t *rpnp, uint64_t *zoid)
  66 {
  67         int error;
  68 
  69         if (zfsvfs->z_norm) {
  70                 matchtype_t mt = MT_FIRST;
  71                 boolean_t conflict = B_FALSE;
  72                 size_t bufsz = 0;
  73                 char *buf = NULL;
  74 
  75                 if (rpnp) {
  76                         buf = rpnp->pn_buf;
  77                         bufsz = rpnp->pn_bufsize;
  78                 }
  79                 if (exact)
  80                         mt = MT_EXACT;
  81                 /*
  82                  * In the non-mixed case we only expect there would ever
  83                  * be one match, but we need to use the normalizing lookup.
  84                  */
  85                 error = zap_lookup_norm(zfsvfs->z_os, dzp->z_id, name, 8, 1,
  86                     zoid, mt, buf, bufsz, &conflict);
  87                 if (!error && deflags)
  88                         *deflags = conflict ? ED_CASE_CONFLICT : 0;
  89         } else {
  90                 error = zap_lookup(zfsvfs->z_os, dzp->z_id, name, 8, 1, zoid);
  91         }
  92         *zoid = ZFS_DIRENT_OBJ(*zoid);
  93 
  94         if (error == ENOENT && update)
  95                 dnlc_update(ZTOV(dzp), name, DNLC_NO_VNODE);
  96 
  97         return (error);
  98 }
  99 
 100 /*
 101  * Lock a directory entry.  A dirlock on <dzp, name> protects that name
 102  * in dzp's directory zap object.  As long as you hold a dirlock, you can
 103  * assume two things: (1) dzp cannot be reaped, and (2) no other thread
 104  * can change the zap entry for (i.e. link or unlink) this name.
 105  *
 106  * Input arguments:
 107  *      dzp     - znode for directory
 108  *      name    - name of entry to lock
 109  *      flag    - ZNEW: if the entry already exists, fail with EEXIST.
 110  *                ZEXISTS: if the entry does not exist, fail with ENOENT.
 111  *                ZSHARED: allow concurrent access with other ZSHARED callers.
 112  *                ZXATTR: we want dzp's xattr directory
 113  *                ZCILOOK: On a mixed sensitivity file system,
 114  *                         this lookup should be case-insensitive.
 115  *                ZCIEXACT: On a purely case-insensitive file system,
 116  *                          this lookup should be case-sensitive.
 117  *                ZRENAMING: we are locking for renaming, force narrow locks
 118  *                ZHAVELOCK: Don't grab the z_name_lock for this call. The
 119  *                           current thread already holds it.
 120  *
 121  * Output arguments:
 122  *      zpp     - pointer to the znode for the entry (NULL if there isn't one)
 123  *      dlpp    - pointer to the dirlock for this entry (NULL on error)
 124  *      direntflags - (case-insensitive lookup only)
 125  *              flags if multiple case-sensitive matches exist in directory
 126  *      realpnp     - (case-insensitive lookup only)
 127  *              actual name matched within the directory
 128  *
 129  * Return value: 0 on success or errno on failure.
 130  *
 131  * NOTE: Always checks for, and rejects, '.' and '..'.
 132  * NOTE: For case-insensitive file systems we take wide locks (see below),
 133  *       but return znode pointers to a single match.
 134  */
 135 int
 136 zfs_dirent_lock(zfs_dirlock_t **dlpp, znode_t *dzp, char *name, znode_t **zpp,
 137     int flag, int *direntflags, pathname_t *realpnp)
 138 {
 139         zfsvfs_t        *zfsvfs = dzp->z_zfsvfs;
 140         zfs_dirlock_t   *dl;
 141         boolean_t       update;
 142         boolean_t       exact;
 143         uint64_t        zoid;
 144         vnode_t         *vp = NULL;
 145         int             error = 0;
 146         int             cmpflags;
 147 
 148         *zpp = NULL;
 149         *dlpp = NULL;
 150 
 151         /*
 152          * Verify that we are not trying to lock '.', '..', or '.zfs'
 153          */
 154         if (name[0] == '.' &&
 155             (name[1] == '\0' || (name[1] == '.' && name[2] == '\0')) ||
 156             zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0)
 157                 return (EEXIST);
 158 
 159         /*
 160          * Case sensitivity and normalization preferences are set when
 161          * the file system is created.  These are stored in the
 162          * zfsvfs->z_case and zfsvfs->z_norm fields.  These choices
 163          * affect what vnodes can be cached in the DNLC, how we
 164          * perform zap lookups, and the "width" of our dirlocks.
 165          *
 166          * A normal dirlock locks a single name.  Note that with
 167          * normalization a name can be composed multiple ways, but
 168          * when normalized, these names all compare equal.  A wide
 169          * dirlock locks multiple names.  We need these when the file
 170          * system is supporting mixed-mode access.  It is sometimes
 171          * necessary to lock all case permutations of file name at
 172          * once so that simultaneous case-insensitive/case-sensitive
 173          * behaves as rationally as possible.
