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
  25 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  26 MODULE_DESCRIPTION("Filesystem in Userspace");
  27 MODULE_LICENSE("GPL");
  28
  29 static struct kmem_cache *fuse_inode_cachep;
  30 struct list_head fuse_conn_list;
  31 DEFINE_MUTEX(fuse_mutex);
  32
  33 static int set_global_limit(const char *val, struct kernel_param *kp);
  34
  35 unsigned max_user_bgreq;
  36 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
  37                   &max_user_bgreq, 0644);
  38 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
  39 MODULE_PARM_DESC(max_user_bgreq,
  40  "Global limit for the maximum number of backgrounded requests an "
  41  "unprivileged user can set");
  42
  43 unsigned max_user_congthresh;
  44 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
  45                   &max_user_congthresh, 0644);
  46 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
  47 MODULE_PARM_DESC(max_user_congthresh,
  48  "Global limit for the maximum congestion threshold an "
  49  "unprivileged user can set");
  50
  51 #define FUSE_SUPER_MAGIC 0x65735546
  52
  53 #define FUSE_DEFAULT_BLKSIZE 512
  54
  55 /** Maximum number of outstanding background requests */
  56 #define FUSE_DEFAULT_MAX_BACKGROUND 12
  57
  58 /** Congestion starts at 75% of maximum */
  59 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
  60
  61 struct fuse_mount_data {
  62         int fd;
  63         unsigned rootmode;
  64         kuid_t user_id;
  65         kgid_t group_id;
  66         unsigned fd_present:1;
  67         unsigned rootmode_present:1;
  68         unsigned user_id_present:1;
  69         unsigned group_id_present:1;
  70         unsigned default_permissions:1;
  71         unsigned allow_other:1;
  72         unsigned max_read;
  73         unsigned blksize;
  74 };
  75
  76 struct fuse_forget_link *fuse_alloc_forget(void)
  77 {
  78         return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL);
  79 }
  80
  81 static struct inode *fuse_alloc_inode(struct super_block *sb)
  82 {
  83         struct inode *inode;
  84         struct fuse_inode *fi;
  85
  86         inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
  87         if (!inode)
  88                 return NULL;
  89
  90         fi = get_fuse_inode(inode);
  91         fi->i_time = 0;
  92         fi->nodeid = 0;
  93         fi->nlookup = 0;
  94         fi->attr_version = 0;
  95         fi->writectr = 0;
  96         fi->orig_ino = 0;
  97         fi->state = 0;
  98         INIT_LIST_HEAD(&fi->write_files);
  99         INIT_LIST_HEAD(&fi->queued_writes);
 100         INIT_LIST_HEAD(&fi->writepages);
 101         init_waitqueue_head(&fi->page_waitq);
 102         mutex_init(&fi->mutex);
 103         fi->forget = fuse_alloc_forget();
 104         if (!fi->forget) {
 105                 kmem_cache_free(fuse_inode_cachep, inode);
 106                 return NULL;
 107         }
 108
 109         return inode;
 110 }
 111
 112 static void fuse_i_callback(struct rcu_head *head)
 113 {
 114         struct inode *inode = container_of(head, struct inode, i_rcu);
 115         kmem_cache_free(fuse_inode_cachep, inode);
 116 }
 117
 118 static void fuse_destroy_inode(struct inode *inode)
 119 {
 120         struct fuse_inode *fi = get_fuse_inode(inode);
 121         BUG_ON(!list_empty(&fi->write_files));
 122         BUG_ON(!list_empty(&fi->queued_writes));
 123         mutex_destroy(&fi->mutex);
 124         kfree(fi->forget);
 125         call_rcu(&inode->i_rcu, fuse_i_callback);
 126 }
 127
 128 static void fuse_evict_inode(struct inode *inode)
 129 {
 130         truncate_inode_pages_final(&inode->i_data);
 131         clear_inode(inode);
 132         if (inode->i_sb->s_flags & MS_ACTIVE) {
 133                 struct fuse_conn *fc = get_fuse_conn(inode);
 134                 struct fuse_inode *fi = get_fuse_inode(inode);
 135                 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
 136                 fi->forget = NULL;
 137         }
 138 }
 139
 140 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
 141 {
 142         sync_filesystem(sb);
 143         if (*flags & MS_MANDLOCK)
 144                 return -EINVAL;
 145
 146         return 0;
 147 }
 148
 149 /*
 150  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
 151  * so that it will fit.
