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->inval_mask = 0;
  94         fi->nodeid = 0;
  95         fi->nlookup = 0;
  96         fi->attr_version = 0;
  97         fi->orig_ino = 0;
  98         fi->state = 0;
  99         mutex_init(&fi->mutex);
 100         spin_lock_init(&fi->lock);
 101         fi->forget = fuse_alloc_forget();
 102         if (!fi->forget) {
 103                 kmem_cache_free(fuse_inode_cachep, inode);
 104                 return NULL;
 105         }
 106
 107         return inode;
 108 }
 109
 110 static void fuse_i_callback(struct rcu_head *head)
 111 {
 112         struct inode *inode = container_of(head, struct inode, i_rcu);
 113         kmem_cache_free(fuse_inode_cachep, inode);
 114 }
 115
 116 static void fuse_destroy_inode(struct inode *inode)
 117 {
 118         struct fuse_inode *fi = get_fuse_inode(inode);
 119         if (S_ISREG(inode->i_mode) && !is_bad_inode(inode)) {
 120                 WARN_ON(!list_empty(&fi->write_files));
 121                 WARN_ON(!list_empty(&fi->queued_writes));
 122         }
 123         mutex_destroy(&fi->mutex);
 124         kfree(fi->forget);
 125         call_rcu(&inode->i_rcu, fuse_i_callback);
 126 }
 127
 128 static void fuse_evict_inode(struct inode *inode)
 129 {
 130         truncate_inode_pages_final(&inode->i_data);
 131         clear_inode(inode);
 132         if (inode->i_sb->s_flags & SB_ACTIVE) {
 133                 struct fuse_conn *fc = get_fuse_conn(inode);
 134                 struct fuse_inode *fi = get_fuse_inode(inode);
 135                 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
 136                 fi->forget = NULL;
 137         }
 138 }
 139
 140 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
 141 {
 142         sync_filesystem(sb);
 143         if (*flags & SB_MANDLOCK)
 144                 return -EINVAL;
 145
 146         return 0;
 147 }
 148
 149 /*
 150  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
 151  * so that it will fit.
 152  */
 153 static ino_t fuse_squash_ino(u64 ino64)
 154 {
 155         ino_t ino = (ino_t) ino64;
 156         if (sizeof(ino_t) < sizeof(u64))
 157                 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
 158         return ino;
 159 }
 160
 161 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
 162                                    u64 attr_valid)
 163 {
 164         struct fuse_conn *fc = get_fuse_conn(inode);
 165         struct fuse_inode *fi = get_fuse_inode(inode);
 166
 167         lockdep_assert_held(&fi->lock);
 168
 169         fi->attr_version = atomic64_inc_return(&fc->attr_version);
 170         fi->i_time = attr_valid;
 171         WRITE_ONCE(fi->inval_mask, 0);
 172
 173         inode->i_ino     = fuse_squash_ino(attr->ino);
 174         inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
 175         set_nlink(inode, attr->nlink);
 176         inode->i_uid     = make_kuid(fc->user_ns, attr->uid);
 177         inode->i_gid     = make_kgid(fc->user_ns, attr->gid);
 178         inode->i_blocks  = attr->blocks;
 179         inode->i_atime.tv_sec   = attr->atime;
 180         inode->i_atime.tv_nsec  = attr->atimensec;
 181         /* mtime from server may be stale due to local buffered write */
 182         if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
 183                 inode->i_mtime.tv_sec   = attr->mtime;
 184                 inode->i_mtime.tv_nsec  = attr->mtimensec;
 185                 inode->i_ctime.tv_sec   = attr->ctime;
 186                 inode->i_ctime.tv_nsec  = attr->ctimensec;
 187         }
 188
 189         if (attr->blksize != 0)
 190                 inode->i_blkbits = ilog2(attr->blksize);
 191         else
 192                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
 193
 194         /*
 195          * Don't set the sticky bit in i_mode, unless we want the VFS
 196          * to check permissions.  This prevents failures due to the
 197          * check in may_delete().
 198          */
 199         fi->orig_i_mode = inode->i_mode;
 200         if (!fc->default_permissions)
 201                 inode->i_mode &= ~S_ISVTX;
 202
 203         fi->orig_ino = attr->ino;
 204 }
 205
 206 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
 207                             u64 attr_valid, u64 attr_version)
 208 {
 209         struct fuse_conn *fc = get_fuse_conn(inode);
 210         struct fuse_inode *fi = get_fuse_inode(inode);
 211         bool is_wb = fc->writeback_cache;
 212         loff_t oldsize;
 213         struct timespec64 old_mtime;
 214
 215         spin_lock(&fi->lock);
 216         if ((attr_version != 0 && fi->attr_version > attr_version) ||
 217             test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
 218                 spin_unlock(&fi->lock);
 219                 return;
 220         }
 221
 222         old_mtime = inode->i_mtime;
 223         fuse_change_attributes_common(inode, attr, attr_valid);
 224
 225         oldsize = inode->i_size;
 226         /*
 227          * In case of writeback_cache enabled, the cached writes beyond EOF
 228          * extend local i_size without keeping userspace server in sync. So,
 229          * attr->size coming from server can be stale. We cannot trust it.
