fs/fuse/inode.c

   1 /*
   2   FUSE: Filesystem in Userspace
   3   Copyright (C) 2001-2008  Miklos Szeredi <miklos@szeredi.hu>
   4
   5   This program can be distributed under the terms of the GNU GPL.
   6   See the file COPYING.
   7 */
   8
   9 #include "fuse_i.h"
  10
  11 #include <linux/pagemap.h>
  12 #include <linux/slab.h>
  13 #include <linux/file.h>
  14 #include <linux/seq_file.h>
  15 #include <linux/init.h>
  16 #include <linux/module.h>
  17 #include <linux/moduleparam.h>
  18 #include <linux/parser.h>
  19 #include <linux/statfs.h>
  20 #include <linux/random.h>
  21 #include <linux/sched.h>
  22 #include <linux/exportfs.h>
  23 #include <linux/posix_acl.h>
  24 #include <linux/pid_namespace.h>
  25
  26 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  27 MODULE_DESCRIPTION("Filesystem in Userspace");
  28 MODULE_LICENSE("GPL");
  29
  30 static struct kmem_cache *fuse_inode_cachep;
  31 struct list_head fuse_conn_list;
  32 DEFINE_MUTEX(fuse_mutex);
  33
  34 static int set_global_limit(const char *val, const struct kernel_param *kp);
  35
  36 unsigned max_user_bgreq;
  37 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
  38                   &max_user_bgreq, 0644);
  39 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
  40 MODULE_PARM_DESC(max_user_bgreq,
  41  "Global limit for the maximum number of backgrounded requests an "
  42  "unprivileged user can set");
  43
  44 unsigned max_user_congthresh;
  45 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
  46                   &max_user_congthresh, 0644);
  47 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
  48 MODULE_PARM_DESC(max_user_congthresh,
  49  "Global limit for the maximum congestion threshold an "
  50  "unprivileged user can set");
  51
  52 #define FUSE_SUPER_MAGIC 0x65735546
  53
  54 #define FUSE_DEFAULT_BLKSIZE 512
  55
  56 /** Maximum number of outstanding background requests */
  57 #define FUSE_DEFAULT_MAX_BACKGROUND 12
  58
  59 /** Congestion starts at 75% of maximum */
  60 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
  61
  62 struct fuse_mount_data {
  63         int fd;
  64         unsigned rootmode;
  65         kuid_t user_id;
  66         kgid_t group_id;
  67         unsigned fd_present:1;
  68         unsigned rootmode_present:1;
  69         unsigned user_id_present:1;
  70         unsigned group_id_present:1;
  71         unsigned default_permissions:1;
  72         unsigned allow_other:1;
  73         unsigned max_read;
  74         unsigned blksize;
  75 };
  76
  77 struct fuse_forget_link *fuse_alloc_forget(void)
  78 {
  79         return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL);
  80 }
  81
  82 static struct inode *fuse_alloc_inode(struct super_block *sb)
  83 {
  84         struct inode *inode;
  85         struct fuse_inode *fi;
  86
  87         inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
  88         if (!inode)
  89                 return NULL;
  90
  91         fi = get_fuse_inode(inode);
  92         fi->i_time = 0;
  93         fi->nodeid = 0;
  94         fi->nlookup = 0;
  95         fi->attr_version = 0;
  96         fi->writectr = 0;
  97         fi->orig_ino = 0;
  98         fi->state = 0;
  99         INIT_LIST_HEAD(&fi->write_files);
 100         INIT_LIST_HEAD(&fi->queued_writes);
 101         INIT_LIST_HEAD(&fi->writepages);
 102         init_waitqueue_head(&fi->page_waitq);
 103         mutex_init(&fi->mutex);
 104         fi->forget = fuse_alloc_forget();
 105         if (!fi->forget) {
 106                 kmem_cache_free(fuse_inode_cachep, inode);
 107                 return NULL;
 108         }
 109
 110         return inode;
 111 }
 112
 113 static void fuse_i_callback(struct rcu_head *head)
 114 {
 115         struct inode *inode = container_of(head, struct inode, i_rcu);
 116         kmem_cache_free(fuse_inode_cachep, inode);
 117 }
 118
 119 static void fuse_destroy_inode(struct inode *inode)
 120 {
 121         struct fuse_inode *fi = get_fuse_inode(inode);
 122         BUG_ON(!list_empty(&fi->write_files));
 123         BUG_ON(!list_empty(&fi->queued_writes));
 124         mutex_destroy(&fi->mutex);
 125         kfree(fi->forget);
 126         call_rcu(&inode->i_rcu, fuse_i_callback);
 127 }
 128
 129 static void fuse_evict_inode(struct inode *inode)
 130 {
 131         truncate_inode_pages_final(&inode->i_data);
 132         clear_inode(inode);
 133         if (inode->i_sb->s_flags & 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         inode->i_atime.tv_sec   = attr->atime;
 178         inode->i_atime.tv_nsec  = attr->atimensec;
 179         /* mtime from server may be stale due to local buffered write */
 180         if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
 181                 inode->i_mtime.tv_sec   = attr->mtime;
 182                 inode->i_mtime.tv_nsec  = attr->mtimensec;
 183                 inode->i_ctime.tv_sec   = attr->ctime;
 184                 inode->i_ctime.tv_nsec  = attr->ctimensec;
 185         }
 186
 187         if (attr->blksize != 0)
 188                 inode->i_blkbits = ilog2(attr->blksize);
 189         else
 190                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
 191
 192         /*
 193          * Don't set the sticky bit in i_mode, unless we want the VFS
 194          * to check permissions.  This prevents failures due to the
 195          * check in may_delete().
