fs/fuse/inode.c

   1 /*
   2   FUSE: Filesystem in Userspace
   3   Copyright (C) 2001-2008  Miklos Szeredi <[email protected]>
   4
   5   This program can be distributed under the terms of the GNU GPL.
   6   See the file COPYING.
   7 */
   8
   9 #include "fuse_i.h"
  10
  11 #include <linux/pagemap.h>
  12 #include <linux/slab.h>
  13 #include <linux/file.h>
  14 #include <linux/seq_file.h>
  15 #include <linux/init.h>
  16 #include <linux/module.h>
  17 #include <linux/moduleparam.h>
  18 #include <linux/fs_context.h>
  19 #include <linux/fs_parser.h>
  20 #include <linux/statfs.h>
  21 #include <linux/random.h>
  22 #include <linux/sched.h>
  23 #include <linux/exportfs.h>
  24 #include <linux/posix_acl.h>
  25 #include <linux/pid_namespace.h>
  26 #include <uapi/linux/magic.h>
  27
  28 MODULE_AUTHOR("Miklos Szeredi <[email protected]>");
  29 MODULE_DESCRIPTION("Filesystem in Userspace");
  30 MODULE_LICENSE("GPL");
  31
  32 static struct kmem_cache *fuse_inode_cachep;
  33 struct list_head fuse_conn_list;
  34 DEFINE_MUTEX(fuse_mutex);
  35
  36 static int set_global_limit(const char *val, const struct kernel_param *kp);
  37
  38 unsigned max_user_bgreq;
  39 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
  40                   &max_user_bgreq, 0644);
  41 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
  42 MODULE_PARM_DESC(max_user_bgreq,
  43  "Global limit for the maximum number of backgrounded requests an "
  44  "unprivileged user can set");
  45
  46 unsigned max_user_congthresh;
  47 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
  48                   &max_user_congthresh, 0644);
  49 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
  50 MODULE_PARM_DESC(max_user_congthresh,
  51  "Global limit for the maximum congestion threshold an "
  52  "unprivileged user can set");
  53
  54 #define FUSE_DEFAULT_BLKSIZE 512
  55
  56 /** Maximum number of outstanding background requests */
  57 #define FUSE_DEFAULT_MAX_BACKGROUND 12
  58
  59 /** Congestion starts at 75% of maximum */
  60 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
  61
  62 #ifdef CONFIG_BLOCK
  63 static struct file_system_type fuseblk_fs_type;
  64 #endif
  65
  66 struct fuse_forget_link *fuse_alloc_forget(void)
  67 {
  68         return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL_ACCOUNT);
  69 }
  70
  71 static struct inode *fuse_alloc_inode(struct super_block *sb)
  72 {
  73         struct fuse_inode *fi;
  74
  75         fi = alloc_inode_sb(sb, fuse_inode_cachep, GFP_KERNEL);
  76         if (!fi)
  77                 return NULL;
  78
  79         fi->i_time = 0;
  80         fi->inval_mask = ~0;
  81         fi->nodeid = 0;
  82         fi->nlookup = 0;
  83         fi->attr_version = 0;
  84         fi->orig_ino = 0;
  85         fi->state = 0;
  86         mutex_init(&fi->mutex);
  87         spin_lock_init(&fi->lock);
  88         fi->forget = fuse_alloc_forget();
  89         if (!fi->forget)
  90                 goto out_free;
  91
  92         if (IS_ENABLED(CONFIG_FUSE_DAX) && !fuse_dax_inode_alloc(sb, fi))
  93                 goto out_free_forget;
  94
  95         return &fi->inode;
  96
  97 out_free_forget:
  98         kfree(fi->forget);
  99 out_free:
 100         kmem_cache_free(fuse_inode_cachep, fi);
 101         return NULL;
 102 }
 103
 104 static void fuse_free_inode(struct inode *inode)
 105 {
 106         struct fuse_inode *fi = get_fuse_inode(inode);
 107
 108         mutex_destroy(&fi->mutex);
 109         kfree(fi->forget);
 110 #ifdef CONFIG_FUSE_DAX
 111         kfree(fi->dax);
 112 #endif
 113         kmem_cache_free(fuse_inode_cachep, fi);
 114 }
 115
 116 static void fuse_evict_inode(struct inode *inode)
 117 {
 118         struct fuse_inode *fi = get_fuse_inode(inode);
 119
 120         /* Will write inode on close/munmap and in all other dirtiers */
 121         WARN_ON(inode->i_state & I_DIRTY_INODE);
 122
 123         truncate_inode_pages_final(&inode->i_data);
 124         clear_inode(inode);
 125         if (inode->i_sb->s_flags & SB_ACTIVE) {
 126                 struct fuse_conn *fc = get_fuse_conn(inode);
 127
 128                 if (FUSE_IS_DAX(inode))
 129                         fuse_dax_inode_cleanup(inode);
 130                 if (fi->nlookup) {
 131                         fuse_queue_forget(fc, fi->forget, fi->nodeid,
 132                                           fi->nlookup);
 133                         fi->forget = NULL;
 134                 }
 135         }
 136         if (S_ISREG(inode->i_mode) && !fuse_is_bad(inode)) {
 137                 WARN_ON(!list_empty(&fi->write_files));
 138                 WARN_ON(!list_empty(&fi->queued_writes));
 139         }
 140 }
 141
 142 static int fuse_reconfigure(struct fs_context *fsc)
 143 {
 144         struct super_block *sb = fsc->root->d_sb;
 145
 146         sync_filesystem(sb);
 147         if (fsc->sb_flags & SB_MANDLOCK)
 148                 return -EINVAL;
 149
 150         return 0;
 151 }
 152
 153 /*
 154  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
 155  * so that it will fit.
 156  */
 157 static ino_t fuse_squash_ino(u64 ino64)
 158 {
 159         ino_t ino = (ino_t) ino64;
 160         if (sizeof(ino_t) < sizeof(u64))
 161                 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
 162         return ino;
 163 }
 164
 165 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
 166                                    struct fuse_statx *sx,
 167                                    u64 attr_valid, u32 cache_mask)
 168 {
 169         struct fuse_conn *fc = get_fuse_conn(inode);
 170         struct fuse_inode *fi = get_fuse_inode(inode);
 171
 172         lockdep_assert_held(&fi->lock);
 173
 174         fi->attr_version = atomic64_inc_return(&fc->attr_version);
 175         fi->i_time = attr_valid;
 176         /* Clear basic stats from invalid mask */
 177         set_mask_bits(&fi->inval_mask, STATX_BASIC_STATS, 0);
 178
 179         inode->i_ino     = fuse_squash_ino(attr->ino);
 180         inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
 181         set_nlink(inode, attr->nlink);
 182         inode->i_uid     = make_kuid(fc->user_ns, attr->uid);
 183         inode->i_gid     = make_kgid(fc->user_ns, attr->gid);
 184         inode->i_blocks  = attr->blocks;
 185
 186         /* Sanitize nsecs */
 187         attr->atimensec = min_t(u32, attr->atimensec, NSEC_PER_SEC - 1);
 188         attr->mtimensec = min_t(u32, attr->mtimensec, NSEC_PER_SEC - 1);
 189         attr->ctimensec = min_t(u32, attr->ctimensec, NSEC_PER_SEC - 1);
 190
 191         inode_set_atime(inode, attr->atime, attr->atimensec);
 192         /* mtime from server may be stale due to local buffered write */
 193         if (!(cache_mask & STATX_MTIME)) {
 194                 inode_set_mtime(inode, attr->mtime, attr->mtimensec);
 195         }
 196         if (!(cache_mask & STATX_CTIME)) {
 197                 inode_set_ctime(inode, attr->ctime, attr->ctimensec);
 198         }
 199         if (sx) {
 200                 /* Sanitize nsecs */
 201                 sx->btime.tv_nsec =
 202                         min_t(u32, sx->btime.tv_nsec, NSEC_PER_SEC - 1);
 203
 204                 /*
 205                  * Btime has been queried, cache is valid (whether or not btime
 206                  * is available or not) so clear STATX_BTIME from inval_mask.
 207                  *
 208                  * Availability of the btime attribute is indicated in
 209                  * FUSE_I_BTIME
 210                  */
 211                 set_mask_bits(&fi->inval_mask, STATX_BTIME, 0);
 212                 if (sx->mask & STATX_BTIME) {
 213                         set_bit(FUSE_I_BTIME, &fi->state);
 214                         fi->i_btime.tv_sec = sx->btime.tv_sec;
 215                         fi->i_btime.tv_nsec = sx->btime.tv_nsec;
 216                 }
 217         }
 218
 219         if (attr->blksize != 0)
 220                 inode->i_blkbits = ilog2(attr->blksize);
 221         else
 222                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
 223
 224         /*
 225          * Don't set the sticky bit in i_mode, unless we want the VFS
 226          * to check permissions.  This prevents failures due to the
 227          * check in may_delete().
