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) 2012, 2015 by Delphix. All rights reserved.
  24  * Copyright (c) 2014 Integros [integros.com]
  25  * Copyright 2016 Nexenta Systems, Inc. All rights reserved.
  26  */
  27 
  28 /* Portions Copyright 2010 Robert Milkowski */
  29 
  30 #include <sys/types.h>
  31 #include <sys/param.h>
  32 #include <sys/systm.h>
  33 #include <sys/sysmacros.h>
  34 #include <sys/kmem.h>
  35 #include <sys/pathname.h>
  36 #include <sys/vnode.h>
  37 #include <sys/vfs.h>
  38 #include <sys/vfs_opreg.h>
  39 #include <sys/mntent.h>
  40 #include <sys/mount.h>
  41 #include <sys/cmn_err.h>
  42 #include "fs/fs_subr.h"
  43 #include <sys/zfs_znode.h>
  44 #include <sys/zfs_dir.h>
  45 #include <sys/zil.h>
  46 #include <sys/fs/zfs.h>
  47 #include <sys/dmu.h>
  48 #include <sys/dsl_prop.h>
  49 #include <sys/dsl_dataset.h>
  50 #include <sys/dsl_deleg.h>
  51 #include <sys/spa.h>
  52 #include <sys/zap.h>
  53 #include <sys/sa.h>
  54 #include <sys/sa_impl.h>
  55 #include <sys/varargs.h>
  56 #include <sys/policy.h>
  57 #include <sys/atomic.h>
  58 #include <sys/mkdev.h>
  59 #include <sys/modctl.h>
  60 #include <sys/refstr.h>
  61 #include <sys/zfs_ioctl.h>
  62 #include <sys/zfs_ctldir.h>
  63 #include <sys/zfs_fuid.h>
  64 #include <sys/bootconf.h>
  65 #include <sys/sunddi.h>
  66 #include <sys/dnlc.h>
  67 #include <sys/dmu_objset.h>
  68 #include <sys/spa_boot.h>
  69 #include "zfs_comutil.h"
  70 
  71 int zfsfstype;
  72 vfsops_t *zfs_vfsops = NULL;
  73 static major_t zfs_major;
  74 static minor_t zfs_minor;
  75 static kmutex_t zfs_dev_mtx;
  76 
  77 extern int sys_shutdown;
  78 
  79 static int zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr);
  80 static int zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr);
  81 static int zfs_mountroot(vfs_t *vfsp, enum whymountroot);
  82 static int zfs_root(vfs_t *vfsp, vnode_t **vpp);
  83 static int zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp);
  84 static int zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp);
  85 static void zfs_freevfs(vfs_t *vfsp);
  86 
  87 static const fs_operation_def_t zfs_vfsops_template[] = {
  88         VFSNAME_MOUNT,          { .vfs_mount = zfs_mount },
  89         VFSNAME_MOUNTROOT,      { .vfs_mountroot = zfs_mountroot },
  90         VFSNAME_UNMOUNT,        { .vfs_unmount = zfs_umount },
  91         VFSNAME_ROOT,           { .vfs_root = zfs_root },
  92         VFSNAME_STATVFS,        { .vfs_statvfs = zfs_statvfs },
  93         VFSNAME_SYNC,           { .vfs_sync = zfs_sync },
  94         VFSNAME_VGET,           { .vfs_vget = zfs_vget },
  95         VFSNAME_FREEVFS,        { .vfs_freevfs = zfs_freevfs },
  96         NULL,                   NULL
  97 };
  98 
  99 /*
 100  * We need to keep a count of active fs's.
 101  * This is necessary to prevent our module
 102  * from being unloaded after a umount -f
 103  */
 104 static uint32_t zfs_active_fs_count = 0;
 105 
 106 static char *noatime_cancel[] = { MNTOPT_ATIME, NULL };
 107 static char *atime_cancel[] = { MNTOPT_NOATIME, NULL };
 108 static char *noxattr_cancel[] = { MNTOPT_XATTR, NULL };
 109 static char *xattr_cancel[] = { MNTOPT_NOXATTR, NULL };
 110 
 111 /*
 112  * MO_DEFAULT is not used since the default value is determined
 113  * by the equivalent property.
 114  */
 115 static mntopt_t mntopts[] = {
 116         { MNTOPT_NOXATTR, noxattr_cancel, NULL, 0, NULL },
 117         { MNTOPT_XATTR, xattr_cancel, NULL, 0, NULL },
 118         { MNTOPT_NOATIME, noatime_cancel, NULL, 0, NULL },
 119         { MNTOPT_ATIME, atime_cancel, NULL, 0, NULL }
 120 };
 121 
 122 static mntopts_t zfs_mntopts = {
 123         sizeof (mntopts) / sizeof (mntopt_t),
 124         mntopts
 125 };
 126 
 127 /*ARGSUSED*/
 128 int
 129 zfs_sync(vfs_t *vfsp, short flag, cred_t *cr)
 130 {
 131         /*
 132          * Data integrity is job one.  We don't want a compromised kernel
 133          * writing to the storage pool, so we never sync during panic.
 134          */
 135         if (panicstr)
 136                 return (0);
 137 
 138         /*
 139          * SYNC_ATTR is used by fsflush() to force old filesystems like UFS
 140          * to sync metadata, which they would otherwise cache indefinitely.
 141          * Semantically, the only requirement is that the sync be initiated.
 142          * The DMU syncs out txgs frequently, so there's nothing to do.
 143          */
 144         if (flag & SYNC_ATTR)
 145                 return (0);
 146 
 147         if (vfsp != NULL) {
 148                 /*
 149                  * Sync a specific filesystem.
 150                  */
 151                 zfsvfs_t *zfsvfs = vfsp->vfs_data;
 152                 dsl_pool_t *dp;
 153 
 154                 ZFS_ENTER(zfsvfs);
 155                 dp = dmu_objset_pool(zfsvfs->z_os);
 156 
 157                 /*
 158                  * If the system is shutting down, then skip any
 159                  * filesystems which may exist on a suspended pool.
 160                  */
 161                 if (sys_shutdown && spa_suspended(dp->dp_spa)) {
 162                         ZFS_EXIT(zfsvfs);
 163                         return (0);
 164                 }
 165 
 166                 if (zfsvfs->z_log != NULL)
 167                         zil_commit(zfsvfs->z_log, 0);
 168 
 169                 ZFS_EXIT(zfsvfs);
 170         } else {
 171                 /*
 172                  * Sync all ZFS filesystems.  This is what happens when you
 173                  * run sync(1M).  Unlike other filesystems, ZFS honors the
 174                  * request by waiting for all pools to commit all dirty data.
 175                  */
 176                 spa_sync_allpools();
 177         }
 178 
 179         return (0);
 180 }
 181 
 182 static int
 183 zfs_create_unique_device(dev_t *dev)
 184 {
 185         major_t new_major;
 186 
 187         do {
 188                 ASSERT3U(zfs_minor, <=, MAXMIN32);
 189                 minor_t start = zfs_minor;
 190                 do {
 191                         mutex_enter(&zfs_dev_mtx);
 192                         if (zfs_minor >= MAXMIN32) {
 193                                 /*
 194                                  * If we're still using the real major
 195                                  * keep out of /dev/zfs and /dev/zvol minor
 196                                  * number space.  If we're using a getudev()'ed
 197                                  * major number, we can use all of its minors.
 198                                  */
 199                                 if (zfs_major == ddi_name_to_major(ZFS_DRIVER))
 200                                         zfs_minor = ZFS_MIN_MINOR;
 201                                 else
 202                                         zfs_minor = 0;
 203                         } else {
 204                                 zfs_minor++;
 205                         }
 206                         *dev = makedevice(zfs_major, zfs_minor);
 207                         mutex_exit(&zfs_dev_mtx);
 208                 } while (vfs_devismounted(*dev) && zfs_minor != start);
 209                 if (zfs_minor == start) {
 210                         /*
 211                          * We are using all ~262,000 minor numbers for the
 212                          * current major number.  Create a new major number.
 213                          */
 214                         if ((new_major = getudev()) == (major_t)-1) {
 215                                 cmn_err(CE_WARN,
 216                                     "zfs_mount: Can't get unique major "
 217                                     "device number.");
 218                                 return (-1);
 219                         }
 220                         mutex_enter(&zfs_dev_mtx);
 221                         zfs_major = new_major;
 222                         zfs_minor = 0;
 223 
 224                         mutex_exit(&zfs_dev_mtx);
 225                 } else {
 226                         break;
 227                 }
 228                 /* CONSTANTCONDITION */
 229         } while (1);
 230 
 231         return (0);
 232 }
 233 
 234 static void
 235 atime_changed_cb(void *arg, uint64_t newval)
 236 {
 237         zfsvfs_t *zfsvfs = arg;
 238 
 239         if (newval == TRUE) {
 240                 zfsvfs->z_atime = TRUE;
 241                 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME);
 242                 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_ATIME, NULL, 0);
 243         } else {
 244                 zfsvfs->z_atime = FALSE;
 245                 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_ATIME);
 246                 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOATIME, NULL, 0);
 247         }
 248 }
 249 
 250 static void
 251 xattr_changed_cb(void *arg, uint64_t newval)
 252 {
 253         zfsvfs_t *zfsvfs = arg;
 254 
 255         if (newval == TRUE) {
 256                 /* XXX locking on vfs_flag? */
 257                 zfsvfs->z_vfs->vfs_flag |= VFS_XATTR;
 258                 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR);
 259                 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_XATTR, NULL, 0);
 260         } else {
 261                 /* XXX locking on vfs_flag? */
 262                 zfsvfs->z_vfs->vfs_flag &= ~VFS_XATTR;
 263                 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_XATTR);
 264                 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOXATTR, NULL, 0);
 265         }
 266 }
 267 
 268 static void
 269 blksz_changed_cb(void *arg, uint64_t newval)
 270 {
 271         zfsvfs_t *zfsvfs = arg;
 272         ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os)));
 273         ASSERT3U(newval, >=, SPA_MINBLOCKSIZE);
 274         ASSERT(ISP2(newval));
 275 
 276         zfsvfs->z_max_blksz = newval;
 277         zfsvfs->z_vfs->vfs_bsize = newval;
 278 }
 279 
 280 static void
 281 readonly_changed_cb(void *arg, uint64_t newval)
 282 {
 283         zfsvfs_t *zfsvfs = arg;
 284 
 285         if (newval) {
 286                 /* XXX locking on vfs_flag? */
 287                 zfsvfs->z_vfs->vfs_flag |= VFS_RDONLY;
 288                 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RW);
 289                 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RO, NULL, 0);
 290         } else {
 291                 /* XXX locking on vfs_flag? */
 292                 zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
 293                 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_RO);
 294                 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_RW, NULL, 0);
 295         }
 296 }
 297 
 298 static void
 299 devices_changed_cb(void *arg, uint64_t newval)
 300 {
 301         zfsvfs_t *zfsvfs = arg;
 302 
 303         if (newval == FALSE) {
 304                 zfsvfs->z_vfs->vfs_flag |= VFS_NODEVICES;
 305                 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES);
 306                 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES, NULL, 0);
 307         } else {
 308                 zfsvfs->z_vfs->vfs_flag &= ~VFS_NODEVICES;
 309                 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NODEVICES);
 310                 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_DEVICES, NULL, 0);
 311         }
 312 }
 313 
 314 static void
 315 setuid_changed_cb(void *arg, uint64_t newval)
 316 {
 317         zfsvfs_t *zfsvfs = arg;
 318 
 319         if (newval == FALSE) {
 320                 zfsvfs->z_vfs->vfs_flag |= VFS_NOSETUID;
 321                 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_SETUID);
 322                 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID, NULL, 0);
 323         } else {
 324                 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOSETUID;
 325                 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOSETUID);
 326                 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_SETUID, NULL, 0);
 327         }
 328 }
 329 
 330 static void
 331 exec_changed_cb(void *arg, uint64_t newval)
 332 {
 333         zfsvfs_t *zfsvfs = arg;
 334 
 335         if (newval == FALSE) {
 336                 zfsvfs->z_vfs->vfs_flag |= VFS_NOEXEC;
 337                 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_EXEC);
 338                 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC, NULL, 0);
 339         } else {
 340                 zfsvfs->z_vfs->vfs_flag &= ~VFS_NOEXEC;
 341                 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NOEXEC);
 342                 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_EXEC, NULL, 0);
 343         }
 344 }
 345 
 346 /*
 347  * The nbmand mount option can be changed at mount time.
