module/os/linux/zfs/zfs_vfsops.c

   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, 2018 by Delphix. All rights reserved.
  24  */
  25
  26 /* Portions Copyright 2010 Robert Milkowski */
  27
  28 #include <sys/types.h>
  29 #include <sys/param.h>
  30 #include <sys/sysmacros.h>
  31 #include <sys/kmem.h>
  32 #include <sys/pathname.h>
  33 #include <sys/vnode.h>
  34 #include <sys/vfs.h>
  35 #include <sys/mntent.h>
  36 #include <sys/cmn_err.h>
  37 #include <sys/zfs_znode.h>
  38 #include <sys/zfs_vnops.h>
  39 #include <sys/zfs_dir.h>
  40 #include <sys/zil.h>
  41 #include <sys/fs/zfs.h>
  42 #include <sys/dmu.h>
  43 #include <sys/dsl_prop.h>
  44 #include <sys/dsl_dataset.h>
  45 #include <sys/dsl_deleg.h>
  46 #include <sys/spa.h>
  47 #include <sys/zap.h>
  48 #include <sys/sa.h>
  49 #include <sys/sa_impl.h>
  50 #include <sys/policy.h>
  51 #include <sys/atomic.h>
  52 #include <sys/zfs_ioctl.h>
  53 #include <sys/zfs_ctldir.h>
  54 #include <sys/zfs_fuid.h>
  55 #include <sys/zfs_quota.h>
  56 #include <sys/sunddi.h>
  57 #include <sys/dmu_objset.h>
  58 #include <sys/dsl_dir.h>
  59 #include <sys/spa_boot.h>
  60 #include <sys/objlist.h>
  61 #include <sys/zpl.h>
  62 #include <linux/vfs_compat.h>
  63 #include "zfs_comutil.h"
  64
  65 enum {
  66         TOKEN_RO,
  67         TOKEN_RW,
  68         TOKEN_SETUID,
  69         TOKEN_NOSETUID,
  70         TOKEN_EXEC,
  71         TOKEN_NOEXEC,
  72         TOKEN_DEVICES,
  73         TOKEN_NODEVICES,
  74         TOKEN_DIRXATTR,
  75         TOKEN_SAXATTR,
  76         TOKEN_XATTR,
  77         TOKEN_NOXATTR,
  78         TOKEN_ATIME,
  79         TOKEN_NOATIME,
  80         TOKEN_RELATIME,
  81         TOKEN_NORELATIME,
  82         TOKEN_NBMAND,
  83         TOKEN_NONBMAND,
  84         TOKEN_MNTPOINT,
  85         TOKEN_LAST,
  86 };
  87
  88 static const match_table_t zpl_tokens = {
  89         { TOKEN_RO,             MNTOPT_RO },
  90         { TOKEN_RW,             MNTOPT_RW },
  91         { TOKEN_SETUID,         MNTOPT_SETUID },
  92         { TOKEN_NOSETUID,       MNTOPT_NOSETUID },
  93         { TOKEN_EXEC,           MNTOPT_EXEC },
  94         { TOKEN_NOEXEC,         MNTOPT_NOEXEC },
  95         { TOKEN_DEVICES,        MNTOPT_DEVICES },
  96         { TOKEN_NODEVICES,      MNTOPT_NODEVICES },
  97         { TOKEN_DIRXATTR,       MNTOPT_DIRXATTR },
  98         { TOKEN_SAXATTR,        MNTOPT_SAXATTR },
  99         { TOKEN_XATTR,          MNTOPT_XATTR },
 100         { TOKEN_NOXATTR,        MNTOPT_NOXATTR },
 101         { TOKEN_ATIME,          MNTOPT_ATIME },
 102         { TOKEN_NOATIME,        MNTOPT_NOATIME },
 103         { TOKEN_RELATIME,       MNTOPT_RELATIME },
 104         { TOKEN_NORELATIME,     MNTOPT_NORELATIME },
 105         { TOKEN_NBMAND,         MNTOPT_NBMAND },
 106         { TOKEN_NONBMAND,       MNTOPT_NONBMAND },
 107         { TOKEN_MNTPOINT,       MNTOPT_MNTPOINT "=%s" },
 108         { TOKEN_LAST,           NULL },
 109 };
 110
 111 static void
 112 zfsvfs_vfs_free(vfs_t *vfsp)
 113 {
 114         if (vfsp != NULL) {
 115                 if (vfsp->vfs_mntpoint != NULL)
 116                         kmem_strfree(vfsp->vfs_mntpoint);
 117
 118                 kmem_free(vfsp, sizeof (vfs_t));
 119         }
 120 }
 121
 122 static int
 123 zfsvfs_parse_option(char *option, int token, substring_t *args, vfs_t *vfsp)
 124 {
 125         switch (token) {
 126         case TOKEN_RO:
 127                 vfsp->vfs_readonly = B_TRUE;
 128                 vfsp->vfs_do_readonly = B_TRUE;
 129                 break;
 130         case TOKEN_RW:
 131                 vfsp->vfs_readonly = B_FALSE;
 132                 vfsp->vfs_do_readonly = B_TRUE;
 133                 break;
 134         case TOKEN_SETUID:
 135                 vfsp->vfs_setuid = B_TRUE;
 136                 vfsp->vfs_do_setuid = B_TRUE;
 137                 break;
 138         case TOKEN_NOSETUID:
 139                 vfsp->vfs_setuid = B_FALSE;
 140                 vfsp->vfs_do_setuid = B_TRUE;
 141                 break;
 142         case TOKEN_EXEC:
 143                 vfsp->vfs_exec = B_TRUE;
 144                 vfsp->vfs_do_exec = B_TRUE;
 145                 break;
 146         case TOKEN_NOEXEC:
 147                 vfsp->vfs_exec = B_FALSE;
 148                 vfsp->vfs_do_exec = B_TRUE;
 149                 break;
 150         case TOKEN_DEVICES:
 151                 vfsp->vfs_devices = B_TRUE;
 152                 vfsp->vfs_do_devices = B_TRUE;
 153                 break;
 154         case TOKEN_NODEVICES:
 155                 vfsp->vfs_devices = B_FALSE;
 156                 vfsp->vfs_do_devices = B_TRUE;
 157                 break;
 158         case TOKEN_DIRXATTR:
 159                 vfsp->vfs_xattr = ZFS_XATTR_DIR;
 160                 vfsp->vfs_do_xattr = B_TRUE;
 161                 break;
 162         case TOKEN_SAXATTR:
 163                 vfsp->vfs_xattr = ZFS_XATTR_SA;
 164                 vfsp->vfs_do_xattr = B_TRUE;
 165                 break;
 166         case TOKEN_XATTR:
 167                 vfsp->vfs_xattr = ZFS_XATTR_DIR;
 168                 vfsp->vfs_do_xattr = B_TRUE;
 169                 break;
 170         case TOKEN_NOXATTR:
 171                 vfsp->vfs_xattr = ZFS_XATTR_OFF;
 172                 vfsp->vfs_do_xattr = B_TRUE;
 173                 break;
 174         case TOKEN_ATIME:
 175                 vfsp->vfs_atime = B_TRUE;
 176                 vfsp->vfs_do_atime = B_TRUE;
 177                 break;
 178         case TOKEN_NOATIME:
 179                 vfsp->vfs_atime = B_FALSE;
 180                 vfsp->vfs_do_atime = B_TRUE;
 181                 break;
 182         case TOKEN_RELATIME:
 183                 vfsp->vfs_relatime = B_TRUE;
 184                 vfsp->vfs_do_relatime = B_TRUE;
 185                 break;
 186         case TOKEN_NORELATIME:
 187                 vfsp->vfs_relatime = B_FALSE;
 188                 vfsp->vfs_do_relatime = B_TRUE;
 189                 break;
 190         case TOKEN_NBMAND:
 191                 vfsp->vfs_nbmand = B_TRUE;
 192                 vfsp->vfs_do_nbmand = B_TRUE;
 193                 break;
 194         case TOKEN_NONBMAND:
 195                 vfsp->vfs_nbmand = B_FALSE;
 196                 vfsp->vfs_do_nbmand = B_TRUE;
 197                 break;
 198         case TOKEN_MNTPOINT:
 199                 vfsp->vfs_mntpoint = match_strdup(&args[0]);
 200                 if (vfsp->vfs_mntpoint == NULL)
 201                         return (SET_ERROR(ENOMEM));
 202
 203                 break;
 204         default:
 205                 break;
 206         }
 207
 208         return (0);
 209 }
 210
 211 /*
 212  * Parse the raw mntopts and return a vfs_t describing the options.
 213  */
 214 static int
 215 zfsvfs_parse_options(char *mntopts, vfs_t **vfsp)
 216 {
 217         vfs_t *tmp_vfsp;
 218         int error;
 219
 220         tmp_vfsp = kmem_zalloc(sizeof (vfs_t), KM_SLEEP);
 221
 222         if (mntopts != NULL) {
 223                 substring_t args[MAX_OPT_ARGS];
 224                 char *tmp_mntopts, *p, *t;
 225                 int token;
 226
 227                 tmp_mntopts = t = kmem_strdup(mntopts);
 228                 if (tmp_mntopts == NULL)
 229                         return (SET_ERROR(ENOMEM));
 230
 231                 while ((p = strsep(&t, ",")) != NULL) {
 232                         if (!*p)
 233                                 continue;
 234
 235                         args[0].to = args[0].from = NULL;
 236                         token = match_token(p, zpl_tokens, args);
 237                         error = zfsvfs_parse_option(p, token, args, tmp_vfsp);
 238                         if (error) {
 239                                 kmem_strfree(tmp_mntopts);
 240                                 zfsvfs_vfs_free(tmp_vfsp);
 241                                 return (error);
 242                         }
 243                 }
 244
 245                 kmem_strfree(tmp_mntopts);
 246         }
 247
 248         *vfsp = tmp_vfsp;
 249
 250         return (0);
 251 }
 252
 253 boolean_t
 254 zfs_is_readonly(zfsvfs_t *zfsvfs)
 255 {
 256         return (!!(zfsvfs->z_sb->s_flags & SB_RDONLY));
 257 }
 258
 259 int
 260 zfs_sync(struct super_block *sb, int wait, cred_t *cr)
 261 {
 262         (void) cr;
 263         zfsvfs_t *zfsvfs = sb->s_fs_info;
 264
 265         /*
 266          * Semantically, the only requirement is that the sync be initiated.
