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 https://opensource.org/licenses/CDDL-1.0.
  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/objlist.h>
  60 #include <sys/zpl.h>
  61 #include <linux/vfs_compat.h>
  62 #include "zfs_comutil.h"
  63
  64 enum {
  65         TOKEN_RO,
  66         TOKEN_RW,
  67         TOKEN_SETUID,
  68         TOKEN_NOSETUID,
  69         TOKEN_EXEC,
  70         TOKEN_NOEXEC,
  71         TOKEN_DEVICES,
  72         TOKEN_NODEVICES,
  73         TOKEN_DIRXATTR,
  74         TOKEN_SAXATTR,
  75         TOKEN_XATTR,
  76         TOKEN_NOXATTR,
  77         TOKEN_ATIME,
  78         TOKEN_NOATIME,
  79         TOKEN_RELATIME,
  80         TOKEN_NORELATIME,
  81         TOKEN_NBMAND,
  82         TOKEN_NONBMAND,
  83         TOKEN_MNTPOINT,
  84         TOKEN_LAST,
  85 };
  86
  87 static const match_table_t zpl_tokens = {
  88         { TOKEN_RO,             MNTOPT_RO },
  89         { TOKEN_RW,             MNTOPT_RW },
  90         { TOKEN_SETUID,         MNTOPT_SETUID },
  91         { TOKEN_NOSETUID,       MNTOPT_NOSETUID },
  92         { TOKEN_EXEC,           MNTOPT_EXEC },
  93         { TOKEN_NOEXEC,         MNTOPT_NOEXEC },
  94         { TOKEN_DEVICES,        MNTOPT_DEVICES },
  95         { TOKEN_NODEVICES,      MNTOPT_NODEVICES },
  96         { TOKEN_DIRXATTR,       MNTOPT_DIRXATTR },
  97         { TOKEN_SAXATTR,        MNTOPT_SAXATTR },
  98         { TOKEN_XATTR,          MNTOPT_XATTR },
  99         { TOKEN_NOXATTR,        MNTOPT_NOXATTR },
 100         { TOKEN_ATIME,          MNTOPT_ATIME },
 101         { TOKEN_NOATIME,        MNTOPT_NOATIME },
 102         { TOKEN_RELATIME,       MNTOPT_RELATIME },
 103         { TOKEN_NORELATIME,     MNTOPT_NORELATIME },
 104         { TOKEN_NBMAND,         MNTOPT_NBMAND },
 105         { TOKEN_NONBMAND,       MNTOPT_NONBMAND },
 106         { TOKEN_MNTPOINT,       MNTOPT_MNTPOINT "=%s" },
 107         { TOKEN_LAST,           NULL },
 108 };
 109
 110 static void
 111 zfsvfs_vfs_free(vfs_t *vfsp)
 112 {
 113         if (vfsp != NULL) {
 114                 if (vfsp->vfs_mntpoint != NULL)
 115                         kmem_strfree(vfsp->vfs_mntpoint);
 116
 117                 kmem_free(vfsp, sizeof (vfs_t));
 118         }
 119 }
 120
 121 static int
 122 zfsvfs_parse_option(char *option, int token, substring_t *args, vfs_t *vfsp)
 123 {
 124         switch (token) {
 125         case TOKEN_RO:
 126                 vfsp->vfs_readonly = B_TRUE;
 127                 vfsp->vfs_do_readonly = B_TRUE;
 128                 break;
 129         case TOKEN_RW:
 130                 vfsp->vfs_readonly = B_FALSE;
 131                 vfsp->vfs_do_readonly = B_TRUE;
 132                 break;
 133         case TOKEN_SETUID:
 134                 vfsp->vfs_setuid = B_TRUE;
 135                 vfsp->vfs_do_setuid = B_TRUE;
 136                 break;
 137         case TOKEN_NOSETUID:
 138                 vfsp->vfs_setuid = B_FALSE;
 139                 vfsp->vfs_do_setuid = B_TRUE;
 140                 break;
 141         case TOKEN_EXEC:
 142                 vfsp->vfs_exec = B_TRUE;
 143                 vfsp->vfs_do_exec = B_TRUE;
 144                 break;
 145         case TOKEN_NOEXEC:
 146                 vfsp->vfs_exec = B_FALSE;
 147                 vfsp->vfs_do_exec = B_TRUE;
 148                 break;
 149         case TOKEN_DEVICES:
 150                 vfsp->vfs_devices = B_TRUE;
 151                 vfsp->vfs_do_devices = B_TRUE;
 152                 break;
 153         case TOKEN_NODEVICES:
 154                 vfsp->vfs_devices = B_FALSE;
 155                 vfsp->vfs_do_devices = B_TRUE;
 156                 break;
 157         case TOKEN_DIRXATTR:
 158                 vfsp->vfs_xattr = ZFS_XATTR_DIR;
 159                 vfsp->vfs_do_xattr = B_TRUE;
 160                 break;
 161         case TOKEN_SAXATTR:
 162                 vfsp->vfs_xattr = ZFS_XATTR_SA;
 163                 vfsp->vfs_do_xattr = B_TRUE;
 164                 break;
 165         case TOKEN_XATTR:
 166                 vfsp->vfs_xattr = ZFS_XATTR_DIR;
 167                 vfsp->vfs_do_xattr = B_TRUE;
 168                 break;
 169         case TOKEN_NOXATTR:
 170                 vfsp->vfs_xattr = ZFS_XATTR_OFF;
 171                 vfsp->vfs_do_xattr = B_TRUE;
 172                 break;
 173         case TOKEN_ATIME:
 174                 vfsp->vfs_atime = B_TRUE;
 175                 vfsp->vfs_do_atime = B_TRUE;
 176                 break;
 177         case TOKEN_NOATIME:
 178                 vfsp->vfs_atime = B_FALSE;
 179                 vfsp->vfs_do_atime = B_TRUE;
 180                 break;
 181         case TOKEN_RELATIME:
 182                 vfsp->vfs_relatime = B_TRUE;
 183                 vfsp->vfs_do_relatime = B_TRUE;
 184                 break;
 185         case TOKEN_NORELATIME:
 186                 vfsp->vfs_relatime = B_FALSE;
 187                 vfsp->vfs_do_relatime = B_TRUE;
 188                 break;
 189         case TOKEN_NBMAND:
 190                 vfsp->vfs_nbmand = B_TRUE;
 191                 vfsp->vfs_do_nbmand = B_TRUE;
 192                 break;
 193         case TOKEN_NONBMAND:
 194                 vfsp->vfs_nbmand = B_FALSE;
 195                 vfsp->vfs_do_nbmand = B_TRUE;
 196                 break;
 197         case TOKEN_MNTPOINT:
 198                 vfsp->vfs_mntpoint = match_strdup(&args[0]);
 199                 if (vfsp->vfs_mntpoint == NULL)
 200                         return (SET_ERROR(ENOMEM));
 201
 202                 break;
 203         default:
 204                 break;
 205         }
 206
 207         return (0);
 208 }
 209
 210 /*
 211  * Parse the raw mntopts and return a vfs_t describing the options.
 212  */
 213 static int
 214 zfsvfs_parse_options(char *mntopts, vfs_t **vfsp)
 215 {
 216         vfs_t *tmp_vfsp;
 217         int error;
 218
 219         tmp_vfsp = kmem_zalloc(sizeof (vfs_t), KM_SLEEP);
 220
 221         if (mntopts != NULL) {
 222                 substring_t args[MAX_OPT_ARGS];
 223                 char *tmp_mntopts, *p, *t;
 224                 int token;
 225
 226                 tmp_mntopts = t = kmem_strdup(mntopts);
 227                 if (tmp_mntopts == NULL)
 228                         return (SET_ERROR(ENOMEM));
 229
 230                 while ((p = strsep(&t, ",")) != NULL) {
 231                         if (!*p)
 232                                 continue;
 233
 234                         args[0].to = args[0].from = NULL;
 235                         token = match_token(p, zpl_tokens, args);
 236                         error = zfsvfs_parse_option(p, token, args, tmp_vfsp);
 237                         if (error) {
 238                                 kmem_strfree(tmp_mntopts);
 239                                 zfsvfs_vfs_free(tmp_vfsp);
 240                                 return (error);
 241                         }
 242                 }
 243
 244                 kmem_strfree(tmp_mntopts);
 245         }
 246
 247         *vfsp = tmp_vfsp;
 248
 249         return (0);
 250 }
 251
 252 boolean_t
 253 zfs_is_readonly(zfsvfs_t *zfsvfs)
 254 {
 255         return (!!(zfsvfs->z_sb->s_flags & SB_RDONLY));
 256 }
 257
 258 int
 259 zfs_sync(struct super_block *sb, int wait, cred_t *cr)
 260 {
 261         (void) cr;
 262         zfsvfs_t *zfsvfs = sb->s_fs_info;
 263
 264         /*
 265          * Semantically, the only requirement is that the sync be initiated.
 266          * The DMU syncs out txgs frequently, so there's nothing to do.
 267          */
 268         if (!wait)
 269                 return (0);
 270
 271         if (zfsvfs != NULL) {
 272                 /*
 273                  * Sync a specific filesystem.
 274                  */
 275                 dsl_pool_t *dp;
 276                 int error;
 277
 278                 if ((error = zfs_enter(zfsvfs, FTAG)) != 0)
 279                         return (error);
 280                 dp = dmu_objset_pool(zfsvfs->z_os);
 281
 282                 /*
 283                  * If the system is shutting down, then skip any
 284                  * filesystems which may exist on a suspended pool.
 285                  */
 286                 if (spa_suspended(dp->dp_spa)) {
 287                         zfs_exit(zfsvfs, FTAG);
 288                         return (0);
 289                 }
 290
 291                 if (zfsvfs->z_log != NULL)
 292                         zil_commit(zfsvfs->z_log, 0);
 293
 294                 zfs_exit(zfsvfs, FTAG);
 295         } else {
 296                 /*
 297                  * Sync all ZFS filesystems.  This is what happens when you
 298                  * run sync(1).  Unlike other filesystems, ZFS honors the
 299                  * request by waiting for all pools to commit all dirty data.
 300                  */
 301                 spa_sync_allpools();
 302         }
 303
 304         return (0);
 305 }
 306
 307 static void
 308 atime_changed_cb(void *arg, uint64_t newval)
 309 {
 310         zfsvfs_t *zfsvfs = arg;
 311         struct super_block *sb = zfsvfs->z_sb;
 312
 313         if (sb == NULL)
 314                 return;
 315         /*
 316          * Update SB_NOATIME bit in VFS super block.  Since atime update is
 317          * determined by atime_needs_update(), atime_needs_update() needs to
 318          * return false if atime is turned off, and not unconditionally return
 319          * false if atime is turned on.
