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         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_bdi_destroy(struct fuse_conn *fc)
 391 {
 392         if (fc->bdi_initialized)
 393                 bdi_destroy(&fc->bdi);
 394 }
 395
 396 static void fuse_put_super(struct super_block *sb)
 397 {
 398         struct fuse_conn *fc = get_fuse_conn_super(sb);
 399
 400         fuse_send_destroy(fc);
 401
 402         fuse_abort_conn(fc);
 403         mutex_lock(&fuse_mutex);
 404         list_del(&fc->entry);
 405         fuse_ctl_remove_conn(fc);
 406         mutex_unlock(&fuse_mutex);
 407         fuse_bdi_destroy(fc);
 408
 409         fuse_conn_put(fc);
 410 }
 411
 412 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
 413 {
 414         stbuf->f_type    = FUSE_SUPER_MAGIC;
 415         stbuf->f_bsize   = attr->bsize;
 416         stbuf->f_frsize  = attr->frsize;
 417         stbuf->f_blocks  = attr->blocks;
 418         stbuf->f_bfree   = attr->bfree;
 419         stbuf->f_bavail  = attr->bavail;
 420         stbuf->f_files   = attr->files;
 421         stbuf->f_ffree   = attr->ffree;
 422         stbuf->f_namelen = attr->namelen;
 423         /* fsid is left zero */
 424 }
 425
 426 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
 427 {
 428         struct super_block *sb = dentry->d_sb;
 429         struct fuse_conn *fc = get_fuse_conn_super(sb);
 430         FUSE_ARGS(args);
 431         struct fuse_statfs_out outarg;
 432         int err;
 433
 434         if (!fuse_allow_current_process(fc)) {
 435                 buf->f_type = FUSE_SUPER_MAGIC;
 436                 return 0;
 437         }
 438
 439         memset(&outarg, 0, sizeof(outarg));
 440         args.in.numargs = 0;
 441         args.in.h.opcode = FUSE_STATFS;
 442         args.in.h.nodeid = get_node_id(d_inode(dentry));
 443         args.out.numargs = 1;
 444         args.out.args[0].size = sizeof(outarg);
 445         args.out.args[0].value = &outarg;
 446         err = fuse_simple_request(fc, &args);
 447         if (!err)
 448                 convert_fuse_statfs(buf, &outarg.st);
 449         return err;
 450 }
 451
 452 enum {
 453         OPT_FD,
 454         OPT_ROOTMODE,
 455         OPT_USER_ID,
 456         OPT_GROUP_ID,
 457         OPT_DEFAULT_PERMISSIONS,
 458         OPT_ALLOW_OTHER,
 459         OPT_MAX_READ,
 460         OPT_BLKSIZE,
 461         OPT_ERR
 462 };
 463
 464 static const match_table_t tokens = {
 465         {OPT_FD,                        "fd=%u"},
 466         {OPT_ROOTMODE,                  "rootmode=%o"},
 467         {OPT_USER_ID,                   "user_id=%u"},
 468         {OPT_GROUP_ID,                  "group_id=%u"},
 469         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 470         {OPT_ALLOW_OTHER,               "allow_other"},
 471         {OPT_MAX_READ,                  "max_read=%u"},
 472         {OPT_BLKSIZE,                   "blksize=%u"},
 473         {OPT_ERR,                       NULL}
 474 };
 475
 476 static int fuse_match_uint(substring_t *s, unsigned int *res)
 477 {
 478         int err = -ENOMEM;
 479         char *buf = match_strdup(s);
 480         if (buf) {
 481                 err = kstrtouint(buf, 10, res);
 482                 kfree(buf);
 483         }
 484         return err;
 485 }
 486
 487 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
 488 {
 489         char *p;
 490         memset(d, 0, sizeof(struct fuse_mount_data));
 491         d->max_read = ~0;
 492         d->blksize = FUSE_DEFAULT_BLKSIZE;
 493
 494         while ((p = strsep(&opt, ",")) != NULL) {
 495                 int token;
 496                 int value;
 497                 unsigned uv;
 498                 substring_t args[MAX_OPT_ARGS];
 499                 if (!*p)
 500                         continue;
 501
 502                 token = match_token(p, tokens, args);
 503                 switch (token) {
 504                 case OPT_FD:
 505                         if (match_int(&args[0], &value))
 506                                 return 0;
 507                         d->fd = value;
 508                         d->fd_present = 1;
 509                         break;
 510
 511                 case OPT_ROOTMODE:
 512                         if (match_octal(&args[0], &value))
 513                                 return 0;
 514                         if (!fuse_valid_type(value))
 515                                 return 0;
 516                         d->rootmode = value;
 517                         d->rootmode_present = 1;
 518                         break;
 519
 520                 case OPT_USER_ID:
 521                         if (fuse_match_uint(&args[0], &uv))