 267          * The DMU syncs out txgs frequently, so there's nothing to do.
 268          */
 269         if (!wait)
 270                 return (0);
 271
 272         if (zfsvfs != NULL) {
 273                 /*
 274                  * Sync a specific filesystem.
 275                  */
 276                 dsl_pool_t *dp;
 277
 278                 ZFS_ENTER(zfsvfs);
 279                 dp = dmu_objset_pool(zfsvfs->z_os);
 280
 281                 /*
 282                  * If the system is shutting down, then skip any
 283                  * filesystems which may exist on a suspended pool.
 284                  */
 285                 if (spa_suspended(dp->dp_spa)) {
 286                         ZFS_EXIT(zfsvfs);
 287                         return (0);
 288                 }
 289
 290                 if (zfsvfs->z_log != NULL)
 291                         zil_commit(zfsvfs->z_log, 0);
 292
 293                 ZFS_EXIT(zfsvfs);
 294         } else {
 295                 /*
 296                  * Sync all ZFS filesystems.  This is what happens when you
 297                  * run sync(1).  Unlike other filesystems, ZFS honors the
 298                  * request by waiting for all pools to commit all dirty data.
 299                  */
 300                 spa_sync_allpools();
 301         }
 302
 303         return (0);
 304 }
 305
 306 static void
 307 atime_changed_cb(void *arg, uint64_t newval)
 308 {
 309         zfsvfs_t *zfsvfs = arg;
 310         struct super_block *sb = zfsvfs->z_sb;
 311
 312         if (sb == NULL)
 313                 return;
 314         /*
 315          * Update SB_NOATIME bit in VFS super block.  Since atime update is
 316          * determined by atime_needs_update(), atime_needs_update() needs to
 317          * return false if atime is turned off, and not unconditionally return
 318          * false if atime is turned on.
 319          */
 320         if (newval)
 321                 sb->s_flags &= ~SB_NOATIME;
 322         else
 323                 sb->s_flags |= SB_NOATIME;
 324 }
 325
 326 static void
 327 relatime_changed_cb(void *arg, uint64_t newval)
 328 {
 329         ((zfsvfs_t *)arg)->z_relatime = newval;
 330 }
 331
 332 static void
 333 xattr_changed_cb(void *arg, uint64_t newval)
 334 {
 335         zfsvfs_t *zfsvfs = arg;
 336
 337         if (newval == ZFS_XATTR_OFF) {
 338                 zfsvfs->z_flags &= ~ZSB_XATTR;
 339         } else {
 340                 zfsvfs->z_flags |= ZSB_XATTR;
 341
 342                 if (newval == ZFS_XATTR_SA)
 343                         zfsvfs->z_xattr_sa = B_TRUE;
 344                 else
 345                         zfsvfs->z_xattr_sa = B_FALSE;
 346         }
 347 }
 348
 349 static void
 350 acltype_changed_cb(void *arg, uint64_t newval)
 351 {
 352         zfsvfs_t *zfsvfs = arg;
 353
 354         switch (newval) {
 355         case ZFS_ACLTYPE_NFSV4:
 356         case ZFS_ACLTYPE_OFF:
 357                 zfsvfs->z_acl_type = ZFS_ACLTYPE_OFF;
 358                 zfsvfs->z_sb->s_flags &= ~SB_POSIXACL;
 359                 break;
 360         case ZFS_ACLTYPE_POSIX:
 361 #ifdef CONFIG_FS_POSIX_ACL
 362                 zfsvfs->z_acl_type = ZFS_ACLTYPE_POSIX;
 363                 zfsvfs->z_sb->s_flags |= SB_POSIXACL;
 364 #else
 365                 zfsvfs->z_acl_type = ZFS_ACLTYPE_OFF;
 366                 zfsvfs->z_sb->s_flags &= ~SB_POSIXACL;
 367 #endif /* CONFIG_FS_POSIX_ACL */
 368                 break;
 369         default:
 370                 break;
 371         }
 372 }
 373
 374 static void
 375 blksz_changed_cb(void *arg, uint64_t newval)
 376 {
 377         zfsvfs_t *zfsvfs = arg;
 378         ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os)));
 379         ASSERT3U(newval, >=, SPA_MINBLOCKSIZE);
 380         ASSERT(ISP2(newval));
 381
 382         zfsvfs->z_max_blksz = newval;
 383 }
 384
 385 static void
 386 readonly_changed_cb(void *arg, uint64_t newval)
 387 {
 388         zfsvfs_t *zfsvfs = arg;
 389         struct super_block *sb = zfsvfs->z_sb;
 390
 391         if (sb == NULL)
 392                 return;
 393
 394         if (newval)
 395                 sb->s_flags |= SB_RDONLY;
 396         else
 397                 sb->s_flags &= ~SB_RDONLY;
 398 }
 399
 400 static void
 401 devices_changed_cb(void *arg, uint64_t newval)
 402 {
 403 }
 404
 405 static void
 406 setuid_changed_cb(void *arg, uint64_t newval)
 407 {
 408 }
 409
 410 static void
 411 exec_changed_cb(void *arg, uint64_t newval)
 412 {
 413 }
 414
 415 static void
 416 nbmand_changed_cb(void *arg, uint64_t newval)
 417 {
 418         zfsvfs_t *zfsvfs = arg;
 419         struct super_block *sb = zfsvfs->z_sb;
 420
 421         if (sb == NULL)
 422                 return;
 423
 424         if (newval == TRUE)
 425                 sb->s_flags |= SB_MANDLOCK;
 426         else
 427                 sb->s_flags &= ~SB_MANDLOCK;
 428 }
 429
 430 static void
 431 snapdir_changed_cb(void *arg, uint64_t newval)
 432 {
 433         ((zfsvfs_t *)arg)->z_show_ctldir = newval;
 434 }
 435
 436 static void
 437 acl_mode_changed_cb(void *arg, uint64_t newval)
 438 {
 439         zfsvfs_t *zfsvfs = arg;
 440
 441         zfsvfs->z_acl_mode = newval;
 442 }
 443
 444 static void
 445 acl_inherit_changed_cb(void *arg, uint64_t newval)
 446 {
 447         ((zfsvfs_t *)arg)->z_acl_inherit = newval;
 448 }
 449
 450 static int
 451 zfs_register_callbacks(vfs_t *vfsp)
 452 {
 453         struct dsl_dataset *ds = NULL;
 454         objset_t *os = NULL;
 455         zfsvfs_t *zfsvfs = NULL;
 456         int error = 0;
 457
 458         ASSERT(vfsp);
 459         zfsvfs = vfsp->vfs_data;
 460         ASSERT(zfsvfs);
 461         os = zfsvfs->z_os;
 462
 463         /*
 464          * The act of registering our callbacks will destroy any mount
 465          * options we may have.  In order to enable temporary overrides
 466          * of mount options, we stash away the current values and
 467          * restore them after we register the callbacks.
 468          */
 469         if (zfs_is_readonly(zfsvfs) || !spa_writeable(dmu_objset_spa(os))) {
 470                 vfsp->vfs_do_readonly = B_TRUE;
 471                 vfsp->vfs_readonly = B_TRUE;
 472         }
 473
 474         /*
 475          * Register property callbacks.
 476          *
 477          * It would probably be fine to just check for i/o error from
 478          * the first prop_register(), but I guess I like to go
 479          * overboard...
 480          */
 481         ds = dmu_objset_ds(os);
 482         dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
 483         error = dsl_prop_register(ds,
 484             zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
 485         error = error ? error : dsl_prop_register(ds,
 486             zfs_prop_to_name(ZFS_PROP_RELATIME), relatime_changed_cb, zfsvfs);
 487         error = error ? error : dsl_prop_register(ds,
 488             zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
 489         error = error ? error : dsl_prop_register(ds,
 490             zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
 491         error = error ? error : dsl_prop_register(ds,
 492             zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
 493         error = error ? error : dsl_prop_register(ds,
 494             zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
 495         error = error ? error : dsl_prop_register(ds,
 496             zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
 497         error = error ? error : dsl_prop_register(ds,
 498             zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
 499         error = error ? error : dsl_prop_register(ds,
 500             zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
 501         error = error ? error : dsl_prop_register(ds,
 502             zfs_prop_to_name(ZFS_PROP_ACLTYPE), acltype_changed_cb, zfsvfs);
 503         error = error ? error : dsl_prop_register(ds,
 504             zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs);
 505         error = error ? error : dsl_prop_register(ds,
 506             zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
 507             zfsvfs);
 508         error = error ? error : dsl_prop_register(ds,
 509             zfs_prop_to_name(ZFS_PROP_NBMAND), nbmand_changed_cb, zfsvfs);
 510         dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
 511         if (error)
 512                 goto unregister;
 513
 514         /*
 515          * Invoke our callbacks to restore temporary mount options.
 516          */
 517         if (vfsp->vfs_do_readonly)
 518                 readonly_changed_cb(zfsvfs, vfsp->vfs_readonly);
 519         if (vfsp->vfs_do_setuid)
 520                 setuid_changed_cb(zfsvfs, vfsp->vfs_setuid);
 521         if (vfsp->vfs_do_exec)
 522                 exec_changed_cb(zfsvfs, vfsp->vfs_exec);
 523         if (vfsp->vfs_do_devices)
 524                 devices_changed_cb(zfsvfs, vfsp->vfs_devices);
 525         if (vfsp->vfs_do_xattr)
 526                 xattr_changed_cb(zfsvfs, vfsp->vfs_xattr);
 527         if (vfsp->vfs_do_atime)
 528                 atime_changed_cb(zfsvfs, vfsp->vfs_atime);
 529         if (vfsp->vfs_do_relatime)
 530                 relatime_changed_cb(zfsvfs, vfsp->vfs_relatime);
 531         if (vfsp->vfs_do_nbmand)
 532                 nbmand_changed_cb(zfsvfs, vfsp->vfs_nbmand);
 533
 534         return (0);
 535
 536 unregister:
 537         dsl_prop_unregister_all(ds, zfsvfs);
 538         return (error);
 539 }
 540
 541 /*
 542  * Takes a dataset, a property, a value and that value's setpoint as
 543  * found in the ZAP. Checks if the property has been changed in the vfs.
 544  * If so, val and setpoint will be overwritten with updated content.