 348  * We can't allow it to be toggled on live file systems or incorrect
 349  * behavior may be seen from cifs clients
 350  *
 351  * This property isn't registered via dsl_prop_register(), but this callback
 352  * will be called when a file system is first mounted
 353  */
 354 static void
 355 nbmand_changed_cb(void *arg, uint64_t newval)
 356 {
 357         zfsvfs_t *zfsvfs = arg;
 358         if (newval == FALSE) {
 359                 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND);
 360                 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND, NULL, 0);
 361         } else {
 362                 vfs_clearmntopt(zfsvfs->z_vfs, MNTOPT_NONBMAND);
 363                 vfs_setmntopt(zfsvfs->z_vfs, MNTOPT_NBMAND, NULL, 0);
 364         }
 365 }
 366 
 367 static void
 368 snapdir_changed_cb(void *arg, uint64_t newval)
 369 {
 370         zfsvfs_t *zfsvfs = arg;
 371 
 372         zfsvfs->z_show_ctldir = newval;
 373 }
 374 
 375 static void
 376 vscan_changed_cb(void *arg, uint64_t newval)
 377 {
 378         zfsvfs_t *zfsvfs = arg;
 379 
 380         zfsvfs->z_vscan = newval;
 381 }
 382 
 383 static void
 384 acl_mode_changed_cb(void *arg, uint64_t newval)
 385 {
 386         zfsvfs_t *zfsvfs = arg;
 387 
 388         zfsvfs->z_acl_mode = newval;
 389 }
 390 
 391 static void
 392 acl_inherit_changed_cb(void *arg, uint64_t newval)
 393 {
 394         zfsvfs_t *zfsvfs = arg;
 395 
 396         zfsvfs->z_acl_inherit = newval;
 397 }
 398 
 399 static int
 400 zfs_register_callbacks(vfs_t *vfsp)
 401 {
 402         struct dsl_dataset *ds = NULL;
 403         objset_t *os = NULL;
 404         zfsvfs_t *zfsvfs = NULL;
 405         uint64_t nbmand;
 406         boolean_t readonly = B_FALSE;
 407         boolean_t do_readonly = B_FALSE;
 408         boolean_t setuid = B_FALSE;
 409         boolean_t do_setuid = B_FALSE;
 410         boolean_t exec = B_FALSE;
 411         boolean_t do_exec = B_FALSE;
 412         boolean_t devices = B_FALSE;
 413         boolean_t do_devices = B_FALSE;
 414         boolean_t xattr = B_FALSE;
 415         boolean_t do_xattr = B_FALSE;
 416         boolean_t atime = B_FALSE;
 417         boolean_t do_atime = B_FALSE;
 418         int error = 0;
 419 
 420         ASSERT(vfsp);
 421         zfsvfs = vfsp->vfs_data;
 422         ASSERT(zfsvfs);
 423         os = zfsvfs->z_os;
 424 
 425         /*
 426          * The act of registering our callbacks will destroy any mount
 427          * options we may have.  In order to enable temporary overrides
 428          * of mount options, we stash away the current values and
 429          * restore them after we register the callbacks.
 430          */
 431         if (vfs_optionisset(vfsp, MNTOPT_RO, NULL) ||
 432             !spa_writeable(dmu_objset_spa(os))) {
 433                 readonly = B_TRUE;
 434                 do_readonly = B_TRUE;
 435         } else if (vfs_optionisset(vfsp, MNTOPT_RW, NULL)) {
 436                 readonly = B_FALSE;
 437                 do_readonly = B_TRUE;
 438         }
 439         if (vfs_optionisset(vfsp, MNTOPT_NOSUID, NULL)) {
 440                 devices = B_FALSE;
 441                 setuid = B_FALSE;
 442                 do_devices = B_TRUE;
 443                 do_setuid = B_TRUE;
 444         } else {
 445                 if (vfs_optionisset(vfsp, MNTOPT_NODEVICES, NULL)) {
 446                         devices = B_FALSE;
 447                         do_devices = B_TRUE;
 448                 } else if (vfs_optionisset(vfsp, MNTOPT_DEVICES, NULL)) {
 449                         devices = B_TRUE;
 450                         do_devices = B_TRUE;
 451                 }
 452 
 453                 if (vfs_optionisset(vfsp, MNTOPT_NOSETUID, NULL)) {
 454                         setuid = B_FALSE;
 455                         do_setuid = B_TRUE;
 456                 } else if (vfs_optionisset(vfsp, MNTOPT_SETUID, NULL)) {
 457                         setuid = B_TRUE;
 458                         do_setuid = B_TRUE;
 459                 }
 460         }
 461         if (vfs_optionisset(vfsp, MNTOPT_NOEXEC, NULL)) {
 462                 exec = B_FALSE;
 463                 do_exec = B_TRUE;
 464         } else if (vfs_optionisset(vfsp, MNTOPT_EXEC, NULL)) {
 465                 exec = B_TRUE;
 466                 do_exec = B_TRUE;
 467         }
 468         if (vfs_optionisset(vfsp, MNTOPT_NOXATTR, NULL)) {
 469                 xattr = B_FALSE;
 470                 do_xattr = B_TRUE;
 471         } else if (vfs_optionisset(vfsp, MNTOPT_XATTR, NULL)) {
 472                 xattr = B_TRUE;
 473                 do_xattr = B_TRUE;
 474         }
 475         if (vfs_optionisset(vfsp, MNTOPT_NOATIME, NULL)) {
 476                 atime = B_FALSE;
 477                 do_atime = B_TRUE;
 478         } else if (vfs_optionisset(vfsp, MNTOPT_ATIME, NULL)) {
 479                 atime = B_TRUE;
 480                 do_atime = B_TRUE;
 481         }
 482 
 483         /*
 484          * nbmand is a special property.  It can only be changed at
 485          * mount time.
 486          *
 487          * This is weird, but it is documented to only be changeable
 488          * at mount time.
 489          */
 490         if (vfs_optionisset(vfsp, MNTOPT_NONBMAND, NULL)) {
 491                 nbmand = B_FALSE;
 492         } else if (vfs_optionisset(vfsp, MNTOPT_NBMAND, NULL)) {
 493                 nbmand = B_TRUE;
 494         } else {
 495                 char osname[ZFS_MAX_DATASET_NAME_LEN];
 496 
 497                 dmu_objset_name(os, osname);
 498                 if (error = dsl_prop_get_integer(osname, "nbmand", &nbmand,
 499                     NULL)) {
 500                         return (error);
 501                 }
 502         }
 503 
 504         /*
 505          * Register property callbacks.
 506          *
 507          * It would probably be fine to just check for i/o error from
 508          * the first prop_register(), but I guess I like to go
 509          * overboard...
 510          */
 511         ds = dmu_objset_ds(os);
 512         dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
 513         error = dsl_prop_register(ds,
 514             zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
 515         error = error ? error : dsl_prop_register(ds,
 516             zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
 517         error = error ? error : dsl_prop_register(ds,
 518             zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
 519         error = error ? error : dsl_prop_register(ds,
 520             zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
 521         error = error ? error : dsl_prop_register(ds,
 522             zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
 523         error = error ? error : dsl_prop_register(ds,
 524             zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
 525         error = error ? error : dsl_prop_register(ds,
 526             zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
 527         error = error ? error : dsl_prop_register(ds,
 528             zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
 529         error = error ? error : dsl_prop_register(ds,
 530             zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs);
 531         error = error ? error : dsl_prop_register(ds,
 532             zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
 533             zfsvfs);
 534         error = error ? error : dsl_prop_register(ds,
 535             zfs_prop_to_name(ZFS_PROP_VSCAN), vscan_changed_cb, zfsvfs);
 536         dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
 537         if (error)
 538                 goto unregister;
 539 
 540         /*
 541          * Invoke our callbacks to restore temporary mount options.
 542          */
 543         if (do_readonly)
 544                 readonly_changed_cb(zfsvfs, readonly);
 545         if (do_setuid)
 546                 setuid_changed_cb(zfsvfs, setuid);
 547         if (do_exec)
 548                 exec_changed_cb(zfsvfs, exec);
 549         if (do_devices)
 550                 devices_changed_cb(zfsvfs, devices);
 551         if (do_xattr)
 552                 xattr_changed_cb(zfsvfs, xattr);
 553         if (do_atime)
 554                 atime_changed_cb(zfsvfs, atime);
 555 
 556         nbmand_changed_cb(zfsvfs, nbmand);
 557 
 558         return (0);
 559 
 560 unregister:
 561         dsl_prop_unregister_all(ds, zfsvfs);
 562         return (error);
 563 }
 564 
 565 static int
 566 zfs_space_delta_cb(dmu_object_type_t bonustype, void *data,
 567     uint64_t *userp, uint64_t *groupp)
 568 {
 569         /*
 570          * Is it a valid type of object to track?
