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  * Copyright (c) 2013 by Delphix. All rights reserved.
  24  */
  25 
  26 /*
  27  * ZFS control directory (a.k.a. ".zfs")
  28  *
  29  * This directory provides a common location for all ZFS meta-objects.
  30  * Currently, this is only the 'snapshot' directory, but this may expand in the
  31  * future.  The elements are built using the GFS primitives, as the hierarchy
  32  * does not actually exist on disk.
  33  *
  34  * For 'snapshot', we don't want to have all snapshots always mounted, because
  35  * this would take up a huge amount of space in /etc/mnttab.  We have three
  36  * types of objects:
  37  *
  38  *      ctldir ------> snapshotdir -------> snapshot
  39  *                                             |
  40  *                                             |
  41  *                                             V
  42  *                                         mounted fs
  43  *
  44  * The 'snapshot' node contains just enough information to lookup '..' and act
  45  * as a mountpoint for the snapshot.  Whenever we lookup a specific snapshot, we
  46  * perform an automount of the underlying filesystem and return the
  47  * corresponding vnode.
  48  *
  49  * All mounts are handled automatically by the kernel, but unmounts are
  50  * (currently) handled from user land.  The main reason is that there is no
  51  * reliable way to auto-unmount the filesystem when it's "no longer in use".
  52  * When the user unmounts a filesystem, we call zfsctl_unmount(), which
  53  * unmounts any snapshots within the snapshot directory.
  54  *
  55  * The '.zfs', '.zfs/snapshot', and all directories created under
  56  * '.zfs/snapshot' (ie: '.zfs/snapshot/<snapname>') are all GFS nodes and
  57  * share the same vfs_t as the head filesystem (what '.zfs' lives under).
  58  *
  59  * File systems mounted ontop of the GFS nodes '.zfs/snapshot/<snapname>'
  60  * (ie: snapshots) are ZFS nodes and have their own unique vfs_t.
  61  * However, vnodes within these mounted on file systems have their v_vfsp
  62  * fields set to the head filesystem to make NFS happy (see
  63  * zfsctl_snapdir_lookup()). We VFS_HOLD the head filesystem's vfs_t
  64  * so that it cannot be freed until all snapshots have been unmounted.
  65  */
  66 
  67 #include <fs/fs_subr.h>
  68 #include <sys/zfs_ctldir.h>
  69 #include <sys/zfs_ioctl.h>
  70 #include <sys/zfs_vfsops.h>
  71 #include <sys/vfs_opreg.h>
  72 #include <sys/gfs.h>
  73 #include <sys/stat.h>
  74 #include <sys/dmu.h>
  75 #include <sys/dsl_destroy.h>
  76 #include <sys/dsl_deleg.h>
  77 #include <sys/mount.h>
  78 #include <sys/sunddi.h>
  79 
  80 #include "zfs_namecheck.h"
  81 
  82 typedef struct zfsctl_node {
  83         gfs_dir_t       zc_gfs_private;
  84         uint64_t        zc_id;
  85         timestruc_t     zc_cmtime;      /* ctime and mtime, always the same */
  86 } zfsctl_node_t;
  87 
  88 typedef struct zfsctl_snapdir {
  89         zfsctl_node_t   sd_node;
  90         kmutex_t        sd_lock;
  91         avl_tree_t      sd_snaps;
  92 } zfsctl_snapdir_t;
  93 
  94 typedef struct {
  95         char            *se_name;
  96         vnode_t         *se_root;
  97         avl_node_t      se_node;
  98 } zfs_snapentry_t;
  99 
 100 static int
 101 snapentry_compare(const void *a, const void *b)
 102 {
 103         const zfs_snapentry_t *sa = a;
 104         const zfs_snapentry_t *sb = b;
 105         int ret = strcmp(sa->se_name, sb->se_name);
 106 
 107         if (ret < 0)
 108                 return (-1);
 109         else if (ret > 0)
 110                 return (1);
 111         else
 112                 return (0);
 113 }
 114 
 115 vnodeops_t *zfsctl_ops_root;
 116 vnodeops_t *zfsctl_ops_snapdir;
 117 vnodeops_t *zfsctl_ops_snapshot;
 118 vnodeops_t *zfsctl_ops_shares;
 119 vnodeops_t *zfsctl_ops_shares_dir;
 120 
 121 static const fs_operation_def_t zfsctl_tops_root[];
 122 static const fs_operation_def_t zfsctl_tops_snapdir[];
 123 static const fs_operation_def_t zfsctl_tops_snapshot[];
 124 static const fs_operation_def_t zfsctl_tops_shares[];
 125 
 126 static vnode_t *zfsctl_mknode_snapdir(vnode_t *);
 127 static vnode_t *zfsctl_mknode_shares(vnode_t *);
 128 static vnode_t *zfsctl_snapshot_mknode(vnode_t *, uint64_t objset);
 129 static int zfsctl_unmount_snap(zfs_snapentry_t *, int, cred_t *);
 130 
 131 static gfs_opsvec_t zfsctl_opsvec[] = {
 132         { ".zfs", zfsctl_tops_root, &zfsctl_ops_root },
 133         { ".zfs/snapshot", zfsctl_tops_snapdir, &zfsctl_ops_snapdir },
 134         { ".zfs/snapshot/vnode", zfsctl_tops_snapshot, &zfsctl_ops_snapshot },
 135         { ".zfs/shares", zfsctl_tops_shares, &zfsctl_ops_shares_dir },
 136         { ".zfs/shares/vnode", zfsctl_tops_shares, &zfsctl_ops_shares },
 137         { NULL }
 138 };
 139 
 140 /*
 141  * Root directory elements.  We only have two entries
 142  * snapshot and shares.
 143  */
 144 static gfs_dirent_t zfsctl_root_entries[] = {
 145         { "snapshot", zfsctl_mknode_snapdir, GFS_CACHE_VNODE },
 146         { "shares", zfsctl_mknode_shares, GFS_CACHE_VNODE },
 147         { NULL }
 148 };
 149 
 150 /* include . and .. in the calculation */
 151 #define NROOT_ENTRIES   ((sizeof (zfsctl_root_entries) / \
 152     sizeof (gfs_dirent_t)) + 1)
 153 
 154 
 155 /*
 156  * Initialize the various GFS pieces we'll need to create and manipulate .zfs
 157  * directories.  This is called from the ZFS init routine, and initializes the
 158  * vnode ops vectors that we'll be using.