 320          */
 321         if (newval)
 322                 sb->s_flags &= ~SB_NOATIME;
 323         else
 324                 sb->s_flags |= SB_NOATIME;
 325 }
 326
 327 static void
 328 relatime_changed_cb(void *arg, uint64_t newval)
 329 {
 330         ((zfsvfs_t *)arg)->z_relatime = newval;
 331 }
 332
 333 static void
 334 xattr_changed_cb(void *arg, uint64_t newval)
 335 {
 336         zfsvfs_t *zfsvfs = arg;
 337
 338         if (newval == ZFS_XATTR_OFF) {
 339                 zfsvfs->z_flags &= ~ZSB_XATTR;
 340         } else {
 341                 zfsvfs->z_flags |= ZSB_XATTR;
 342
 343                 if (newval == ZFS_XATTR_SA)
 344                         zfsvfs->z_xattr_sa = B_TRUE;
 345                 else
 346                         zfsvfs->z_xattr_sa = B_FALSE;
 347         }
 348 }
 349
 350 static void
 351 acltype_changed_cb(void *arg, uint64_t newval)
 352 {
 353         zfsvfs_t *zfsvfs = arg;
 354
 355         switch (newval) {
 356         case ZFS_ACLTYPE_NFSV4:
 357         case ZFS_ACLTYPE_OFF:
 358                 zfsvfs->z_acl_type = ZFS_ACLTYPE_OFF;
 359                 zfsvfs->z_sb->s_flags &= ~SB_POSIXACL;
 360                 break;
 361         case ZFS_ACLTYPE_POSIX:
 362 #ifdef CONFIG_FS_POSIX_ACL
 363                 zfsvfs->z_acl_type = ZFS_ACLTYPE_POSIX;
 364                 zfsvfs->z_sb->s_flags |= SB_POSIXACL;
 365 #else
 366                 zfsvfs->z_acl_type = ZFS_ACLTYPE_OFF;
 367                 zfsvfs->z_sb->s_flags &= ~SB_POSIXACL;
 368 #endif /* CONFIG_FS_POSIX_ACL */
 369                 break;
 370         default:
 371                 break;
 372         }
 373 }
 374
 375 static void
 376 blksz_changed_cb(void *arg, uint64_t newval)
 377 {
 378         zfsvfs_t *zfsvfs = arg;
 379         ASSERT3U(newval, <=, spa_maxblocksize(dmu_objset_spa(zfsvfs->z_os)));
 380         ASSERT3U(newval, >=, SPA_MINBLOCKSIZE);
 381         ASSERT(ISP2(newval));
 382
 383         zfsvfs->z_max_blksz = newval;
 384 }
 385
 386 static void
 387 readonly_changed_cb(void *arg, uint64_t newval)
 388 {
 389         zfsvfs_t *zfsvfs = arg;
 390         struct super_block *sb = zfsvfs->z_sb;
 391
 392         if (sb == NULL)
 393                 return;
 394
 395         if (newval)
 396                 sb->s_flags |= SB_RDONLY;
 397         else
 398                 sb->s_flags &= ~SB_RDONLY;
 399 }
 400
 401 static void
 402 devices_changed_cb(void *arg, uint64_t newval)
 403 {
 404 }
 405
 406 static void
 407 setuid_changed_cb(void *arg, uint64_t newval)
 408 {
 409 }
 410
 411 static void
 412 exec_changed_cb(void *arg, uint64_t newval)
 413 {
 414 }
 415
 416 static void
 417 nbmand_changed_cb(void *arg, uint64_t newval)
 418 {
 419         zfsvfs_t *zfsvfs = arg;
 420         struct super_block *sb = zfsvfs->z_sb;
 421
 422         if (sb == NULL)
 423                 return;
 424
 425         if (newval == TRUE)
 426                 sb->s_flags |= SB_MANDLOCK;
 427         else
 428                 sb->s_flags &= ~SB_MANDLOCK;
 429 }
 430
 431 static void
 432 snapdir_changed_cb(void *arg, uint64_t newval)
 433 {
 434         ((zfsvfs_t *)arg)->z_show_ctldir = newval;
 435 }
 436
 437 static void
 438 acl_mode_changed_cb(void *arg, uint64_t newval)
 439 {
 440         zfsvfs_t *zfsvfs = arg;
 441
 442         zfsvfs->z_acl_mode = newval;
 443 }
 444
 445 static void
 446 acl_inherit_changed_cb(void *arg, uint64_t newval)
 447 {
 448         ((zfsvfs_t *)arg)->z_acl_inherit = newval;
 449 }
 450
 451 static int
 452 zfs_register_callbacks(vfs_t *vfsp)
 453 {
 454         struct dsl_dataset *ds = NULL;
 455         objset_t *os = NULL;
 456         zfsvfs_t *zfsvfs = NULL;
 457         int error = 0;
 458
 459         ASSERT(vfsp);
 460         zfsvfs = vfsp->vfs_data;
 461         ASSERT(zfsvfs);
 462         os = zfsvfs->z_os;
 463
 464         /*
 465          * The act of registering our callbacks will destroy any mount
 466          * options we may have.  In order to enable temporary overrides
 467          * of mount options, we stash away the current values and
 468          * restore them after we register the callbacks.
 469          */
 470         if (zfs_is_readonly(zfsvfs) || !spa_writeable(dmu_objset_spa(os))) {
 471                 vfsp->vfs_do_readonly = B_TRUE;
 472                 vfsp->vfs_readonly = B_TRUE;
 473         }
 474
 475         /*
 476          * Register property callbacks.
 477          *
 478          * It would probably be fine to just check for i/o error from
 479          * the first prop_register(), but I guess I like to go
 480          * overboard...
 481          */
 482         ds = dmu_objset_ds(os);
 483         dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
 484         error = dsl_prop_register(ds,
 485             zfs_prop_to_name(ZFS_PROP_ATIME), atime_changed_cb, zfsvfs);
 486         error = error ? error : dsl_prop_register(ds,
 487             zfs_prop_to_name(ZFS_PROP_RELATIME), relatime_changed_cb, zfsvfs);
 488         error = error ? error : dsl_prop_register(ds,
 489             zfs_prop_to_name(ZFS_PROP_XATTR), xattr_changed_cb, zfsvfs);
 490         error = error ? error : dsl_prop_register(ds,
 491             zfs_prop_to_name(ZFS_PROP_RECORDSIZE), blksz_changed_cb, zfsvfs);
 492         error = error ? error : dsl_prop_register(ds,
 493             zfs_prop_to_name(ZFS_PROP_READONLY), readonly_changed_cb, zfsvfs);
 494         error = error ? error : dsl_prop_register(ds,
 495             zfs_prop_to_name(ZFS_PROP_DEVICES), devices_changed_cb, zfsvfs);
 496         error = error ? error : dsl_prop_register(ds,
 497             zfs_prop_to_name(ZFS_PROP_SETUID), setuid_changed_cb, zfsvfs);
 498         error = error ? error : dsl_prop_register(ds,
 499             zfs_prop_to_name(ZFS_PROP_EXEC), exec_changed_cb, zfsvfs);
 500         error = error ? error : dsl_prop_register(ds,
 501             zfs_prop_to_name(ZFS_PROP_SNAPDIR), snapdir_changed_cb, zfsvfs);
 502         error = error ? error : dsl_prop_register(ds,
 503             zfs_prop_to_name(ZFS_PROP_ACLTYPE), acltype_changed_cb, zfsvfs);
 504         error = error ? error : dsl_prop_register(ds,
 505             zfs_prop_to_name(ZFS_PROP_ACLMODE), acl_mode_changed_cb, zfsvfs);
 506         error = error ? error : dsl_prop_register(ds,
 507             zfs_prop_to_name(ZFS_PROP_ACLINHERIT), acl_inherit_changed_cb,
 508             zfsvfs);
 509         error = error ? error : dsl_prop_register(ds,
 510             zfs_prop_to_name(ZFS_PROP_NBMAND), nbmand_changed_cb, zfsvfs);
 511         dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
 512         if (error)
 513                 goto unregister;
 514
 515         /*
 516          * Invoke our callbacks to restore temporary mount options.
 517          */
 518         if (vfsp->vfs_do_readonly)
 519                 readonly_changed_cb(zfsvfs, vfsp->vfs_readonly);
 520         if (vfsp->vfs_do_setuid)
 521                 setuid_changed_cb(zfsvfs, vfsp->vfs_setuid);
 522         if (vfsp->vfs_do_exec)
 523                 exec_changed_cb(zfsvfs, vfsp->vfs_exec);
 524         if (vfsp->vfs_do_devices)
 525                 devices_changed_cb(zfsvfs, vfsp->vfs_devices);
 526         if (vfsp->vfs_do_xattr)
 527                 xattr_changed_cb(zfsvfs, vfsp->vfs_xattr);
 528         if (vfsp->vfs_do_atime)
 529                 atime_changed_cb(zfsvfs, vfsp->vfs_atime);
 530         if (vfsp->vfs_do_relatime)
 531                 relatime_changed_cb(zfsvfs, vfsp->vfs_relatime);
 532         if (vfsp->vfs_do_nbmand)
 533                 nbmand_changed_cb(zfsvfs, vfsp->vfs_nbmand);
 534
 535         return (0);
 536
 537 unregister:
 538         dsl_prop_unregister_all(ds, zfsvfs);
 539         return (error);
 540 }
 541
 542 /*
 543  * Takes a dataset, a property, a value and that value's setpoint as
 544  * found in the ZAP. Checks if the property has been changed in the vfs.
 545  * If so, val and setpoint will be overwritten with updated content.
 546  * Otherwise, they are left unchanged.
 547  */
 548 int
 549 zfs_get_temporary_prop(dsl_dataset_t *ds, zfs_prop_t zfs_prop, uint64_t *val,
 550     char *setpoint)
 551 {
 552         int error;
 553         zfsvfs_t *zfvp;
 554         vfs_t *vfsp;
 555         objset_t *os;
 556         uint64_t tmp = *val;
 557
 558         error = dmu_objset_from_ds(ds, &os);
 559         if (error != 0)
 560                 return (error);
 561
 562         if (dmu_objset_type(os) != DMU_OST_ZFS)
 563                 return (EINVAL);
 564
 565         mutex_enter(&os->os_user_ptr_lock);
 566         zfvp = dmu_objset_get_user(os);
 567         mutex_exit(&os->os_user_ptr_lock);
 568         if (zfvp == NULL)
 569                 return (ESRCH);
 570
 571         vfsp = zfvp->z_vfs;
 572
 573         switch (zfs_prop) {
 574         case ZFS_PROP_ATIME:
 575                 if (vfsp->vfs_do_atime)
 576                         tmp = vfsp->vfs_atime;
 577                 break;
 578         case ZFS_PROP_RELATIME:
 579                 if (vfsp->vfs_do_relatime)
 580                         tmp = vfsp->vfs_relatime;
 581                 break;
 582         case ZFS_PROP_DEVICES:
 583                 if (vfsp->vfs_do_devices)
 584                         tmp = vfsp->vfs_devices;
 585                 break;
 586         case ZFS_PROP_EXEC:
 587                 if (vfsp->vfs_do_exec)
 588                         tmp = vfsp->vfs_exec;
 589                 break;
 590         case ZFS_PROP_SETUID:
 591                 if (vfsp->vfs_do_setuid)
 592                         tmp = vfsp->vfs_setuid;
 593                 break;
 594         case ZFS_PROP_READONLY:
 595                 if (vfsp->vfs_do_readonly)
 596                         tmp = vfsp->vfs_readonly;
 597                 break;
 598         case ZFS_PROP_XATTR:
 599                 if (vfsp->vfs_do_xattr)
 600                         tmp = vfsp->vfs_xattr;
 601                 break;
 602         case ZFS_PROP_NBMAND:
 603                 if (vfsp->vfs_do_nbmand)
 604                         tmp = vfsp->vfs_nbmand;
 605                 break;
 606         default:
 607                 return (ENOENT);
 608         }
 609
 610         if (tmp != *val) {
 611                 (void) strcpy(setpoint, "temporary");
 612                 *val = tmp;
 613         }
 614         return (0);
 615 }
 616
 617 /*
 618  * Associate this zfsvfs with the given objset, which must be owned.