 230          */
 231         if (!is_wb || !S_ISREG(inode->i_mode))
 232                 i_size_write(inode, attr->size);
 233         spin_unlock(&fi->lock);
 234
 235         if (!is_wb && S_ISREG(inode->i_mode)) {
 236                 bool inval = false;
 237
 238                 if (oldsize != attr->size) {
 239                         truncate_pagecache(inode, attr->size);
 240                         inval = true;
 241                 } else if (fc->auto_inval_data) {
 242                         struct timespec64 new_mtime = {
 243                                 .tv_sec = attr->mtime,
 244                                 .tv_nsec = attr->mtimensec,
 245                         };
 246
 247                         /*
 248                          * Auto inval mode also checks and invalidates if mtime
 249                          * has changed.
 250                          */
 251                         if (!timespec64_equal(&old_mtime, &new_mtime))
 252                                 inval = true;
 253                 }
 254
 255                 if (inval)
 256                         invalidate_inode_pages2(inode->i_mapping);
 257         }
 258 }
 259
 260 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
 261 {
 262         inode->i_mode = attr->mode & S_IFMT;
 263         inode->i_size = attr->size;
 264         inode->i_mtime.tv_sec  = attr->mtime;
 265         inode->i_mtime.tv_nsec = attr->mtimensec;
 266         inode->i_ctime.tv_sec  = attr->ctime;
 267         inode->i_ctime.tv_nsec = attr->ctimensec;
 268         if (S_ISREG(inode->i_mode)) {
 269                 fuse_init_common(inode);
 270                 fuse_init_file_inode(inode);
 271         } else if (S_ISDIR(inode->i_mode))
 272                 fuse_init_dir(inode);
 273         else if (S_ISLNK(inode->i_mode))
 274                 fuse_init_symlink(inode);
 275         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
 276                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
 277                 fuse_init_common(inode);
 278                 init_special_inode(inode, inode->i_mode,
 279                                    new_decode_dev(attr->rdev));
 280         } else
 281                 BUG();
 282 }
 283
 284 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
 285 {
 286         u64 nodeid = *(u64 *) _nodeidp;
 287         if (get_node_id(inode) == nodeid)
 288                 return 1;
 289         else
 290                 return 0;
 291 }
 292
 293 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
 294 {
 295         u64 nodeid = *(u64 *) _nodeidp;
 296         get_fuse_inode(inode)->nodeid = nodeid;
 297         return 0;
 298 }
 299
 300 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
 301                         int generation, struct fuse_attr *attr,
 302                         u64 attr_valid, u64 attr_version)
 303 {
 304         struct inode *inode;
 305         struct fuse_inode *fi;
 306         struct fuse_conn *fc = get_fuse_conn_super(sb);
 307
 308  retry:
 309         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
 310         if (!inode)
 311                 return NULL;
 312
 313         if ((inode->i_state & I_NEW)) {
 314                 inode->i_flags |= S_NOATIME;
 315                 if (!fc->writeback_cache || !S_ISREG(attr->mode))
 316                         inode->i_flags |= S_NOCMTIME;
 317                 inode->i_generation = generation;
 318                 fuse_init_inode(inode, attr);
 319                 unlock_new_inode(inode);
 320         } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
 321                 /* Inode has changed type, any I/O on the old should fail */
 322                 make_bad_inode(inode);
 323                 iput(inode);
 324                 goto retry;
 325         }
 326
 327         fi = get_fuse_inode(inode);
 328         spin_lock(&fi->lock);
 329         fi->nlookup++;
 330         spin_unlock(&fi->lock);
 331         fuse_change_attributes(inode, attr, attr_valid, attr_version);
 332
 333         return inode;
 334 }
 335
 336 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
 337                              loff_t offset, loff_t len)
 338 {
 339         struct inode *inode;
 340         pgoff_t pg_start;
 341         pgoff_t pg_end;
 342
 343         inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
 344         if (!inode)
 345                 return -ENOENT;
 346
 347         fuse_invalidate_attr(inode);
 348         forget_all_cached_acls(inode);
 349         if (offset >= 0) {
 350                 pg_start = offset >> PAGE_SHIFT;
 351                 if (len <= 0)
 352                         pg_end = -1;
 353                 else
 354                         pg_end = (offset + len - 1) >> PAGE_SHIFT;
 355                 invalidate_inode_pages2_range(inode->i_mapping,
 356                                               pg_start, pg_end);
 357         }
 358         iput(inode);
 359         return 0;
 360 }
 361
 362 bool fuse_lock_inode(struct inode *inode)
 363 {
 364         bool locked = false;
 365
