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  * Copyright (c) 2015, Joyent, Inc.
  27  */
  28 
  29 #include <sys/types.h>
  30 #include <sys/param.h>
  31 #include <sys/time.h>
  32 #include <sys/systm.h>
  33 #include <sys/sysmacros.h>
  34 #include <sys/resource.h>
  35 #include <sys/vfs.h>
  36 #include <sys/vnode.h>
  37 #include <sys/file.h>
  38 #include <sys/mode.h>
  39 #include <sys/kmem.h>
  40 #include <sys/uio.h>
  41 #include <sys/pathname.h>
  42 #include <sys/cmn_err.h>
  43 #include <sys/errno.h>
  44 #include <sys/stat.h>
  45 #include <sys/unistd.h>
  46 #include <sys/sunddi.h>
  47 #include <sys/random.h>
  48 #include <sys/policy.h>
  49 #include <sys/zfs_dir.h>
  50 #include <sys/zfs_acl.h>
  51 #include <sys/fs/zfs.h>
  52 #include "fs/fs_subr.h"
  53 #include <sys/zap.h>
  54 #include <sys/dmu.h>
  55 #include <sys/dmu_objset.h>
  56 #include <sys/atomic.h>
  57 #include <sys/zfs_ctldir.h>
  58 #include <sys/zfs_fuid.h>
  59 #include <sys/sa.h>
  60 #include <sys/zfs_sa.h>
  61 #include <sys/dnlc.h>
  62 #include <sys/extdirent.h>
  63 
  64 /*
  65  * zfs_match_find() is used by zfs_dirent_lock() to peform zap lookups
  66  * of names after deciding which is the appropriate lookup interface.
  67  */
  68 static int
  69 zfs_match_find(zfsvfs_t *zfsvfs, znode_t *dzp, char *name, matchtype_t mt,
  70     boolean_t update, int *deflags, pathname_t *rpnp, uint64_t *zoid)
  71 {
  72         int error;
  73 
  74         if (zfsvfs->z_norm) {
  75                 boolean_t conflict = B_FALSE;
  76                 size_t bufsz = 0;
  77                 char *buf = NULL;
  78 
  79                 if (rpnp) {
  80                         buf = rpnp->pn_buf;
  81                         bufsz = rpnp->pn_bufsize;
  82                 }
  83 
  84                 /*
  85                  * In the non-mixed case we only expect there would ever
  86                  * be one match, but we need to use the normalizing lookup.
  87                  */
  88                 error = zap_lookup_norm(zfsvfs->z_os, dzp->z_id, name, 8, 1,
  89                     zoid, mt, buf, bufsz, &conflict);
  90                 if (!error && deflags)
  91                         *deflags = conflict ? ED_CASE_CONFLICT : 0;
  92         } else {
  93                 error = zap_lookup(zfsvfs->z_os, dzp->z_id, name, 8, 1, zoid);
  94         }
  95         *zoid = ZFS_DIRENT_OBJ(*zoid);
  96 
  97         if (error == ENOENT && update)
  98                 dnlc_update(ZTOV(dzp), name, DNLC_NO_VNODE);
  99 
 100         return (error);
 101 }
 102 
 103 /*
 104  * Lock a directory entry.  A dirlock on <dzp, name> protects that name
 105  * in dzp's directory zap object.  As long as you hold a dirlock, you can
 106  * assume two things: (1) dzp cannot be reaped, and (2) no other thread
 107  * can change the zap entry for (i.e. link or unlink) this name.
 108  *
 109  * Input arguments:
 110  *      dzp     - znode for directory
 111  *      name    - name of entry to lock
 112  *      flag    - ZNEW: if the entry already exists, fail with EEXIST.
 113  *                ZEXISTS: if the entry does not exist, fail with ENOENT.
 114  *                ZSHARED: allow concurrent access with other ZSHARED callers.
 115  *                ZXATTR: we want dzp's xattr directory
 116  *                ZCILOOK: On a mixed sensitivity file system,
 117  *                         this lookup should be case-insensitive.
 118  *                ZCIEXACT: On a purely case-insensitive file system,
 119  *                          this lookup should be case-sensitive.
 120  *                ZRENAMING: we are locking for renaming, force narrow locks
 121  *                ZHAVELOCK: Don't grab the z_name_lock for this call. The
 122  *                           current thread already holds it.
 123  *
 124  * Output arguments:
 125  *      zpp     - pointer to the znode for the entry (NULL if there isn't one)
 126  *      dlpp    - pointer to the dirlock for this entry (NULL on error)
 127  *      direntflags - (case-insensitive lookup only)
 128  *              flags if multiple case-sensitive matches exist in directory
 129  *      realpnp     - (case-insensitive lookup only)
 130  *              actual name matched within the directory
 131  *
 132  * Return value: 0 on success or errno on failure.
 133  *
 134  * NOTE: Always checks for, and rejects, '.' and '..'.
 135  * NOTE: For case-insensitive file systems we take wide locks (see below),
 136  *       but return znode pointers to a single match.
 137  */
 138 int
 139 zfs_dirent_lock(zfs_dirlock_t **dlpp, znode_t *dzp, char *name, znode_t **zpp,
 140     int flag, int *direntflags, pathname_t *realpnp)
 141 {
 142         zfsvfs_t        *zfsvfs = dzp->z_zfsvfs;
 143         zfs_dirlock_t   *dl;
 144         boolean_t       update;
 145         matchtype_t     mt = 0;
 146         uint64_t        zoid;
 147         vnode_t         *vp = NULL;
 148         int             error = 0;
 149         int             cmpflags;
 150 
 151         *zpp = NULL;
 152         *dlpp = NULL;
 153 
 154         /*
 155          * Verify that we are not trying to lock '.', '..', or '.zfs'
 156          */
 157         if (name[0] == '.' &&
 158             (name[1] == '\0' || (name[1] == '.' && name[2] == '\0')) ||
 159             zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0)
 160                 return (SET_ERROR(EEXIST));
 161 
 162         /*
 163          * Case sensitivity and normalization preferences are set when
 164          * the file system is created.  These are stored in the
 165          * zfsvfs->z_case and zfsvfs->z_norm fields.  These choices
 166          * affect what vnodes can be cached in the DNLC, how we
 167          * perform zap lookups, and the "width" of our dirlocks.