 228          */
 229         fi->orig_i_mode = inode->i_mode;
 230         if (!fc->default_permissions)
 231                 inode->i_mode &= ~S_ISVTX;
 232
 233         fi->orig_ino = attr->ino;
 234
 235         /*
 236          * We are refreshing inode data and it is possible that another
 237          * client set suid/sgid or security.capability xattr. So clear
 238          * S_NOSEC. Ideally, we could have cleared it only if suid/sgid
 239          * was set or if security.capability xattr was set. But we don't
 240          * know if security.capability has been set or not. So clear it
 241          * anyway. Its less efficient but should be safe.
 242          */
 243         inode->i_flags &= ~S_NOSEC;
 244 }
 245
 246 u32 fuse_get_cache_mask(struct inode *inode)
 247 {
 248         struct fuse_conn *fc = get_fuse_conn(inode);
 249
 250         if (!fc->writeback_cache || !S_ISREG(inode->i_mode))
 251                 return 0;
 252
 253         return STATX_MTIME | STATX_CTIME | STATX_SIZE;
 254 }
 255
 256 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
 257                             struct fuse_statx *sx,
 258                             u64 attr_valid, u64 attr_version)
 259 {
 260         struct fuse_conn *fc = get_fuse_conn(inode);
 261         struct fuse_inode *fi = get_fuse_inode(inode);
 262         u32 cache_mask;
 263         loff_t oldsize;
 264         struct timespec64 old_mtime;
 265
 266         spin_lock(&fi->lock);
 267         /*
 268          * In case of writeback_cache enabled, writes update mtime, ctime and
 269          * may update i_size.  In these cases trust the cached value in the
 270          * inode.
 271          */
 272         cache_mask = fuse_get_cache_mask(inode);
 273         if (cache_mask & STATX_SIZE)
 274                 attr->size = i_size_read(inode);
 275
 276         if (cache_mask & STATX_MTIME) {
 277                 attr->mtime = inode_get_mtime_sec(inode);
 278                 attr->mtimensec = inode_get_mtime_nsec(inode);
 279         }
 280         if (cache_mask & STATX_CTIME) {
 281                 attr->ctime = inode_get_ctime_sec(inode);
 282                 attr->ctimensec = inode_get_ctime_nsec(inode);
 283         }
 284
 285         if ((attr_version != 0 && fi->attr_version > attr_version) ||
 286             test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
 287                 spin_unlock(&fi->lock);
 288                 return;
 289         }
 290
 291         old_mtime = inode_get_mtime(inode);
 292         fuse_change_attributes_common(inode, attr, sx, attr_valid, cache_mask);
 293
 294         oldsize = inode->i_size;
 295         /*
 296          * In case of writeback_cache enabled, the cached writes beyond EOF
 297          * extend local i_size without keeping userspace server in sync. So,
 298          * attr->size coming from server can be stale. We cannot trust it.
 299          */
 300         if (!(cache_mask & STATX_SIZE))
 301                 i_size_write(inode, attr->size);
 302         spin_unlock(&fi->lock);
 303
 304         if (!cache_mask && S_ISREG(inode->i_mode)) {
 305                 bool inval = false;
 306
 307                 if (oldsize != attr->size) {
 308                         truncate_pagecache(inode, attr->size);
 309                         if (!fc->explicit_inval_data)
 310                                 inval = true;
 311                 } else if (fc->auto_inval_data) {
 312                         struct timespec64 new_mtime = {
 313                                 .tv_sec = attr->mtime,
 314                                 .tv_nsec = attr->mtimensec,
 315                         };
 316
 317                         /*
 318                          * Auto inval mode also checks and invalidates if mtime
 319                          * has changed.
 320                          */
 321                         if (!timespec64_equal(&old_mtime, &new_mtime))
 322                                 inval = true;
 323                 }
 324
 325                 if (inval)
 326                         invalidate_inode_pages2(inode->i_mapping);
 327         }
 328
 329         if (IS_ENABLED(CONFIG_FUSE_DAX))
 330                 fuse_dax_dontcache(inode, attr->flags);
 331 }
 332
 333 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr,
 334                             struct fuse_conn *fc)
 335 {
 336         inode->i_mode = attr->mode & S_IFMT;
 337         inode->i_size = attr->size;
 338         inode_set_mtime(inode, attr->mtime, attr->mtimensec);
 339         inode_set_ctime(inode, attr->ctime, attr->ctimensec);
 340         if (S_ISREG(inode->i_mode)) {
 341                 fuse_init_common(inode);
 342                 fuse_init_file_inode(inode, attr->flags);
 343         } else if (S_ISDIR(inode->i_mode))
 344                 fuse_init_dir(inode);
 345         else if (S_ISLNK(inode->i_mode))
 346                 fuse_init_symlink(inode);
 347         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
 348                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
 349                 fuse_init_common(inode);
 350                 init_special_inode(inode, inode->i_mode,
 351                                    new_decode_dev(attr->rdev));
 352         } else
 353                 BUG();
 354         /*
 355          * Ensure that we don't cache acls for daemons without FUSE_POSIX_ACL
 356          * so they see the exact same behavior as before.
 357          */
 358         if (!fc->posix_acl)
 359                 inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
 360 }
 361
 362 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
 363 {
 364         u64 nodeid = *(u64 *) _nodeidp;
 365         if (get_node_id(inode) == nodeid)
 366                 return 1;
 367         else
 368                 return 0;
 369 }
 370
 371 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
 372 {
 373         u64 nodeid = *(u64 *) _nodeidp;
 374         get_fuse_inode(inode)->nodeid = nodeid;
 375         return 0;
 376 }
 377
 378 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
 379                         int generation, struct fuse_attr *attr,
 380                         u64 attr_valid, u64 attr_version)
 381 {
 382         struct inode *inode;
 383         struct fuse_inode *fi;
 384         struct fuse_conn *fc = get_fuse_conn_super(sb);
 385
 386         /*
 387          * Auto mount points get their node id from the submount root, which is
 388          * not a unique identifier within this filesystem.
 389          *
 390          * To avoid conflicts, do not place submount points into the inode hash
 391          * table.
 392          */
 393         if (fc->auto_submounts && (attr->flags & FUSE_ATTR_SUBMOUNT) &&
 394             S_ISDIR(attr->mode)) {
 395                 inode = new_inode(sb);
 396                 if (!inode)
 397                         return NULL;
 398
 399                 fuse_init_inode(inode, attr, fc);
 400                 get_fuse_inode(inode)->nodeid = nodeid;
 401                 inode->i_flags |= S_AUTOMOUNT;
 402                 goto done;
 403         }
 404
 405 retry:
 406         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
 407         if (!inode)
 408                 return NULL;
 409
 410         if ((inode->i_state & I_NEW)) {
 411                 inode->i_flags |= S_NOATIME;
 412                 if (!fc->writeback_cache || !S_ISREG(attr->mode))
 413                         inode->i_flags |= S_NOCMTIME;
 414                 inode->i_generation = generation;
 415                 fuse_init_inode(inode, attr, fc);
 416                 unlock_new_inode(inode);
 417         } else if (fuse_stale_inode(inode, generation, attr)) {
 418                 /* nodeid was reused, any I/O on the old inode should fail */
 419                 fuse_make_bad(inode);
 420                 iput(inode);
 421                 goto retry;
 422         }
 423 done:
 424         fi = get_fuse_inode(inode);
 425         spin_lock(&fi->lock);
 426         fi->nlookup++;
 427         spin_unlock(&fi->lock);
 428         fuse_change_attributes(inode, attr, NULL, attr_valid, attr_version);
 429
 430         return inode;
 431 }
 432
 433 struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid,
 434                            struct fuse_mount **fm)
 435 {
 436         struct fuse_mount *fm_iter;
 437         struct inode *inode;
 438
 439         WARN_ON(!rwsem_is_locked(&fc->killsb));
 440         list_for_each_entry(fm_iter, &fc->mounts, fc_entry) {
 441                 if (!fm_iter->sb)
 442                         continue;
 443
 444                 inode = ilookup5(fm_iter->sb, nodeid, fuse_inode_eq, &nodeid);
 445                 if (inode) {
 446                         if (fm)
 447                                 *fm = fm_iter;
 448                         return inode;
 449                 }
 450         }
 451
 452         return NULL;
 453 }
 454
 455 int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
 456                              loff_t offset, loff_t len)
 457 {
 458         struct fuse_inode *fi;
 459         struct inode *inode;
 460         pgoff_t pg_start;
 461         pgoff_t pg_end;
 462
 463         inode = fuse_ilookup(fc, nodeid, NULL);
 464         if (!inode)
 465                 return -ENOENT;
 466
 467         fi = get_fuse_inode(inode);
 468         spin_lock(&fi->lock);
 469         fi->attr_version = atomic64_inc_return(&fc->attr_version);
 470         spin_unlock(&fi->lock);
 471
 472         fuse_invalidate_attr(inode);
 473         forget_all_cached_acls(inode);
 474         if (offset >= 0) {
 475                 pg_start = offset >> PAGE_SHIFT;
 476                 if (len <= 0)
 477                         pg_end = -1;
 478                 else
 479                         pg_end = (offset + len - 1) >> PAGE_SHIFT;
 480                 invalidate_inode_pages2_range(inode->i_mapping,
 481                                               pg_start, pg_end);
 482         }
 483         iput(inode);
 484         return 0;
 485 }
 486
 487 bool fuse_lock_inode(struct inode *inode)
 488 {
 489         bool locked = false;
 490
 491         if (!get_fuse_conn(inode)->parallel_dirops) {
 492                 mutex_lock(&get_fuse_inode(inode)->mutex);
 493                 locked = true;
 494         }
 495
 496         return locked;
 497 }
 498
 499 void fuse_unlock_inode(struct inode *inode, bool locked)
 500 {
 501         if (locked)
 502                 mutex_unlock(&get_fuse_inode(inode)->mutex);
 503 }
 504
 505 static void fuse_umount_begin(struct super_block *sb)
 506 {
 507         struct fuse_conn *fc = get_fuse_conn_super(sb);
 508
 509         if (fc->no_force_umount)
 510                 return;
 511
 512         fuse_abort_conn(fc);
 513
 514         // Only retire block-device-based superblocks.