 366         if (!get_fuse_conn(inode)->parallel_dirops) {
 367                 mutex_lock(&get_fuse_inode(inode)->mutex);
 368                 locked = true;
 369         }
 370
 371         return locked;
 372 }
 373
 374 void fuse_unlock_inode(struct inode *inode, bool locked)
 375 {
 376         if (locked)
 377                 mutex_unlock(&get_fuse_inode(inode)->mutex);
 378 }
 379
 380 static void fuse_umount_begin(struct super_block *sb)
 381 {
 382         fuse_abort_conn(get_fuse_conn_super(sb));
 383 }
 384
 385 static void fuse_send_destroy(struct fuse_conn *fc)
 386 {
 387         struct fuse_req *req = fc->destroy_req;
 388         if (req && fc->conn_init) {
 389                 fc->destroy_req = NULL;
 390                 req->in.h.opcode = FUSE_DESTROY;
 391                 __set_bit(FR_FORCE, &req->flags);
 392                 __clear_bit(FR_BACKGROUND, &req->flags);
 393                 fuse_request_send(fc, req);
 394                 fuse_put_request(fc, req);
 395         }
 396 }
 397
 398 static void fuse_put_super(struct super_block *sb)
 399 {
 400         struct fuse_conn *fc = get_fuse_conn_super(sb);
 401
 402         mutex_lock(&fuse_mutex);
 403         list_del(&fc->entry);
 404         fuse_ctl_remove_conn(fc);
 405         mutex_unlock(&fuse_mutex);
 406
 407         fuse_conn_put(fc);
 408 }
 409
 410 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
 411 {
 412         stbuf->f_type    = FUSE_SUPER_MAGIC;
 413         stbuf->f_bsize   = attr->bsize;
 414         stbuf->f_frsize  = attr->frsize;
 415         stbuf->f_blocks  = attr->blocks;
 416         stbuf->f_bfree   = attr->bfree;
 417         stbuf->f_bavail  = attr->bavail;
 418         stbuf->f_files   = attr->files;
 419         stbuf->f_ffree   = attr->ffree;
 420         stbuf->f_namelen = attr->namelen;
 421         /* fsid is left zero */
 422 }
 423
 424 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
 425 {
 426         struct super_block *sb = dentry->d_sb;
 427         struct fuse_conn *fc = get_fuse_conn_super(sb);
 428         FUSE_ARGS(args);
 429         struct fuse_statfs_out outarg;
 430         int err;
 431
 432         if (!fuse_allow_current_process(fc)) {
 433                 buf->f_type = FUSE_SUPER_MAGIC;
 434                 return 0;
 435         }
 436
 437         memset(&outarg, 0, sizeof(outarg));
 438         args.in.numargs = 0;
 439         args.in.h.opcode = FUSE_STATFS;
 440         args.in.h.nodeid = get_node_id(d_inode(dentry));
 441         args.out.numargs = 1;
 442         args.out.args[0].size = sizeof(outarg);
 443         args.out.args[0].value = &outarg;
 444         err = fuse_simple_request(fc, &args);
 445         if (!err)
 446                 convert_fuse_statfs(buf, &outarg.st);
 447         return err;
 448 }
 449
 450 enum {
 451         OPT_FD,
 452         OPT_ROOTMODE,
 453         OPT_USER_ID,
 454         OPT_GROUP_ID,
 455         OPT_DEFAULT_PERMISSIONS,
 456         OPT_ALLOW_OTHER,
 457         OPT_MAX_READ,
 458         OPT_BLKSIZE,
 459         OPT_ERR
 460 };
 461
 462 static const match_table_t tokens = {
 463         {OPT_FD,                        "fd=%u"},
 464         {OPT_ROOTMODE,                  "rootmode=%o"},
 465         {OPT_USER_ID,                   "user_id=%u"},
 466         {OPT_GROUP_ID,                  "group_id=%u"},
 467         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 468         {OPT_ALLOW_OTHER,               "allow_other"},
 469         {OPT_MAX_READ,                  "max_read=%u"},
 470         {OPT_BLKSIZE,                   "blksize=%u"},
 471         {OPT_ERR,                       NULL}
 472 };
 473
 474 static int fuse_match_uint(substring_t *s, unsigned int *res)
 475 {
 476         int err = -ENOMEM;
 477         char *buf = match_strdup(s);
 478         if (buf) {
 479                 err = kstrtouint(buf, 10, res);
 480                 kfree(buf);
 481         }
 482         return err;
 483 }
 484
 485 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev,
 486                           struct user_namespace *user_ns)
 487 {
 488         char *p;
 489         memset(d, 0, sizeof(struct fuse_mount_data));
 490         d->max_read = ~0;
 491         d->blksize = FUSE_DEFAULT_BLKSIZE;
 492
 493         while ((p = strsep(&opt, ",")) != NULL) {
 494                 int token;
 495                 int value;
 496                 unsigned uv;
 497                 substring_t args[MAX_OPT_ARGS];
 498                 if (!*p)
 499                         continue;
 500
 501                 token = match_token(p, tokens, args);
 502                 switch (token) {
 503                 case OPT_FD:
 504                         if (match_int(&args[0], &value))
 505                                 return 0;
 506                         d->fd = value;
 507                         d->fd_present = 1;
 508                         break;
 509
 510                 case OPT_ROOTMODE:
 511                         if (match_octal(&args[0], &value))
 512                                 return 0;
 513                         if (!fuse_valid_type(value))