 159  */
 160 void
 161 zfsctl_init(void)
 162 {
 163         VERIFY(gfs_make_opsvec(zfsctl_opsvec) == 0);
 164 }
 165 
 166 void
 167 zfsctl_fini(void)
 168 {
 169         /*
 170          * Remove vfsctl vnode ops
 171          */
 172         if (zfsctl_ops_root)
 173                 vn_freevnodeops(zfsctl_ops_root);
 174         if (zfsctl_ops_snapdir)
 175                 vn_freevnodeops(zfsctl_ops_snapdir);
 176         if (zfsctl_ops_snapshot)
 177                 vn_freevnodeops(zfsctl_ops_snapshot);
 178         if (zfsctl_ops_shares)
 179                 vn_freevnodeops(zfsctl_ops_shares);
 180         if (zfsctl_ops_shares_dir)
 181                 vn_freevnodeops(zfsctl_ops_shares_dir);
 182 
 183         zfsctl_ops_root = NULL;
 184         zfsctl_ops_snapdir = NULL;
 185         zfsctl_ops_snapshot = NULL;
 186         zfsctl_ops_shares = NULL;
 187         zfsctl_ops_shares_dir = NULL;
 188 }
 189 
 190 boolean_t
 191 zfsctl_is_node(vnode_t *vp)
 192 {
 193         return (vn_matchops(vp, zfsctl_ops_root) ||
 194             vn_matchops(vp, zfsctl_ops_snapdir) ||
 195             vn_matchops(vp, zfsctl_ops_snapshot) ||
 196             vn_matchops(vp, zfsctl_ops_shares) ||
 197             vn_matchops(vp, zfsctl_ops_shares_dir));
 198 
 199 }
 200 
 201 /*
 202  * Return the inode number associated with the 'snapshot' or
 203  * 'shares' directory.
 204  */
 205 /* ARGSUSED */
 206 static ino64_t
 207 zfsctl_root_inode_cb(vnode_t *vp, int index)
 208 {
 209         zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
 210 
 211         ASSERT(index <= 2);
 212 
 213         if (index == 0)
 214                 return (ZFSCTL_INO_SNAPDIR);
 215 
 216         return (zfsvfs->z_shares_dir);
 217 }
 218 
 219 /*
 220  * Create the '.zfs' directory.  This directory is cached as part of the VFS
 221  * structure.  This results in a hold on the vfs_t.  The code in zfs_umount()
 222  * therefore checks against a vfs_count of 2 instead of 1.  This reference
 223  * is removed when the ctldir is destroyed in the unmount.
 224  */
 225 void
 226 zfsctl_create(zfsvfs_t *zfsvfs)
 227 {
 228         vnode_t *vp, *rvp;
 229         zfsctl_node_t *zcp;
 230         uint64_t crtime[2];
 231 
 232         ASSERT(zfsvfs->z_ctldir == NULL);
 233 
 234         vp = gfs_root_create(sizeof (zfsctl_node_t), zfsvfs->z_vfs,
 235             zfsctl_ops_root, ZFSCTL_INO_ROOT, zfsctl_root_entries,
 236             zfsctl_root_inode_cb, MAXNAMELEN, NULL, NULL);
 237         zcp = vp->v_data;
 238         zcp->zc_id = ZFSCTL_INO_ROOT;
 239 
 240         VERIFY(VFS_ROOT(zfsvfs->z_vfs, &rvp) == 0);
 241         VERIFY(0 == sa_lookup(VTOZ(rvp)->z_sa_hdl, SA_ZPL_CRTIME(zfsvfs),
 242             &crtime, sizeof (crtime)));
 243         ZFS_TIME_DECODE(&zcp->zc_cmtime, crtime);
 244         VN_RELE(rvp);
 245 
 246         /*
 247          * We're only faking the fact that we have a root of a filesystem for
 248          * the sake of the GFS interfaces.  Undo the flag manipulation it did
 249          * for us.
 250          */
 251         vp->v_flag &= ~(VROOT | VNOCACHE | VNOMAP | VNOSWAP | VNOMOUNT);
 252 
 253         zfsvfs->z_ctldir = vp;
 254 }
 255 
 256 /*
 257  * Destroy the '.zfs' directory.  Only called when the filesystem is unmounted.
 258  * There might still be more references if we were force unmounted, but only
 259  * new zfs_inactive() calls can occur and they don't reference .zfs
 260  */
 261 void
 262 zfsctl_destroy(zfsvfs_t *zfsvfs)
 263 {
 264         VN_RELE(zfsvfs->z_ctldir);
 265         zfsvfs->z_ctldir = NULL;
 266 }
 267 
 268 /*
 269  * Given a root znode, retrieve the associated .zfs directory.
 270  * Add a hold to the vnode and return it.
 271  */
 272 vnode_t *
 273 zfsctl_root(znode_t *zp)
 274 {
 275         ASSERT(zfs_has_ctldir(zp));
 276         VN_HOLD(zp->z_zfsvfs->z_ctldir);
 277         return (zp->z_zfsvfs->z_ctldir);
 278 }
 279 
 280 /*
 281  * Common open routine.  Disallow any write access.
 282  */
 283 /* ARGSUSED */
 284 static int
 285 zfsctl_common_open(vnode_t **vpp, int flags, cred_t *cr, caller_context_t *ct)
 286 {
 287         if (flags & FWRITE)
 288                 return (SET_ERROR(EACCES));
 289 
 290         return (0);
 291 }
 292 
 293 /*
 294  * Common close routine.  Nothing to do here.
 295  */
 296 /* ARGSUSED */
 297 static int
 298 zfsctl_common_close(vnode_t *vpp, int flags, int count, offset_t off,
 299     cred_t *cr, caller_context_t *ct)
 300 {
 301         return (0);
 302 }
 303 
 304 /*
 305  * Common access routine.  Disallow writes.
 306  */
 307 /* ARGSUSED */
 308 static int
 309 zfsctl_common_access(vnode_t *vp, int mode, int flags, cred_t *cr,
 310     caller_context_t *ct)
 311 {
 312         if (flags & V_ACE_MASK) {
 313                 if (mode & ACE_ALL_WRITE_PERMS)
 314                         return (SET_ERROR(EACCES));
 315         } else {
 316                 if (mode & VWRITE)
 317                         return (SET_ERROR(EACCES));
 318         }
 319 
 320         return (0);
 321 }
 322 
 323 /*
 324  * Common getattr function.  Fill in basic information.
 325  */
 326 static void
 327 zfsctl_common_getattr(vnode_t *vp, vattr_t *vap)
 328 {
 329         timestruc_t     now;
 330 
 331         vap->va_uid = 0;
 332         vap->va_gid = 0;
 333         vap->va_rdev = 0;
 334         /*
 335          * We are a purely virtual object, so we have no
 336          * blocksize or allocated blocks.
 337          */
 338         vap->va_blksize = 0;
 339         vap->va_nblocks = 0;
 340         vap->va_seq = 0;
 341         vap->va_fsid = vp->v_vfsp->vfs_dev;
 342         vap->va_mode = S_IRUSR | S_IXUSR | S_IRGRP | S_IXGRP |
 343             S_IROTH | S_IXOTH;
 344         vap->va_type = VDIR;
 345         /*
 346          * We live in the now (for atime).