 571          */
 572         if (bonustype != DMU_OT_ZNODE && bonustype != DMU_OT_SA)
 573                 return (SET_ERROR(ENOENT));
 574 
 575         /*
 576          * If we have a NULL data pointer
 577          * then assume the id's aren't changing and
 578          * return EEXIST to the dmu to let it know to
 579          * use the same ids
 580          */
 581         if (data == NULL)
 582                 return (SET_ERROR(EEXIST));
 583 
 584         if (bonustype == DMU_OT_ZNODE) {
 585                 znode_phys_t *znp = data;
 586                 *userp = znp->zp_uid;
 587                 *groupp = znp->zp_gid;
 588         } else {
 589                 int hdrsize;
 590                 sa_hdr_phys_t *sap = data;
 591                 sa_hdr_phys_t sa = *sap;
 592                 boolean_t swap = B_FALSE;
 593 
 594                 ASSERT(bonustype == DMU_OT_SA);
 595 
 596                 if (sa.sa_magic == 0) {
 597                         /*
 598                          * This should only happen for newly created
 599                          * files that haven't had the znode data filled
 600                          * in yet.
 601                          */
 602                         *userp = 0;
 603                         *groupp = 0;
 604                         return (0);
 605                 }
 606                 if (sa.sa_magic == BSWAP_32(SA_MAGIC)) {
 607                         sa.sa_magic = SA_MAGIC;
 608                         sa.sa_layout_info = BSWAP_16(sa.sa_layout_info);
 609                         swap = B_TRUE;
 610                 } else {
 611                         VERIFY3U(sa.sa_magic, ==, SA_MAGIC);
 612                 }
 613 
 614                 hdrsize = sa_hdrsize(&sa);
 615                 VERIFY3U(hdrsize, >=, sizeof (sa_hdr_phys_t));
 616                 *userp = *((uint64_t *)((uintptr_t)data + hdrsize +
 617                     SA_UID_OFFSET));
 618                 *groupp = *((uint64_t *)((uintptr_t)data + hdrsize +
 619                     SA_GID_OFFSET));
 620                 if (swap) {
 621                         *userp = BSWAP_64(*userp);
 622                         *groupp = BSWAP_64(*groupp);
 623                 }
 624         }
 625         return (0);
 626 }
 627 
 628 static void
 629 fuidstr_to_sid(zfsvfs_t *zfsvfs, const char *fuidstr,
 630     char *domainbuf, int buflen, uid_t *ridp)
 631 {
 632         uint64_t fuid;
 633         const char *domain;
 634 
 635         fuid = zfs_strtonum(fuidstr, NULL);
 636 
 637         domain = zfs_fuid_find_by_idx(zfsvfs, FUID_INDEX(fuid));
 638         if (domain)
 639                 (void) strlcpy(domainbuf, domain, buflen);
 640         else
 641                 domainbuf[0] = '\0';
 642         *ridp = FUID_RID(fuid);
 643 }
 644 
 645 static uint64_t
 646 zfs_userquota_prop_to_obj(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type)
 647 {
 648         switch (type) {
 649         case ZFS_PROP_USERUSED:
 650                 return (DMU_USERUSED_OBJECT);
 651         case ZFS_PROP_GROUPUSED:
 652                 return (DMU_GROUPUSED_OBJECT);
 653         case ZFS_PROP_USERQUOTA:
 654                 return (zfsvfs->z_userquota_obj);
 655         case ZFS_PROP_GROUPQUOTA:
 656                 return (zfsvfs->z_groupquota_obj);
 657         }
 658         return (0);
 659 }
 660 
 661 int
 662 zfs_userspace_many(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
 663     uint64_t *cookiep, void *vbuf, uint64_t *bufsizep)
 664 {
 665         int error;
 666         zap_cursor_t zc;
 667         zap_attribute_t za;
 668         zfs_useracct_t *buf = vbuf;
 669         uint64_t obj;
 670 
 671         if (!dmu_objset_userspace_present(zfsvfs->z_os))
 672                 return (SET_ERROR(ENOTSUP));
 673 
 674         obj = zfs_userquota_prop_to_obj(zfsvfs, type);
 675         if (obj == 0) {
 676                 *bufsizep = 0;
 677                 return (0);
 678         }
 679 
 680         for (zap_cursor_init_serialized(&zc, zfsvfs->z_os, obj, *cookiep);
 681             (error = zap_cursor_retrieve(&zc, &za)) == 0;
 682             zap_cursor_advance(&zc)) {
 683                 if ((uintptr_t)buf - (uintptr_t)vbuf + sizeof (zfs_useracct_t) >
 684                     *bufsizep)
 685                         break;
 686 
 687                 fuidstr_to_sid(zfsvfs, za.za_name,
 688                     buf->zu_domain, sizeof (buf->zu_domain), &buf->zu_rid);
 689 
 690                 buf->zu_space = za.za_first_integer;
 691                 buf++;
 692         }
 693         if (error == ENOENT)
 694                 error = 0;
 695 
 696         ASSERT3U((uintptr_t)buf - (uintptr_t)vbuf, <=, *bufsizep);
 697         *bufsizep = (uintptr_t)buf - (uintptr_t)vbuf;
 698         *cookiep = zap_cursor_serialize(&zc);
 699         zap_cursor_fini(&zc);
 700         return (error);
 701 }
 702 
 703 /*
 704  * buf must be big enough (eg, 32 bytes)
 705  */
 706 static int
 707 id_to_fuidstr(zfsvfs_t *zfsvfs, const char *domain, uid_t rid,
 708     char *buf, boolean_t addok)
 709 {
 710         uint64_t fuid;
 711         int domainid = 0;
 712 
 713         if (domain && domain[0]) {
 714                 domainid = zfs_fuid_find_by_domain(zfsvfs, domain, NULL, addok);
 715                 if (domainid == -1)
 716                         return (SET_ERROR(ENOENT));
 717         }
 718         fuid = FUID_ENCODE(domainid, rid);
 719         (void) sprintf(buf, "%llx", (longlong_t)fuid);
 720         return (0);
 721 }
 722 
 723 int
 724 zfs_userspace_one(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
 725     const char *domain, uint64_t rid, uint64_t *valp)
 726 {
 727         char buf[32];
 728         int err;
 729         uint64_t obj;
 730 
 731         *valp = 0;
 732 
 733         if (!dmu_objset_userspace_present(zfsvfs->z_os))
 734                 return (SET_ERROR(ENOTSUP));
 735 
 736         obj = zfs_userquota_prop_to_obj(zfsvfs, type);
 737         if (obj == 0)
 738                 return (0);
 739 
 740         err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_FALSE);
 741         if (err)
 742                 return (err);
 743 
 744         err = zap_lookup(zfsvfs->z_os, obj, buf, 8, 1, valp);
 745         if (err == ENOENT)
 746                 err = 0;
 747         return (err);
 748 }
 749 
 750 int
 751 zfs_set_userquota(zfsvfs_t *zfsvfs, zfs_userquota_prop_t type,
 752     const char *domain, uint64_t rid, uint64_t quota)
 753 {
 754         char buf[32];
 755         int err;
 756         dmu_tx_t *tx;
 757         uint64_t *objp;
 758         boolean_t fuid_dirtied;
 759 
 760         if (type != ZFS_PROP_USERQUOTA && type != ZFS_PROP_GROUPQUOTA)
 761                 return (SET_ERROR(EINVAL));
 762 
 763         if (zfsvfs->z_version < ZPL_VERSION_USERSPACE)
 764                 return (SET_ERROR(ENOTSUP));
 765 
 766         objp = (type == ZFS_PROP_USERQUOTA) ? &zfsvfs->z_userquota_obj :
 767             &zfsvfs->z_groupquota_obj;
 768 
 769         err = id_to_fuidstr(zfsvfs, domain, rid, buf, B_TRUE);
 770         if (err)
 771                 return (err);
 772         fuid_dirtied = zfsvfs->z_fuid_dirty;
 773 
 774         tx = dmu_tx_create(zfsvfs->z_os);
 775         dmu_tx_hold_zap(tx, *objp ? *objp : DMU_NEW_OBJECT, B_TRUE, NULL);
 776         if (*objp == 0) {
 777                 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
 778                     zfs_userquota_prop_prefixes[type]);
 779         }
 780         if (fuid_dirtied)
 781                 zfs_fuid_txhold(zfsvfs, tx);
 782         err = dmu_tx_assign(tx, TXG_WAIT);
 783         if (err) {
 784                 dmu_tx_abort(tx);
 785                 return (err);
 786         }
 787 
 788         mutex_enter(&zfsvfs->z_lock);
 789         if (*objp == 0) {
 790                 *objp = zap_create(zfsvfs->z_os, DMU_OT_USERGROUP_QUOTA,
 791                     DMU_OT_NONE, 0, tx);
 792                 VERIFY(0 == zap_add(zfsvfs->z_os, MASTER_NODE_OBJ,
 793                     zfs_userquota_prop_prefixes[type], 8, 1, objp, tx));
 794         }
 795         mutex_exit(&zfsvfs->z_lock);
 796 
 797         if (quota == 0) {
 798                 err = zap_remove(zfsvfs->z_os, *objp, buf, tx);
 799                 if (err == ENOENT)
 800                         err = 0;
 801         } else {
 802                 err = zap_update(zfsvfs->z_os, *objp, buf, 8, 1, &quota, tx);
 803         }
 804         ASSERT(err == 0);
 805         if (fuid_dirtied)
 806                 zfs_fuid_sync(zfsvfs, tx);
 807         dmu_tx_commit(tx);
 808         return (err);
 809 }
 810 
 811 boolean_t
 812 zfs_fuid_overquota(zfsvfs_t *zfsvfs, boolean_t isgroup, uint64_t fuid)
 813 {
 814         char buf[32];
 815         uint64_t used, quota, usedobj, quotaobj;
 816         int err;
 817 
 818         usedobj = isgroup ? DMU_GROUPUSED_OBJECT : DMU_USERUSED_OBJECT;
 819         quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
 820 
 821         if (quotaobj == 0 || zfsvfs->z_replay)
 822                 return (B_FALSE);
 823 
 824         (void) sprintf(buf, "%llx", (longlong_t)fuid);
 825         err = zap_lookup(zfsvfs->z_os, quotaobj, buf, 8, 1, &quota);
 826         if (err != 0)
 827                 return (B_FALSE);
 828 
 829         err = zap_lookup(zfsvfs->z_os, usedobj, buf, 8, 1, &used);
 830         if (err != 0)
 831                 return (B_FALSE);
 832         return (used >= quota);
 833 }
 834 
 835 boolean_t
 836 zfs_owner_overquota(zfsvfs_t *zfsvfs, znode_t *zp, boolean_t isgroup)
 837 {
 838         uint64_t fuid;
 839         uint64_t quotaobj;
 840 
 841         quotaobj = isgroup ? zfsvfs->z_groupquota_obj : zfsvfs->z_userquota_obj;
 842 
 843         fuid = isgroup ? zp->z_gid : zp->z_uid;
 844 
 845         if (quotaobj == 0 || zfsvfs->z_replay)
 846                 return (B_FALSE);
 847 
 848         return (zfs_fuid_overquota(zfsvfs, isgroup, fuid));
 849 }
 850 
 851 /*
 852  * Associate this zfsvfs with the given objset, which must be owned.
