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, 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 & MS_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 & MS_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(&init_user_ns, attr->uid);
 175         inode->i_gid     = make_kgid(&init_user_ns, attr->gid);
 176         inode->i_blocks  = attr->blocks;
 177
 178         /* Sanitize nsecs */
 179         attr->atimensec = min_t(u32, attr->atimensec, NSEC_PER_SEC - 1);
 180         attr->mtimensec = min_t(u32, attr->mtimensec, NSEC_PER_SEC - 1);
 181         attr->ctimensec = min_t(u32, attr->ctimensec, NSEC_PER_SEC - 1);
 182
 183         inode->i_atime.tv_sec   = attr->atime;
 184         inode->i_atime.tv_nsec  = attr->atimensec;
 185         /* mtime from server may be stale due to local buffered write */
 186         if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
 187                 inode->i_mtime.tv_sec   = attr->mtime;
 188                 inode->i_mtime.tv_nsec  = attr->mtimensec;
 189                 inode->i_ctime.tv_sec   = attr->ctime;
 190                 inode->i_ctime.tv_nsec  = attr->ctimensec;
 191         }
 192
 193         if (attr->blksize != 0)
 194                 inode->i_blkbits = ilog2(attr->blksize);
 195         else
 196                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
 197
 198         /*
 199          * Don't set the sticky bit in i_mode, unless we want the VFS
 200          * to check permissions.  This prevents failures due to the
 201          * check in may_delete().
 202          */
 203         fi->orig_i_mode = inode->i_mode;
 204         if (!fc->default_permissions)
 205                 inode->i_mode &= ~S_ISVTX;
 206
 207         fi->orig_ino = attr->ino;
 208 }
 209
 210 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
 211                             u64 attr_valid, u64 attr_version)
 212 {
 213         struct fuse_conn *fc = get_fuse_conn(inode);
 214         struct fuse_inode *fi = get_fuse_inode(inode);
 215         bool is_wb = fc->writeback_cache;
 216         loff_t oldsize;
 217         struct timespec old_mtime;
 218
 219         spin_lock(&fc->lock);
 220         if ((attr_version != 0 && fi->attr_version > attr_version) ||
 221             test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
 222                 spin_unlock(&fc->lock);
 223                 return;
 224         }
 225
 226         old_mtime = inode->i_mtime;
 227         fuse_change_attributes_common(inode, attr, attr_valid);
 228
 229         oldsize = inode->i_size;
 230         /*
 231          * In case of writeback_cache enabled, the cached writes beyond EOF
 232          * extend local i_size without keeping userspace server in sync. So,
 233          * attr->size coming from server can be stale. We cannot trust it.
 234          */
 235         if (!is_wb || !S_ISREG(inode->i_mode))
 236                 i_size_write(inode, attr->size);
 237         spin_unlock(&fc->lock);
 238
 239         if (!is_wb && S_ISREG(inode->i_mode)) {
 240                 bool inval = false;
 241
 242                 if (oldsize != attr->size) {
 243                         truncate_pagecache(inode, attr->size);
 244                         inval = true;
 245                 } else if (fc->auto_inval_data) {
 246                         struct timespec new_mtime = {
 247                                 .tv_sec = attr->mtime,
 248                                 .tv_nsec = attr->mtimensec,
 249                         };
 250
 251                         /*
 252                          * Auto inval mode also checks and invalidates if mtime
 253                          * has changed.