 174          */
 175 
 176         /*
 177          * Decide if exact matches should be requested when performing
 178          * a zap lookup on file systems supporting case-insensitive
 179          * access.
 180          */
 181         exact =
 182             ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE) && (flag & ZCIEXACT)) ||
 183             ((zfsvfs->z_case == ZFS_CASE_MIXED) && !(flag & ZCILOOK));
 184 
 185         /*
 186          * Only look in or update the DNLC if we are looking for the
 187          * name on a file system that does not require normalization
 188          * or case folding.  We can also look there if we happen to be
 189          * on a non-normalizing, mixed sensitivity file system IF we
 190          * are looking for the exact name.
 191          *
 192          * Maybe can add TO-UPPERed version of name to dnlc in ci-only
 193          * case for performance improvement?
 194          */
 195         update = !zfsvfs->z_norm ||
 196             ((zfsvfs->z_case == ZFS_CASE_MIXED) &&
 197             !(zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER) && !(flag & ZCILOOK));
 198 
 199         /*
 200          * ZRENAMING indicates we are in a situation where we should
 201          * take narrow locks regardless of the file system's
 202          * preferences for normalizing and case folding.  This will
 203          * prevent us deadlocking trying to grab the same wide lock
 204          * twice if the two names happen to be case-insensitive
 205          * matches.
 206          */
 207         if (flag & ZRENAMING)
 208                 cmpflags = 0;
 209         else
 210                 cmpflags = zfsvfs->z_norm;
 211 
 212         /*
 213          * Wait until there are no locks on this name.
 214          *
 215          * Don't grab the the lock if it is already held. However, cannot
 216          * have both ZSHARED and ZHAVELOCK together.
 217          */
 218         ASSERT(!(flag & ZSHARED) || !(flag & ZHAVELOCK));
 219         if (!(flag & ZHAVELOCK))
 220                 rw_enter(&dzp->z_name_lock, RW_READER);
 221 
 222         mutex_enter(&dzp->z_lock);
 223         for (;;) {
 224                 if (dzp->z_unlinked) {
 225                         mutex_exit(&dzp->z_lock);
 226                         if (!(flag & ZHAVELOCK))
 227                                 rw_exit(&dzp->z_name_lock);
 228                         return (ENOENT);
 229                 }
 230                 for (dl = dzp->z_dirlocks; dl != NULL; dl = dl->dl_next) {
 231                         if ((u8_strcmp(name, dl->dl_name, 0, cmpflags,
 232                             U8_UNICODE_LATEST, &error) == 0) || error != 0)
 233                                 break;
 234                 }
 235                 if (error != 0) {
 236                         mutex_exit(&dzp->z_lock);
 237                         if (!(flag & ZHAVELOCK))
 238                                 rw_exit(&dzp->z_name_lock);
 239                         return (ENOENT);
 240                 }
 241                 if (dl == NULL) {
 242                         /*
 243                          * Allocate a new dirlock and add it to the list.
 244                          */
 245                         dl = kmem_alloc(sizeof (zfs_dirlock_t), KM_SLEEP);
 246                         cv_init(&dl->dl_cv, NULL, CV_DEFAULT, NULL);
 247                         dl->dl_name = name;
 248                         dl->dl_sharecnt = 0;
 249                         dl->dl_namelock = 0;
 250                         dl->dl_namesize = 0;
 251                         dl->dl_dzp = dzp;
 252                         dl->dl_next = dzp->z_dirlocks;
 253                         dzp->z_dirlocks = dl;
 254                         break;
 255                 }
 256                 if ((flag & ZSHARED) && dl->dl_sharecnt != 0)
 257                         break;
 258                 cv_wait(&dl->dl_cv, &dzp->z_lock);
 259         }
 260 
 261         /*
 262          * If the z_name_lock was NOT held for this dirlock record it.
 263          */
 264         if (flag & ZHAVELOCK)
 265                 dl->dl_namelock = 1;
 266 
 267         if ((flag & ZSHARED) && ++dl->dl_sharecnt > 1 && dl->dl_namesize == 0) {
 268                 /*
 269                  * We're the second shared reference to dl.  Make a copy of
 270                  * dl_name in case the first thread goes away before we do.
 271                  * Note that we initialize the new name before storing its
 272                  * pointer into dl_name, because the first thread may load
 273                  * dl->dl_name at any time.  He'll either see the old value,
 274                  * which is his, or the new shared copy; either is OK.
 275                  */
 276                 dl->dl_namesize = strlen(dl->dl_name) + 1;
 277                 name = kmem_alloc(dl->dl_namesize, KM_SLEEP);
 278                 bcopy(dl->dl_name, name, dl->dl_namesize);
 279                 dl->dl_name = name;
 280         }
 281 
 282         mutex_exit(&dzp->z_lock);
 283 
 284         /*
 285          * We have a dirlock on the name.  (Note that it is the dirlock,
 286          * not the dzp's z_lock, that protects the name in the zap object.)
 287          * See if there's an object by this name; if so, put a hold on it.