 347          */
 348         gethrestime(&now);
 349         vap->va_atime = now;
 350 }
 351 
 352 /*ARGSUSED*/
 353 static int
 354 zfsctl_common_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
 355 {
 356         zfsvfs_t        *zfsvfs = vp->v_vfsp->vfs_data;
 357         zfsctl_node_t   *zcp = vp->v_data;
 358         uint64_t        object = zcp->zc_id;
 359         zfid_short_t    *zfid;
 360         int             i;
 361 
 362         ZFS_ENTER(zfsvfs);
 363 
 364         if (fidp->fid_len < SHORT_FID_LEN) {
 365                 fidp->fid_len = SHORT_FID_LEN;
 366                 ZFS_EXIT(zfsvfs);
 367                 return (SET_ERROR(ENOSPC));
 368         }
 369 
 370         zfid = (zfid_short_t *)fidp;
 371 
 372         zfid->zf_len = SHORT_FID_LEN;
 373 
 374         for (i = 0; i < sizeof (zfid->zf_object); i++)
 375                 zfid->zf_object[i] = (uint8_t)(object >> (8 * i));
 376 
 377         /* .zfs znodes always have a generation number of 0 */
 378         for (i = 0; i < sizeof (zfid->zf_gen); i++)
 379                 zfid->zf_gen[i] = 0;
 380 
 381         ZFS_EXIT(zfsvfs);
 382         return (0);
 383 }
 384 
 385 
 386 /*ARGSUSED*/
 387 static int
 388 zfsctl_shares_fid(vnode_t *vp, fid_t *fidp, caller_context_t *ct)
 389 {
 390         zfsvfs_t        *zfsvfs = vp->v_vfsp->vfs_data;
 391         znode_t         *dzp;
 392         int             error;
 393 
 394         ZFS_ENTER(zfsvfs);
 395 
 396         if (zfsvfs->z_shares_dir == 0) {
 397                 ZFS_EXIT(zfsvfs);
 398                 return (SET_ERROR(ENOTSUP));
 399         }
 400 
 401         if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
 402                 error = VOP_FID(ZTOV(dzp), fidp, ct);
 403                 VN_RELE(ZTOV(dzp));
 404         }
 405 
 406         ZFS_EXIT(zfsvfs);
 407         return (error);
 408 }
 409 /*
 410  * .zfs inode namespace
 411  *
 412  * We need to generate unique inode numbers for all files and directories
 413  * within the .zfs pseudo-filesystem.  We use the following scheme:
 414  *
 415  *      ENTRY                   ZFSCTL_INODE
 416  *      .zfs                    1
 417  *      .zfs/snapshot           2
 418  *      .zfs/snapshot/<snap>      objectid(snap)
 419  */
 420 
 421 #define ZFSCTL_INO_SNAP(id)     (id)
 422 
 423 /*
 424  * Get root directory attributes.
 425  */
 426 /* ARGSUSED */
 427 static int
 428 zfsctl_root_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
 429     caller_context_t *ct)
 430 {
 431         zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
 432         zfsctl_node_t *zcp = vp->v_data;
 433 
 434         ZFS_ENTER(zfsvfs);
 435         vap->va_nodeid = ZFSCTL_INO_ROOT;
 436         vap->va_nlink = vap->va_size = NROOT_ENTRIES;
 437         vap->va_mtime = vap->va_ctime = zcp->zc_cmtime;
 438 
 439         zfsctl_common_getattr(vp, vap);
 440         ZFS_EXIT(zfsvfs);
 441 
 442         return (0);
 443 }
 444 
 445 /*
 446  * Special case the handling of "..".
 447  */
 448 /* ARGSUSED */
 449 int
 450 zfsctl_root_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp,
 451     int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
 452     int *direntflags, pathname_t *realpnp)
 453 {
 454         zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
 455         int err;
 456 
 457         /*
 458          * No extended attributes allowed under .zfs
 459          */
 460         if (flags & LOOKUP_XATTR)
 461                 return (SET_ERROR(EINVAL));
 462 
 463         ZFS_ENTER(zfsvfs);
 464 
 465         if (strcmp(nm, "..") == 0) {
 466                 err = VFS_ROOT(dvp->v_vfsp, vpp);
 467         } else {
 468                 err = gfs_vop_lookup(dvp, nm, vpp, pnp, flags, rdir,
 469                     cr, ct, direntflags, realpnp);
 470         }
 471 
 472         ZFS_EXIT(zfsvfs);
 473 
 474         return (err);
 475 }
 476 
 477 static int
 478 zfsctl_pathconf(vnode_t *vp, int cmd, ulong_t *valp, cred_t *cr,
 479     caller_context_t *ct)
 480 {
 481         /*
 482          * We only care about ACL_ENABLED so that libsec can
 483          * display ACL correctly and not default to POSIX draft.
 484          */
 485         if (cmd == _PC_ACL_ENABLED) {
 486                 *valp = _ACL_ACE_ENABLED;
 487                 return (0);
 488         }
 489 
 490         return (fs_pathconf(vp, cmd, valp, cr, ct));
 491 }
 492 
 493 static const fs_operation_def_t zfsctl_tops_root[] = {
 494         { VOPNAME_OPEN,         { .vop_open = zfsctl_common_open }      },
 495         { VOPNAME_CLOSE,        { .vop_close = zfsctl_common_close }    },
 496         { VOPNAME_IOCTL,        { .error = fs_inval }                   },
 497         { VOPNAME_GETATTR,      { .vop_getattr = zfsctl_root_getattr }  },
 498         { VOPNAME_ACCESS,       { .vop_access = zfsctl_common_access }  },
 499         { VOPNAME_READDIR,      { .vop_readdir = gfs_vop_readdir }      },
 500         { VOPNAME_LOOKUP,       { .vop_lookup = zfsctl_root_lookup }    },
 501         { VOPNAME_SEEK,         { .vop_seek = fs_seek }                 },
 502         { VOPNAME_INACTIVE,     { .vop_inactive = gfs_vop_inactive }    },
 503         { VOPNAME_PATHCONF,     { .vop_pathconf = zfsctl_pathconf }     },
 504         { VOPNAME_FID,          { .vop_fid = zfsctl_common_fid  }       },
 505         { NULL }
 506 };
 507 
 508 /*
 509  * Gets the full dataset name that corresponds to the given snapshot name
 510  * Example:
 511  *      zfsctl_snapshot_zname("snap1") -> "mypool/myfs@snap1"
 512  */
 513 static int
 514 zfsctl_snapshot_zname(vnode_t *vp, const char *name, int len, char *zname)
 515 {
 516         objset_t *os = ((zfsvfs_t *)((vp)->v_vfsp->vfs_data))->z_os;
 517 
 518         if (snapshot_namecheck(name, NULL, NULL) != 0)
 519                 return (SET_ERROR(EILSEQ));
 520         dmu_objset_name(os, zname);
 521         if (strlen(zname) + 1 + strlen(name) >= len)
 522                 return (SET_ERROR(ENAMETOOLONG));
 523         (void) strcat(zname, "@");
 524         (void) strcat(zname, name);
 525         return (0);
 526 }
 527 
 528 static int
 529 zfsctl_unmount_snap(zfs_snapentry_t *sep, int fflags, cred_t *cr)
 530 {
 531         vnode_t *svp = sep->se_root;
 532         int error;
 533 
 534         ASSERT(vn_ismntpt(svp));
 535 
 536         /* this will be dropped by dounmount() */
 537         if ((error = vn_vfswlock(svp)) != 0)
 538                 return (error);
 539 
 540         VN_HOLD(svp);
 541         error = dounmount(vn_mountedvfs(svp), fflags, cr);
 542         if (error) {
 543                 VN_RELE(svp);
 544                 return (error);
 545         }
 546 
 547         /*
 548          * We can't use VN_RELE(), as that will try to invoke
 549          * zfsctl_snapdir_inactive(), which would cause us to destroy
 550          * the sd_lock mutex held by our caller.
