fs/fuse/inode.c

   1 /*
   2   FUSE: Filesystem in Userspace
   3   Copyright (C) 2001-2008  Miklos Szeredi <miklos@szeredi.hu>
   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/parser.h>
  19 #include <linux/statfs.h>
  20 #include <linux/random.h>
  21 #include <linux/sched.h>
  22 #include <linux/exportfs.h>
  23 #include <linux/posix_acl.h>
  24 #include <linux/pid_namespace.h>
  25
  26 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  27 MODULE_DESCRIPTION("Filesystem in Userspace");
  28 MODULE_LICENSE("GPL");
  29
  30 static struct kmem_cache *fuse_inode_cachep;
  31 struct list_head fuse_conn_list;
  32 DEFINE_MUTEX(fuse_mutex);
  33
  34 static int set_global_limit(const char *val, const struct kernel_param *kp);
  35
  36 unsigned max_user_bgreq;
  37 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
  38                   &max_user_bgreq, 0644);
  39 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
  40 MODULE_PARM_DESC(max_user_bgreq,
  41  "Global limit for the maximum number of backgrounded requests an "
  42  "unprivileged user can set");
  43
  44 unsigned max_user_congthresh;
  45 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
  46                   &max_user_congthresh, 0644);
  47 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
  48 MODULE_PARM_DESC(max_user_congthresh,
  49  "Global limit for the maximum congestion threshold an "
  50  "unprivileged user can set");
  51
  52 #define FUSE_SUPER_MAGIC 0x65735546
  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 struct fuse_mount_data {
  63         int fd;
  64         unsigned rootmode;
  65         kuid_t user_id;
  66         kgid_t group_id;
  67         unsigned fd_present:1;
  68         unsigned rootmode_present:1;
  69         unsigned user_id_present:1;
  70         unsigned group_id_present:1;
  71         unsigned default_permissions:1;
  72         unsigned allow_other:1;
  73         unsigned max_read;
  74         unsigned blksize;
  75 };
  76
  77 struct fuse_forget_link *fuse_alloc_forget(void)
  78 {
  79         return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL);
  80 }
  81
  82 static struct inode *fuse_alloc_inode(struct super_block *sb)
  83 {
  84         struct inode *inode;
  85         struct fuse_inode *fi;
  86
  87         inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
  88         if (!inode)
  89                 return NULL;
  90
  91         fi = get_fuse_inode(inode);
  92         fi->i_time = 0;
  93         fi->nodeid = 0;
  94         fi->nlookup = 0;
  95         fi->attr_version = 0;
  96         fi->writectr = 0;
  97         fi->orig_ino = 0;
  98         fi->state = 0;
  99         INIT_LIST_HEAD(&fi->write_files);
 100         INIT_LIST_HEAD(&fi->queued_writes);
 101         INIT_LIST_HEAD(&fi->writepages);
 102         init_waitqueue_head(&fi->page_waitq);
 103         mutex_init(&fi->mutex);
 104         fi->forget = fuse_alloc_forget();
 105         if (!fi->forget) {
 106                 kmem_cache_free(fuse_inode_cachep, inode);
 107                 return NULL;
 108         }
 109
 110         return inode;
 111 }
 112
 113 static void fuse_i_callback(struct rcu_head *head)
 114 {
 115         struct inode *inode = container_of(head, struct inode, i_rcu);
 116         kmem_cache_free(fuse_inode_cachep, inode);
 117 }
 118
 119 static void fuse_destroy_inode(struct inode *inode)
 120 {
 121         struct fuse_inode *fi = get_fuse_inode(inode);
 122         BUG_ON(!list_empty(&fi->write_files));
 123         BUG_ON(!list_empty(&fi->queued_writes));
 124         mutex_destroy(&fi->mutex);
 125         kfree(fi->forget);
 126         call_rcu(&inode->i_rcu, fuse_i_callback);
 127 }
 128
 129 static void fuse_evict_inode(struct inode *inode)
 130 {
 131         truncate_inode_pages_final(&inode->i_data);
 132         clear_inode(inode);
 133         if (inode->i_sb->s_flags & SB_ACTIVE) {
 134                 struct fuse_conn *fc = get_fuse_conn(inode);
 135                 struct fuse_inode *fi = get_fuse_inode(inode);
 136                 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
 137                 fi->forget = NULL;
 138         }
 139 }
 140
 141 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
 142 {
 143         sync_filesystem(sb);
 144         if (*flags & SB_MANDLOCK)
 145                 return -EINVAL;
 146
 147         return 0;
 148 }
 149
 150 /*
 151  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
 152  * so that it will fit.
 153  */
 154 static ino_t fuse_squash_ino(u64 ino64)
 155 {
 156         ino_t ino = (ino_t) ino64;
 157         if (sizeof(ino_t) < sizeof(u64))
 158                 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
 159         return ino;
 160 }
 161
 162 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
 163                                    u64 attr_valid)
 164 {
 165         struct fuse_conn *fc = get_fuse_conn(inode);
 166         struct fuse_inode *fi = get_fuse_inode(inode);
 167
 168         fi->attr_version = ++fc->attr_version;
 169         fi->i_time = attr_valid;
 170
 171         inode->i_ino     = fuse_squash_ino(attr->ino);
 172         inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
 173         set_nlink(inode, attr->nlink);
 174         inode->i_uid     = make_kuid(fc->user_ns, attr->uid);
 175         inode->i_gid     = make_kgid(fc->user_ns, attr->gid);
 176         inode->i_blocks  = attr->blocks;
 177         inode->i_atime.tv_sec   = attr->atime;
 178         inode->i_atime.tv_nsec  = attr->atimensec;
 179         /* mtime from server may be stale due to local buffered write */
 180         if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
 181                 inode->i_mtime.tv_sec   = attr->mtime;
 182                 inode->i_mtime.tv_nsec  = attr->mtimensec;
 183                 inode->i_ctime.tv_sec   = attr->ctime;
 184                 inode->i_ctime.tv_nsec  = attr->ctimensec;
 185         }
 186
 187         if (attr->blksize != 0)
 188                 inode->i_blkbits = ilog2(attr->blksize);
 189         else
 190                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
 191
 192         /*
 193          * Don't set the sticky bit in i_mode, unless we want the VFS
 194          * to check permissions.  This prevents failures due to the
 195          * check in may_delete().