 522                                 return 0;
 523                         d->user_id = make_kuid(current_user_ns(), uv);
 524                         if (!uid_valid(d->user_id))
 525                                 return 0;
 526                         d->user_id_present = 1;
 527                         break;
 528
 529                 case OPT_GROUP_ID:
 530                         if (fuse_match_uint(&args[0], &uv))
 531                                 return 0;
 532                         d->group_id = make_kgid(current_user_ns(), uv);
 533                         if (!gid_valid(d->group_id))
 534                                 return 0;
 535                         d->group_id_present = 1;
 536                         break;
 537
 538                 case OPT_DEFAULT_PERMISSIONS:
 539                         d->default_permissions = 1;
 540                         break;
 541
 542                 case OPT_ALLOW_OTHER:
 543                         d->allow_other = 1;
 544                         break;
 545
 546                 case OPT_MAX_READ:
 547                         if (match_int(&args[0], &value))
 548                                 return 0;
 549                         d->max_read = value;
 550                         break;
 551
 552                 case OPT_BLKSIZE:
 553                         if (!is_bdev || match_int(&args[0], &value))
 554                                 return 0;
 555                         d->blksize = value;
 556                         break;
 557
 558                 default:
 559                         return 0;
 560                 }
 561         }
 562
 563         if (!d->fd_present || !d->rootmode_present ||
 564             !d->user_id_present || !d->group_id_present)
 565                 return 0;
 566
 567         return 1;
 568 }
 569
 570 static int fuse_show_options(struct seq_file *m, struct dentry *root)
 571 {
 572         struct super_block *sb = root->d_sb;
 573         struct fuse_conn *fc = get_fuse_conn_super(sb);
 574
 575         seq_printf(m, ",user_id=%u", from_kuid_munged(&init_user_ns, fc->user_id));
 576         seq_printf(m, ",group_id=%u", from_kgid_munged(&init_user_ns, fc->group_id));
 577         if (fc->default_permissions)
 578                 seq_puts(m, ",default_permissions");
 579         if (fc->allow_other)
 580                 seq_puts(m, ",allow_other");
 581         if (fc->max_read != ~0)
 582                 seq_printf(m, ",max_read=%u", fc->max_read);
 583         if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
 584                 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
 585         return 0;
 586 }
 587
 588 static void fuse_iqueue_init(struct fuse_iqueue *fiq)
 589 {
 590         memset(fiq, 0, sizeof(struct fuse_iqueue));
 591         init_waitqueue_head(&fiq->waitq);
 592         INIT_LIST_HEAD(&fiq->pending);
 593         INIT_LIST_HEAD(&fiq->interrupts);
 594         fiq->forget_list_tail = &fiq->forget_list_head;
 595         fiq->connected = 1;
 596 }
 597
 598 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
 599 {
 600         memset(fpq, 0, sizeof(struct fuse_pqueue));
 601         spin_lock_init(&fpq->lock);
 602         INIT_LIST_HEAD(&fpq->processing);
 603         INIT_LIST_HEAD(&fpq->io);
 604         fpq->connected = 1;
 605 }
 606
 607 void fuse_conn_init(struct fuse_conn *fc)
 608 {
 609         memset(fc, 0, sizeof(*fc));
 610         spin_lock_init(&fc->lock);
 611         init_rwsem(&fc->killsb);
 612         refcount_set(&fc->count, 1);
 613         atomic_set(&fc->dev_count, 1);
 614         init_waitqueue_head(&fc->blocked_waitq);
 615         init_waitqueue_head(&fc->reserved_req_waitq);
 616         fuse_iqueue_init(&fc->iq);
 617         INIT_LIST_HEAD(&fc->bg_queue);
 618         INIT_LIST_HEAD(&fc->entry);
 619         INIT_LIST_HEAD(&fc->devices);
 620         atomic_set(&fc->num_waiting, 0);
 621         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
 622         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
 623         fc->khctr = 0;
 624         fc->polled_files = RB_ROOT;
 625         fc->blocked = 0;
 626         fc->initialized = 0;
 627         fc->connected = 1;
 628         fc->attr_version = 1;
 629         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
 630         fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
 631 }
 632 EXPORT_SYMBOL_GPL(fuse_conn_init);
 633
 634 void fuse_conn_put(struct fuse_conn *fc)
 635 {
 636         if (refcount_dec_and_test(&fc->count)) {
 637                 if (fc->destroy_req)
 638                         fuse_request_free(fc->destroy_req);
 639                 put_pid_ns(fc->pid_ns);
 640                 fc->release(fc);