 853  * This will cache a bunch of on-disk state from the objset in the
 854  * zfsvfs.
 855  */
 856 static int
 857 zfsvfs_init(zfsvfs_t *zfsvfs, objset_t *os)
 858 {
 859         int error;
 860         uint64_t val;
 861 
 862         zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE;
 863         zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
 864         zfsvfs->z_os = os;
 865 
 866         error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
 867         if (error != 0)
 868                 return (error);
 869         if (zfsvfs->z_version >
 870             zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
 871                 (void) printf("Can't mount a version %lld file system "
 872                     "on a version %lld pool\n. Pool must be upgraded to mount "
 873                     "this file system.", (u_longlong_t)zfsvfs->z_version,
 874                     (u_longlong_t)spa_version(dmu_objset_spa(os)));
 875                 return (SET_ERROR(ENOTSUP));
 876         }
 877         error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &val);
 878         if (error != 0)
 879                 return (error);
 880         zfsvfs->z_norm = (int)val;
 881 
 882         error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &val);
 883         if (error != 0)
 884                 return (error);
 885         zfsvfs->z_utf8 = (val != 0);
 886 
 887         error = zfs_get_zplprop(os, ZFS_PROP_CASE, &val);
 888         if (error != 0)
 889                 return (error);
 890         zfsvfs->z_case = (uint_t)val;
 891 
 892         /*
 893          * Fold case on file systems that are always or sometimes case
 894          * insensitive.
 895          */
 896         if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
 897             zfsvfs->z_case == ZFS_CASE_MIXED)
 898                 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
 899 
 900         zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
 901         zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
 902 
 903         uint64_t sa_obj = 0;
 904         if (zfsvfs->z_use_sa) {
 905                 /* should either have both of these objects or none */
 906                 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
 907                     &sa_obj);
 908                 if (error != 0)
 909                         return (error);
 910         }
 911 
 912         error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
 913             &zfsvfs->z_attr_table);
 914         if (error != 0)
 915                 return (error);
 916 
 917         if (zfsvfs->z_version >= ZPL_VERSION_SA)
 918                 sa_register_update_callback(os, zfs_sa_upgrade);
 919 
 920         error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
 921             &zfsvfs->z_root);
 922         if (error != 0)
 923                 return (error);
 924         ASSERT(zfsvfs->z_root != 0);
 925 
 926         error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
 927             &zfsvfs->z_unlinkedobj);
 928         if (error != 0)
 929                 return (error);
 930 
 931         error = zap_lookup(os, MASTER_NODE_OBJ,
 932             zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
 933             8, 1, &zfsvfs->z_userquota_obj);
 934         if (error == ENOENT)
 935                 zfsvfs->z_userquota_obj = 0;
 936         else if (error != 0)
 937                 return (error);
 938 
 939         error = zap_lookup(os, MASTER_NODE_OBJ,
 940             zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
 941             8, 1, &zfsvfs->z_groupquota_obj);
 942         if (error == ENOENT)
 943                 zfsvfs->z_groupquota_obj = 0;
 944         else if (error != 0)
 945                 return (error);
 946 
 947         error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
 948             &zfsvfs->z_fuid_obj);
 949         if (error == ENOENT)
 950                 zfsvfs->z_fuid_obj = 0;
 951         else if (error != 0)
 952                 return (error);
 953 
 954         error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
 955             &zfsvfs->z_shares_dir);
 956         if (error == ENOENT)
 957                 zfsvfs->z_shares_dir = 0;
 958         else if (error != 0)
 959                 return (error);
 960 
 961         return (0);
 962 }
 963 
 964 int
 965 zfsvfs_create(const char *osname, zfsvfs_t **zfvp)
 966 {
 967         objset_t *os;
 968         zfsvfs_t *zfsvfs;
 969         int error;
 970 
 971         zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
 972 
 973         /*
 974          * We claim to always be readonly so we can open snapshots;
 975          * other ZPL code will prevent us from writing to snapshots.
 976          */
 977 
 978         error = dmu_objset_own(osname, DMU_OST_ZFS, B_TRUE, zfsvfs, &os);
 979         if (error != 0) {
 980                 kmem_free(zfsvfs, sizeof (zfsvfs_t));
 981                 return (error);
 982         }
 983 
 984         error = zfsvfs_create_impl(zfvp, zfsvfs, os);
 985         if (error != 0) {
 986                 dmu_objset_disown(os, zfsvfs);
 987         }
 988         return (error);
 989 }
 990 
 991 
 992 int
 993 zfsvfs_create_impl(zfsvfs_t **zfvp, zfsvfs_t *zfsvfs, objset_t *os)
 994 {
 995         int error;
 996 
 997         zfsvfs->z_vfs = NULL;
 998         zfsvfs->z_parent = zfsvfs;
 999 
1000         mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
1001         mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
1002         list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
1003             offsetof(znode_t, z_link_node));
1004         rrm_init(&zfsvfs->z_teardown_lock, B_FALSE);
1005         rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
1006         rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
1007         for (int i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1008                 mutex_init(&zfsvfs->z_hold_mtx[i], NULL, MUTEX_DEFAULT, NULL);
1009 
1010         error = zfsvfs_init(zfsvfs, os);
1011         if (error != 0) {
1012                 *zfvp = NULL;
1013                 kmem_free(zfsvfs, sizeof (zfsvfs_t));
1014                 return (error);
1015         }
1016 
1017         *zfvp = zfsvfs;
1018         return (0);
1019 }
1020 
1021 static int
1022 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
1023 {
1024         int error;
1025 
1026         error = zfs_register_callbacks(zfsvfs->z_vfs);
1027         if (error)
1028                 return (error);
1029 
1030         zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
1031 
1032         /*
1033          * If we are not mounting (ie: online recv), then we don't
1034          * have to worry about replaying the log as we blocked all
1035          * operations out since we closed the ZIL.
1036          */
1037         if (mounting) {
1038                 boolean_t readonly;
1039 
1040                 /*
1041                  * During replay we remove the read only flag to
1042                  * allow replays to succeed.
1043                  */
1044                 readonly = zfsvfs->z_vfs->vfs_flag & VFS_RDONLY;
1045                 if (readonly != 0)
1046                         zfsvfs->z_vfs->vfs_flag &= ~VFS_RDONLY;
1047                 else
1048                         zfs_unlinked_drain(zfsvfs);
1049 
1050                 /*
1051                  * Parse and replay the intent log.
1052                  *
1053                  * Because of ziltest, this must be done after
1054                  * zfs_unlinked_drain().  (Further note: ziltest
1055                  * doesn't use readonly mounts, where
1056                  * zfs_unlinked_drain() isn't called.)  This is because
1057                  * ziltest causes spa_sync() to think it's committed,
1058                  * but actually it is not, so the intent log contains
1059                  * many txg's worth of changes.
1060                  *
1061                  * In particular, if object N is in the unlinked set in
1062                  * the last txg to actually sync, then it could be
1063                  * actually freed in a later txg and then reallocated
1064                  * in a yet later txg.  This would write a "create
1065                  * object N" record to the intent log.  Normally, this
1066                  * would be fine because the spa_sync() would have
1067                  * written out the fact that object N is free, before
1068                  * we could write the "create object N" intent log
1069                  * record.
1070                  *
1071                  * But when we are in ziltest mode, we advance the "open
1072                  * txg" without actually spa_sync()-ing the changes to
1073                  * disk.  So we would see that object N is still
1074                  * allocated and in the unlinked set, and there is an
1075                  * intent log record saying to allocate it.
1076                  */
1077                 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
1078                         if (zil_replay_disable) {
1079                                 zil_destroy(zfsvfs->z_log, B_FALSE);
1080                         } else {
1081                                 zfsvfs->z_replay = B_TRUE;
1082                                 zil_replay(zfsvfs->z_os, zfsvfs,
1083                                     zfs_replay_vector);
1084                                 zfsvfs->z_replay = B_FALSE;
1085                         }
1086                 }
1087                 zfsvfs->z_vfs->vfs_flag |= readonly; /* restore readonly bit */
1088         }
1089 
1090         /*
1091          * Set the objset user_ptr to track its zfsvfs.
1092          */
1093         mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1094         dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1095         mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1096 
1097         return (0);
1098 }
1099 
1100 void
1101 zfsvfs_free(zfsvfs_t *zfsvfs)
1102 {
1103         int i;
1104         extern krwlock_t zfsvfs_lock; /* in zfs_znode.c */
1105 
1106         /*
1107          * This is a barrier to prevent the filesystem from going away in
1108          * zfs_znode_move() until we can safely ensure that the filesystem is
1109          * not unmounted. We consider the filesystem valid before the barrier
1110          * and invalid after the barrier.