 288          */
 289         if (flag & ZXATTR) {
 290                 error = sa_lookup(dzp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &zoid,
 291                     sizeof (zoid));
 292                 if (error == 0)
 293                         error = (zoid == 0 ? ENOENT : 0);
 294         } else {
 295                 if (update)
 296                         vp = dnlc_lookup(ZTOV(dzp), name);
 297                 if (vp == DNLC_NO_VNODE) {
 298                         VN_RELE(vp);
 299                         error = ENOENT;
 300                 } else if (vp) {
 301                         if (flag & ZNEW) {
 302                                 zfs_dirent_unlock(dl);
 303                                 VN_RELE(vp);
 304                                 return (EEXIST);
 305                         }
 306                         *dlpp = dl;
 307                         *zpp = VTOZ(vp);
 308                         return (0);
 309                 } else {
 310                         error = zfs_match_find(zfsvfs, dzp, name, exact,
 311                             update, direntflags, realpnp, &zoid);
 312                 }
 313         }
 314         if (error) {
 315                 if (error != ENOENT || (flag & ZEXISTS)) {
 316                         zfs_dirent_unlock(dl);
 317                         return (error);
 318                 }
 319         } else {
 320                 if (flag & ZNEW) {
 321                         zfs_dirent_unlock(dl);
 322                         return (EEXIST);
 323                 }
 324                 error = zfs_zget(zfsvfs, zoid, zpp);
 325                 if (error) {
 326                         zfs_dirent_unlock(dl);
 327                         return (error);
 328                 }
 329                 if (!(flag & ZXATTR) && update)
 330                         dnlc_update(ZTOV(dzp), name, ZTOV(*zpp));
 331         }
 332 
 333         *dlpp = dl;
 334 
 335         return (0);
 336 }
 337 
 338 /*
 339  * Unlock this directory entry and wake anyone who was waiting for it.
 340  */
 341 void
 342 zfs_dirent_unlock(zfs_dirlock_t *dl)
 343 {
 344         znode_t *dzp = dl->dl_dzp;
 345         zfs_dirlock_t **prev_dl, *cur_dl;
 346 
 347         mutex_enter(&dzp->z_lock);
 348 
 349         if (!dl->dl_namelock)
 350                 rw_exit(&dzp->z_name_lock);
 351 
 352         if (dl->dl_sharecnt > 1) {
 353                 dl->dl_sharecnt--;
 354                 mutex_exit(&dzp->z_lock);
 355                 return;
 356         }
 357         prev_dl = &dzp->z_dirlocks;
 358         while ((cur_dl = *prev_dl) != dl)
 359                 prev_dl = &cur_dl->dl_next;
 360         *prev_dl = dl->dl_next;
 361         cv_broadcast(&dl->dl_cv);
 362         mutex_exit(&dzp->z_lock);
 363 
 364         if (dl->dl_namesize != 0)
 365                 kmem_free(dl->dl_name, dl->dl_namesize);
 366         cv_destroy(&dl->dl_cv);
 367         kmem_free(dl, sizeof (*dl));
 368 }
 369 
 370 /*
 371  * Look up an entry in a directory.
 372  *
 373  * NOTE: '.' and '..' are handled as special cases because
 374  *      no directory entries are actually stored for them.  If this is
 375  *      the root of a filesystem, then '.zfs' is also treated as a
 376  *      special pseudo-directory.
 377  */
 378 int
 379 zfs_dirlook(znode_t *dzp, char *name, vnode_t **vpp, int flags,
 380     int *deflg, pathname_t *rpnp)
 381 {
 382         zfs_dirlock_t *dl;
 383         znode_t *zp;
 384         int error = 0;
 385         uint64_t parent;
 386 
 387         if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) {
 388                 *vpp = ZTOV(dzp);
 389                 VN_HOLD(*vpp);
 390         } else if (name[0] == '.' && name[1] == '.' && name[2] == 0) {
 391                 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
 392 
 393                 /*
 394                  * If we are a snapshot mounted under .zfs, return
 395                  * the vp for the snapshot directory.
 396                  */
 397                 if ((error = sa_lookup(dzp->z_sa_hdl,
 398                     SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0)
 399                         return (error);
 400                 if (parent == dzp->z_id && zfsvfs->z_parent != zfsvfs) {
 401                         error = zfsctl_root_lookup(zfsvfs->z_parent->z_ctldir,
 402                             "snapshot", vpp, NULL, 0, NULL, kcred,
 403                             NULL, NULL, NULL);
 404                         return (error);
 405                 }
 406                 rw_enter(&dzp->z_parent_lock, RW_READER);
 407                 error = zfs_zget(zfsvfs, parent, &zp);
 408                 if (error == 0)
 409                         *vpp = ZTOV(zp);
 410                 rw_exit(&dzp->z_parent_lock);
 411         } else if (zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0) {
 412                 *vpp = zfsctl_root(dzp);
 413         } else {
 414                 int zf;
 415 
 416                 zf = ZEXISTS | ZSHARED;
 417                 if (flags & FIGNORECASE)
 418                         zf |= ZCILOOK;
 419 
 420                 error = zfs_dirent_lock(&dl, dzp, name, &zp, zf, deflg, rpnp);
 421                 if (error == 0) {
 422                         *vpp = ZTOV(zp);
 423                         zfs_dirent_unlock(dl);
 424                         dzp->z_zn_prefetch = B_TRUE; /* enable prefetching */
 425                 }
 426                 rpnp = NULL;
 427         }
 428 
 429         if ((flags & FIGNORECASE) && rpnp && !error)
 430                 (void) strlcpy(rpnp->pn_buf, name, rpnp->pn_bufsize);
 431 
 432         return (error);
 433 }
 434 
 435 /*
 436  * unlinked Set (formerly known as the "delete queue") Error Handling
 437  *
 438  * When dealing with the unlinked set, we dmu_tx_hold_zap(), but we
 439  * don't specify the name of the entry that we will be manipulating.  We
 440  * also fib and say that we won't be adding any new entries to the
 441  * unlinked set, even though we might (this is to lower the minimum file
 442  * size that can be deleted in a full filesystem).  So on the small
 443  * chance that the nlink list is using a fat zap (ie. has more than
 444  * 2000 entries), we *may* not pre-read a block that's needed.