 514                                 return 0;
 515                         d->rootmode = value;
 516                         d->rootmode_present = 1;
 517                         break;
 518
 519                 case OPT_USER_ID:
 520                         if (fuse_match_uint(&args[0], &uv))
 521                                 return 0;
 522                         d->user_id = make_kuid(user_ns, uv);
 523                         if (!uid_valid(d->user_id))
 524                                 return 0;
 525                         d->user_id_present = 1;
 526                         break;
 527
 528                 case OPT_GROUP_ID:
 529                         if (fuse_match_uint(&args[0], &uv))
 530                                 return 0;
 531                         d->group_id = make_kgid(user_ns, uv);
 532                         if (!gid_valid(d->group_id))
 533                                 return 0;
 534                         d->group_id_present = 1;
 535                         break;
 536
 537                 case OPT_DEFAULT_PERMISSIONS:
 538                         d->default_permissions = 1;
 539                         break;
 540
 541                 case OPT_ALLOW_OTHER:
 542                         d->allow_other = 1;
 543                         break;
 544
 545                 case OPT_MAX_READ:
 546                         if (match_int(&args[0], &value))
 547                                 return 0;
 548                         d->max_read = value;
 549                         break;
 550
 551                 case OPT_BLKSIZE:
 552                         if (!is_bdev || match_int(&args[0], &value))
 553                                 return 0;
 554                         d->blksize = value;
 555                         break;
 556
 557                 default:
 558                         return 0;
 559                 }
 560         }
 561
 562         if (!d->fd_present || !d->rootmode_present ||
 563             !d->user_id_present || !d->group_id_present)
 564                 return 0;
 565
 566         return 1;
 567 }
 568
 569 static int fuse_show_options(struct seq_file *m, struct dentry *root)
 570 {
 571         struct super_block *sb = root->d_sb;
 572         struct fuse_conn *fc = get_fuse_conn_super(sb);
 573
 574         seq_printf(m, ",user_id=%u", from_kuid_munged(fc->user_ns, fc->user_id));
 575         seq_printf(m, ",group_id=%u", from_kgid_munged(fc->user_ns, fc->group_id));
 576         if (fc->default_permissions)
 577                 seq_puts(m, ",default_permissions");
 578         if (fc->allow_other)
 579                 seq_puts(m, ",allow_other");
 580         if (fc->max_read != ~0)
 581                 seq_printf(m, ",max_read=%u", fc->max_read);
 582         if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
 583                 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
 584         return 0;
 585 }
 586
 587 static void fuse_iqueue_init(struct fuse_iqueue *fiq)
 588 {
 589         memset(fiq, 0, sizeof(struct fuse_iqueue));
 590         init_waitqueue_head(&fiq->waitq);
 591         INIT_LIST_HEAD(&fiq->pending);
 592         INIT_LIST_HEAD(&fiq->interrupts);
 593         fiq->forget_list_tail = &fiq->forget_list_head;
 594         fiq->connected = 1;
 595 }
 596
 597 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
 598 {
 599         unsigned int i;
 600
 601         spin_lock_init(&fpq->lock);
 602         for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
 603                 INIT_LIST_HEAD(&fpq->processing[i]);
 604         INIT_LIST_HEAD(&fpq->io);
 605         fpq->connected = 1;
 606 }
 607
 608 void fuse_conn_init(struct fuse_conn *fc, struct user_namespace *user_ns)
 609 {
 610         memset(fc, 0, sizeof(*fc));
 611         spin_lock_init(&fc->lock);
 612         spin_lock_init(&fc->bg_lock);
 613         init_rwsem(&fc->killsb);
 614         refcount_set(&fc->count, 1);
 615         atomic_set(&fc->dev_count, 1);
 616         init_waitqueue_head(&fc->blocked_waitq);
 617         init_waitqueue_head(&fc->reserved_req_waitq);
 618         fuse_iqueue_init(&fc->iq);
 619         INIT_LIST_HEAD(&fc->bg_queue);
 620         INIT_LIST_HEAD(&fc->entry);
 621         INIT_LIST_HEAD(&fc->devices);
 622         atomic_set(&fc->num_waiting, 0);
 623         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
 624         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
 625         atomic64_set(&fc->khctr, 0);
 626         fc->polled_files = RB_ROOT;
 627         fc->blocked = 0;
 628         fc->initialized = 0;
 629         fc->connected = 1;
 630         atomic64_set(&fc->attr_version, 1);
 631         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
 632         fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
 633         fc->user_ns = get_user_ns(user_ns);
 634         fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
 635 }
 636 EXPORT_SYMBOL_GPL(fuse_conn_init);
 637
 638 void fuse_conn_put(struct fuse_conn *fc)
 639 {
 640         if (refcount_dec_and_test(&fc->count)) {
 641                 if (fc->destroy_req)