 515         if (sb->s_bdev != NULL)
 516                 retire_super(sb);
 517 }
 518
 519 static void fuse_send_destroy(struct fuse_mount *fm)
 520 {
 521         if (fm->fc->conn_init) {
 522                 FUSE_ARGS(args);
 523
 524                 args.opcode = FUSE_DESTROY;
 525                 args.force = true;
 526                 args.nocreds = true;
 527                 fuse_simple_request(fm, &args);
 528         }
 529 }
 530
 531 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
 532 {
 533         stbuf->f_type    = FUSE_SUPER_MAGIC;
 534         stbuf->f_bsize   = attr->bsize;
 535         stbuf->f_frsize  = attr->frsize;
 536         stbuf->f_blocks  = attr->blocks;
 537         stbuf->f_bfree   = attr->bfree;
 538         stbuf->f_bavail  = attr->bavail;
 539         stbuf->f_files   = attr->files;
 540         stbuf->f_ffree   = attr->ffree;
 541         stbuf->f_namelen = attr->namelen;
 542         /* fsid is left zero */
 543 }
 544
 545 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
 546 {
 547         struct super_block *sb = dentry->d_sb;
 548         struct fuse_mount *fm = get_fuse_mount_super(sb);
 549         FUSE_ARGS(args);
 550         struct fuse_statfs_out outarg;
 551         int err;
 552
 553         if (!fuse_allow_current_process(fm->fc)) {
 554                 buf->f_type = FUSE_SUPER_MAGIC;
 555                 return 0;
 556         }
 557
 558         memset(&outarg, 0, sizeof(outarg));
 559         args.in_numargs = 0;
 560         args.opcode = FUSE_STATFS;
 561         args.nodeid = get_node_id(d_inode(dentry));
 562         args.out_numargs = 1;
 563         args.out_args[0].size = sizeof(outarg);
 564         args.out_args[0].value = &outarg;
 565         err = fuse_simple_request(fm, &args);
 566         if (!err)
 567                 convert_fuse_statfs(buf, &outarg.st);
 568         return err;
 569 }
 570
 571 static struct fuse_sync_bucket *fuse_sync_bucket_alloc(void)
 572 {
 573         struct fuse_sync_bucket *bucket;
 574
 575         bucket = kzalloc(sizeof(*bucket), GFP_KERNEL | __GFP_NOFAIL);
 576         if (bucket) {
 577                 init_waitqueue_head(&bucket->waitq);
 578                 /* Initial active count */
 579                 atomic_set(&bucket->count, 1);
 580         }
 581         return bucket;
 582 }
 583
 584 static void fuse_sync_fs_writes(struct fuse_conn *fc)
 585 {
 586         struct fuse_sync_bucket *bucket, *new_bucket;
 587         int count;
 588
 589         new_bucket = fuse_sync_bucket_alloc();
 590         spin_lock(&fc->lock);
 591         bucket = rcu_dereference_protected(fc->curr_bucket, 1);
 592         count = atomic_read(&bucket->count);
 593         WARN_ON(count < 1);
 594         /* No outstanding writes? */
 595         if (count == 1) {
 596                 spin_unlock(&fc->lock);
 597                 kfree(new_bucket);
 598                 return;
 599         }
 600
 601         /*
 602          * Completion of new bucket depends on completion of this bucket, so add
 603          * one more count.
 604          */
 605         atomic_inc(&new_bucket->count);
 606         rcu_assign_pointer(fc->curr_bucket, new_bucket);
 607         spin_unlock(&fc->lock);
 608         /*
 609          * Drop initial active count.  At this point if all writes in this and
 610          * ancestor buckets complete, the count will go to zero and this task
 611          * will be woken up.
 612          */
 613         atomic_dec(&bucket->count);
 614
 615         wait_event(bucket->waitq, atomic_read(&bucket->count) == 0);
 616
 617         /* Drop temp count on descendant bucket */
 618         fuse_sync_bucket_dec(new_bucket);
 619         kfree_rcu(bucket, rcu);
 620 }
 621
 622 static int fuse_sync_fs(struct super_block *sb, int wait)
 623 {
 624         struct fuse_mount *fm = get_fuse_mount_super(sb);
 625         struct fuse_conn *fc = fm->fc;
 626         struct fuse_syncfs_in inarg;
 627         FUSE_ARGS(args);
 628         int err;
 629
 630         /*
 631          * Userspace cannot handle the wait == 0 case.  Avoid a
 632          * gratuitous roundtrip.
 633          */
 634         if (!wait)
 635                 return 0;
 636
 637         /* The filesystem is being unmounted.  Nothing to do. */
 638         if (!sb->s_root)
 639                 return 0;
 640
 641         if (!fc->sync_fs)
 642                 return 0;
 643
 644         fuse_sync_fs_writes(fc);
 645
 646         memset(&inarg, 0, sizeof(inarg));
 647         args.in_numargs = 1;
 648         args.in_args[0].size = sizeof(inarg);
 649         args.in_args[0].value = &inarg;
 650         args.opcode = FUSE_SYNCFS;
 651         args.nodeid = get_node_id(sb->s_root->d_inode);
 652         args.out_numargs = 0;
 653
 654         err = fuse_simple_request(fm, &args);
 655         if (err == -ENOSYS) {
 656                 fc->sync_fs = 0;
 657                 err = 0;
 658         }
 659
 660         return err;
 661 }
 662
 663 enum {
 664         OPT_SOURCE,
 665         OPT_SUBTYPE,
 666         OPT_FD,
 667         OPT_ROOTMODE,
 668         OPT_USER_ID,
 669         OPT_GROUP_ID,
 670         OPT_DEFAULT_PERMISSIONS,
 671         OPT_ALLOW_OTHER,
 672         OPT_MAX_READ,
 673         OPT_BLKSIZE,
 674         OPT_ERR
 675 };
 676
 677 static const struct fs_parameter_spec fuse_fs_parameters[] = {
 678         fsparam_string  ("source",              OPT_SOURCE),
 679         fsparam_u32     ("fd",                  OPT_FD),
 680         fsparam_u32oct  ("rootmode",            OPT_ROOTMODE),
 681         fsparam_u32     ("user_id",             OPT_USER_ID),
 682         fsparam_u32     ("group_id",            OPT_GROUP_ID),
 683         fsparam_flag    ("default_permissions", OPT_DEFAULT_PERMISSIONS),
 684         fsparam_flag    ("allow_other",         OPT_ALLOW_OTHER),
 685         fsparam_u32     ("max_read",            OPT_MAX_READ),
 686         fsparam_u32     ("blksize",             OPT_BLKSIZE),
 687         fsparam_string  ("subtype",             OPT_SUBTYPE),
 688         {}
 689 };
 690
 691 static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param)
 692 {
 693         struct fs_parse_result result;
 694         struct fuse_fs_context *ctx = fsc->fs_private;
 695         int opt;
 696
 697         if (fsc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
 698                 /*
 699                  * Ignore options coming from mount(MS_REMOUNT) for backward
 700                  * compatibility.