 196          */
 197         fi->orig_i_mode = inode->i_mode;
 198         if (!fc->default_permissions)
 199                 inode->i_mode &= ~S_ISVTX;
 200
 201         fi->orig_ino = attr->ino;
 202 }
 203
 204 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
 205                             u64 attr_valid, u64 attr_version)
 206 {
 207         struct fuse_conn *fc = get_fuse_conn(inode);
 208         struct fuse_inode *fi = get_fuse_inode(inode);
 209         bool is_wb = fc->writeback_cache;
 210         loff_t oldsize;
 211         struct timespec64 old_mtime;
 212
 213         spin_lock(&fc->lock);
 214         if ((attr_version != 0 && fi->attr_version > attr_version) ||
 215             test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
 216                 spin_unlock(&fc->lock);
 217                 return;
 218         }
 219
 220         old_mtime = inode->i_mtime;
 221         fuse_change_attributes_common(inode, attr, attr_valid);
 222
 223         oldsize = inode->i_size;
 224         /*
 225          * In case of writeback_cache enabled, the cached writes beyond EOF
 226          * extend local i_size without keeping userspace server in sync. So,
 227          * attr->size coming from server can be stale. We cannot trust it.
 228          */
 229         if (!is_wb || !S_ISREG(inode->i_mode))
 230                 i_size_write(inode, attr->size);
 231         spin_unlock(&fc->lock);
 232
 233         if (!is_wb && S_ISREG(inode->i_mode)) {
 234                 bool inval = false;
 235
 236                 if (oldsize != attr->size) {
 237                         truncate_pagecache(inode, attr->size);
 238                         inval = true;
 239                 } else if (fc->auto_inval_data) {
 240                         struct timespec64 new_mtime = {
 241                                 .tv_sec = attr->mtime,
 242                                 .tv_nsec = attr->mtimensec,
 243                         };
 244
 245                         /*
 246                          * Auto inval mode also checks and invalidates if mtime
 247                          * has changed.
 248                          */
 249                         if (!timespec64_equal(&old_mtime, &new_mtime))
 250                                 inval = true;
 251                 }
 252
 253                 if (inval)
 254                         invalidate_inode_pages2(inode->i_mapping);
 255         }
 256 }
 257
 258 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
 259 {
 260         inode->i_mode = attr->mode & S_IFMT;
 261         inode->i_size = attr->size;
 262         inode->i_mtime.tv_sec  = attr->mtime;
 263         inode->i_mtime.tv_nsec = attr->mtimensec;
 264         inode->i_ctime.tv_sec  = attr->ctime;
 265         inode->i_ctime.tv_nsec = attr->ctimensec;
 266         if (S_ISREG(inode->i_mode)) {
 267                 fuse_init_common(inode);
 268                 fuse_init_file_inode(inode);
 269         } else if (S_ISDIR(inode->i_mode))
 270                 fuse_init_dir(inode);
 271         else if (S_ISLNK(inode->i_mode))
 272                 fuse_init_symlink(inode);
 273         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
 274                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
 275                 fuse_init_common(inode);
 276                 init_special_inode(inode, inode->i_mode,
 277                                    new_decode_dev(attr->rdev));
 278         } else
 279                 BUG();
 280 }
 281
 282 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
 283 {
 284         u64 nodeid = *(u64 *) _nodeidp;
 285         if (get_node_id(inode) == nodeid)
 286                 return 1;
 287         else
 288                 return 0;
 289 }
 290
 291 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
 292 {
 293         u64 nodeid = *(u64 *) _nodeidp;
 294         get_fuse_inode(inode)->nodeid = nodeid;
 295         return 0;
 296 }
 297
 298 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
 299                         int generation, struct fuse_attr *attr,
 300                         u64 attr_valid, u64 attr_version)
 301 {
 302         struct inode *inode;
 303         struct fuse_inode *fi;
 304         struct fuse_conn *fc = get_fuse_conn_super(sb);
 305
 306  retry:
 307         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
 308         if (!inode)
 309                 return NULL;
 310
 311         if ((inode->i_state & I_NEW)) {
 312                 inode->i_flags |= S_NOATIME;
 313                 if (!fc->writeback_cache || !S_ISREG(attr->mode))
 314                         inode->i_flags |= S_NOCMTIME;
 315                 inode->i_generation = generation;
 316                 fuse_init_inode(inode, attr);
 317                 unlock_new_inode(inode);
 318         } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
 319                 /* Inode has changed type, any I/O on the old should fail */
 320                 make_bad_inode(inode);
 321                 iput(inode);
 322                 goto retry;
 323         }
 324
 325         fi = get_fuse_inode(inode);
 326         spin_lock(&fc->lock);
 327         fi->nlookup++;
 328         spin_unlock(&fc->lock);
 329         fuse_change_attributes(inode, attr, attr_valid, attr_version);
 330
 331         return inode;
 332 }
 333
 334 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
 335                              loff_t offset, loff_t len)
 336 {
 337         struct inode *inode;
 338         pgoff_t pg_start;
 339         pgoff_t pg_end;
 340
 341         inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
 342         if (!inode)
 343                 return -ENOENT;
 344
 345         fuse_invalidate_attr(inode);
 346         forget_all_cached_acls(inode);
 347         if (offset >= 0) {
 348                 pg_start = offset >> PAGE_SHIFT;
 349                 if (len <= 0)
 350                         pg_end = -1;
 351                 else
 352                         pg_end = (offset + len - 1) >> PAGE_SHIFT;
 353                 invalidate_inode_pages2_range(inode->i_mapping,
 354                                               pg_start, pg_end);
 355         }
 356         iput(inode);