 642                         fuse_request_free(fc->destroy_req);
 643                 put_pid_ns(fc->pid_ns);
 644                 put_user_ns(fc->user_ns);
 645                 fc->release(fc);
 646         }
 647 }
 648 EXPORT_SYMBOL_GPL(fuse_conn_put);
 649
 650 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
 651 {
 652         refcount_inc(&fc->count);
 653         return fc;
 654 }
 655 EXPORT_SYMBOL_GPL(fuse_conn_get);
 656
 657 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
 658 {
 659         struct fuse_attr attr;
 660         memset(&attr, 0, sizeof(attr));
 661
 662         attr.mode = mode;
 663         attr.ino = FUSE_ROOT_ID;
 664         attr.nlink = 1;
 665         return fuse_iget(sb, 1, 0, &attr, 0, 0);
 666 }
 667
 668 struct fuse_inode_handle {
 669         u64 nodeid;
 670         u32 generation;
 671 };
 672
 673 static struct dentry *fuse_get_dentry(struct super_block *sb,
 674                                       struct fuse_inode_handle *handle)
 675 {
 676         struct fuse_conn *fc = get_fuse_conn_super(sb);
 677         struct inode *inode;
 678         struct dentry *entry;
 679         int err = -ESTALE;
 680
 681         if (handle->nodeid == 0)
 682                 goto out_err;
 683
 684         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
 685         if (!inode) {
 686                 struct fuse_entry_out outarg;
 687                 const struct qstr name = QSTR_INIT(".", 1);
 688
 689                 if (!fc->export_support)
 690                         goto out_err;
 691
 692                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
 693                                        &inode);
 694                 if (err && err != -ENOENT)
 695                         goto out_err;
 696                 if (err || !inode) {
 697                         err = -ESTALE;
 698                         goto out_err;
 699                 }
 700                 err = -EIO;
 701                 if (get_node_id(inode) != handle->nodeid)
 702                         goto out_iput;
 703         }
 704         err = -ESTALE;
 705         if (inode->i_generation != handle->generation)
 706                 goto out_iput;
 707
 708         entry = d_obtain_alias(inode);
 709         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
 710                 fuse_invalidate_entry_cache(entry);
 711
 712         return entry;
 713
 714  out_iput:
 715         iput(inode);
 716  out_err:
 717         return ERR_PTR(err);
 718 }
 719
 720 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
 721                            struct inode *parent)
 722 {
 723         int len = parent ? 6 : 3;
 724         u64 nodeid;
 725         u32 generation;
 726
 727         if (*max_len < len) {
 728                 *max_len = len;
 729                 return  FILEID_INVALID;
 730         }
 731
 732         nodeid = get_fuse_inode(inode)->nodeid;
 733         generation = inode->i_generation;
 734
 735         fh[0] = (u32)(nodeid >> 32);
 736         fh[1] = (u32)(nodeid & 0xffffffff);
 737         fh[2] = generation;
 738
 739         if (parent) {
 740                 nodeid = get_fuse_inode(parent)->nodeid;
 741                 generation = parent->i_generation;
 742
 743                 fh[3] = (u32)(nodeid >> 32);
 744                 fh[4] = (u32)(nodeid & 0xffffffff);
 745                 fh[5] = generation;
 746         }
 747
 748         *max_len = len;
 749         return parent ? 0x82 : 0x81;
 750 }
 751
 752 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
 753                 struct fid *fid, int fh_len, int fh_type)
 754 {
 755         struct fuse_inode_handle handle;
 756
 757         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
 758                 return NULL;
 759
 760         handle.nodeid = (u64) fid->raw[0] << 32;
 761         handle.nodeid |= (u64) fid->raw[1];
 762         handle.generation = fid->raw[2];
 763         return fuse_get_dentry(sb, &handle);
 764 }
 765
 766 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
 767                 struct fid *fid, int fh_len, int fh_type)
 768 {
 769         struct fuse_inode_handle parent;
 770
 771         if (fh_type != 0x82 || fh_len < 6)
 772                 return NULL;
 773
 774         parent.nodeid = (u64) fid->raw[3] << 32;
 775         parent.nodeid |= (u64) fid->raw[4];
 776         parent.generation = fid->raw[5];
 777         return fuse_get_dentry(sb, &parent);
 778 }
 779
 780 static struct dentry *fuse_get_parent(struct dentry *child)
 781 {
 782         struct inode *child_inode = d_inode(child);
 783         struct fuse_conn *fc = get_fuse_conn(child_inode);
 784         struct inode *inode;
 785         struct dentry *parent;
 786         struct fuse_entry_out outarg;
 787         const struct qstr name = QSTR_INIT("..", 2);
 788         int err;
 789
 790         if (!fc->export_support)
 791                 return ERR_PTR(-ESTALE);
 792
 793         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
 794                                &name, &outarg, &inode);
 795         if (err) {
 796                 if (err == -ENOENT)