 619  * This will cache a bunch of on-disk state from the objset in the
 620  * zfsvfs.
 621  */
 622 static int
 623 zfsvfs_init(zfsvfs_t *zfsvfs, objset_t *os)
 624 {
 625         int error;
 626         uint64_t val;
 627
 628         zfsvfs->z_max_blksz = SPA_OLD_MAXBLOCKSIZE;
 629         zfsvfs->z_show_ctldir = ZFS_SNAPDIR_VISIBLE;
 630         zfsvfs->z_os = os;
 631
 632         error = zfs_get_zplprop(os, ZFS_PROP_VERSION, &zfsvfs->z_version);
 633         if (error != 0)
 634                 return (error);
 635         if (zfsvfs->z_version >
 636             zfs_zpl_version_map(spa_version(dmu_objset_spa(os)))) {
 637                 (void) printk("Can't mount a version %lld file system "
 638                     "on a version %lld pool\n. Pool must be upgraded to mount "
 639                     "this file system.\n", (u_longlong_t)zfsvfs->z_version,
 640                     (u_longlong_t)spa_version(dmu_objset_spa(os)));
 641                 return (SET_ERROR(ENOTSUP));
 642         }
 643         error = zfs_get_zplprop(os, ZFS_PROP_NORMALIZE, &val);
 644         if (error != 0)
 645                 return (error);
 646         zfsvfs->z_norm = (int)val;
 647
 648         error = zfs_get_zplprop(os, ZFS_PROP_UTF8ONLY, &val);
 649         if (error != 0)
 650                 return (error);
 651         zfsvfs->z_utf8 = (val != 0);
 652
 653         error = zfs_get_zplprop(os, ZFS_PROP_CASE, &val);
 654         if (error != 0)
 655                 return (error);
 656         zfsvfs->z_case = (uint_t)val;
 657
 658         if ((error = zfs_get_zplprop(os, ZFS_PROP_ACLTYPE, &val)) != 0)
 659                 return (error);
 660         zfsvfs->z_acl_type = (uint_t)val;
 661
 662         /*
 663          * Fold case on file systems that are always or sometimes case
 664          * insensitive.
 665          */
 666         if (zfsvfs->z_case == ZFS_CASE_INSENSITIVE ||
 667             zfsvfs->z_case == ZFS_CASE_MIXED)
 668                 zfsvfs->z_norm |= U8_TEXTPREP_TOUPPER;
 669
 670         zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
 671         zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
 672
 673         uint64_t sa_obj = 0;
 674         if (zfsvfs->z_use_sa) {
 675                 /* should either have both of these objects or none */
 676                 error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SA_ATTRS, 8, 1,
 677                     &sa_obj);
 678                 if (error != 0)
 679                         return (error);
 680
 681                 error = zfs_get_zplprop(os, ZFS_PROP_XATTR, &val);
 682                 if ((error == 0) && (val == ZFS_XATTR_SA))
 683                         zfsvfs->z_xattr_sa = B_TRUE;
 684         }
 685
 686         error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1,
 687             &zfsvfs->z_root);
 688         if (error != 0)
 689                 return (error);
 690         ASSERT(zfsvfs->z_root != 0);
 691
 692         error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_UNLINKED_SET, 8, 1,
 693             &zfsvfs->z_unlinkedobj);
 694         if (error != 0)
 695                 return (error);
 696
 697         error = zap_lookup(os, MASTER_NODE_OBJ,
 698             zfs_userquota_prop_prefixes[ZFS_PROP_USERQUOTA],
 699             8, 1, &zfsvfs->z_userquota_obj);
 700         if (error == ENOENT)
 701                 zfsvfs->z_userquota_obj = 0;
 702         else if (error != 0)
 703                 return (error);
 704
 705         error = zap_lookup(os, MASTER_NODE_OBJ,
 706             zfs_userquota_prop_prefixes[ZFS_PROP_GROUPQUOTA],
 707             8, 1, &zfsvfs->z_groupquota_obj);
 708         if (error == ENOENT)
 709                 zfsvfs->z_groupquota_obj = 0;
 710         else if (error != 0)
 711                 return (error);
 712
 713         error = zap_lookup(os, MASTER_NODE_OBJ,
 714             zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTQUOTA],
 715             8, 1, &zfsvfs->z_projectquota_obj);
 716         if (error == ENOENT)
 717                 zfsvfs->z_projectquota_obj = 0;
 718         else if (error != 0)
 719                 return (error);
 720
 721         error = zap_lookup(os, MASTER_NODE_OBJ,
 722             zfs_userquota_prop_prefixes[ZFS_PROP_USEROBJQUOTA],
 723             8, 1, &zfsvfs->z_userobjquota_obj);
 724         if (error == ENOENT)
 725                 zfsvfs->z_userobjquota_obj = 0;
 726         else if (error != 0)
 727                 return (error);
 728
 729         error = zap_lookup(os, MASTER_NODE_OBJ,
 730             zfs_userquota_prop_prefixes[ZFS_PROP_GROUPOBJQUOTA],
 731             8, 1, &zfsvfs->z_groupobjquota_obj);
 732         if (error == ENOENT)
 733                 zfsvfs->z_groupobjquota_obj = 0;
 734         else if (error != 0)
 735                 return (error);
 736
 737         error = zap_lookup(os, MASTER_NODE_OBJ,
 738             zfs_userquota_prop_prefixes[ZFS_PROP_PROJECTOBJQUOTA],
 739             8, 1, &zfsvfs->z_projectobjquota_obj);
 740         if (error == ENOENT)
 741                 zfsvfs->z_projectobjquota_obj = 0;
 742         else if (error != 0)
 743                 return (error);
 744
 745         error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES, 8, 1,
 746             &zfsvfs->z_fuid_obj);
 747         if (error == ENOENT)
 748                 zfsvfs->z_fuid_obj = 0;
 749         else if (error != 0)
 750                 return (error);
 751
 752         error = zap_lookup(os, MASTER_NODE_OBJ, ZFS_SHARES_DIR, 8, 1,
 753             &zfsvfs->z_shares_dir);
 754         if (error == ENOENT)
 755                 zfsvfs->z_shares_dir = 0;
 756         else if (error != 0)
 757                 return (error);
 758
 759         error = sa_setup(os, sa_obj, zfs_attr_table, ZPL_END,
 760             &zfsvfs->z_attr_table);
 761         if (error != 0)
 762                 return (error);
 763
 764         if (zfsvfs->z_version >= ZPL_VERSION_SA)
 765                 sa_register_update_callback(os, zfs_sa_upgrade);
 766
 767         return (0);
 768 }
 769
 770 int
 771 zfsvfs_create(const char *osname, boolean_t readonly, zfsvfs_t **zfvp)
 772 {
 773         objset_t *os;
 774         zfsvfs_t *zfsvfs;
 775         int error;
 776         boolean_t ro = (readonly || (strchr(osname, '@') != NULL));
 777
 778         zfsvfs = kmem_zalloc(sizeof (zfsvfs_t), KM_SLEEP);
 779
 780         error = dmu_objset_own(osname, DMU_OST_ZFS, ro, B_TRUE, zfsvfs, &os);
 781         if (error != 0) {
 782                 kmem_free(zfsvfs, sizeof (zfsvfs_t));
 783                 return (error);
 784         }
 785
 786         error = zfsvfs_create_impl(zfvp, zfsvfs, os);
 787
 788         return (error);
 789 }
 790
 791
 792 /*
 793  * Note: zfsvfs is assumed to be malloc'd, and will be freed by this function
 794  * on a failure.  Do not pass in a statically allocated zfsvfs.
 795  */
 796 int
 797 zfsvfs_create_impl(zfsvfs_t **zfvp, zfsvfs_t *zfsvfs, objset_t *os)
 798 {
 799         int error;
 800
 801         zfsvfs->z_vfs = NULL;
 802         zfsvfs->z_sb = NULL;
 803         zfsvfs->z_parent = zfsvfs;
 804
 805         mutex_init(&zfsvfs->z_znodes_lock, NULL, MUTEX_DEFAULT, NULL);
 806         mutex_init(&zfsvfs->z_lock, NULL, MUTEX_DEFAULT, NULL);
 807         list_create(&zfsvfs->z_all_znodes, sizeof (znode_t),
 808             offsetof(znode_t, z_link_node));
 809         ZFS_TEARDOWN_INIT(zfsvfs);
 810         rw_init(&zfsvfs->z_teardown_inactive_lock, NULL, RW_DEFAULT, NULL);
 811         rw_init(&zfsvfs->z_fuid_lock, NULL, RW_DEFAULT, NULL);
 812
 813         int size = MIN(1 << (highbit64(zfs_object_mutex_size) - 1),
 814             ZFS_OBJ_MTX_MAX);
 815         zfsvfs->z_hold_size = size;
 816         zfsvfs->z_hold_trees = vmem_zalloc(sizeof (avl_tree_t) * size,
 817             KM_SLEEP);
 818         zfsvfs->z_hold_locks = vmem_zalloc(sizeof (kmutex_t) * size, KM_SLEEP);
 819         for (int i = 0; i != size; i++) {
 820                 avl_create(&zfsvfs->z_hold_trees[i], zfs_znode_hold_compare,
 821                     sizeof (znode_hold_t), offsetof(znode_hold_t, zh_node));
 822                 mutex_init(&zfsvfs->z_hold_locks[i], NULL, MUTEX_DEFAULT, NULL);
 823         }
 824
 825         error = zfsvfs_init(zfsvfs, os);
 826         if (error != 0) {
 827                 dmu_objset_disown(os, B_TRUE, zfsvfs);
 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         /*
 852          * If we are not mounting (ie: online recv), then we don't
 853          * have to worry about replaying the log as we blocked all
 854          * operations out since we closed the ZIL.
 855          */
 856         if (mounting) {
 857                 ASSERT3P(zfsvfs->z_kstat.dk_kstats, ==, NULL);
 858                 error = dataset_kstats_create(&zfsvfs->z_kstat, zfsvfs->z_os);
 859                 if (error)
 860                         return (error);
 861                 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data,
 862                     &zfsvfs->z_kstat.dk_zil_sums);
 863
 864                 /*
 865                  * During replay we remove the read only flag to
 866                  * allow replays to succeed.