 152  */
 153 static ino_t fuse_squash_ino(u64 ino64)
 154 {
 155         ino_t ino = (ino_t) ino64;
 156         if (sizeof(ino_t) < sizeof(u64))
 157                 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
 158         return ino;
 159 }
 160
 161 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
 162                                    u64 attr_valid)
 163 {
 164         struct fuse_conn *fc = get_fuse_conn(inode);
 165         struct fuse_inode *fi = get_fuse_inode(inode);
 166
 167         fi->attr_version = ++fc->attr_version;
 168         fi->i_time = attr_valid;
 169
 170         inode->i_ino     = fuse_squash_ino(attr->ino);
 171         inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
 172         set_nlink(inode, attr->nlink);
 173         inode->i_uid     = make_kuid(&init_user_ns, attr->uid);
 174         inode->i_gid     = make_kgid(&init_user_ns, attr->gid);
 175         inode->i_blocks  = attr->blocks;
 176         inode->i_atime.tv_sec   = attr->atime;
 177         inode->i_atime.tv_nsec  = attr->atimensec;
 178         /* mtime from server may be stale due to local buffered write */
 179         if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
 180                 inode->i_mtime.tv_sec   = attr->mtime;
 181                 inode->i_mtime.tv_nsec  = attr->mtimensec;
 182                 inode->i_ctime.tv_sec   = attr->ctime;
 183                 inode->i_ctime.tv_nsec  = attr->ctimensec;
 184         }
 185
 186         if (attr->blksize != 0)
 187                 inode->i_blkbits = ilog2(attr->blksize);
 188         else
 189                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
 190
 191         /*
 192          * Don't set the sticky bit in i_mode, unless we want the VFS
 193          * to check permissions.  This prevents failures due to the
 194          * check in may_delete().
 195          */
 196         fi->orig_i_mode = inode->i_mode;
 197         if (!fc->default_permissions)
 198                 inode->i_mode &= ~S_ISVTX;
 199
 200         fi->orig_ino = attr->ino;
 201 }
 202
 203 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
 204                             u64 attr_valid, u64 attr_version)
 205 {
 206         struct fuse_conn *fc = get_fuse_conn(inode);
 207         struct fuse_inode *fi = get_fuse_inode(inode);
 208         bool is_wb = fc->writeback_cache;
 209         loff_t oldsize;
 210         struct timespec old_mtime;
 211
 212         spin_lock(&fc->lock);
 213         if ((attr_version != 0 && fi->attr_version > attr_version) ||
 214             test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
 215                 spin_unlock(&fc->lock);
 216                 return;
 217         }
 218
 219         old_mtime = inode->i_mtime;
 220         fuse_change_attributes_common(inode, attr, attr_valid);
 221
 222         oldsize = inode->i_size;
 223         /*
 224          * In case of writeback_cache enabled, the cached writes beyond EOF
 225          * extend local i_size without keeping userspace server in sync. So,
 226          * attr->size coming from server can be stale. We cannot trust it.
 227          */
 228         if (!is_wb || !S_ISREG(inode->i_mode))
 229                 i_size_write(inode, attr->size);
 230         spin_unlock(&fc->lock);
 231
 232         if (!is_wb && S_ISREG(inode->i_mode)) {
 233                 bool inval = false;
 234
 235                 if (oldsize != attr->size) {
 236                         truncate_pagecache(inode, attr->size);
 237                         inval = true;
 238                 } else if (fc->auto_inval_data) {
 239                         struct timespec new_mtime = {
 240                                 .tv_sec = attr->mtime,
 241                                 .tv_nsec = attr->mtimensec,
 242                         };
 243
 244                         /*
 245                          * Auto inval mode also checks and invalidates if mtime
 246                          * has changed.