 797                         return ERR_PTR(-ESTALE);
 798                 return ERR_PTR(err);
 799         }
 800
 801         parent = d_obtain_alias(inode);
 802         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
 803                 fuse_invalidate_entry_cache(parent);
 804
 805         return parent;
 806 }
 807
 808 static const struct export_operations fuse_export_operations = {
 809         .fh_to_dentry   = fuse_fh_to_dentry,
 810         .fh_to_parent   = fuse_fh_to_parent,
 811         .encode_fh      = fuse_encode_fh,
 812         .get_parent     = fuse_get_parent,
 813 };
 814
 815 static const struct super_operations fuse_super_operations = {
 816         .alloc_inode    = fuse_alloc_inode,
 817         .destroy_inode  = fuse_destroy_inode,
 818         .evict_inode    = fuse_evict_inode,
 819         .write_inode    = fuse_write_inode,
 820         .drop_inode     = generic_delete_inode,
 821         .remount_fs     = fuse_remount_fs,
 822         .put_super      = fuse_put_super,
 823         .umount_begin   = fuse_umount_begin,
 824         .statfs         = fuse_statfs,
 825         .show_options   = fuse_show_options,
 826 };
 827
 828 static void sanitize_global_limit(unsigned *limit)
 829 {
 830         if (*limit == 0)
 831                 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) /
 832                          sizeof(struct fuse_req);
 833
 834         if (*limit >= 1 << 16)
 835                 *limit = (1 << 16) - 1;
 836 }
 837
 838 static int set_global_limit(const char *val, const struct kernel_param *kp)
 839 {
 840         int rv;
 841
 842         rv = param_set_uint(val, kp);
 843         if (rv)
 844                 return rv;
 845
 846         sanitize_global_limit((unsigned *)kp->arg);
 847
 848         return 0;
 849 }
 850
 851 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
 852 {
 853         int cap_sys_admin = capable(CAP_SYS_ADMIN);
 854
 855         if (arg->minor < 13)
 856                 return;
 857
 858         sanitize_global_limit(&max_user_bgreq);
 859         sanitize_global_limit(&max_user_congthresh);
 860
 861         spin_lock(&fc->bg_lock);
 862         if (arg->max_background) {
 863                 fc->max_background = arg->max_background;
 864
 865                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
 866                         fc->max_background = max_user_bgreq;
 867         }
 868         if (arg->congestion_threshold) {
 869                 fc->congestion_threshold = arg->congestion_threshold;
 870
 871                 if (!cap_sys_admin &&
 872                     fc->congestion_threshold > max_user_congthresh)
 873                         fc->congestion_threshold = max_user_congthresh;
 874         }
 875         spin_unlock(&fc->bg_lock);
 876 }
 877
 878 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
 879 {
 880         struct fuse_init_out *arg = &req->misc.init_out;
 881
 882         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
 883                 fc->conn_error = 1;
 884         else {
 885                 unsigned long ra_pages;
 886
 887                 process_init_limits(fc, arg);
 888
 889                 if (arg->minor >= 6) {
 890                         ra_pages = arg->max_readahead / PAGE_SIZE;
 891                         if (arg->flags & FUSE_ASYNC_READ)
 892                                 fc->async_read = 1;
 893                         if (!(arg->flags & FUSE_POSIX_LOCKS))
 894                                 fc->no_lock = 1;
 895                         if (arg->minor >= 17) {
 896                                 if (!(arg->flags & FUSE_FLOCK_LOCKS))
 897                                         fc->no_flock = 1;
 898                         } else {
 899                                 if (!(arg->flags & FUSE_POSIX_LOCKS))
 900                                         fc->no_flock = 1;
 901                         }
 902                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
 903                                 fc->atomic_o_trunc = 1;
 904                         if (arg->minor >= 9) {
 905                                 /* LOOKUP has dependency on proto version */
 906                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
 907                                         fc->export_support = 1;
 908                         }
 909                         if (arg->flags & FUSE_BIG_WRITES)
 910                                 fc->big_writes = 1;
 911                         if (arg->flags & FUSE_DONT_MASK)
 912                                 fc->dont_mask = 1;
 913                         if (arg->flags & FUSE_AUTO_INVAL_DATA)
 914                                 fc->auto_inval_data = 1;
 915                         if (arg->flags & FUSE_DO_READDIRPLUS) {
 916                                 fc->do_readdirplus = 1;
 917                                 if (arg->flags & FUSE_READDIRPLUS_AUTO)
 918                                         fc->readdirplus_auto = 1;
 919                         }
 920                         if (arg->flags & FUSE_ASYNC_DIO)