 545  * Otherwise, they are left unchanged.
 546  */
 547 int
 548 zfs_get_temporary_prop(dsl_dataset_t *ds, zfs_prop_t zfs_prop, uint64_t *val,
 549     char *setpoint)
 550 {
 551         int error;
 552         zfsvfs_t *zfvp;
 553         vfs_t *vfsp;
 554         objset_t *os;
 555         uint64_t tmp = *val;
 556
 557         error = dmu_objset_from_ds(ds, &os);
 558         if (error != 0)
 559                 return (error);
 560
 561         if (dmu_objset_type(os) != DMU_OST_ZFS)
 562                 return (EINVAL);
 563
 564         mutex_enter(&os->os_user_ptr_lock);
 565         zfvp = dmu_objset_get_user(os);
 566         mutex_exit(&os->os_user_ptr_lock);
 567         if (zfvp == NULL)
 568                 return (ESRCH);
 569
 570         vfsp = zfvp->z_vfs;
 571
 572         switch (zfs_prop) {
 573         case ZFS_PROP_ATIME:
 574                 if (vfsp->vfs_do_atime)
 575                         tmp = vfsp->vfs_atime;
 576                 break;
 577         case ZFS_PROP_RELATIME:
 578                 if (vfsp->vfs_do_relatime)
 579                         tmp = vfsp->vfs_relatime;
 580                 break;
 581         case ZFS_PROP_DEVICES:
 582                 if (vfsp->vfs_do_devices)
 583                         tmp = vfsp->vfs_devices;
 584                 break;
 585         case ZFS_PROP_EXEC:
 586                 if (vfsp->vfs_do_exec)
 587                         tmp = vfsp->vfs_exec;
 588                 break;
 589         case ZFS_PROP_SETUID:
 590                 if (vfsp->vfs_do_setuid)
 591                         tmp = vfsp->vfs_setuid;
 592                 break;
 593         case ZFS_PROP_READONLY:
 594                 if (vfsp->vfs_do_readonly)
 595                         tmp = vfsp->vfs_readonly;
 596                 break;
 597         case ZFS_PROP_XATTR:
 598                 if (vfsp->vfs_do_xattr)
 599                         tmp = vfsp->vfs_xattr;
 600                 break;
 601         case ZFS_PROP_NBMAND:
 602                 if (vfsp->vfs_do_nbmand)
 603                         tmp = vfsp->vfs_nbmand;
 604                 break;
 605         default:
 606                 return (ENOENT);
 607         }
 608
 609         if (tmp != *val) {
 610                 (void) strcpy(setpoint, "temporary");
 611                 *val = tmp;
 612         }
 613         return (0);
 614 }
 615
 616 /*
 617  * Associate this zfsvfs with the given objset, which must be owned.
 618  * This will cache a bunch of on-disk state from the objset in the
 619  * zfsvfs.
 620  */
 621 static int
 622 zfsvfs_init(zfsvfs_t *zfsvfs, objset_t *os)
 623 {
 624         int error;
 625         uint64_t val;
 626
 627         zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE;
 628         zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
 629         zfsvfs->z_os = os;
 630
 631         error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
 632         if (error != 0)
 633                 return (error);
 634         if (zfsvfs->z_version >
 635             zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
 636                 (void) printk("Can't mount a version %lld file system "
 637                     "on a version %lld pool\n. Pool must be upgraded to mount "
 638                     "this file system.\n", (u_longlong_t)zfsvfs->z_version,
 639                     (u_longlong_t)spa_version(dmu_objset_spa(os)));
 640                 return (SET_ERROR(ENOTSUP));
 641         }
 642         error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &val);
 643         if (error != 0)
 644                 return (error);
 645         zfsvfs->z_norm = (int)val;
 646
 647         error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &val);
 648         if (error != 0)
 649                 return (error);
 650         zfsvfs->z_utf8 = (val != 0);
 651
 652         error = zfs_get_zplprop(os, ZFS_PROP_CASE, &val);
 653         if (error != 0)
 654                 return (error);
 655         zfsvfs->z_case = (uint_t)val;
 656
 657         if ((error = zfs_get_zplprop(os, ZFS_PROP_ACLTYPE, &val)) != 0)
 658                 return (error);
 659         zfsvfs->z_acl_type = (uint_t)val;
 660
 661         /*
 662          * Fold case on file systems that are always or sometimes case
 663          * insensitive.
 664          */
 665         if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
 666             zfsvfs->z_case == ZFS_CASE_MIXED)
 667                 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
 668
 669         zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
 670         zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
 671
 672         uint64_t sa_obj = 0;
 673         if (zfsvfs->z_use_sa) {
 674                 /* should either have both of these objects or none */
 675                 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
 676                     &sa_obj);
 677                 if (error != 0)
 678                         return (error);
 679
 680                 error = zfs_get_zplprop(os, ZFS_PROP_XATTR, &val);
 681                 if ((error == 0) && (val == ZFS_XATTR_SA))
 682                         zfsvfs->z_xattr_sa = B_TRUE;
 683         }
 684
 685         error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
 686             &zfsvfs->z_root);
 687         if (error != 0)
 688                 return (error);
 689         ASSERT(zfsvfs->z_root != 0);
 690
 691         error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
 692             &zfsvfs->z_unlinkedobj);
 693         if (error != 0)
 694                 return (error);
 695
 696         error = zap_lookup(os, MASTER_NODE_OBJ,
 697             zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
 698             8, 1, &zfsvfs->z_userquota_obj);
 699         if (error == ENOENT)
 700                 zfsvfs->z_userquota_obj = 0;
 701         else if (error != 0)
 702                 return (error);
 703
 704         error = zap_lookup(os, MASTER_NODE_OBJ,
 705             zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
 706             8, 1, &zfsvfs->z_groupquota_obj);
 707         if (error == ENOENT)
 708                 zfsvfs->z_groupquota_obj = 0;
 709         else if (error != 0)
 710                 return (error);
 711
 712         error = zap_lookup(os, MASTER_NODE_OBJ,
 713             zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTQUOTA],
 714             8, 1, &zfsvfs->z_projectquota_obj);
 715         if (error == ENOENT)
 716                 zfsvfs->z_projectquota_obj = 0;
 717         else if (error != 0)
 718                 return (error);
 719
 720         error = zap_lookup(os, MASTER_NODE_OBJ,
 721             zfs_userquota_prop_prefixes[ZFS_PROP_USEROBJQUOTA],
 722             8, 1, &zfsvfs->z_userobjquota_obj);
 723         if (error == ENOENT)
 724                 zfsvfs->z_userobjquota_obj = 0;
 725         else if (error != 0)
 726                 return (error);
 727
 728         error = zap_lookup(os, MASTER_NODE_OBJ,
 729             zfs_userquota_prop_prefixes[ZFS_PROP_GROUPOBJQUOTA],
 730             8, 1, &zfsvfs->z_groupobjquota_obj);
 731         if (error == ENOENT)
 732                 zfsvfs->z_groupobjquota_obj = 0;
 733         else if (error != 0)
 734                 return (error);
 735
 736         error = zap_lookup(os, MASTER_NODE_OBJ,
 737             zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTOBJQUOTA],
 738             8, 1, &zfsvfs->z_projectobjquota_obj);
 739         if (error == ENOENT)
 740                 zfsvfs->z_projectobjquota_obj = 0;
 741         else if (error != 0)
 742                 return (error);
 743
 744         error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
 745             &zfsvfs->z_fuid_obj);
 746         if (error == ENOENT)
 747                 zfsvfs->z_fuid_obj = 0;
 748         else if (error != 0)
 749                 return (error);
 750
 751         error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
 752             &zfsvfs->z_shares_dir);
 753         if (error == ENOENT)
 754                 zfsvfs->z_shares_dir = 0;
 755         else if (error != 0)
 756                 return (error);
 757
 758         error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
 759             &zfsvfs->z_attr_table);
 760         if (error != 0)
 761                 return (error);
 762
 763         if (zfsvfs->z_version >= ZPL_VERSION_SA)
 764                 sa_register_update_callback(os, zfs_sa_upgrade);
 765
 766         return (0);
 767 }
 768
 769 int
 770 zfsvfs_create(const char *osname, boolean_t readonly, zfsvfs_t **zfvp)
 771 {
 772         objset_t *os;
 773         zfsvfs_t *zfsvfs;
 774         int error;
 775         boolean_t ro = (readonly || (strchr(osname, '@') != NULL));
 776
 777         zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
 778
 779         error = dmu_objset_own(osname, DMU_OST_ZFS, ro, B_TRUE, zfsvfs, &os);
 780         if (error != 0) {
 781                 kmem_free(zfsvfs, sizeof (zfsvfs_t));
 782                 return (error);
 783         }
 784
 785         error = zfsvfs_create_impl(zfvp, zfsvfs, os);
 786         if (error != 0) {
 787                 dmu_objset_disown(os, B_TRUE, zfsvfs);
 788         }
 789         return (error);
 790 }
 791
 792
 793 /*
 794  * Note: zfsvfs is assumed to be malloc'd, and will be freed by this function
 795  * on a failure.  Do not pass in a statically allocated zfsvfs.