 254                          */
 255                         if (!timespec_equal(&old_mtime, &new_mtime))
 256                                 inval = true;
 257                 }
 258
 259                 if (inval)
 260                         invalidate_inode_pages2(inode->i_mapping);
 261         }
 262 }
 263
 264 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
 265 {
 266         inode->i_mode = attr->mode & S_IFMT;
 267         inode->i_size = attr->size;
 268         inode->i_mtime.tv_sec  = attr->mtime;
 269         inode->i_mtime.tv_nsec = attr->mtimensec;
 270         inode->i_ctime.tv_sec  = attr->ctime;
 271         inode->i_ctime.tv_nsec = attr->ctimensec;
 272         if (S_ISREG(inode->i_mode)) {
 273                 fuse_init_common(inode);
 274                 fuse_init_file_inode(inode);
 275         } else if (S_ISDIR(inode->i_mode))
 276                 fuse_init_dir(inode);
 277         else if (S_ISLNK(inode->i_mode))
 278                 fuse_init_symlink(inode);
 279         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
 280                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
 281                 fuse_init_common(inode);
 282                 init_special_inode(inode, inode->i_mode,
 283                                    new_decode_dev(attr->rdev));
 284         } else
 285                 BUG();
 286 }
 287
 288 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
 289 {
 290         u64 nodeid = *(u64 *) _nodeidp;
 291         if (get_node_id(inode) == nodeid)
 292                 return 1;
 293         else
 294                 return 0;
 295 }
 296
 297 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
 298 {
 299         u64 nodeid = *(u64 *) _nodeidp;
 300         get_fuse_inode(inode)->nodeid = nodeid;
 301         return 0;
 302 }
 303
 304 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
 305                         int generation, struct fuse_attr *attr,
 306                         u64 attr_valid, u64 attr_version)
 307 {
 308         struct inode *inode;
 309         struct fuse_inode *fi;
 310         struct fuse_conn *fc = get_fuse_conn_super(sb);
 311
 312  retry:
 313         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
 314         if (!inode)
 315                 return NULL;
 316
 317         if ((inode->i_state & I_NEW)) {
 318                 inode->i_flags |= S_NOATIME;
 319                 if (!fc->writeback_cache || !S_ISREG(attr->mode))
 320                         inode->i_flags |= S_NOCMTIME;
 321                 inode->i_generation = generation;
 322                 fuse_init_inode(inode, attr);
 323                 unlock_new_inode(inode);
 324         } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
 325                 /* Inode has changed type, any I/O on the old should fail */
 326                 fuse_make_bad(inode);
 327                 iput(inode);
 328                 goto retry;
 329         }
 330
 331         fi = get_fuse_inode(inode);
 332         spin_lock(&fc->lock);
 333         fi->nlookup++;
 334         spin_unlock(&fc->lock);
 335         fuse_change_attributes(inode, attr, attr_valid, attr_version);
 336
 337         return inode;
 338 }
 339
 340 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
 341                              loff_t offset, loff_t len)
 342 {
 343         struct inode *inode;
 344         pgoff_t pg_start;
 345         pgoff_t pg_end;
 346
 347         inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
 348         if (!inode)
 349                 return -ENOENT;
 350
 351         fuse_invalidate_attr(inode);
 352         forget_all_cached_acls(inode);
 353         if (offset >= 0) {
 354                 pg_start = offset >> PAGE_SHIFT;
 355                 if (len <= 0)
 356                         pg_end = -1;
 357                 else
 358                         pg_end = (offset + len - 1) >> PAGE_SHIFT;
 359                 invalidate_inode_pages2_range(inode->i_mapping,
 360                                               pg_start, pg_end);
 361         }
 362         iput(inode);
 363         return 0;
 364 }
 365
 366 bool fuse_lock_inode(struct inode *inode)
 367 {
 368         bool locked = false;
 369
 370         if (!get_fuse_conn(inode)->parallel_dirops) {
 371                 mutex_lock(&get_fuse_inode(inode)->mutex);
 372                 locked = true;
 373         }
 374
 375         return locked;
 376 }
 377
 378 void fuse_unlock_inode(struct inode *inode, bool locked)
 379 {
 380         if (locked)
 381                 mutex_unlock(&get_fuse_inode(inode)->mutex);
 382 }
 383
 384 static void fuse_umount_begin(struct super_block *sb)
 385 {