 921                                 fc->async_dio = 1;
 922                         if (arg->flags & FUSE_WRITEBACK_CACHE)
 923                                 fc->writeback_cache = 1;
 924                         if (arg->flags & FUSE_PARALLEL_DIROPS)
 925                                 fc->parallel_dirops = 1;
 926                         if (arg->flags & FUSE_HANDLE_KILLPRIV)
 927                                 fc->handle_killpriv = 1;
 928                         if (arg->time_gran && arg->time_gran <= 1000000000)
 929                                 fc->sb->s_time_gran = arg->time_gran;
 930                         if ((arg->flags & FUSE_POSIX_ACL)) {
 931                                 fc->default_permissions = 1;
 932                                 fc->posix_acl = 1;
 933                                 fc->sb->s_xattr = fuse_acl_xattr_handlers;
 934                         }
 935                         if (arg->flags & FUSE_CACHE_SYMLINKS)
 936                                 fc->cache_symlinks = 1;
 937                         if (arg->flags & FUSE_ABORT_ERROR)
 938                                 fc->abort_err = 1;
 939                         if (arg->flags & FUSE_MAX_PAGES) {
 940                                 fc->max_pages =
 941                                         min_t(unsigned int, FUSE_MAX_MAX_PAGES,
 942                                         max_t(unsigned int, arg->max_pages, 1));
 943                         }
 944                 } else {
 945                         ra_pages = fc->max_read / PAGE_SIZE;
 946                         fc->no_lock = 1;
 947                         fc->no_flock = 1;
 948                 }
 949
 950                 fc->sb->s_bdi->ra_pages =
 951                                 min(fc->sb->s_bdi->ra_pages, ra_pages);
 952                 fc->minor = arg->minor;
 953                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
 954                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
 955                 fc->conn_init = 1;
 956         }
 957         fuse_set_initialized(fc);
 958         wake_up_all(&fc->blocked_waitq);
 959 }
 960
 961 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
 962 {
 963         struct fuse_init_in *arg = &req->misc.init_in;
 964
 965         arg->major = FUSE_KERNEL_VERSION;
 966         arg->minor = FUSE_KERNEL_MINOR_VERSION;
 967         arg->max_readahead = fc->sb->s_bdi->ra_pages * PAGE_SIZE;
 968         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
 969                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
 970                 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
 971                 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
 972                 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
 973                 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
 974                 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
 975                 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
 976                 FUSE_NO_OPENDIR_SUPPORT;
 977         req->in.h.opcode = FUSE_INIT;
 978         req->in.numargs = 1;
 979         req->in.args[0].size = sizeof(*arg);
 980         req->in.args[0].value = arg;
 981         req->out.numargs = 1;
 982         /* Variable length argument used for backward compatibility
 983            with interface version < 7.5.  Rest of init_out is zeroed
 984            by do_get_request(), so a short reply is not a problem */
 985         req->out.argvar = 1;
 986         req->out.args[0].size = sizeof(struct fuse_init_out);
 987         req->out.args[0].value = &req->misc.init_out;
 988         req->end = process_init_reply;
 989         fuse_request_send_background(fc, req);
 990 }
 991
 992 static void fuse_free_conn(struct fuse_conn *fc)
 993 {
 994         WARN_ON(!list_empty(&fc->devices));
 995         kfree_rcu(fc, rcu);
 996 }
 997
 998 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
 999 {
1000         int err;
1001         char *suffix = "";
1002
1003         if (sb->s_bdev) {
1004                 suffix = "-fuseblk";
1005                 /*
1006                  * sb->s_bdi points to blkdev's bdi however we want to redirect
1007                  * it to our private bdi...
1008                  */
1009                 bdi_put(sb->s_bdi);
1010                 sb->s_bdi = &noop_backing_dev_info;
1011         }
1012         err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1013                                    MINOR(fc->dev), suffix);
1014         if (err)
1015                 return err;
1016
1017         sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
1018         /* fuse does it's own writeback accounting */
1019         sb->s_bdi->capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
1020
1021         /*
1022          * For a single fuse filesystem use max 1% of dirty +
1023          * writeback threshold.
1024          *
1025          * This gives about 1M of write buffer for memory maps on a
1026          * machine with 1G and 10% dirty_ratio, which should be more
1027          * than enough.