 196          */
 197         fi->orig_i_mode = inode->i_mode;
 198         if (!fc->default_permissions)
 199                 inode->i_mode &= ~S_ISVTX;
 200
 201         fi->orig_ino = attr->ino;
 202 }
 203
 204 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
 205                             u64 attr_valid, u64 attr_version)
 206 {
 207         struct fuse_conn *fc = get_fuse_conn(inode);
 208         struct fuse_inode *fi = get_fuse_inode(inode);
 209         bool is_wb = fc->writeback_cache;
 210         loff_t oldsize;
 211         struct timespec old_mtime;
 212
 213         spin_lock(&fc->lock);
 214         if ((attr_version != 0 && fi->attr_version > attr_version) ||
 215             test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
 216                 spin_unlock(&fc->lock);
 217                 return;
 218         }
 219
 220         old_mtime = inode->i_mtime;
 221         fuse_change_attributes_common(inode, attr, attr_valid);
 222
 223         oldsize = inode->i_size;
 224         /*
 225          * In case of writeback_cache enabled, the cached writes beyond EOF
 226          * extend local i_size without keeping userspace server in sync. So,
 227          * attr->size coming from server can be stale. We cannot trust it.
 228          */
 229         if (!is_wb || !S_ISREG(inode->i_mode))
 230                 i_size_write(inode, attr->size);
 231         spin_unlock(&fc->lock);
 232
 233         if (!is_wb && S_ISREG(inode->i_mode)) {
 234                 bool inval = false;
 235
 236                 if (oldsize != attr->size) {
 237                         truncate_pagecache(inode, attr->size);
 238                         inval = true;
 239                 } else if (fc->auto_inval_data) {
 240                         struct timespec new_mtime = {
 241                                 .tv_sec = attr->mtime,
 242                                 .tv_nsec = attr->mtimensec,
 243                         };
 244
 245                         /*
 246                          * Auto inval mode also checks and invalidates if mtime
 247                          * has changed.
 248                          */
 249                         if (!timespec_equal(&old_mtime, &new_mtime))
 250                                 inval = true;
 251                 }
 252
 253                 if (inval)
 254                         invalidate_inode_pages2(inode->i_mapping);
 255         }
 256 }
 257
 258 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
 259 {
 260         inode->i_mode = attr->mode & S_IFMT;
 261         inode->i_size = attr->size;
 262         inode->i_mtime.tv_sec  = attr->mtime;
 263         inode->i_mtime.tv_nsec = attr->mtimensec;
 264         inode->i_ctime.tv_sec  = attr->ctime;
 265         inode->i_ctime.tv_nsec = attr->ctimensec;
 266         if (S_ISREG(inode->i_mode)) {
 267                 fuse_init_common(inode);
 268                 fuse_init_file_inode(inode);
 269         } else if (S_ISDIR(inode->i_mode))
 270                 fuse_init_dir(inode);
 271         else if (S_ISLNK(inode->i_mode))
 272                 fuse_init_symlink(inode);
 273         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
 274                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
 275                 fuse_init_common(inode);
 276                 init_special_inode(inode, inode->i_mode,
 277                                    new_decode_dev(attr->rdev));
 278         } else
 279                 BUG();
 280 }
 281
 282 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
 283 {
 284         u64 nodeid = *(u64 *) _nodeidp;
 285         if (get_node_id(inode) == nodeid)
 286                 return 1;
 287         else
 288                 return 0;
 289 }
 290
 291 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
 292 {
 293         u64 nodeid = *(u64 *) _nodeidp;
 294         get_fuse_inode(inode)->nodeid = nodeid;
 295         return 0;
 296 }
 297
 298 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
 299                         int generation, struct fuse_attr *attr,
 300                         u64 attr_valid, u64 attr_version)
 301 {
 302         struct inode *inode;
 303         struct fuse_inode *fi;
 304         struct fuse_conn *fc = get_fuse_conn_super(sb);
 305
 306  retry:
 307         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
 308         if (!inode)
 309                 return NULL;
 310
 311         if ((inode->i_state & I_NEW)) {
 312                 inode->i_flags |= S_NOATIME;
 313                 if (!fc->writeback_cache || !S_ISREG(attr->mode))
 314                         inode->i_flags |= S_NOCMTIME;
 315                 inode->i_generation = generation;
 316                 fuse_init_inode(inode, attr);
 317                 unlock_new_inode(inode);
 318         } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
 319                 /* Inode has changed type, any I/O on the old should fail */
 320                 make_bad_inode(inode);
 321                 iput(inode);
 322                 goto retry;
 323         }
 324
 325         fi = get_fuse_inode(inode);
 326         spin_lock(&fc->lock);
 327         fi->nlookup++;
 328         spin_unlock(&fc->lock);
 329         fuse_change_attributes(inode, attr, attr_valid, attr_version);
 330
 331         return inode;
 332 }
 333
 334 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
 335                              loff_t offset, loff_t len)
 336 {
 337         struct inode *inode;
 338         pgoff_t pg_start;
 339         pgoff_t pg_end;
 340
 341         inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
 342         if (!inode)
 343                 return -ENOENT;
 344
 345         fuse_invalidate_attr(inode);
 346         forget_all_cached_acls(inode);
 347         if (offset >= 0) {
 348                 pg_start = offset >> PAGE_SHIFT;
 349                 if (len <= 0)
 350                         pg_end = -1;
 351                 else
 352                         pg_end = (offset + len - 1) >> PAGE_SHIFT;
 353                 invalidate_inode_pages2_range(inode->i_mapping,
 354                                               pg_start, pg_end);