 796  */
 797 int
 798 zfsvfs_create_impl(zfsvfs_t **zfvp, zfsvfs_t *zfsvfs, objset_t *os)
 799 {
 800         int error;
 801
 802         zfsvfs->z_vfs = NULL;
 803         zfsvfs->z_sb = NULL;
 804         zfsvfs->z_parent = zfsvfs;
 805
 806         mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
 807         mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
 808         list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
 809             offsetof(znode_t, z_link_node));
 810         ZFS_TEARDOWN_INIT(zfsvfs);
 811         rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
 812         rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
 813
 814         int size = MIN(1 << (highbit64(zfs_object_mutex_size) - 1),
 815             ZFS_OBJ_MTX_MAX);
 816         zfsvfs->z_hold_size = size;
 817         zfsvfs->z_hold_trees = vmem_zalloc(sizeof (avl_tree_t) * size,
 818             KM_SLEEP);
 819         zfsvfs->z_hold_locks = vmem_zalloc(sizeof (kmutex_t) * size, KM_SLEEP);
 820         for (int i = 0; i != size; i++) {
 821                 avl_create(&zfsvfs->z_hold_trees[i], zfs_znode_hold_compare,
 822                     sizeof (znode_hold_t), offsetof(znode_hold_t, zh_node));
 823                 mutex_init(&zfsvfs->z_hold_locks[i], NULL, MUTEX_DEFAULT, NULL);
 824         }
 825
 826         error = zfsvfs_init(zfsvfs, os);
 827         if (error != 0) {
 828                 *zfvp = NULL;
 829                 zfsvfs_free(zfsvfs);
 830                 return (error);
 831         }
 832
 833         zfsvfs->z_drain_task = TASKQID_INVALID;
 834         zfsvfs->z_draining = B_FALSE;
 835         zfsvfs->z_drain_cancel = B_TRUE;
 836
 837         *zfvp = zfsvfs;
 838         return (0);
 839 }
 840
 841 static int
 842 zfsvfs_setup(zfsvfs_t *zfsvfs, boolean_t mounting)
 843 {
 844         int error;
 845         boolean_t readonly = zfs_is_readonly(zfsvfs);
 846
 847         error = zfs_register_callbacks(zfsvfs->z_vfs);
 848         if (error)
 849                 return (error);
 850
 851         zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data);
 852
 853         /*
 854          * If we are not mounting (ie: online recv), then we don't
 855          * have to worry about replaying the log as we blocked all
 856          * operations out since we closed the ZIL.
 857          */
 858         if (mounting) {
 859                 ASSERT3P(zfsvfs->z_kstat.dk_kstats, ==, NULL);
 860                 dataset_kstats_create(&zfsvfs->z_kstat, zfsvfs->z_os);
 861
 862                 /*
 863                  * During replay we remove the read only flag to
 864                  * allow replays to succeed.
 865                  */
 866                 if (readonly != 0) {
 867                         readonly_changed_cb(zfsvfs, B_FALSE);
 868                 } else {
 869                         zap_stats_t zs;
 870                         if (zap_get_stats(zfsvfs->z_os, zfsvfs->z_unlinkedobj,
 871                             &zs) == 0) {
 872                                 dataset_kstats_update_nunlinks_kstat(
 873                                     &zfsvfs->z_kstat, zs.zs_num_entries);
 874                                 dprintf_ds(zfsvfs->z_os->os_dsl_dataset,
 875                                     "num_entries in unlinked set: %llu",
 876                                     zs.zs_num_entries);
 877                         }
 878                         zfs_unlinked_drain(zfsvfs);
 879                         dsl_dir_t *dd = zfsvfs->z_os->os_dsl_dataset->ds_dir;
 880                         dd->dd_activity_cancelled = B_FALSE;
 881                 }
 882
 883                 /*
 884                  * Parse and replay the intent log.
 885                  *
 886                  * Because of ziltest, this must be done after
 887                  * zfs_unlinked_drain().  (Further note: ziltest
 888                  * doesn't use readonly mounts, where
 889                  * zfs_unlinked_drain() isn't called.)  This is because
 890                  * ziltest causes spa_sync() to think it's committed,
 891                  * but actually it is not, so the intent log contains
 892                  * many txg's worth of changes.
 893                  *
 894                  * In particular, if object N is in the unlinked set in
 895                  * the last txg to actually sync, then it could be
 896                  * actually freed in a later txg and then reallocated
 897                  * in a yet later txg.  This would write a "create
 898                  * object N" record to the intent log.  Normally, this
 899                  * would be fine because the spa_sync() would have
 900                  * written out the fact that object N is free, before
 901                  * we could write the "create object N" intent log
 902                  * record.
 903                  *
 904                  * But when we are in ziltest mode, we advance the "open
 905                  * txg" without actually spa_sync()-ing the changes to
 906                  * disk.  So we would see that object N is still
 907                  * allocated and in the unlinked set, and there is an
 908                  * intent log record saying to allocate it.
 909                  */
 910                 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
 911                         if (zil_replay_disable) {
 912                                 zil_destroy(zfsvfs->z_log, B_FALSE);
 913                         } else {
 914                                 zfsvfs->z_replay = B_TRUE;
 915                                 zil_replay(zfsvfs->z_os, zfsvfs,
 916                                     zfs_replay_vector);
 917                                 zfsvfs->z_replay = B_FALSE;
 918                         }
 919                 }
 920
 921                 /* restore readonly bit */
 922                 if (readonly != 0)
 923                         readonly_changed_cb(zfsvfs, B_TRUE);
 924         }
 925
 926         /*
 927          * Set the objset user_ptr to track its zfsvfs.
 928          */
 929         mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
 930         dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
 931         mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
 932
 933         return (0);
 934 }
 935
 936 void
 937 zfsvfs_free(zfsvfs_t *zfsvfs)
 938 {
 939         int i, size = zfsvfs->z_hold_size;
 940
 941         zfs_fuid_destroy(zfsvfs);
 942
 943         mutex_destroy(&zfsvfs->z_znodes_lock);
 944         mutex_destroy(&zfsvfs->z_lock);
 945         list_destroy(&zfsvfs->z_all_znodes);
 946         ZFS_TEARDOWN_DESTROY(zfsvfs);
 947         rw_destroy(&zfsvfs->z_teardown_inactive_lock);
 948         rw_destroy(&zfsvfs->z_fuid_lock);
 949         for (i = 0; i != size; i++) {
 950                 avl_destroy(&zfsvfs->z_hold_trees[i]);
 951                 mutex_destroy(&zfsvfs->z_hold_locks[i]);
 952         }
 953         vmem_free(zfsvfs->z_hold_trees, sizeof (avl_tree_t) * size);
 954         vmem_free(zfsvfs->z_hold_locks, sizeof (kmutex_t) * size);
 955         zfsvfs_vfs_free(zfsvfs->z_vfs);
 956         dataset_kstats_destroy(&zfsvfs->z_kstat);
 957         kmem_free(zfsvfs, sizeof (zfsvfs_t));
 958 }
 959
 960 static void
 961 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
 962 {
 963         zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
 964         zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
 965 }
 966
 967 static void
 968 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
 969 {
 970         objset_t *os = zfsvfs->z_os;
 971
 972         if (!dmu_objset_is_snapshot(os))
 973                 dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs);
 974 }
 975
 976 #ifdef HAVE_MLSLABEL
 977 /*
 978  * Check that the hex label string is appropriate for the dataset being
 979  * mounted into the global_zone proper.
 980  *
 981  * Return an error if the hex label string is not default or
 982  * admin_low/admin_high.  For admin_low labels, the corresponding
 983  * dataset must be readonly.
 984  */
 985 int
 986 zfs_check_global_label(const char *dsname, const char *hexsl)
 987 {
 988         if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
 989                 return (0);
 990         if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
 991                 return (0);
 992         if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
 993                 /* must be readonly */
 994                 uint64_t rdonly;
 995
 996                 if (dsl_prop_get_integer(dsname,
 997                     zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
 998                         return (SET_ERROR(EACCES));
 999                 return (rdonly ? 0 : SET_ERROR(EACCES));
1000         }
1001         return (SET_ERROR(EACCES));
1002 }
1003 #endif /* HAVE_MLSLABEL */
1004
1005 static int
1006 zfs_statfs_project(zfsvfs_t *zfsvfs, znode_t *zp, struct kstatfs *statp,
1007     uint32_t bshift)
1008 {
1009         char buf[20 + DMU_OBJACCT_PREFIX_LEN];
1010         uint64_t offset = DMU_OBJACCT_PREFIX_LEN;
1011         uint64_t quota;
1012         uint64_t used;
1013         int err;
1014
1015         strlcpy(buf, DMU_OBJACCT_PREFIX, DMU_OBJACCT_PREFIX_LEN + 1);
1016         err = zfs_id_to_fuidstr(zfsvfs, NULL, zp->z_projid, buf + offset,
1017             sizeof (buf) - offset, B_FALSE);
1018         if (err)
1019                 return (err);
1020
1021         if (zfsvfs->z_projectquota_obj == 0)
1022                 goto objs;
1023
1024         err = zap_lookup(zfsvfs->z_os, zfsvfs->z_projectquota_obj,
1025             buf + offset, 8, 1, &quota);
1026         if (err == ENOENT)
1027                 goto objs;
1028         else if (err)
1029                 return (err);
1030
1031         err = zap_lookup(zfsvfs->z_os, DMU_PROJECTUSED_OBJECT,
1032             buf + offset, 8, 1, &used);
1033         if (unlikely(err == ENOENT)) {
1034                 uint32_t blksize;
1035                 u_longlong_t nblocks;
1036
1037                 /*
1038                  * Quota accounting is async, so it is possible race case.
1039                  * There is at least one object with the given project ID.
1040                  */
1041                 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
1042                 if (unlikely(zp->z_blksz == 0))
1043                         blksize = zfsvfs->z_max_blksz;
1044
1045                 used = blksize * nblocks;
1046         } else if (err) {
1047                 return (err);
1048         }
1049
1050         statp->f_blocks = quota >> bshift;
1051         statp->f_bfree = (quota > used) ? ((quota - used) >> bshift) : 0;
1052         statp->f_bavail = statp->f_bfree;
1053
1054 objs:
1055         if (zfsvfs->z_projectobjquota_obj == 0)
1056                 return (0);
1057
1058         err = zap_lookup(zfsvfs->z_os, zfsvfs->z_projectobjquota_obj,
1059             buf + offset, 8, 1, &quota);
1060         if (err == ENOENT)
1061                 return (0);
1062         else if (err)
1063                 return (err);
1064
1065         err = zap_lookup(zfsvfs->z_os, DMU_PROJECTUSED_OBJECT,
1066             buf, 8, 1, &used);
1067         if (unlikely(err == ENOENT)) {
1068                 /*
1069                  * Quota accounting is async, so it is possible race case.
1070                  * There is at least one object with the given project ID.
1071                  */
1072                 used = 1;
1073         } else if (err) {
1074                 return (err);
1075         }
1076
1077         statp->f_files = quota;
1078         statp->f_ffree = (quota > used) ? (quota - used) : 0;
1079
1080         return (0);
1081 }
1082
1083 int
1084 zfs_statvfs(struct inode *ip, struct kstatfs *statp)
1085 {
1086         zfsvfs_t *zfsvfs = ITOZSB(ip);
1087         uint64_t refdbytes, availbytes, usedobjs, availobjs;
1088         int err = 0;
1089
1090         ZFS_ENTER(zfsvfs);
1091
1092         dmu_objset_space(zfsvfs->z_os,
1093             &refdbytes, &availbytes, &usedobjs, &availobjs);
1094
1095         uint64_t fsid = dmu_objset_fsid_guid(zfsvfs->z_os);
1096         /*
1097          * The underlying storage pool actually uses multiple block
1098          * size.  Under Solaris frsize (fragment size) is reported as
1099          * the smallest block size we support, and bsize (block size)
1100          * as the filesystem's maximum block size.  Unfortunately,
1101          * under Linux the fragment size and block size are often used
1102          * interchangeably.  Thus we are forced to report both of them
1103          * as the filesystem's maximum block size.