 641         }
 642 }
 643 EXPORT_SYMBOL_GPL(fuse_conn_put);
 644
 645 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
 646 {
 647         refcount_inc(&fc->count);
 648         return fc;
 649 }
 650 EXPORT_SYMBOL_GPL(fuse_conn_get);
 651
 652 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
 653 {
 654         struct fuse_attr attr;
 655         memset(&attr, 0, sizeof(attr));
 656
 657         attr.mode = mode;
 658         attr.ino = FUSE_ROOT_ID;
 659         attr.nlink = 1;
 660         return fuse_iget(sb, 1, 0, &attr, 0, 0);
 661 }
 662
 663 struct fuse_inode_handle {
 664         u64 nodeid;
 665         u32 generation;
 666 };
 667
 668 static struct dentry *fuse_get_dentry(struct super_block *sb,
 669                                       struct fuse_inode_handle *handle)
 670 {
 671         struct fuse_conn *fc = get_fuse_conn_super(sb);
 672         struct inode *inode;
 673         struct dentry *entry;
 674         int err = -ESTALE;
 675
 676         if (handle->nodeid == 0)
 677                 goto out_err;
 678
 679         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
 680         if (!inode) {
 681                 struct fuse_entry_out outarg;
 682                 const struct qstr name = QSTR_INIT(".", 1);
 683
 684                 if (!fc->export_support)
 685                         goto out_err;
 686
 687                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
 688                                        &inode);
 689                 if (err && err != -ENOENT)
 690                         goto out_err;
 691                 if (err || !inode) {
 692                         err = -ESTALE;
 693                         goto out_err;
 694                 }
 695                 err = -EIO;
 696                 if (get_node_id(inode) != handle->nodeid)
 697                         goto out_iput;
 698         }
 699         err = -ESTALE;
 700         if (inode->i_generation != handle->generation)
 701                 goto out_iput;
 702
 703         entry = d_obtain_alias(inode);
 704         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
 705                 fuse_invalidate_entry_cache(entry);
 706
 707         return entry;
 708
 709  out_iput:
 710         iput(inode);
 711  out_err:
 712         return ERR_PTR(err);
 713 }
 714
 715 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
 716                            struct inode *parent)
 717 {
 718         int len = parent ? 6 : 3;
 719         u64 nodeid;
 720         u32 generation;
 721
 722         if (*max_len < len) {
 723                 *max_len = len;
 724                 return  FILEID_INVALID;
 725         }
 726
 727         nodeid = get_fuse_inode(inode)->nodeid;
 728         generation = inode->i_generation;
 729
 730         fh[0] = (u32)(nodeid >> 32);
 731         fh[1] = (u32)(nodeid & 0xffffffff);
 732         fh[2] = generation;
 733
 734         if (parent) {
 735                 nodeid = get_fuse_inode(parent)->nodeid;
 736                 generation = parent->i_generation;
 737
 738                 fh[3] = (u32)(nodeid >> 32);
 739                 fh[4] = (u32)(nodeid & 0xffffffff);
 740                 fh[5] = generation;
 741         }
 742
 743         *max_len = len;
 744         return parent ? 0x82 : 0x81;
 745 }
 746
 747 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
 748                 struct fid *fid, int fh_len, int fh_type)
 749 {
 750         struct fuse_inode_handle handle;
 751
 752         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
 753                 return NULL;
 754
 755         handle.nodeid = (u64) fid->raw[0] << 32;
 756         handle.nodeid |= (u64) fid->raw[1];
 757         handle.generation = fid->raw[2];
 758         return fuse_get_dentry(sb, &handle);
 759 }
 760
 761 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
 762                 struct fid *fid, int fh_len, int fh_type)
 763 {
 764         struct fuse_inode_handle parent;
 765
 766         if (fh_type != 0x82 || fh_len < 6)
 767                 return NULL;
 768
 769         parent.nodeid = (u64) fid->raw[3] << 32;
 770         parent.nodeid |= (u64) fid->raw[4];
 771         parent.generation = fid->raw[5];
 772         return fuse_get_dentry(sb, &parent);
 773 }
 774
 775 static struct dentry *fuse_get_parent(struct dentry *child)
 776 {
 777         struct inode *child_inode = d_inode(child);
 778         struct fuse_conn *fc = get_fuse_conn(child_inode);
 779         struct inode *inode;
 780         struct dentry *parent;
 781         struct fuse_entry_out outarg;
 782         const struct qstr name = QSTR_INIT("..", 2);
 783         int err;
 784
 785         if (!fc->export_support)