 355         }
 356         iput(inode);
 357         return 0;
 358 }
 359
 360 void fuse_lock_inode(struct inode *inode)
 361 {
 362         if (!get_fuse_conn(inode)->parallel_dirops)
 363                 mutex_lock(&get_fuse_inode(inode)->mutex);
 364 }
 365
 366 void fuse_unlock_inode(struct inode *inode)
 367 {
 368         if (!get_fuse_conn(inode)->parallel_dirops)
 369                 mutex_unlock(&get_fuse_inode(inode)->mutex);
 370 }
 371
 372 static void fuse_umount_begin(struct super_block *sb)
 373 {
 374         fuse_abort_conn(get_fuse_conn_super(sb));
 375 }
 376
 377 static void fuse_send_destroy(struct fuse_conn *fc)
 378 {
 379         struct fuse_req *req = fc->destroy_req;
 380         if (req && fc->conn_init) {
 381                 fc->destroy_req = NULL;
 382                 req->in.h.opcode = FUSE_DESTROY;
 383                 __set_bit(FR_FORCE, &req->flags);
 384                 __clear_bit(FR_BACKGROUND, &req->flags);
 385                 fuse_request_send(fc, req);
 386                 fuse_put_request(fc, req);
 387         }
 388 }
 389
 390 static void fuse_put_super(struct super_block *sb)
 391 {
 392         struct fuse_conn *fc = get_fuse_conn_super(sb);
 393
 394         fuse_send_destroy(fc);
 395
 396         fuse_abort_conn(fc);
 397         mutex_lock(&fuse_mutex);
 398         list_del(&fc->entry);
 399         fuse_ctl_remove_conn(fc);
 400         mutex_unlock(&fuse_mutex);
 401
 402         fuse_conn_put(fc);
 403 }
 404
 405 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
 406 {
 407         stbuf->f_type    = FUSE_SUPER_MAGIC;
 408         stbuf->f_bsize   = attr->bsize;
 409         stbuf->f_frsize  = attr->frsize;
 410         stbuf->f_blocks  = attr->blocks;
 411         stbuf->f_bfree   = attr->bfree;
 412         stbuf->f_bavail  = attr->bavail;
 413         stbuf->f_files   = attr->files;
 414         stbuf->f_ffree   = attr->ffree;
 415         stbuf->f_namelen = attr->namelen;
 416         /* fsid is left zero */
 417 }
 418
 419 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
 420 {
 421         struct super_block *sb = dentry->d_sb;
 422         struct fuse_conn *fc = get_fuse_conn_super(sb);
 423         FUSE_ARGS(args);
 424         struct fuse_statfs_out outarg;
 425         int err;
 426
 427         if (!fuse_allow_current_process(fc)) {
 428                 buf->f_type = FUSE_SUPER_MAGIC;
 429                 return 0;
 430         }
 431
 432         memset(&outarg, 0, sizeof(outarg));
 433         args.in.numargs = 0;
 434         args.in.h.opcode = FUSE_STATFS;
 435         args.in.h.nodeid = get_node_id(d_inode(dentry));
 436         args.out.numargs = 1;
 437         args.out.args[0].size = sizeof(outarg);
 438         args.out.args[0].value = &outarg;
 439         err = fuse_simple_request(fc, &args);
 440         if (!err)
 441                 convert_fuse_statfs(buf, &outarg.st);
 442         return err;
 443 }
 444
 445 enum {
 446         OPT_FD,
 447         OPT_ROOTMODE,
 448         OPT_USER_ID,
 449         OPT_GROUP_ID,
 450         OPT_DEFAULT_PERMISSIONS,
 451         OPT_ALLOW_OTHER,
 452         OPT_MAX_READ,
 453         OPT_BLKSIZE,
 454         OPT_ERR
 455 };
 456
 457 static const match_table_t tokens = {
 458         {OPT_FD,                        "fd=%u"},
 459         {OPT_ROOTMODE,                  "rootmode=%o"},
 460         {OPT_USER_ID,                   "user_id=%u"},
 461         {OPT_GROUP_ID,                  "group_id=%u"},
 462         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 463         {OPT_ALLOW_OTHER,               "allow_other"},
 464         {OPT_MAX_READ,                  "max_read=%u"},
 465         {OPT_BLKSIZE,                   "blksize=%u"},
 466         {OPT_ERR,                       NULL}
 467 };
 468
 469 static int fuse_match_uint(substring_t *s, unsigned int *res)
 470 {
 471         int err = -ENOMEM;
 472         char *buf = match_strdup(s);
 473         if (buf) {
 474                 err = kstrtouint(buf, 10, res);
 475                 kfree(buf);
 476         }
 477         return err;
 478 }
 479
 480 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
 481 {
 482         char *p;
 483         memset(d, 0, sizeof(struct fuse_mount_data));
 484         d->max_read = ~0;
 485         d->blksize = FUSE_DEFAULT_BLKSIZE;
 486
 487         while ((p = strsep(&opt, ",")) != NULL) {
 488                 int token;
 489                 int value;
 490                 unsigned uv;
 491                 substring_t args[MAX_OPT_ARGS];
 492                 if (!*p)
 493                         continue;
 494
 495                 token = match_token(p, tokens, args);
 496                 switch (token) {
 497                 case OPT_FD:
 498                         if (match_int(&args[0], &value))
 499                                 return 0;
 500                         d->fd = value;
 501                         d->fd_present = 1;
 502                         break;
 503
 504                 case OPT_ROOTMODE:
 505                         if (match_octal(&args[0], &value))
 506                                 return 0;
 507                         if (!fuse_valid_type(value))
 508                                 return 0;
 509                         d->rootmode = value;
 510                         d->rootmode_present = 1;
 511                         break;
 512
 513                 case OPT_USER_ID:
 514                         if (fuse_match_uint(&args[0], &uv))
 515                                 return 0;
 516                         d->user_id = make_kuid(current_user_ns(), uv);
 517                         if (!uid_valid(d->user_id))
 518                                 return 0;
 519                         d->user_id_present = 1;
 520                         break;
 521
 522                 case OPT_GROUP_ID:
 523                         if (fuse_match_uint(&args[0], &uv))
 524                                 return 0;
 525                         d->group_id = make_kgid(current_user_ns(), uv);
 526                         if (!gid_valid(d->group_id))
 527                                 return 0;