1104          */
1105         statp->f_frsize = zfsvfs->z_max_blksz;
1106         statp->f_bsize = zfsvfs->z_max_blksz;
1107         uint32_t bshift = fls(statp->f_bsize) - 1;
1108
1109         /*
1110          * The following report "total" blocks of various kinds in
1111          * the file system, but reported in terms of f_bsize - the
1112          * "preferred" size.
1113          */
1114
1115         /* Round up so we never have a filesystem using 0 blocks. */
1116         refdbytes = P2ROUNDUP(refdbytes, statp->f_bsize);
1117         statp->f_blocks = (refdbytes + availbytes) >> bshift;
1118         statp->f_bfree = availbytes >> bshift;
1119         statp->f_bavail = statp->f_bfree; /* no root reservation */
1120
1121         /*
1122          * statvfs() should really be called statufs(), because it assumes
1123          * static metadata.  ZFS doesn't preallocate files, so the best
1124          * we can do is report the max that could possibly fit in f_files,
1125          * and that minus the number actually used in f_ffree.
1126          * For f_ffree, report the smaller of the number of objects available
1127          * and the number of blocks (each object will take at least a block).
1128          */
1129         statp->f_ffree = MIN(availobjs, availbytes >> DNODE_SHIFT);
1130         statp->f_files = statp->f_ffree + usedobjs;
1131         statp->f_fsid.val[0] = (uint32_t)fsid;
1132         statp->f_fsid.val[1] = (uint32_t)(fsid >> 32);
1133         statp->f_type = ZFS_SUPER_MAGIC;
1134         statp->f_namelen = MAXNAMELEN - 1;
1135
1136         /*
1137          * We have all of 40 characters to stuff a string here.
1138          * Is there anything useful we could/should provide?
1139          */
1140         memset(statp->f_spare, 0, sizeof (statp->f_spare));
1141
1142         if (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
1143             dmu_objset_projectquota_present(zfsvfs->z_os)) {
1144                 znode_t *zp = ITOZ(ip);
1145
1146                 if (zp->z_pflags & ZFS_PROJINHERIT && zp->z_projid &&
1147                     zpl_is_valid_projid(zp->z_projid))
1148                         err = zfs_statfs_project(zfsvfs, zp, statp, bshift);
1149         }
1150
1151         ZFS_EXIT(zfsvfs);
1152         return (err);
1153 }
1154
1155 static int
1156 zfs_root(zfsvfs_t *zfsvfs, struct inode **ipp)
1157 {
1158         znode_t *rootzp;
1159         int error;
1160
1161         ZFS_ENTER(zfsvfs);
1162
1163         error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1164         if (error == 0)
1165                 *ipp = ZTOI(rootzp);
1166
1167         ZFS_EXIT(zfsvfs);
1168         return (error);
1169 }
1170
1171 /*
1172  * Linux kernels older than 3.1 do not support a per-filesystem shrinker.
1173  * To accommodate this we must improvise and manually walk the list of znodes
1174  * attempting to prune dentries in order to be able to drop the inodes.
1175  *
1176  * To avoid scanning the same znodes multiple times they are always rotated
1177  * to the end of the z_all_znodes list.  New znodes are inserted at the
1178  * end of the list so we're always scanning the oldest znodes first.
1179  */
1180 static int
1181 zfs_prune_aliases(zfsvfs_t *zfsvfs, unsigned long nr_to_scan)
1182 {
1183         znode_t **zp_array, *zp;
1184         int max_array = MIN(nr_to_scan, PAGE_SIZE * 8 / sizeof (znode_t *));
1185         int objects = 0;
1186         int i = 0, j = 0;
1187
1188         zp_array = kmem_zalloc(max_array * sizeof (znode_t *), KM_SLEEP);
1189
1190         mutex_enter(&zfsvfs->z_znodes_lock);
1191         while ((zp = list_head(&zfsvfs->z_all_znodes)) != NULL) {
1192
1193                 if ((i++ > nr_to_scan) || (j >= max_array))
1194                         break;
1195
1196                 ASSERT(list_link_active(&zp->z_link_node));
1197                 list_remove(&zfsvfs->z_all_znodes, zp);
1198                 list_insert_tail(&zfsvfs->z_all_znodes, zp);
1199
1200                 /* Skip active znodes and .zfs entries */
1201                 if (MUTEX_HELD(&zp->z_lock) || zp->z_is_ctldir)
1202                         continue;
1203
1204                 if (igrab(ZTOI(zp)) == NULL)
1205                         continue;
1206
1207                 zp_array[j] = zp;
1208                 j++;
1209         }
1210         mutex_exit(&zfsvfs->z_znodes_lock);
1211
1212         for (i = 0; i < j; i++) {
1213                 zp = zp_array[i];
1214
1215                 ASSERT3P(zp, !=, NULL);
1216                 d_prune_aliases(ZTOI(zp));
1217
1218                 if (atomic_read(&ZTOI(zp)->i_count) == 1)
1219                         objects++;
1220
1221                 zrele(zp);
1222         }
1223
1224         kmem_free(zp_array, max_array * sizeof (znode_t *));
1225
1226         return (objects);
1227 }
1228
1229 /*
1230  * The ARC has requested that the filesystem drop entries from the dentry
1231  * and inode caches.  This can occur when the ARC needs to free meta data
1232  * blocks but can't because they are all pinned by entries in these caches.
1233  */
1234 int
1235 zfs_prune(struct super_block *sb, unsigned long nr_to_scan, int *objects)
1236 {
1237         zfsvfs_t *zfsvfs = sb->s_fs_info;
1238         int error = 0;
1239         struct shrinker *shrinker = &sb->s_shrink;
1240         struct shrink_control sc = {
1241                 .nr_to_scan = nr_to_scan,
1242                 .gfp_mask = GFP_KERNEL,
1243         };
1244
1245         ZFS_ENTER(zfsvfs);
1246
1247 #if defined(HAVE_SPLIT_SHRINKER_CALLBACK) && \
1248         defined(SHRINK_CONTROL_HAS_NID) && \
1249         defined(SHRINKER_NUMA_AWARE)
1250         if (sb->s_shrink.flags & SHRINKER_NUMA_AWARE) {
1251                 *objects = 0;
1252                 for_each_online_node(sc.nid) {
1253                         *objects += (*shrinker->scan_objects)(shrinker, &sc);
1254                         /*
1255                          * reset sc.nr_to_scan, modified by
1256                          * scan_objects == super_cache_scan
1257                          */
1258                         sc.nr_to_scan = nr_to_scan;
1259                 }
1260         } else {
1261                         *objects = (*shrinker->scan_objects)(shrinker, &sc);
1262         }
1263
1264 #elif defined(HAVE_SPLIT_SHRINKER_CALLBACK)
1265         *objects = (*shrinker->scan_objects)(shrinker, &sc);
1266 #elif defined(HAVE_SINGLE_SHRINKER_CALLBACK)
1267         *objects = (*shrinker->shrink)(shrinker, &sc);
1268 #elif defined(HAVE_D_PRUNE_ALIASES)
1269 #define D_PRUNE_ALIASES_IS_DEFAULT
1270         *objects = zfs_prune_aliases(zfsvfs, nr_to_scan);
1271 #else
1272 #error "No available dentry and inode cache pruning mechanism."
1273 #endif
1274
1275 #if defined(HAVE_D_PRUNE_ALIASES) && !defined(D_PRUNE_ALIASES_IS_DEFAULT)
1276 #undef  D_PRUNE_ALIASES_IS_DEFAULT
1277         /*
1278          * Fall back to zfs_prune_aliases if the kernel's per-superblock
1279          * shrinker couldn't free anything, possibly due to the inodes being
1280          * allocated in a different memcg.
1281          */
1282         if (*objects == 0)
1283                 *objects = zfs_prune_aliases(zfsvfs, nr_to_scan);
1284 #endif
1285
1286         ZFS_EXIT(zfsvfs);
1287
1288         dprintf_ds(zfsvfs->z_os->os_dsl_dataset,
1289             "pruning, nr_to_scan=%lu objects=%d error=%d\n",
1290             nr_to_scan, *objects, error);
1291
1292         return (error);
1293 }
1294
1295 /*
1296  * Teardown the zfsvfs_t.
1297  *
1298  * Note, if 'unmounting' is FALSE, we return with the 'z_teardown_lock'
1299  * and 'z_teardown_inactive_lock' held.
1300  */
1301 static int
1302 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1303 {
1304         znode_t *zp;
1305
1306         zfs_unlinked_drain_stop_wait(zfsvfs);
1307
1308         /*
1309          * If someone has not already unmounted this file system,
1310          * drain the zrele_taskq to ensure all active references to the
1311          * zfsvfs_t have been handled only then can it be safely destroyed.
1312          */
1313         if (zfsvfs->z_os) {
1314                 /*
1315                  * If we're unmounting we have to wait for the list to
1316                  * drain completely.
1317                  *
1318                  * If we're not unmounting there's no guarantee the list
1319                  * will drain completely, but iputs run from the taskq
1320                  * may add the parents of dir-based xattrs to the taskq
1321                  * so we want to wait for these.
1322                  *
1323                  * We can safely read z_nr_znodes without locking because the
1324                  * VFS has already blocked operations which add to the
1325                  * z_all_znodes list and thus increment z_nr_znodes.
1326                  */
1327                 int round = 0;
1328                 while (zfsvfs->z_nr_znodes > 0) {
1329                         taskq_wait_outstanding(dsl_pool_zrele_taskq(
1330                             dmu_objset_pool(zfsvfs->z_os)), 0);
1331                         if (++round > 1 && !unmounting)
1332                                 break;
1333                 }
1334         }
1335
1336         ZFS_TEARDOWN_ENTER_WRITE(zfsvfs, FTAG);
1337
1338         if (!unmounting) {
1339                 /*
1340                  * We purge the parent filesystem's super block as the
1341                  * parent filesystem and all of its snapshots have their
1342                  * inode's super block set to the parent's filesystem's
1343                  * super block.  Note,  'z_parent' is self referential
1344                  * for non-snapshots.