 786                 return ERR_PTR(-ESTALE);
 787
 788         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
 789                                &name, &outarg, &inode);
 790         if (err) {
 791                 if (err == -ENOENT)
 792                         return ERR_PTR(-ESTALE);
 793                 return ERR_PTR(err);
 794         }
 795
 796         parent = d_obtain_alias(inode);
 797         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
 798                 fuse_invalidate_entry_cache(parent);
 799
 800         return parent;
 801 }
 802
 803 static const struct export_operations fuse_export_operations = {
 804         .fh_to_dentry   = fuse_fh_to_dentry,
 805         .fh_to_parent   = fuse_fh_to_parent,
 806         .encode_fh      = fuse_encode_fh,
 807         .get_parent     = fuse_get_parent,
 808 };
 809
 810 static const struct super_operations fuse_super_operations = {
 811         .alloc_inode    = fuse_alloc_inode,
 812         .destroy_inode  = fuse_destroy_inode,
 813         .evict_inode    = fuse_evict_inode,
 814         .write_inode    = fuse_write_inode,
 815         .drop_inode     = generic_delete_inode,
 816         .remount_fs     = fuse_remount_fs,
 817         .put_super      = fuse_put_super,
 818         .umount_begin   = fuse_umount_begin,
 819         .statfs         = fuse_statfs,
 820         .show_options   = fuse_show_options,
 821 };
 822
 823 static void sanitize_global_limit(unsigned *limit)
 824 {
 825         if (*limit == 0)
 826                 *limit = ((totalram_pages << PAGE_SHIFT) >> 13) /
 827                          sizeof(struct fuse_req);
 828
 829         if (*limit >= 1 << 16)
 830                 *limit = (1 << 16) - 1;
 831 }
 832
 833 static int set_global_limit(const char *val, struct kernel_param *kp)
 834 {
 835         int rv;
 836
 837         rv = param_set_uint(val, kp);
 838         if (rv)
 839                 return rv;
 840
 841         sanitize_global_limit((unsigned *)kp->arg);
 842
 843         return 0;
 844 }
 845
 846 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
 847 {
 848         int cap_sys_admin = capable(CAP_SYS_ADMIN);
 849
 850         if (arg->minor < 13)
 851                 return;
 852
 853         sanitize_global_limit(&max_user_bgreq);
 854         sanitize_global_limit(&max_user_congthresh);
 855
 856         if (arg->max_background) {
 857                 fc->max_background = arg->max_background;
 858
 859                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
 860                         fc->max_background = max_user_bgreq;
 861         }
 862         if (arg->congestion_threshold) {
 863                 fc->congestion_threshold = arg->congestion_threshold;
 864
 865                 if (!cap_sys_admin &&
 866                     fc->congestion_threshold > max_user_congthresh)
 867                         fc->congestion_threshold = max_user_congthresh;
 868         }
 869 }
 870
 871 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
 872 {
 873         struct fuse_init_out *arg = &req->misc.init_out;
 874
 875         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
 876                 fc->conn_error = 1;
 877         else {
 878                 unsigned long ra_pages;
 879
 880                 process_init_limits(fc, arg);
 881
 882                 if (arg->minor >= 6) {
 883                         ra_pages = arg->max_readahead / PAGE_SIZE;
 884                         if (arg->flags & FUSE_ASYNC_READ)
 885                                 fc->async_read = 1;
 886                         if (!(arg->flags & FUSE_POSIX_LOCKS))
 887                                 fc->no_lock = 1;
 888                         if (arg->minor >= 17) {
 889                                 if (!(arg->flags & FUSE_FLOCK_LOCKS))
 890                                         fc->no_flock = 1;
 891                         } else {
 892                                 if (!(arg->flags & FUSE_POSIX_LOCKS))
 893                                         fc->no_flock = 1;
 894                         }
 895                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
 896                                 fc->atomic_o_trunc = 1;
 897                         if (arg->minor >= 9) {
 898                                 /* LOOKUP has dependency on proto version */
 899                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
 900                                         fc->export_support = 1;
 901                         }
 902                         if (arg->flags & FUSE_BIG_WRITES)
 903                                 fc->big_writes = 1;
 904                         if (arg->flags & FUSE_DONT_MASK)