 528                         d->group_id_present = 1;
 529                         break;
 530
 531                 case OPT_DEFAULT_PERMISSIONS:
 532                         d->default_permissions = 1;
 533                         break;
 534
 535                 case OPT_ALLOW_OTHER:
 536                         d->allow_other = 1;
 537                         break;
 538
 539                 case OPT_MAX_READ:
 540                         if (match_int(&args[0], &value))
 541                                 return 0;
 542                         d->max_read = value;
 543                         break;
 544
 545                 case OPT_BLKSIZE:
 546                         if (!is_bdev || match_int(&args[0], &value))
 547                                 return 0;
 548                         d->blksize = value;
 549                         break;
 550
 551                 default:
 552                         return 0;
 553                 }
 554         }
 555
 556         if (!d->fd_present || !d->rootmode_present ||
 557             !d->user_id_present || !d->group_id_present)
 558                 return 0;
 559
 560         return 1;
 561 }
 562
 563 static int fuse_show_options(struct seq_file *m, struct dentry *root)
 564 {
 565         struct super_block *sb = root->d_sb;
 566         struct fuse_conn *fc = get_fuse_conn_super(sb);
 567
 568         seq_printf(m, ",user_id=%u", from_kuid_munged(&init_user_ns, fc->user_id));
 569         seq_printf(m, ",group_id=%u", from_kgid_munged(&init_user_ns, fc->group_id));
 570         if (fc->default_permissions)
 571                 seq_puts(m, ",default_permissions");
 572         if (fc->allow_other)
 573                 seq_puts(m, ",allow_other");
 574         if (fc->max_read != ~0)
 575                 seq_printf(m, ",max_read=%u", fc->max_read);
 576         if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
 577                 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
 578         return 0;
 579 }
 580
 581 static void fuse_iqueue_init(struct fuse_iqueue *fiq)
 582 {
 583         memset(fiq, 0, sizeof(struct fuse_iqueue));
 584         init_waitqueue_head(&fiq->waitq);
 585         INIT_LIST_HEAD(&fiq->pending);
 586         INIT_LIST_HEAD(&fiq->interrupts);
 587         fiq->forget_list_tail = &fiq->forget_list_head;
 588         fiq->connected = 1;
 589 }
 590
 591 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
 592 {
 593         memset(fpq, 0, sizeof(struct fuse_pqueue));
 594         spin_lock_init(&fpq->lock);
 595         INIT_LIST_HEAD(&fpq->processing);
 596         INIT_LIST_HEAD(&fpq->io);
 597         fpq->connected = 1;
 598 }
 599
 600 void fuse_conn_init(struct fuse_conn *fc)
 601 {
 602         memset(fc, 0, sizeof(*fc));
 603         spin_lock_init(&fc->lock);
 604         init_rwsem(&fc->killsb);
 605         refcount_set(&fc->count, 1);
 606         atomic_set(&fc->dev_count, 1);
 607         init_waitqueue_head(&fc->blocked_waitq);
 608         init_waitqueue_head(&fc->reserved_req_waitq);
 609         fuse_iqueue_init(&fc->iq);
 610         INIT_LIST_HEAD(&fc->bg_queue);
 611         INIT_LIST_HEAD(&fc->entry);
 612         INIT_LIST_HEAD(&fc->devices);
 613         atomic_set(&fc->num_waiting, 0);
 614         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
 615         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
 616         fc->khctr = 0;
 617         fc->polled_files = RB_ROOT;
 618         fc->blocked = 0;
 619         fc->initialized = 0;
 620         fc->connected = 1;
 621         fc->attr_version = 1;
 622         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
 623         fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
 624 }
 625 EXPORT_SYMBOL_GPL(fuse_conn_init);
 626
 627 void fuse_conn_put(struct fuse_conn *fc)
 628 {
 629         if (refcount_dec_and_test(&fc->count)) {
 630                 if (fc->destroy_req)
 631                         fuse_request_free(fc->destroy_req);
 632                 put_pid_ns(fc->pid_ns);
 633                 fc->release(fc);
 634         }
 635 }
 636 EXPORT_SYMBOL_GPL(fuse_conn_put);
 637
 638 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
 639 {
 640         refcount_inc(&fc->count);
 641         return fc;
 642 }
 643 EXPORT_SYMBOL_GPL(fuse_conn_get);
 644
 645 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
 646 {
 647         struct fuse_attr attr;
 648         memset(&attr, 0, sizeof(attr));
 649
 650         attr.mode = mode;
 651         attr.ino = FUSE_ROOT_ID;
 652         attr.nlink = 1;
 653         return fuse_iget(sb, 1, 0, &attr, 0, 0);
 654 }
 655
 656 struct fuse_inode_handle {
 657         u64 nodeid;
 658         u32 generation;
 659 };
 660
 661 static struct dentry *fuse_get_dentry(struct super_block *sb,
 662                                       struct fuse_inode_handle *handle)
 663 {
 664         struct fuse_conn *fc = get_fuse_conn_super(sb);
 665         struct inode *inode;
 666         struct dentry *entry;
 667         int err = -ESTALE;
 668
 669         if (handle->nodeid == 0)
 670                 goto out_err;
 671
 672         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
 673         if (!inode) {
 674                 struct fuse_entry_out outarg;
 675                 const struct qstr name = QSTR_INIT(".", 1);
 676
 677                 if (!fc->export_support)
 678                         goto out_err;
 679
 680                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
 681                                        &inode);
 682                 if (err && err != -ENOENT)
 683                         goto out_err;
 684                 if (err || !inode) {
 685                         err = -ESTALE;
 686                         goto out_err;
 687                 }
 688                 err = -EIO;
 689                 if (get_node_id(inode) != handle->nodeid)
 690                         goto out_iput;
 691         }
 692         err = -ESTALE;
 693         if (inode->i_generation != handle->generation)
 694                 goto out_iput;
 695
 696         entry = d_obtain_alias(inode);
 697         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