 247                          */
 248                         if (!timespec_equal(&old_mtime, &new_mtime))
 249                                 inval = true;
 250                 }
 251
 252                 if (inval)
 253                         invalidate_inode_pages2(inode->i_mapping);
 254         }
 255 }
 256
 257 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
 258 {
 259         inode->i_mode = attr->mode & S_IFMT;
 260         inode->i_size = attr->size;
 261         inode->i_mtime.tv_sec  = attr->mtime;
 262         inode->i_mtime.tv_nsec = attr->mtimensec;
 263         inode->i_ctime.tv_sec  = attr->ctime;
 264         inode->i_ctime.tv_nsec = attr->ctimensec;
 265         if (S_ISREG(inode->i_mode)) {
 266                 fuse_init_common(inode);
 267                 fuse_init_file_inode(inode);
 268         } else if (S_ISDIR(inode->i_mode))
 269                 fuse_init_dir(inode);
 270         else if (S_ISLNK(inode->i_mode))
 271                 fuse_init_symlink(inode);
 272         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
 273                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
 274                 fuse_init_common(inode);
 275                 init_special_inode(inode, inode->i_mode,
 276                                    new_decode_dev(attr->rdev));
 277         } else
 278                 BUG();
 279 }
 280
 281 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
 282 {
 283         u64 nodeid = *(u64 *) _nodeidp;
 284         if (get_node_id(inode) == nodeid)
 285                 return 1;
 286         else
 287                 return 0;
 288 }
 289
 290 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
 291 {
 292         u64 nodeid = *(u64 *) _nodeidp;
 293         get_fuse_inode(inode)->nodeid = nodeid;
 294         return 0;
 295 }
 296
 297 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
 298                         int generation, struct fuse_attr *attr,
 299                         u64 attr_valid, u64 attr_version)
 300 {
 301         struct inode *inode;
 302         struct fuse_inode *fi;
 303         struct fuse_conn *fc = get_fuse_conn_super(sb);
 304
 305  retry:
 306         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
 307         if (!inode)
 308                 return NULL;
 309
 310         if ((inode->i_state & I_NEW)) {
 311                 inode->i_flags |= S_NOATIME;
 312                 if (!fc->writeback_cache || !S_ISREG(attr->mode))
 313                         inode->i_flags |= S_NOCMTIME;
 314                 inode->i_generation = generation;
 315                 fuse_init_inode(inode, attr);
 316                 unlock_new_inode(inode);
 317         } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
 318                 /* Inode has changed type, any I/O on the old should fail */
 319                 fuse_make_bad(inode);
 320                 iput(inode);
 321                 goto retry;
 322         }
 323
 324         fi = get_fuse_inode(inode);
 325         spin_lock(&fc->lock);
 326         fi->nlookup++;
 327         spin_unlock(&fc->lock);
 328         fuse_change_attributes(inode, attr, attr_valid, attr_version);
 329
 330         return inode;
 331 }
 332
 333 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
 334                              loff_t offset, loff_t len)
 335 {
 336         struct inode *inode;
 337         pgoff_t pg_start;
 338         pgoff_t pg_end;
 339
 340         inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
 341         if (!inode)
 342                 return -ENOENT;
 343
 344         fuse_invalidate_attr(inode);
 345         forget_all_cached_acls(inode);
 346         if (offset >= 0) {
 347                 pg_start = offset >> PAGE_SHIFT;
 348                 if (len <= 0)
 349                         pg_end = -1;
 350                 else
 351                         pg_end = (offset + len - 1) >> PAGE_SHIFT;
 352                 invalidate_inode_pages2_range(inode->i_mapping,
 353                                               pg_start, pg_end);
 354         }
 355         iput(inode);
 356         return 0;
 357 }
 358
 359 bool fuse_lock_inode(struct inode *inode)
 360 {
 361         bool locked = false;
 362
 363         if (!get_fuse_conn(inode)->parallel_dirops) {
 364                 mutex_lock(&get_fuse_inode(inode)->mutex);
 365                 locked = true;
 366         }
 367
 368         return locked;
 369 }
 370
 371 void fuse_unlock_inode(struct inode *inode, bool locked)
 372 {
 373         if (locked)
 374                 mutex_unlock(&get_fuse_inode(inode)->mutex);
 375 }
 376
 377 static void fuse_umount_begin(struct super_block *sb)
 378 {
 379         fuse_abort_conn(get_fuse_conn_super(sb));
 380 }
 381
 382 static void fuse_send_destroy(struct fuse_conn *fc)
 383 {
 384         struct fuse_req *req = fc->destroy_req;
 385         if (req && fc->conn_init) {