 905                                 fc->dont_mask = 1;
 906                         if (arg->flags & FUSE_AUTO_INVAL_DATA)
 907                                 fc->auto_inval_data = 1;
 908                         if (arg->flags & FUSE_DO_READDIRPLUS) {
 909                                 fc->do_readdirplus = 1;
 910                                 if (arg->flags & FUSE_READDIRPLUS_AUTO)
 911                                         fc->readdirplus_auto = 1;
 912                         }
 913                         if (arg->flags & FUSE_ASYNC_DIO)
 914                                 fc->async_dio = 1;
 915                         if (arg->flags & FUSE_WRITEBACK_CACHE)
 916                                 fc->writeback_cache = 1;
 917                         if (arg->flags & FUSE_PARALLEL_DIROPS)
 918                                 fc->parallel_dirops = 1;
 919                         if (arg->flags & FUSE_HANDLE_KILLPRIV)
 920                                 fc->handle_killpriv = 1;
 921                         if (arg->time_gran && arg->time_gran <= 1000000000)
 922                                 fc->sb->s_time_gran = arg->time_gran;
 923                         if ((arg->flags & FUSE_POSIX_ACL)) {
 924                                 fc->default_permissions = 1;
 925                                 fc->posix_acl = 1;
 926                                 fc->sb->s_xattr = fuse_acl_xattr_handlers;
 927                         }
 928                 } else {
 929                         ra_pages = fc->max_read / PAGE_SIZE;
 930                         fc->no_lock = 1;
 931                         fc->no_flock = 1;
 932                 }
 933
 934                 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
 935                 fc->minor = arg->minor;
 936                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
 937                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
 938                 fc->conn_init = 1;
 939         }
 940         fuse_set_initialized(fc);
 941         wake_up_all(&fc->blocked_waitq);
 942 }
 943
 944 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
 945 {
 946         struct fuse_init_in *arg = &req->misc.init_in;
 947
 948         arg->major = FUSE_KERNEL_VERSION;
 949         arg->minor = FUSE_KERNEL_MINOR_VERSION;
 950         arg->max_readahead = fc->bdi.ra_pages * PAGE_SIZE;
 951         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
 952                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
 953                 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
 954                 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
 955                 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
 956                 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
 957                 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL;
 958         req->in.h.opcode = FUSE_INIT;
 959         req->in.numargs = 1;
 960         req->in.args[0].size = sizeof(*arg);
 961         req->in.args[0].value = arg;
 962         req->out.numargs = 1;
 963         /* Variable length argument used for backward compatibility
 964            with interface version < 7.5.  Rest of init_out is zeroed
 965            by do_get_request(), so a short reply is not a problem */
 966         req->out.argvar = 1;
 967         req->out.args[0].size = sizeof(struct fuse_init_out);
 968         req->out.args[0].value = &req->misc.init_out;
 969         req->end = process_init_reply;
 970         fuse_request_send_background(fc, req);
 971 }
 972
 973 static void fuse_free_conn(struct fuse_conn *fc)
 974 {
 975         WARN_ON(!list_empty(&fc->devices));
 976         kfree_rcu(fc, rcu);
 977 }
 978
 979 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
 980 {
 981         int err;
 982
 983         fc->bdi.name = "fuse";
 984         fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
 985         /* fuse does it's own writeback accounting */
 986         fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
 987
 988         err = bdi_init(&fc->bdi);
 989         if (err)
 990                 return err;
 991
 992         fc->bdi_initialized = 1;
 993
 994         if (sb->s_bdev) {
 995                 err =  bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
 996                                     MAJOR(fc->dev), MINOR(fc->dev));
 997         } else {
 998                 err = bdi_register_dev(&fc->bdi, fc->dev);
 999         }
1000
1001         if (err)
1002                 return err;
1003
1004         /*
1005          * For a single fuse filesystem use max 1% of dirty +
1006          * writeback threshold.