 698                 fuse_invalidate_entry_cache(entry);
 699
 700         return entry;
 701
 702  out_iput:
 703         iput(inode);
 704  out_err:
 705         return ERR_PTR(err);
 706 }
 707
 708 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
 709                            struct inode *parent)
 710 {
 711         int len = parent ? 6 : 3;
 712         u64 nodeid;
 713         u32 generation;
 714
 715         if (*max_len < len) {
 716                 *max_len = len;
 717                 return  FILEID_INVALID;
 718         }
 719
 720         nodeid = get_fuse_inode(inode)->nodeid;
 721         generation = inode->i_generation;
 722
 723         fh[0] = (u32)(nodeid >> 32);
 724         fh[1] = (u32)(nodeid & 0xffffffff);
 725         fh[2] = generation;
 726
 727         if (parent) {
 728                 nodeid = get_fuse_inode(parent)->nodeid;
 729                 generation = parent->i_generation;
 730
 731                 fh[3] = (u32)(nodeid >> 32);
 732                 fh[4] = (u32)(nodeid & 0xffffffff);
 733                 fh[5] = generation;
 734         }
 735
 736         *max_len = len;
 737         return parent ? 0x82 : 0x81;
 738 }
 739
 740 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
 741                 struct fid *fid, int fh_len, int fh_type)
 742 {
 743         struct fuse_inode_handle handle;
 744
 745         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
 746                 return NULL;
 747
 748         handle.nodeid = (u64) fid->raw[0] << 32;
 749         handle.nodeid |= (u64) fid->raw[1];
 750         handle.generation = fid->raw[2];
 751         return fuse_get_dentry(sb, &handle);
 752 }
 753
 754 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
 755                 struct fid *fid, int fh_len, int fh_type)
 756 {
 757         struct fuse_inode_handle parent;
 758
 759         if (fh_type != 0x82 || fh_len < 6)
 760                 return NULL;
 761
 762         parent.nodeid = (u64) fid->raw[3] << 32;
 763         parent.nodeid |= (u64) fid->raw[4];
 764         parent.generation = fid->raw[5];
 765         return fuse_get_dentry(sb, &parent);
 766 }
 767
 768 static struct dentry *fuse_get_parent(struct dentry *child)
 769 {
 770         struct inode *child_inode = d_inode(child);
 771         struct fuse_conn *fc = get_fuse_conn(child_inode);
 772         struct inode *inode;
 773         struct dentry *parent;
 774         struct fuse_entry_out outarg;
 775         const struct qstr name = QSTR_INIT("..", 2);
 776         int err;
 777
 778         if (!fc->export_support)
 779                 return ERR_PTR(-ESTALE);
 780
 781         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
 782                                &name, &outarg, &inode);
 783         if (err) {
 784                 if (err == -ENOENT)
 785                         return ERR_PTR(-ESTALE);
 786                 return ERR_PTR(err);
 787         }
 788
 789         parent = d_obtain_alias(inode);
 790         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
 791                 fuse_invalidate_entry_cache(parent);
 792
 793         return parent;
 794 }
 795
 796 static const struct export_operations fuse_export_operations = {
 797         .fh_to_dentry   = fuse_fh_to_dentry,
 798         .fh_to_parent   = fuse_fh_to_parent,
 799         .encode_fh      = fuse_encode_fh,
 800         .get_parent     = fuse_get_parent,
 801 };
 802
 803 static const struct super_operations fuse_super_operations = {
 804         .alloc_inode    = fuse_alloc_inode,
 805         .destroy_inode  = fuse_destroy_inode,
 806         .evict_inode    = fuse_evict_inode,
 807         .write_inode    = fuse_write_inode,
 808         .drop_inode     = generic_delete_inode,
 809         .remount_fs     = fuse_remount_fs,
 810         .put_super      = fuse_put_super,
 811         .umount_begin   = fuse_umount_begin,
 812         .statfs         = fuse_statfs,
 813         .show_options   = fuse_show_options,
 814 };
 815
 816 static void sanitize_global_limit(unsigned *limit)
 817 {
 818         if (*limit == 0)
 819                 *limit = ((totalram_pages << PAGE_SHIFT) >> 13) /
 820                          sizeof(struct fuse_req);
 821
 822         if (*limit >= 1 << 16)
 823                 *limit = (1 << 16) - 1;
 824 }
 825
 826 static int set_global_limit(const char *val, const struct kernel_param *kp)
 827 {
 828         int rv;
 829
 830         rv = param_set_uint(val, kp);
 831         if (rv)
 832                 return rv;
 833
 834         sanitize_global_limit((unsigned *)kp->arg);
 835
 836         return 0;
 837 }
 838
 839 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
 840 {
 841         int cap_sys_admin = capable(CAP_SYS_ADMIN);
 842
 843         if (arg->minor < 13)
 844                 return;
 845
 846         sanitize_global_limit(&max_user_bgreq);
 847         sanitize_global_limit(&max_user_congthresh);
 848
 849         if (arg->max_background) {
 850                 fc->max_background = arg->max_background;
 851
 852                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
 853                         fc->max_background = max_user_bgreq;
 854         }
 855         if (arg->congestion_threshold) {
 856                 fc->congestion_threshold = arg->congestion_threshold;
 857
 858                 if (!cap_sys_admin &&
 859                     fc->congestion_threshold > max_user_congthresh)
 860                         fc->congestion_threshold = max_user_congthresh;
 861         }
 862 }
 863
 864 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
 865 {
 866         struct fuse_init_out *arg = &req->misc.init_out;
 867
 868         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
 869                 fc->conn_error = 1;
 870         else {
 871                 unsigned long ra_pages;
 872
 873                 process_init_limits(fc, arg);
 874
 875                 if (arg->minor >= 6) {
 876                         ra_pages = arg->max_readahead / PAGE_SIZE;
 877                         if (arg->flags & FUSE_ASYNC_READ)
 878                                 fc->async_read = 1;