 357         return 0;
 358 }
 359
 360 bool fuse_lock_inode(struct inode *inode)
 361 {
 362         bool locked = false;
 363
 364         if (!get_fuse_conn(inode)->parallel_dirops) {
 365                 mutex_lock(&get_fuse_inode(inode)->mutex);
 366                 locked = true;
 367         }
 368
 369         return locked;
 370 }
 371
 372 void fuse_unlock_inode(struct inode *inode, bool locked)
 373 {
 374         if (locked)
 375                 mutex_unlock(&get_fuse_inode(inode)->mutex);
 376 }
 377
 378 static void fuse_umount_begin(struct super_block *sb)
 379 {
 380         fuse_abort_conn(get_fuse_conn_super(sb), false);
 381 }
 382
 383 static void fuse_send_destroy(struct fuse_conn *fc)
 384 {
 385         struct fuse_req *req = fc->destroy_req;
 386         if (req && fc->conn_init) {
 387                 fc->destroy_req = NULL;
 388                 req->in.h.opcode = FUSE_DESTROY;
 389                 __set_bit(FR_FORCE, &req->flags);
 390                 __clear_bit(FR_BACKGROUND, &req->flags);
 391                 fuse_request_send(fc, req);
 392                 fuse_put_request(fc, req);
 393         }
 394 }
 395
 396 static void fuse_put_super(struct super_block *sb)
 397 {
 398         struct fuse_conn *fc = get_fuse_conn_super(sb);
 399
 400         mutex_lock(&fuse_mutex);
 401         list_del(&fc->entry);
 402         fuse_ctl_remove_conn(fc);
 403         mutex_unlock(&fuse_mutex);
 404
 405         fuse_conn_put(fc);
 406 }
 407
 408 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
 409 {
 410         stbuf->f_type    = FUSE_SUPER_MAGIC;
 411         stbuf->f_bsize   = attr->bsize;
 412         stbuf->f_frsize  = attr->frsize;
 413         stbuf->f_blocks  = attr->blocks;
 414         stbuf->f_bfree   = attr->bfree;
 415         stbuf->f_bavail  = attr->bavail;
 416         stbuf->f_files   = attr->files;
 417         stbuf->f_ffree   = attr->ffree;
 418         stbuf->f_namelen = attr->namelen;
 419         /* fsid is left zero */
 420 }
 421
 422 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
 423 {
 424         struct super_block *sb = dentry->d_sb;
 425         struct fuse_conn *fc = get_fuse_conn_super(sb);
 426         FUSE_ARGS(args);
 427         struct fuse_statfs_out outarg;
 428         int err;
 429
 430         if (!fuse_allow_current_process(fc)) {
 431                 buf->f_type = FUSE_SUPER_MAGIC;
 432                 return 0;
 433         }
 434
 435         memset(&outarg, 0, sizeof(outarg));
 436         args.in.numargs = 0;
 437         args.in.h.opcode = FUSE_STATFS;
 438         args.in.h.nodeid = get_node_id(d_inode(dentry));
 439         args.out.numargs = 1;
 440         args.out.args[0].size = sizeof(outarg);
 441         args.out.args[0].value = &outarg;
 442         err = fuse_simple_request(fc, &args);
 443         if (!err)
 444                 convert_fuse_statfs(buf, &outarg.st);
 445         return err;
 446 }
 447
 448 enum {
 449         OPT_FD,
 450         OPT_ROOTMODE,
 451         OPT_USER_ID,
 452         OPT_GROUP_ID,
 453         OPT_DEFAULT_PERMISSIONS,
 454         OPT_ALLOW_OTHER,
 455         OPT_MAX_READ,
 456         OPT_BLKSIZE,
 457         OPT_ERR
 458 };
 459
 460 static const match_table_t tokens = {
 461         {OPT_FD,                        "fd=%u"},
 462         {OPT_ROOTMODE,                  "rootmode=%o"},
 463         {OPT_USER_ID,                   "user_id=%u"},
 464         {OPT_GROUP_ID,                  "group_id=%u"},
 465         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 466         {OPT_ALLOW_OTHER,               "allow_other"},
 467         {OPT_MAX_READ,                  "max_read=%u"},
 468         {OPT_BLKSIZE,                   "blksize=%u"},
 469         {OPT_ERR,                       NULL}
 470 };
 471
 472 static int fuse_match_uint(substring_t *s, unsigned int *res)
 473 {
 474         int err = -ENOMEM;
 475         char *buf = match_strdup(s);
 476         if (buf) {
 477                 err = kstrtouint(buf, 10, res);
 478                 kfree(buf);
 479         }
 480         return err;
 481 }
 482
 483 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev,
 484                           struct user_namespace *user_ns)
 485 {
 486         char *p;
 487         memset(d, 0, sizeof(struct fuse_mount_data));
 488         d->max_read = ~0;
 489         d->blksize = FUSE_DEFAULT_BLKSIZE;
 490
 491         while ((p = strsep(&opt, ",")) != NULL) {
 492                 int token;
 493                 int value;
 494                 unsigned uv;
 495                 substring_t args[MAX_OPT_ARGS];
 496                 if (!*p)
 497                         continue;
 498
 499                 token = match_token(p, tokens, args);
 500                 switch (token) {
 501                 case OPT_FD:
 502                         if (match_int(&args[0], &value))
 503                                 return 0;
 504                         d->fd = value;
 505                         d->fd_present = 1;
 506                         break;
 507
 508                 case OPT_ROOTMODE:
 509                         if (match_octal(&args[0], &value))
 510                                 return 0;
 511                         if (!fuse_valid_type(value))
 512                                 return 0;
 513                         d->rootmode = value;
 514                         d->rootmode_present = 1;
 515                         break;
 516
 517                 case OPT_USER_ID:
 518                         if (fuse_match_uint(&args[0], &uv))
 519                                 return 0;
 520                         d->user_id = make_kuid(user_ns, uv);
 521                         if (!uid_valid(d->user_id))
 522                                 return 0;
 523                         d->user_id_present = 1;
 524                         break;
 525
 526                 case OPT_GROUP_ID:
 527                         if (fuse_match_uint(&args[0], &uv))
 528                                 return 0;
 529                         d->group_id = make_kgid(user_ns, uv);
 530                         if (!gid_valid(d->group_id))
 531                                 return 0;
 532                         d->group_id_present = 1;