 168          *
 169          * A normal dirlock locks a single name.  Note that with
 170          * normalization a name can be composed multiple ways, but
 171          * when normalized, these names all compare equal.  A wide
 172          * dirlock locks multiple names.  We need these when the file
 173          * system is supporting mixed-mode access.  It is sometimes
 174          * necessary to lock all case permutations of file name at
 175          * once so that simultaneous case-insensitive/case-sensitive
 176          * behaves as rationally as possible.
 177          */
 178 
 179         /*
 180          * When matching we may need to normalize & change case according to
 181          * FS settings.
 182          *
 183          * Note that a normalized match is necessary for a case insensitive
 184          * filesystem when the lookup request is not exact because normalization
 185          * can fold case independent of normalizing code point sequences.
 186          *
 187          * See the table above zfs_dropname().
 188          */
 189         if (zfsvfs->z_norm != 0) {
 190                 mt = MT_NORMALIZE;
 191 
 192                 /*
 193                  * Determine if the match needs to honor the case specified in
 194                  * lookup, and if so keep track of that so that during
 195                  * normalization we don't fold case.
 196                  */
 197                 if ((zfsvfs->z_case == ZFS_CASE_INSENSITIVE &&
 198                     (flag & ZCIEXACT)) ||
 199                     (zfsvfs->z_case == ZFS_CASE_MIXED && !(flag & ZCILOOK))) {
 200                         mt |= MT_MATCH_CASE;
 201                 }
 202         }
 203 
 204         /*
 205          * Only look in or update the DNLC if we are looking for the
 206          * name on a file system that does not require normalization
 207          * or case folding.  We can also look there if we happen to be
 208          * on a non-normalizing, mixed sensitivity file system IF we
 209          * are looking for the exact name.
 210          *
 211          * Maybe can add TO-UPPERed version of name to dnlc in ci-only
 212          * case for performance improvement?
 213          */
 214         update = !zfsvfs->z_norm ||
 215             (zfsvfs->z_case == ZFS_CASE_MIXED &&
 216             !(zfsvfs->z_norm & ~U8_TEXTPREP_TOUPPER) && !(flag & ZCILOOK));
 217 
 218         /*
 219          * ZRENAMING indicates we are in a situation where we should
 220          * take narrow locks regardless of the file system's
 221          * preferences for normalizing and case folding.  This will
 222          * prevent us deadlocking trying to grab the same wide lock
 223          * twice if the two names happen to be case-insensitive
 224          * matches.
 225          */
 226         if (flag & ZRENAMING)
 227                 cmpflags = 0;
 228         else
 229                 cmpflags = zfsvfs->z_norm;
 230 
 231         /*
 232          * Wait until there are no locks on this name.
 233          *
 234          * Don't grab the the lock if it is already held. However, cannot
 235          * have both ZSHARED and ZHAVELOCK together.
 236          */
 237         ASSERT(!(flag & ZSHARED) || !(flag & ZHAVELOCK));
 238         if (!(flag & ZHAVELOCK))
 239                 rw_enter(&dzp->z_name_lock, RW_READER);
 240 
 241         mutex_enter(&dzp->z_lock);
 242         for (;;) {
 243                 if (dzp->z_unlinked) {
 244                         mutex_exit(&dzp->z_lock);
 245                         if (!(flag & ZHAVELOCK))
 246                                 rw_exit(&dzp->z_name_lock);
 247                         return (SET_ERROR(ENOENT));
 248                 }
 249                 for (dl = dzp->z_dirlocks; dl != NULL; dl = dl->dl_next) {
 250                         if ((u8_strcmp(name, dl->dl_name, 0, cmpflags,
 251                             U8_UNICODE_LATEST, &error) == 0) || error != 0)
 252                                 break;
 253                 }
 254                 if (error != 0) {
 255                         mutex_exit(&dzp->z_lock);
 256                         if (!(flag & ZHAVELOCK))
 257                                 rw_exit(&dzp->z_name_lock);
 258                         return (SET_ERROR(ENOENT));
 259                 }
 260                 if (dl == NULL) {
 261                         /*
 262                          * Allocate a new dirlock and add it to the list.
 263                          */
 264                         dl = kmem_alloc(sizeof (zfs_dirlock_t), KM_SLEEP);
 265                         cv_init(&dl->dl_cv, NULL, CV_DEFAULT, NULL);
 266                         dl->dl_name = name;
 267                         dl->dl_sharecnt = 0;
 268                         dl->dl_namelock = 0;
 269                         dl->dl_namesize = 0;
 270                         dl->dl_dzp = dzp;
 271                         dl->dl_next = dzp->z_dirlocks;
 272                         dzp->z_dirlocks = dl;
 273                         break;
 274                 }
 275                 if ((flag & ZSHARED) && dl->dl_sharecnt != 0)
 276                         break;
 277                 cv_wait(&dl->dl_cv, &dzp->z_lock);
 278         }
 279 
 280         /*
 281          * If the z_name_lock was NOT held for this dirlock record it.
 282          */
 283         if (flag & ZHAVELOCK)
 284                 dl->dl_namelock = 1;
 285 
 286         if ((flag & ZSHARED) && ++dl->dl_sharecnt > 1 && dl->dl_namesize == 0) {
 287                 /*
 288                  * We're the second shared reference to dl.  Make a copy of
 289                  * dl_name in case the first thread goes away before we do.
 290                  * Note that we initialize the new name before storing its
 291                  * pointer into dl_name, because the first thread may load
 292                  * dl->dl_name at any time.  It'll either see the old value,
 293                  * which belongs to it, or the new shared copy; either is OK.
 294                  */
 295                 dl->dl_namesize = strlen(dl->dl_name) + 1;
 296                 name = kmem_alloc(dl->dl_namesize, KM_SLEEP);
 297                 bcopy(dl->dl_name, name, dl->dl_namesize);
 298                 dl->dl_name = name;
 299         }
 300 
 301         mutex_exit(&dzp->z_lock);
 302 
 303         /*
 304          * We have a dirlock on the name.  (Note that it is the dirlock,
 305          * not the dzp's z_lock, that protects the name in the zap object.)
 306          * See if there's an object by this name; if so, put a hold on it.