 551          */
 552         ASSERT(svp->v_count == 1);
 553         gfs_vop_inactive(svp, cr, NULL);
 554 
 555         kmem_free(sep->se_name, strlen(sep->se_name) + 1);
 556         kmem_free(sep, sizeof (zfs_snapentry_t));
 557 
 558         return (0);
 559 }
 560 
 561 static void
 562 zfsctl_rename_snap(zfsctl_snapdir_t *sdp, zfs_snapentry_t *sep, const char *nm)
 563 {
 564         avl_index_t where;
 565         vfs_t *vfsp;
 566         refstr_t *pathref;
 567         char newpath[MAXNAMELEN];
 568         char *tail;
 569 
 570         ASSERT(MUTEX_HELD(&sdp->sd_lock));
 571         ASSERT(sep != NULL);
 572 
 573         vfsp = vn_mountedvfs(sep->se_root);
 574         ASSERT(vfsp != NULL);
 575 
 576         vfs_lock_wait(vfsp);
 577 
 578         /*
 579          * Change the name in the AVL tree.
 580          */
 581         avl_remove(&sdp->sd_snaps, sep);
 582         kmem_free(sep->se_name, strlen(sep->se_name) + 1);
 583         sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP);
 584         (void) strcpy(sep->se_name, nm);
 585         VERIFY(avl_find(&sdp->sd_snaps, sep, &where) == NULL);
 586         avl_insert(&sdp->sd_snaps, sep, where);
 587 
 588         /*
 589          * Change the current mountpoint info:
 590          *      - update the tail of the mntpoint path
 591          *      - update the tail of the resource path
 592          */
 593         pathref = vfs_getmntpoint(vfsp);
 594         (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath));
 595         VERIFY((tail = strrchr(newpath, '/')) != NULL);
 596         *(tail+1) = '\0';
 597         ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath));
 598         (void) strcat(newpath, nm);
 599         refstr_rele(pathref);
 600         vfs_setmntpoint(vfsp, newpath, 0);
 601 
 602         pathref = vfs_getresource(vfsp);
 603         (void) strncpy(newpath, refstr_value(pathref), sizeof (newpath));
 604         VERIFY((tail = strrchr(newpath, '@')) != NULL);
 605         *(tail+1) = '\0';
 606         ASSERT3U(strlen(newpath) + strlen(nm), <, sizeof (newpath));
 607         (void) strcat(newpath, nm);
 608         refstr_rele(pathref);
 609         vfs_setresource(vfsp, newpath, 0);
 610 
 611         vfs_unlock(vfsp);
 612 }
 613 
 614 /*ARGSUSED*/
 615 static int
 616 zfsctl_snapdir_rename(vnode_t *sdvp, char *snm, vnode_t *tdvp, char *tnm,
 617     cred_t *cr, caller_context_t *ct, int flags)
 618 {
 619         zfsctl_snapdir_t *sdp = sdvp->v_data;
 620         zfs_snapentry_t search, *sep;
 621         zfsvfs_t *zfsvfs;
 622         avl_index_t where;
 623         char from[MAXNAMELEN], to[MAXNAMELEN];
 624         char real[MAXNAMELEN], fsname[MAXNAMELEN];
 625         int err;
 626 
 627         zfsvfs = sdvp->v_vfsp->vfs_data;
 628         ZFS_ENTER(zfsvfs);
 629 
 630         if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
 631                 err = dmu_snapshot_realname(zfsvfs->z_os, snm, real,
 632                     MAXNAMELEN, NULL);
 633                 if (err == 0) {
 634                         snm = real;
 635                 } else if (err != ENOTSUP) {
 636                         ZFS_EXIT(zfsvfs);
 637                         return (err);
 638                 }
 639         }
 640 
 641         ZFS_EXIT(zfsvfs);
 642 
 643         dmu_objset_name(zfsvfs->z_os, fsname);
 644 
 645         err = zfsctl_snapshot_zname(sdvp, snm, MAXNAMELEN, from);
 646         if (err == 0)
 647                 err = zfsctl_snapshot_zname(tdvp, tnm, MAXNAMELEN, to);
 648         if (err == 0)
 649                 err = zfs_secpolicy_rename_perms(from, to, cr);
 650         if (err != 0)
 651                 return (err);
 652 
 653         /*
 654          * Cannot move snapshots out of the snapdir.
 655          */
 656         if (sdvp != tdvp)
 657                 return (SET_ERROR(EINVAL));
 658 
 659         if (strcmp(snm, tnm) == 0)
 660                 return (0);
 661 
 662         mutex_enter(&sdp->sd_lock);
 663 
 664         search.se_name = (char *)snm;
 665         if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) == NULL) {
 666                 mutex_exit(&sdp->sd_lock);
 667                 return (SET_ERROR(ENOENT));
 668         }
 669 
 670         err = dsl_dataset_rename_snapshot(fsname, snm, tnm, B_FALSE);
 671         if (err == 0)
 672                 zfsctl_rename_snap(sdp, sep, tnm);
 673 
 674         mutex_exit(&sdp->sd_lock);
 675 
 676         return (err);
 677 }
 678 
 679 /* ARGSUSED */
 680 static int
 681 zfsctl_snapdir_remove(vnode_t *dvp, char *name, vnode_t *cwd, cred_t *cr,
 682     caller_context_t *ct, int flags)
 683 {
 684         zfsctl_snapdir_t *sdp = dvp->v_data;
 685         zfs_snapentry_t *sep;
 686         zfs_snapentry_t search;
 687         zfsvfs_t *zfsvfs;
 688         char snapname[MAXNAMELEN];
 689         char real[MAXNAMELEN];
 690         int err;
 691 
 692         zfsvfs = dvp->v_vfsp->vfs_data;
 693         ZFS_ENTER(zfsvfs);
 694 
 695         if ((flags & FIGNORECASE) || zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
 696 
 697                 err = dmu_snapshot_realname(zfsvfs->z_os, name, real,
 698                     MAXNAMELEN, NULL);
 699                 if (err == 0) {
 700                         name = real;
 701                 } else if (err != ENOTSUP) {
 702                         ZFS_EXIT(zfsvfs);
 703                         return (err);
 704                 }
 705         }
 706 
 707         ZFS_EXIT(zfsvfs);
 708 
 709         err = zfsctl_snapshot_zname(dvp, name, MAXNAMELEN, snapname);
 710         if (err == 0)
 711                 err = zfs_secpolicy_destroy_perms(snapname, cr);
 712         if (err != 0)
 713                 return (err);
 714 
 715         mutex_enter(&sdp->sd_lock);
 716 
 717         search.se_name = name;
 718         sep = avl_find(&sdp->sd_snaps, &search, NULL);
 719         if (sep) {
 720                 avl_remove(&sdp->sd_snaps, sep);
 721                 err = zfsctl_unmount_snap(sep, MS_FORCE, cr);
 722                 if (err != 0)
 723                         avl_add(&sdp->sd_snaps, sep);
 724                 else
 725                         err = dsl_destroy_snapshot(snapname, B_FALSE);
 726         } else {
 727                 err = SET_ERROR(ENOENT);
 728         }
 729 
 730         mutex_exit(&sdp->sd_lock);
 731 
 732         return (err);
 733 }
 734 
 735 /*
 736  * This creates a snapshot under '.zfs/snapshot'.