 386         fuse_abort_conn(get_fuse_conn_super(sb));
 387 }
 388
 389 static void fuse_send_destroy(struct fuse_conn *fc)
 390 {
 391         struct fuse_req *req = fc->destroy_req;
 392         if (req && fc->conn_init) {
 393                 fc->destroy_req = NULL;
 394                 req->in.h.opcode = FUSE_DESTROY;
 395                 __set_bit(FR_FORCE, &req->flags);
 396                 __clear_bit(FR_BACKGROUND, &req->flags);
 397                 fuse_request_send(fc, req);
 398                 fuse_put_request(fc, req);
 399         }
 400 }
 401
 402 static void fuse_put_super(struct super_block *sb)
 403 {
 404         struct fuse_conn *fc = get_fuse_conn_super(sb);
 405
 406         mutex_lock(&fuse_mutex);
 407         list_del(&fc->entry);
 408         fuse_ctl_remove_conn(fc);
 409         mutex_unlock(&fuse_mutex);
 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         refcount_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         fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
 633 }
 634 EXPORT_SYMBOL_GPL(fuse_conn_init);
 635
 636 void fuse_conn_put(struct fuse_conn *fc)
 637 {
 638         if (refcount_dec_and_test(&fc->count)) {
 639                 if (fc->destroy_req)
 640                         fuse_request_free(fc->destroy_req);
 641                 put_pid_ns(fc->pid_ns);
 642                 fc->release(fc);
 643         }
 644 }
 645 EXPORT_SYMBOL_GPL(fuse_conn_put);
 646
 647 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
 648 {
 649         refcount_inc(&fc->count);
 650         return fc;
 651 }
 652 EXPORT_SYMBOL_GPL(fuse_conn_get);
 653
 654 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
 655 {
 656         struct fuse_attr attr;
 657         memset(&attr, 0, sizeof(attr));
 658
 659         attr.mode = mode;
 660         attr.ino = FUSE_ROOT_ID;
 661         attr.nlink = 1;
 662         return fuse_iget(sb, 1, 0, &attr, 0, 0);
 663 }
 664
 665 struct fuse_inode_handle {
 666         u64 nodeid;
 667         u32 generation;
 668 };
 669
 670 static struct dentry *fuse_get_dentry(struct super_block *sb,
 671                                       struct fuse_inode_handle *handle)
 672 {
 673         struct fuse_conn *fc = get_fuse_conn_super(sb);
 674         struct inode *inode;
 675         struct dentry *entry;
 676         int err = -ESTALE;
 677
 678         if (handle->nodeid == 0)
 679                 goto out_err;
 680
 681         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
 682         if (!inode) {
 683                 struct fuse_entry_out outarg;
 684                 const struct qstr name = QSTR_INIT(".", 1);
 685
 686                 if (!fc->export_support)
 687                         goto out_err;
 688
 689                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
 690                                        &inode);
 691                 if (err && err != -ENOENT)
 692                         goto out_err;
 693                 if (err || !inode) {
 694                         err = -ESTALE;
 695                         goto out_err;
 696                 }
 697                 err = -EIO;
 698                 if (get_node_id(inode) != handle->nodeid)
 699                         goto out_iput;
 700         }
 701         err = -ESTALE;
 702         if (inode->i_generation != handle->generation)
 703                 goto out_iput;
 704
 705         entry = d_obtain_alias(inode);
 706         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
 707                 fuse_invalidate_entry_cache(entry);
 708
 709         return entry;
 710
 711  out_iput:
 712         iput(inode);
 713  out_err:
 714         return ERR_PTR(err);
 715 }
 716
 717 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
 718                            struct inode *parent)
 719 {
 720         int len = parent ? 6 : 3;
 721         u64 nodeid;
 722         u32 generation;
 723
 724         if (*max_len < len) {
 725                 *max_len = len;
 726                 return  FILEID_INVALID;
 727         }
 728
 729         nodeid = get_fuse_inode(inode)->nodeid;
 730         generation = inode->i_generation;
 731
 732         fh[0] = (u32)(nodeid >> 32);
 733         fh[1] = (u32)(nodeid & 0xffffffff);
 734         fh[2] = generation;
 735
 736         if (parent) {
 737                 nodeid = get_fuse_inode(parent)->nodeid;
 738                 generation = parent->i_generation;
 739
 740                 fh[3] = (u32)(nodeid >> 32);
 741                 fh[4] = (u32)(nodeid & 0xffffffff);
 742                 fh[5] = generation;
 743         }
 744
 745         *max_len = len;
 746         return parent ? 0x82 : 0x81;
 747 }
 748
 749 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