 867                  */
 868                 if (readonly != 0) {
 869                         readonly_changed_cb(zfsvfs, B_FALSE);
 870                 } else {
 871                         zap_stats_t zs;
 872                         if (zap_get_stats(zfsvfs->z_os, zfsvfs->z_unlinkedobj,
 873                             &zs) == 0) {
 874                                 dataset_kstats_update_nunlinks_kstat(
 875                                     &zfsvfs->z_kstat, zs.zs_num_entries);
 876                                 dprintf_ds(zfsvfs->z_os->os_dsl_dataset,
 877                                     "num_entries in unlinked set: %llu",
 878                                     zs.zs_num_entries);
 879                         }
 880                         zfs_unlinked_drain(zfsvfs);
 881                         dsl_dir_t *dd = zfsvfs->z_os->os_dsl_dataset->ds_dir;
 882                         dd->dd_activity_cancelled = B_FALSE;
 883                 }
 884
 885                 /*
 886                  * Parse and replay the intent log.
 887                  *
 888                  * Because of ziltest, this must be done after
 889                  * zfs_unlinked_drain().  (Further note: ziltest
 890                  * doesn't use readonly mounts, where
 891                  * zfs_unlinked_drain() isn't called.)  This is because
 892                  * ziltest causes spa_sync() to think it's committed,
 893                  * but actually it is not, so the intent log contains
 894                  * many txg's worth of changes.
 895                  *
 896                  * In particular, if object N is in the unlinked set in
 897                  * the last txg to actually sync, then it could be
 898                  * actually freed in a later txg and then reallocated
 899                  * in a yet later txg.  This would write a "create
 900                  * object N" record to the intent log.  Normally, this
 901                  * would be fine because the spa_sync() would have
 902                  * written out the fact that object N is free, before
 903                  * we could write the "create object N" intent log
 904                  * record.
 905                  *
 906                  * But when we are in ziltest mode, we advance the "open
 907                  * txg" without actually spa_sync()-ing the changes to
 908                  * disk.  So we would see that object N is still
 909                  * allocated and in the unlinked set, and there is an
 910                  * intent log record saying to allocate it.
 911                  */
 912                 if (spa_writeable(dmu_objset_spa(zfsvfs->z_os))) {
 913                         if (zil_replay_disable) {
 914                                 zil_destroy(zfsvfs->z_log, B_FALSE);
 915                         } else {
 916                                 zfsvfs->z_replay = B_TRUE;
 917                                 zil_replay(zfsvfs->z_os, zfsvfs,
 918                                     zfs_replay_vector);
 919                                 zfsvfs->z_replay = B_FALSE;
 920                         }
 921                 }
 922
 923                 /* restore readonly bit */
 924                 if (readonly != 0)
 925                         readonly_changed_cb(zfsvfs, B_TRUE);
 926         } else {
 927                 ASSERT3P(zfsvfs->z_kstat.dk_kstats, !=, NULL);
 928                 zfsvfs->z_log = zil_open(zfsvfs->z_os, zfs_get_data,
 929                     &zfsvfs->z_kstat.dk_zil_sums);
 930         }
 931
 932         /*
 933          * Set the objset user_ptr to track its zfsvfs.
 934          */
 935         mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
 936         dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
 937         mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
 938
 939         return (0);
 940 }
 941
 942 void
 943 zfsvfs_free(zfsvfs_t *zfsvfs)
 944 {
 945         int i, size = zfsvfs->z_hold_size;
 946
 947         zfs_fuid_destroy(zfsvfs);
 948
 949         mutex_destroy(&zfsvfs->z_znodes_lock);
 950         mutex_destroy(&zfsvfs->z_lock);
 951         list_destroy(&zfsvfs->z_all_znodes);
 952         ZFS_TEARDOWN_DESTROY(zfsvfs);
 953         rw_destroy(&zfsvfs->z_teardown_inactive_lock);
 954         rw_destroy(&zfsvfs->z_fuid_lock);
 955         for (i = 0; i != size; i++) {
 956                 avl_destroy(&zfsvfs->z_hold_trees[i]);
 957                 mutex_destroy(&zfsvfs->z_hold_locks[i]);
 958         }
 959         vmem_free(zfsvfs->z_hold_trees, sizeof (avl_tree_t) * size);
 960         vmem_free(zfsvfs->z_hold_locks, sizeof (kmutex_t) * size);
 961         zfsvfs_vfs_free(zfsvfs->z_vfs);
 962         dataset_kstats_destroy(&zfsvfs->z_kstat);
 963         kmem_free(zfsvfs, sizeof (zfsvfs_t));
 964 }
 965
 966 static void
 967 zfs_set_fuid_feature(zfsvfs_t *zfsvfs)
 968 {
 969         zfsvfs->z_use_fuids = USE_FUIDS(zfsvfs->z_version, zfsvfs->z_os);
 970         zfsvfs->z_use_sa = USE_SA(zfsvfs->z_version, zfsvfs->z_os);
 971 }
 972
 973 static void
 974 zfs_unregister_callbacks(zfsvfs_t *zfsvfs)
 975 {
 976         objset_t *os = zfsvfs->z_os;
 977
 978         if (!dmu_objset_is_snapshot(os))
 979                 dsl_prop_unregister_all(dmu_objset_ds(os), zfsvfs);
 980 }
 981
 982 #ifdef HAVE_MLSLABEL
 983 /*
 984  * Check that the hex label string is appropriate for the dataset being
 985  * mounted into the global_zone proper.
 986  *
 987  * Return an error if the hex label string is not default or
 988  * admin_low/admin_high.  For admin_low labels, the corresponding
 989  * dataset must be readonly.
 990  */
 991 int
 992 zfs_check_global_label(const char *dsname, const char *hexsl)
 993 {
 994         if (strcasecmp(hexsl, ZFS_MLSLABEL_DEFAULT) == 0)
 995                 return (0);
 996         if (strcasecmp(hexsl, ADMIN_HIGH) == 0)
 997                 return (0);
 998         if (strcasecmp(hexsl, ADMIN_LOW) == 0) {
 999                 /* must be readonly */
1000                 uint64_t rdonly;
1001
1002                 if (dsl_prop_get_integer(dsname,
1003                     zfs_prop_to_name(ZFS_PROP_READONLY), &rdonly, NULL))
1004                         return (SET_ERROR(EACCES));
1005                 return (rdonly ? 0 : SET_ERROR(EACCES));
1006         }
1007         return (SET_ERROR(EACCES));
1008 }
1009 #endif /* HAVE_MLSLABEL */
1010
1011 static int
1012 zfs_statfs_project(zfsvfs_t *zfsvfs, znode_t *zp, struct kstatfs *statp,
1013     uint32_t bshift)
1014 {
1015         char buf[20 + DMU_OBJACCT_PREFIX_LEN];
1016         uint64_t offset = DMU_OBJACCT_PREFIX_LEN;
1017         uint64_t quota;
1018         uint64_t used;
1019         int err;
1020
1021         strlcpy(buf, DMU_OBJACCT_PREFIX, DMU_OBJACCT_PREFIX_LEN + 1);
1022         err = zfs_id_to_fuidstr(zfsvfs, NULL, zp->z_projid, buf + offset,
1023             sizeof (buf) - offset, B_FALSE);
1024         if (err)
1025                 return (err);
1026
1027         if (zfsvfs->z_projectquota_obj == 0)
1028                 goto objs;
1029
1030         err = zap_lookup(zfsvfs->z_os, zfsvfs->z_projectquota_obj,
1031             buf + offset, 8, 1, &quota);
1032         if (err == ENOENT)
1033                 goto objs;
1034         else if (err)
1035                 return (err);
1036
1037         err = zap_lookup(zfsvfs->z_os, DMU_PROJECTUSED_OBJECT,
1038             buf + offset, 8, 1, &used);
1039         if (unlikely(err == ENOENT)) {
1040                 uint32_t blksize;
1041                 u_longlong_t nblocks;
1042
1043                 /*
1044                  * Quota accounting is async, so it is possible race case.
1045                  * There is at least one object with the given project ID.
1046                  */
1047                 sa_object_size(zp->z_sa_hdl, &blksize, &nblocks);
1048                 if (unlikely(zp->z_blksz == 0))
1049                         blksize = zfsvfs->z_max_blksz;
1050
1051                 used = blksize * nblocks;
1052         } else if (err) {
1053                 return (err);
1054         }
1055
1056         statp->f_blocks = quota >> bshift;
1057         statp->f_bfree = (quota > used) ? ((quota - used) >> bshift) : 0;
1058         statp->f_bavail = statp->f_bfree;
1059
1060 objs:
1061         if (zfsvfs->z_projectobjquota_obj == 0)
1062                 return (0);
1063
1064         err = zap_lookup(zfsvfs->z_os, zfsvfs->z_projectobjquota_obj,
1065             buf + offset, 8, 1, &quota);
1066         if (err == ENOENT)
1067                 return (0);
1068         else if (err)
1069                 return (err);
1070
1071         err = zap_lookup(zfsvfs->z_os, DMU_PROJECTUSED_OBJECT,
1072             buf, 8, 1, &used);
1073         if (unlikely(err == ENOENT)) {
1074                 /*
1075                  * Quota accounting is async, so it is possible race case.
1076                  * There is at least one object with the given project ID.
1077                  */
1078                 used = 1;
1079         } else if (err) {
1080                 return (err);
1081         }
1082
1083         statp->f_files = quota;
1084         statp->f_ffree = (quota > used) ? (quota - used) : 0;
1085
1086         return (0);
1087 }
1088
1089 int
1090 zfs_statvfs(struct inode *ip, struct kstatfs *statp)
1091 {
1092         zfsvfs_t *zfsvfs = ITOZSB(ip);
1093         uint64_t refdbytes, availbytes, usedobjs, availobjs;
1094         int err = 0;
1095
1096         if ((err = zfs_enter(zfsvfs, FTAG)) != 0)
1097                 return (err);
1098
1099         dmu_objset_space(zfsvfs->z_os,
1100             &refdbytes, &availbytes, &usedobjs, &availobjs);
1101
1102         uint64_t fsid = dmu_objset_fsid_guid(zfsvfs->z_os);
1103         /*
1104          * The underlying storage pool actually uses multiple block
1105          * size.  Under Solaris frsize (fragment size) is reported as
1106          * the smallest block size we support, and bsize (block size)
1107          * as the filesystem's maximum block size.  Unfortunately,
1108          * under Linux the fragment size and block size are often used
1109          * interchangeably.  Thus we are forced to report both of them
1110          * as the filesystem's maximum block size.
1111          */
1112         statp->f_frsize = zfsvfs->z_max_blksz;
1113         statp->f_bsize = zfsvfs->z_max_blksz;
1114         uint32_t bshift = fls(statp->f_bsize) - 1;
1115
1116         /*
1117          * The following report "total" blocks of various kinds in
1118          * the file system, but reported in terms of f_bsize - the
1119          * "preferred" size.