 445  * Therefore it is remotely possible for some of the assertions
 446  * regarding the unlinked set below to fail due to i/o error.  On a
 447  * nondebug system, this will result in the space being leaked.
 448  */
 449 void
 450 zfs_unlinked_add(znode_t *zp, dmu_tx_t *tx)
 451 {
 452         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
 453 
 454         ASSERT(zp->z_unlinked);
 455         ASSERT(zp->z_links == 0);
 456 
 457         VERIFY3U(0, ==,
 458             zap_add_int(zfsvfs->z_os, zfsvfs->z_unlinkedobj, zp->z_id, tx));
 459 }
 460 
 461 /*
 462  * Clean up any znodes that had no links when we either crashed or
 463  * (force) umounted the file system.
 464  */
 465 void
 466 zfs_unlinked_drain(zfsvfs_t *zfsvfs)
 467 {
 468         zap_cursor_t    zc;
 469         zap_attribute_t zap;
 470         dmu_object_info_t doi;
 471         znode_t         *zp;
 472         int             error;
 473 
 474         /*
 475          * Interate over the contents of the unlinked set.
 476          */
 477         for (zap_cursor_init(&zc, zfsvfs->z_os, zfsvfs->z_unlinkedobj);
 478             zap_cursor_retrieve(&zc, &zap) == 0;
 479             zap_cursor_advance(&zc)) {
 480 
 481                 /*
 482                  * See what kind of object we have in list
 483                  */
 484 
 485                 error = dmu_object_info(zfsvfs->z_os,
 486                     zap.za_first_integer, &doi);
 487                 if (error != 0)
 488                         continue;
 489 
 490                 ASSERT((doi.doi_type == DMU_OT_PLAIN_FILE_CONTENTS) ||
 491                     (doi.doi_type == DMU_OT_DIRECTORY_CONTENTS));
 492                 /*
 493                  * We need to re-mark these list entries for deletion,
 494                  * so we pull them back into core and set zp->z_unlinked.
 495                  */
 496                 error = zfs_zget(zfsvfs, zap.za_first_integer, &zp);
 497 
 498                 /*
 499                  * We may pick up znodes that are already marked for deletion.
 500                  * This could happen during the purge of an extended attribute
 501                  * directory.  All we need to do is skip over them, since they
 502                  * are already in the system marked z_unlinked.
 503                  */
 504                 if (error != 0)
 505                         continue;
 506 
 507                 zp->z_unlinked = B_TRUE;
 508                 VN_RELE(ZTOV(zp));
 509         }
 510         zap_cursor_fini(&zc);
 511 }
 512 
 513 /*
 514  * Delete the entire contents of a directory.  Return a count
 515  * of the number of entries that could not be deleted. If we encounter
 516  * an error, return a count of at least one so that the directory stays
 517  * in the unlinked set.
 518  *
 519  * NOTE: this function assumes that the directory is inactive,
 520  *      so there is no need to lock its entries before deletion.
 521  *      Also, it assumes the directory contents is *only* regular
 522  *      files.