 701                  */
 702                 if (fsc->oldapi)
 703                         return 0;
 704
 705                 return invalfc(fsc, "No changes allowed in reconfigure");
 706         }
 707
 708         opt = fs_parse(fsc, fuse_fs_parameters, param, &result);
 709         if (opt < 0)
 710                 return opt;
 711
 712         switch (opt) {
 713         case OPT_SOURCE:
 714                 if (fsc->source)
 715                         return invalfc(fsc, "Multiple sources specified");
 716                 fsc->source = param->string;
 717                 param->string = NULL;
 718                 break;
 719
 720         case OPT_SUBTYPE:
 721                 if (ctx->subtype)
 722                         return invalfc(fsc, "Multiple subtypes specified");
 723                 ctx->subtype = param->string;
 724                 param->string = NULL;
 725                 return 0;
 726
 727         case OPT_FD:
 728                 ctx->fd = result.uint_32;
 729                 ctx->fd_present = true;
 730                 break;
 731
 732         case OPT_ROOTMODE:
 733                 if (!fuse_valid_type(result.uint_32))
 734                         return invalfc(fsc, "Invalid rootmode");
 735                 ctx->rootmode = result.uint_32;
 736                 ctx->rootmode_present = true;
 737                 break;
 738
 739         case OPT_USER_ID:
 740                 ctx->user_id = make_kuid(fsc->user_ns, result.uint_32);
 741                 if (!uid_valid(ctx->user_id))
 742                         return invalfc(fsc, "Invalid user_id");
 743                 ctx->user_id_present = true;
 744                 break;
 745
 746         case OPT_GROUP_ID:
 747                 ctx->group_id = make_kgid(fsc->user_ns, result.uint_32);
 748                 if (!gid_valid(ctx->group_id))
 749                         return invalfc(fsc, "Invalid group_id");
 750                 ctx->group_id_present = true;
 751                 break;
 752
 753         case OPT_DEFAULT_PERMISSIONS:
 754                 ctx->default_permissions = true;
 755                 break;
 756
 757         case OPT_ALLOW_OTHER:
 758                 ctx->allow_other = true;
 759                 break;
 760
 761         case OPT_MAX_READ:
 762                 ctx->max_read = result.uint_32;
 763                 break;
 764
 765         case OPT_BLKSIZE:
 766                 if (!ctx->is_bdev)
 767                         return invalfc(fsc, "blksize only supported for fuseblk");
 768                 ctx->blksize = result.uint_32;
 769                 break;
 770
 771         default:
 772                 return -EINVAL;
 773         }
 774
 775         return 0;
 776 }
 777
 778 static void fuse_free_fsc(struct fs_context *fsc)
 779 {
 780         struct fuse_fs_context *ctx = fsc->fs_private;
 781
 782         if (ctx) {
 783                 kfree(ctx->subtype);
 784                 kfree(ctx);
 785         }
 786 }
 787
 788 static int fuse_show_options(struct seq_file *m, struct dentry *root)
 789 {
 790         struct super_block *sb = root->d_sb;
 791         struct fuse_conn *fc = get_fuse_conn_super(sb);
 792
 793         if (fc->legacy_opts_show) {
 794                 seq_printf(m, ",user_id=%u",
 795                            from_kuid_munged(fc->user_ns, fc->user_id));
 796                 seq_printf(m, ",group_id=%u",
 797                            from_kgid_munged(fc->user_ns, fc->group_id));
 798                 if (fc->default_permissions)
 799                         seq_puts(m, ",default_permissions");
 800                 if (fc->allow_other)
 801                         seq_puts(m, ",allow_other");
 802                 if (fc->max_read != ~0)
 803                         seq_printf(m, ",max_read=%u", fc->max_read);
 804                 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
 805                         seq_printf(m, ",blksize=%lu", sb->s_blocksize);
 806         }
 807 #ifdef CONFIG_FUSE_DAX
 808         if (fc->dax_mode == FUSE_DAX_ALWAYS)
 809                 seq_puts(m, ",dax=always");
 810         else if (fc->dax_mode == FUSE_DAX_NEVER)
 811                 seq_puts(m, ",dax=never");
 812         else if (fc->dax_mode == FUSE_DAX_INODE_USER)
 813                 seq_puts(m, ",dax=inode");
 814 #endif
 815
 816         return 0;
 817 }
 818
 819 static void fuse_iqueue_init(struct fuse_iqueue *fiq,
 820                              const struct fuse_iqueue_ops *ops,
 821                              void *priv)
 822 {
 823         memset(fiq, 0, sizeof(struct fuse_iqueue));
 824         spin_lock_init(&fiq->lock);
 825         init_waitqueue_head(&fiq->waitq);
 826         INIT_LIST_HEAD(&fiq->pending);
 827         INIT_LIST_HEAD(&fiq->interrupts);
 828         fiq->forget_list_tail = &fiq->forget_list_head;
 829         fiq->connected = 1;
 830         fiq->ops = ops;
 831         fiq->priv = priv;
 832 }
 833
 834 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
 835 {
 836         unsigned int i;
 837
 838         spin_lock_init(&fpq->lock);
 839         for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
 840                 INIT_LIST_HEAD(&fpq->processing[i]);
 841         INIT_LIST_HEAD(&fpq->io);
 842         fpq->connected = 1;
 843 }
 844
 845 void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm,
 846                     struct user_namespace *user_ns,
 847                     const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
 848 {
 849         memset(fc, 0, sizeof(*fc));
 850         spin_lock_init(&fc->lock);
 851         spin_lock_init(&fc->bg_lock);
 852         init_rwsem(&fc->killsb);
 853         refcount_set(&fc->count, 1);
 854         atomic_set(&fc->dev_count, 1);
 855         init_waitqueue_head(&fc->blocked_waitq);
 856         fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
 857         INIT_LIST_HEAD(&fc->bg_queue);
 858         INIT_LIST_HEAD(&fc->entry);
 859         INIT_LIST_HEAD(&fc->devices);
 860         atomic_set(&fc->num_waiting, 0);
 861         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
 862         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
 863         atomic64_set(&fc->khctr, 0);
 864         fc->polled_files = RB_ROOT;
 865         fc->blocked = 0;
 866         fc->initialized = 0;
 867         fc->connected = 1;
 868         atomic64_set(&fc->attr_version, 1);
 869         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
 870         fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
 871         fc->user_ns = get_user_ns(user_ns);
 872         fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
 873         fc->max_pages_limit = FUSE_MAX_MAX_PAGES;
 874
 875         INIT_LIST_HEAD(&fc->mounts);
 876         list_add(&fm->fc_entry, &fc->mounts);
 877         fm->fc = fc;
 878 }
 879 EXPORT_SYMBOL_GPL(fuse_conn_init);
 880
 881 void fuse_conn_put(struct fuse_conn *fc)
 882 {
 883         if (refcount_dec_and_test(&fc->count)) {
 884                 struct fuse_iqueue *fiq = &fc->iq;
 885                 struct fuse_sync_bucket *bucket;
 886
 887                 if (IS_ENABLED(CONFIG_FUSE_DAX))
 888                         fuse_dax_conn_free(fc);
 889                 if (fiq->ops->release)
 890                         fiq->ops->release(fiq);
 891                 put_pid_ns(fc->pid_ns);
 892                 put_user_ns(fc->user_ns);
 893                 bucket = rcu_dereference_protected(fc->curr_bucket, 1);
 894                 if (bucket) {
 895                         WARN_ON(atomic_read(&bucket->count) != 1);
 896                         kfree(bucket);
 897                 }
 898                 fc->release(fc);
 899         }
 900 }
 901 EXPORT_SYMBOL_GPL(fuse_conn_put);
 902
 903 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
 904 {
 905         refcount_inc(&fc->count);
 906         return fc;
 907 }
 908 EXPORT_SYMBOL_GPL(fuse_conn_get);
 909
 910 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
 911 {
 912         struct fuse_attr attr;
 913         memset(&attr, 0, sizeof(attr));
 914
 915         attr.mode = mode;
 916         attr.ino = FUSE_ROOT_ID;
 917         attr.nlink = 1;
 918         return fuse_iget(sb, 1, 0, &attr, 0, 0);
 919 }
 920
 921 struct fuse_inode_handle {
 922         u64 nodeid;
 923         u32 generation;
 924 };
 925
 926 static struct dentry *fuse_get_dentry(struct super_block *sb,
 927                                       struct fuse_inode_handle *handle)
 928 {
 929         struct fuse_conn *fc = get_fuse_conn_super(sb);
 930         struct inode *inode;
 931         struct dentry *entry;
 932         int err = -ESTALE;
 933
 934         if (handle->nodeid == 0)
 935                 goto out_err;
 936
 937         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
 938         if (!inode) {
 939                 struct fuse_entry_out outarg;
 940                 const struct qstr name = QSTR_INIT(".", 1);
 941
 942                 if (!fc->export_support)
 943                         goto out_err;
 944
 945                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
 946                                        &inode);
 947                 if (err && err != -ENOENT)
 948                         goto out_err;
 949                 if (err || !inode) {
 950                         err = -ESTALE;
 951                         goto out_err;
 952                 }
 953                 err = -EIO;
 954                 if (get_node_id(inode) != handle->nodeid)
 955                         goto out_iput;
 956         }
 957         err = -ESTALE;
 958         if (inode->i_generation != handle->generation)
 959                 goto out_iput;
 960
 961         entry = d_obtain_alias(inode);
 962         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
 963                 fuse_invalidate_entry_cache(entry);
 964
 965         return entry;
 966
 967  out_iput:
 968         iput(inode);
 969  out_err:
 970         return ERR_PTR(err);
 971 }
 972
 973 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
 974                            struct inode *parent)
 975 {
 976         int len = parent ? 6 : 3;
 977         u64 nodeid;
 978         u32 generation;
 979
 980         if (*max_len < len) {
 981                 *max_len = len;
 982                 return  FILEID_INVALID;