 386                 fc->destroy_req = NULL;
 387                 req->in.h.opcode = FUSE_DESTROY;
 388                 __set_bit(FR_FORCE, &req->flags);
 389                 __clear_bit(FR_BACKGROUND, &req->flags);
 390                 fuse_request_send(fc, req);
 391                 fuse_put_request(fc, req);
 392         }
 393 }
 394
 395 static void fuse_bdi_destroy(struct fuse_conn *fc)
 396 {
 397         if (fc->bdi_initialized)
 398                 bdi_destroy(&fc->bdi);
 399 }
 400
 401 static void fuse_put_super(struct super_block *sb)
 402 {
 403         struct fuse_conn *fc = get_fuse_conn_super(sb);
 404
 405         mutex_lock(&fuse_mutex);
 406         list_del(&fc->entry);
 407         fuse_ctl_remove_conn(fc);
 408         mutex_unlock(&fuse_mutex);
 409         fuse_bdi_destroy(fc);
 410
 411         fuse_conn_put(fc);
 412 }
 413
 414 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
 415 {
 416         stbuf->f_type    = FUSE_SUPER_MAGIC;
 417         stbuf->f_bsize   = attr->bsize;
 418         stbuf->f_frsize  = attr->frsize;
 419         stbuf->f_blocks  = attr->blocks;
 420         stbuf->f_bfree   = attr->bfree;
 421         stbuf->f_bavail  = attr->bavail;
 422         stbuf->f_files   = attr->files;
 423         stbuf->f_ffree   = attr->ffree;
 424         stbuf->f_namelen = attr->namelen;
 425         /* fsid is left zero */
 426 }
 427
 428 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
 429 {
 430         struct super_block *sb = dentry->d_sb;
 431         struct fuse_conn *fc = get_fuse_conn_super(sb);
 432         FUSE_ARGS(args);
 433         struct fuse_statfs_out outarg;
 434         int err;
 435
 436         if (!fuse_allow_current_process(fc)) {
 437                 buf->f_type = FUSE_SUPER_MAGIC;
 438                 return 0;
 439         }
 440
 441         memset(&outarg, 0, sizeof(outarg));
 442         args.in.numargs = 0;
 443         args.in.h.opcode = FUSE_STATFS;
 444         args.in.h.nodeid = get_node_id(d_inode(dentry));
 445         args.out.numargs = 1;
 446         args.out.args[0].size = sizeof(outarg);
 447         args.out.args[0].value = &outarg;
 448         err = fuse_simple_request(fc, &args);
 449         if (!err)
 450                 convert_fuse_statfs(buf, &outarg.st);
 451         return err;
 452 }
 453
 454 enum {
 455         OPT_FD,
 456         OPT_ROOTMODE,
 457         OPT_USER_ID,
 458         OPT_GROUP_ID,
 459         OPT_DEFAULT_PERMISSIONS,
 460         OPT_ALLOW_OTHER,
 461         OPT_MAX_READ,
 462         OPT_BLKSIZE,
 463         OPT_ERR
 464 };
 465
 466 static const match_table_t tokens = {
 467         {OPT_FD,                        "fd=%u"},
 468         {OPT_ROOTMODE,                  "rootmode=%o"},
 469         {OPT_USER_ID,                   "user_id=%u"},
 470         {OPT_GROUP_ID,                  "group_id=%u"},
 471         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 472         {OPT_ALLOW_OTHER,               "allow_other"},
 473         {OPT_MAX_READ,                  "max_read=%u"},
 474         {OPT_BLKSIZE,                   "blksize=%u"},
 475         {OPT_ERR,                       NULL}
 476 };
 477
 478 static int fuse_match_uint(substring_t *s, unsigned int *res)
 479 {
 480         int err = -ENOMEM;
 481         char *buf = match_strdup(s);
 482         if (buf) {
 483                 err = kstrtouint(buf, 10, res);
 484                 kfree(buf);
 485         }
 486         return err;
 487 }
 488
 489 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
 490 {
 491         char *p;
 492         memset(d, 0, sizeof(struct fuse_mount_data));
 493         d->max_read = ~0;
 494         d->blksize = FUSE_DEFAULT_BLKSIZE;
 495
 496         while ((p = strsep(&opt, ",")) != NULL) {
 497                 int token;
 498                 int value;
 499                 unsigned uv;
 500                 substring_t args[MAX_OPT_ARGS];
 501                 if (!*p)
 502                         continue;
 503
 504                 token = match_token(p, tokens, args);
 505                 switch (token) {
 506                 case OPT_FD:
 507                         if (match_int(&args[0], &value))
 508                                 return 0;
 509                         d->fd = value;
 510                         d->fd_present = 1;
 511                         break;
 512
 513                 case OPT_ROOTMODE:
 514                         if (match_octal(&args[0], &value))
 515                                 return 0;
 516                         if (!fuse_valid_type(value))
 517                                 return 0;
 518                         d->rootmode = value;
 519                         d->rootmode_present = 1;
 520                         break;
 521
 522                 case OPT_USER_ID:
 523                         if (fuse_match_uint(&args[0], &uv))
 524                                 return 0;