1120          */
1121
1122         /* Round up so we never have a filesystem using 0 blocks. */
1123         refdbytes = P2ROUNDUP(refdbytes, statp->f_bsize);
1124         statp->f_blocks = (refdbytes + availbytes) >> bshift;
1125         statp->f_bfree = availbytes >> bshift;
1126         statp->f_bavail = statp->f_bfree; /* no root reservation */
1127
1128         /*
1129          * statvfs() should really be called statufs(), because it assumes
1130          * static metadata.  ZFS doesn't preallocate files, so the best
1131          * we can do is report the max that could possibly fit in f_files,
1132          * and that minus the number actually used in f_ffree.
1133          * For f_ffree, report the smaller of the number of objects available
1134          * and the number of blocks (each object will take at least a block).
1135          */
1136         statp->f_ffree = MIN(availobjs, availbytes >> DNODE_SHIFT);
1137         statp->f_files = statp->f_ffree + usedobjs;
1138         statp->f_fsid.val[0] = (uint32_t)fsid;
1139         statp->f_fsid.val[1] = (uint32_t)(fsid >> 32);
1140         statp->f_type = ZFS_SUPER_MAGIC;
1141         statp->f_namelen = MAXNAMELEN - 1;
1142
1143         /*
1144          * We have all of 40 characters to stuff a string here.
1145          * Is there anything useful we could/should provide?
1146          */
1147         memset(statp->f_spare, 0, sizeof (statp->f_spare));
1148
1149         if (dmu_objset_projectquota_enabled(zfsvfs->z_os) &&
1150             dmu_objset_projectquota_present(zfsvfs->z_os)) {
1151                 znode_t *zp = ITOZ(ip);
1152
1153                 if (zp->z_pflags & ZFS_PROJINHERIT && zp->z_projid &&
1154                     zpl_is_valid_projid(zp->z_projid))
1155                         err = zfs_statfs_project(zfsvfs, zp, statp, bshift);
1156         }
1157
1158         zfs_exit(zfsvfs, FTAG);
1159         return (err);
1160 }
1161
1162 static int
1163 zfs_root(zfsvfs_t *zfsvfs, struct inode **ipp)
1164 {
1165         znode_t *rootzp;
1166         int error;
1167
1168         if ((error = zfs_enter(zfsvfs, FTAG)) != 0)
1169                 return (error);
1170
1171         error = zfs_zget(zfsvfs, zfsvfs->z_root, &rootzp);
1172         if (error == 0)
1173                 *ipp = ZTOI(rootzp);
1174
1175         zfs_exit(zfsvfs, FTAG);
1176         return (error);
1177 }
1178
1179 /*
1180  * Linux kernels older than 3.1 do not support a per-filesystem shrinker.
1181  * To accommodate this we must improvise and manually walk the list of znodes
1182  * attempting to prune dentries in order to be able to drop the inodes.
1183  *
1184  * To avoid scanning the same znodes multiple times they are always rotated
1185  * to the end of the z_all_znodes list.  New znodes are inserted at the
1186  * end of the list so we're always scanning the oldest znodes first.
1187  */
1188 static int
1189 zfs_prune_aliases(zfsvfs_t *zfsvfs, unsigned long nr_to_scan)
1190 {
1191         znode_t **zp_array, *zp;
1192         int max_array = MIN(nr_to_scan, PAGE_SIZE * 8 / sizeof (znode_t *));
1193         int objects = 0;
1194         int i = 0, j = 0;
1195
1196         zp_array = kmem_zalloc(max_array * sizeof (znode_t *), KM_SLEEP);
1197
1198         mutex_enter(&zfsvfs->z_znodes_lock);
1199         while ((zp = list_head(&zfsvfs->z_all_znodes)) != NULL) {
1200
1201                 if ((i++ > nr_to_scan) || (j >= max_array))
1202                         break;
1203
1204                 ASSERT(list_link_active(&zp->z_link_node));
1205                 list_remove(&zfsvfs->z_all_znodes, zp);
1206                 list_insert_tail(&zfsvfs->z_all_znodes, zp);
1207
1208                 /* Skip active znodes and .zfs entries */
1209                 if (MUTEX_HELD(&zp->z_lock) || zp->z_is_ctldir)
1210                         continue;
1211
1212                 if (igrab(ZTOI(zp)) == NULL)
1213                         continue;
1214
1215                 zp_array[j] = zp;
1216                 j++;
1217         }
1218         mutex_exit(&zfsvfs->z_znodes_lock);
1219
1220         for (i = 0; i < j; i++) {
1221                 zp = zp_array[i];
1222
1223                 ASSERT3P(zp, !=, NULL);
1224                 d_prune_aliases(ZTOI(zp));
1225
1226                 if (atomic_read(&ZTOI(zp)->i_count) == 1)
1227                         objects++;
1228
1229                 zrele(zp);
1230         }
1231
1232         kmem_free(zp_array, max_array * sizeof (znode_t *));
1233
1234         return (objects);
1235 }
1236
1237 /*
1238  * The ARC has requested that the filesystem drop entries from the dentry
1239  * and inode caches.  This can occur when the ARC needs to free meta data
1240  * blocks but can't because they are all pinned by entries in these caches.
1241  */
1242 int
1243 zfs_prune(struct super_block *sb, unsigned long nr_to_scan, int *objects)
1244 {
1245         zfsvfs_t *zfsvfs = sb->s_fs_info;
1246         int error = 0;
1247         struct shrinker *shrinker = &sb->s_shrink;
1248         struct shrink_control sc = {
1249                 .nr_to_scan = nr_to_scan,
1250                 .gfp_mask = GFP_KERNEL,
1251         };
1252
1253         if ((error = zfs_enter(zfsvfs, FTAG)) != 0)
1254                 return (error);
1255
1256 #if defined(HAVE_SPLIT_SHRINKER_CALLBACK) && \
1257         defined(SHRINK_CONTROL_HAS_NID) && \
1258         defined(SHRINKER_NUMA_AWARE)
1259         if (sb->s_shrink.flags & SHRINKER_NUMA_AWARE) {
1260                 *objects = 0;
1261                 for_each_online_node(sc.nid) {
1262                         *objects += (*shrinker->scan_objects)(shrinker, &sc);
1263                         /*
1264                          * reset sc.nr_to_scan, modified by
1265                          * scan_objects == super_cache_scan
1266                          */
1267                         sc.nr_to_scan = nr_to_scan;
1268                 }
1269         } else {
1270                         *objects = (*shrinker->scan_objects)(shrinker, &sc);
1271         }
1272
1273 #elif defined(HAVE_SPLIT_SHRINKER_CALLBACK)
1274         *objects = (*shrinker->scan_objects)(shrinker, &sc);
1275 #elif defined(HAVE_SINGLE_SHRINKER_CALLBACK)
1276         *objects = (*shrinker->shrink)(shrinker, &sc);
1277 #elif defined(HAVE_D_PRUNE_ALIASES)
1278 #define D_PRUNE_ALIASES_IS_DEFAULT
1279         *objects = zfs_prune_aliases(zfsvfs, nr_to_scan);
1280 #else
1281 #error "No available dentry and inode cache pruning mechanism."
1282 #endif
1283
1284 #if defined(HAVE_D_PRUNE_ALIASES) && !defined(D_PRUNE_ALIASES_IS_DEFAULT)
1285 #undef  D_PRUNE_ALIASES_IS_DEFAULT
1286         /*
1287          * Fall back to zfs_prune_aliases if the kernel's per-superblock
1288          * shrinker couldn't free anything, possibly due to the inodes being
1289          * allocated in a different memcg.
1290          */
1291         if (*objects == 0)
1292                 *objects = zfs_prune_aliases(zfsvfs, nr_to_scan);
1293 #endif
1294
1295         zfs_exit(zfsvfs, FTAG);
1296
1297         dprintf_ds(zfsvfs->z_os->os_dsl_dataset,
1298             "pruning, nr_to_scan=%lu objects=%d error=%d\n",
1299             nr_to_scan, *objects, error);
1300
1301         return (error);
1302 }
1303
1304 /*
1305  * Teardown the zfsvfs_t.
1306  *
1307  * Note, if 'unmounting' is FALSE, we return with the 'z_teardown_lock'
1308  * and 'z_teardown_inactive_lock' held.
1309  */
1310 static int
1311 zfsvfs_teardown(zfsvfs_t *zfsvfs, boolean_t unmounting)
1312 {
1313         znode_t *zp;
1314
1315         zfs_unlinked_drain_stop_wait(zfsvfs);
1316
1317         /*
1318          * If someone has not already unmounted this file system,
1319          * drain the zrele_taskq to ensure all active references to the
1320          * zfsvfs_t have been handled only then can it be safely destroyed.
1321          */
1322         if (zfsvfs->z_os) {
1323                 /*
1324                  * If we're unmounting we have to wait for the list to
1325                  * drain completely.
1326                  *
1327                  * If we're not unmounting there's no guarantee the list
1328                  * will drain completely, but iputs run from the taskq
1329                  * may add the parents of dir-based xattrs to the taskq
1330                  * so we want to wait for these.
1331                  *
1332                  * We can safely read z_nr_znodes without locking because the
1333                  * VFS has already blocked operations which add to the
1334                  * z_all_znodes list and thus increment z_nr_znodes.
1335                  */
1336                 int round = 0;
1337                 while (zfsvfs->z_nr_znodes > 0) {
1338                         taskq_wait_outstanding(dsl_pool_zrele_taskq(
1339                             dmu_objset_pool(zfsvfs->z_os)), 0);
1340                         if (++round > 1 && !unmounting)
1341                                 break;
1342                 }
1343         }
1344
1345         ZFS_TEARDOWN_ENTER_WRITE(zfsvfs, FTAG);
1346
1347         if (!unmounting) {
1348                 /*
1349                  * We purge the parent filesystem's super block as the
1350                  * parent filesystem and all of its snapshots have their
1351                  * inode's super block set to the parent's filesystem's
1352                  * super block.  Note,  'z_parent' is self referential
1353                  * for non-snapshots.
1354                  */
1355                 shrink_dcache_sb(zfsvfs->z_parent->z_sb);
1356         }
1357
1358         /*
1359          * Close the zil. NB: Can't close the zil while zfs_inactive
1360          * threads are blocked as zil_close can call zfs_inactive.
1361          */
1362         if (zfsvfs->z_log) {
1363                 zil_close(zfsvfs->z_log);
1364                 zfsvfs->z_log = NULL;
1365         }
1366
1367         rw_enter(&zfsvfs->z_teardown_inactive_lock, RW_WRITER);
1368
1369         /*
1370          * If we are not unmounting (ie: online recv) and someone already
1371          * unmounted this file system while we were doing the switcheroo,
1372          * or a reopen of z_os failed then just bail out now.
1373          */
1374         if (!unmounting && (zfsvfs->z_unmounted || zfsvfs->z_os == NULL)) {
1375                 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1376                 ZFS_TEARDOWN_EXIT(zfsvfs, FTAG);
1377                 return (SET_ERROR(EIO));
1378         }
1379
1380         /*
1381          * At this point there are no VFS ops active, and any new VFS ops
1382          * will fail with EIO since we have z_teardown_lock for writer (only
1383          * relevant for forced unmount).