1111          */
1112         rw_enter(&zfsvfs_lock, RW_READER);
1113         rw_exit(&zfsvfs_lock);
1114 
1115         zfs_fuid_destroy(zfsvfs);
1116 
1117         mutex_destroy(&zfsvfs->z_znodes_lock);
1118         mutex_destroy(&zfsvfs->z_lock);
1119         list_destroy(&zfsvfs->z_all_znodes);
1120         rrm_destroy(&zfsvfs->z_teardown_lock);
1121         rw_destroy(&zfsvfs->z_teardown_inactive_lock);
1122         rw_destroy(&zfsvfs->z_fuid_lock);
1123         for (i = 0; i != ZFS_OBJ_MTX_SZ; i++)
1124                 mutex_destroy(&zfsvfs->z_hold_mtx[i]);
1125         kmem_free(zfsvfs, sizeof (zfsvfs_t));
1126 }
1127 
1128 static void
1129 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
1130 {
1131         zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
1132         if (zfsvfs->z_vfs) {
1133                 if (zfsvfs->z_use_fuids) {
1134                         vfs_set_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1135                         vfs_set_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1136                         vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1137                         vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1138                         vfs_set_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1139                         vfs_set_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1140                 } else {
1141                         vfs_clear_feature(zfsvfs->z_vfs, VFSFT_XVATTR);
1142                         vfs_clear_feature(zfsvfs->z_vfs, VFSFT_SYSATTR_VIEWS);
1143                         vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACEMASKONACCESS);
1144                         vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACLONCREATE);
1145                         vfs_clear_feature(zfsvfs->z_vfs, VFSFT_ACCESS_FILTER);
1146                         vfs_clear_feature(zfsvfs->z_vfs, VFSFT_REPARSE);
1147                 }
1148         }
1149         zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
1150 }
1151 
1152 static int
1153 zfs_domount(vfs_t *vfsp, char *osname)
1154 {
1155         dev_t mount_dev;
1156         uint64_t recordsize, fsid_guid;
1157         int error = 0;
1158         zfsvfs_t *zfsvfs;
1159 
1160         ASSERT(vfsp);
1161         ASSERT(osname);
1162 
1163         error = zfsvfs_create(osname, &zfsvfs);
1164         if (error)
1165                 return (error);
1166         zfsvfs->z_vfs = vfsp;
1167 
1168         /* Initialize the generic filesystem structure. */
1169         vfsp->vfs_bcount = 0;
1170         vfsp->vfs_data = NULL;
1171 
1172         if (zfs_create_unique_device(&mount_dev) == -1) {
1173                 error = SET_ERROR(ENODEV);
1174                 goto out;
1175         }
1176         ASSERT(vfs_devismounted(mount_dev) == 0);
1177 
1178         if (error = dsl_prop_get_integer(osname, "recordsize", &recordsize,
1179             NULL))
1180                 goto out;
1181 
1182         vfsp->vfs_dev = mount_dev;
1183         vfsp->vfs_fstype = zfsfstype;
1184         vfsp->vfs_bsize = recordsize;
1185         vfsp->vfs_flag |= VFS_NOTRUNC;
1186         vfsp->vfs_data = zfsvfs;
1187 
1188         /*
1189          * The fsid is 64 bits, composed of an 8-bit fs type, which
1190          * separates our fsid from any other filesystem types, and a
1191          * 56-bit objset unique ID.  The objset unique ID is unique to
1192          * all objsets open on this system, provided by unique_create().
1193          * The 8-bit fs type must be put in the low bits of fsid[1]
1194          * because that's where other Solaris filesystems put it.
1195          */
1196         fsid_guid = dmu_objset_fsid_guid(zfsvfs->z_os);
1197         ASSERT((fsid_guid & ~((1ULL<<56)-1)) == 0);
1198         vfsp->vfs_fsid.val[0] = fsid_guid;
1199         vfsp->vfs_fsid.val[1] = ((fsid_guid>>32) << 8) |
1200             zfsfstype & 0xFF;
1201 
1202         /*
1203          * Set features for file system.
1204          */
1205         zfs_set_fuid_feature(zfsvfs);
1206         if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE) {
1207                 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1208                 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1209                 vfs_set_feature(vfsp, VFSFT_NOCASESENSITIVE);
1210         } else if (zfsvfs->z_case == ZFS_CASE_MIXED) {
1211                 vfs_set_feature(vfsp, VFSFT_DIRENTFLAGS);
1212                 vfs_set_feature(vfsp, VFSFT_CASEINSENSITIVE);
1213         }
1214         vfs_set_feature(vfsp, VFSFT_ZEROCOPY_SUPPORTED);
1215 
1216         if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1217                 uint64_t pval;
1218 
1219                 atime_changed_cb(zfsvfs, B_FALSE);
1220                 readonly_changed_cb(zfsvfs, B_TRUE);
1221                 if (error = dsl_prop_get_integer(osname, "xattr", &pval, NULL))
1222                         goto out;
1223                 xattr_changed_cb(zfsvfs, pval);
1224                 zfsvfs->z_issnap = B_TRUE;
1225                 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1226 
1227                 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1228                 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1229                 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1230         } else {
1231                 error = zfsvfs_setup(zfsvfs, B_TRUE);
1232         }
1233 
1234         if (!zfsvfs->z_issnap)
1235                 zfsctl_create(zfsvfs);
1236 out:
1237         if (error) {
1238                 dmu_objset_disown(zfsvfs->z_os, zfsvfs);
1239                 zfsvfs_free(zfsvfs);
1240         } else {
1241                 atomic_inc_32(&zfs_active_fs_count);
1242         }
1243 
1244         return (error);
1245 }
1246 
1247 void
1248 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
1249 {
1250         objset_t *os = zfsvfs->z_os;
1251 
1252         if (!dmu_objset_is_snapshot(os))
1253                 dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs);
1254 }
1255 
1256 /*
1257  * Convert a decimal digit string to a uint64_t integer.
1258  */
1259 static int
1260 str_to_uint64(char *str, uint64_t *objnum)
1261 {
1262         uint64_t num = 0;
1263 
1264         while (*str) {
1265                 if (*str < '0' || *str > '9')
1266                         return (SET_ERROR(EINVAL));
1267 
1268                 num = num*10 + *str++ - '0';
1269         }
1270 
1271         *objnum = num;
1272         return (0);
1273 }
1274 
1275 /*
1276  * The boot path passed from the boot loader is in the form of
1277  * "rootpool-name/root-filesystem-object-number'. Convert this
1278  * string to a dataset name: "rootpool-name/root-filesystem-name".
1279  */
1280 static int
1281 zfs_parse_bootfs(char *bpath, char *outpath)
1282 {
1283         char *slashp;
1284         uint64_t objnum;
1285         int error;
1286 
1287         if (*bpath == 0 || *bpath == '/')
1288                 return (SET_ERROR(EINVAL));
1289 
1290         (void) strcpy(outpath, bpath);
1291 
1292         slashp = strchr(bpath, '/');
1293 
1294         /* if no '/', just return the pool name */
1295         if (slashp == NULL) {
1296                 return (0);
1297         }
1298 
1299         /* if not a number, just return the root dataset name */
1300         if (str_to_uint64(slashp+1, &objnum)) {
1301                 return (0);
1302         }
1303 
1304         *slashp = '\0';
1305         error = dsl_dsobj_to_dsname(bpath, objnum, outpath);
1306         *slashp = '/';
1307 
1308         return (error);
1309 }
1310 
1311 /*
1312  * Check that the hex label string is appropriate for the dataset being
1313  * mounted into the global_zone proper.
1314  *
1315  * Return an error if the hex label string is not default or
1316  * admin_low/admin_high.  For admin_low labels, the corresponding
1317  * dataset must be readonly.
1318  */
1319 int
1320 zfs_check_global_label(const char *dsname, const char *hexsl)
1321 {
1322         if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1323                 return (0);
1324         if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
1325                 return (0);
1326         if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
1327                 /* must be readonly */
1328                 uint64_t rdonly;
1329 
1330                 if (dsl_prop_get_integer(dsname,
1331                     zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1332                         return (SET_ERROR(EACCES));
1333                 return (rdonly ? 0 : EACCES);
1334         }
1335         return (SET_ERROR(EACCES));
1336 }
1337 
1338 /*
1339  * Determine whether the mount is allowed according to MAC check.
1340  * by comparing (where appropriate) label of the dataset against
1341  * the label of the zone being mounted into.  If the dataset has
1342  * no label, create one.
1343  *
1344  * Returns 0 if access allowed, error otherwise (e.g. EACCES)
1345  */
1346 static int
1347 zfs_mount_label_policy(vfs_t *vfsp, char *osname)
1348 {
1349         int             error, retv;
1350         zone_t          *mntzone = NULL;
1351         ts_label_t      *mnt_tsl;
1352         bslabel_t       *mnt_sl;
1353         bslabel_t       ds_sl;
1354         char            ds_hexsl[MAXNAMELEN];
1355 
1356         retv = EACCES;                          /* assume the worst */
1357 
1358         /*
1359          * Start by getting the dataset label if it exists.
1360          */
1361         error = dsl_prop_get(osname, zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1362             1, sizeof (ds_hexsl), &ds_hexsl, NULL);
1363         if (error)
1364                 return (SET_ERROR(EACCES));
1365 
1366         /*
1367          * If labeling is NOT enabled, then disallow the mount of datasets
1368          * which have a non-default label already.  No other label checks
1369          * are needed.
1370          */
1371         if (!is_system_labeled()) {
1372                 if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
1373                         return (0);
1374                 return (SET_ERROR(EACCES));
1375         }
1376 
1377         /*
1378          * Get the label of the mountpoint.  If mounting into the global
1379          * zone (i.e. mountpoint is not within an active zone and the
1380          * zoned property is off), the label must be default or
1381          * admin_low/admin_high only; no other checks are needed.
1382          */
1383         mntzone = zone_find_by_any_path(refstr_value(vfsp->vfs_mntpt), B_FALSE);
1384         if (mntzone->zone_id == GLOBAL_ZONEID) {
1385                 uint64_t zoned;
1386 
1387                 zone_rele(mntzone);
1388 
1389                 if (dsl_prop_get_integer(osname,
1390                     zfs_prop_to_name(ZFS_PROP_ZONED), &zoned, NULL))
1391                         return (SET_ERROR(EACCES));
1392                 if (!zoned)
1393                         return (zfs_check_global_label(osname, ds_hexsl));
1394                 else
1395                         /*
1396                          * This is the case of a zone dataset being mounted
1397                          * initially, before the zone has been fully created;
1398                          * allow this mount into global zone.
1399                          */
1400                         return (0);
1401         }
1402 
1403         mnt_tsl = mntzone->zone_slabel;
1404         ASSERT(mnt_tsl != NULL);
1405         label_hold(mnt_tsl);
1406         mnt_sl = label2bslabel(mnt_tsl);
1407 
1408         if (strcasecmp(ds_hexsl, ZFS_MLSLABEL_DEFAULT) == 0) {
1409                 /*
1410                  * The dataset doesn't have a real label, so fabricate one.
1411                  */
1412                 char *str = NULL;
1413 
1414                 if (l_to_str_internal(mnt_sl, &str) == 0 &&
1415                     dsl_prop_set_string(osname,
1416                     zfs_prop_to_name(ZFS_PROP_MLSLABEL),
1417                     ZPROP_SRC_LOCAL, str) == 0)
1418                         retv = 0;
1419                 if (str != NULL)
1420                         kmem_free(str, strlen(str) + 1);
1421         } else if (hexstr_to_label(ds_hexsl, &ds_sl) == 0) {
1422                 /*
1423                  * Now compare labels to complete the MAC check.  If the
1424                  * labels are equal then allow access.  If the mountpoint
1425                  * label dominates the dataset label, allow readonly access.
1426                  * Otherwise, access is denied.