 879                         if (!(arg->flags & FUSE_POSIX_LOCKS))
 880                                 fc->no_lock = 1;
 881                         if (arg->minor >= 17) {
 882                                 if (!(arg->flags & FUSE_FLOCK_LOCKS))
 883                                         fc->no_flock = 1;
 884                         } else {
 885                                 if (!(arg->flags & FUSE_POSIX_LOCKS))
 886                                         fc->no_flock = 1;
 887                         }
 888                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
 889                                 fc->atomic_o_trunc = 1;
 890                         if (arg->minor >= 9) {
 891                                 /* LOOKUP has dependency on proto version */
 892                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
 893                                         fc->export_support = 1;
 894                         }
 895                         if (arg->flags & FUSE_BIG_WRITES)
 896                                 fc->big_writes = 1;
 897                         if (arg->flags & FUSE_DONT_MASK)
 898                                 fc->dont_mask = 1;
 899                         if (arg->flags & FUSE_AUTO_INVAL_DATA)
 900                                 fc->auto_inval_data = 1;
 901                         if (arg->flags & FUSE_DO_READDIRPLUS) {
 902                                 fc->do_readdirplus = 1;
 903                                 if (arg->flags & FUSE_READDIRPLUS_AUTO)
 904                                         fc->readdirplus_auto = 1;
 905                         }
 906                         if (arg->flags & FUSE_ASYNC_DIO)
 907                                 fc->async_dio = 1;
 908                         if (arg->flags & FUSE_WRITEBACK_CACHE)
 909                                 fc->writeback_cache = 1;
 910                         if (arg->flags & FUSE_PARALLEL_DIROPS)
 911                                 fc->parallel_dirops = 1;
 912                         if (arg->flags & FUSE_HANDLE_KILLPRIV)
 913                                 fc->handle_killpriv = 1;
 914                         if (arg->time_gran && arg->time_gran <= 1000000000)
 915                                 fc->sb->s_time_gran = arg->time_gran;
 916                         if ((arg->flags & FUSE_POSIX_ACL)) {
 917                                 fc->default_permissions = 1;
 918                                 fc->posix_acl = 1;
 919                                 fc->sb->s_xattr = fuse_acl_xattr_handlers;
 920                         }
 921                 } else {
 922                         ra_pages = fc->max_read / PAGE_SIZE;
 923                         fc->no_lock = 1;
 924                         fc->no_flock = 1;
 925                 }
 926
 927                 fc->sb->s_bdi->ra_pages =
 928                                 min(fc->sb->s_bdi->ra_pages, ra_pages);
 929                 fc->minor = arg->minor;
 930                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
 931                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
 932                 fc->conn_init = 1;
 933         }
 934         fuse_set_initialized(fc);
 935         wake_up_all(&fc->blocked_waitq);
 936 }
 937
 938 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
 939 {
 940         struct fuse_init_in *arg = &req->misc.init_in;
 941
 942         arg->major = FUSE_KERNEL_VERSION;
 943         arg->minor = FUSE_KERNEL_MINOR_VERSION;
 944         arg->max_readahead = fc->sb->s_bdi->ra_pages * PAGE_SIZE;
 945         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
 946                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
 947                 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
 948                 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
 949                 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
 950                 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
 951                 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL;
 952         req->in.h.opcode = FUSE_INIT;
 953         req->in.numargs = 1;
 954         req->in.args[0].size = sizeof(*arg);
 955         req->in.args[0].value = arg;
 956         req->out.numargs = 1;
 957         /* Variable length argument used for backward compatibility
 958            with interface version < 7.5.  Rest of init_out is zeroed
 959            by do_get_request(), so a short reply is not a problem */
 960         req->out.argvar = 1;
 961         req->out.args[0].size = sizeof(struct fuse_init_out);
 962         req->out.args[0].value = &req->misc.init_out;
 963         req->end = process_init_reply;
 964         fuse_request_send_background(fc, req);
 965 }
 966
 967 static void fuse_free_conn(struct fuse_conn *fc)
 968 {
 969         WARN_ON(!list_empty(&fc->devices));
 970         kfree_rcu(fc, rcu);
 971 }
 972
 973 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
 974 {
 975         int err;
 976         char *suffix = "";
 977
 978         if (sb->s_bdev) {
 979                 suffix = "-fuseblk";
 980                 /*
 981                  * sb->s_bdi points to blkdev's bdi however we want to redirect
 982                  * it to our private bdi...
 983                  */
 984                 bdi_put(sb->s_bdi);
 985                 sb->s_bdi = &noop_backing_dev_info;
 986         }
 987         err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
 988                                    MINOR(fc->dev), suffix);
 989         if (err)
 990                 return err;
 991
 992         sb->s_bdi->ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
 993         /* fuse does it's own writeback accounting */
 994         sb->s_bdi->capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
 995
 996         /*
 997          * For a single fuse filesystem use max 1% of dirty +
 998          * writeback threshold.
 999          *
1000          * This gives about 1M of write buffer for memory maps on a
1001          * machine with 1G and 10% dirty_ratio, which should be more
1002          * than enough.