1384          *
1385          * Release all holds on dbufs. We also grab an extra reference to all
1386          * the remaining inodes so that the kernel does not attempt to free
1387          * any inodes of a suspended fs. This can cause deadlocks since the
1388          * zfs_resume_fs() process may involve starting threads, which might
1389          * attempt to free unreferenced inodes to free up memory for the new
1390          * thread.
1391          */
1392         if (!unmounting) {
1393                 mutex_enter(&zfsvfs->z_znodes_lock);
1394                 for (zp = list_head(&zfsvfs->z_all_znodes); zp != NULL;
1395                     zp = list_next(&zfsvfs->z_all_znodes, zp)) {
1396                         if (zp->z_sa_hdl)
1397                                 zfs_znode_dmu_fini(zp);
1398                         if (igrab(ZTOI(zp)) != NULL)
1399                                 zp->z_suspended = B_TRUE;
1400
1401                 }
1402                 mutex_exit(&zfsvfs->z_znodes_lock);
1403         }
1404
1405         /*
1406          * If we are unmounting, set the unmounted flag and let new VFS ops
1407          * unblock.  zfs_inactive will have the unmounted behavior, and all
1408          * other VFS ops will fail with EIO.
1409          */
1410         if (unmounting) {
1411                 zfsvfs->z_unmounted = B_TRUE;
1412                 rw_exit(&zfsvfs->z_teardown_inactive_lock);
1413                 ZFS_TEARDOWN_EXIT(zfsvfs, FTAG);
1414         }
1415
1416         /*
1417          * z_os will be NULL if there was an error in attempting to reopen
1418          * zfsvfs, so just return as the properties had already been
1419          *
1420          * unregistered and cached data had been evicted before.
1421          */
1422         if (zfsvfs->z_os == NULL)
1423                 return (0);
1424
1425         /*
1426          * Unregister properties.
1427          */
1428         zfs_unregister_callbacks(zfsvfs);
1429
1430         /*
1431          * Evict cached data. We must write out any dirty data before
1432          * disowning the dataset.
1433          */
1434         objset_t *os = zfsvfs->z_os;
1435         boolean_t os_dirty = B_FALSE;
1436         for (int t = 0; t < TXG_SIZE; t++) {
1437                 if (dmu_objset_is_dirty(os, t)) {
1438                         os_dirty = B_TRUE;
1439                         break;
1440                 }
1441         }
1442         if (!zfs_is_readonly(zfsvfs) && os_dirty) {
1443                 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1444         }
1445         dmu_objset_evict_dbufs(zfsvfs->z_os);
1446         dsl_dir_t *dd = os->os_dsl_dataset->ds_dir;
1447         dsl_dir_cancel_waiters(dd);
1448
1449         return (0);
1450 }
1451
1452 #if defined(HAVE_SUPER_SETUP_BDI_NAME)
1453 atomic_long_t zfs_bdi_seq = ATOMIC_LONG_INIT(0);
1454 #endif
1455
1456 int
1457 zfs_domount(struct super_block *sb, zfs_mnt_t *zm, int silent)
1458 {
1459         const char *osname = zm->mnt_osname;
1460         struct inode *root_inode = NULL;
1461         uint64_t recordsize;
1462         int error = 0;
1463         zfsvfs_t *zfsvfs = NULL;
1464         vfs_t *vfs = NULL;
1465         int canwrite;
1466         int dataset_visible_zone;
1467
1468         ASSERT(zm);
1469         ASSERT(osname);
1470
1471         dataset_visible_zone = zone_dataset_visible(osname, &canwrite);
1472
1473         /*
1474          * Refuse to mount a filesystem if we are in a namespace and the
1475          * dataset is not visible or writable in that namespace.
1476          */
1477         if (!INGLOBALZONE(curproc) &&
1478             (!dataset_visible_zone || !canwrite)) {
1479                 return (SET_ERROR(EPERM));
1480         }
1481
1482         error = zfsvfs_parse_options(zm->mnt_data, &vfs);
1483         if (error)
1484                 return (error);
1485
1486         /*
1487          * If a non-writable filesystem is being mounted without the
1488          * read-only flag, pretend it was set, as done for snapshots.
1489          */
1490         if (!canwrite)
1491                 vfs->vfs_readonly = true;
1492
1493         error = zfsvfs_create(osname, vfs->vfs_readonly, &zfsvfs);
1494         if (error) {
1495                 zfsvfs_vfs_free(vfs);
1496                 goto out;
1497         }
1498
1499         if ((error = dsl_prop_get_integer(osname, "recordsize",
1500             &recordsize, NULL))) {
1501                 zfsvfs_vfs_free(vfs);
1502                 goto out;
1503         }
1504
1505         vfs->vfs_data = zfsvfs;
1506         zfsvfs->z_vfs = vfs;
1507         zfsvfs->z_sb = sb;
1508         sb->s_fs_info = zfsvfs;
1509         sb->s_magic = ZFS_SUPER_MAGIC;
1510         sb->s_maxbytes = MAX_LFS_FILESIZE;
1511         sb->s_time_gran = 1;
1512         sb->s_blocksize = recordsize;
1513         sb->s_blocksize_bits = ilog2(recordsize);
1514
1515         error = -zpl_bdi_setup(sb, "zfs");
1516         if (error)
1517                 goto out;
1518
1519         sb->s_bdi->ra_pages = 0;
1520
1521         /* Set callback operations for the file system. */
1522         sb->s_op = &zpl_super_operations;
1523         sb->s_xattr = zpl_xattr_handlers;
1524         sb->s_export_op = &zpl_export_operations;
1525         sb->s_d_op = &zpl_dentry_operations;
1526
1527         /* Set features for file system. */
1528         zfs_set_fuid_feature(zfsvfs);
1529
1530         if (dmu_objset_is_snapshot(zfsvfs->z_os)) {
1531                 uint64_t pval;
1532
1533                 atime_changed_cb(zfsvfs, B_FALSE);
1534                 readonly_changed_cb(zfsvfs, B_TRUE);
1535                 if ((error = dsl_prop_get_integer(osname,
1536                     "xattr", &pval, NULL)))
1537                         goto out;
1538                 xattr_changed_cb(zfsvfs, pval);
1539                 if ((error = dsl_prop_get_integer(osname,
1540                     "acltype", &pval, NULL)))
1541                         goto out;
1542                 acltype_changed_cb(zfsvfs, pval);
1543                 zfsvfs->z_issnap = B_TRUE;
1544                 zfsvfs->z_os->os_sync = ZFS_SYNC_DISABLED;
1545                 zfsvfs->z_snap_defer_time = jiffies;
1546
1547                 mutex_enter(&zfsvfs->z_os->os_user_ptr_lock);
1548                 dmu_objset_set_user(zfsvfs->z_os, zfsvfs);
1549                 mutex_exit(&zfsvfs->z_os->os_user_ptr_lock);
1550         } else {
1551                 if ((error = zfsvfs_setup(zfsvfs, B_TRUE)))
1552                         goto out;
1553         }
1554
1555         /* Allocate a root inode for the filesystem. */
1556         error = zfs_root(zfsvfs, &root_inode);
1557         if (error) {
1558                 (void) zfs_umount(sb);
1559                 goto out;
1560         }
1561
1562         /* Allocate a root dentry for the filesystem */
1563         sb->s_root = d_make_root(root_inode);
1564         if (sb->s_root == NULL) {
1565                 (void) zfs_umount(sb);
1566                 error = SET_ERROR(ENOMEM);
1567                 goto out;
1568         }
1569
1570         if (!zfsvfs->z_issnap)
1571                 zfsctl_create(zfsvfs);
1572
1573         zfsvfs->z_arc_prune = arc_add_prune_callback(zpl_prune_sb, sb);
1574 out:
1575         if (error) {
1576                 if (zfsvfs != NULL) {
1577                         dmu_objset_disown(zfsvfs->z_os, B_TRUE, zfsvfs);
1578                         zfsvfs_free(zfsvfs);
1579                 }
1580                 /*
1581                  * make sure we don't have dangling sb->s_fs_info which
1582                  * zfs_preumount will use.
1583                  */
1584                 sb->s_fs_info = NULL;
1585         }
1586
1587         return (error);
1588 }
1589
1590 /*
1591  * Called when an unmount is requested and certain sanity checks have
1592  * already passed.  At this point no dentries or inodes have been reclaimed
1593  * from their respective caches.  We drop the extra reference on the .zfs
1594  * control directory to allow everything to be reclaimed.  All snapshots
1595  * must already have been unmounted to reach this point.
1596  */
1597 void
1598 zfs_preumount(struct super_block *sb)
1599 {
1600         zfsvfs_t *zfsvfs = sb->s_fs_info;
1601
1602         /* zfsvfs is NULL when zfs_domount fails during mount */
1603         if (zfsvfs) {
1604                 zfs_unlinked_drain_stop_wait(zfsvfs);
1605                 zfsctl_destroy(sb->s_fs_info);
1606                 /*
1607                  * Wait for zrele_async before entering evict_inodes in
1608                  * generic_shutdown_super. The reason we must finish before
1609                  * evict_inodes is when lazytime is on, or when zfs_purgedir
1610                  * calls zfs_zget, zrele would bump i_count from 0 to 1. This
1611                  * would race with the i_count check in evict_inodes. This means
1612                  * it could destroy the inode while we are still using it.
1613                  *
1614                  * We wait for two passes. xattr directories in the first pass
1615                  * may add xattr entries in zfs_purgedir, so in the second pass
1616                  * we wait for them. We don't use taskq_wait here because it is
1617                  * a pool wide taskq. Other mounted filesystems can constantly
1618                  * do zrele_async and there's no guarantee when taskq will be
1619                  * empty.
1620                  */
1621                 taskq_wait_outstanding(dsl_pool_zrele_taskq(
1622                     dmu_objset_pool(zfsvfs->z_os)), 0);
1623                 taskq_wait_outstanding(dsl_pool_zrele_taskq(
1624                     dmu_objset_pool(zfsvfs->z_os)), 0);
1625         }
1626 }
1627
1628 /*
1629  * Called once all other unmount released tear down has occurred.
1630  * It is our responsibility to release any remaining infrastructure.
1631  */
1632 int
1633 zfs_umount(struct super_block *sb)
1634 {
1635         zfsvfs_t *zfsvfs = sb->s_fs_info;
1636         objset_t *os;
1637
1638         if (zfsvfs->z_arc_prune != NULL)
1639                 arc_remove_prune_callback(zfsvfs->z_arc_prune);
1640         VERIFY(zfsvfs_teardown(zfsvfs, B_TRUE) == 0);
1641         os = zfsvfs->z_os;
1642         zpl_bdi_destroy(sb);
1643
1644         /*
1645          * z_os will be NULL if there was an error in
1646          * attempting to reopen zfsvfs.
1647          */
1648         if (os != NULL) {
1649                 /*
1650                  * Unset the objset user_ptr.