 307          */
 308         if (flag & ZXATTR) {
 309                 error = sa_lookup(dzp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &zoid,
 310                     sizeof (zoid));
 311                 if (error == 0)
 312                         error = (zoid == 0 ? ENOENT : 0);
 313         } else {
 314                 if (update)
 315                         vp = dnlc_lookup(ZTOV(dzp), name);
 316                 if (vp == DNLC_NO_VNODE) {
 317                         VN_RELE(vp);
 318                         error = SET_ERROR(ENOENT);
 319                 } else if (vp) {
 320                         if (flag & ZNEW) {
 321                                 zfs_dirent_unlock(dl);
 322                                 VN_RELE(vp);
 323                                 return (SET_ERROR(EEXIST));
 324                         }
 325                         *dlpp = dl;
 326                         *zpp = VTOZ(vp);
 327                         return (0);
 328                 } else {
 329                         error = zfs_match_find(zfsvfs, dzp, name, mt,
 330                             update, direntflags, realpnp, &zoid);
 331                 }
 332         }
 333         if (error) {
 334                 if (error != ENOENT || (flag & ZEXISTS)) {
 335                         zfs_dirent_unlock(dl);
 336                         return (error);
 337                 }
 338         } else {
 339                 if (flag & ZNEW) {
 340                         zfs_dirent_unlock(dl);
 341                         return (SET_ERROR(EEXIST));
 342                 }
 343                 error = zfs_zget(zfsvfs, zoid, zpp);
 344                 if (error) {
 345                         zfs_dirent_unlock(dl);
 346                         return (error);
 347                 }
 348                 if (!(flag & ZXATTR) && update)
 349                         dnlc_update(ZTOV(dzp), name, ZTOV(*zpp));
 350         }
 351 
 352         *dlpp = dl;
 353 
 354         return (0);
 355 }
 356 
 357 /*
 358  * Unlock this directory entry and wake anyone who was waiting for it.
 359  */
 360 void
 361 zfs_dirent_unlock(zfs_dirlock_t *dl)
 362 {
 363         znode_t *dzp = dl->dl_dzp;
 364         zfs_dirlock_t **prev_dl, *cur_dl;
 365 
 366         mutex_enter(&dzp->z_lock);
 367 
 368         if (!dl->dl_namelock)
 369                 rw_exit(&dzp->z_name_lock);
 370 
 371         if (dl->dl_sharecnt > 1) {
 372                 dl->dl_sharecnt--;
 373                 mutex_exit(&dzp->z_lock);
 374                 return;
 375         }
 376         prev_dl = &dzp->z_dirlocks;
 377         while ((cur_dl = *prev_dl) != dl)
 378                 prev_dl = &cur_dl->dl_next;
 379         *prev_dl = dl->dl_next;
 380         cv_broadcast(&dl->dl_cv);
 381         mutex_exit(&dzp->z_lock);
 382 
 383         if (dl->dl_namesize != 0)
 384                 kmem_free(dl->dl_name, dl->dl_namesize);
 385         cv_destroy(&dl->dl_cv);
 386         kmem_free(dl, sizeof (*dl));
 387 }
 388 
 389 /*
 390  * Look up an entry in a directory.
 391  *
 392  * NOTE: '.' and '..' are handled as special cases because
 393  *      no directory entries are actually stored for them.  If this is
 394  *      the root of a filesystem, then '.zfs' is also treated as a
 395  *      special pseudo-directory.
 396  */
 397 int
 398 zfs_dirlook(znode_t *dzp, char *name, vnode_t **vpp, int flags,
 399     int *deflg, pathname_t *rpnp)
 400 {
 401         zfs_dirlock_t *dl;
 402         znode_t *zp;
 403         int error = 0;
 404         uint64_t parent;
 405 
 406         if (name[0] == 0 || (name[0] == '.' && name[1] == 0)) {
 407                 *vpp = ZTOV(dzp);
 408                 VN_HOLD(*vpp);
 409         } else if (name[0] == '.' && name[1] == '.' && name[2] == 0) {
 410                 zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
 411 
 412                 /*
 413                  * If we are a snapshot mounted under .zfs, return
 414                  * the vp for the snapshot directory.
 415                  */
 416                 if ((error = sa_lookup(dzp->z_sa_hdl,
 417                     SA_ZPL_PARENT(zfsvfs), &parent, sizeof (parent))) != 0)
 418                         return (error);
 419                 if (parent == dzp->z_id && zfsvfs->z_parent != zfsvfs) {
 420                         error = zfsctl_root_lookup(zfsvfs->z_parent->z_ctldir,
 421                             "snapshot", vpp, NULL, 0, NULL, kcred,
 422                             NULL, NULL, NULL);
 423                         return (error);
 424                 }
 425                 rw_enter(&dzp->z_parent_lock, RW_READER);
 426                 error = zfs_zget(zfsvfs, parent, &zp);
 427                 if (error == 0)
 428                         *vpp = ZTOV(zp);
 429                 rw_exit(&dzp->z_parent_lock);
 430         } else if (zfs_has_ctldir(dzp) && strcmp(name, ZFS_CTLDIR_NAME) == 0) {
 431                 *vpp = zfsctl_root(dzp);
 432         } else {
 433                 int zf;
 434 
 435                 zf = ZEXISTS | ZSHARED;
 436                 if (flags & FIGNORECASE)
 437                         zf |= ZCILOOK;
 438 
 439                 error = zfs_dirent_lock(&dl, dzp, name, &zp, zf, deflg, rpnp);
 440                 if (error == 0) {
 441                         *vpp = ZTOV(zp);
 442                         zfs_dirent_unlock(dl);
 443                         dzp->z_zn_prefetch = B_TRUE; /* enable prefetching */
 444                 }
 445                 rpnp = NULL;
 446         }
 447 
 448         if ((flags & FIGNORECASE) && rpnp && !error)
 449                 (void) strlcpy(rpnp->pn_buf, name, rpnp->pn_bufsize);
 450 
 451         return (error);
 452 }
 453 
 454 /*
 455  * unlinked Set (formerly known as the "delete queue") Error Handling
 456  *
 457  * When dealing with the unlinked set, we dmu_tx_hold_zap(), but we
 458  * don't specify the name of the entry that we will be manipulating.  We
 459  * also fib and say that we won't be adding any new entries to the
 460  * unlinked set, even though we might (this is to lower the minimum file
 461  * size that can be deleted in a full filesystem).  So on the small
 462  * chance that the nlink list is using a fat zap (ie. has more than
 463  * 2000 entries), we *may* not pre-read a block that's needed.