1345                  */
1346                 shrink_dcache_sb(zfsvfs->z_parent->z_sb);
1347         }
1348
1349         /*
1350          * Close the zil. NB: Can't close the zil while zfs_inactive
1351          * threads are blocked as zil_close can call zfs_inactive.
1352          */
1353         if (zfsvfs->z_log) {
1354                 zil_close(zfsvfs->z_log);
1355                 zfsvfs->z_log = NULL;
1356         }
1357
1358         rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1359
1360         /*
1361          * If we are not unmounting (ie: online recv) and someone already
1362          * unmounted this file system while we were doing the switcheroo,
1363          * or a reopen of z_os failed then just bail out now.
1364          */
1365         if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1366                 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1367                 ZFS_TEARDOWN_EXIT(zfsvfs, FTAG);
1368                 return (SET_ERROR(EIO));
1369         }
1370
1371         /*
1372          * At this point there are no VFS ops active, and any new VFS ops
1373          * will fail with EIO since we have z_teardown_lock for writer (only
1374          * relevant for forced unmount).
1375          *
1376          * Release all holds on dbufs. We also grab an extra reference to all
1377          * the remaining inodes so that the kernel does not attempt to free
1378          * any inodes of a suspended fs. This can cause deadlocks since the
1379          * zfs_resume_fs() process may involve starting threads, which might
1380          * attempt to free unreferenced inodes to free up memory for the new
1381          * thread.
1382          */
1383         if (!unmounting) {
1384                 mutex_enter(&zfsvfs->z_znodes_lock);
1385                 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1386                     zp = list_next(&zfsvfs->z_all_znodes, zp)) {
1387                         if (zp->z_sa_hdl)
1388                                 zfs_znode_dmu_fini(zp);
1389                         if (igrab(ZTOI(zp)) != NULL)
1390                                 zp->z_suspended = B_TRUE;
1391
1392                 }
1393                 mutex_exit(&zfsvfs->z_znodes_lock);
1394         }
1395
1396         /*
1397          * If we are unmounting, set the unmounted flag and let new VFS ops
1398          * unblock.  zfs_inactive will have the unmounted behavior, and all
1399          * other VFS ops will fail with EIO.
1400          */
1401         if (unmounting) {
1402                 zfsvfs->z_unmounted = B_TRUE;
1403                 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1404                 ZFS_TEARDOWN_EXIT(zfsvfs, FTAG);
1405         }
1406
1407         /*
1408          * z_os will be NULL if there was an error in attempting to reopen
1409          * zfsvfs, so just return as the properties had already been
1410          *
1411          * unregistered and cached data had been evicted before.
1412          */
1413         if (zfsvfs->z_os == NULL)
1414                 return (0);
1415
1416         /*
1417          * Unregister properties.
1418          */
1419         zfs_unregister_callbacks(zfsvfs);
1420
1421         /*
1422          * Evict cached data. We must write out any dirty data before
1423          * disowning the dataset.
1424          */
1425         objset_t *os = zfsvfs->z_os;
1426         boolean_t os_dirty = B_FALSE;
1427         for (int t = 0; t < TXG_SIZE; t++) {
1428                 if (dmu_objset_is_dirty(os, t)) {
1429                         os_dirty = B_TRUE;
1430                         break;
1431                 }
1432         }
1433         if (!zfs_is_readonly(zfsvfs) && os_dirty) {
1434                 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1435         }
1436         dmu_objset_evict_dbufs(zfsvfs->z_os);
1437         dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1438         dsl_dir_cancel_waiters(dd);
1439
1440         return (0);
1441 }
1442
1443 #if defined(HAVE_SUPER_SETUP_BDI_NAME)
1444 atomic_long_t zfs_bdi_seq = ATOMIC_LONG_INIT(0);
1445 #endif
1446
1447 int
1448 zfs_domount(struct super_block *sb, zfs_mnt_t *zm, int silent)
1449 {
1450         const char *osname = zm->mnt_osname;
1451         struct inode *root_inode = NULL;
1452         uint64_t recordsize;
1453         int error = 0;
1454         zfsvfs_t *zfsvfs = NULL;
1455         vfs_t *vfs = NULL;
1456
1457         ASSERT(zm);
1458         ASSERT(osname);
1459
1460         error = zfsvfs_parse_options(zm->mnt_data, &vfs);
1461         if (error)
1462                 return (error);
1463
1464         error = zfsvfs_create(osname, vfs->vfs_readonly, &zfsvfs);
1465         if (error) {
1466                 zfsvfs_vfs_free(vfs);
1467                 goto out;
1468         }
1469
1470         if ((error = dsl_prop_get_integer(osname, "recordsize",
1471             &recordsize, NULL))) {
1472                 zfsvfs_vfs_free(vfs);
1473                 goto out;
1474         }
1475
1476         vfs->vfs_data = zfsvfs;
1477         zfsvfs->z_vfs = vfs;
1478         zfsvfs->z_sb = sb;
1479         sb->s_fs_info = zfsvfs;
1480         sb->s_magic = ZFS_SUPER_MAGIC;
1481         sb->s_maxbytes = MAX_LFS_FILESIZE;
1482         sb->s_time_gran = 1;
1483         sb->s_blocksize = recordsize;
1484         sb->s_blocksize_bits = ilog2(recordsize);
1485
1486         error = -zpl_bdi_setup(sb, "zfs");
1487         if (error)
1488                 goto out;
1489
1490         sb->s_bdi->ra_pages = 0;
1491
1492         /* Set callback operations for the file system. */
1493         sb->s_op = &zpl_super_operations;
1494         sb->s_xattr = zpl_xattr_handlers;
1495         sb->s_export_op = &zpl_export_operations;
1496         sb->s_d_op = &zpl_dentry_operations;
1497
1498         /* Set features for file system. */
1499         zfs_set_fuid_feature(zfsvfs);
1500
1501         if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1502                 uint64_t pval;
1503
1504                 atime_changed_cb(zfsvfs, B_FALSE);
1505                 readonly_changed_cb(zfsvfs, B_TRUE);
1506                 if ((error = dsl_prop_get_integer(osname,
1507                     "xattr", &pval, NULL)))
1508                         goto out;
1509                 xattr_changed_cb(zfsvfs, pval);
1510                 if ((error = dsl_prop_get_integer(osname,
1511                     "acltype", &pval, NULL)))
1512                         goto out;
1513                 acltype_changed_cb(zfsvfs, pval);
1514                 zfsvfs->z_issnap = B_TRUE;
1515                 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1516                 zfsvfs->z_snap_defer_time = jiffies;
1517
1518                 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1519                 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1520                 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1521         } else {
1522                 if ((error = zfsvfs_setup(zfsvfs, B_TRUE)))
1523                         goto out;
1524         }
1525
1526         /* Allocate a root inode for the filesystem. */
1527         error = zfs_root(zfsvfs, &root_inode);
1528         if (error) {
1529                 (void) zfs_umount(sb);
1530                 goto out;
1531         }
1532
1533         /* Allocate a root dentry for the filesystem */
1534         sb->s_root = d_make_root(root_inode);
1535         if (sb->s_root == NULL) {
1536                 (void) zfs_umount(sb);
1537                 error = SET_ERROR(ENOMEM);
1538                 goto out;
1539         }
1540
1541         if (!zfsvfs->z_issnap)
1542                 zfsctl_create(zfsvfs);
1543
1544         zfsvfs->z_arc_prune = arc_add_prune_callback(zpl_prune_sb, sb);
1545 out:
1546         if (error) {
1547                 if (zfsvfs != NULL) {
1548                         dmu_objset_disown(zfsvfs->z_os, B_TRUE, zfsvfs);
1549                         zfsvfs_free(zfsvfs);
1550                 }
1551                 /*
1552                  * make sure we don't have dangling sb->s_fs_info which
1553                  * zfs_preumount will use.
1554                  */
1555                 sb->s_fs_info = NULL;
1556         }
1557
1558         return (error);
1559 }
1560
1561 /*
1562  * Called when an unmount is requested and certain sanity checks have
1563  * already passed.  At this point no dentries or inodes have been reclaimed
1564  * from their respective caches.  We drop the extra reference on the .zfs
1565  * control directory to allow everything to be reclaimed.  All snapshots
1566  * must already have been unmounted to reach this point.
1567  */
1568 void
1569 zfs_preumount(struct super_block *sb)
1570 {
1571         zfsvfs_t *zfsvfs = sb->s_fs_info;
1572
1573         /* zfsvfs is NULL when zfs_domount fails during mount */
1574         if (zfsvfs) {
1575                 zfs_unlinked_drain_stop_wait(zfsvfs);
1576                 zfsctl_destroy(sb->s_fs_info);
1577                 /*
1578                  * Wait for zrele_async before entering evict_inodes in
1579                  * generic_shutdown_super. The reason we must finish before
1580                  * evict_inodes is when lazytime is on, or when zfs_purgedir
1581                  * calls zfs_zget, zrele would bump i_count from 0 to 1. This
1582                  * would race with the i_count check in evict_inodes. This means
1583                  * it could destroy the inode while we are still using it.
1584                  *
1585                  * We wait for two passes. xattr directories in the first pass
1586                  * may add xattr entries in zfs_purgedir, so in the second pass
1587                  * we wait for them. We don't use taskq_wait here because it is
1588                  * a pool wide taskq. Other mounted filesystems can constantly
1589                  * do zrele_async and there's no guarantee when taskq will be
1590                  * empty.
1591                  */
1592                 taskq_wait_outstanding(dsl_pool_zrele_taskq(
1593                     dmu_objset_pool(zfsvfs->z_os)), 0);
1594                 taskq_wait_outstanding(dsl_pool_zrele_taskq(
1595                     dmu_objset_pool(zfsvfs->z_os)), 0);
1596         }
1597 }
1598
1599 /*
1600  * Called once all other unmount released tear down has occurred.
1601  * It is our responsibility to release any remaining infrastructure.