 525                         d->user_id = make_kuid(current_user_ns(), uv);
 526                         if (!uid_valid(d->user_id))
 527                                 return 0;
 528                         d->user_id_present = 1;
 529                         break;
 530
 531                 case OPT_GROUP_ID:
 532                         if (fuse_match_uint(&args[0], &uv))
 533                                 return 0;
 534                         d->group_id = make_kgid(current_user_ns(), uv);
 535                         if (!gid_valid(d->group_id))
 536                                 return 0;
 537                         d->group_id_present = 1;
 538                         break;
 539
 540                 case OPT_DEFAULT_PERMISSIONS:
 541                         d->default_permissions = 1;
 542                         break;
 543
 544                 case OPT_ALLOW_OTHER:
 545                         d->allow_other = 1;
 546                         break;
 547
 548                 case OPT_MAX_READ:
 549                         if (match_int(&args[0], &value))
 550                                 return 0;
 551                         d->max_read = value;
 552                         break;
 553
 554                 case OPT_BLKSIZE:
 555                         if (!is_bdev || match_int(&args[0], &value))
 556                                 return 0;
 557                         d->blksize = value;
 558                         break;
 559
 560                 default:
 561                         return 0;
 562                 }
 563         }
 564
 565         if (!d->fd_present || !d->rootmode_present ||
 566             !d->user_id_present || !d->group_id_present)
 567                 return 0;
 568
 569         return 1;
 570 }
 571
 572 static int fuse_show_options(struct seq_file *m, struct dentry *root)
 573 {
 574         struct super_block *sb = root->d_sb;
 575         struct fuse_conn *fc = get_fuse_conn_super(sb);
 576
 577         seq_printf(m, ",user_id=%u", from_kuid_munged(&init_user_ns, fc->user_id));
 578         seq_printf(m, ",group_id=%u", from_kgid_munged(&init_user_ns, fc->group_id));
 579         if (fc->default_permissions)
 580                 seq_puts(m, ",default_permissions");
 581         if (fc->allow_other)
 582                 seq_puts(m, ",allow_other");
 583         if (fc->max_read != ~0)
 584                 seq_printf(m, ",max_read=%u", fc->max_read);
 585         if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
 586                 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
 587         return 0;
 588 }
 589
 590 static void fuse_iqueue_init(struct fuse_iqueue *fiq)
 591 {
 592         memset(fiq, 0, sizeof(struct fuse_iqueue));
 593         init_waitqueue_head(&fiq->waitq);
 594         INIT_LIST_HEAD(&fiq->pending);
 595         INIT_LIST_HEAD(&fiq->interrupts);
 596         fiq->forget_list_tail = &fiq->forget_list_head;
 597         fiq->connected = 1;
 598 }
 599
 600 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
 601 {
 602         memset(fpq, 0, sizeof(struct fuse_pqueue));
 603         spin_lock_init(&fpq->lock);
 604         INIT_LIST_HEAD(&fpq->processing);
 605         INIT_LIST_HEAD(&fpq->io);
 606         fpq->connected = 1;
 607 }
 608
 609 void fuse_conn_init(struct fuse_conn *fc)
 610 {
 611         memset(fc, 0, sizeof(*fc));
 612         spin_lock_init(&fc->lock);
 613         init_rwsem(&fc->killsb);
 614         atomic_set(&fc->count, 1);
 615         atomic_set(&fc->dev_count, 1);
 616         init_waitqueue_head(&fc->blocked_waitq);
 617         init_waitqueue_head(&fc->reserved_req_waitq);
 618         fuse_iqueue_init(&fc->iq);
 619         INIT_LIST_HEAD(&fc->bg_queue);
 620         INIT_LIST_HEAD(&fc->entry);
 621         INIT_LIST_HEAD(&fc->devices);
 622         atomic_set(&fc->num_waiting, 0);
 623         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
 624         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
 625         fc->khctr = 0;
 626         fc->polled_files = RB_ROOT;
 627         fc->blocked = 0;
 628         fc->initialized = 0;
 629         fc->connected = 1;
 630         fc->attr_version = 1;
 631         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
 632 }
 633 EXPORT_SYMBOL_GPL(fuse_conn_init);
 634
 635 void fuse_conn_put(struct fuse_conn *fc)
 636 {
 637         if (atomic_dec_and_test(&fc->count)) {
 638                 if (fc->destroy_req)
 639                         fuse_request_free(fc->destroy_req);
 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         atomic_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, 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                 } else {
 929                         ra_pages = fc->max_read / PAGE_SIZE;
 930                         fc->no_lock = 1;
 931                         fc->no_flock = 1;
 932                 }
 933
 934                 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