 464  * Therefore it is remotely possible for some of the assertions
 465  * regarding the unlinked set below to fail due to i/o error.  On a
 466  * nondebug system, this will result in the space being leaked.
 467  */
 468 void
 469 zfs_unlinked_add(znode_t *zp, dmu_tx_t *tx)
 470 {
 471         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
 472 
 473         ASSERT(zp->z_unlinked);
 474         ASSERT(zp->z_links == 0);
 475 
 476         VERIFY3U(0, ==,
 477             zap_add_int(zfsvfs->z_os, zfsvfs->z_unlinkedobj, zp->z_id, tx));
 478 }
 479 
 480 /*
 481  * Clean up any znodes that had no links when we either crashed or
 482  * (force) umounted the file system.
 483  */
 484 static void
 485 zfs_unlinked_drain_impl(zfsvfs_t *zfsvfs)
 486 {
 487         zap_cursor_t    zc;
 488         zap_attribute_t zap;
 489         dmu_object_info_t doi;
 490         znode_t         *zp;
 491         int             error;
 492 
 493         ASSERT(zfsvfs->z_drain_state != ZFS_DRAIN_SHUTDOWN);
 494         /*
 495          * Interate over the contents of the unlinked set.
 496          */
 497         for (zap_cursor_init(&zc, zfsvfs->z_os, zfsvfs->z_unlinkedobj);
 498             zap_cursor_retrieve(&zc, &zap) == 0 &&
 499             /* Only checking for a shutdown request, so no locking reqd. */
 500             zfsvfs->z_drain_state == ZFS_DRAIN_RUNNING;
 501             zap_cursor_advance(&zc)) {
 502 
 503                 /*
 504                  * See what kind of object we have in list
 505                  */
 506 
 507                 error = dmu_object_info(zfsvfs->z_os,
 508                     zap.za_first_integer, &doi);
 509                 if (error != 0)
 510                         continue;
 511 
 512                 ASSERT((doi.doi_type == DMU_OT_PLAIN_FILE_CONTENTS) ||
 513                     (doi.doi_type == DMU_OT_DIRECTORY_CONTENTS));
 514                 /*
 515                  * We need to re-mark these list entries for deletion,
 516                  * so we pull them back into core and set zp->z_unlinked.
 517                  */
 518                 error = zfs_zget(zfsvfs, zap.za_first_integer, &zp);
 519 
 520                 /*
 521                  * We may pick up znodes that are already marked for deletion.
 522                  * This could happen during the purge of an extended attribute
 523                  * directory.  All we need to do is skip over them, since they
 524                  * are already in the system marked z_unlinked.
 525                  */
 526                 if (error != 0)
 527                         continue;
 528 
 529                 zp->z_unlinked = B_TRUE;
 530                 VN_RELE(ZTOV(zp));
 531 
 532                 ASSERT(!zfsvfs->z_unmounted);
 533         }
 534         zap_cursor_fini(&zc);
 535 
 536         mutex_enter(&zfsvfs->z_drain_lock);
 537         zfsvfs->z_drain_state = ZFS_DRAIN_SHUTDOWN;
 538         cv_broadcast(&zfsvfs->z_drain_cv);
 539         mutex_exit(&zfsvfs->z_drain_lock);
 540 }
 541 
 542 /*
 543  * Setup required hold. After that tries to dispatch
 544  * async unlinked drain logic. Otherwise executes
 545  * the logic synchronously.
 546  */
 547 void
 548 zfs_unlinked_drain(zfsvfs_t *zfsvfs)
 549 {
 550         ASSERT(!zfsvfs->z_unmounted);
 551 
 552         mutex_enter(&zfsvfs->z_drain_lock);
 553         ASSERT(zfsvfs->z_drain_state == ZFS_DRAIN_SHUTDOWN);
 554         zfsvfs->z_drain_state = ZFS_DRAIN_RUNNING;
 555         mutex_exit(&zfsvfs->z_drain_lock);
 556 
 557         if (taskq_dispatch(dsl_pool_vnrele_taskq(
 558             spa_get_dsl(zfsvfs->z_os->os_spa)),
 559             (void (*)(void *))zfs_unlinked_drain_impl, zfsvfs,
 560             TQ_NOSLEEP) == NULL) {
 561                 cmn_err(CE_WARN, "async zfs_unlinked_drain dispatch failed");
 562                 zfs_unlinked_drain_impl(zfsvfs);
 563         }
 564 }
 565 
 566 /*
 567  * Stops an asynchronous zfs_unlinked_drain. This must be called prior to
 568  * destroying the zfsvfs_t, as the drain doesn't hold the z_teardown_lock.
 569  */
 570 void
 571 zfs_unlinked_drain_stop_wait(zfsvfs_t *zfsvfs)
 572 {
 573         ASSERT(!zfsvfs->z_unmounted);
 574 
 575         mutex_enter(&zfsvfs->z_drain_lock);
 576         while (zfsvfs->z_drain_state != ZFS_DRAIN_SHUTDOWN) {
 577                 zfsvfs->z_drain_state = ZFS_DRAIN_SHUTDOWN_REQ;
 578                 cv_wait(&zfsvfs->z_drain_cv, &zfsvfs->z_drain_lock);
 579         }
 580         mutex_exit(&zfsvfs->z_drain_lock);
 581 }
 582 
 583 /*
 584  * Delete the entire contents of a directory.  Return a count
 585  * of the number of entries that could not be deleted. If we encounter
 586  * an error, return a count of at least one so that the directory stays
 587  * in the unlinked set.
 588  *
 589  * NOTE: this function assumes that the directory is inactive,
 590  *      so there is no need to lock its entries before deletion.
 591  *      Also, it assumes the directory contents is *only* regular
 592  *      files.