 983         }
 984
 985         nodeid = get_fuse_inode(inode)->nodeid;
 986         generation = inode->i_generation;
 987
 988         fh[0] = (u32)(nodeid >> 32);
 989         fh[1] = (u32)(nodeid & 0xffffffff);
 990         fh[2] = generation;
 991
 992         if (parent) {
 993                 nodeid = get_fuse_inode(parent)->nodeid;
 994                 generation = parent->i_generation;
 995
 996                 fh[3] = (u32)(nodeid >> 32);
 997                 fh[4] = (u32)(nodeid & 0xffffffff);
 998                 fh[5] = generation;
 999         }
1000
1001         *max_len = len;
1002         return parent ? FILEID_INO64_GEN_PARENT : FILEID_INO64_GEN;
1003 }
1004
1005 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
1006                 struct fid *fid, int fh_len, int fh_type)
1007 {
1008         struct fuse_inode_handle handle;
1009
1010         if ((fh_type != FILEID_INO64_GEN &&
1011              fh_type != FILEID_INO64_GEN_PARENT) || fh_len < 3)
1012                 return NULL;
1013
1014         handle.nodeid = (u64) fid->raw[0] << 32;
1015         handle.nodeid |= (u64) fid->raw[1];
1016         handle.generation = fid->raw[2];
1017         return fuse_get_dentry(sb, &handle);
1018 }
1019
1020 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
1021                 struct fid *fid, int fh_len, int fh_type)
1022 {
1023         struct fuse_inode_handle parent;
1024
1025         if (fh_type != FILEID_INO64_GEN_PARENT || fh_len < 6)
1026                 return NULL;
1027
1028         parent.nodeid = (u64) fid->raw[3] << 32;
1029         parent.nodeid |= (u64) fid->raw[4];
1030         parent.generation = fid->raw[5];
1031         return fuse_get_dentry(sb, &parent);
1032 }
1033
1034 static struct dentry *fuse_get_parent(struct dentry *child)
1035 {
1036         struct inode *child_inode = d_inode(child);
1037         struct fuse_conn *fc = get_fuse_conn(child_inode);
1038         struct inode *inode;
1039         struct dentry *parent;
1040         struct fuse_entry_out outarg;
1041         int err;
1042
1043         if (!fc->export_support)
1044                 return ERR_PTR(-ESTALE);
1045
1046         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
1047                                &dotdot_name, &outarg, &inode);
1048         if (err) {
1049                 if (err == -ENOENT)
1050                         return ERR_PTR(-ESTALE);
1051                 return ERR_PTR(err);
1052         }
1053
1054         parent = d_obtain_alias(inode);
1055         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
1056                 fuse_invalidate_entry_cache(parent);
1057
1058         return parent;
1059 }
1060
1061 static const struct export_operations fuse_export_operations = {
1062         .fh_to_dentry   = fuse_fh_to_dentry,
1063         .fh_to_parent   = fuse_fh_to_parent,
1064         .encode_fh      = fuse_encode_fh,
1065         .get_parent     = fuse_get_parent,
1066 };
1067
1068 static const struct super_operations fuse_super_operations = {
1069         .alloc_inode    = fuse_alloc_inode,
1070         .free_inode     = fuse_free_inode,
1071         .evict_inode    = fuse_evict_inode,
1072         .write_inode    = fuse_write_inode,
1073         .drop_inode     = generic_delete_inode,
1074         .umount_begin   = fuse_umount_begin,
1075         .statfs         = fuse_statfs,
1076         .sync_fs        = fuse_sync_fs,
1077         .show_options   = fuse_show_options,
1078 };
1079
1080 static void sanitize_global_limit(unsigned *limit)
1081 {
1082         /*
1083          * The default maximum number of async requests is calculated to consume
1084          * 1/2^13 of the total memory, assuming 392 bytes per request.
1085          */
1086         if (*limit == 0)
1087                 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
1088
1089         if (*limit >= 1 << 16)
1090                 *limit = (1 << 16) - 1;
1091 }
1092
1093 static int set_global_limit(const char *val, const struct kernel_param *kp)
1094 {
1095         int rv;
1096
1097         rv = param_set_uint(val, kp);
1098         if (rv)
1099                 return rv;
1100
1101         sanitize_global_limit((unsigned *)kp->arg);
1102
1103         return 0;
1104 }
1105
1106 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
1107 {
1108         int cap_sys_admin = capable(CAP_SYS_ADMIN);
1109
1110         if (arg->minor < 13)
1111                 return;
1112
1113         sanitize_global_limit(&max_user_bgreq);
1114         sanitize_global_limit(&max_user_congthresh);
1115
1116         spin_lock(&fc->bg_lock);
1117         if (arg->max_background) {
1118                 fc->max_background = arg->max_background;
1119
1120                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
1121                         fc->max_background = max_user_bgreq;
1122         }
1123         if (arg->congestion_threshold) {
1124                 fc->congestion_threshold = arg->congestion_threshold;
1125
1126                 if (!cap_sys_admin &&
1127                     fc->congestion_threshold > max_user_congthresh)
1128                         fc->congestion_threshold = max_user_congthresh;
1129         }
1130         spin_unlock(&fc->bg_lock);
1131 }
1132
1133 struct fuse_init_args {
1134         struct fuse_args args;
1135         struct fuse_init_in in;
1136         struct fuse_init_out out;
1137 };
1138
1139 static void process_init_reply(struct fuse_mount *fm, struct fuse_args *args,
1140                                int error)
1141 {
1142         struct fuse_conn *fc = fm->fc;
1143         struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
1144         struct fuse_init_out *arg = &ia->out;
1145         bool ok = true;
1146
1147         if (error || arg->major != FUSE_KERNEL_VERSION)
1148                 ok = false;
1149         else {
1150                 unsigned long ra_pages;
1151
1152                 process_init_limits(fc, arg);
1153
1154                 if (arg->minor >= 6) {
1155                         u64 flags = arg->flags;
1156
1157                         if (flags & FUSE_INIT_EXT)
1158                                 flags |= (u64) arg->flags2 << 32;
1159
1160                         ra_pages = arg->max_readahead / PAGE_SIZE;
1161                         if (flags & FUSE_ASYNC_READ)
1162                                 fc->async_read = 1;
1163                         if (!(flags & FUSE_POSIX_LOCKS))
1164                                 fc->no_lock = 1;
1165                         if (arg->minor >= 17) {
1166                                 if (!(flags & FUSE_FLOCK_LOCKS))
1167                                         fc->no_flock = 1;
1168                         } else {
1169                                 if (!(flags & FUSE_POSIX_LOCKS))
1170                                         fc->no_flock = 1;
1171                         }
1172                         if (flags & FUSE_ATOMIC_O_TRUNC)
1173                                 fc->atomic_o_trunc = 1;
1174                         if (arg->minor >= 9) {
1175                                 /* LOOKUP has dependency on proto version */
1176                                 if (flags & FUSE_EXPORT_SUPPORT)
1177                                         fc->export_support = 1;
1178                         }
1179                         if (flags & FUSE_BIG_WRITES)
1180                                 fc->big_writes = 1;
1181                         if (flags & FUSE_DONT_MASK)
1182                                 fc->dont_mask = 1;
1183                         if (flags & FUSE_AUTO_INVAL_DATA)
1184                                 fc->auto_inval_data = 1;
1185                         else if (flags & FUSE_EXPLICIT_INVAL_DATA)
1186                                 fc->explicit_inval_data = 1;
1187                         if (flags & FUSE_DO_READDIRPLUS) {
1188                                 fc->do_readdirplus = 1;
1189                                 if (flags & FUSE_READDIRPLUS_AUTO)
1190                                         fc->readdirplus_auto = 1;
1191                         }
1192                         if (flags & FUSE_ASYNC_DIO)
1193                                 fc->async_dio = 1;
1194                         if (flags & FUSE_WRITEBACK_CACHE)
1195                                 fc->writeback_cache = 1;
1196                         if (flags & FUSE_PARALLEL_DIROPS)
1197                                 fc->parallel_dirops = 1;
1198                         if (flags & FUSE_HANDLE_KILLPRIV)
1199                                 fc->handle_killpriv = 1;
1200                         if (arg->time_gran && arg->time_gran <= 1000000000)
1201                                 fm->sb->s_time_gran = arg->time_gran;
1202                         if ((flags & FUSE_POSIX_ACL)) {
1203                                 fc->default_permissions = 1;
1204                                 fc->posix_acl = 1;
1205                         }
1206                         if (flags & FUSE_CACHE_SYMLINKS)
1207                                 fc->cache_symlinks = 1;
1208                         if (flags & FUSE_ABORT_ERROR)
1209                                 fc->abort_err = 1;
1210                         if (flags & FUSE_MAX_PAGES) {
1211                                 fc->max_pages =
1212                                         min_t(unsigned int, fc->max_pages_limit,
1213                                         max_t(unsigned int, arg->max_pages, 1));
1214                         }
1215                         if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1216                                 if (flags & FUSE_MAP_ALIGNMENT &&
1217                                     !fuse_dax_check_alignment(fc, arg->map_alignment)) {
1218                                         ok = false;
1219                                 }
1220                                 if (flags & FUSE_HAS_INODE_DAX)
1221                                         fc->inode_dax = 1;
1222                         }
1223                         if (flags & FUSE_HANDLE_KILLPRIV_V2) {
1224                                 fc->handle_killpriv_v2 = 1;
1225                                 fm->sb->s_flags |= SB_NOSEC;
1226                         }
1227                         if (flags & FUSE_SETXATTR_EXT)