 533                         break;
 534
 535                 case OPT_DEFAULT_PERMISSIONS:
 536                         d->default_permissions = 1;
 537                         break;
 538
 539                 case OPT_ALLOW_OTHER:
 540                         d->allow_other = 1;
 541                         break;
 542
 543                 case OPT_MAX_READ:
 544                         if (match_int(&args[0], &value))
 545                                 return 0;
 546                         d->max_read = value;
 547                         break;
 548
 549                 case OPT_BLKSIZE:
 550                         if (!is_bdev || match_int(&args[0], &value))
 551                                 return 0;
 552                         d->blksize = value;
 553                         break;
 554
 555                 default:
 556                         return 0;
 557                 }
 558         }
 559
 560         if (!d->fd_present || !d->rootmode_present ||
 561             !d->user_id_present || !d->group_id_present)
 562                 return 0;
 563
 564         return 1;
 565 }
 566
 567 static int fuse_show_options(struct seq_file *m, struct dentry *root)
 568 {
 569         struct super_block *sb = root->d_sb;
 570         struct fuse_conn *fc = get_fuse_conn_super(sb);
 571
 572         seq_printf(m, ",user_id=%u", from_kuid_munged(fc->user_ns, fc->user_id));
 573         seq_printf(m, ",group_id=%u", from_kgid_munged(fc->user_ns, fc->group_id));
 574         if (fc->default_permissions)
 575                 seq_puts(m, ",default_permissions");
 576         if (fc->allow_other)
 577                 seq_puts(m, ",allow_other");
 578         if (fc->max_read != ~0)
 579                 seq_printf(m, ",max_read=%u", fc->max_read);
 580         if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
 581                 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
 582         return 0;
 583 }
 584
 585 static void fuse_iqueue_init(struct fuse_iqueue *fiq)
 586 {
 587         memset(fiq, 0, sizeof(struct fuse_iqueue));
 588         spin_lock_init(&fiq->lock);
 589         init_waitqueue_head(&fiq->waitq);
 590         INIT_LIST_HEAD(&fiq->pending);
 591         INIT_LIST_HEAD(&fiq->interrupts);
 592         fiq->forget_list_tail = &fiq->forget_list_head;
 593         fiq->connected = 1;
 594 }
 595
 596 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
 597 {
 598         memset(fpq, 0, sizeof(struct fuse_pqueue));
 599         spin_lock_init(&fpq->lock);
 600         INIT_LIST_HEAD(&fpq->processing);
 601         INIT_LIST_HEAD(&fpq->io);
 602         fpq->connected = 1;
 603 }
 604
 605 void fuse_conn_init(struct fuse_conn *fc, struct user_namespace *user_ns)
 606 {
 607         memset(fc, 0, sizeof(*fc));
 608         spin_lock_init(&fc->lock);
 609         init_rwsem(&fc->killsb);
 610         refcount_set(&fc->count, 1);
 611         atomic_set(&fc->dev_count, 1);
 612         init_waitqueue_head(&fc->blocked_waitq);
 613         init_waitqueue_head(&fc->reserved_req_waitq);
 614         fuse_iqueue_init(&fc->iq);
 615         INIT_LIST_HEAD(&fc->bg_queue);
 616         INIT_LIST_HEAD(&fc->entry);
 617         INIT_LIST_HEAD(&fc->devices);
 618         atomic_set(&fc->num_waiting, 0);
 619         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
 620         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
 621         fc->khctr = 0;
 622         fc->polled_files = RB_ROOT;
 623         fc->blocked = 0;
 624         fc->initialized = 0;
 625         fc->connected = 1;
 626         fc->attr_version = 1;
 627         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
 628         fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
 629         fc->user_ns = get_user_ns(user_ns);
 630 }
 631 EXPORT_SYMBOL_GPL(fuse_conn_init);
 632
 633 void fuse_conn_put(struct fuse_conn *fc)
 634 {
 635         if (refcount_dec_and_test(&fc->count)) {
 636                 if (fc->destroy_req)
 637                         fuse_request_free(fc->destroy_req);
 638                 put_pid_ns(fc->pid_ns);
 639                 put_user_ns(fc->user_ns);
 640                 fc->release(fc);
 641         }
 642 }
 643 EXPORT_SYMBOL_GPL(fuse_conn_put);
 644
 645 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
 646 {
 647         refcount_inc(&fc->count);
 648         return fc;
 649 }
 650 EXPORT_SYMBOL_GPL(fuse_conn_get);
 651
 652 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
 653 {
 654         struct fuse_attr attr;
 655         memset(&attr, 0, sizeof(attr));
 656
 657         attr.mode = mode;
 658         attr.ino = FUSE_ROOT_ID;
 659         attr.nlink = 1;
 660         return fuse_iget(sb, 1, 0, &attr, 0, 0);
 661 }
 662
 663 struct fuse_inode_handle {
 664         u64 nodeid;
 665         u32 generation;
 666 };
 667
 668 static struct dentry *fuse_get_dentry(struct super_block *sb,
 669                                       struct fuse_inode_handle *handle)
 670 {
 671         struct fuse_conn *fc = get_fuse_conn_super(sb);
 672         struct inode *inode;
 673         struct dentry *entry;
 674         int err = -ESTALE;
 675
 676         if (handle->nodeid == 0)
 677                 goto out_err;
 678
 679         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
 680         if (!inode) {
 681                 struct fuse_entry_out outarg;
 682                 const struct qstr name = QSTR_INIT(".", 1);
 683
 684                 if (!fc->export_support)
 685                         goto out_err;
 686
 687                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
 688                                        &inode);
 689                 if (err && err != -ENOENT)
 690                         goto out_err;
 691                 if (err || !inode) {
 692                         err = -ESTALE;
 693                         goto out_err;
 694                 }
 695                 err = -EIO;
 696                 if (get_node_id(inode) != handle->nodeid)
 697                         goto out_iput;
 698         }
 699         err = -ESTALE;
 700         if (inode->i_generation != handle->generation)
 701                 goto out_iput;
 702
 703         entry = d_obtain_alias(inode);