1602  */
1603 int
1604 zfs_umount(struct super_block *sb)
1605 {
1606         zfsvfs_t *zfsvfs = sb->s_fs_info;
1607         objset_t *os;
1608
1609         if (zfsvfs->z_arc_prune != NULL)
1610                 arc_remove_prune_callback(zfsvfs->z_arc_prune);
1611         VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
1612         os = zfsvfs->z_os;
1613         zpl_bdi_destroy(sb);
1614
1615         /*
1616          * z_os will be NULL if there was an error in
1617          * attempting to reopen zfsvfs.
1618          */
1619         if (os != NULL) {
1620                 /*
1621                  * Unset the objset user_ptr.
1622                  */
1623                 mutex_enter(&os->os_user_ptr_lock);
1624                 dmu_objset_set_user(os, NULL);
1625                 mutex_exit(&os->os_user_ptr_lock);
1626
1627                 /*
1628                  * Finally release the objset
1629                  */
1630                 dmu_objset_disown(os, B_TRUE, zfsvfs);
1631         }
1632
1633         zfsvfs_free(zfsvfs);
1634         return (0);
1635 }
1636
1637 int
1638 zfs_remount(struct super_block *sb, int *flags, zfs_mnt_t *zm)
1639 {
1640         zfsvfs_t *zfsvfs = sb->s_fs_info;
1641         vfs_t *vfsp;
1642         boolean_t issnap = dmu_objset_is_snapshot(zfsvfs->z_os);
1643         int error;
1644
1645         if ((issnap || !spa_writeable(dmu_objset_spa(zfsvfs->z_os))) &&
1646             !(*flags & SB_RDONLY)) {
1647                 *flags |= SB_RDONLY;
1648                 return (EROFS);
1649         }
1650
1651         error = zfsvfs_parse_options(zm->mnt_data, &vfsp);
1652         if (error)
1653                 return (error);
1654
1655         if (!zfs_is_readonly(zfsvfs) && (*flags & SB_RDONLY))
1656                 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1657
1658         zfs_unregister_callbacks(zfsvfs);
1659         zfsvfs_vfs_free(zfsvfs->z_vfs);
1660
1661         vfsp->vfs_data = zfsvfs;
1662         zfsvfs->z_vfs = vfsp;
1663         if (!issnap)
1664                 (void) zfs_register_callbacks(vfsp);
1665
1666         return (error);
1667 }
1668
1669 int
1670 zfs_vget(struct super_block *sb, struct inode **ipp, fid_t *fidp)
1671 {
1672         zfsvfs_t        *zfsvfs = sb->s_fs_info;
1673         znode_t         *zp;
1674         uint64_t        object = 0;
1675         uint64_t        fid_gen = 0;
1676         uint64_t        gen_mask;
1677         uint64_t        zp_gen;
1678         int             i, err;
1679
1680         *ipp = NULL;
1681
1682         if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
1683                 zfid_short_t    *zfid = (zfid_short_t *)fidp;
1684
1685                 for (i = 0; i < sizeof (zfid->zf_object); i++)
1686                         object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
1687
1688                 for (i = 0; i < sizeof (zfid->zf_gen); i++)
1689                         fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
1690         } else {
1691                 return (SET_ERROR(EINVAL));
1692         }
1693
1694         /* LONG_FID_LEN means snapdirs */
1695         if (fidp->fid_len == LONG_FID_LEN) {
1696                 zfid_long_t     *zlfid = (zfid_long_t *)fidp;
1697                 uint64_t        objsetid = 0;
1698                 uint64_t        setgen = 0;
1699
1700                 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
1701                         objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
1702
1703                 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
1704                         setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
1705
1706                 if (objsetid != ZFSCTL_INO_SNAPDIRS - object) {
1707                         dprintf("snapdir fid: objsetid (%llu) != "
1708                             "ZFSCTL_INO_SNAPDIRS (%llu) - object (%llu)\n",
1709                             objsetid, ZFSCTL_INO_SNAPDIRS, object);
1710
1711                         return (SET_ERROR(EINVAL));
1712                 }
1713
1714                 if (fid_gen > 1 || setgen != 0) {
1715                         dprintf("snapdir fid: fid_gen (%llu) and setgen "
1716                             "(%llu)\n", fid_gen, setgen);
1717                         return (SET_ERROR(EINVAL));
1718                 }
1719
1720                 return (zfsctl_snapdir_vget(sb, objsetid, fid_gen, ipp));
1721         }
1722
1723         ZFS_ENTER(zfsvfs);
1724         /* A zero fid_gen means we are in the .zfs control directories */
1725         if (fid_gen == 0 &&
1726             (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
1727                 *ipp = zfsvfs->z_ctldir;
1728                 ASSERT(*ipp != NULL);
1729                 if (object == ZFSCTL_INO_SNAPDIR) {
1730                         VERIFY(zfsctl_root_lookup(*ipp, "snapshot", ipp,
1731                             0, kcred, NULL, NULL) == 0);
1732                 } else {
1733                         /*
1734                          * Must have an existing ref, so igrab()
1735                          * cannot return NULL
1736                          */
1737                         VERIFY3P(igrab(*ipp), !=, NULL);
1738                 }
1739                 ZFS_EXIT(zfsvfs);
1740                 return (0);
1741         }
1742
1743         gen_mask = -1ULL >> (64 - 8 * i);
1744
1745         dprintf("getting %llu [%llu mask %llx]\n", object, fid_gen, gen_mask);
1746         if ((err = zfs_zget(zfsvfs, object, &zp))) {
1747                 ZFS_EXIT(zfsvfs);
1748                 return (err);
1749         }
1750
1751         /* Don't export xattr stuff */
1752         if (zp->z_pflags & ZFS_XATTR) {
1753                 zrele(zp);
1754                 ZFS_EXIT(zfsvfs);
1755                 return (SET_ERROR(ENOENT));
1756         }
1757
1758         (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
1759             sizeof (uint64_t));
1760         zp_gen = zp_gen & gen_mask;
1761         if (zp_gen == 0)
1762                 zp_gen = 1;
1763         if ((fid_gen == 0) && (zfsvfs->z_root == object))
1764                 fid_gen = zp_gen;
1765         if (zp->z_unlinked || zp_gen != fid_gen) {
1766                 dprintf("znode gen (%llu) != fid gen (%llu)\n", zp_gen,
1767                     fid_gen);
1768                 zrele(zp);
1769                 ZFS_EXIT(zfsvfs);
1770                 return (SET_ERROR(ENOENT));
1771         }
1772
1773         *ipp = ZTOI(zp);
1774         if (*ipp)
1775                 zfs_znode_update_vfs(ITOZ(*ipp));
1776
1777         ZFS_EXIT(zfsvfs);
1778         return (0);
1779 }
1780
1781 /*
1782  * Block out VFS ops and close zfsvfs_t
1783  *
1784  * Note, if successful, then we return with the 'z_teardown_lock' and
1785  * 'z_teardown_inactive_lock' write held.  We leave ownership of the underlying
1786  * dataset and objset intact so that they can be atomically handed off during
1787  * a subsequent rollback or recv operation and the resume thereafter.
1788  */
1789 int
1790 zfs_suspend_fs(zfsvfs_t *zfsvfs)
1791 {
1792         int error;
1793
1794         if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
1795                 return (error);
1796
1797         return (0);
1798 }
1799
1800 /*
1801  * Rebuild SA and release VOPs.  Note that ownership of the underlying dataset
1802  * is an invariant across any of the operations that can be performed while the
1803  * filesystem was suspended.  Whether it succeeded or failed, the preconditions
1804  * are the same: the relevant objset and associated dataset are owned by
1805  * zfsvfs, held, and long held on entry.
1806  */
1807 int
1808 zfs_resume_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds)
1809 {
1810         int err, err2;
1811         znode_t *zp;
1812
1813         ASSERT(ZFS_TEARDOWN_WRITE_HELD(zfsvfs));
1814         ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
1815
1816         /*
1817          * We already own this, so just update the objset_t, as the one we
1818          * had before may have been evicted.
1819          */
1820         objset_t *os;
1821         VERIFY3P(ds->ds_owner, ==, zfsvfs);
1822         VERIFY(dsl_dataset_long_held(ds));
1823         dsl_pool_t *dp = spa_get_dsl(dsl_dataset_get_spa(ds));
1824         dsl_pool_config_enter(dp, FTAG);
1825         VERIFY0(dmu_objset_from_ds(ds, &os));
1826         dsl_pool_config_exit(dp, FTAG);
1827
1828         err = zfsvfs_init(zfsvfs, os);
1829         if (err != 0)
1830                 goto bail;
1831
1832         ds->ds_dir->dd_activity_cancelled = B_FALSE;
1833         VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
1834
1835         zfs_set_fuid_feature(zfsvfs);
1836         zfsvfs->z_rollback_time = jiffies;
1837
1838         /*
1839          * Attempt to re-establish all the active inodes with their
1840          * dbufs.  If a zfs_rezget() fails, then we unhash the inode
1841          * and mark it stale.  This prevents a collision if a new
1842          * inode/object is created which must use the same inode
1843          * number.  The stale inode will be be released when the
1844          * VFS prunes the dentry holding the remaining references
1845          * on the stale inode.
1846          */
1847         mutex_enter(&zfsvfs->z_znodes_lock);
1848         for (zp = list_head(&zfsvfs->z_all_znodes); zp;
1849             zp = list_next(&zfsvfs->z_all_znodes, zp)) {
1850                 err2 = zfs_rezget(zp);
1851                 if (err2) {
1852                         remove_inode_hash(ZTOI(zp));
1853                         zp->z_is_stale = B_TRUE;
1854                 }
1855
1856                 /* see comment in zfs_suspend_fs() */
1857                 if (zp->z_suspended) {
1858                         zfs_zrele_async(zp);
1859                         zp->z_suspended = B_FALSE;
1860                 }
1861         }
1862         mutex_exit(&zfsvfs->z_znodes_lock);
1863
1864         if (!zfs_is_readonly(zfsvfs) && !zfsvfs->z_unmounted) {
1865                 /*
1866                  * zfs_suspend_fs() could have interrupted freeing
1867                  * of dnodes. We need to restart this freeing so
1868                  * that we don't "leak" the space.