 593  */
 594 static int
 595 zfs_purgedir(znode_t *dzp)
 596 {
 597         zap_cursor_t    zc;
 598         zap_attribute_t zap;
 599         znode_t         *xzp;
 600         dmu_tx_t        *tx;
 601         zfsvfs_t        *zfsvfs = dzp->z_zfsvfs;
 602         zfs_dirlock_t   dl;
 603         int skipped = 0;
 604         int error;
 605 
 606         for (zap_cursor_init(&zc, zfsvfs->z_os, dzp->z_id);
 607             (error = zap_cursor_retrieve(&zc, &zap)) == 0;
 608             zap_cursor_advance(&zc)) {
 609                 error = zfs_zget(zfsvfs,
 610                     ZFS_DIRENT_OBJ(zap.za_first_integer), &xzp);
 611                 if (error) {
 612                         skipped += 1;
 613                         continue;
 614                 }
 615 
 616                 ASSERT((ZTOV(xzp)->v_type == VREG) ||
 617                     (ZTOV(xzp)->v_type == VLNK));
 618 
 619                 tx = dmu_tx_create(zfsvfs->z_os);
 620                 dmu_tx_hold_sa(tx, dzp->z_sa_hdl, B_FALSE);
 621                 dmu_tx_hold_zap(tx, dzp->z_id, FALSE, zap.za_name);
 622                 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
 623                 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
 624                 /* Is this really needed ? */
 625                 zfs_sa_upgrade_txholds(tx, xzp);
 626                 dmu_tx_mark_netfree(tx);
 627                 error = dmu_tx_assign(tx, TXG_WAIT);
 628                 if (error) {
 629                         dmu_tx_abort(tx);
 630                         VN_RELE(ZTOV(xzp));
 631                         skipped += 1;
 632                         continue;
 633                 }
 634                 bzero(&dl, sizeof (dl));
 635                 dl.dl_dzp = dzp;
 636                 dl.dl_name = zap.za_name;
 637 
 638                 error = zfs_link_destroy(&dl, xzp, tx, 0, NULL);
 639                 if (error)
 640                         skipped += 1;
 641                 dmu_tx_commit(tx);
 642 
 643                 VN_RELE(ZTOV(xzp));
 644         }
 645         zap_cursor_fini(&zc);
 646         if (error != ENOENT)
 647                 skipped += 1;
 648         return (skipped);
 649 }
 650 
 651 void
 652 zfs_rmnode(znode_t *zp)
 653 {
 654         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
 655         objset_t        *os = zfsvfs->z_os;
 656         znode_t         *xzp = NULL;
 657         dmu_tx_t        *tx;
 658         uint64_t        acl_obj;
 659         uint64_t        xattr_obj;
 660         int             error;
 661 
 662         ASSERT(zp->z_links == 0);
 663         ASSERT(ZTOV(zp)->v_count == 0);
 664 
 665         /*
 666          * If this is an attribute directory, purge its contents.
 667          */
 668         if (ZTOV(zp)->v_type == VDIR && (zp->z_pflags & ZFS_XATTR)) {
 669                 if (zfs_purgedir(zp) != 0) {
 670                         /*
 671                          * Not enough space to delete some xattrs.
 672                          * Leave it in the unlinked set.
 673                          */
 674                         zfs_znode_dmu_fini(zp);
 675                         zfs_znode_free(zp);
 676                         return;
 677                 }
 678         } else {
 679                 /*
 680                  * Free up all the data in the file.  We don't do this for
 681                  * XATTR directories because we need truncate and remove to be
 682                  * in the same tx, like in zfs_znode_delete(). Otherwise, if
 683                  * we crash here we'll end up with an inconsistent truncated
 684                  * zap object in the delete queue.  Note a truncated file is
 685                  * harmless since it only contains user data.
 686                  */
 687                 error = dmu_free_long_range(os, zp->z_id, 0, DMU_OBJECT_END);
 688                 if (error) {
 689                         /*
 690                          * Not enough space or we were interrupted by unmount.
 691                          * Leave the file in the unlinked set.
 692                          */
 693                         zfs_znode_dmu_fini(zp);
 694                         zfs_znode_free(zp);
 695                         return;
 696                 }
 697         }
 698 
 699         /*
 700          * If the file has extended attributes, we're going to unlink
 701          * the xattr dir.
 702          */
 703         error = sa_lookup(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs),
 704             &xattr_obj, sizeof (xattr_obj));
 705         if (error == 0 && xattr_obj) {
 706                 error = zfs_zget(zfsvfs, xattr_obj, &xzp);
 707                 ASSERT(error == 0);
 708         }
 709 
 710         acl_obj = zfs_external_acl(zp);
 711 
 712         /*
 713          * Set up the final transaction.
 714          */
 715         tx = dmu_tx_create(os);
 716         dmu_tx_hold_free(tx, zp->z_id, 0, DMU_OBJECT_END);
 717         dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, FALSE, NULL);
 718         if (xzp) {
 719                 dmu_tx_hold_zap(tx, zfsvfs->z_unlinkedobj, TRUE, NULL);
 720                 dmu_tx_hold_sa(tx, xzp->z_sa_hdl, B_FALSE);
 721         }
 722         if (acl_obj)
 723                 dmu_tx_hold_free(tx, acl_obj, 0, DMU_OBJECT_END);
 724 
 725         zfs_sa_upgrade_txholds(tx, zp);
 726         error = dmu_tx_assign(tx, TXG_WAIT);
 727         if (error) {
 728                 /*
 729                  * Not enough space to delete the file.  Leave it in the
 730                  * unlinked set, leaking it until the fs is remounted (at
 731                  * which point we'll call zfs_unlinked_drain() to process it).