 704         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
 705                 fuse_invalidate_entry_cache(entry);
 706
 707         return entry;
 708
 709  out_iput:
 710         iput(inode);
 711  out_err:
 712         return ERR_PTR(err);
 713 }
 714
 715 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
 716                            struct inode *parent)
 717 {
 718         int len = parent ? 6 : 3;
 719         u64 nodeid;
 720         u32 generation;
 721
 722         if (*max_len < len) {
 723                 *max_len = len;
 724                 return  FILEID_INVALID;
 725         }
 726
 727         nodeid = get_fuse_inode(inode)->nodeid;
 728         generation = inode->i_generation;
 729
 730         fh[0] = (u32)(nodeid >> 32);
 731         fh[1] = (u32)(nodeid & 0xffffffff);
 732         fh[2] = generation;
 733
 734         if (parent) {
 735                 nodeid = get_fuse_inode(parent)->nodeid;
 736                 generation = parent->i_generation;
 737
 738                 fh[3] = (u32)(nodeid >> 32);
 739                 fh[4] = (u32)(nodeid & 0xffffffff);
 740                 fh[5] = generation;
 741         }
 742
 743         *max_len = len;
 744         return parent ? 0x82 : 0x81;
 745 }
 746
 747 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
 748                 struct fid *fid, int fh_len, int fh_type)
 749 {
 750         struct fuse_inode_handle handle;
 751
 752         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
 753                 return NULL;
 754
 755         handle.nodeid = (u64) fid->raw[0] << 32;
 756         handle.nodeid |= (u64) fid->raw[1];
 757         handle.generation = fid->raw[2];
 758         return fuse_get_dentry(sb, &handle);
 759 }
 760
 761 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
 762                 struct fid *fid, int fh_len, int fh_type)
 763 {
 764         struct fuse_inode_handle parent;
 765
 766         if (fh_type != 0x82 || fh_len < 6)
 767                 return NULL;
 768
 769         parent.nodeid = (u64) fid->raw[3] << 32;
 770         parent.nodeid |= (u64) fid->raw[4];
 771         parent.generation = fid->raw[5];
 772         return fuse_get_dentry(sb, &parent);
 773 }
 774
 775 static struct dentry *fuse_get_parent(struct dentry *child)
 776 {
 777         struct inode *child_inode = d_inode(child);
 778         struct fuse_conn *fc = get_fuse_conn(child_inode);
 779         struct inode *inode;
 780         struct dentry *parent;
 781         struct fuse_entry_out outarg;
 782         const struct qstr name = QSTR_INIT("..", 2);
 783         int err;
 784
 785         if (!fc->export_support)
 786                 return ERR_PTR(-ESTALE);
 787
 788         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
 789                                &name, &outarg, &inode);
 790         if (err) {
 791                 if (err == -ENOENT)
 792                         return ERR_PTR(-ESTALE);
 793                 return ERR_PTR(err);
 794         }
 795
 796         parent = d_obtain_alias(inode);
 797         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
 798                 fuse_invalidate_entry_cache(parent);
 799
 800         return parent;
 801 }
 802
 803 static const struct export_operations fuse_export_operations = {
 804         .fh_to_dentry   = fuse_fh_to_dentry,
 805         .fh_to_parent   = fuse_fh_to_parent,
 806         .encode_fh      = fuse_encode_fh,
 807         .get_parent     = fuse_get_parent,
 808 };
 809
 810 static const struct super_operations fuse_super_operations = {
 811         .alloc_inode    = fuse_alloc_inode,
 812         .destroy_inode  = fuse_destroy_inode,
 813         .evict_inode    = fuse_evict_inode,
 814         .write_inode    = fuse_write_inode,
 815         .drop_inode     = generic_delete_inode,
 816         .remount_fs     = fuse_remount_fs,
 817         .put_super      = fuse_put_super,
 818         .umount_begin   = fuse_umount_begin,
 819         .statfs         = fuse_statfs,
 820         .show_options   = fuse_show_options,
 821 };
 822
 823 static void sanitize_global_limit(unsigned *limit)
 824 {
 825         if (*limit == 0)
 826                 *limit = ((totalram_pages << PAGE_SHIFT) >> 13) /
 827                          sizeof(struct fuse_req);
 828
 829         if (*limit >= 1 << 16)
 830                 *limit = (1 << 16) - 1;
 831 }
 832
 833 static int set_global_limit(const char *val, const struct kernel_param *kp)
 834 {
 835         int rv;
 836
 837         rv = param_set_uint(val, kp);
 838         if (rv)
 839                 return rv;
 840
 841         sanitize_global_limit((unsigned *)kp->arg);
 842
 843         return 0;
 844 }
 845
 846 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
 847 {
 848         int cap_sys_admin = capable(CAP_SYS_ADMIN);
 849
 850         if (arg->minor < 13)
 851                 return;
 852
 853         sanitize_global_limit(&max_user_bgreq);
 854         sanitize_global_limit(&max_user_congthresh);
 855
 856         if (arg->max_background) {
 857                 fc->max_background = arg->max_background;
 858
 859                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
 860                         fc->max_background = max_user_bgreq;
 861         }
 862         if (arg->congestion_threshold) {
 863                 fc->congestion_threshold = arg->congestion_threshold;
 864
 865                 if (!cap_sys_admin &&
 866                     fc->congestion_threshold > max_user_congthresh)
 867                         fc->congestion_threshold = max_user_congthresh;
 868         }
 869 }
 870
 871 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
 872 {
 873         struct fuse_init_out *arg = &req->misc.init_out;
 874
 875         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
 876                 fc->conn_error = 1;
 877         else {
 878                 unsigned long ra_pages;
 879
 880                 process_init_limits(fc, arg);
 881
 882                 if (arg->minor >= 6) {
 883                         ra_pages = arg->max_readahead / PAGE_SIZE;
 884                         if (arg->flags & FUSE_ASYNC_READ)
 885                                 fc->async_read = 1;
 886                         if (!(arg->flags & FUSE_POSIX_LOCKS))