1228                                 fc->setxattr_ext = 1;
1229                         if (flags & FUSE_SECURITY_CTX)
1230                                 fc->init_security = 1;
1231                         if (flags & FUSE_CREATE_SUPP_GROUP)
1232                                 fc->create_supp_group = 1;
1233                         if (flags & FUSE_DIRECT_IO_RELAX)
1234                                 fc->direct_io_relax = 1;
1235                 } else {
1236                         ra_pages = fc->max_read / PAGE_SIZE;
1237                         fc->no_lock = 1;
1238                         fc->no_flock = 1;
1239                 }
1240
1241                 fm->sb->s_bdi->ra_pages =
1242                                 min(fm->sb->s_bdi->ra_pages, ra_pages);
1243                 fc->minor = arg->minor;
1244                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
1245                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
1246                 fc->conn_init = 1;
1247         }
1248         kfree(ia);
1249
1250         if (!ok) {
1251                 fc->conn_init = 0;
1252                 fc->conn_error = 1;
1253         }
1254
1255         fuse_set_initialized(fc);
1256         wake_up_all(&fc->blocked_waitq);
1257 }
1258
1259 void fuse_send_init(struct fuse_mount *fm)
1260 {
1261         struct fuse_init_args *ia;
1262         u64 flags;
1263
1264         ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
1265
1266         ia->in.major = FUSE_KERNEL_VERSION;
1267         ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
1268         ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE;
1269         flags =
1270                 FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
1271                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
1272                 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1273                 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1274                 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1275                 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1276                 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1277                 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1278                 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA |
1279                 FUSE_HANDLE_KILLPRIV_V2 | FUSE_SETXATTR_EXT | FUSE_INIT_EXT |
1280                 FUSE_SECURITY_CTX | FUSE_CREATE_SUPP_GROUP |
1281                 FUSE_HAS_EXPIRE_ONLY | FUSE_DIRECT_IO_RELAX;
1282 #ifdef CONFIG_FUSE_DAX
1283         if (fm->fc->dax)
1284                 flags |= FUSE_MAP_ALIGNMENT;
1285         if (fuse_is_inode_dax_mode(fm->fc->dax_mode))
1286                 flags |= FUSE_HAS_INODE_DAX;
1287 #endif
1288         if (fm->fc->auto_submounts)
1289                 flags |= FUSE_SUBMOUNTS;
1290
1291         ia->in.flags = flags;
1292         ia->in.flags2 = flags >> 32;
1293
1294         ia->args.opcode = FUSE_INIT;
1295         ia->args.in_numargs = 1;
1296         ia->args.in_args[0].size = sizeof(ia->in);
1297         ia->args.in_args[0].value = &ia->in;
1298         ia->args.out_numargs = 1;
1299         /* Variable length argument used for backward compatibility
1300            with interface version < 7.5.  Rest of init_out is zeroed
1301            by do_get_request(), so a short reply is not a problem */
1302         ia->args.out_argvar = true;
1303         ia->args.out_args[0].size = sizeof(ia->out);
1304         ia->args.out_args[0].value = &ia->out;
1305         ia->args.force = true;
1306         ia->args.nocreds = true;
1307         ia->args.end = process_init_reply;
1308
1309         if (fuse_simple_background(fm, &ia->args, GFP_KERNEL) != 0)
1310                 process_init_reply(fm, &ia->args, -ENOTCONN);
1311 }
1312 EXPORT_SYMBOL_GPL(fuse_send_init);
1313
1314 void fuse_free_conn(struct fuse_conn *fc)
1315 {
1316         WARN_ON(!list_empty(&fc->devices));
1317         kfree_rcu(fc, rcu);
1318 }
1319 EXPORT_SYMBOL_GPL(fuse_free_conn);
1320
1321 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1322 {
1323         int err;
1324         char *suffix = "";
1325
1326         if (sb->s_bdev) {
1327                 suffix = "-fuseblk";
1328                 /*
1329                  * sb->s_bdi points to blkdev's bdi however we want to redirect
1330                  * it to our private bdi...
1331                  */
1332                 bdi_put(sb->s_bdi);
1333                 sb->s_bdi = &noop_backing_dev_info;
1334         }
1335         err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1336                                    MINOR(fc->dev), suffix);
1337         if (err)
1338                 return err;
1339
1340         /* fuse does it's own writeback accounting */
1341         sb->s_bdi->capabilities &= ~BDI_CAP_WRITEBACK_ACCT;
1342         sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT;
1343
1344         /*
1345          * For a single fuse filesystem use max 1% of dirty +
1346          * writeback threshold.
1347          *
1348          * This gives about 1M of write buffer for memory maps on a
1349          * machine with 1G and 10% dirty_ratio, which should be more
1350          * than enough.
1351          *
1352          * Privileged users can raise it by writing to
1353          *
1354          *    /sys/class/bdi/<bdi>/max_ratio
1355          */
1356         bdi_set_max_ratio(sb->s_bdi, 1);
1357
1358         return 0;
1359 }
1360
1361 struct fuse_dev *fuse_dev_alloc(void)
1362 {
1363         struct fuse_dev *fud;
1364         struct list_head *pq;
1365
1366         fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1367         if (!fud)
1368                 return NULL;
1369
1370         pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1371         if (!pq) {
1372                 kfree(fud);
1373                 return NULL;
1374         }
1375
1376         fud->pq.processing = pq;
1377         fuse_pqueue_init(&fud->pq);
1378
1379         return fud;
1380 }
1381 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1382
1383 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1384 {
1385         fud->fc = fuse_conn_get(fc);
1386         spin_lock(&fc->lock);
1387         list_add_tail(&fud->entry, &fc->devices);
1388         spin_unlock(&fc->lock);
1389 }
1390 EXPORT_SYMBOL_GPL(fuse_dev_install);
1391
1392 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1393 {
1394         struct fuse_dev *fud;
1395
1396         fud = fuse_dev_alloc();
1397         if (!fud)
1398                 return NULL;
1399
1400         fuse_dev_install(fud, fc);
1401         return fud;
1402 }
1403 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1404
1405 void fuse_dev_free(struct fuse_dev *fud)
1406 {
1407         struct fuse_conn *fc = fud->fc;
1408
1409         if (fc) {
1410                 spin_lock(&fc->lock);
1411                 list_del(&fud->entry);
1412                 spin_unlock(&fc->lock);
1413
1414                 fuse_conn_put(fc);
1415         }
1416         kfree(fud->pq.processing);
1417         kfree(fud);
1418 }
1419 EXPORT_SYMBOL_GPL(fuse_dev_free);
1420
1421 static void fuse_fill_attr_from_inode(struct fuse_attr *attr,
1422                                       const struct fuse_inode *fi)
1423 {
1424         struct timespec64 atime = inode_get_atime(&fi->inode);
1425         struct timespec64 mtime = inode_get_mtime(&fi->inode);
1426         struct timespec64 ctime = inode_get_ctime(&fi->inode);
1427
1428         *attr = (struct fuse_attr){
1429                 .ino            = fi->inode.i_ino,
1430                 .size           = fi->inode.i_size,
1431                 .blocks         = fi->inode.i_blocks,
1432                 .atime          = atime.tv_sec,
1433                 .mtime          = mtime.tv_sec,
1434                 .ctime          = ctime.tv_sec,
1435                 .atimensec      = atime.tv_nsec,
1436                 .mtimensec      = mtime.tv_nsec,
1437                 .ctimensec      = ctime.tv_nsec,
1438                 .mode           = fi->inode.i_mode,
1439                 .nlink          = fi->inode.i_nlink,
1440                 .uid            = fi->inode.i_uid.val,
1441                 .gid            = fi->inode.i_gid.val,
1442                 .rdev           = fi->inode.i_rdev,
1443                 .blksize        = 1u << fi->inode.i_blkbits,
1444         };
1445 }
1446
1447 static void fuse_sb_defaults(struct super_block *sb)
1448 {
1449         sb->s_magic = FUSE_SUPER_MAGIC;
1450         sb->s_op = &fuse_super_operations;
1451         sb->s_xattr = fuse_xattr_handlers;
1452         sb->s_maxbytes = MAX_LFS_FILESIZE;
1453         sb->s_time_gran = 1;
1454         sb->s_export_op = &fuse_export_operations;
1455         sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1456         if (sb->s_user_ns != &init_user_ns)
1457                 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1458         sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1459 }
1460
1461 static int fuse_fill_super_submount(struct super_block *sb,
1462                                     struct fuse_inode *parent_fi)
1463 {
1464         struct fuse_mount *fm = get_fuse_mount_super(sb);
1465         struct super_block *parent_sb = parent_fi->inode.i_sb;
1466         struct fuse_attr root_attr;
1467         struct inode *root;
1468
1469         fuse_sb_defaults(sb);
1470         fm->sb = sb;
1471
1472         WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1473         sb->s_bdi = bdi_get(parent_sb->s_bdi);
1474
1475         sb->s_xattr = parent_sb->s_xattr;
1476         sb->s_time_gran = parent_sb->s_time_gran;
1477         sb->s_blocksize = parent_sb->s_blocksize;
1478         sb->s_blocksize_bits = parent_sb->s_blocksize_bits;
1479         sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL);
1480         if (parent_sb->s_subtype && !sb->s_subtype)
1481                 return -ENOMEM;
1482
1483         fuse_fill_attr_from_inode(&root_attr, parent_fi);
1484         root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0);
1485         /*
1486          * This inode is just a duplicate, so it is not looked up and
1487          * its nlookup should not be incremented.  fuse_iget() does
1488          * that, though, so undo it here.