1007          *
1008          * This gives about 1M of write buffer for memory maps on a
1009          * machine with 1G and 10% dirty_ratio, which should be more
1010          * than enough.
1011          *
1012          * Privileged users can raise it by writing to
1013          *
1014          *    /sys/class/bdi/<bdi>/max_ratio
1015          */
1016         bdi_set_max_ratio(&fc->bdi, 1);
1017
1018         return 0;
1019 }
1020
1021 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc)
1022 {
1023         struct fuse_dev *fud;
1024
1025         fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1026         if (fud) {
1027                 fud->fc = fuse_conn_get(fc);
1028                 fuse_pqueue_init(&fud->pq);
1029
1030                 spin_lock(&fc->lock);
1031                 list_add_tail(&fud->entry, &fc->devices);
1032                 spin_unlock(&fc->lock);
1033         }
1034
1035         return fud;
1036 }
1037 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1038
1039 void fuse_dev_free(struct fuse_dev *fud)
1040 {
1041         struct fuse_conn *fc = fud->fc;
1042
1043         if (fc) {
1044                 spin_lock(&fc->lock);
1045                 list_del(&fud->entry);
1046                 spin_unlock(&fc->lock);
1047
1048                 fuse_conn_put(fc);
1049         }
1050         kfree(fud);
1051 }
1052 EXPORT_SYMBOL_GPL(fuse_dev_free);
1053
1054 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
1055 {
1056         struct fuse_dev *fud;
1057         struct fuse_conn *fc;
1058         struct inode *root;
1059         struct fuse_mount_data d;
1060         struct file *file;
1061         struct dentry *root_dentry;
1062         struct fuse_req *init_req;
1063         int err;
1064         int is_bdev = sb->s_bdev != NULL;
1065
1066         err = -EINVAL;
1067         if (sb->s_flags & MS_MANDLOCK)
1068                 goto err;
1069
1070         sb->s_flags &= ~(MS_NOSEC | MS_I_VERSION);
1071
1072         if (!parse_fuse_opt(data, &d, is_bdev))
1073                 goto err;
1074
1075         if (is_bdev) {
1076 #ifdef CONFIG_BLOCK
1077                 err = -EINVAL;
1078                 if (!sb_set_blocksize(sb, d.blksize))
1079                         goto err;
1080 #endif
1081         } else {
1082                 sb->s_blocksize = PAGE_SIZE;
1083                 sb->s_blocksize_bits = PAGE_SHIFT;
1084         }
1085         sb->s_magic = FUSE_SUPER_MAGIC;
1086         sb->s_op = &fuse_super_operations;
1087         sb->s_xattr = fuse_xattr_handlers;
1088         sb->s_maxbytes = MAX_LFS_FILESIZE;
1089         sb->s_time_gran = 1;
1090         sb->s_export_op = &fuse_export_operations;
1091
1092         file = fget(d.fd);
1093         err = -EINVAL;
1094         if (!file)
1095                 goto err;
1096
1097         if ((file->f_op != &fuse_dev_operations) ||
1098             (file->f_cred->user_ns != &init_user_ns))
1099                 goto err_fput;
1100
1101         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1102         err = -ENOMEM;
1103         if (!fc)
1104                 goto err_fput;
1105
1106         fuse_conn_init(fc);
1107         fc->release = fuse_free_conn;
1108
1109         fud = fuse_dev_alloc(fc);
1110         if (!fud)
1111                 goto err_put_conn;
1112
1113         fc->dev = sb->s_dev;
1114         fc->sb = sb;
1115         err = fuse_bdi_init(fc, sb);
1116         if (err)
1117                 goto err_dev_free;
1118
1119         sb->s_bdi = &fc->bdi;
1120
1121         /* Handle umasking inside the fuse code */
1122         if (sb->s_flags & MS_POSIXACL)
1123                 fc->dont_mask = 1;
1124         sb->s_flags |= MS_POSIXACL;
1125
1126         fc->default_permissions = d.default_permissions;
1127         fc->allow_other = d.allow_other;
1128         fc->user_id = d.user_id;
1129         fc->group_id = d.group_id;
1130         fc->max_read = max_t(unsigned, 4096, d.max_read);