 750                 struct fid *fid, int fh_len, int fh_type)
 751 {
 752         struct fuse_inode_handle handle;
 753
 754         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
 755                 return NULL;
 756
 757         handle.nodeid = (u64) fid->raw[0] << 32;
 758         handle.nodeid |= (u64) fid->raw[1];
 759         handle.generation = fid->raw[2];
 760         return fuse_get_dentry(sb, &handle);
 761 }
 762
 763 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
 764                 struct fid *fid, int fh_len, int fh_type)
 765 {
 766         struct fuse_inode_handle parent;
 767
 768         if (fh_type != 0x82 || fh_len < 6)
 769                 return NULL;
 770
 771         parent.nodeid = (u64) fid->raw[3] << 32;
 772         parent.nodeid |= (u64) fid->raw[4];
 773         parent.generation = fid->raw[5];
 774         return fuse_get_dentry(sb, &parent);
 775 }
 776
 777 static struct dentry *fuse_get_parent(struct dentry *child)
 778 {
 779         struct inode *child_inode = d_inode(child);
 780         struct fuse_conn *fc = get_fuse_conn(child_inode);
 781         struct inode *inode;
 782         struct dentry *parent;
 783         struct fuse_entry_out outarg;
 784         const struct qstr name = QSTR_INIT("..", 2);
 785         int err;
 786
 787         if (!fc->export_support)
 788                 return ERR_PTR(-ESTALE);
 789
 790         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
 791                                &name, &outarg, &inode);
 792         if (err) {
 793                 if (err == -ENOENT)
 794                         return ERR_PTR(-ESTALE);
 795                 return ERR_PTR(err);
 796         }
 797
 798         parent = d_obtain_alias(inode);
 799         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
 800                 fuse_invalidate_entry_cache(parent);
 801
 802         return parent;
 803 }
 804
 805 static const struct export_operations fuse_export_operations = {
 806         .fh_to_dentry   = fuse_fh_to_dentry,
 807         .fh_to_parent   = fuse_fh_to_parent,
 808         .encode_fh      = fuse_encode_fh,
 809         .get_parent     = fuse_get_parent,
 810 };
 811
 812 static const struct super_operations fuse_super_operations = {
 813         .alloc_inode    = fuse_alloc_inode,
 814         .destroy_inode  = fuse_destroy_inode,
 815         .evict_inode    = fuse_evict_inode,
 816         .write_inode    = fuse_write_inode,
 817         .drop_inode     = generic_delete_inode,
 818         .remount_fs     = fuse_remount_fs,
 819         .put_super      = fuse_put_super,
 820         .umount_begin   = fuse_umount_begin,
 821         .statfs         = fuse_statfs,
 822         .show_options   = fuse_show_options,
 823 };
 824
 825 static void sanitize_global_limit(unsigned *limit)
 826 {
 827         if (*limit == 0)
 828                 *limit = ((totalram_pages << PAGE_SHIFT) >> 13) /
 829                          sizeof(struct fuse_req);
 830
 831         if (*limit >= 1 << 16)
 832                 *limit = (1 << 16) - 1;
 833 }
 834
 835 static int set_global_limit(const char *val, struct kernel_param *kp)
 836 {
 837         int rv;
 838
 839         rv = param_set_uint(val, kp);
 840         if (rv)
 841                 return rv;
 842
 843         sanitize_global_limit((unsigned *)kp->arg);
 844
 845         return 0;
 846 }
 847
 848 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
 849 {
 850         int cap_sys_admin = capable(CAP_SYS_ADMIN);
 851
 852         if (arg->minor < 13)
 853                 return;
 854
 855         sanitize_global_limit(&max_user_bgreq);
 856         sanitize_global_limit(&max_user_congthresh);
 857
 858         if (arg->max_background) {
 859                 fc->max_background = arg->max_background;
 860
 861                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
 862                         fc->max_background = max_user_bgreq;
 863         }
 864         if (arg->congestion_threshold) {
 865                 fc->congestion_threshold = arg->congestion_threshold;
 866
 867                 if (!cap_sys_admin &&
 868                     fc->congestion_threshold > max_user_congthresh)
 869                         fc->congestion_threshold = max_user_congthresh;
 870         }
 871 }
 872
 873 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
 874 {
 875         struct fuse_init_out *arg = &req->misc.init_out;
 876
 877         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
 878                 fc->conn_error = 1;
 879         else {
 880                 unsigned long ra_pages;
 881
 882                 process_init_limits(fc, arg);
 883
 884                 if (arg->minor >= 6) {