 732                  */
 733                 dmu_tx_abort(tx);
 734                 zfs_znode_dmu_fini(zp);
 735                 zfs_znode_free(zp);
 736                 goto out;
 737         }
 738 
 739         if (xzp) {
 740                 ASSERT(error == 0);
 741                 mutex_enter(&xzp->z_lock);
 742                 xzp->z_unlinked = B_TRUE;    /* mark xzp for deletion */
 743                 xzp->z_links = 0;    /* no more links to it */
 744                 VERIFY(0 == sa_update(xzp->z_sa_hdl, SA_ZPL_LINKS(zfsvfs),
 745                     &xzp->z_links, sizeof (xzp->z_links), tx));
 746                 mutex_exit(&xzp->z_lock);
 747                 zfs_unlinked_add(xzp, tx);
 748         }
 749 
 750         /* Remove this znode from the unlinked set */
 751         VERIFY3U(0, ==,
 752             zap_remove_int(zfsvfs->z_os, zfsvfs->z_unlinkedobj, zp->z_id, tx));
 753 
 754         zfs_znode_delete(zp, tx);
 755 
 756         dmu_tx_commit(tx);
 757 out:
 758         if (xzp)
 759                 VN_RELE(ZTOV(xzp));
 760 }
 761 
 762 static uint64_t
 763 zfs_dirent(znode_t *zp, uint64_t mode)
 764 {
 765         uint64_t de = zp->z_id;
 766 
 767         if (zp->z_zfsvfs->z_version >= ZPL_VERSION_DIRENT_TYPE)
 768                 de |= IFTODT(mode) << 60;
 769         return (de);
 770 }
 771 
 772 /*
 773  * Link zp into dl.  Can only fail if zp has been unlinked.
 774  */
 775 int
 776 zfs_link_create(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag)
 777 {
 778         znode_t *dzp = dl->dl_dzp;
 779         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
 780         vnode_t *vp = ZTOV(zp);
 781         uint64_t value;
 782         int zp_is_dir = (vp->v_type == VDIR);
 783         sa_bulk_attr_t bulk[5];
 784         uint64_t mtime[2], ctime[2];
 785         int count = 0;
 786         int error;
 787 
 788         mutex_enter(&zp->z_lock);
 789 
 790         if (!(flag & ZRENAMING)) {
 791                 if (zp->z_unlinked) {        /* no new links to unlinked zp */
 792                         ASSERT(!(flag & (ZNEW | ZEXISTS)));
 793                         mutex_exit(&zp->z_lock);
 794                         return (SET_ERROR(ENOENT));
 795                 }
 796                 zp->z_links++;
 797                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
 798                     &zp->z_links, sizeof (zp->z_links));
 799 
 800         }
 801         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_PARENT(zfsvfs), NULL,
 802             &dzp->z_id, sizeof (dzp->z_id));
 803         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
 804             &zp->z_pflags, sizeof (zp->z_pflags));
 805 
 806         if (!(flag & ZNEW)) {
 807                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
 808                     ctime, sizeof (ctime));
 809                 zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime,
 810                     ctime, B_TRUE);
 811         }
 812         error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
 813         ASSERT(error == 0);
 814 
 815         mutex_exit(&zp->z_lock);
 816 
 817         mutex_enter(&dzp->z_lock);
 818         dzp->z_size++;
 819         dzp->z_links += zp_is_dir;
 820         count = 0;
 821         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs), NULL,
 822             &dzp->z_size, sizeof (dzp->z_size));
 823         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs), NULL,
 824             &dzp->z_links, sizeof (dzp->z_links));
 825         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs), NULL,
 826             mtime, sizeof (mtime));
 827         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs), NULL,
 828             ctime, sizeof (ctime));
 829         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs), NULL,
 830             &dzp->z_pflags, sizeof (dzp->z_pflags));
 831         zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
 832         error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx);
 833         ASSERT(error == 0);
 834         mutex_exit(&dzp->z_lock);
 835 
 836         value = zfs_dirent(zp, zp->z_mode);
 837         error = zap_add(zp->z_zfsvfs->z_os, dzp->z_id, dl->dl_name,
 838             8, 1, &value, tx);
 839         ASSERT(error == 0);
 840 
 841         dnlc_update(ZTOV(dzp), dl->dl_name, vp);
 842 
 843         return (0);
 844 }
 845 
 846 /*
 847  * The match type in the code for this function should conform to:
 848  *
 849  * ------------------------------------------------------------------------
 850  * fs type  | z_norm      | lookup type | match type
 851  * ---------|-------------|-------------|----------------------------------
 852  * CS !norm | 0           |           0 | 0 (exact)
 853  * CS  norm | formX       |           0 | MT_NORMALIZE
 854  * CI !norm | upper       |   !ZCIEXACT | MT_NORMALIZE
 855  * CI !norm | upper       |    ZCIEXACT | MT_NORMALIZE | MT_MATCH_CASE
 856  * CI  norm | upper|formX |   !ZCIEXACT | MT_NORMALIZE
 857  * CI  norm | upper|formX |    ZCIEXACT | MT_NORMALIZE | MT_MATCH_CASE
 858  * CM !norm | upper       |    !ZCILOOK | MT_NORMALIZE | MT_MATCH_CASE
 859  * CM !norm | upper       |     ZCILOOK | MT_NORMALIZE
 860  * CM  norm | upper|formX |    !ZCILOOK | MT_NORMALIZE | MT_MATCH_CASE
 861  * CM  norm | upper|formX |     ZCILOOK | MT_NORMALIZE
 862  *
 863  * Abbreviations:
 864  *    CS = Case Sensitive, CI = Case Insensitive, CM = Case Mixed
 865  *    upper = case folding set by fs type on creation (U8_TEXTPREP_TOUPPER)
 866  *    formX = unicode normalization form set on fs creation
 867  */
 868 static int
 869 zfs_dropname(zfs_dirlock_t *dl, znode_t *zp, znode_t *dzp, dmu_tx_t *tx,
 870     int flag)
 871 {
 872         int error;
 873 
 874         if (zp->z_zfsvfs->z_norm) {
 875                 matchtype_t mt = MT_NORMALIZE;
 876 
 877                 if ((zp->z_zfsvfs->z_case == ZFS_CASE_INSENSITIVE &&
 878                     (flag & ZCIEXACT)) ||
 879                     (zp->z_zfsvfs->z_case == ZFS_CASE_MIXED &&
 880                     !(flag & ZCILOOK))) {
 881                         mt |= MT_MATCH_CASE;
 882                 }
 883 
 884                 error = zap_remove_norm(zp->z_zfsvfs->z_os, dzp->z_id,
 885                     dl->dl_name, mt, tx);
 886         } else {
 887                 error = zap_remove(zp->z_zfsvfs->z_os, dzp->z_id, dl->dl_name,
 888                     tx);
 889         }
 890 
 891         return (error);
 892 }
 893 
 894 /*
 895  * Unlink zp from dl, and mark zp for deletion if this was the last link.
 896  * Can fail if zp is a mount point (EBUSY) or a non-empty directory (EEXIST).