1003          *
1004          * Privileged users can raise it by writing to
1005          *
1006          *    /sys/class/bdi/<bdi>/max_ratio
1007          */
1008         bdi_set_max_ratio(sb->s_bdi, 1);
1009
1010         return 0;
1011 }
1012
1013 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc)
1014 {
1015         struct fuse_dev *fud;
1016
1017         fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1018         if (fud) {
1019                 fud->fc = fuse_conn_get(fc);
1020                 fuse_pqueue_init(&fud->pq);
1021
1022                 spin_lock(&fc->lock);
1023                 list_add_tail(&fud->entry, &fc->devices);
1024                 spin_unlock(&fc->lock);
1025         }
1026
1027         return fud;
1028 }
1029 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1030
1031 void fuse_dev_free(struct fuse_dev *fud)
1032 {
1033         struct fuse_conn *fc = fud->fc;
1034
1035         if (fc) {
1036                 spin_lock(&fc->lock);
1037                 list_del(&fud->entry);
1038                 spin_unlock(&fc->lock);
1039
1040                 fuse_conn_put(fc);
1041         }
1042         kfree(fud);
1043 }
1044 EXPORT_SYMBOL_GPL(fuse_dev_free);
1045
1046 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
1047 {
1048         struct fuse_dev *fud;
1049         struct fuse_conn *fc;
1050         struct inode *root;
1051         struct fuse_mount_data d;
1052         struct file *file;
1053         struct dentry *root_dentry;
1054         struct fuse_req *init_req;
1055         int err;
1056         int is_bdev = sb->s_bdev != NULL;
1057
1058         err = -EINVAL;
1059         if (sb->s_flags & MS_MANDLOCK)
1060                 goto err;
1061
1062         sb->s_flags &= ~(MS_NOSEC | SB_I_VERSION);
1063
1064         if (!parse_fuse_opt(data, &d, is_bdev))
1065                 goto err;
1066
1067         if (is_bdev) {
1068 #ifdef CONFIG_BLOCK
1069                 err = -EINVAL;
1070                 if (!sb_set_blocksize(sb, d.blksize))
1071                         goto err;
1072 #endif
1073         } else {
1074                 sb->s_blocksize = PAGE_SIZE;
1075                 sb->s_blocksize_bits = PAGE_SHIFT;
1076         }
1077         sb->s_magic = FUSE_SUPER_MAGIC;
1078         sb->s_op = &fuse_super_operations;
1079         sb->s_xattr = fuse_xattr_handlers;
1080         sb->s_maxbytes = MAX_LFS_FILESIZE;
1081         sb->s_time_gran = 1;
1082         sb->s_export_op = &fuse_export_operations;
1083
1084         file = fget(d.fd);
1085         err = -EINVAL;
1086         if (!file)
1087                 goto err;
1088
1089         if ((file->f_op != &fuse_dev_operations) ||
1090             (file->f_cred->user_ns != &init_user_ns))
1091                 goto err_fput;
1092
1093         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1094         err = -ENOMEM;
1095         if (!fc)
1096                 goto err_fput;
1097
1098         fuse_conn_init(fc);
1099         fc->release = fuse_free_conn;
1100
1101         fud = fuse_dev_alloc(fc);
1102         if (!fud)
1103                 goto err_put_conn;
1104
1105         fc->dev = sb->s_dev;
1106         fc->sb = sb;
1107         err = fuse_bdi_init(fc, sb);
1108         if (err)
1109                 goto err_dev_free;
1110
1111         /* Handle umasking inside the fuse code */
1112         if (sb->s_flags & MS_POSIXACL)
1113                 fc->dont_mask = 1;
1114         sb->s_flags |= MS_POSIXACL;
1115
1116         fc->default_permissions = d.default_permissions;
1117         fc->allow_other = d.allow_other;
1118         fc->user_id = d.user_id;
1119         fc->group_id = d.group_id;
1120         fc->max_read = max_t(unsigned, 4096, d.max_read);
1121
1122         /* Used by get_root_inode() */
1123         sb->s_fs_info = fc;
1124
1125         err = -ENOMEM;
1126         root = fuse_get_root_inode(sb, d.rootmode);
1127         sb->s_d_op = &fuse_root_dentry_operations;
1128         root_dentry = d_make_root(root);
1129         if (!root_dentry)
1130                 goto err_dev_free;
1131         /* Root dentry doesn't have .d_revalidate */
1132         sb->s_d_op = &fuse_dentry_operations;
1133
1134         init_req = fuse_request_alloc(0);
1135         if (!init_req)
1136                 goto err_put_root;
1137         __set_bit(FR_BACKGROUND, &init_req->flags);
1138
1139         if (is_bdev) {
1140                 fc->destroy_req = fuse_request_alloc(0);
1141                 if (!fc->destroy_req)
1142                         goto err_free_init_req;
1143         }
1144
1145         mutex_lock(&fuse_mutex);
1146         err = -EINVAL;
1147         if (file->private_data)
1148                 goto err_unlock;
1149
1150         err = fuse_ctl_add_conn(fc);
1151         if (err)
1152                 goto err_unlock;
1153
1154         list_add_tail(&fc->entry, &fuse_conn_list);
1155         sb->s_root = root_dentry;
1156         file->private_data = fud;
1157         mutex_unlock(&fuse_mutex);
1158         /*
1159          * atomic_dec_and_test() in fput() provides the necessary
1160          * memory barrier for file->private_data to be visible on all
1161          * CPUs after this
1162          */
1163         fput(file);
1164
1165         fuse_send_init(fc, init_req);
1166
1167         return 0;
1168
1169  err_unlock:
1170         mutex_unlock(&fuse_mutex);
1171  err_free_init_req:
1172         fuse_request_free(init_req);
1173  err_put_root:
1174         dput(root_dentry);
1175  err_dev_free:
1176         fuse_dev_free(fud);
1177  err_put_conn:
1178         fuse_conn_put(fc);
1179  err_fput:
1180         fput(file);
1181  err:
1182         return err;
1183 }
1184
1185 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1186                        int flags, const char *dev_name,