 885                         ra_pages = arg->max_readahead / PAGE_SIZE;
 886                         if (arg->flags & FUSE_ASYNC_READ)
 887                                 fc->async_read = 1;
 888                         if (!(arg->flags & FUSE_POSIX_LOCKS))
 889                                 fc->no_lock = 1;
 890                         if (arg->minor >= 17) {
 891                                 if (!(arg->flags & FUSE_FLOCK_LOCKS))
 892                                         fc->no_flock = 1;
 893                         } else {
 894                                 if (!(arg->flags & FUSE_POSIX_LOCKS))
 895                                         fc->no_flock = 1;
 896                         }
 897                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
 898                                 fc->atomic_o_trunc = 1;
 899                         if (arg->minor >= 9) {
 900                                 /* LOOKUP has dependency on proto version */
 901                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
 902                                         fc->export_support = 1;
 903                         }
 904                         if (arg->flags & FUSE_BIG_WRITES)
 905                                 fc->big_writes = 1;
 906                         if (arg->flags & FUSE_DONT_MASK)
 907                                 fc->dont_mask = 1;
 908                         if (arg->flags & FUSE_AUTO_INVAL_DATA)
 909                                 fc->auto_inval_data = 1;
 910                         if (arg->flags & FUSE_DO_READDIRPLUS) {
 911                                 fc->do_readdirplus = 1;
 912                                 if (arg->flags & FUSE_READDIRPLUS_AUTO)
 913                                         fc->readdirplus_auto = 1;
 914                         }
 915                         if (arg->flags & FUSE_ASYNC_DIO)
 916                                 fc->async_dio = 1;
 917                         if (arg->flags & FUSE_WRITEBACK_CACHE)
 918                                 fc->writeback_cache = 1;
 919                         if (arg->flags & FUSE_PARALLEL_DIROPS)
 920                                 fc->parallel_dirops = 1;
 921                         if (arg->flags & FUSE_HANDLE_KILLPRIV)
 922                                 fc->handle_killpriv = 1;
 923                         if (arg->time_gran && arg->time_gran <= 1000000000)
 924                                 fc->sb->s_time_gran = arg->time_gran;
 925                         if ((arg->flags & FUSE_POSIX_ACL)) {
 926                                 fc->default_permissions = 1;
 927                                 fc->posix_acl = 1;
 928                                 fc->sb->s_xattr = fuse_acl_xattr_handlers;
 929                         }
 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         req->in.h.opcode = FUSE_INIT;
 962         req->in.numargs = 1;
 963         req->in.args[0].size = sizeof(*arg);
 964         req->in.args[0].value = arg;
 965         req->out.numargs = 1;
 966         /* Variable length argument used for backward compatibility
 967            with interface version < 7.5.  Rest of init_out is zeroed
 968            by do_get_request(), so a short reply is not a problem */
 969         req->out.argvar = 1;
 970         req->out.args[0].size = sizeof(struct fuse_init_out);
 971         req->out.args[0].value = &req->misc.init_out;
 972         req->end = process_init_reply;
 973         fuse_request_send_background(fc, req);
 974 }
 975
 976 static void fuse_free_conn(struct fuse_conn *fc)
 977 {
 978         WARN_ON(!list_empty(&fc->devices));
 979         kfree_rcu(fc, rcu);
 980 }
 981
 982 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
 983 {
 984         int err;
 985         char *suffix = "";
 986
 987         if (sb->s_bdev) {
 988                 suffix = "-fuseblk";
 989                 /*
 990                  * sb->s_bdi points to blkdev's bdi however we want to redirect
 991                  * it to our private bdi...
 992                  */
 993                 bdi_put(sb->s_bdi);
 994                 sb->s_bdi = &noop_backing_dev_info;
 995         }
 996         err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
 997                                    MINOR(fc->dev), suffix);
 998         if (err)
 999                 return err;
1000
1001         sb->s_bdi->ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
1002         /* fuse does it's own writeback accounting */
1003         sb->s_bdi->capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
1004
1005         /*
1006          * For a single fuse filesystem use max 1% of dirty +
1007          * writeback threshold.