 523  */
 524 static int
 525 zfs_purgedir(znode_t *dzp)
 526 {
 527         zap_cursor_t    zc;
 528         zap_attribute_t zap;
 529         znode_t         *xzp;
 530         dmu_tx_t        *tx;
 531         zfsvfs_t        *zfsvfs = dzp->z_zfsvfs;
 532         zfs_dirlock_t   dl;
 533         int skipped = 0;
 534         int error;
 535 
 536         for (zap_cursor_init(&zc, zfsvfs->z_os, dzp->z_id);
 537             (error = zap_cursor_retrieve(&zc, &zap)) == 0;
 538             zap_cursor_advance(&zc)) {
 539                 error = zfs_zget(zfsvfs,
 540                     ZFS_DIRENT_OBJ(zap.za_first_integer), &xzp);
 541                 if (error) {
 542                         skipped += 1;
 543                         continue;
 544                 }
 545 
 546                 ASSERT((ZTOV(xzp)->v_type == VREG) ||
 547                     (ZTOV(xzp)->v_type == VLNK));
 548 
 549                 tx = dmu_tx_create(zfsvfs->z_os);
 550                 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
 551                 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, zap.za_name);
 552                 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
 553                 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
 554                 /* Is this really needed ? */
 555                 zfs_sa_upgrade_txholds(tx, xzp);
 556                 error = dmu_tx_assign(tx, TXG_WAIT);
 557                 if (error) {
 558                         dmu_tx_abort(tx);
 559                         VN_RELE(ZTOV(xzp));
 560                         skipped += 1;
 561                         continue;
 562                 }
 563                 bzero(&dl, sizeof (dl));
 564                 dl.dl_dzp = dzp;
 565                 dl.dl_name = zap.za_name;
 566 
 567                 error = zfs_link_destroy(&dl, xzp, tx, 0, NULL);
 568                 if (error)
 569                         skipped += 1;
 570                 dmu_tx_commit(tx);
 571 
 572                 VN_RELE(ZTOV(xzp));
 573         }
 574         zap_cursor_fini(&zc);
 575         if (error != ENOENT)
 576                 skipped += 1;
 577         return (skipped);
 578 }
 579 
 580 void
 581 zfs_rmnode(znode_t *zp)
 582 {
 583         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
 584         objset_t        *os = zfsvfs->z_os;
 585         znode_t         *xzp = NULL;
 586         dmu_tx_t        *tx;
 587         uint64_t        acl_obj;
 588         uint64_t        xattr_obj;
 589         int             error;
 590 
 591         ASSERT(zp->z_links == 0);
 592         ASSERT(ZTOV(zp)->v_count == 0);
 593 
 594         /*
 595          * If this is an attribute directory, purge its contents.
 596          */
 597         if (ZTOV(zp)->v_type == VDIR && (zp->z_pflags & ZFS_XATTR)) {
 598                 if (zfs_purgedir(zp) != 0) {
 599                         /*
 600                          * Not enough space to delete some xattrs.
 601                          * Leave it in the unlinked set.
 602                          */
 603                         zfs_znode_dmu_fini(zp);
 604                         zfs_znode_free(zp);
 605                         return;
 606                 }
 607         }
 608 
 609         /*
 610          * Free up all the data in the file.
 611          */
 612         error = dmu_free_long_range(os, zp->z_id, 0, DMU_OBJECT_END);
 613         if (error) {
 614                 /*
 615                  * Not enough space.  Leave the file in the unlinked set.
 616                  */
 617                 zfs_znode_dmu_fini(zp);
 618                 zfs_znode_free(zp);
 619                 return;
 620         }
 621 
 622         /*
 623          * If the file has extended attributes, we're going to unlink
 624          * the xattr dir.
 625          */
 626         error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
 627             &xattr_obj, sizeof (xattr_obj));
 628         if (error == 0 && xattr_obj) {
 629                 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
 630                 ASSERT(error == 0);
 631         }
 632 
 633         acl_obj = zfs_external_acl(zp);
 634 
 635         /*
 636          * Set up the final transaction.
 637          */
 638         tx = dmu_tx_create(os);
 639         dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END);
 640         dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
 641         if (xzp) {
 642                 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, TRUE, NULL);
 643                 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
 644         }
 645         if (acl_obj)
 646                 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
 647 
 648         zfs_sa_upgrade_txholds(tx, zp);
 649         error = dmu_tx_assign(tx, TXG_WAIT);
 650         if (error) {
 651                 /*
 652                  * Not enough space to delete the file.  Leave it in the
 653                  * unlinked set, leaking it until the fs is remounted (at
 654                  * which point we'll call zfs_unlinked_drain() to process it).
 655                  */
 656                 dmu_tx_abort(tx);
 657                 zfs_znode_dmu_fini(zp);
 658                 zfs_znode_free(zp);
 659                 goto out;
 660         }
 661 
 662         if (xzp) {
 663                 ASSERT(error == 0);
 664                 mutex_enter(&xzp->z_lock);
 665                 xzp->z_unlinked = B_TRUE;    /* mark xzp for deletion */
 666                 xzp->z_links = 0;    /* no more links to it */
 667                 VERIFY(0 == sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
 668                     &xzp->z_links, sizeof (xzp->z_links), tx));
 669                 mutex_exit(&xzp->z_lock);
 670                 zfs_unlinked_add(xzp, tx);
 671         }
 672 
 673         /* Remove this znode from the unlinked set */
 674         VERIFY3U(0, ==,
 675             zap_remove_int(zfsvfs->z_os, zfsvfs->z_unlinkedobj, zp->z_id, tx));
 676 
 677         zfs_znode_delete(zp, tx);
 678 
 679         dmu_tx_commit(tx);
 680 out:
 681         if (xzp)
 682                 VN_RELE(ZTOV(xzp));
 683 }
 684 
 685 static uint64_t
 686 zfs_dirent(znode_t *zp, uint64_t mode)
 687 {
 688         uint64_t de = zp->z_id;
 689 
 690         if (zp->z_zfsvfs->z_version >= ZPL_VERSION_DIRENT_TYPE)
 691                 de |= IFTODT(mode) << 60;
 692         return (de);
 693 }
 694 
 695 /*
 696  * Link zp into dl.  Can only fail if zp has been unlinked.