 935                 fc->minor = arg->minor;
 936                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
 937                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
 938                 fc->conn_init = 1;
 939         }
 940         fuse_set_initialized(fc);
 941         wake_up_all(&fc->blocked_waitq);
 942 }
 943
 944 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
 945 {
 946         struct fuse_init_in *arg = &req->misc.init_in;
 947
 948         arg->major = FUSE_KERNEL_VERSION;
 949         arg->minor = FUSE_KERNEL_MINOR_VERSION;
 950         arg->max_readahead = fc->bdi.ra_pages * PAGE_SIZE;
 951         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
 952                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
 953                 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
 954                 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
 955                 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
 956                 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
 957                 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL;
 958         req->in.h.opcode = FUSE_INIT;
 959         req->in.numargs = 1;
 960         req->in.args[0].size = sizeof(*arg);
 961         req->in.args[0].value = arg;
 962         req->out.numargs = 1;
 963         /* Variable length argument used for backward compatibility
 964            with interface version < 7.5.  Rest of init_out is zeroed
 965            by do_get_request(), so a short reply is not a problem */
 966         req->out.argvar = 1;
 967         req->out.args[0].size = sizeof(struct fuse_init_out);
 968         req->out.args[0].value = &req->misc.init_out;
 969         req->end = process_init_reply;
 970         fuse_request_send_background(fc, req);
 971 }
 972
 973 static void fuse_free_conn(struct fuse_conn *fc)
 974 {
 975         WARN_ON(!list_empty(&fc->devices));
 976         kfree_rcu(fc, rcu);
 977 }
 978
 979 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
 980 {
 981         int err;
 982
 983         fc->bdi.name = "fuse";
 984         fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
 985         /* fuse does it's own writeback accounting */
 986         fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
 987
 988         err = bdi_init(&fc->bdi);
 989         if (err)
 990                 return err;
 991
 992         fc->bdi_initialized = 1;
 993
 994         if (sb->s_bdev) {
 995                 err =  bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
 996                                     MAJOR(fc->dev), MINOR(fc->dev));
 997         } else {
 998                 err = bdi_register_dev(&fc->bdi, fc->dev);
 999         }
1000
1001         if (err)
1002                 return err;
1003
1004         /*
1005          * For a single fuse filesystem use max 1% of dirty +
1006          * writeback threshold.
1007          *
1008          * This gives about 1M of write buffer for memory maps on a
1009          * machine with 1G and 10% dirty_ratio, which should be more
1010          * than enough.
1011          *
1012          * Privileged users can raise it by writing to
1013          *
1014          *    /sys/class/bdi/<bdi>/max_ratio
1015          */
1016         bdi_set_max_ratio(&fc->bdi, 1);
1017
1018         return 0;
1019 }
1020
1021 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc)
1022 {
1023         struct fuse_dev *fud;
1024
1025         fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1026         if (fud) {
1027                 fud->fc = fuse_conn_get(fc);
1028                 fuse_pqueue_init(&fud->pq);
1029
1030                 spin_lock(&fc->lock);
1031                 list_add_tail(&fud->entry, &fc->devices);
1032                 spin_unlock(&fc->lock);
1033         }
1034
1035         return fud;
1036 }
1037 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1038
1039 void fuse_dev_free(struct fuse_dev *fud)
1040 {
1041         struct fuse_conn *fc = fud->fc;
1042
1043         if (fc) {
1044                 spin_lock(&fc->lock);
1045                 list_del(&fud->entry);
1046                 spin_unlock(&fc->lock);
1047
1048                 fuse_conn_put(fc);
1049         }
1050         kfree(fud);
1051 }
1052 EXPORT_SYMBOL_GPL(fuse_dev_free);
1053
1054 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
1055 {
1056         struct fuse_dev *fud;
1057         struct fuse_conn *fc;
1058         struct inode *root;
1059         struct fuse_mount_data d;
1060         struct file *file;
1061         struct dentry *root_dentry;
1062         struct fuse_req *init_req;
1063         int err;
1064         int is_bdev = sb->s_bdev != NULL;
1065
1066         err = -EINVAL;
1067         if (sb->s_flags & MS_MANDLOCK)
1068                 goto err;
1069
1070         sb->s_flags &= ~(MS_NOSEC | MS_I_VERSION);
1071