1187                        void *raw_data)
1188 {
1189         return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1190 }
1191
1192 static void fuse_kill_sb_anon(struct super_block *sb)
1193 {
1194         struct fuse_conn *fc = get_fuse_conn_super(sb);
1195
1196         if (fc) {
1197                 down_write(&fc->killsb);
1198                 fc->sb = NULL;
1199                 up_write(&fc->killsb);
1200         }
1201
1202         kill_anon_super(sb);
1203 }
1204
1205 static struct file_system_type fuse_fs_type = {
1206         .owner          = THIS_MODULE,
1207         .name           = "fuse",
1208         .fs_flags       = FS_HAS_SUBTYPE,
1209         .mount          = fuse_mount,
1210         .kill_sb        = fuse_kill_sb_anon,
1211 };
1212 MODULE_ALIAS_FS("fuse");
1213
1214 #ifdef CONFIG_BLOCK
1215 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1216                            int flags, const char *dev_name,
1217                            void *raw_data)
1218 {
1219         return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1220 }
1221
1222 static void fuse_kill_sb_blk(struct super_block *sb)
1223 {
1224         struct fuse_conn *fc = get_fuse_conn_super(sb);
1225
1226         if (fc) {
1227                 down_write(&fc->killsb);
1228                 fc->sb = NULL;
1229                 up_write(&fc->killsb);
1230         }
1231
1232         kill_block_super(sb);
1233 }
1234
1235 static struct file_system_type fuseblk_fs_type = {
1236         .owner          = THIS_MODULE,
1237         .name           = "fuseblk",
1238         .mount          = fuse_mount_blk,
1239         .kill_sb        = fuse_kill_sb_blk,
1240         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1241 };
1242 MODULE_ALIAS_FS("fuseblk");
1243
1244 static inline int register_fuseblk(void)
1245 {
1246         return register_filesystem(&fuseblk_fs_type);
1247 }
1248
1249 static inline void unregister_fuseblk(void)
1250 {
1251         unregister_filesystem(&fuseblk_fs_type);
1252 }
1253 #else
1254 static inline int register_fuseblk(void)
1255 {
1256         return 0;
1257 }
1258
1259 static inline void unregister_fuseblk(void)
1260 {
1261 }
1262 #endif
1263
1264 static void fuse_inode_init_once(void *foo)
1265 {
1266         struct inode *inode = foo;
1267
1268         inode_init_once(inode);
1269 }
1270
1271 static int __init fuse_fs_init(void)
1272 {
1273         int err;
1274
1275         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1276                                               sizeof(struct fuse_inode), 0,
1277                                               SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT,
1278                                               fuse_inode_init_once);
1279         err = -ENOMEM;
1280         if (!fuse_inode_cachep)
1281                 goto out;
1282
1283         err = register_fuseblk();
1284         if (err)
1285                 goto out2;
1286
1287         err = register_filesystem(&fuse_fs_type);
1288         if (err)
1289                 goto out3;
1290
1291         return 0;
1292
1293  out3:
1294         unregister_fuseblk();
1295  out2:
1296         kmem_cache_destroy(fuse_inode_cachep);
1297  out:
1298         return err;
1299 }
1300
1301 static void fuse_fs_cleanup(void)
1302 {
1303         unregister_filesystem(&fuse_fs_type);
1304         unregister_fuseblk();
1305
1306         /*
1307          * Make sure all delayed rcu free inodes are flushed before we
1308          * destroy cache.
1309          */
1310         rcu_barrier();
1311         kmem_cache_destroy(fuse_inode_cachep);
1312 }
1313
1314 static struct kobject *fuse_kobj;
1315
1316 static int fuse_sysfs_init(void)
1317 {
1318         int err;
1319
1320         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1321         if (!fuse_kobj) {
1322                 err = -ENOMEM;
1323                 goto out_err;
1324         }
1325
1326         err = sysfs_create_mount_point(fuse_kobj, "connections");
1327         if (err)
1328                 goto out_fuse_unregister;
1329
1330         return 0;
1331
1332  out_fuse_unregister:
1333         kobject_put(fuse_kobj);
1334  out_err:
1335         return err;
1336 }
1337
1338 static void fuse_sysfs_cleanup(void)
1339 {
1340         sysfs_remove_mount_point(fuse_kobj, "connections");
1341         kobject_put(fuse_kobj);
1342 }
1343
1344 static int __init fuse_init(void)
1345 {
1346         int res;
1347
1348         printk(KERN_INFO "fuse init (API version %i.%i)\n",
1349                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1350
1351         INIT_LIST_HEAD(&fuse_conn_list);
1352         res = fuse_fs_init();
1353         if (res)
1354                 goto err;
1355
1356         res = fuse_dev_init();
1357         if (res)
1358                 goto err_fs_cleanup;
1359
1360         res = fuse_sysfs_init();
1361         if (res)
1362                 goto err_dev_cleanup;
1363
1364         res = fuse_ctl_init();
1365         if (res)
1366                 goto err_sysfs_cleanup;
1367
1368         sanitize_global_limit(&max_user_bgreq);
1369         sanitize_global_limit(&max_user_congthresh);
1370
1371         return 0;
1372
1373  err_sysfs_cleanup:
1374         fuse_sysfs_cleanup();
1375  err_dev_cleanup:
1376         fuse_dev_cleanup();
1377  err_fs_cleanup:
1378         fuse_fs_cleanup();
1379  err:
1380         return res;
1381 }
1382
1383 static void __exit fuse_exit(void)
1384 {
1385         printk(KERN_DEBUG "fuse exit\n");
1386
1387         fuse_ctl_cleanup();
1388         fuse_sysfs_cleanup();
1389         fuse_fs_cleanup();
1390         fuse_dev_cleanup();
1391 }
1392
1393 module_init(fuse_init);
1394 module_exit(fuse_exit);