1008          *
1009          * This gives about 1M of write buffer for memory maps on a
1010          * machine with 1G and 10% dirty_ratio, which should be more
1011          * than enough.
1012          *
1013          * Privileged users can raise it by writing to
1014          *
1015          *    /sys/class/bdi/<bdi>/max_ratio
1016          */
1017         bdi_set_max_ratio(sb->s_bdi, 1);
1018
1019         return 0;
1020 }
1021
1022 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc)
1023 {
1024         struct fuse_dev *fud;
1025
1026         fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1027         if (fud) {
1028                 fud->fc = fuse_conn_get(fc);
1029                 fuse_pqueue_init(&fud->pq);
1030
1031                 spin_lock(&fc->lock);
1032                 list_add_tail(&fud->entry, &fc->devices);
1033                 spin_unlock(&fc->lock);
1034         }
1035
1036         return fud;
1037 }
1038 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1039
1040 void fuse_dev_free(struct fuse_dev *fud)
1041 {
1042         struct fuse_conn *fc = fud->fc;
1043
1044         if (fc) {
1045                 spin_lock(&fc->lock);
1046                 list_del(&fud->entry);
1047                 spin_unlock(&fc->lock);
1048
1049                 fuse_conn_put(fc);
1050         }
1051         kfree(fud);
1052 }
1053 EXPORT_SYMBOL_GPL(fuse_dev_free);
1054
1055 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
1056 {
1057         struct fuse_dev *fud;
1058         struct fuse_conn *fc;
1059         struct inode *root;
1060         struct fuse_mount_data d;
1061         struct file *file;
1062         struct dentry *root_dentry;
1063         struct fuse_req *init_req;
1064         int err;
1065         int is_bdev = sb->s_bdev != NULL;
1066
1067         err = -EINVAL;
1068         if (sb->s_flags & MS_MANDLOCK)
1069                 goto err;
1070
1071         sb->s_flags &= ~(MS_NOSEC | SB_I_VERSION);
1072
1073         if (!parse_fuse_opt(data, &d, is_bdev))
1074                 goto err;
1075
1076         if (is_bdev) {
1077 #ifdef CONFIG_BLOCK
1078                 err = -EINVAL;
1079                 if (!sb_set_blocksize(sb, d.blksize))
1080                         goto err;
1081 #endif
1082         } else {
1083                 sb->s_blocksize = PAGE_SIZE;
1084                 sb->s_blocksize_bits = PAGE_SHIFT;
1085         }
1086         sb->s_magic = FUSE_SUPER_MAGIC;
1087         sb->s_op = &fuse_super_operations;
1088         sb->s_xattr = fuse_xattr_handlers;
1089         sb->s_maxbytes = MAX_LFS_FILESIZE;
1090         sb->s_time_gran = 1;
1091         sb->s_export_op = &fuse_export_operations;
1092
1093         file = fget(d.fd);
1094         err = -EINVAL;
1095         if (!file)
1096                 goto err;
1097
1098         if ((file->f_op != &fuse_dev_operations) ||
1099             (file->f_cred->user_ns != &init_user_ns))
1100                 goto err_fput;
1101
1102         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1103         err = -ENOMEM;
1104         if (!fc)
1105                 goto err_fput;
1106
1107         fuse_conn_init(fc);
1108         fc->release = fuse_free_conn;
1109
1110         fud = fuse_dev_alloc(fc);
1111         if (!fud)
1112                 goto err_put_conn;
1113
1114         fc->dev = sb->s_dev;
1115         fc->sb = sb;
1116         err = fuse_bdi_init(fc, sb);
1117         if (err)
1118                 goto err_dev_free;
1119
1120         /* Handle umasking inside the fuse code */
1121         if (sb->s_flags & MS_POSIXACL)
1122                 fc->dont_mask = 1;
1123         sb->s_flags |= MS_POSIXACL;
1124
1125         fc->default_permissions = d.default_permissions;
1126         fc->allow_other = d.allow_other;
1127         fc->user_id = d.user_id;
1128         fc->group_id = d.group_id;
1129         fc->max_read = max_t(unsigned, 4096, d.max_read);
1130