 897  * If 'unlinkedp' is NULL, we put unlinked znodes on the unlinked list.
 898  * If it's non-NULL, we use it to indicate whether the znode needs deletion,
 899  * and it's the caller's job to do it.
 900  */
 901 int
 902 zfs_link_destroy(zfs_dirlock_t *dl, znode_t *zp, dmu_tx_t *tx, int flag,
 903     boolean_t *unlinkedp)
 904 {
 905         znode_t *dzp = dl->dl_dzp;
 906         zfsvfs_t *zfsvfs = dzp->z_zfsvfs;
 907         vnode_t *vp = ZTOV(zp);
 908         int zp_is_dir = (vp->v_type == VDIR);
 909         boolean_t unlinked = B_FALSE;
 910         sa_bulk_attr_t bulk[5];
 911         uint64_t mtime[2], ctime[2];
 912         int count = 0;
 913         int error;
 914 
 915         dnlc_remove(ZTOV(dzp), dl->dl_name);
 916 
 917         if (!(flag & ZRENAMING)) {
 918                 if (vn_vfswlock(vp))            /* prevent new mounts on zp */
 919                         return (SET_ERROR(EBUSY));
 920 
 921                 if (vn_ismntpt(vp)) {           /* don't remove mount point */
 922                         vn_vfsunlock(vp);
 923                         return (SET_ERROR(EBUSY));
 924                 }
 925 
 926                 mutex_enter(&zp->z_lock);
 927 
 928                 if (zp_is_dir && !zfs_dirempty(zp)) {
 929                         mutex_exit(&zp->z_lock);
 930                         vn_vfsunlock(vp);
 931                         return (SET_ERROR(EEXIST));
 932                 }
 933 
 934                 /*
 935                  * If we get here, we are going to try to remove the object.
 936                  * First try removing the name from the directory; if that
 937                  * fails, return the error.
 938                  */
 939                 error = zfs_dropname(dl, zp, dzp, tx, flag);
 940                 if (error != 0) {
 941                         mutex_exit(&zp->z_lock);
 942                         vn_vfsunlock(vp);
 943                         return (error);
 944                 }
 945 
 946                 if (zp->z_links <= zp_is_dir) {
 947                         zfs_panic_recover("zfs: link count on %s is %u, "
 948                             "should be at least %u",
 949                             zp->z_vnode->v_path != vn_vpath_empty ?
 950                             zp->z_vnode->v_path : "<unknown>",
 951                             (int)zp->z_links, zp_is_dir + 1);
 952                         zp->z_links = zp_is_dir + 1;
 953                 }
 954                 if (--zp->z_links == zp_is_dir) {
 955                         zp->z_unlinked = B_TRUE;
 956                         zp->z_links = 0;
 957                         unlinked = B_TRUE;
 958                 } else {
 959                         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs),
 960                             NULL, &ctime, sizeof (ctime));
 961                         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
 962                             NULL, &zp->z_pflags, sizeof (zp->z_pflags));
 963                         zfs_tstamp_update_setup(zp, STATE_CHANGED, mtime, ctime,
 964                             B_TRUE);
 965                 }
 966                 SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs),
 967                     NULL, &zp->z_links, sizeof (zp->z_links));
 968                 error = sa_bulk_update(zp->z_sa_hdl, bulk, count, tx);
 969                 count = 0;
 970                 ASSERT(error == 0);
 971                 mutex_exit(&zp->z_lock);
 972                 vn_vfsunlock(vp);
 973         } else {
 974                 error = zfs_dropname(dl, zp, dzp, tx, flag);
 975                 if (error != 0)
 976                         return (error);
 977         }
 978 
 979         mutex_enter(&dzp->z_lock);
 980         dzp->z_size--;               /* one dirent removed */
 981         dzp->z_links -= zp_is_dir;   /* ".." link from zp */
 982         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_LINKS(zfsvfs),
 983             NULL, &dzp->z_links, sizeof (dzp->z_links));
 984         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_SIZE(zfsvfs),
 985             NULL, &dzp->z_size, sizeof (dzp->z_size));
 986         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_CTIME(zfsvfs),
 987             NULL, ctime, sizeof (ctime));
 988         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_MTIME(zfsvfs),
 989             NULL, mtime, sizeof (mtime));
 990         SA_ADD_BULK_ATTR(bulk, count, SA_ZPL_FLAGS(zfsvfs),
 991             NULL, &dzp->z_pflags, sizeof (dzp->z_pflags));
 992         zfs_tstamp_update_setup(dzp, CONTENT_MODIFIED, mtime, ctime, B_TRUE);
 993         error = sa_bulk_update(dzp->z_sa_hdl, bulk, count, tx);
 994         ASSERT(error == 0);
 995         mutex_exit(&dzp->z_lock);
 996 
 997         if (unlinkedp != NULL)
 998                 *unlinkedp = unlinked;
 999         else if (unlinked)
1000                 zfs_unlinked_add(zp, tx);
1001 
1002         return (0);
1003 }
1004 
1005 /*
1006  * Indicate whether the directory is empty.  Works with or without z_lock
1007  * held, but can only be consider a hint in the latter case.  Returns true
1008  * if only "." and ".." remain and there's no work in progress.