 697  */
 698 int
 699 zfs_link_create(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag)
 700 {
 701         znode_t *dzp = dl->dl_dzp;
 702         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
 703         vnode_t *vp = ZTOV(zp);
 704         uint64_t value;
 705         int zp_is_dir = (vp->v_type == VDIR);
 706         sa_bulk_attr_t bulk[5];
 707         uint64_t mtime[2], ctime[2];
 708         int count = 0;
 709         int error;
 710 
 711         mutex_enter(&zp->z_lock);
 712 
 713         if (!(flag & ZRENAMING)) {
 714                 if (zp->z_unlinked) {        /* no new links to unlinked zp */
 715                         ASSERT(!(flag & (ZNEW | ZEXISTS)));
 716                         mutex_exit(&zp->z_lock);
 717                         return (ENOENT);
 718                 }
 719                 zp->z_links++;
 720                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
 721                     &zp->z_links, sizeof (zp->z_links));
 722 
 723         }
 724         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL,
 725             &dzp->z_id, sizeof (dzp->z_id));
 726         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
 727             &zp->z_pflags, sizeof (zp->z_pflags));
 728 
 729         if (!(flag & ZNEW)) {
 730                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
 731                     ctime, sizeof (ctime));
 732                 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime,
 733                     ctime, B_TRUE);
 734         }
 735         error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
 736         ASSERT(error == 0);
 737 
 738         mutex_exit(&zp->z_lock);
 739 
 740         mutex_enter(&dzp->z_lock);
 741         dzp->z_size++;
 742         dzp->z_links += zp_is_dir;
 743         count = 0;
 744         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
 745             &dzp->z_size, sizeof (dzp->z_size));
 746         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
 747             &dzp->z_links, sizeof (dzp->z_links));
 748         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
 749             mtime, sizeof (mtime));
 750         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
 751             ctime, sizeof (ctime));
 752         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
 753             &dzp->z_pflags, sizeof (dzp->z_pflags));
 754         zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
 755         error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx);
 756         ASSERT(error == 0);
 757         mutex_exit(&dzp->z_lock);
 758 
 759         value = zfs_dirent(zp, zp->z_mode);
 760         error = zap_add(zp->z_zfsvfs->z_os, dzp->z_id, dl->dl_name,
 761             8, 1, &value, tx);
 762         ASSERT(error == 0);
 763 
 764         dnlc_update(ZTOV(dzp), dl->dl_name, vp);
 765 
 766         return (0);
 767 }
 768 
 769 static int
 770 zfs_dropname(zfs_dirlock_t *dl, znode_t *zp, znode_t *dzp, dmu_tx_t *tx,
 771     int flag)
 772 {
 773         int error;
 774 
 775         if (zp->z_zfsvfs->z_norm) {
 776                 if (((zp->z_zfsvfs->z_case == ZFS_CASE_INSENSITIVE) &&
 777                     (flag & ZCIEXACT)) ||
 778                     ((zp->z_zfsvfs->z_case == ZFS_CASE_MIXED) &&
 779                     !(flag & ZCILOOK)))
 780                         error = zap_remove_norm(zp->z_zfsvfs->z_os,
 781                             dzp->z_id, dl->dl_name, MT_EXACT, tx);
 782                 else
 783                         error = zap_remove_norm(zp->z_zfsvfs->z_os,
 784                             dzp->z_id, dl->dl_name, MT_FIRST, tx);
 785         } else {
 786                 error = zap_remove(zp->z_zfsvfs->z_os,
 787                     dzp->z_id, dl->dl_name, tx);
 788         }
 789 
 790         return (error);
 791 }
 792 
 793 /*
 794  * Unlink zp from dl, and mark zp for deletion if this was the last link.
 795  * Can fail if zp is a mount point (EBUSY) or a non-empty directory (EEXIST).
 796  * If 'unlinkedp' is NULL, we put unlinked znodes on the unlinked list.
 797  * If it's non-NULL, we use it to indicate whether the znode needs deletion,
 798  * and it's the caller's job to do it.
 799  */
 800 int
 801 zfs_link_destroy(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag,
 802         boolean_t *unlinkedp)
 803 {
 804         znode_t *dzp = dl->dl_dzp;
 805         zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
 806         vnode_t *vp = ZTOV(zp);
 807         int zp_is_dir = (vp->v_type == VDIR);
 808         boolean_t unlinked = B_FALSE;
 809         sa_bulk_attr_t bulk[5];
 810         uint64_t mtime[2], ctime[2];
 811         int count = 0;
 812         int error;
 813 
 814         dnlc_remove(ZTOV(dzp), dl->dl_name);
 815 
 816         if (!(flag & ZRENAMING)) {
 817                 if (vn_vfswlock(vp))            /* prevent new mounts on zp */
 818                         return (EBUSY);
 819 
 820                 if (vn_ismntpt(vp)) {           /* don't remove mount point */
 821                         vn_vfsunlock(vp);
 822                         return (EBUSY);
 823                 }
 824 
 825                 mutex_enter(&zp->z_lock);
 826 
 827                 if (zp_is_dir && !zfs_dirempty(zp)) {
 828                         mutex_exit(&zp->z_lock);
 829                         vn_vfsunlock(vp);
 830                         return (EEXIST);
 831                 }
 832 
 833                 /*
 834                  * If we get here, we are going to try to remove the object.