1131         /* Used by get_root_inode() */
1132         sb->s_fs_info = fc;
1133
1134         err = -ENOMEM;
1135         root = fuse_get_root_inode(sb, d.rootmode);
1136         sb->s_d_op = &fuse_root_dentry_operations;
1137         root_dentry = d_make_root(root);
1138         if (!root_dentry)
1139                 goto err_dev_free;
1140         /* Root dentry doesn't have .d_revalidate */
1141         sb->s_d_op = &fuse_dentry_operations;
1142
1143         init_req = fuse_request_alloc(0);
1144         if (!init_req)
1145                 goto err_put_root;
1146         __set_bit(FR_BACKGROUND, &init_req->flags);
1147
1148         if (is_bdev) {
1149                 fc->destroy_req = fuse_request_alloc(0);
1150                 if (!fc->destroy_req)
1151                         goto err_free_init_req;
1152         }
1153
1154         mutex_lock(&fuse_mutex);
1155         err = -EINVAL;
1156         if (file->private_data)
1157                 goto err_unlock;
1158
1159         err = fuse_ctl_add_conn(fc);
1160         if (err)
1161                 goto err_unlock;
1162
1163         list_add_tail(&fc->entry, &fuse_conn_list);
1164         sb->s_root = root_dentry;
1165         file->private_data = fud;
1166         mutex_unlock(&fuse_mutex);
1167         /*
1168          * atomic_dec_and_test() in fput() provides the necessary
1169          * memory barrier for file->private_data to be visible on all
1170          * CPUs after this
1171          */
1172         fput(file);
1173
1174         fuse_send_init(fc, init_req);
1175
1176         return 0;
1177
1178  err_unlock:
1179         mutex_unlock(&fuse_mutex);
1180  err_free_init_req:
1181         fuse_request_free(init_req);
1182  err_put_root:
1183         dput(root_dentry);
1184  err_dev_free:
1185         fuse_dev_free(fud);
1186  err_put_conn:
1187         fuse_conn_put(fc);
1188         sb->s_fs_info = NULL;
1189  err_fput:
1190         fput(file);
1191  err:
1192         return err;
1193 }
1194
1195 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1196                        int flags, const char *dev_name,
1197                        void *raw_data)
1198 {
1199         return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1200 }
1201
1202 static void fuse_sb_destroy(struct super_block *sb)
1203 {
1204         struct fuse_conn *fc = get_fuse_conn_super(sb);
1205
1206         if (fc) {
1207                 fuse_send_destroy(fc);
1208
1209                 fuse_abort_conn(fc);
1210                 fuse_wait_aborted(fc);
1211
1212                 down_write(&fc->killsb);
1213                 fc->sb = NULL;
1214                 up_write(&fc->killsb);
1215         }
1216 }
1217
1218 static void fuse_kill_sb_anon(struct super_block *sb)
1219 {
1220         fuse_sb_destroy(sb);
1221         kill_anon_super(sb);
1222 }
1223
1224 static struct file_system_type fuse_fs_type = {
1225         .owner          = THIS_MODULE,
1226         .name           = "fuse",
1227         .fs_flags       = FS_HAS_SUBTYPE,
1228         .mount          = fuse_mount,
1229         .kill_sb        = fuse_kill_sb_anon,
1230 };
1231 MODULE_ALIAS_FS("fuse");
1232
1233 #ifdef CONFIG_BLOCK
1234 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1235                            int flags, const char *dev_name,
1236                            void *raw_data)
1237 {
1238         return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1239 }
1240
1241 static void fuse_kill_sb_blk(struct super_block *sb)
1242 {
1243         fuse_sb_destroy(sb);
1244         kill_block_super(sb);
1245 }
1246
1247 static struct file_system_type fuseblk_fs_type = {
1248         .owner          = THIS_MODULE,
1249         .name           = "fuseblk",
1250         .mount          = fuse_mount_blk,
1251         .kill_sb        = fuse_kill_sb_blk,
1252         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1253 };
1254 MODULE_ALIAS_FS("fuseblk");
1255