1131
1132         /* Used by get_root_inode() */
1133         sb->s_fs_info = fc;
1134
1135         err = -ENOMEM;
1136         root = fuse_get_root_inode(sb, d.rootmode);
1137         sb->s_d_op = &fuse_root_dentry_operations;
1138         root_dentry = d_make_root(root);
1139         if (!root_dentry)
1140                 goto err_dev_free;
1141         /* Root dentry doesn't have .d_revalidate */
1142         sb->s_d_op = &fuse_dentry_operations;
1143
1144         init_req = fuse_request_alloc(0);
1145         if (!init_req)
1146                 goto err_put_root;
1147         __set_bit(FR_BACKGROUND, &init_req->flags);
1148
1149         if (is_bdev) {
1150                 fc->destroy_req = fuse_request_alloc(0);
1151                 if (!fc->destroy_req)
1152                         goto err_free_init_req;
1153         }
1154
1155         mutex_lock(&fuse_mutex);
1156         err = -EINVAL;
1157         if (file->private_data)
1158                 goto err_unlock;
1159
1160         err = fuse_ctl_add_conn(fc);
1161         if (err)
1162                 goto err_unlock;
1163
1164         list_add_tail(&fc->entry, &fuse_conn_list);
1165         sb->s_root = root_dentry;
1166         file->private_data = fud;
1167         mutex_unlock(&fuse_mutex);
1168         /*
1169          * atomic_dec_and_test() in fput() provides the necessary
1170          * memory barrier for file->private_data to be visible on all
1171          * CPUs after this
1172          */
1173         fput(file);
1174
1175         fuse_send_init(fc, init_req);
1176
1177         return 0;
1178
1179  err_unlock:
1180         mutex_unlock(&fuse_mutex);
1181  err_free_init_req:
1182         fuse_request_free(init_req);
1183  err_put_root:
1184         dput(root_dentry);
1185  err_dev_free:
1186         fuse_dev_free(fud);
1187  err_put_conn:
1188         fuse_bdi_destroy(fc);
1189         fuse_conn_put(fc);
1190  err_fput:
1191         fput(file);
1192  err:
1193         return err;
1194 }
1195
1196 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1197                        int flags, const char *dev_name,
1198                        void *raw_data)
1199 {
1200         return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1201 }
1202
1203 static void fuse_kill_sb_anon(struct super_block *sb)
1204 {
1205         struct fuse_conn *fc = get_fuse_conn_super(sb);
1206
1207         if (fc) {
1208                 down_write(&fc->killsb);
1209                 fc->sb = NULL;
1210                 up_write(&fc->killsb);
1211         }
1212
1213         kill_anon_super(sb);
1214 }
1215
1216 static struct file_system_type fuse_fs_type = {
1217         .owner          = THIS_MODULE,
1218         .name           = "fuse",
1219         .fs_flags       = FS_HAS_SUBTYPE,
1220         .mount          = fuse_mount,
1221         .kill_sb        = fuse_kill_sb_anon,
1222 };
1223 MODULE_ALIAS_FS("fuse");
1224
1225 #ifdef CONFIG_BLOCK
1226 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1227                            int flags, const char *dev_name,
1228                            void *raw_data)
1229 {
1230         return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1231 }
1232
1233 static void fuse_kill_sb_blk(struct super_block *sb)
1234 {
1235         struct fuse_conn *fc = get_fuse_conn_super(sb);
1236
1237         if (fc) {
1238                 down_write(&fc->killsb);
1239                 fc->sb = NULL;
1240                 up_write(&fc->killsb);
1241         }
1242
1243         kill_block_super(sb);
1244 }
1245
1246 static struct file_system_type fuseblk_fs_type = {
1247         .owner          = THIS_MODULE,
1248         .name           = "fuseblk",
1249         .mount          = fuse_mount_blk,
1250         .kill_sb        = fuse_kill_sb_blk,
1251         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1252 };
1253 MODULE_ALIAS_FS("fuseblk");
1254