1427                  */
1428                 if (blequal(mnt_sl, &ds_sl))
1429                         retv = 0;
1430                 else if (bldominates(mnt_sl, &ds_sl)) {
1431                         vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
1432                         retv = 0;
1433                 }
1434         }
1435 
1436         label_rele(mnt_tsl);
1437         zone_rele(mntzone);
1438         return (retv);
1439 }
1440 
1441 static int
1442 zfs_mountroot(vfs_t *vfsp, enum whymountroot why)
1443 {
1444         int error = 0;
1445         static int zfsrootdone = 0;
1446         zfsvfs_t *zfsvfs = NULL;
1447         znode_t *zp = NULL;
1448         vnode_t *vp = NULL;
1449         char *zfs_bootfs;
1450         char *zfs_devid;
1451 
1452         ASSERT(vfsp);
1453 
1454         /*
1455          * The filesystem that we mount as root is defined in the
1456          * boot property "zfs-bootfs" with a format of
1457          * "poolname/root-dataset-objnum".
1458          */
1459         if (why == ROOT_INIT) {
1460                 if (zfsrootdone++)
1461                         return (SET_ERROR(EBUSY));
1462                 /*
1463                  * the process of doing a spa_load will require the
1464                  * clock to be set before we could (for example) do
1465                  * something better by looking at the timestamp on
1466                  * an uberblock, so just set it to -1.
1467                  */
1468                 clkset(-1);
1469 
1470                 if ((zfs_bootfs = spa_get_bootprop("zfs-bootfs")) == NULL) {
1471                         cmn_err(CE_NOTE, "spa_get_bootfs: can not get "
1472                             "bootfs name");
1473                         return (SET_ERROR(EINVAL));
1474                 }
1475                 zfs_devid = spa_get_bootprop("diskdevid");
1476                 error = spa_import_rootpool(rootfs.bo_name, zfs_devid);
1477                 if (zfs_devid)
1478                         spa_free_bootprop(zfs_devid);
1479                 if (error) {
1480                         spa_free_bootprop(zfs_bootfs);
1481                         cmn_err(CE_NOTE, "spa_import_rootpool: error %d",
1482                             error);
1483                         return (error);
1484                 }
1485                 if (error = zfs_parse_bootfs(zfs_bootfs, rootfs.bo_name)) {
1486                         spa_free_bootprop(zfs_bootfs);
1487                         cmn_err(CE_NOTE, "zfs_parse_bootfs: error %d",
1488                             error);
1489                         return (error);
1490                 }
1491 
1492                 spa_free_bootprop(zfs_bootfs);
1493 
1494                 if (error = vfs_lock(vfsp))
1495                         return (error);
1496 
1497                 if (error = zfs_domount(vfsp, rootfs.bo_name)) {
1498                         cmn_err(CE_NOTE, "zfs_domount: error %d", error);
1499                         goto out;
1500                 }
1501 
1502                 zfsvfs = (zfsvfs_t *)vfsp->vfs_data;
1503                 ASSERT(zfsvfs);
1504                 if (error = zfs_zget(zfsvfs, zfsvfs->z_root, &zp)) {
1505                         cmn_err(CE_NOTE, "zfs_zget: error %d", error);
1506                         goto out;
1507                 }
1508 
1509                 vp = ZTOV(zp);
1510                 mutex_enter(&vp->v_lock);
1511                 vp->v_flag |= VROOT;
1512                 mutex_exit(&vp->v_lock);
1513                 rootvp = vp;
1514 
1515                 /*
1516                  * Leave rootvp held.  The root file system is never unmounted.
1517                  */
1518 
1519                 vfs_add((struct vnode *)0, vfsp,
1520                     (vfsp->vfs_flag & VFS_RDONLY) ? MS_RDONLY : 0);
1521 out:
1522                 vfs_unlock(vfsp);
1523                 return (error);
1524         } else if (why == ROOT_REMOUNT) {
1525                 readonly_changed_cb(vfsp->vfs_data, B_FALSE);
1526                 vfsp->vfs_flag |= VFS_REMOUNT;
1527 
1528                 /* refresh mount options */
1529                 zfs_unregister_callbacks(vfsp->vfs_data);
1530                 return (zfs_register_callbacks(vfsp));
1531 
1532         } else if (why == ROOT_UNMOUNT) {
1533                 zfs_unregister_callbacks((zfsvfs_t *)vfsp->vfs_data);
1534                 (void) zfs_sync(vfsp, 0, 0);
1535                 return (0);
1536         }
1537 
1538         /*
1539          * if "why" is equal to anything else other than ROOT_INIT,
1540          * ROOT_REMOUNT, or ROOT_UNMOUNT, we do not support it.
1541          */
1542         return (SET_ERROR(ENOTSUP));
1543 }
1544 
1545 /*ARGSUSED*/
1546 static int
1547 zfs_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
1548 {
1549         char            *osname;
1550         pathname_t      spn;
1551         int             error = 0;
1552         uio_seg_t       fromspace = (uap->flags & MS_SYSSPACE) ?
1553             UIO_SYSSPACE : UIO_USERSPACE;
1554         int             canwrite;
1555 
1556         if (mvp->v_type != VDIR)
1557                 return (SET_ERROR(ENOTDIR));
1558 
1559         mutex_enter(&mvp->v_lock);
1560         if ((uap->flags & MS_REMOUNT) == 0 &&
1561             (uap->flags & MS_OVERLAY) == 0 &&
1562             (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
1563                 mutex_exit(&mvp->v_lock);
1564                 return (SET_ERROR(EBUSY));
1565         }
1566         mutex_exit(&mvp->v_lock);
1567 
1568         /*
1569          * ZFS does not support passing unparsed data in via MS_DATA.
1570          * Users should use the MS_OPTIONSTR interface; this means
1571          * that all option parsing is already done and the options struct
1572          * can be interrogated.
1573          */
1574         if ((uap->flags & MS_DATA) && uap->datalen > 0)
1575                 return (SET_ERROR(EINVAL));
1576 
1577         /*
1578          * Get the objset name (the "special" mount argument).
1579          */
1580         if (error = pn_get(uap->spec, fromspace, &spn))
1581                 return (error);
1582 
1583         osname = spn.pn_path;
1584 
1585         /*
1586          * Check for mount privilege?
1587          *
1588          * If we don't have privilege then see if
1589          * we have local permission to allow it
1590          */
1591         error = secpolicy_fs_mount(cr, mvp, vfsp);
1592         if (error) {
1593                 if (dsl_deleg_access(osname, ZFS_DELEG_PERM_MOUNT, cr) == 0) {
1594                         vattr_t         vattr;
1595 
1596                         /*
1597                          * Make sure user is the owner of the mount point
1598                          * or has sufficient privileges.
1599                          */
1600 
1601                         vattr.va_mask = AT_UID;
1602 
1603                         if (VOP_GETATTR(mvp, &vattr, 0, cr, NULL)) {
1604                                 goto out;
1605                         }
1606 
1607                         if (secpolicy_vnode_owner(cr, vattr.va_uid) != 0 &&
1608                             VOP_ACCESS(mvp, VWRITE, 0, cr, NULL) != 0) {
1609                                 goto out;
1610                         }
1611                         secpolicy_fs_mount_clearopts(cr, vfsp);
1612                 } else {
1613                         goto out;
1614                 }
1615         }
1616 
1617         /*
1618          * Refuse to mount a filesystem if we are in a local zone and the
1619          * dataset is not visible.
1620          */
1621         if (!INGLOBALZONE(curproc) &&
1622             (!zone_dataset_visible(osname, &canwrite) || !canwrite)) {
1623                 error = SET_ERROR(EPERM);
1624                 goto out;
1625         }
1626 
1627         error = zfs_mount_label_policy(vfsp, osname);
1628         if (error)
1629                 goto out;
1630 
1631         /*
1632          * When doing a remount, we simply refresh our temporary properties
1633          * according to those options set in the current VFS options.
1634          */
1635         if (uap->flags & MS_REMOUNT) {
1636                 /* refresh mount options */
1637                 zfs_unregister_callbacks(vfsp->vfs_data);
1638                 error = zfs_register_callbacks(vfsp);
1639                 goto out;
1640         }
1641 
1642         error = zfs_domount(vfsp, osname);
1643 
1644         /*
1645          * Add an extra VFS_HOLD on our parent vfs so that it can't
1646          * disappear due to a forced unmount.
1647          */
1648         if (error == 0 && ((zfsvfs_t *)vfsp->vfs_data)->z_issnap)
1649                 VFS_HOLD(mvp->v_vfsp);
1650 
1651 out:
1652         pn_free(&spn);
1653         return (error);
1654 }
1655 
1656 static int
1657 zfs_statvfs(vfs_t *vfsp, struct statvfs64 *statp)
1658 {
1659         zfsvfs_t *zfsvfs = vfsp->vfs_data;
1660         dev32_t d32;
1661         uint64_t refdbytes, availbytes, usedobjs, availobjs;
1662 
1663         ZFS_ENTER(zfsvfs);
1664 
1665         dmu_objset_space(zfsvfs->z_os,
1666             &refdbytes, &availbytes, &usedobjs, &availobjs);
1667 
1668         /*
1669          * The underlying storage pool actually uses multiple block sizes.
1670          * We report the fragsize as the smallest block size we support,
1671          * and we report our blocksize as the filesystem's maximum blocksize.
1672          */
1673         statp->f_frsize = 1UL << SPA_MINBLOCKSHIFT;
1674         statp->f_bsize = zfsvfs->z_max_blksz;
1675 
1676         /*
1677          * The following report "total" blocks of various kinds in the
1678          * file system, but reported in terms of f_frsize - the
1679          * "fragment" size.
1680          */
1681 
1682         statp->f_blocks = (refdbytes + availbytes) >> SPA_MINBLOCKSHIFT;
1683         statp->f_bfree = availbytes >> SPA_MINBLOCKSHIFT;
1684         statp->f_bavail = statp->f_bfree; /* no root reservation */
1685 
1686         /*
1687          * statvfs() should really be called statufs(), because it assumes
1688          * static metadata.  ZFS doesn't preallocate files, so the best
1689          * we can do is report the max that could possibly fit in f_files,
1690          * and that minus the number actually used in f_ffree.
1691          * For f_ffree, report the smaller of the number of object available
1692          * and the number of blocks (each object will take at least a block).
1693          */
1694         statp->f_ffree = MIN(availobjs, statp->f_bfree);
1695         statp->f_favail = statp->f_ffree; /* no "root reservation" */
1696         statp->f_files = statp->f_ffree + usedobjs;
1697 
1698         (void) cmpldev(&d32, vfsp->vfs_dev);
1699         statp->f_fsid = d32;
1700 
1701         /*
1702          * We're a zfs filesystem.