1256 static inline int register_fuseblk(void)
1257 {
1258         return register_filesystem(&fuseblk_fs_type);
1259 }
1260
1261 static inline void unregister_fuseblk(void)
1262 {
1263         unregister_filesystem(&fuseblk_fs_type);
1264 }
1265 #else
1266 static inline int register_fuseblk(void)
1267 {
1268         return 0;
1269 }
1270
1271 static inline void unregister_fuseblk(void)
1272 {
1273 }
1274 #endif
1275
1276 static void fuse_inode_init_once(void *foo)
1277 {
1278         struct inode *inode = foo;
1279
1280         inode_init_once(inode);
1281 }
1282
1283 static int __init fuse_fs_init(void)
1284 {
1285         int err;
1286
1287         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1288                                               sizeof(struct fuse_inode), 0,
1289                                               SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT,
1290                                               fuse_inode_init_once);
1291         err = -ENOMEM;
1292         if (!fuse_inode_cachep)
1293                 goto out;
1294
1295         err = register_fuseblk();
1296         if (err)
1297                 goto out2;
1298
1299         err = register_filesystem(&fuse_fs_type);
1300         if (err)
1301                 goto out3;
1302
1303         return 0;
1304
1305  out3:
1306         unregister_fuseblk();
1307  out2:
1308         kmem_cache_destroy(fuse_inode_cachep);
1309  out:
1310         return err;
1311 }
1312
1313 static void fuse_fs_cleanup(void)
1314 {
1315         unregister_filesystem(&fuse_fs_type);
1316         unregister_fuseblk();
1317
1318         /*
1319          * Make sure all delayed rcu free inodes are flushed before we
1320          * destroy cache.
1321          */
1322         rcu_barrier();
1323         kmem_cache_destroy(fuse_inode_cachep);
1324 }
1325
1326 static struct kobject *fuse_kobj;
1327
1328 static int fuse_sysfs_init(void)
1329 {
1330         int err;
1331
1332         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1333         if (!fuse_kobj) {
1334                 err = -ENOMEM;
1335                 goto out_err;
1336         }
1337
1338         err = sysfs_create_mount_point(fuse_kobj, "connections");
1339         if (err)
1340                 goto out_fuse_unregister;
1341
1342         return 0;
1343
1344  out_fuse_unregister:
1345         kobject_put(fuse_kobj);
1346  out_err:
1347         return err;
1348 }
1349
1350 static void fuse_sysfs_cleanup(void)
1351 {
1352         sysfs_remove_mount_point(fuse_kobj, "connections");
1353         kobject_put(fuse_kobj);
1354 }
1355
1356 static int __init fuse_init(void)
1357 {
1358         int res;
1359
1360         printk(KERN_INFO "fuse init (API version %i.%i)\n",
1361                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1362
1363         INIT_LIST_HEAD(&fuse_conn_list);
1364         res = fuse_fs_init();
1365         if (res)
1366                 goto err;
1367
1368         res = fuse_dev_init();
1369         if (res)
1370                 goto err_fs_cleanup;
1371
1372         res = fuse_sysfs_init();
1373         if (res)
1374                 goto err_dev_cleanup;
1375
1376         res = fuse_ctl_init();
1377         if (res)
1378                 goto err_sysfs_cleanup;
1379
1380         sanitize_global_limit(&max_user_bgreq);
1381         sanitize_global_limit(&max_user_congthresh);
1382
1383         return 0;
1384
1385  err_sysfs_cleanup:
1386         fuse_sysfs_cleanup();
1387  err_dev_cleanup:
1388         fuse_dev_cleanup();
1389  err_fs_cleanup:
1390         fuse_fs_cleanup();
1391  err:
1392         return res;
1393 }
1394
1395 static void __exit fuse_exit(void)
1396 {
1397         printk(KERN_DEBUG "fuse exit\n");
1398
1399         fuse_ctl_cleanup();
1400         fuse_sysfs_cleanup();
1401         fuse_fs_cleanup();
1402         fuse_dev_cleanup();
1403 }
1404
1405 module_init(fuse_init);
1406 module_exit(fuse_exit);