1651                  */
1652                 mutex_enter(&os->os_user_ptr_lock);
1653                 dmu_objset_set_user(os, NULL);
1654                 mutex_exit(&os->os_user_ptr_lock);
1655
1656                 /*
1657                  * Finally release the objset
1658                  */
1659                 dmu_objset_disown(os, B_TRUE, zfsvfs);
1660         }
1661
1662         zfsvfs_free(zfsvfs);
1663         return (0);
1664 }
1665
1666 int
1667 zfs_remount(struct super_block *sb, int *flags, zfs_mnt_t *zm)
1668 {
1669         zfsvfs_t *zfsvfs = sb->s_fs_info;
1670         vfs_t *vfsp;
1671         boolean_t issnap = dmu_objset_is_snapshot(zfsvfs->z_os);
1672         int error;
1673
1674         if ((issnap || !spa_writeable(dmu_objset_spa(zfsvfs->z_os))) &&
1675             !(*flags & SB_RDONLY)) {
1676                 *flags |= SB_RDONLY;
1677                 return (EROFS);
1678         }
1679
1680         error = zfsvfs_parse_options(zm->mnt_data, &vfsp);
1681         if (error)
1682                 return (error);
1683
1684         if (!zfs_is_readonly(zfsvfs) && (*flags & SB_RDONLY))
1685                 txg_wait_synced(dmu_objset_pool(zfsvfs->z_os), 0);
1686
1687         zfs_unregister_callbacks(zfsvfs);
1688         zfsvfs_vfs_free(zfsvfs->z_vfs);
1689
1690         vfsp->vfs_data = zfsvfs;
1691         zfsvfs->z_vfs = vfsp;
1692         if (!issnap)
1693                 (void) zfs_register_callbacks(vfsp);
1694
1695         return (error);
1696 }
1697
1698 int
1699 zfs_vget(struct super_block *sb, struct inode **ipp, fid_t *fidp)
1700 {
1701         zfsvfs_t        *zfsvfs = sb->s_fs_info;
1702         znode_t         *zp;
1703         uint64_t        object = 0;
1704         uint64_t        fid_gen = 0;
1705         uint64_t        gen_mask;
1706         uint64_t        zp_gen;
1707         int             i, err;
1708
1709         *ipp = NULL;
1710
1711         if (fidp->fid_len == SHORT_FID_LEN || fidp->fid_len == LONG_FID_LEN) {
1712                 zfid_short_t    *zfid = (zfid_short_t *)fidp;
1713
1714                 for (i = 0; i < sizeof (zfid->zf_object); i++)
1715                         object |= ((uint64_t)zfid->zf_object[i]) << (8 * i);
1716
1717                 for (i = 0; i < sizeof (zfid->zf_gen); i++)
1718                         fid_gen |= ((uint64_t)zfid->zf_gen[i]) << (8 * i);
1719         } else {
1720                 return (SET_ERROR(EINVAL));
1721         }
1722
1723         /* LONG_FID_LEN means snapdirs */
1724         if (fidp->fid_len == LONG_FID_LEN) {
1725                 zfid_long_t     *zlfid = (zfid_long_t *)fidp;
1726                 uint64_t        objsetid = 0;
1727                 uint64_t        setgen = 0;
1728
1729                 for (i = 0; i < sizeof (zlfid->zf_setid); i++)
1730                         objsetid |= ((uint64_t)zlfid->zf_setid[i]) << (8 * i);
1731
1732                 for (i = 0; i < sizeof (zlfid->zf_setgen); i++)
1733                         setgen |= ((uint64_t)zlfid->zf_setgen[i]) << (8 * i);
1734
1735                 if (objsetid != ZFSCTL_INO_SNAPDIRS - object) {
1736                         dprintf("snapdir fid: objsetid (%llu) != "
1737                             "ZFSCTL_INO_SNAPDIRS (%llu) - object (%llu)\n",
1738                             objsetid, ZFSCTL_INO_SNAPDIRS, object);
1739
1740                         return (SET_ERROR(EINVAL));
1741                 }
1742
1743                 if (fid_gen > 1 || setgen != 0) {
1744                         dprintf("snapdir fid: fid_gen (%llu) and setgen "
1745                             "(%llu)\n", fid_gen, setgen);
1746                         return (SET_ERROR(EINVAL));
1747                 }
1748
1749                 return (zfsctl_snapdir_vget(sb, objsetid, fid_gen, ipp));
1750         }
1751
1752         if ((err = zfs_enter(zfsvfs, FTAG)) != 0)
1753                 return (err);
1754         /* A zero fid_gen means we are in the .zfs control directories */
1755         if (fid_gen == 0 &&
1756             (object == ZFSCTL_INO_ROOT || object == ZFSCTL_INO_SNAPDIR)) {
1757                 *ipp = zfsvfs->z_ctldir;
1758                 ASSERT(*ipp != NULL);
1759                 if (object == ZFSCTL_INO_SNAPDIR) {
1760                         VERIFY(zfsctl_root_lookup(*ipp, "snapshot", ipp,
1761                             0, kcred, NULL, NULL) == 0);
1762                 } else {
1763                         /*
1764                          * Must have an existing ref, so igrab()
1765                          * cannot return NULL
1766                          */
1767                         VERIFY3P(igrab(*ipp), !=, NULL);
1768                 }
1769                 zfs_exit(zfsvfs, FTAG);
1770                 return (0);
1771         }
1772
1773         gen_mask = -1ULL >> (64 - 8 * i);
1774
1775         dprintf("getting %llu [%llu mask %llx]\n", object, fid_gen, gen_mask);
1776         if ((err = zfs_zget(zfsvfs, object, &zp))) {
1777                 zfs_exit(zfsvfs, FTAG);
1778                 return (err);
1779         }
1780
1781         /* Don't export xattr stuff */
1782         if (zp->z_pflags & ZFS_XATTR) {
1783                 zrele(zp);
1784                 zfs_exit(zfsvfs, FTAG);
1785                 return (SET_ERROR(ENOENT));
1786         }
1787
1788         (void) sa_lookup(zp->z_sa_hdl, SA_ZPL_GEN(zfsvfs), &zp_gen,
1789             sizeof (uint64_t));
1790         zp_gen = zp_gen & gen_mask;
1791         if (zp_gen == 0)
1792                 zp_gen = 1;
1793         if ((fid_gen == 0) && (zfsvfs->z_root == object))
1794                 fid_gen = zp_gen;
1795         if (zp->z_unlinked || zp_gen != fid_gen) {
1796                 dprintf("znode gen (%llu) != fid gen (%llu)\n", zp_gen,
1797                     fid_gen);
1798                 zrele(zp);
1799                 zfs_exit(zfsvfs, FTAG);
1800                 return (SET_ERROR(ENOENT));
1801         }
1802
1803         *ipp = ZTOI(zp);
1804         if (*ipp)
1805                 zfs_znode_update_vfs(ITOZ(*ipp));
1806
1807         zfs_exit(zfsvfs, FTAG);
1808         return (0);
1809 }
1810
1811 /*
1812  * Block out VFS ops and close zfsvfs_t
1813  *
1814  * Note, if successful, then we return with the 'z_teardown_lock' and
1815  * 'z_teardown_inactive_lock' write held.  We leave ownership of the underlying
1816  * dataset and objset intact so that they can be atomically handed off during
1817  * a subsequent rollback or recv operation and the resume thereafter.
1818  */
1819 int
1820 zfs_suspend_fs(zfsvfs_t *zfsvfs)
1821 {
1822         int error;
1823
1824         if ((error = zfsvfs_teardown(zfsvfs, B_FALSE)) != 0)
1825                 return (error);
1826
1827         return (0);
1828 }
1829
1830 /*
1831  * Rebuild SA and release VOPs.  Note that ownership of the underlying dataset
1832  * is an invariant across any of the operations that can be performed while the
1833  * filesystem was suspended.  Whether it succeeded or failed, the preconditions
1834  * are the same: the relevant objset and associated dataset are owned by
1835  * zfsvfs, held, and long held on entry.
1836  */
1837 int
1838 zfs_resume_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds)
1839 {
1840         int err, err2;
1841         znode_t *zp;
1842
1843         ASSERT(ZFS_TEARDOWN_WRITE_HELD(zfsvfs));
1844         ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
1845
1846         /*
1847          * We already own this, so just update the objset_t, as the one we
1848          * had before may have been evicted.
1849          */
1850         objset_t *os;
1851         VERIFY3P(ds->ds_owner, ==, zfsvfs);
1852         VERIFY(dsl_dataset_long_held(ds));
1853         dsl_pool_t *dp = spa_get_dsl(dsl_dataset_get_spa(ds));
1854         dsl_pool_config_enter(dp, FTAG);
1855         VERIFY0(dmu_objset_from_ds(ds, &os));
1856         dsl_pool_config_exit(dp, FTAG);
1857
1858         err = zfsvfs_init(zfsvfs, os);
1859         if (err != 0)
1860                 goto bail;
1861
1862         ds->ds_dir->dd_activity_cancelled = B_FALSE;
1863         VERIFY(zfsvfs_setup(zfsvfs, B_FALSE) == 0);
1864
1865         zfs_set_fuid_feature(zfsvfs);
1866         zfsvfs->z_rollback_time = jiffies;
1867
1868         /*
1869          * Attempt to re-establish all the active inodes with their
1870          * dbufs.  If a zfs_rezget() fails, then we unhash the inode
1871          * and mark it stale.  This prevents a collision if a new
1872          * inode/object is created which must use the same inode
1873          * number.  The stale inode will be be released when the
1874          * VFS prunes the dentry holding the remaining references
1875          * on the stale inode.
1876          */
1877         mutex_enter(&zfsvfs->z_znodes_lock);
1878         for (zp = list_head(&zfsvfs->z_all_znodes); zp;
1879             zp = list_next(&zfsvfs->z_all_znodes, zp)) {
1880                 err2 = zfs_rezget(zp);
1881                 if (err2) {
1882                         remove_inode_hash(ZTOI(zp));
1883                         zp->z_is_stale = B_TRUE;
1884                 }
1885
1886                 /* see comment in zfs_suspend_fs() */
1887                 if (zp->z_suspended) {
1888                         zfs_zrele_async(zp);
1889                         zp->z_suspended = B_FALSE;
1890                 }
1891         }
1892         mutex_exit(&zfsvfs->z_znodes_lock);
1893
1894         if (!zfs_is_readonly(zfsvfs) && !zfsvfs->z_unmounted) {
1895                 /*
1896                  * zfs_suspend_fs() could have interrupted freeing
1897                  * of dnodes. We need to restart this freeing so
1898                  * that we don't "leak" the space.
1899                  */
1900                 zfs_unlinked_drain(zfsvfs);
1901         }
1902
1903         /*
1904          * Most of the time zfs_suspend_fs is used for changing the contents
1905          * of the underlying dataset. ZFS rollback and receive operations
1906          * might create files for which negative dentries are present in
1907          * the cache. Since walking the dcache would require a lot of GPL-only
1908          * code duplication, it's much easier on these rather rare occasions
1909          * just to flush the whole dcache for the given dataset/filesystem.