1703          */
1704         (void) strcpy(statp->f_basetype, vfssw[vfsp->vfs_fstype].vsw_name);
1705 
1706         statp->f_flag = vf_to_stf(vfsp->vfs_flag);
1707 
1708         statp->f_namemax = MAXNAMELEN - 1;
1709 
1710         /*
1711          * We have all of 32 characters to stuff a string here.
1712          * Is there anything useful we could/should provide?
1713          */
1714         bzero(statp->f_fstr, sizeof (statp->f_fstr));
1715 
1716         ZFS_EXIT(zfsvfs);
1717         return (0);
1718 }
1719 
1720 static int
1721 zfs_root(vfs_t *vfsp, vnode_t **vpp)
1722 {
1723         zfsvfs_t *zfsvfs = vfsp->vfs_data;
1724         znode_t *rootzp;
1725         int error;
1726 
1727         ZFS_ENTER(zfsvfs);
1728 
1729         error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1730         if (error == 0)
1731                 *vpp = ZTOV(rootzp);
1732 
1733         ZFS_EXIT(zfsvfs);
1734         return (error);
1735 }
1736 
1737 /*
1738  * Teardown the zfsvfs::z_os.
1739  *
1740  * Note, if 'unmounting' is FALSE, we return with the 'z_teardown_lock'
1741  * and 'z_teardown_inactive_lock' held.
1742  */
1743 static int
1744 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1745 {
1746         znode_t *zp;
1747 
1748         rrm_enter(&zfsvfs->z_teardown_lock, RW_WRITER, FTAG);
1749 
1750         if (!unmounting) {
1751                 /*
1752                  * We purge the parent filesystem's vfsp as the parent
1753                  * filesystem and all of its snapshots have their vnode's
1754                  * v_vfsp set to the parent's filesystem's vfsp.  Note,
1755                  * 'z_parent' is self referential for non-snapshots.
1756                  */
1757                 (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1758         }
1759 
1760         /*
1761          * Close the zil. NB: Can't close the zil while zfs_inactive
1762          * threads are blocked as zil_close can call zfs_inactive.
1763          */
1764         if (zfsvfs->z_log) {
1765                 zil_close(zfsvfs->z_log);
1766                 zfsvfs->z_log = NULL;
1767         }
1768 
1769         rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1770 
1771         /*
1772          * If we are not unmounting (ie: online recv) and someone already
1773          * unmounted this file system while we were doing the switcheroo,
1774          * or a reopen of z_os failed then just bail out now.
1775          */
1776         if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1777                 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1778                 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1779                 return (SET_ERROR(EIO));
1780         }
1781 
1782         /*
1783          * At this point there are no vops active, and any new vops will
1784          * fail with EIO since we have z_teardown_lock for writer (only
1785          * relavent for forced unmount).
1786          *
1787          * Release all holds on dbufs.
1788          */
1789         mutex_enter(&zfsvfs->z_znodes_lock);
1790         for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1791             zp = list_next(&zfsvfs->z_all_znodes, zp))
1792                 if (zp->z_sa_hdl) {
1793                         ASSERT(ZTOV(zp)->v_count > 0);
1794                         zfs_znode_dmu_fini(zp);
1795                 }
1796         mutex_exit(&zfsvfs->z_znodes_lock);
1797 
1798         /*
1799          * If we are unmounting, set the unmounted flag and let new vops
1800          * unblock.  zfs_inactive will have the unmounted behavior, and all
1801          * other vops will fail with EIO.
1802          */
1803         if (unmounting) {
1804                 zfsvfs->z_unmounted = B_TRUE;
1805                 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1806                 rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
1807         }
1808 
1809         /*
1810          * z_os will be NULL if there was an error in attempting to reopen
1811          * zfsvfs, so just return as the properties had already been
1812          * unregistered and cached data had been evicted before.
1813          */
1814         if (zfsvfs->z_os == NULL)
1815                 return (0);
1816 
1817         /*
1818          * Unregister properties.
1819          */
1820         zfs_unregister_callbacks(zfsvfs);
1821 
1822         /*
1823          * Evict cached data
1824          */
1825         if (dsl_dataset_is_dirty(dmu_objset_ds(zfsvfs->z_os)) &&
1826             !(zfsvfs->z_vfs->vfs_flag & VFS_RDONLY))
1827                 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1828         dmu_objset_evict_dbufs(zfsvfs->z_os);
1829 
1830         return (0);
1831 }
1832 
1833 /*ARGSUSED*/
1834 static int
1835 zfs_umount(vfs_t *vfsp, int fflag, cred_t *cr)
1836 {
1837         zfsvfs_t *zfsvfs = vfsp->vfs_data;
1838         objset_t *os;
1839         int ret;
1840 
1841         ret = secpolicy_fs_unmount(cr, vfsp);
1842         if (ret) {
1843                 if (dsl_deleg_access((char *)refstr_value(vfsp->vfs_resource),
1844                     ZFS_DELEG_PERM_MOUNT, cr))
1845                         return (ret);
1846         }
1847 
1848         /*
1849          * We purge the parent filesystem's vfsp as the parent filesystem
1850          * and all of its snapshots have their vnode's v_vfsp set to the
1851          * parent's filesystem's vfsp.  Note, 'z_parent' is self
1852          * referential for non-snapshots.
1853          */
1854         (void) dnlc_purge_vfsp(zfsvfs->z_parent->z_vfs, 0);
1855 
1856         /*
1857          * Unmount any snapshots mounted under .zfs before unmounting the
1858          * dataset itself.
1859          */
1860         if (zfsvfs->z_ctldir != NULL &&
1861             (ret = zfsctl_umount_snapshots(vfsp, fflag, cr)) != 0) {
1862                 return (ret);
1863         }
1864 
1865         if (!(fflag & MS_FORCE)) {
1866                 /*
1867                  * Check the number of active vnodes in the file system.
1868                  * Our count is maintained in the vfs structure, but the
1869                  * number is off by 1 to indicate a hold on the vfs
1870                  * structure itself.
1871                  *
1872                  * The '.zfs' directory maintains a reference of its
1873                  * own, and any active references underneath are
1874                  * reflected in the vnode count.
1875                  */
1876                 if (zfsvfs->z_ctldir == NULL) {
1877                         if (vfsp->vfs_count > 1)
1878                                 return (SET_ERROR(EBUSY));
1879                 } else {
1880                         if (vfsp->vfs_count > 2 ||
1881                             zfsvfs->z_ctldir->v_count > 1)
1882                                 return (SET_ERROR(EBUSY));
1883                 }
1884         }
1885 
1886         vfsp->vfs_flag |= VFS_UNMOUNTED;
1887 
1888         VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
1889         os = zfsvfs->z_os;
1890 
1891         /*
1892          * z_os will be NULL if there was an error in
1893          * attempting to reopen zfsvfs.
1894          */
1895         if (os != NULL) {
1896                 /*
1897                  * Unset the objset user_ptr.
1898                  */
1899                 mutex_enter(&os->os_user_ptr_lock);
1900                 dmu_objset_set_user(os, NULL);
1901                 mutex_exit(&os->os_user_ptr_lock);
1902 
1903                 /*
1904                  * Finally release the objset
1905                  */
1906                 dmu_objset_disown(os, zfsvfs);
1907         }
1908 
1909         /*
1910          * We can now safely destroy the '.zfs' directory node.
1911          */
1912         if (zfsvfs->z_ctldir != NULL)
1913                 zfsctl_destroy(zfsvfs);
1914 
1915         return (0);
1916 }
1917 
1918 static int
1919 zfs_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp)
1920 {
1921         zfsvfs_t        *zfsvfs = vfsp->vfs_data;
1922         znode_t         *zp;
1923         uint64_t        object = 0;
1924         uint64_t        fid_gen = 0;
1925         uint64_t        gen_mask;
1926         uint64_t        zp_gen;
1927         int             i, err;
1928 
1929         *vpp = NULL;
1930 
1931         ZFS_ENTER(zfsvfs);
1932 
1933         if (fidp->fid_len == LONG_FID_LEN) {
1934                 zfid_long_t     *zlfid = (zfid_long_t *)fidp;
1935                 uint64_t        objsetid = 0;
1936                 uint64_t        setgen = 0;
1937 
1938                 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
1939                         objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
1940 
1941                 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
1942                         setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
1943 
1944                 ZFS_EXIT(zfsvfs);
1945 
1946                 err = zfsctl_lookup_objset(vfsp, objsetid, &zfsvfs);
1947                 if (err)
1948                         return (SET_ERROR(EINVAL));
1949                 ZFS_ENTER(zfsvfs);
1950         }
1951 
1952         if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
1953                 zfid_short_t    *zfid = (zfid_short_t *)fidp;
1954 
1955                 for (i = 0; i < sizeof (zfid->zf_object); i++)
1956                         object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
1957 
1958                 for (i = 0; i < sizeof (zfid->zf_gen); i++)
1959                         fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
1960         } else {
1961                 ZFS_EXIT(zfsvfs);
1962                 return (SET_ERROR(EINVAL));
1963         }
1964 
1965         /* A zero fid_gen means we are in the .zfs control directories */
1966         if (fid_gen == 0 &&
1967             (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
1968                 *vpp = zfsvfs->z_ctldir;
1969                 ASSERT(*vpp != NULL);
1970                 if (object == ZFSCTL_INO_SNAPDIR) {
1971                         VERIFY(zfsctl_root_lookup(*vpp, "snapshot", vpp, NULL,
1972                             0, NULL, NULL, NULL, NULL, NULL) == 0);
1973                 } else {
1974                         VN_HOLD(*vpp);
1975                 }
1976                 ZFS_EXIT(zfsvfs);
1977                 return (0);
1978         }
1979 
1980         gen_mask = -1ULL >> (64 - 8 * i);
1981 
1982         dprintf("getting %llu [%u mask %llx]\n", object, fid_gen, gen_mask);
1983         if (err = zfs_zget(zfsvfs, object, &zp)) {
1984                 ZFS_EXIT(zfsvfs);
1985                 return (err);
1986         }
1987         (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
1988             sizeof (uint64_t));
1989         zp_gen = zp_gen & gen_mask;
1990         if (zp_gen == 0)
1991                 zp_gen = 1;
1992         if (zp->z_unlinked || zp_gen != fid_gen) {
1993                 dprintf("znode gen (%u) != fid gen (%u)\n", zp_gen, fid_gen);
1994                 VN_RELE(ZTOV(zp));
1995                 ZFS_EXIT(zfsvfs);
1996                 return (SET_ERROR(EINVAL));
1997         }
1998 
1999         *vpp = ZTOV(zp);
2000         ZFS_EXIT(zfsvfs);
2001         return (0);
2002 }
2003 
2004 /*
2005  * Block out VOPs and close zfsvfs_t::z_os
2006  *
2007  * Note, if successful, then we return with the 'z_teardown_lock' and
2008  * 'z_teardown_inactive_lock' write held.  We leave ownership of the underlying
2009  * dataset and objset intact so that they can be atomically handed off during
2010  * a subsequent rollback or recv operation and the resume thereafter.