 835                  * First try removing the name from the directory; if that
 836                  * fails, return the error.
 837                  */
 838                 error = zfs_dropname(dl, zp, dzp, tx, flag);
 839                 if (error != 0) {
 840                         mutex_exit(&zp->z_lock);
 841                         vn_vfsunlock(vp);
 842                         return (error);
 843                 }
 844 
 845                 if (zp->z_links <= zp_is_dir) {
 846                         zfs_panic_recover("zfs: link count on %s is %u, "
 847                             "should be at least %u",
 848                             zp->z_vnode->v_path ? zp->z_vnode->v_path :
 849                             "<unknown>", (int)zp->z_links,
 850                             zp_is_dir + 1);
 851                         zp->z_links = zp_is_dir + 1;
 852                 }
 853                 if (--zp->z_links == zp_is_dir) {
 854                         zp->z_unlinked = B_TRUE;
 855                         zp->z_links = 0;
 856                         unlinked = B_TRUE;
 857                 } else {
 858                         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs),
 859                             NULL, &ctime, sizeof (ctime));
 860                         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
 861                             NULL, &zp->z_pflags, sizeof (zp->z_pflags));
 862                         zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
 863                             B_TRUE);
 864                 }
 865                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs),
 866                     NULL, &zp->z_links, sizeof (zp->z_links));
 867                 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
 868                 count = 0;
 869                 ASSERT(error == 0);
 870                 mutex_exit(&zp->z_lock);
 871                 vn_vfsunlock(vp);
 872         } else {
 873                 error = zfs_dropname(dl, zp, dzp, tx, flag);
 874                 if (error != 0)
 875                         return (error);
 876         }
 877 
 878         mutex_enter(&dzp->z_lock);
 879         dzp->z_size--;               /* one dirent removed */
 880         dzp->z_links -= zp_is_dir;   /* ".." link from zp */
 881         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs),
 882             NULL, &dzp->z_links, sizeof (dzp->z_links));
 883         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
 884             NULL, &dzp->z_size, sizeof (dzp->z_size));
 885         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs),
 886             NULL, ctime, sizeof (ctime));
 887         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
 888             NULL, mtime, sizeof (mtime));
 889         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
 890             NULL, &dzp->z_pflags, sizeof (dzp->z_pflags));
 891         zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
 892         error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx);
 893         ASSERT(error == 0);
 894         mutex_exit(&dzp->z_lock);
 895 
 896         if (unlinkedp != NULL)
 897                 *unlinkedp = unlinked;
 898         else if (unlinked)
 899                 zfs_unlinked_add(zp, tx);
 900 
 901         return (0);
 902 }
 903 
 904 /*
 905  * Indicate whether the directory is empty.  Works with or without z_lock
 906  * held, but can only be consider a hint in the latter case.  Returns true
 907  * if only "." and ".." remain and there's no work in progress.
 908  */
 909 boolean_t
 910 zfs_dirempty(znode_t *dzp)
 911 {
 912         return (dzp->z_size == 2 && dzp->z_dirlocks == 0);
 913 }
 914 
 915 int
 916 zfs_make_xattrdir(znode_t *zp, vattr_t *vap, vnode_t **xvpp, cred_t *cr)
 917 {
 918         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
 919         znode_t *xzp;
 920         dmu_tx_t *tx;
 921         int error;
 922         zfs_acl_ids_t acl_ids;
 923         boolean_t fuid_dirtied;
 924         uint64_t parent;
 925 
 926         *xvpp = NULL;
 927 
 928         if (error = zfs_zaccess(zp, ACE_WRITE_NAMED_ATTRS, 0, B_FALSE, cr))
 929                 return (error);
 930 
 931         if ((error = zfs_acl_ids_create(zp, IS_XATTR, vap, cr, NULL,
 932             &acl_ids)) != 0)
 933                 return (error);
 934         if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
 935                 zfs_acl_ids_free(&acl_ids);
 936                 return (EDQUOT);
 937         }
 938 
 939 top:
 940         tx = dmu_tx_create(zfsvfs->z_os);
 941         dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
 942             ZFS_SA_BASE_ATTR_SIZE);
 943         dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
 944         dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
 945         fuid_dirtied = zfsvfs->z_fuid_dirty;
 946         if (fuid_dirtied)
 947                 zfs_fuid_txhold(zfsvfs, tx);
 948         error = dmu_tx_assign(tx, TXG_NOWAIT);
 949         if (error) {
 950                 if (error == ERESTART) {
 951                         dmu_tx_wait(tx);
 952                         dmu_tx_abort(tx);
 953                         goto top;
 954                 }
 955                 zfs_acl_ids_free(&acl_ids);
 956                 dmu_tx_abort(tx);
 957                 return (error);
 958         }
 959         zfs_mknode(zp, vap, tx, cr, IS_XATTR, &xzp, &acl_ids);
 960 
 961         if (fuid_dirtied)
 962                 zfs_fuid_sync(zfsvfs, tx);
 963 
 964 #ifdef DEBUG
 965         error = sa_lookup(xzp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
 966             &parent, sizeof (parent));
 967         ASSERT(error == 0 && parent == zp->z_id);
 968 #endif
 969 
 970         VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &xzp->z_id,
 971             sizeof (xzp->z_id), tx));
 972 
 973         (void) zfs_log_create(zfsvfs->z_log, tx, TX_MKXATTR, zp,
 974             xzp, "", NULL, acl_ids.z_fuidp, vap);
 975 
 976         zfs_acl_ids_free(&acl_ids);
 977         dmu_tx_commit(tx);
 978 
 979         *xvpp = ZTOV(xzp);
 980 
 981         return (0);
 982 }
 983 
 984 /*
 985  * Return a znode for the extended attribute directory for zp.