1489          */
1490         get_fuse_inode(root)->nlookup--;
1491         sb->s_d_op = &fuse_dentry_operations;
1492         sb->s_root = d_make_root(root);
1493         if (!sb->s_root)
1494                 return -ENOMEM;
1495
1496         return 0;
1497 }
1498
1499 /* Filesystem context private data holds the FUSE inode of the mount point */
1500 static int fuse_get_tree_submount(struct fs_context *fsc)
1501 {
1502         struct fuse_mount *fm;
1503         struct fuse_inode *mp_fi = fsc->fs_private;
1504         struct fuse_conn *fc = get_fuse_conn(&mp_fi->inode);
1505         struct super_block *sb;
1506         int err;
1507
1508         fm = kzalloc(sizeof(struct fuse_mount), GFP_KERNEL);
1509         if (!fm)
1510                 return -ENOMEM;
1511
1512         fm->fc = fuse_conn_get(fc);
1513         fsc->s_fs_info = fm;
1514         sb = sget_fc(fsc, NULL, set_anon_super_fc);
1515         if (fsc->s_fs_info)
1516                 fuse_mount_destroy(fm);
1517         if (IS_ERR(sb))
1518                 return PTR_ERR(sb);
1519
1520         /* Initialize superblock, making @mp_fi its root */
1521         err = fuse_fill_super_submount(sb, mp_fi);
1522         if (err) {
1523                 deactivate_locked_super(sb);
1524                 return err;
1525         }
1526
1527         down_write(&fc->killsb);
1528         list_add_tail(&fm->fc_entry, &fc->mounts);
1529         up_write(&fc->killsb);
1530
1531         sb->s_flags |= SB_ACTIVE;
1532         fsc->root = dget(sb->s_root);
1533
1534         return 0;
1535 }
1536
1537 static const struct fs_context_operations fuse_context_submount_ops = {
1538         .get_tree       = fuse_get_tree_submount,
1539 };
1540
1541 int fuse_init_fs_context_submount(struct fs_context *fsc)
1542 {
1543         fsc->ops = &fuse_context_submount_ops;
1544         return 0;
1545 }
1546 EXPORT_SYMBOL_GPL(fuse_init_fs_context_submount);
1547
1548 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1549 {
1550         struct fuse_dev *fud = NULL;
1551         struct fuse_mount *fm = get_fuse_mount_super(sb);
1552         struct fuse_conn *fc = fm->fc;
1553         struct inode *root;
1554         struct dentry *root_dentry;
1555         int err;
1556
1557         err = -EINVAL;
1558         if (sb->s_flags & SB_MANDLOCK)
1559                 goto err;
1560
1561         rcu_assign_pointer(fc->curr_bucket, fuse_sync_bucket_alloc());
1562         fuse_sb_defaults(sb);
1563
1564         if (ctx->is_bdev) {
1565 #ifdef CONFIG_BLOCK
1566                 err = -EINVAL;
1567                 if (!sb_set_blocksize(sb, ctx->blksize))
1568                         goto err;
1569 #endif
1570         } else {
1571                 sb->s_blocksize = PAGE_SIZE;
1572                 sb->s_blocksize_bits = PAGE_SHIFT;
1573         }
1574
1575         sb->s_subtype = ctx->subtype;
1576         ctx->subtype = NULL;
1577         if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1578                 err = fuse_dax_conn_alloc(fc, ctx->dax_mode, ctx->dax_dev);
1579                 if (err)
1580                         goto err;
1581         }
1582
1583         if (ctx->fudptr) {
1584                 err = -ENOMEM;
1585                 fud = fuse_dev_alloc_install(fc);
1586                 if (!fud)
1587                         goto err_free_dax;
1588         }
1589
1590         fc->dev = sb->s_dev;
1591         fm->sb = sb;
1592         err = fuse_bdi_init(fc, sb);
1593         if (err)
1594                 goto err_dev_free;
1595
1596         /* Handle umasking inside the fuse code */
1597         if (sb->s_flags & SB_POSIXACL)
1598                 fc->dont_mask = 1;
1599         sb->s_flags |= SB_POSIXACL;
1600
1601         fc->default_permissions = ctx->default_permissions;
1602         fc->allow_other = ctx->allow_other;
1603         fc->user_id = ctx->user_id;
1604         fc->group_id = ctx->group_id;
1605         fc->legacy_opts_show = ctx->legacy_opts_show;
1606         fc->max_read = max_t(unsigned int, 4096, ctx->max_read);
1607         fc->destroy = ctx->destroy;
1608         fc->no_control = ctx->no_control;
1609         fc->no_force_umount = ctx->no_force_umount;
1610
1611         err = -ENOMEM;
1612         root = fuse_get_root_inode(sb, ctx->rootmode);
1613         sb->s_d_op = &fuse_root_dentry_operations;
1614         root_dentry = d_make_root(root);
1615         if (!root_dentry)
1616                 goto err_dev_free;
1617         /* Root dentry doesn't have .d_revalidate */
1618         sb->s_d_op = &fuse_dentry_operations;
1619
1620         mutex_lock(&fuse_mutex);
1621         err = -EINVAL;
1622         if (ctx->fudptr && *ctx->fudptr)
1623                 goto err_unlock;
1624
1625         err = fuse_ctl_add_conn(fc);
1626         if (err)
1627                 goto err_unlock;
1628
1629         list_add_tail(&fc->entry, &fuse_conn_list);
1630         sb->s_root = root_dentry;
1631         if (ctx->fudptr)
1632                 *ctx->fudptr = fud;
1633         mutex_unlock(&fuse_mutex);
1634         return 0;
1635
1636  err_unlock:
1637         mutex_unlock(&fuse_mutex);
1638         dput(root_dentry);
1639  err_dev_free:
1640         if (fud)
1641                 fuse_dev_free(fud);
1642  err_free_dax:
1643         if (IS_ENABLED(CONFIG_FUSE_DAX))
1644                 fuse_dax_conn_free(fc);
1645  err:
1646         return err;
1647 }
1648 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1649
1650 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1651 {
1652         struct fuse_fs_context *ctx = fsc->fs_private;
1653         int err;
1654
1655         if (!ctx->file || !ctx->rootmode_present ||
1656             !ctx->user_id_present || !ctx->group_id_present)
1657                 return -EINVAL;
1658
1659         /*
1660          * Require mount to happen from the same user namespace which
1661          * opened /dev/fuse to prevent potential attacks.
1662          */
1663         if ((ctx->file->f_op != &fuse_dev_operations) ||
1664             (ctx->file->f_cred->user_ns != sb->s_user_ns))
1665                 return -EINVAL;
1666         ctx->fudptr = &ctx->file->private_data;
1667
1668         err = fuse_fill_super_common(sb, ctx);
1669         if (err)
1670                 return err;
1671         /* file->private_data shall be visible on all CPUs after this */
1672         smp_mb();
1673         fuse_send_init(get_fuse_mount_super(sb));
1674         return 0;
1675 }
1676
1677 /*
1678  * This is the path where user supplied an already initialized fuse dev.  In
1679  * this case never create a new super if the old one is gone.
1680  */
1681 static int fuse_set_no_super(struct super_block *sb, struct fs_context *fsc)
1682 {
1683         return -ENOTCONN;
1684 }
1685
1686 static int fuse_test_super(struct super_block *sb, struct fs_context *fsc)
1687 {
1688
1689         return fsc->sget_key == get_fuse_conn_super(sb);
1690 }
1691
1692 static int fuse_get_tree(struct fs_context *fsc)
1693 {
1694         struct fuse_fs_context *ctx = fsc->fs_private;
1695         struct fuse_dev *fud;
1696         struct fuse_conn *fc;
1697         struct fuse_mount *fm;
1698         struct super_block *sb;
1699         int err;
1700
1701         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1702         if (!fc)
1703                 return -ENOMEM;
1704
1705         fm = kzalloc(sizeof(*fm), GFP_KERNEL);
1706         if (!fm) {
1707                 kfree(fc);
1708                 return -ENOMEM;
1709         }
1710
1711         fuse_conn_init(fc, fm, fsc->user_ns, &fuse_dev_fiq_ops, NULL);
1712         fc->release = fuse_free_conn;
1713
1714         fsc->s_fs_info = fm;
1715
1716         if (ctx->fd_present)
1717                 ctx->file = fget(ctx->fd);
1718
1719         if (IS_ENABLED(CONFIG_BLOCK) && ctx->is_bdev) {
1720                 err = get_tree_bdev(fsc, fuse_fill_super);
1721                 goto out;
1722         }
1723         /*
1724          * While block dev mount can be initialized with a dummy device fd
1725          * (found by device name), normal fuse mounts can't
1726          */
1727         err = -EINVAL;
1728         if (!ctx->file)
1729                 goto out;
1730
1731         /*
1732          * Allow creating a fuse mount with an already initialized fuse
1733          * connection
1734          */
1735         fud = READ_ONCE(ctx->file->private_data);
1736         if (ctx->file->f_op == &fuse_dev_operations && fud) {
1737                 fsc->sget_key = fud->fc;
1738                 sb = sget_fc(fsc, fuse_test_super, fuse_set_no_super);
1739                 err = PTR_ERR_OR_ZERO(sb);
1740                 if (!IS_ERR(sb))
1741                         fsc->root = dget(sb->s_root);
1742         } else {
1743                 err = get_tree_nodev(fsc, fuse_fill_super);
1744         }
1745 out:
1746         if (fsc->s_fs_info)
1747                 fuse_mount_destroy(fm);
1748         if (ctx->file)
1749                 fput(ctx->file);
1750         return err;
1751 }
1752
1753 static const struct fs_context_operations fuse_context_ops = {
1754         .free           = fuse_free_fsc,
1755         .parse_param    = fuse_parse_param,
1756         .reconfigure    = fuse_reconfigure,
1757         .get_tree       = fuse_get_tree,
1758 };
1759
1760 /*
1761  * Set up the filesystem mount context.