 737  */
 738 /* ARGSUSED */
 739 static int
 740 zfsctl_snapdir_mkdir(vnode_t *dvp, char *dirname, vattr_t *vap, vnode_t  **vpp,
 741     cred_t *cr, caller_context_t *cc, int flags, vsecattr_t *vsecp)
 742 {
 743         zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
 744         char name[MAXNAMELEN];
 745         int err;
 746         static enum symfollow follow = NO_FOLLOW;
 747         static enum uio_seg seg = UIO_SYSSPACE;
 748 
 749         if (snapshot_namecheck(dirname, NULL, NULL) != 0)
 750                 return (SET_ERROR(EILSEQ));
 751 
 752         dmu_objset_name(zfsvfs->z_os, name);
 753 
 754         *vpp = NULL;
 755 
 756         err = zfs_secpolicy_snapshot_perms(name, cr);
 757         if (err != 0)
 758                 return (err);
 759 
 760         if (err == 0) {
 761                 err = dmu_objset_snapshot_one(name, dirname);
 762                 if (err != 0)
 763                         return (err);
 764                 err = lookupnameat(dirname, seg, follow, NULL, vpp, dvp);
 765         }
 766 
 767         return (err);
 768 }
 769 
 770 /*
 771  * Lookup entry point for the 'snapshot' directory.  Try to open the
 772  * snapshot if it exist, creating the pseudo filesystem vnode as necessary.
 773  * Perform a mount of the associated dataset on top of the vnode.
 774  */
 775 /* ARGSUSED */
 776 static int
 777 zfsctl_snapdir_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp,
 778     int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
 779     int *direntflags, pathname_t *realpnp)
 780 {
 781         zfsctl_snapdir_t *sdp = dvp->v_data;
 782         objset_t *snap;
 783         char snapname[MAXNAMELEN];
 784         char real[MAXNAMELEN];
 785         char *mountpoint;
 786         zfs_snapentry_t *sep, search;
 787         struct mounta margs;
 788         vfs_t *vfsp;
 789         size_t mountpoint_len;
 790         avl_index_t where;
 791         zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
 792         int err;
 793 
 794         /*
 795          * No extended attributes allowed under .zfs
 796          */
 797         if (flags & LOOKUP_XATTR)
 798                 return (SET_ERROR(EINVAL));
 799 
 800         ASSERT(dvp->v_type == VDIR);
 801 
 802         /*
 803          * If we get a recursive call, that means we got called
 804          * from the domount() code while it was trying to look up the
 805          * spec (which looks like a local path for zfs).  We need to
 806          * add some flag to domount() to tell it not to do this lookup.
 807          */
 808         if (MUTEX_HELD(&sdp->sd_lock))
 809                 return (SET_ERROR(ENOENT));
 810 
 811         ZFS_ENTER(zfsvfs);
 812 
 813         if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) {
 814                 ZFS_EXIT(zfsvfs);
 815                 return (0);
 816         }
 817 
 818         if (flags & FIGNORECASE) {
 819                 boolean_t conflict = B_FALSE;
 820 
 821                 err = dmu_snapshot_realname(zfsvfs->z_os, nm, real,
 822                     MAXNAMELEN, &conflict);
 823                 if (err == 0) {
 824                         nm = real;
 825                 } else if (err != ENOTSUP) {
 826                         ZFS_EXIT(zfsvfs);
 827                         return (err);
 828                 }
 829                 if (realpnp)
 830                         (void) strlcpy(realpnp->pn_buf, nm,
 831                             realpnp->pn_bufsize);
 832                 if (conflict && direntflags)
 833                         *direntflags = ED_CASE_CONFLICT;
 834         }
 835 
 836         mutex_enter(&sdp->sd_lock);
 837         search.se_name = (char *)nm;
 838         if ((sep = avl_find(&sdp->sd_snaps, &search, &where)) != NULL) {
 839                 *vpp = sep->se_root;
 840                 VN_HOLD(*vpp);
 841                 err = traverse(vpp);
 842                 if (err != 0) {
 843                         VN_RELE(*vpp);
 844                         *vpp = NULL;
 845                 } else if (*vpp == sep->se_root) {
 846                         /*
 847                          * The snapshot was unmounted behind our backs,
 848                          * try to remount it.
 849                          */
 850                         goto domount;
 851                 } else {
 852                         /*
 853                          * VROOT was set during the traverse call.  We need
 854                          * to clear it since we're pretending to be part
 855                          * of our parent's vfs.
 856                          */
 857                         (*vpp)->v_flag &= ~VROOT;
 858                 }
 859                 mutex_exit(&sdp->sd_lock);
 860                 ZFS_EXIT(zfsvfs);
 861                 return (err);
 862         }
 863 
 864         /*
 865          * The requested snapshot is not currently mounted, look it up.
 866          */
 867         err = zfsctl_snapshot_zname(dvp, nm, MAXNAMELEN, snapname);
 868         if (err != 0) {
 869                 mutex_exit(&sdp->sd_lock);
 870                 ZFS_EXIT(zfsvfs);
 871                 /*
 872                  * handle "ls *" or "?" in a graceful manner,
 873                  * forcing EILSEQ to ENOENT.