 986  * ** If the directory does not already exist, it is created **
 987  *
 988  *      IN:     zp      - znode to obtain attribute directory from
 989  *              cr      - credentials of caller
 990  *              flags   - flags from the VOP_LOOKUP call
 991  *
 992  *      OUT:    xzpp    - pointer to extended attribute znode
 993  *
 994  *      RETURN: 0 on success
 995  *              error number on failure
 996  */
 997 int
 998 zfs_get_xattrdir(znode_t *zp, vnode_t **xvpp, cred_t *cr, int flags)
 999 {
1000         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
1001         znode_t         *xzp;
1002         zfs_dirlock_t   *dl;
1003         vattr_t         va;
1004         int             error;
1005 top:
1006         error = zfs_dirent_lock(&dl, zp, "", &xzp, ZXATTR, NULL, NULL);
1007         if (error)
1008                 return (error);
1009 
1010         if (xzp != NULL) {
1011                 *xvpp = ZTOV(xzp);
1012                 zfs_dirent_unlock(dl);
1013                 return (0);
1014         }
1015 
1016 
1017         if (!(flags & CREATE_XATTR_DIR)) {
1018                 zfs_dirent_unlock(dl);
1019                 return (ENOENT);
1020         }
1021 
1022         if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
1023                 zfs_dirent_unlock(dl);
1024                 return (EROFS);
1025         }
1026 
1027         /*
1028          * The ability to 'create' files in an attribute
1029          * directory comes from the write_xattr permission on the base file.
1030          *
1031          * The ability to 'search' an attribute directory requires
1032          * read_xattr permission on the base file.
1033          *
1034          * Once in a directory the ability to read/write attributes
1035          * is controlled by the permissions on the attribute file.
1036          */
1037         va.va_mask = AT_TYPE | AT_MODE | AT_UID | AT_GID;
1038         va.va_type = VDIR;
1039         va.va_mode = S_IFDIR | S_ISVTX | 0777;
1040         zfs_fuid_map_ids(zp, cr, &va.va_uid, &va.va_gid);
1041 
1042         error = zfs_make_xattrdir(zp, &va, xvpp, cr);
1043         zfs_dirent_unlock(dl);
1044 
1045         if (error == ERESTART) {
1046                 /* NB: we already did dmu_tx_wait() if necessary */
1047                 goto top;
1048         }
1049 
1050         return (error);
1051 }
1052 
1053 /*
1054  * Decide whether it is okay to remove within a sticky directory.
1055  *
1056  * In sticky directories, write access is not sufficient;
1057  * you can remove entries from a directory only if:
1058  *
1059  *      you own the directory,
1060  *      you own the entry,
1061  *      the entry is a plain file and you have write access,
1062  *      or you are privileged (checked in secpolicy...).
1063  *
1064  * The function returns 0 if remove access is granted.
1065  */
1066 int
1067 zfs_sticky_remove_access(znode_t *zdp, znode_t *zp, cred_t *cr)
1068 {
1069         uid_t           uid;
1070         uid_t           downer;
1071         uid_t           fowner;
1072         zfsvfs_t        *zfsvfs = zdp->z_zfsvfs;
1073 
1074         if (zdp->z_zfsvfs->z_replay)
1075                 return (0);
1076 
1077         if ((zdp->z_mode & S_ISVTX) == 0)
1078                 return (0);
1079 
1080         downer = zfs_fuid_map_id(zfsvfs, zdp->z_uid, cr, ZFS_OWNER);
1081         fowner = zfs_fuid_map_id(zfsvfs, zp->z_uid, cr, ZFS_OWNER);
1082 
1083         if ((uid = crgetuid(cr)) == downer || uid == fowner ||
1084             (ZTOV(zp)->v_type == VREG &&
1085             zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr) == 0))
1086                 return (0);
1087         else
1088                 return (secpolicy_vnode_remove(cr));
1089 }