 887                                 fc->no_lock = 1;
 888                         if (arg->minor >= 17) {
 889                                 if (!(arg->flags & FUSE_FLOCK_LOCKS))
 890                                         fc->no_flock = 1;
 891                         } else {
 892                                 if (!(arg->flags & FUSE_POSIX_LOCKS))
 893                                         fc->no_flock = 1;
 894                         }
 895                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
 896                                 fc->atomic_o_trunc = 1;
 897                         if (arg->minor >= 9) {
 898                                 /* LOOKUP has dependency on proto version */
 899                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
 900                                         fc->export_support = 1;
 901                         }
 902                         if (arg->flags & FUSE_BIG_WRITES)
 903                                 fc->big_writes = 1;
 904                         if (arg->flags & FUSE_DONT_MASK)
 905                                 fc->dont_mask = 1;
 906                         if (arg->flags & FUSE_AUTO_INVAL_DATA)
 907                                 fc->auto_inval_data = 1;
 908                         if (arg->flags & FUSE_DO_READDIRPLUS) {
 909                                 fc->do_readdirplus = 1;
 910                                 if (arg->flags & FUSE_READDIRPLUS_AUTO)
 911                                         fc->readdirplus_auto = 1;
 912                         }
 913                         if (arg->flags & FUSE_ASYNC_DIO)
 914                                 fc->async_dio = 1;
 915                         if (arg->flags & FUSE_WRITEBACK_CACHE)
 916                                 fc->writeback_cache = 1;
 917                         if (arg->flags & FUSE_PARALLEL_DIROPS)
 918                                 fc->parallel_dirops = 1;
 919                         if (arg->flags & FUSE_HANDLE_KILLPRIV)
 920                                 fc->handle_killpriv = 1;
 921                         if (arg->time_gran && arg->time_gran <= 1000000000)
 922                                 fc->sb->s_time_gran = arg->time_gran;
 923                         if ((arg->flags & FUSE_POSIX_ACL)) {
 924                                 fc->default_permissions = 1;
 925                                 fc->posix_acl = 1;
 926                                 fc->sb->s_xattr = fuse_acl_xattr_handlers;
 927                         }
 928                         if (arg->flags & FUSE_ABORT_ERROR)
 929                                 fc->abort_err = 1;
 930                 } else {
 931                         ra_pages = fc->max_read / PAGE_SIZE;
 932                         fc->no_lock = 1;
 933                         fc->no_flock = 1;
 934                 }
 935
 936                 fc->sb->s_bdi->ra_pages =
 937                                 min(fc->sb->s_bdi->ra_pages, ra_pages);
 938                 fc->minor = arg->minor;
 939                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
 940                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
 941                 fc->conn_init = 1;
 942         }
 943         fuse_set_initialized(fc);
 944         wake_up_all(&fc->blocked_waitq);
 945 }
 946
 947 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
 948 {
 949         struct fuse_init_in *arg = &req->misc.init_in;
 950
 951         arg->major = FUSE_KERNEL_VERSION;
 952         arg->minor = FUSE_KERNEL_MINOR_VERSION;
 953         arg->max_readahead = fc->sb->s_bdi->ra_pages * PAGE_SIZE;
 954         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
 955                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
 956                 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
 957                 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
 958                 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
 959                 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
 960                 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
 961                 FUSE_ABORT_ERROR;
 962         req->in.h.opcode = FUSE_INIT;
 963         req->in.numargs = 1;
 964         req->in.args[0].size = sizeof(*arg);
 965         req->in.args[0].value = arg;
 966         req->out.numargs = 1;
 967         /* Variable length argument used for backward compatibility
 968            with interface version < 7.5.  Rest of init_out is zeroed
 969            by do_get_request(), so a short reply is not a problem */
 970         req->out.argvar = 1;
 971         req->out.args[0].size = sizeof(struct fuse_init_out);
 972         req->out.args[0].value = &req->misc.init_out;
 973         req->end = process_init_reply;
 974         fuse_request_send_background(fc, req);
 975 }
 976
 977 static void fuse_free_conn(struct fuse_conn *fc)
 978 {
 979         WARN_ON(!list_empty(&fc->devices));
 980         kfree_rcu(fc, rcu);
 981 }
 982
 983 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
 984 {
 985         int err;
 986         char *suffix = "";
 987
 988         if (sb->s_bdev) {
 989                 suffix = "-fuseblk";
 990                 /*
 991                  * sb->s_bdi points to blkdev's bdi however we want to redirect
 992                  * it to our private bdi...
 993                  */
 994                 bdi_put(sb->s_bdi);
 995                 sb->s_bdi = &noop_backing_dev_info;
 996         }
 997         err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
 998                                    MINOR(fc->dev), suffix);
 999         if (err)
1000                 return err;
1001
1002         sb->s_bdi->ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
1003         /* fuse does it's own writeback accounting */
1004         sb->s_bdi->capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
1005
1006         /*
1007          * For a single fuse filesystem use max 1% of dirty +
1008          * writeback threshold.
1009          *
1010          * This gives about 1M of write buffer for memory maps on a
1011          * machine with 1G and 10% dirty_ratio, which should be more
1012          * than enough.