 874                  * Since shell ultimately passes "*" or "?" as name to lookup
 875                  */
 876                 return (err == EILSEQ ? ENOENT : err);
 877         }
 878         if (dmu_objset_hold(snapname, FTAG, &snap) != 0) {
 879                 mutex_exit(&sdp->sd_lock);
 880                 ZFS_EXIT(zfsvfs);
 881                 return (SET_ERROR(ENOENT));
 882         }
 883 
 884         sep = kmem_alloc(sizeof (zfs_snapentry_t), KM_SLEEP);
 885         sep->se_name = kmem_alloc(strlen(nm) + 1, KM_SLEEP);
 886         (void) strcpy(sep->se_name, nm);
 887         *vpp = sep->se_root = zfsctl_snapshot_mknode(dvp, dmu_objset_id(snap));
 888         avl_insert(&sdp->sd_snaps, sep, where);
 889 
 890         dmu_objset_rele(snap, FTAG);
 891 domount:
 892         mountpoint_len = strlen(refstr_value(dvp->v_vfsp->vfs_mntpt)) +
 893             strlen("/.zfs/snapshot/") + strlen(nm) + 1;
 894         mountpoint = kmem_alloc(mountpoint_len, KM_SLEEP);
 895         (void) snprintf(mountpoint, mountpoint_len, "%s/.zfs/snapshot/%s",
 896             refstr_value(dvp->v_vfsp->vfs_mntpt), nm);
 897 
 898         margs.spec = snapname;
 899         margs.dir = mountpoint;
 900         margs.flags = MS_SYSSPACE | MS_NOMNTTAB;
 901         margs.fstype = "zfs";
 902         margs.dataptr = NULL;
 903         margs.datalen = 0;
 904         margs.optptr = NULL;
 905         margs.optlen = 0;
 906 
 907         err = domount("zfs", &margs, *vpp, kcred, &vfsp);
 908         kmem_free(mountpoint, mountpoint_len);
 909 
 910         if (err == 0) {
 911                 /*
 912                  * Return the mounted root rather than the covered mount point.
 913                  * Takes the GFS vnode at .zfs/snapshot/<snapname> and returns
 914                  * the ZFS vnode mounted on top of the GFS node.  This ZFS
 915                  * vnode is the root of the newly created vfsp.
 916                  */
 917                 VFS_RELE(vfsp);
 918                 err = traverse(vpp);
 919         }
 920 
 921         if (err == 0) {
 922                 /*
 923                  * Fix up the root vnode mounted on .zfs/snapshot/<snapname>.
 924                  *
 925                  * This is where we lie about our v_vfsp in order to
 926                  * make .zfs/snapshot/<snapname> accessible over NFS
 927                  * without requiring manual mounts of <snapname>.
 928                  */
 929                 ASSERT(VTOZ(*vpp)->z_zfsvfs != zfsvfs);
 930                 VTOZ(*vpp)->z_zfsvfs->z_parent = zfsvfs;
 931                 (*vpp)->v_vfsp = zfsvfs->z_vfs;
 932                 (*vpp)->v_flag &= ~VROOT;
 933         }
 934         mutex_exit(&sdp->sd_lock);
 935         ZFS_EXIT(zfsvfs);
 936 
 937         /*
 938          * If we had an error, drop our hold on the vnode and
 939          * zfsctl_snapshot_inactive() will clean up.
 940          */
 941         if (err != 0) {
 942                 VN_RELE(*vpp);
 943                 *vpp = NULL;
 944         }
 945         return (err);
 946 }
 947 
 948 /* ARGSUSED */
 949 static int
 950 zfsctl_shares_lookup(vnode_t *dvp, char *nm, vnode_t **vpp, pathname_t *pnp,
 951     int flags, vnode_t *rdir, cred_t *cr, caller_context_t *ct,
 952     int *direntflags, pathname_t *realpnp)
 953 {
 954         zfsvfs_t *zfsvfs = dvp->v_vfsp->vfs_data;
 955         znode_t *dzp;
 956         int error;
 957 
 958         ZFS_ENTER(zfsvfs);
 959 
 960         if (gfs_lookup_dot(vpp, dvp, zfsvfs->z_ctldir, nm) == 0) {
 961                 ZFS_EXIT(zfsvfs);
 962                 return (0);
 963         }
 964 
 965         if (zfsvfs->z_shares_dir == 0) {
 966                 ZFS_EXIT(zfsvfs);
 967                 return (SET_ERROR(ENOTSUP));
 968         }
 969         if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0)
 970                 error = VOP_LOOKUP(ZTOV(dzp), nm, vpp, pnp,
 971                     flags, rdir, cr, ct, direntflags, realpnp);
 972 
 973         VN_RELE(ZTOV(dzp));
 974         ZFS_EXIT(zfsvfs);
 975 
 976         return (error);
 977 }
 978 
 979 /* ARGSUSED */
 980 static int
 981 zfsctl_snapdir_readdir_cb(vnode_t *vp, void *dp, int *eofp,
 982     offset_t *offp, offset_t *nextp, void *data, int flags)
 983 {
 984         zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
 985         char snapname[MAXNAMELEN];
 986         uint64_t id, cookie;
 987         boolean_t case_conflict;
 988         int error;
 989 
 990         ZFS_ENTER(zfsvfs);
 991 
 992         cookie = *offp;
 993         dsl_pool_config_enter(dmu_objset_pool(zfsvfs->z_os), FTAG);
 994         error = dmu_snapshot_list_next(zfsvfs->z_os, MAXNAMELEN, snapname, &id,
 995             &cookie, &case_conflict);
 996         dsl_pool_config_exit(dmu_objset_pool(zfsvfs->z_os), FTAG);
 997         if (error) {
 998                 ZFS_EXIT(zfsvfs);
 999                 if (error == ENOENT) {
1000                         *eofp = 1;
1001                         return (0);
1002                 }
1003                 return (error);
1004         }
1005 
1006         if (flags & V_RDDIR_ENTFLAGS) {
1007                 edirent_t *eodp = dp;
1008 
1009                 (void) strcpy(eodp->ed_name, snapname);
1010                 eodp->ed_ino = ZFSCTL_INO_SNAP(id);
1011                 eodp->ed_eflags = case_conflict ? ED_CASE_CONFLICT : 0;
1012         } else {
1013                 struct dirent64 *odp = dp;
1014 
1015                 (void) strcpy(odp->d_name, snapname);
1016                 odp->d_ino = ZFSCTL_INO_SNAP(id);
1017         }
1018         *nextp = cookie;
1019 
1020         ZFS_EXIT(zfsvfs);
1021 
1022         return (0);
1023 }
1024 
1025 /* ARGSUSED */
1026 static int
1027 zfsctl_shares_readdir(vnode_t *vp, uio_t *uiop, cred_t *cr, int *eofp,
1028     caller_context_t *ct, int flags)
1029 {
1030         zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1031         znode_t *dzp;
1032         int error;
1033 
1034         ZFS_ENTER(zfsvfs);
1035 
1036         if (zfsvfs->z_shares_dir == 0) {
1037                 ZFS_EXIT(zfsvfs);
1038                 return (SET_ERROR(ENOTSUP));
1039         }
1040         if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
1041                 error = VOP_READDIR(ZTOV(dzp), uiop, cr, eofp, ct, flags);
1042                 VN_RELE(ZTOV(dzp));
1043         } else {
1044                 *eofp = 1;
1045                 error = SET_ERROR(ENOENT);
1046         }
1047 
1048         ZFS_EXIT(zfsvfs);
1049         return (error);
1050 }
1051 
1052 /*
1053  * pvp is the '.zfs' directory (zfsctl_node_t).