1762  */
1763 static int fuse_init_fs_context(struct fs_context *fsc)
1764 {
1765         struct fuse_fs_context *ctx;
1766
1767         ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1768         if (!ctx)
1769                 return -ENOMEM;
1770
1771         ctx->max_read = ~0;
1772         ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1773         ctx->legacy_opts_show = true;
1774
1775 #ifdef CONFIG_BLOCK
1776         if (fsc->fs_type == &fuseblk_fs_type) {
1777                 ctx->is_bdev = true;
1778                 ctx->destroy = true;
1779         }
1780 #endif
1781
1782         fsc->fs_private = ctx;
1783         fsc->ops = &fuse_context_ops;
1784         return 0;
1785 }
1786
1787 bool fuse_mount_remove(struct fuse_mount *fm)
1788 {
1789         struct fuse_conn *fc = fm->fc;
1790         bool last = false;
1791
1792         down_write(&fc->killsb);
1793         list_del_init(&fm->fc_entry);
1794         if (list_empty(&fc->mounts))
1795                 last = true;
1796         up_write(&fc->killsb);
1797
1798         return last;
1799 }
1800 EXPORT_SYMBOL_GPL(fuse_mount_remove);
1801
1802 void fuse_conn_destroy(struct fuse_mount *fm)
1803 {
1804         struct fuse_conn *fc = fm->fc;
1805
1806         if (fc->destroy)
1807                 fuse_send_destroy(fm);
1808
1809         fuse_abort_conn(fc);
1810         fuse_wait_aborted(fc);
1811
1812         if (!list_empty(&fc->entry)) {
1813                 mutex_lock(&fuse_mutex);
1814                 list_del(&fc->entry);
1815                 fuse_ctl_remove_conn(fc);
1816                 mutex_unlock(&fuse_mutex);
1817         }
1818 }
1819 EXPORT_SYMBOL_GPL(fuse_conn_destroy);
1820
1821 static void fuse_sb_destroy(struct super_block *sb)
1822 {
1823         struct fuse_mount *fm = get_fuse_mount_super(sb);
1824         bool last;
1825
1826         if (sb->s_root) {
1827                 last = fuse_mount_remove(fm);
1828                 if (last)
1829                         fuse_conn_destroy(fm);
1830         }
1831 }
1832
1833 void fuse_mount_destroy(struct fuse_mount *fm)
1834 {
1835         fuse_conn_put(fm->fc);
1836         kfree(fm);
1837 }
1838 EXPORT_SYMBOL(fuse_mount_destroy);
1839
1840 static void fuse_kill_sb_anon(struct super_block *sb)
1841 {
1842         fuse_sb_destroy(sb);
1843         kill_anon_super(sb);
1844         fuse_mount_destroy(get_fuse_mount_super(sb));
1845 }
1846
1847 static struct file_system_type fuse_fs_type = {
1848         .owner          = THIS_MODULE,
1849         .name           = "fuse",
1850         .fs_flags       = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1851         .init_fs_context = fuse_init_fs_context,
1852         .parameters     = fuse_fs_parameters,
1853         .kill_sb        = fuse_kill_sb_anon,
1854 };
1855 MODULE_ALIAS_FS("fuse");
1856
1857 #ifdef CONFIG_BLOCK
1858 static void fuse_kill_sb_blk(struct super_block *sb)
1859 {
1860         fuse_sb_destroy(sb);
1861         kill_block_super(sb);
1862         fuse_mount_destroy(get_fuse_mount_super(sb));
1863 }
1864
1865 static struct file_system_type fuseblk_fs_type = {
1866         .owner          = THIS_MODULE,
1867         .name           = "fuseblk",
1868         .init_fs_context = fuse_init_fs_context,
1869         .parameters     = fuse_fs_parameters,
1870         .kill_sb        = fuse_kill_sb_blk,
1871         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1872 };
1873 MODULE_ALIAS_FS("fuseblk");
1874
1875 static inline int register_fuseblk(void)
1876 {
1877         return register_filesystem(&fuseblk_fs_type);
1878 }
1879
1880 static inline void unregister_fuseblk(void)
1881 {
1882         unregister_filesystem(&fuseblk_fs_type);
1883 }
1884 #else
1885 static inline int register_fuseblk(void)
1886 {
1887         return 0;
1888 }
1889
1890 static inline void unregister_fuseblk(void)
1891 {
1892 }
1893 #endif
1894
1895 static void fuse_inode_init_once(void *foo)
1896 {
1897         struct inode *inode = foo;
1898
1899         inode_init_once(inode);
1900 }
1901
1902 static int __init fuse_fs_init(void)
1903 {
1904         int err;
1905
1906         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1907                         sizeof(struct fuse_inode), 0,
1908                         SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1909                         fuse_inode_init_once);
1910         err = -ENOMEM;
1911         if (!fuse_inode_cachep)
1912                 goto out;
1913
1914         err = register_fuseblk();
1915         if (err)
1916                 goto out2;
1917
1918         err = register_filesystem(&fuse_fs_type);
1919         if (err)
1920                 goto out3;
1921
1922         return 0;
1923
1924  out3:
1925         unregister_fuseblk();
1926  out2:
1927         kmem_cache_destroy(fuse_inode_cachep);
1928  out:
1929         return err;
1930 }
1931
1932 static void fuse_fs_cleanup(void)
1933 {
1934         unregister_filesystem(&fuse_fs_type);
1935         unregister_fuseblk();
1936
1937         /*
1938          * Make sure all delayed rcu free inodes are flushed before we
1939          * destroy cache.
1940          */
1941         rcu_barrier();
1942         kmem_cache_destroy(fuse_inode_cachep);
1943 }
1944
1945 static struct kobject *fuse_kobj;
1946
1947 static int fuse_sysfs_init(void)
1948 {
1949         int err;
1950
1951         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1952         if (!fuse_kobj) {
1953                 err = -ENOMEM;
1954                 goto out_err;
1955         }
1956
1957         err = sysfs_create_mount_point(fuse_kobj, "connections");
1958         if (err)
1959                 goto out_fuse_unregister;
1960
1961         return 0;
1962
1963  out_fuse_unregister:
1964         kobject_put(fuse_kobj);
1965  out_err:
1966         return err;
1967 }
1968
1969 static void fuse_sysfs_cleanup(void)
1970 {
1971         sysfs_remove_mount_point(fuse_kobj, "connections");
1972         kobject_put(fuse_kobj);
1973 }
1974
1975 static int __init fuse_init(void)
1976 {
1977         int res;
1978
1979         pr_info("init (API version %i.%i)\n",
1980                 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1981
1982         INIT_LIST_HEAD(&fuse_conn_list);
1983         res = fuse_fs_init();
1984         if (res)
1985                 goto err;
1986
1987         res = fuse_dev_init();
1988         if (res)
1989                 goto err_fs_cleanup;
1990
1991         res = fuse_sysfs_init();
1992         if (res)
1993                 goto err_dev_cleanup;
1994
1995         res = fuse_ctl_init();
1996         if (res)
1997                 goto err_sysfs_cleanup;
1998
1999         sanitize_global_limit(&max_user_bgreq);
2000         sanitize_global_limit(&max_user_congthresh);
2001
2002         return 0;
2003
2004  err_sysfs_cleanup:
2005         fuse_sysfs_cleanup();
2006  err_dev_cleanup:
2007         fuse_dev_cleanup();
2008  err_fs_cleanup:
2009         fuse_fs_cleanup();
2010  err:
2011         return res;
2012 }
2013
2014 static void __exit fuse_exit(void)
2015 {
2016         pr_debug("exit\n");
2017
2018         fuse_ctl_cleanup();
2019         fuse_sysfs_cleanup();
2020         fuse_fs_cleanup();
2021         fuse_dev_cleanup();
2022 }
2023
2024 module_init(fuse_init);
2025 module_exit(fuse_exit);