1009  */
1010 boolean_t
1011 zfs_dirempty(znode_t *dzp)
1012 {
1013         return (dzp->z_size == 2 && dzp->z_dirlocks == 0);
1014 }
1015 
1016 int
1017 zfs_make_xattrdir(znode_t *zp, vattr_t *vap, vnode_t **xvpp, cred_t *cr)
1018 {
1019         zfsvfs_t *zfsvfs = zp->z_zfsvfs;
1020         znode_t *xzp;
1021         dmu_tx_t *tx;
1022         int error;
1023         zfs_acl_ids_t acl_ids;
1024         boolean_t fuid_dirtied;
1025         uint64_t parent;
1026 
1027         *xvpp = NULL;
1028 
1029         if (error = zfs_zaccess(zp, ACE_WRITE_NAMED_ATTRS, 0, B_FALSE, cr))
1030                 return (error);
1031 
1032         if ((error = zfs_acl_ids_create(zp, IS_XATTR, vap, cr, NULL,
1033             &acl_ids)) != 0)
1034                 return (error);
1035         if (zfs_acl_ids_overquota(zfsvfs, &acl_ids)) {
1036                 zfs_acl_ids_free(&acl_ids);
1037                 return (SET_ERROR(EDQUOT));
1038         }
1039 
1040         tx = dmu_tx_create(zfsvfs->z_os);
1041         dmu_tx_hold_sa_create(tx, acl_ids.z_aclp->z_acl_bytes +
1042             ZFS_SA_BASE_ATTR_SIZE);
1043         dmu_tx_hold_sa(tx, zp->z_sa_hdl, B_TRUE);
1044         dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1045         fuid_dirtied = zfsvfs->z_fuid_dirty;
1046         if (fuid_dirtied)
1047                 zfs_fuid_txhold(zfsvfs, tx);
1048         error = dmu_tx_assign(tx, TXG_WAIT);
1049         if (error) {
1050                 zfs_acl_ids_free(&acl_ids);
1051                 dmu_tx_abort(tx);
1052                 return (error);
1053         }
1054         zfs_mknode(zp, vap, tx, cr, IS_XATTR, &xzp, &acl_ids);
1055 
1056         if (fuid_dirtied)
1057                 zfs_fuid_sync(zfsvfs, tx);
1058 
1059 #ifdef DEBUG
1060         error = sa_lookup(xzp->z_sa_hdl, SA_ZPL_PARENT(zfsvfs),
1061             &parent, sizeof (parent));
1062         ASSERT(error == 0 && parent == zp->z_id);
1063 #endif
1064 
1065         VERIFY(0 == sa_update(zp->z_sa_hdl, SA_ZPL_XATTR(zfsvfs), &xzp->z_id,
1066             sizeof (xzp->z_id), tx));
1067 
1068         (void) zfs_log_create(zfsvfs->z_log, tx, TX_MKXATTR, zp,
1069             xzp, "", NULL, acl_ids.z_fuidp, vap);
1070 
1071         zfs_acl_ids_free(&acl_ids);
1072         dmu_tx_commit(tx);
1073 
1074         *xvpp = ZTOV(xzp);
1075 
1076         return (0);
1077 }
1078 
1079 /*
1080  * Return a znode for the extended attribute directory for zp.
1081  * ** If the directory does not already exist, it is created **
1082  *
1083  *      IN:     zp      - znode to obtain attribute directory from
1084  *              cr      - credentials of caller
1085  *              flags   - flags from the VOP_LOOKUP call
1086  *
1087  *      OUT:    xzpp    - pointer to extended attribute znode
1088  *
1089  *      RETURN: 0 on success
1090  *              error number on failure
1091  */
1092 int
1093 zfs_get_xattrdir(znode_t *zp, vnode_t **xvpp, cred_t *cr, int flags)
1094 {
1095         zfsvfs_t        *zfsvfs = zp->z_zfsvfs;
1096         znode_t         *xzp;
1097         zfs_dirlock_t   *dl;
1098         vattr_t         va;
1099         int             error;
1100 top:
1101         error = zfs_dirent_lock(&dl, zp, "", &xzp, ZXATTR, NULL, NULL);
1102         if (error)
1103                 return (error);
1104 
1105         if (xzp != NULL) {
1106                 *xvpp = ZTOV(xzp);
1107                 zfs_dirent_unlock(dl);
1108                 return (0);
1109         }
1110 
1111 
1112         if (!(flags & CREATE_XATTR_DIR)) {
1113                 zfs_dirent_unlock(dl);
1114                 return (SET_ERROR(ENOENT));
1115         }
1116 
1117         if (zfsvfs->z_vfs->vfs_flag & VFS_RDONLY) {
1118                 zfs_dirent_unlock(dl);
1119                 return (SET_ERROR(EROFS));
1120         }
1121 
1122         /*
1123          * The ability to 'create' files in an attribute
1124          * directory comes from the write_xattr permission on the base file.
1125          *
1126          * The ability to 'search' an attribute directory requires
1127          * read_xattr permission on the base file.
1128          *
1129          * Once in a directory the ability to read/write attributes
1130          * is controlled by the permissions on the attribute file.
1131          */
1132         va.va_mask = AT_TYPE | AT_MODE | AT_UID | AT_GID;
1133         va.va_type = VDIR;
1134         va.va_mode = S_IFDIR | S_ISVTX | 0777;
1135         zfs_fuid_map_ids(zp, cr, &va.va_uid, &va.va_gid);
1136 
1137         error = zfs_make_xattrdir(zp, &va, xvpp, cr);
1138         zfs_dirent_unlock(dl);
1139 
1140         if (error == ERESTART) {
1141                 /* NB: we already did dmu_tx_wait() if necessary */
1142                 goto top;
1143         }
1144 
1145         return (error);
1146 }
1147 
1148 /*
1149  * Decide whether it is okay to remove within a sticky directory.
1150  *
1151  * In sticky directories, write access is not sufficient;
1152  * you can remove entries from a directory only if:
1153  *
1154  *      you own the directory,
1155  *      you own the entry,
1156  *      the entry is a plain file and you have write access,
1157  *      or you are privileged (checked in secpolicy...).
1158  *
1159  * The function returns 0 if remove access is granted.
1160  */
1161 int
1162 zfs_sticky_remove_access(znode_t *zdp, znode_t *zp, cred_t *cr)
1163 {
1164         uid_t           uid;
1165         uid_t           downer;
1166         uid_t           fowner;
1167         zfsvfs_t        *zfsvfs = zdp->z_zfsvfs;
1168 
1169         if (zdp->z_zfsvfs->z_replay)
1170                 return (0);
1171 
1172         if ((zdp->z_mode & S_ISVTX) == 0)
1173                 return (0);
1174 
1175         downer = zfs_fuid_map_id(zfsvfs, zdp->z_uid, cr, ZFS_OWNER);
1176         fowner = zfs_fuid_map_id(zfsvfs, zp->z_uid, cr, ZFS_OWNER);
1177 
1178         if ((uid = crgetuid(cr)) == downer || uid == fowner ||
1179             (ZTOV(zp)->v_type == VREG &&
1180             zfs_zaccess(zp, ACE_WRITE_DATA, 0, B_FALSE, cr) == 0))
1181                 return (0);
1182         else
1183                 return (secpolicy_vnode_remove(cr));
1184 }