1054  *
1055  * Creates vp, which is '.zfs/snapshot' (zfsctl_snapdir_t).
1056  *
1057  * This function is the callback to create a GFS vnode for '.zfs/snapshot'
1058  * when a lookup is performed on .zfs for "snapshot".
1059  */
1060 vnode_t *
1061 zfsctl_mknode_snapdir(vnode_t *pvp)
1062 {
1063         vnode_t *vp;
1064         zfsctl_snapdir_t *sdp;
1065 
1066         vp = gfs_dir_create(sizeof (zfsctl_snapdir_t), pvp,
1067             zfsctl_ops_snapdir, NULL, NULL, MAXNAMELEN,
1068             zfsctl_snapdir_readdir_cb, NULL);
1069         sdp = vp->v_data;
1070         sdp->sd_node.zc_id = ZFSCTL_INO_SNAPDIR;
1071         sdp->sd_node.zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime;
1072         mutex_init(&sdp->sd_lock, NULL, MUTEX_DEFAULT, NULL);
1073         avl_create(&sdp->sd_snaps, snapentry_compare,
1074             sizeof (zfs_snapentry_t), offsetof(zfs_snapentry_t, se_node));
1075         return (vp);
1076 }
1077 
1078 vnode_t *
1079 zfsctl_mknode_shares(vnode_t *pvp)
1080 {
1081         vnode_t *vp;
1082         zfsctl_node_t *sdp;
1083 
1084         vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp,
1085             zfsctl_ops_shares, NULL, NULL, MAXNAMELEN,
1086             NULL, NULL);
1087         sdp = vp->v_data;
1088         sdp->zc_cmtime = ((zfsctl_node_t *)pvp->v_data)->zc_cmtime;
1089         return (vp);
1090 
1091 }
1092 
1093 /* ARGSUSED */
1094 static int
1095 zfsctl_shares_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
1096     caller_context_t *ct)
1097 {
1098         zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1099         znode_t *dzp;
1100         int error;
1101 
1102         ZFS_ENTER(zfsvfs);
1103         if (zfsvfs->z_shares_dir == 0) {
1104                 ZFS_EXIT(zfsvfs);
1105                 return (SET_ERROR(ENOTSUP));
1106         }
1107         if ((error = zfs_zget(zfsvfs, zfsvfs->z_shares_dir, &dzp)) == 0) {
1108                 error = VOP_GETATTR(ZTOV(dzp), vap, flags, cr, ct);
1109                 VN_RELE(ZTOV(dzp));
1110         }
1111         ZFS_EXIT(zfsvfs);
1112         return (error);
1113 
1114 
1115 }
1116 
1117 /* ARGSUSED */
1118 static int
1119 zfsctl_snapdir_getattr(vnode_t *vp, vattr_t *vap, int flags, cred_t *cr,
1120     caller_context_t *ct)
1121 {
1122         zfsvfs_t *zfsvfs = vp->v_vfsp->vfs_data;
1123         zfsctl_snapdir_t *sdp = vp->v_data;
1124 
1125         ZFS_ENTER(zfsvfs);
1126         zfsctl_common_getattr(vp, vap);
1127         vap->va_nodeid = gfs_file_inode(vp);
1128         vap->va_nlink = vap->va_size = avl_numnodes(&sdp->sd_snaps) + 2;
1129         vap->va_ctime = vap->va_mtime = dmu_objset_snap_cmtime(zfsvfs->z_os);
1130         ZFS_EXIT(zfsvfs);
1131 
1132         return (0);
1133 }
1134 
1135 /* ARGSUSED */
1136 static void
1137 zfsctl_snapdir_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
1138 {
1139         zfsctl_snapdir_t *sdp = vp->v_data;
1140         void *private;
1141 
1142         private = gfs_dir_inactive(vp);
1143         if (private != NULL) {
1144                 ASSERT(avl_numnodes(&sdp->sd_snaps) == 0);
1145                 mutex_destroy(&sdp->sd_lock);
1146                 avl_destroy(&sdp->sd_snaps);
1147                 kmem_free(private, sizeof (zfsctl_snapdir_t));
1148         }
1149 }
1150 
1151 static const fs_operation_def_t zfsctl_tops_snapdir[] = {
1152         { VOPNAME_OPEN,         { .vop_open = zfsctl_common_open }      },
1153         { VOPNAME_CLOSE,        { .vop_close = zfsctl_common_close }    },
1154         { VOPNAME_IOCTL,        { .error = fs_inval }                   },
1155         { VOPNAME_GETATTR,      { .vop_getattr = zfsctl_snapdir_getattr } },
1156         { VOPNAME_ACCESS,       { .vop_access = zfsctl_common_access }  },
1157         { VOPNAME_RENAME,       { .vop_rename = zfsctl_snapdir_rename } },
1158         { VOPNAME_RMDIR,        { .vop_rmdir = zfsctl_snapdir_remove }  },
1159         { VOPNAME_MKDIR,        { .vop_mkdir = zfsctl_snapdir_mkdir }   },
1160         { VOPNAME_READDIR,      { .vop_readdir = gfs_vop_readdir }      },
1161         { VOPNAME_LOOKUP,       { .vop_lookup = zfsctl_snapdir_lookup } },
1162         { VOPNAME_SEEK,         { .vop_seek = fs_seek }                 },
1163         { VOPNAME_INACTIVE,     { .vop_inactive = zfsctl_snapdir_inactive } },
1164         { VOPNAME_FID,          { .vop_fid = zfsctl_common_fid }        },
1165         { NULL }
1166 };
1167 
1168 static const fs_operation_def_t zfsctl_tops_shares[] = {
1169         { VOPNAME_OPEN,         { .vop_open = zfsctl_common_open }      },
1170         { VOPNAME_CLOSE,        { .vop_close = zfsctl_common_close }    },
1171         { VOPNAME_IOCTL,        { .error = fs_inval }                   },
1172         { VOPNAME_GETATTR,      { .vop_getattr = zfsctl_shares_getattr } },
1173         { VOPNAME_ACCESS,       { .vop_access = zfsctl_common_access }  },
1174         { VOPNAME_READDIR,      { .vop_readdir = zfsctl_shares_readdir } },
1175         { VOPNAME_LOOKUP,       { .vop_lookup = zfsctl_shares_lookup }  },
1176         { VOPNAME_SEEK,         { .vop_seek = fs_seek }                 },
1177         { VOPNAME_INACTIVE,     { .vop_inactive = gfs_vop_inactive } },
1178         { VOPNAME_FID,          { .vop_fid = zfsctl_shares_fid } },
1179         { NULL }
1180 };
1181 
1182 /*
1183  * pvp is the GFS vnode '.zfs/snapshot'.