1869                  */
1870                 zfs_unlinked_drain(zfsvfs);
1871         }
1872
1873         /*
1874          * Most of the time zfs_suspend_fs is used for changing the contents
1875          * of the underlying dataset. ZFS rollback and receive operations
1876          * might create files for which negative dentries are present in
1877          * the cache. Since walking the dcache would require a lot of GPL-only
1878          * code duplication, it's much easier on these rather rare occasions
1879          * just to flush the whole dcache for the given dataset/filesystem.
1880          */
1881         shrink_dcache_sb(zfsvfs->z_sb);
1882
1883 bail:
1884         if (err != 0)
1885                 zfsvfs->z_unmounted = B_TRUE;
1886
1887         /* release the VFS ops */
1888         rw_exit(&zfsvfs->z_teardown_inactive_lock);
1889         ZFS_TEARDOWN_EXIT(zfsvfs, FTAG);
1890
1891         if (err != 0) {
1892                 /*
1893                  * Since we couldn't setup the sa framework, try to force
1894                  * unmount this file system.
1895                  */
1896                 if (zfsvfs->z_os)
1897                         (void) zfs_umount(zfsvfs->z_sb);
1898         }
1899         return (err);
1900 }
1901
1902 /*
1903  * Release VOPs and unmount a suspended filesystem.
1904  */
1905 int
1906 zfs_end_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds)
1907 {
1908         ASSERT(ZFS_TEARDOWN_WRITE_HELD(zfsvfs));
1909         ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
1910
1911         /*
1912          * We already own this, so just hold and rele it to update the
1913          * objset_t, as the one we had before may have been evicted.
1914          */
1915         objset_t *os;
1916         VERIFY3P(ds->ds_owner, ==, zfsvfs);
1917         VERIFY(dsl_dataset_long_held(ds));
1918         dsl_pool_t *dp = spa_get_dsl(dsl_dataset_get_spa(ds));
1919         dsl_pool_config_enter(dp, FTAG);
1920         VERIFY0(dmu_objset_from_ds(ds, &os));
1921         dsl_pool_config_exit(dp, FTAG);
1922         zfsvfs->z_os = os;
1923
1924         /* release the VOPs */
1925         rw_exit(&zfsvfs->z_teardown_inactive_lock);
1926         ZFS_TEARDOWN_EXIT(zfsvfs, FTAG);
1927
1928         /*
1929          * Try to force unmount this file system.
1930          */
1931         (void) zfs_umount(zfsvfs->z_sb);
1932         zfsvfs->z_unmounted = B_TRUE;
1933         return (0);
1934 }
1935
1936 /*
1937  * Automounted snapshots rely on periodic revalidation
1938  * to defer snapshots from being automatically unmounted.
1939  */
1940
1941 inline void
1942 zfs_exit_fs(zfsvfs_t *zfsvfs)
1943 {
1944         if (!zfsvfs->z_issnap)
1945                 return;
1946
1947         if (time_after(jiffies, zfsvfs->z_snap_defer_time +
1948             MAX(zfs_expire_snapshot * HZ / 2, HZ))) {
1949                 zfsvfs->z_snap_defer_time = jiffies;
1950                 zfsctl_snapshot_unmount_delay(zfsvfs->z_os->os_spa,
1951                     dmu_objset_id(zfsvfs->z_os),
1952                     zfs_expire_snapshot);
1953         }
1954 }
1955
1956 int
1957 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
1958 {
1959         int error;
1960         objset_t *os = zfsvfs->z_os;
1961         dmu_tx_t *tx;
1962
1963         if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
1964                 return (SET_ERROR(EINVAL));
1965
1966         if (newvers < zfsvfs->z_version)
1967                 return (SET_ERROR(EINVAL));
1968
1969         if (zfs_spa_version_map(newvers) >
1970             spa_version(dmu_objset_spa(zfsvfs->z_os)))
1971                 return (SET_ERROR(ENOTSUP));
1972
1973         tx = dmu_tx_create(os);
1974         dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
1975         if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
1976                 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
1977                     ZFS_SA_ATTRS);
1978                 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
1979         }
1980         error = dmu_tx_assign(tx, TXG_WAIT);
1981         if (error) {
1982                 dmu_tx_abort(tx);
1983                 return (error);
1984         }
1985
1986         error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
1987             8, 1, &newvers, tx);
1988
1989         if (error) {
1990                 dmu_tx_commit(tx);
1991                 return (error);
1992         }
1993
1994         if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
1995                 uint64_t sa_obj;
1996
1997                 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
1998                     SPA_VERSION_SA);
1999                 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2000                     DMU_OT_NONE, 0, tx);
2001
2002                 error = zap_add(os, MASTER_NODE_OBJ,
2003                     ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2004                 ASSERT0(error);
2005
2006                 VERIFY(0 == sa_set_sa_object(os, sa_obj));
2007                 sa_register_update_callback(os, zfs_sa_upgrade);
2008         }
2009
2010         spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2011             "from %llu to %llu", zfsvfs->z_version, newvers);
2012
2013         dmu_tx_commit(tx);
2014
2015         zfsvfs->z_version = newvers;
2016         os->os_version = newvers;
2017
2018         zfs_set_fuid_feature(zfsvfs);
2019
2020         return (0);
2021 }
2022
2023 /*
2024  * Read a property stored within the master node.
2025  */
2026 int
2027 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2028 {
2029         uint64_t *cached_copy = NULL;
2030
2031         /*
2032          * Figure out where in the objset_t the cached copy would live, if it
2033          * is available for the requested property.
2034          */
2035         if (os != NULL) {
2036                 switch (prop) {
2037                 case ZFS_PROP_VERSION:
2038                         cached_copy = &os->os_version;
2039                         break;
2040                 case ZFS_PROP_NORMALIZE:
2041                         cached_copy = &os->os_normalization;
2042                         break;
2043                 case ZFS_PROP_UTF8ONLY:
2044                         cached_copy = &os->os_utf8only;
2045                         break;
2046                 case ZFS_PROP_CASE:
2047                         cached_copy = &os->os_casesensitivity;
2048                         break;
2049                 default:
2050                         break;
2051                 }
2052         }
2053         if (cached_copy != NULL && *cached_copy != OBJSET_PROP_UNINITIALIZED) {
2054                 *value = *cached_copy;
2055                 return (0);
2056         }
2057
2058         /*
2059          * If the property wasn't cached, look up the file system's value for
2060          * the property. For the version property, we look up a slightly
2061          * different string.
2062          */
2063         const char *pname;
2064         int error = ENOENT;
2065         if (prop == ZFS_PROP_VERSION)
2066                 pname = ZPL_VERSION_STR;
2067         else
2068                 pname = zfs_prop_to_name(prop);
2069
2070         if (os != NULL) {
2071                 ASSERT3U(os->os_phys->os_type, ==, DMU_OST_ZFS);
2072                 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2073         }
2074
2075         if (error == ENOENT) {
2076                 /* No value set, use the default value */
2077                 switch (prop) {
2078                 case ZFS_PROP_VERSION:
2079                         *value = ZPL_VERSION;
2080                         break;
2081                 case ZFS_PROP_NORMALIZE:
2082                 case ZFS_PROP_UTF8ONLY:
2083                         *value = 0;
2084                         break;
2085                 case ZFS_PROP_CASE:
2086                         *value = ZFS_CASE_SENSITIVE;
2087                         break;
2088                 case ZFS_PROP_ACLTYPE:
2089                         *value = ZFS_ACLTYPE_OFF;
2090                         break;
2091                 default:
2092                         return (error);
2093                 }
2094                 error = 0;
2095         }
2096
2097         /*
2098          * If one of the methods for getting the property value above worked,
2099          * copy it into the objset_t's cache.
2100          */
2101         if (error == 0 && cached_copy != NULL) {
2102                 *cached_copy = *value;
2103         }
2104
2105         return (error);
2106 }
2107
2108 /*
2109  * Return true if the corresponding vfs's unmounted flag is set.
2110  * Otherwise return false.
2111  * If this function returns true we know VFS unmount has been initiated.
2112  */
2113 boolean_t
2114 zfs_get_vfs_flag_unmounted(objset_t *os)
2115 {
2116         zfsvfs_t *zfvp;
2117         boolean_t unmounted = B_FALSE;
2118
2119         ASSERT(dmu_objset_type(os) == DMU_OST_ZFS);
2120
2121         mutex_enter(&os->os_user_ptr_lock);
2122         zfvp = dmu_objset_get_user(os);
2123         if (zfvp != NULL && zfvp->z_unmounted)
2124                 unmounted = B_TRUE;
2125         mutex_exit(&os->os_user_ptr_lock);
2126
2127         return (unmounted);
2128 }
2129
2130 void
2131 zfsvfs_update_fromname(const char *oldname, const char *newname)
2132 {
2133         /*
2134          * We don't need to do anything here, the devname is always current by
2135          * virtue of zfsvfs->z_sb->s_op->show_devname.
2136          */
2137         (void) oldname, (void) newname;
2138 }
2139
2140 void
2141 zfs_init(void)
2142 {
2143         zfsctl_init();
2144         zfs_znode_init();
2145         dmu_objset_register_type(DMU_OST_ZFS, zpl_get_file_info);
2146         register_filesystem(&zpl_fs_type);
2147 }
2148
2149 void
2150 zfs_fini(void)
2151 {
2152         /*
2153          * we don't use outstanding because zpl_posix_acl_free might add more.
2154          */
2155         taskq_wait(system_delay_taskq);
2156         taskq_wait(system_taskq);
2157         unregister_filesystem(&zpl_fs_type);
2158         zfs_znode_fini();
2159         zfsctl_fini();
2160 }
2161
2162 #if defined(_KERNEL)
2163 EXPORT_SYMBOL(zfs_suspend_fs);
2164 EXPORT_SYMBOL(zfs_resume_fs);
2165 EXPORT_SYMBOL(zfs_set_version);
2166 EXPORT_SYMBOL(zfsvfs_create);
2167 EXPORT_SYMBOL(zfsvfs_free);
2168 EXPORT_SYMBOL(zfs_is_readonly);
2169 EXPORT_SYMBOL(zfs_domount);
2170 EXPORT_SYMBOL(zfs_preumount);
2171 EXPORT_SYMBOL(zfs_umount);
2172 EXPORT_SYMBOL(zfs_remount);
2173 EXPORT_SYMBOL(zfs_statvfs);
2174 EXPORT_SYMBOL(zfs_vget);
2175 EXPORT_SYMBOL(zfs_prune);
2176 #endif