1072         if (!parse_fuse_opt(data, &d, is_bdev))
1073                 goto err;
1074
1075         if (is_bdev) {
1076 #ifdef CONFIG_BLOCK
1077                 err = -EINVAL;
1078                 if (!sb_set_blocksize(sb, d.blksize))
1079                         goto err;
1080 #endif
1081         } else {
1082                 sb->s_blocksize = PAGE_SIZE;
1083                 sb->s_blocksize_bits = PAGE_SHIFT;
1084         }
1085         sb->s_magic = FUSE_SUPER_MAGIC;
1086         sb->s_op = &fuse_super_operations;
1087         sb->s_xattr = fuse_xattr_handlers;
1088         sb->s_maxbytes = MAX_LFS_FILESIZE;
1089         sb->s_time_gran = 1;
1090         sb->s_export_op = &fuse_export_operations;
1091
1092         file = fget(d.fd);
1093         err = -EINVAL;
1094         if (!file)
1095                 goto err;
1096
1097         if ((file->f_op != &fuse_dev_operations) ||
1098             (file->f_cred->user_ns != &init_user_ns))
1099                 goto err_fput;
1100
1101         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1102         err = -ENOMEM;
1103         if (!fc)
1104                 goto err_fput;
1105
1106         fuse_conn_init(fc);
1107         fc->release = fuse_free_conn;
1108
1109         fud = fuse_dev_alloc(fc);
1110         if (!fud)
1111                 goto err_put_conn;
1112
1113         fc->dev = sb->s_dev;
1114         fc->sb = sb;
1115         err = fuse_bdi_init(fc, sb);
1116         if (err)
1117                 goto err_dev_free;
1118
1119         sb->s_bdi = &fc->bdi;
1120
1121         /* Handle umasking inside the fuse code */
1122         if (sb->s_flags & MS_POSIXACL)
1123                 fc->dont_mask = 1;
1124         sb->s_flags |= MS_POSIXACL;
1125
1126         fc->default_permissions = d.default_permissions;
1127         fc->allow_other = d.allow_other;
1128         fc->user_id = d.user_id;
1129         fc->group_id = d.group_id;
1130         fc->max_read = max_t(unsigned, 4096, d.max_read);
1131
1132         /* Used by get_root_inode() */
1133         sb->s_fs_info = fc;
1134
1135         err = -ENOMEM;
1136         root = fuse_get_root_inode(sb, d.rootmode);
1137         sb->s_d_op = &fuse_root_dentry_operations;
1138         root_dentry = d_make_root(root);
1139         if (!root_dentry)
1140                 goto err_dev_free;
1141         /* Root dentry doesn't have .d_revalidate */
1142         sb->s_d_op = &fuse_dentry_operations;
1143
1144         init_req = fuse_request_alloc(0);
1145         if (!init_req)
1146                 goto err_put_root;
1147         __set_bit(FR_BACKGROUND, &init_req->flags);
1148
1149         if (is_bdev) {
1150                 fc->destroy_req = fuse_request_alloc(0);
1151                 if (!fc->destroy_req)
1152                         goto err_free_init_req;
1153         }
1154
1155         mutex_lock(&fuse_mutex);
1156         err = -EINVAL;
1157         if (file->private_data)
1158                 goto err_unlock;
1159
1160         err = fuse_ctl_add_conn(fc);
1161         if (err)
1162                 goto err_unlock;
1163
1164         list_add_tail(&fc->entry, &fuse_conn_list);
1165         sb->s_root = root_dentry;
1166         file->private_data = fud;
1167         mutex_unlock(&fuse_mutex);
1168         /*
1169          * atomic_dec_and_test() in fput() provides the necessary
1170          * memory barrier for file->private_data to be visible on all
1171          * CPUs after this
1172          */
1173         fput(file);
1174
1175         fuse_send_init(fc, init_req);
1176
1177         return 0;
1178
1179  err_unlock:
1180         mutex_unlock(&fuse_mutex);
1181  err_free_init_req:
1182         fuse_request_free(init_req);
1183  err_put_root:
1184         dput(root_dentry);
1185  err_dev_free:
1186         fuse_dev_free(fud);
1187  err_put_conn:
1188         fuse_bdi_destroy(fc);
1189         fuse_conn_put(fc);
1190         sb->s_fs_info = NULL;
1191  err_fput:
1192         fput(file);
1193  err:
1194         return err;
1195 }
1196
1197 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1198                        int flags, const char *dev_name,
1199                        void *raw_data)
1200 {
1201         return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1202 }
1203
1204 static void fuse_sb_destroy(struct super_block *sb)
1205 {
1206         struct fuse_conn *fc = get_fuse_conn_super(sb);
1207
1208         if (fc) {
1209                 fuse_send_destroy(fc);
1210
1211                 fuse_abort_conn(fc);
1212                 fuse_wait_aborted(fc);
1213
1214                 down_write(&fc->killsb);
1215                 fc->sb = NULL;
1216                 up_write(&fc->killsb);
1217         }
1218 }
1219
1220 static void fuse_kill_sb_anon(struct super_block *sb)
1221 {
1222         fuse_sb_destroy(sb);
1223         kill_anon_super(sb);
1224 }
1225
1226 static struct file_system_type fuse_fs_type = {
1227         .owner          = THIS_MODULE,