1184  *
1185  * This creates a GFS node under '.zfs/snapshot' representing each
1186  * snapshot.  This newly created GFS node is what we mount snapshot
1187  * vfs_t's ontop of.
1188  */
1189 static vnode_t *
1190 zfsctl_snapshot_mknode(vnode_t *pvp, uint64_t objset)
1191 {
1192         vnode_t *vp;
1193         zfsctl_node_t *zcp;
1194 
1195         vp = gfs_dir_create(sizeof (zfsctl_node_t), pvp,
1196             zfsctl_ops_snapshot, NULL, NULL, MAXNAMELEN, NULL, NULL);
1197         zcp = vp->v_data;
1198         zcp->zc_id = objset;
1199 
1200         return (vp);
1201 }
1202 
1203 static void
1204 zfsctl_snapshot_inactive(vnode_t *vp, cred_t *cr, caller_context_t *ct)
1205 {
1206         zfsctl_snapdir_t *sdp;
1207         zfs_snapentry_t *sep, *next;
1208         vnode_t *dvp;
1209 
1210         VERIFY(gfs_dir_lookup(vp, "..", &dvp, cr, 0, NULL, NULL) == 0);
1211         sdp = dvp->v_data;
1212 
1213         mutex_enter(&sdp->sd_lock);
1214 
1215         if (vp->v_count > 1) {
1216                 mutex_exit(&sdp->sd_lock);
1217                 return;
1218         }
1219         ASSERT(!vn_ismntpt(vp));
1220 
1221         sep = avl_first(&sdp->sd_snaps);
1222         while (sep != NULL) {
1223                 next = AVL_NEXT(&sdp->sd_snaps, sep);
1224 
1225                 if (sep->se_root == vp) {
1226                         avl_remove(&sdp->sd_snaps, sep);
1227                         kmem_free(sep->se_name, strlen(sep->se_name) + 1);
1228                         kmem_free(sep, sizeof (zfs_snapentry_t));
1229                         break;
1230                 }
1231                 sep = next;
1232         }
1233         ASSERT(sep != NULL);
1234 
1235         mutex_exit(&sdp->sd_lock);
1236         VN_RELE(dvp);
1237 
1238         /*
1239          * Dispose of the vnode for the snapshot mount point.
1240          * This is safe to do because once this entry has been removed
1241          * from the AVL tree, it can't be found again, so cannot become
1242          * "active".  If we lookup the same name again we will end up
1243          * creating a new vnode.
1244          */
1245         gfs_vop_inactive(vp, cr, ct);
1246 }
1247 
1248 
1249 /*
1250  * These VP's should never see the light of day.  They should always
1251  * be covered.
1252  */
1253 static const fs_operation_def_t zfsctl_tops_snapshot[] = {
1254         VOPNAME_INACTIVE, { .vop_inactive =  zfsctl_snapshot_inactive },
1255         NULL, NULL
1256 };
1257 
1258 int
1259 zfsctl_lookup_objset(vfs_t *vfsp, uint64_t objsetid, zfsvfs_t **zfsvfsp)
1260 {
1261         zfsvfs_t *zfsvfs = vfsp->vfs_data;
1262         vnode_t *dvp, *vp;
1263         zfsctl_snapdir_t *sdp;
1264         zfsctl_node_t *zcp;
1265         zfs_snapentry_t *sep;
1266         int error;
1267 
1268         ASSERT(zfsvfs->z_ctldir != NULL);
1269         error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp,
1270             NULL, 0, NULL, kcred, NULL, NULL, NULL);
1271         if (error != 0)
1272                 return (error);
1273         sdp = dvp->v_data;
1274 
1275         mutex_enter(&sdp->sd_lock);
1276         sep = avl_first(&sdp->sd_snaps);
1277         while (sep != NULL) {
1278                 vp = sep->se_root;
1279                 zcp = vp->v_data;
1280                 if (zcp->zc_id == objsetid)
1281                         break;
1282 
1283                 sep = AVL_NEXT(&sdp->sd_snaps, sep);
1284         }
1285 
1286         if (sep != NULL) {
1287                 VN_HOLD(vp);
1288                 /*
1289                  * Return the mounted root rather than the covered mount point.
1290                  * Takes the GFS vnode at .zfs/snapshot/<snapshot objsetid>
1291                  * and returns the ZFS vnode mounted on top of the GFS node.
1292                  * This ZFS vnode is the root of the vfs for objset 'objsetid'.
1293                  */
1294                 error = traverse(&vp);
1295                 if (error == 0) {
1296                         if (vp == sep->se_root)
1297                                 error = SET_ERROR(EINVAL);
1298                         else
1299                                 *zfsvfsp = VTOZ(vp)->z_zfsvfs;
1300                 }
1301                 mutex_exit(&sdp->sd_lock);
1302                 VN_RELE(vp);
1303         } else {
1304                 error = SET_ERROR(EINVAL);
1305                 mutex_exit(&sdp->sd_lock);
1306         }
1307 
1308         VN_RELE(dvp);
1309 
1310         return (error);
1311 }
1312 
1313 /*
1314  * Unmount any snapshots for the given filesystem.  This is called from
1315  * zfs_umount() - if we have a ctldir, then go through and unmount all the
1316  * snapshots.
1317  */
1318 int
1319 zfsctl_umount_snapshots(vfs_t *vfsp, int fflags, cred_t *cr)
1320 {
1321         zfsvfs_t *zfsvfs = vfsp->vfs_data;
1322         vnode_t *dvp;
1323         zfsctl_snapdir_t *sdp;
1324         zfs_snapentry_t *sep, *next;
1325         int error;
1326 
1327         ASSERT(zfsvfs->z_ctldir != NULL);
1328         error = zfsctl_root_lookup(zfsvfs->z_ctldir, "snapshot", &dvp,
1329             NULL, 0, NULL, cr, NULL, NULL, NULL);
1330         if (error != 0)
1331                 return (error);
1332         sdp = dvp->v_data;
1333 
1334         mutex_enter(&sdp->sd_lock);
1335 
1336         sep = avl_first(&sdp->sd_snaps);
1337         while (sep != NULL) {
1338                 next = AVL_NEXT(&sdp->sd_snaps, sep);
1339 
1340                 /*
1341                  * If this snapshot is not mounted, then it must
1342                  * have just been unmounted by somebody else, and
1343                  * will be cleaned up by zfsctl_snapdir_inactive().
1344                  */
1345                 if (vn_ismntpt(sep->se_root)) {
1346                         avl_remove(&sdp->sd_snaps, sep);
1347                         error = zfsctl_unmount_snap(sep, fflags, cr);
1348                         if (error) {
1349                                 avl_add(&sdp->sd_snaps, sep);
1350                                 break;
1351                         }
1352                 }
1353                 sep = next;
1354         }
1355 
1356         mutex_exit(&sdp->sd_lock);
1357         VN_RELE(dvp);
1358 
1359         return (error);
1360 }