1910          */
1911         shrink_dcache_sb(zfsvfs->z_sb);
1912
1913 bail:
1914         if (err != 0)
1915                 zfsvfs->z_unmounted = B_TRUE;
1916
1917         /* release the VFS ops */
1918         rw_exit(&zfsvfs->z_teardown_inactive_lock);
1919         ZFS_TEARDOWN_EXIT(zfsvfs, FTAG);
1920
1921         if (err != 0) {
1922                 /*
1923                  * Since we couldn't setup the sa framework, try to force
1924                  * unmount this file system.
1925                  */
1926                 if (zfsvfs->z_os)
1927                         (void) zfs_umount(zfsvfs->z_sb);
1928         }
1929         return (err);
1930 }
1931
1932 /*
1933  * Release VOPs and unmount a suspended filesystem.
1934  */
1935 int
1936 zfs_end_fs(zfsvfs_t *zfsvfs, dsl_dataset_t *ds)
1937 {
1938         ASSERT(ZFS_TEARDOWN_WRITE_HELD(zfsvfs));
1939         ASSERT(RW_WRITE_HELD(&zfsvfs->z_teardown_inactive_lock));
1940
1941         /*
1942          * We already own this, so just hold and rele it to update the
1943          * objset_t, as the one we had before may have been evicted.
1944          */
1945         objset_t *os;
1946         VERIFY3P(ds->ds_owner, ==, zfsvfs);
1947         VERIFY(dsl_dataset_long_held(ds));
1948         dsl_pool_t *dp = spa_get_dsl(dsl_dataset_get_spa(ds));
1949         dsl_pool_config_enter(dp, FTAG);
1950         VERIFY0(dmu_objset_from_ds(ds, &os));
1951         dsl_pool_config_exit(dp, FTAG);
1952         zfsvfs->z_os = os;
1953
1954         /* release the VOPs */
1955         rw_exit(&zfsvfs->z_teardown_inactive_lock);
1956         ZFS_TEARDOWN_EXIT(zfsvfs, FTAG);
1957
1958         /*
1959          * Try to force unmount this file system.
1960          */
1961         (void) zfs_umount(zfsvfs->z_sb);
1962         zfsvfs->z_unmounted = B_TRUE;
1963         return (0);
1964 }
1965
1966 /*
1967  * Automounted snapshots rely on periodic revalidation
1968  * to defer snapshots from being automatically unmounted.
1969  */
1970
1971 inline void
1972 zfs_exit_fs(zfsvfs_t *zfsvfs)
1973 {
1974         if (!zfsvfs->z_issnap)
1975                 return;
1976
1977         if (time_after(jiffies, zfsvfs->z_snap_defer_time +
1978             MAX(zfs_expire_snapshot * HZ / 2, HZ))) {
1979                 zfsvfs->z_snap_defer_time = jiffies;
1980                 zfsctl_snapshot_unmount_delay(zfsvfs->z_os->os_spa,
1981                     dmu_objset_id(zfsvfs->z_os),
1982                     zfs_expire_snapshot);
1983         }
1984 }
1985
1986 int
1987 zfs_set_version(zfsvfs_t *zfsvfs, uint64_t newvers)
1988 {
1989         int error;
1990         objset_t *os = zfsvfs->z_os;
1991         dmu_tx_t *tx;
1992
1993         if (newvers < ZPL_VERSION_INITIAL || newvers > ZPL_VERSION)
1994                 return (SET_ERROR(EINVAL));
1995
1996         if (newvers < zfsvfs->z_version)
1997                 return (SET_ERROR(EINVAL));
1998
1999         if (zfs_spa_version_map(newvers) >
2000             spa_version(dmu_objset_spa(zfsvfs->z_os)))
2001                 return (SET_ERROR(ENOTSUP));
2002
2003         tx = dmu_tx_create(os);
2004         dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_FALSE, ZPL_VERSION_STR);
2005         if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2006                 dmu_tx_hold_zap(tx, MASTER_NODE_OBJ, B_TRUE,
2007                     ZFS_SA_ATTRS);
2008                 dmu_tx_hold_zap(tx, DMU_NEW_OBJECT, FALSE, NULL);
2009         }
2010         error = dmu_tx_assign(tx, TXG_WAIT);
2011         if (error) {
2012                 dmu_tx_abort(tx);
2013                 return (error);
2014         }
2015
2016         error = zap_update(os, MASTER_NODE_OBJ, ZPL_VERSION_STR,
2017             8, 1, &newvers, tx);
2018
2019         if (error) {
2020                 dmu_tx_commit(tx);
2021                 return (error);
2022         }
2023
2024         if (newvers >= ZPL_VERSION_SA && !zfsvfs->z_use_sa) {
2025                 uint64_t sa_obj;
2026
2027                 ASSERT3U(spa_version(dmu_objset_spa(zfsvfs->z_os)), >=,
2028                     SPA_VERSION_SA);
2029                 sa_obj = zap_create(os, DMU_OT_SA_MASTER_NODE,
2030                     DMU_OT_NONE, 0, tx);
2031
2032                 error = zap_add(os, MASTER_NODE_OBJ,
2033                     ZFS_SA_ATTRS, 8, 1, &sa_obj, tx);
2034                 ASSERT0(error);
2035
2036                 VERIFY(0 == sa_set_sa_object(os, sa_obj));
2037                 sa_register_update_callback(os, zfs_sa_upgrade);
2038         }
2039
2040         spa_history_log_internal_ds(dmu_objset_ds(os), "upgrade", tx,
2041             "from %llu to %llu", zfsvfs->z_version, newvers);
2042
2043         dmu_tx_commit(tx);
2044
2045         zfsvfs->z_version = newvers;
2046         os->os_version = newvers;
2047
2048         zfs_set_fuid_feature(zfsvfs);
2049
2050         return (0);
2051 }
2052
2053 /*
2054  * Read a property stored within the master node.
2055  */
2056 int
2057 zfs_get_zplprop(objset_t *os, zfs_prop_t prop, uint64_t *value)
2058 {
2059         uint64_t *cached_copy = NULL;
2060
2061         /*
2062          * Figure out where in the objset_t the cached copy would live, if it
2063          * is available for the requested property.
2064          */
2065         if (os != NULL) {
2066                 switch (prop) {
2067                 case ZFS_PROP_VERSION:
2068                         cached_copy = &os->os_version;
2069                         break;
2070                 case ZFS_PROP_NORMALIZE:
2071                         cached_copy = &os->os_normalization;
2072                         break;
2073                 case ZFS_PROP_UTF8ONLY:
2074                         cached_copy = &os->os_utf8only;
2075                         break;
2076                 case ZFS_PROP_CASE:
2077                         cached_copy = &os->os_casesensitivity;
2078                         break;
2079                 default:
2080                         break;
2081                 }
2082         }
2083         if (cached_copy != NULL && *cached_copy != OBJSET_PROP_UNINITIALIZED) {
2084                 *value = *cached_copy;
2085                 return (0);
2086         }
2087
2088         /*
2089          * If the property wasn't cached, look up the file system's value for
2090          * the property. For the version property, we look up a slightly
2091          * different string.
2092          */
2093         const char *pname;
2094         int error = ENOENT;
2095         if (prop == ZFS_PROP_VERSION)
2096                 pname = ZPL_VERSION_STR;
2097         else
2098                 pname = zfs_prop_to_name(prop);
2099
2100         if (os != NULL) {
2101                 ASSERT3U(os->os_phys->os_type, ==, DMU_OST_ZFS);
2102                 error = zap_lookup(os, MASTER_NODE_OBJ, pname, 8, 1, value);
2103         }
2104
2105         if (error == ENOENT) {
2106                 /* No value set, use the default value */
2107                 switch (prop) {
2108                 case ZFS_PROP_VERSION:
2109                         *value = ZPL_VERSION;
2110                         break;
2111                 case ZFS_PROP_NORMALIZE:
2112                 case ZFS_PROP_UTF8ONLY:
2113                         *value = 0;
2114                         break;
2115                 case ZFS_PROP_CASE:
2116                         *value = ZFS_CASE_SENSITIVE;
2117                         break;
2118                 case ZFS_PROP_ACLTYPE:
2119                         *value = ZFS_ACLTYPE_OFF;
2120                         break;
2121                 default:
2122                         return (error);
2123                 }
2124                 error = 0;
2125         }
2126
2127         /*
2128          * If one of the methods for getting the property value above worked,
2129          * copy it into the objset_t's cache.
2130          */
2131         if (error == 0 && cached_copy != NULL) {
2132                 *cached_copy = *value;
2133         }
2134
2135         return (error);
2136 }
2137
2138 /*
2139  * Return true if the corresponding vfs's unmounted flag is set.
2140  * Otherwise return false.
2141  * If this function returns true we know VFS unmount has been initiated.
2142  */
2143 boolean_t
2144 zfs_get_vfs_flag_unmounted(objset_t *os)
2145 {
2146         zfsvfs_t *zfvp;
2147         boolean_t unmounted = B_FALSE;
2148
2149         ASSERT(dmu_objset_type(os) == DMU_OST_ZFS);
2150
2151         mutex_enter(&os->os_user_ptr_lock);
2152         zfvp = dmu_objset_get_user(os);
2153         if (zfvp != NULL && zfvp->z_unmounted)
2154                 unmounted = B_TRUE;
2155         mutex_exit(&os->os_user_ptr_lock);
2156
2157         return (unmounted);
2158 }
2159
2160 void
2161 zfsvfs_update_fromname(const char *oldname, const char *newname)
2162 {
2163         /*
2164          * We don't need to do anything here, the devname is always current by
2165          * virtue of zfsvfs->z_sb->s_op->show_devname.
2166          */
2167         (void) oldname, (void) newname;
2168 }
2169
2170 void
2171 zfs_init(void)
2172 {
2173         zfsctl_init();
2174         zfs_znode_init();
2175         dmu_objset_register_type(DMU_OST_ZFS, zpl_get_file_info);
2176         register_filesystem(&zpl_fs_type);
2177 }
2178
2179 void
2180 zfs_fini(void)
2181 {
2182         /*
2183          * we don't use outstanding because zpl_posix_acl_free might add more.
2184          */
2185         taskq_wait(system_delay_taskq);
2186         taskq_wait(system_taskq);
2187         unregister_filesystem(&zpl_fs_type);
2188         zfs_znode_fini();
2189         zfsctl_fini();
2190 }
2191
2192 #if defined(_KERNEL)
2193 EXPORT_SYMBOL(zfs_suspend_fs);
2194 EXPORT_SYMBOL(zfs_resume_fs);
2195 EXPORT_SYMBOL(zfs_set_version);
2196 EXPORT_SYMBOL(zfsvfs_create);
2197 EXPORT_SYMBOL(zfsvfs_free);
2198 EXPORT_SYMBOL(zfs_is_readonly);
2199 EXPORT_SYMBOL(zfs_domount);
2200 EXPORT_SYMBOL(zfs_preumount);
2201 EXPORT_SYMBOL(zfs_umount);
2202 EXPORT_SYMBOL(zfs_remount);
2203 EXPORT_SYMBOL(zfs_statvfs);
2204 EXPORT_SYMBOL(zfs_vget);
2205 EXPORT_SYMBOL(zfs_prune);
2206 #endif