1255 static inline int register_fuseblk(void)
1256 {
1257         return register_filesystem(&fuseblk_fs_type);
1258 }
1259
1260 static inline void unregister_fuseblk(void)
1261 {
1262         unregister_filesystem(&fuseblk_fs_type);
1263 }
1264 #else
1265 static inline int register_fuseblk(void)
1266 {
1267         return 0;
1268 }
1269
1270 static inline void unregister_fuseblk(void)
1271 {
1272 }
1273 #endif
1274
1275 static void fuse_inode_init_once(void *foo)
1276 {
1277         struct inode *inode = foo;
1278
1279         inode_init_once(inode);
1280 }
1281
1282 static int __init fuse_fs_init(void)
1283 {
1284         int err;
1285
1286         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1287                                               sizeof(struct fuse_inode), 0,
1288                                               SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT,
1289                                               fuse_inode_init_once);
1290         err = -ENOMEM;
1291         if (!fuse_inode_cachep)
1292                 goto out;
1293
1294         err = register_fuseblk();
1295         if (err)
1296                 goto out2;
1297
1298         err = register_filesystem(&fuse_fs_type);
1299         if (err)
1300                 goto out3;
1301
1302         return 0;
1303
1304  out3:
1305         unregister_fuseblk();
1306  out2:
1307         kmem_cache_destroy(fuse_inode_cachep);
1308  out:
1309         return err;
1310 }
1311
1312 static void fuse_fs_cleanup(void)
1313 {
1314         unregister_filesystem(&fuse_fs_type);
1315         unregister_fuseblk();
1316
1317         /*
1318          * Make sure all delayed rcu free inodes are flushed before we
1319          * destroy cache.
1320          */
1321         rcu_barrier();
1322         kmem_cache_destroy(fuse_inode_cachep);
1323 }
1324
1325 static struct kobject *fuse_kobj;
1326
1327 static int fuse_sysfs_init(void)
1328 {
1329         int err;
1330
1331         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1332         if (!fuse_kobj) {
1333                 err = -ENOMEM;
1334                 goto out_err;
1335         }
1336
1337         err = sysfs_create_mount_point(fuse_kobj, "connections");
1338         if (err)
1339                 goto out_fuse_unregister;
1340
1341         return 0;
1342
1343  out_fuse_unregister:
1344         kobject_put(fuse_kobj);
1345  out_err:
1346         return err;
1347 }
1348
1349 static void fuse_sysfs_cleanup(void)
1350 {
1351         sysfs_remove_mount_point(fuse_kobj, "connections");
1352         kobject_put(fuse_kobj);
1353 }
1354
1355 static int __init fuse_init(void)
1356 {
1357         int res;
1358
1359         printk(KERN_INFO "fuse init (API version %i.%i)\n",
1360                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1361
1362         INIT_LIST_HEAD(&fuse_conn_list);
1363         res = fuse_fs_init();
1364         if (res)
1365                 goto err;
1366
1367         res = fuse_dev_init();
1368         if (res)
1369                 goto err_fs_cleanup;
1370
1371         res = fuse_sysfs_init();
1372         if (res)
1373                 goto err_dev_cleanup;
1374
1375         res = fuse_ctl_init();
1376         if (res)
1377                 goto err_sysfs_cleanup;
1378
1379         sanitize_global_limit(&max_user_bgreq);
1380         sanitize_global_limit(&max_user_congthresh);
1381
1382         return 0;
1383
1384  err_sysfs_cleanup:
1385         fuse_sysfs_cleanup();
1386  err_dev_cleanup:
1387         fuse_dev_cleanup();
1388  err_fs_cleanup:
1389         fuse_fs_cleanup();
1390  err:
1391         return res;
1392 }
1393
1394 static void __exit fuse_exit(void)
1395 {
1396         printk(KERN_DEBUG "fuse exit\n");
1397
1398         fuse_ctl_cleanup();
1399         fuse_sysfs_cleanup();
1400         fuse_fs_cleanup();
1401         fuse_dev_cleanup();
1402 }
1403
1404 module_init(fuse_init);
1405 module_exit(fuse_exit);