1228         .name           = "fuse",
1229         .fs_flags       = FS_HAS_SUBTYPE,
1230         .mount          = fuse_mount,
1231         .kill_sb        = fuse_kill_sb_anon,
1232 };
1233 MODULE_ALIAS_FS("fuse");
1234
1235 #ifdef CONFIG_BLOCK
1236 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1237                            int flags, const char *dev_name,
1238                            void *raw_data)
1239 {
1240         return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1241 }
1242
1243 static void fuse_kill_sb_blk(struct super_block *sb)
1244 {
1245         fuse_sb_destroy(sb);
1246         kill_block_super(sb);
1247 }
1248
1249 static struct file_system_type fuseblk_fs_type = {
1250         .owner          = THIS_MODULE,
1251         .name           = "fuseblk",
1252         .mount          = fuse_mount_blk,
1253         .kill_sb        = fuse_kill_sb_blk,
1254         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1255 };
1256 MODULE_ALIAS_FS("fuseblk");
1257
1258 static inline int register_fuseblk(void)
1259 {
1260         return register_filesystem(&fuseblk_fs_type);
1261 }
1262
1263 static inline void unregister_fuseblk(void)
1264 {
1265         unregister_filesystem(&fuseblk_fs_type);
1266 }
1267 #else
1268 static inline int register_fuseblk(void)
1269 {
1270         return 0;
1271 }
1272
1273 static inline void unregister_fuseblk(void)
1274 {
1275 }
1276 #endif
1277
1278 static void fuse_inode_init_once(void *foo)
1279 {
1280         struct inode *inode = foo;
1281
1282         inode_init_once(inode);
1283 }
1284
1285 static int __init fuse_fs_init(void)
1286 {
1287         int err;
1288
1289         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1290                                               sizeof(struct fuse_inode), 0,
1291                                               SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT,
1292                                               fuse_inode_init_once);
1293         err = -ENOMEM;
1294         if (!fuse_inode_cachep)
1295                 goto out;
1296
1297         err = register_fuseblk();
1298         if (err)
1299                 goto out2;
1300
1301         err = register_filesystem(&fuse_fs_type);
1302         if (err)
1303                 goto out3;
1304
1305         return 0;
1306
1307  out3:
1308         unregister_fuseblk();
1309  out2:
1310         kmem_cache_destroy(fuse_inode_cachep);
1311  out:
1312         return err;
1313 }
1314
1315 static void fuse_fs_cleanup(void)
1316 {
1317         unregister_filesystem(&fuse_fs_type);
1318         unregister_fuseblk();
1319
1320         /*
1321          * Make sure all delayed rcu free inodes are flushed before we
1322          * destroy cache.
1323          */
1324         rcu_barrier();
1325         kmem_cache_destroy(fuse_inode_cachep);
1326 }
1327
1328 static struct kobject *fuse_kobj;
1329
1330 static int fuse_sysfs_init(void)
1331 {
1332         int err;
1333
1334         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1335         if (!fuse_kobj) {
1336                 err = -ENOMEM;
1337                 goto out_err;
1338         }
1339
1340         err = sysfs_create_mount_point(fuse_kobj, "connections");
1341         if (err)
1342                 goto out_fuse_unregister;
1343
1344         return 0;
1345
1346  out_fuse_unregister:
1347         kobject_put(fuse_kobj);
1348  out_err:
1349         return err;
1350 }
1351
1352 static void fuse_sysfs_cleanup(void)
1353 {
1354         sysfs_remove_mount_point(fuse_kobj, "connections");
1355         kobject_put(fuse_kobj);
1356 }
1357
1358 static int __init fuse_init(void)
1359 {
1360         int res;
1361
1362         printk(KERN_INFO "fuse init (API version %i.%i)\n",
1363                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1364
1365         INIT_LIST_HEAD(&fuse_conn_list);
1366         res = fuse_fs_init();
1367         if (res)
1368                 goto err;
1369
1370         res = fuse_dev_init();
1371         if (res)
1372                 goto err_fs_cleanup;
1373
1374         res = fuse_sysfs_init();
1375         if (res)
1376                 goto err_dev_cleanup;
1377
1378         res = fuse_ctl_init();
1379         if (res)
1380                 goto err_sysfs_cleanup;
1381
1382         sanitize_global_limit(&max_user_bgreq);
1383         sanitize_global_limit(&max_user_congthresh);
1384
1385         return 0;
1386
1387  err_sysfs_cleanup:
1388         fuse_sysfs_cleanup();
1389  err_dev_cleanup:
1390         fuse_dev_cleanup();
1391  err_fs_cleanup:
1392         fuse_fs_cleanup();
1393  err:
1394         return res;
1395 }
1396
1397 static void __exit fuse_exit(void)
1398 {
1399         printk(KERN_DEBUG "fuse exit\n");
1400
1401         fuse_ctl_cleanup();
1402         fuse_sysfs_cleanup();
1403         fuse_fs_cleanup();
1404         fuse_dev_cleanup();
1405 }
1406
1407 module_init(fuse_init);
1408 module_exit(fuse_exit);