1013          *
1014          * Privileged users can raise it by writing to
1015          *
1016          *    /sys/class/bdi/<bdi>/max_ratio
1017          */
1018         bdi_set_max_ratio(sb->s_bdi, 1);
1019
1020         return 0;
1021 }
1022
1023 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc)
1024 {
1025         struct fuse_dev *fud;
1026
1027         fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1028         if (fud) {
1029                 fud->fc = fuse_conn_get(fc);
1030                 fuse_pqueue_init(&fud->pq);
1031
1032                 spin_lock(&fc->lock);
1033                 list_add_tail(&fud->entry, &fc->devices);
1034                 spin_unlock(&fc->lock);
1035         }
1036
1037         return fud;
1038 }
1039 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1040
1041 void fuse_dev_free(struct fuse_dev *fud)
1042 {
1043         struct fuse_conn *fc = fud->fc;
1044
1045         if (fc) {
1046                 spin_lock(&fc->lock);
1047                 list_del(&fud->entry);
1048                 spin_unlock(&fc->lock);
1049
1050                 fuse_conn_put(fc);
1051         }
1052         kfree(fud);
1053 }
1054 EXPORT_SYMBOL_GPL(fuse_dev_free);
1055
1056 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
1057 {
1058         struct fuse_dev *fud;
1059         struct fuse_conn *fc;
1060         struct inode *root;
1061         struct fuse_mount_data d;
1062         struct file *file;
1063         struct dentry *root_dentry;
1064         struct fuse_req *init_req;
1065         int err;
1066         int is_bdev = sb->s_bdev != NULL;
1067
1068         err = -EINVAL;
1069         if (sb->s_flags & SB_MANDLOCK)
1070                 goto err;
1071
1072         sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1073
1074         if (!parse_fuse_opt(data, &d, is_bdev, sb->s_user_ns))
1075                 goto err;
1076
1077         if (is_bdev) {
1078 #ifdef CONFIG_BLOCK
1079                 err = -EINVAL;
1080                 if (!sb_set_blocksize(sb, d.blksize))
1081                         goto err;
1082 #endif
1083         } else {
1084                 sb->s_blocksize = PAGE_SIZE;
1085                 sb->s_blocksize_bits = PAGE_SHIFT;
1086         }
1087         sb->s_magic = FUSE_SUPER_MAGIC;
1088         sb->s_op = &fuse_super_operations;
1089         sb->s_xattr = fuse_xattr_handlers;
1090         sb->s_maxbytes = MAX_LFS_FILESIZE;
1091         sb->s_time_gran = 1;
1092         sb->s_export_op = &fuse_export_operations;
1093         sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1094         if (sb->s_user_ns != &init_user_ns)
1095                 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1096
1097         file = fget(d.fd);
1098         err = -EINVAL;
1099         if (!file)
1100                 goto err;
1101
1102         /*
1103          * Require mount to happen from the same user namespace which
1104          * opened /dev/fuse to prevent potential attacks.
1105          */
1106         if (file->f_op != &fuse_dev_operations ||
1107             file->f_cred->user_ns != sb->s_user_ns)
1108                 goto err_fput;
1109
1110         /*
1111          * If we are not in the initial user namespace posix
1112          * acls must be translated.
1113          */
1114         if (sb->s_user_ns != &init_user_ns)
1115                 sb->s_xattr = fuse_no_acl_xattr_handlers;
1116
1117         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1118         err = -ENOMEM;
1119         if (!fc)
1120                 goto err_fput;
1121
1122         fuse_conn_init(fc, sb->s_user_ns);
1123         fc->release = fuse_free_conn;
1124
1125         fud = fuse_dev_alloc(fc);
1126         if (!fud)
1127                 goto err_put_conn;
1128
1129         fc->dev = sb->s_dev;
1130         fc->sb = sb;
1131         err = fuse_bdi_init(fc, sb);
1132         if (err)
1133                 goto err_dev_free;
1134
1135         /* Handle umasking inside the fuse code */
1136         if (sb->s_flags & SB_POSIXACL)
1137                 fc->dont_mask = 1;
1138         sb->s_flags |= SB_POSIXACL;
1139
1140         fc->default_permissions = d.default_permissions;
1141         fc->allow_other = d.allow_other;
1142         fc->user_id = d.user_id;
1143         fc->group_id = d.group_id;
1144         fc->max_read = max_t(unsigned, 4096, d.max_read);
1145
1146         /* Used by get_root_inode() */
1147         sb->s_fs_info = fc;
1148
1149         err = -ENOMEM;
1150         root = fuse_get_root_inode(sb, d.rootmode);
1151         sb->s_d_op = &fuse_root_dentry_operations;
1152         root_dentry = d_make_root(root);
1153         if (!root_dentry)
1154                 goto err_dev_free;
1155         /* Root dentry doesn't have .d_revalidate */
1156         sb->s_d_op = &fuse_dentry_operations;
1157
1158         init_req = fuse_request_alloc(0);
1159         if (!init_req)
1160                 goto err_put_root;
1161         __set_bit(FR_BACKGROUND, &init_req->flags);
1162
1163         if (is_bdev) {
1164                 fc->destroy_req = fuse_request_alloc(0);
1165                 if (!fc->destroy_req)
1166                         goto err_free_init_req;
1167         }
1168
1169         mutex_lock(&fuse_mutex);
1170         err = -EINVAL;
1171         if (file->private_data)
1172                 goto err_unlock;
1173
1174         err = fuse_ctl_add_conn(fc);
1175         if (err)
1176                 goto err_unlock;
1177
1178         list_add_tail(&fc->entry, &fuse_conn_list);
1179         sb->s_root = root_dentry;
1180         file->private_data = fud;
1181         mutex_unlock(&fuse_mutex);
1182         /*
1183          * atomic_dec_and_test() in fput() provides the necessary
1184          * memory barrier for file->private_data to be visible on all
1185          * CPUs after this
1186          */
1187         fput(file);
1188
1189         fuse_send_init(fc, init_req);
1190
1191         return 0;
1192
1193  err_unlock:
1194         mutex_unlock(&fuse_mutex);
1195  err_free_init_req:
1196         fuse_request_free(init_req);
1197  err_put_root:
1198         dput(root_dentry);
1199  err_dev_free:
1200         fuse_dev_free(fud);
1201  err_put_conn:
1202         fuse_conn_put(fc);
1203         sb->s_fs_info = NULL;
1204  err_fput:
1205         fput(file);
1206  err:
1207         return err;
1208 }
1209
1210 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1211                        int flags, const char *dev_name,
1212                        void *raw_data)
1213 {
1214         return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1215 }
1216
1217 static void fuse_sb_destroy(struct super_block *sb)
1218 {
1219         struct fuse_conn *fc = get_fuse_conn_super(sb);
1220
1221         if (fc) {
1222                 fuse_send_destroy(fc);
1223
1224                 fuse_abort_conn(fc, false);
1225                 fuse_wait_aborted(fc);
1226
1227                 down_write(&fc->killsb);
1228                 fc->sb = NULL;
1229                 up_write(&fc->killsb);
1230         }
1231 }
1232
1233 static void fuse_kill_sb_anon(struct super_block *sb)
1234 {
1235         fuse_sb_destroy(sb);
1236         kill_anon_super(sb);
1237 }
1238
1239 static struct file_system_type fuse_fs_type = {
1240         .owner          = THIS_MODULE,
1241         .name           = "fuse",
1242         .fs_flags       = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1243         .mount          = fuse_mount,
1244         .kill_sb        = fuse_kill_sb_anon,
1245 };
1246 MODULE_ALIAS_FS("fuse");
1247
1248 #ifdef CONFIG_BLOCK
1249 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1250                            int flags, const char *dev_name,
1251                            void *raw_data)
1252 {
1253         return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1254 }
1255
1256 static void fuse_kill_sb_blk(struct super_block *sb)
1257 {
1258         fuse_sb_destroy(sb);
1259         kill_block_super(sb);
1260 }
1261
1262 static struct file_system_type fuseblk_fs_type = {
1263         .owner          = THIS_MODULE,
1264         .name           = "fuseblk",
1265         .mount          = fuse_mount_blk,
1266         .kill_sb        = fuse_kill_sb_blk,
1267         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1268 };
1269 MODULE_ALIAS_FS("fuseblk");
1270
1271 static inline int register_fuseblk(void)
1272 {
1273         return register_filesystem(&fuseblk_fs_type);
1274 }
1275
1276 static inline void unregister_fuseblk(void)
1277 {
1278         unregister_filesystem(&fuseblk_fs_type);
1279 }
1280 #else
1281 static inline int register_fuseblk(void)
1282 {
1283         return 0;
1284 }
1285
1286 static inline void unregister_fuseblk(void)
1287 {
1288 }
1289 #endif
1290
1291 static void fuse_inode_init_once(void *foo)
1292 {
1293         struct inode *inode = foo;
1294
1295         inode_init_once(inode);
1296 }
1297
1298 static int __init fuse_fs_init(void)
1299 {
1300         int err;
1301
1302         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1303                         sizeof(struct fuse_inode), 0,
1304                         SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1305                         fuse_inode_init_once);
1306         err = -ENOMEM;
1307         if (!fuse_inode_cachep)
1308                 goto out;
1309
1310         err = register_fuseblk();
1311         if (err)
1312                 goto out2;
1313
1314         err = register_filesystem(&fuse_fs_type);
1315         if (err)
1316                 goto out3;
1317
1318         return 0;
1319
1320  out3:
1321         unregister_fuseblk();
1322  out2:
1323         kmem_cache_destroy(fuse_inode_cachep);
1324  out:
1325         return err;
1326 }
1327
1328 static void fuse_fs_cleanup(void)
1329 {
1330         unregister_filesystem(&fuse_fs_type);
1331         unregister_fuseblk();
1332
1333         /*
1334          * Make sure all delayed rcu free inodes are flushed before we
1335          * destroy cache.
1336          */
1337         rcu_barrier();
1338         kmem_cache_destroy(fuse_inode_cachep);
1339 }
1340
1341 static struct kobject *fuse_kobj;
1342
1343 static int fuse_sysfs_init(void)
1344 {
1345         int err;
1346
1347         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1348         if (!fuse_kobj) {
1349                 err = -ENOMEM;
1350                 goto out_err;
1351         }
1352
1353         err = sysfs_create_mount_point(fuse_kobj, "connections");
1354         if (err)
1355                 goto out_fuse_unregister;
1356
1357         return 0;
1358
1359  out_fuse_unregister:
1360         kobject_put(fuse_kobj);
1361  out_err:
1362         return err;
1363 }
1364
1365 static void fuse_sysfs_cleanup(void)
1366 {
1367         sysfs_remove_mount_point(fuse_kobj, "connections");
1368         kobject_put(fuse_kobj);
1369 }
1370
1371 static int __init fuse_init(void)
1372 {
1373         int res;
1374
1375         printk(KERN_INFO "fuse init (API version %i.%i)\n",
1376                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1377
1378         INIT_LIST_HEAD(&fuse_conn_list);
1379         res = fuse_fs_init();
1380         if (res)
1381                 goto err;
1382
1383         res = fuse_dev_init();
1384         if (res)
1385                 goto err_fs_cleanup;
1386
1387         res = fuse_sysfs_init();
1388         if (res)
1389                 goto err_dev_cleanup;
1390
1391         res = fuse_ctl_init();
1392         if (res)
1393                 goto err_sysfs_cleanup;
1394
1395         sanitize_global_limit(&max_user_bgreq);
1396         sanitize_global_limit(&max_user_congthresh);
1397
1398         return 0;
1399
1400  err_sysfs_cleanup:
1401         fuse_sysfs_cleanup();
1402  err_dev_cleanup:
1403         fuse_dev_cleanup();
1404  err_fs_cleanup:
1405         fuse_fs_cleanup();
1406  err:
1407         return res;
1408 }
1409
1410 static void __exit fuse_exit(void)
1411 {
1412         printk(KERN_DEBUG "fuse exit\n");
1413
1414         fuse_ctl_cleanup();
1415         fuse_sysfs_cleanup();
1416         fuse_fs_cleanup();
1417         fuse_dev_cleanup();
1418 }
1419
1420 module_init(fuse_init);
1421 module_exit(fuse_exit);