2011  */
2012 int
2013 zfs_suspend_fs(zfsvfs_t *zfsvfs)
2014 {
2015         int error;
2016 
2017         if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
2018                 return (error);
2019 
2020         return (0);
2021 }
2022 
2023 /*
2024  * Rebuild SA and release VOPs.  Note that ownership of the underlying dataset
2025  * is an invariant across any of the operations that can be performed while the
2026  * filesystem was suspended.  Whether it succeeded or failed, the preconditions
2027  * are the same: the relevant objset and associated dataset are owned by
2028  * zfsvfs, held, and long held on entry.
2029  */
2030 int
2031 zfs_resume_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds)
2032 {
2033         int err;
2034         znode_t *zp;
2035 
2036         ASSERT(RRM_WRITE_HELD(&zfsvfs->z_teardown_lock));
2037         ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
2038 
2039         /*
2040          * We already own this, so just update the objset_t, as the one we
2041          * had before may have been evicted.
2042          */
2043         objset_t *os;
2044         VERIFY3P(ds->ds_owner, ==, zfsvfs);
2045         VERIFY(dsl_dataset_long_held(ds));
2046         VERIFY0(dmu_objset_from_ds(ds, &os));
2047 
2048         err = zfsvfs_init(zfsvfs, os);
2049         if (err != 0)
2050                 goto bail;
2051 
2052         VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
2053 
2054         zfs_set_fuid_feature(zfsvfs);
2055 
2056         /*
2057          * Attempt to re-establish all the active znodes with
2058          * their dbufs.  If a zfs_rezget() fails, then we'll let
2059          * any potential callers discover that via ZFS_ENTER_VERIFY_VP
2060          * when they try to use their znode.
2061          */
2062         mutex_enter(&zfsvfs->z_znodes_lock);
2063         for (zp = list_head(&zfsvfs->z_all_znodes); zp;
2064             zp = list_next(&zfsvfs->z_all_znodes, zp)) {
2065                 (void) zfs_rezget(zp);
2066         }
2067         mutex_exit(&zfsvfs->z_znodes_lock);
2068 
2069 bail:
2070         /* release the VOPs */
2071         rw_exit(&zfsvfs->z_teardown_inactive_lock);
2072         rrm_exit(&zfsvfs->z_teardown_lock, FTAG);
2073 
2074         if (err) {
2075                 /*
2076                  * Since we couldn't setup the sa framework, try to force
2077                  * unmount this file system.
2078                  */
2079                 if (vn_vfswlock(zfsvfs->z_vfs->vfs_vnodecovered) == 0)
2080                         (void) dounmount(zfsvfs->z_vfs, MS_FORCE, CRED());
2081         }
2082         return (err);
2083 }
2084 
2085 static void
2086 zfs_freevfs(vfs_t *vfsp)
2087 {
2088         zfsvfs_t *zfsvfs = vfsp->vfs_data;
2089 
2090         /*
2091          * If this is a snapshot, we have an extra VFS_HOLD on our parent
2092          * from zfs_mount().  Release it here.  If we came through
2093          * zfs_mountroot() instead, we didn't grab an extra hold, so
2094          * skip the VFS_RELE for rootvfs.
2095          */
2096         if (zfsvfs->z_issnap && (vfsp != rootvfs))
2097                 VFS_RELE(zfsvfs->z_parent->z_vfs);
2098 
2099         zfsvfs_free(zfsvfs);
2100 
2101         atomic_dec_32(&zfs_active_fs_count);
2102 }
2103 
2104 /*
2105  * VFS_INIT() initialization.  Note that there is no VFS_FINI(),
2106  * so we can't safely do any non-idempotent initialization here.
2107  * Leave that to zfs_init() and zfs_fini(), which are called
2108  * from the module's _init() and _fini() entry points.
2109  */
2110 /*ARGSUSED*/
2111 static int
2112 zfs_vfsinit(int fstype, char *name)
2113 {
2114         int error;
2115 
2116         zfsfstype = fstype;
2117 
2118         /*
2119          * Setup vfsops and vnodeops tables.
2120          */
2121         error = vfs_setfsops(fstype, zfs_vfsops_template, &zfs_vfsops);
2122         if (error != 0) {
2123                 cmn_err(CE_WARN, "zfs: bad vfs ops template");
2124         }
2125 
2126         error = zfs_create_op_tables();
2127         if (error) {
2128                 zfs_remove_op_tables();
2129                 cmn_err(CE_WARN, "zfs: bad vnode ops template");
2130                 (void) vfs_freevfsops_by_type(zfsfstype);
2131                 return (error);
2132         }
2133 
2134         mutex_init(&zfs_dev_mtx, NULL, MUTEX_DEFAULT, NULL);
2135 
2136         /*
2137          * Unique major number for all zfs mounts.
2138          * If we run out of 32-bit minors, we'll getudev() another major.
2139          */
2140         zfs_major = ddi_name_to_major(ZFS_DRIVER);
2141         zfs_minor = ZFS_MIN_MINOR;
2142 
2143         return (0);
2144 }
2145 
2146 void
2147 zfs_init(void)
2148 {
2149         /*
2150          * Initialize .zfs directory structures
2151          */
2152         zfsctl_init();
2153 
2154         /*
2155          * Initialize znode cache, vnode ops, etc...
2156          */
2157         zfs_znode_init();
2158 
2159         dmu_objset_register_type(DMU_OST_ZFS, zfs_space_delta_cb);
2160 }
2161 
2162 void
2163 zfs_fini(void)
2164 {
2165         zfsctl_fini();
2166         zfs_znode_fini();
2167 }
2168 
2169 int
2170 zfs_busy(void)
2171 {
2172         return (zfs_active_fs_count != 0);
2173 }
2174 
2175 int
2176 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
2177 {
2178         int error;
2179         objset_t *os = zfsvfs->z_os;
2180         dmu_tx_t *tx;
2181 
2182         if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
2183                 return (SET_ERROR(EINVAL));
2184 
2185         if (newvers < zfsvfs->z_version)
2186                 return (SET_ERROR(EINVAL));
2187 
2188         if (zfs_spa_version_map(newvers) >
2189             spa_version(dmu_objset_spa(zfsvfs->z_os)))
2190                 return (SET_ERROR(ENOTSUP));
2191 
2192         tx = dmu_tx_create(os);
2193         dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2194         if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2195                 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2196                     ZFS_SA_ATTRS);
2197                 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2198         }
2199         error = dmu_tx_assign(tx, TXG_WAIT);
2200         if (error) {
2201                 dmu_tx_abort(tx);
2202                 return (error);
2203         }
2204 
2205         error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2206             8, 1, &newvers, tx);
2207 
2208         if (error) {
2209                 dmu_tx_commit(tx);
2210                 return (error);
2211         }
2212 
2213         if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2214                 uint64_t sa_obj;
2215 
2216                 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2217                     SPA_VERSION_SA);
2218                 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2219                     DMU_OT_NONE, 0, tx);
2220 
2221                 error = zap_add(os, MASTER_NODE_OBJ,
2222                     ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2223                 ASSERT0(error);
2224 
2225                 VERIFY(0 == sa_set_sa_object(os, sa_obj));
2226                 sa_register_update_callback(os, zfs_sa_upgrade);
2227         }
2228 
2229         spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2230             "from %llu to %llu", zfsvfs->z_version, newvers);
2231 
2232         dmu_tx_commit(tx);
2233 
2234         zfsvfs->z_version = newvers;
2235 
2236         zfs_set_fuid_feature(zfsvfs);
2237 
2238         return (0);
2239 }
2240 
2241 /*
2242  * Read a property stored within the master node.
2243  */
2244 int
2245 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2246 {
2247         const char *pname;
2248         int error = ENOENT;
2249 
2250         /*
2251          * Look up the file system's value for the property.  For the
2252          * version property, we look up a slightly different string.
2253          */
2254         if (prop == ZFS_PROP_VERSION)
2255                 pname = ZPL_VERSION_STR;
2256         else
2257                 pname = zfs_prop_to_name(prop);
2258 
2259         if (os != NULL) {
2260                 ASSERT3U(os->os_phys->os_type, ==, DMU_OST_ZFS);
2261                 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2262         }
2263 
2264         if (error == ENOENT) {
2265                 /* No value set, use the default value */
2266                 switch (prop) {
2267                 case ZFS_PROP_VERSION:
2268                         *value = ZPL_VERSION;
2269                         break;
2270                 case ZFS_PROP_NORMALIZE:
2271                 case ZFS_PROP_UTF8ONLY:
2272                         *value = 0;
2273                         break;
2274                 case ZFS_PROP_CASE:
2275                         *value = ZFS_CASE_SENSITIVE;
2276                         break;
2277                 default:
2278                         return (error);
2279                 }
2280                 error = 0;
2281         }
2282         return (error);
2283 }
2284 
2285 /*
2286  * Return true if the coresponding vfs's unmounted flag is set.
2287  * Otherwise return false.
2288  * If this function returns true we know VFS unmount has been initiated.
2289  */
2290 boolean_t
2291 zfs_get_vfs_flag_unmounted(objset_t *os)
2292 {
2293         zfsvfs_t *zfvp;
2294         boolean_t unmounted = B_FALSE;
2295 
2296         ASSERT(dmu_objset_type(os) == DMU_OST_ZFS);
2297 
2298         mutex_enter(&os->os_user_ptr_lock);
2299         zfvp = dmu_objset_get_user(os);
2300         if (zfvp != NULL && zfvp->z_vfs != NULL &&
2301             (zfvp->z_vfs->vfs_flag & VFS_UNMOUNTED))
2302                 unmounted = B_TRUE;
2303         mutex_exit(&os->os_user_ptr_lock);
2304 
2305         return (unmounted);
2306 }
2307 
2308 static vfsdef_t vfw = {
2309         VFSDEF_VERSION,
2310         MNTTYPE_ZFS,
2311         zfs_vfsinit,
2312         VSW_HASPROTO|VSW_CANRWRO|VSW_CANREMOUNT|VSW_VOLATILEDEV|VSW_STATS|
2313             VSW_XID|VSW_ZMOUNT,
2314         &zfs_mntopts
2315 };
2316 
2317 struct modlfs zfs_modlfs = {
2318         &mod_fsops, "ZFS filesystem version " SPA_VERSION_STRING, &vfw
2319 };