1028          *
1029          * Privileged users can raise it by writing to
1030          *
1031          *    /sys/class/bdi/<bdi>/max_ratio
1032          */
1033         bdi_set_max_ratio(sb->s_bdi, 1);
1034
1035         return 0;
1036 }
1037
1038 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc)
1039 {
1040         struct fuse_dev *fud;
1041         struct list_head *pq;
1042
1043         fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1044         if (!fud)
1045                 return NULL;
1046
1047         pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1048         if (!pq) {
1049                 kfree(fud);
1050                 return NULL;
1051         }
1052
1053         fud->pq.processing = pq;
1054         fud->fc = fuse_conn_get(fc);
1055         fuse_pqueue_init(&fud->pq);
1056
1057         spin_lock(&fc->lock);
1058         list_add_tail(&fud->entry, &fc->devices);
1059         spin_unlock(&fc->lock);
1060
1061         return fud;
1062 }
1063 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1064
1065 void fuse_dev_free(struct fuse_dev *fud)
1066 {
1067         struct fuse_conn *fc = fud->fc;
1068
1069         if (fc) {
1070                 spin_lock(&fc->lock);
1071                 list_del(&fud->entry);
1072                 spin_unlock(&fc->lock);
1073
1074                 fuse_conn_put(fc);
1075         }
1076         kfree(fud->pq.processing);
1077         kfree(fud);
1078 }
1079 EXPORT_SYMBOL_GPL(fuse_dev_free);
1080
1081 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
1082 {
1083         struct fuse_dev *fud;
1084         struct fuse_conn *fc;
1085         struct inode *root;
1086         struct fuse_mount_data d;
1087         struct file *file;
1088         struct dentry *root_dentry;
1089         struct fuse_req *init_req;
1090         int err;
1091         int is_bdev = sb->s_bdev != NULL;
1092
1093         err = -EINVAL;
1094         if (sb->s_flags & SB_MANDLOCK)
1095                 goto err;
1096
1097         sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1098
1099         if (!parse_fuse_opt(data, &d, is_bdev, sb->s_user_ns))
1100                 goto err;
1101
1102         if (is_bdev) {
1103 #ifdef CONFIG_BLOCK
1104                 err = -EINVAL;
1105                 if (!sb_set_blocksize(sb, d.blksize))
1106                         goto err;
1107 #endif
1108         } else {
1109                 sb->s_blocksize = PAGE_SIZE;
1110                 sb->s_blocksize_bits = PAGE_SHIFT;
1111         }
1112         sb->s_magic = FUSE_SUPER_MAGIC;
1113         sb->s_op = &fuse_super_operations;
1114         sb->s_xattr = fuse_xattr_handlers;
1115         sb->s_maxbytes = MAX_LFS_FILESIZE;
1116         sb->s_time_gran = 1;
1117         sb->s_export_op = &fuse_export_operations;
1118         sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1119         if (sb->s_user_ns != &init_user_ns)
1120                 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1121
1122         file = fget(d.fd);
1123         err = -EINVAL;
1124         if (!file)
1125                 goto err;
1126
1127         /*
1128          * Require mount to happen from the same user namespace which
1129          * opened /dev/fuse to prevent potential attacks.
1130          */
1131         if (file->f_op != &fuse_dev_operations ||
1132             file->f_cred->user_ns != sb->s_user_ns)
1133                 goto err_fput;
1134
1135         /*
1136          * If we are not in the initial user namespace posix
1137          * acls must be translated.
1138          */
1139         if (sb->s_user_ns != &init_user_ns)
1140                 sb->s_xattr = fuse_no_acl_xattr_handlers;
1141
1142         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1143         err = -ENOMEM;
1144         if (!fc)
1145                 goto err_fput;
1146
1147         fuse_conn_init(fc, sb->s_user_ns);
1148         fc->release = fuse_free_conn;
1149
1150         fud = fuse_dev_alloc(fc);
1151         if (!fud)
1152                 goto err_put_conn;
1153
1154         fc->dev = sb->s_dev;
1155         fc->sb = sb;
1156         err = fuse_bdi_init(fc, sb);
1157         if (err)
1158                 goto err_dev_free;
1159
1160         /* Handle umasking inside the fuse code */
1161         if (sb->s_flags & SB_POSIXACL)
1162                 fc->dont_mask = 1;
1163         sb->s_flags |= SB_POSIXACL;
1164
1165         fc->default_permissions = d.default_permissions;
1166         fc->allow_other = d.allow_other;
1167         fc->user_id = d.user_id;
1168         fc->group_id = d.group_id;
1169         fc->max_read = max_t(unsigned, 4096, d.max_read);
1170
1171         /* Used by get_root_inode() */
1172         sb->s_fs_info = fc;
1173
1174         err = -ENOMEM;
1175         root = fuse_get_root_inode(sb, d.rootmode);
1176         sb->s_d_op = &fuse_root_dentry_operations;
1177         root_dentry = d_make_root(root);
1178         if (!root_dentry)
1179                 goto err_dev_free;
1180         /* Root dentry doesn't have .d_revalidate */
1181         sb->s_d_op = &fuse_dentry_operations;
1182
1183         init_req = fuse_request_alloc(0);
1184         if (!init_req)
1185                 goto err_put_root;
1186         __set_bit(FR_BACKGROUND, &init_req->flags);
1187
1188         if (is_bdev) {
1189                 fc->destroy_req = fuse_request_alloc(0);
1190                 if (!fc->destroy_req)
1191                         goto err_free_init_req;
1192         }
1193
1194         mutex_lock(&fuse_mutex);
1195         err = -EINVAL;
1196         if (file->private_data)
1197                 goto err_unlock;
1198
1199         err = fuse_ctl_add_conn(fc);
1200         if (err)
1201                 goto err_unlock;
1202
1203         list_add_tail(&fc->entry, &fuse_conn_list);
1204         sb->s_root = root_dentry;
1205         file->private_data = fud;
1206         mutex_unlock(&fuse_mutex);
1207         /*
1208          * atomic_dec_and_test() in fput() provides the necessary
1209          * memory barrier for file->private_data to be visible on all
1210          * CPUs after this
1211          */
1212         fput(file);
1213
1214         fuse_send_init(fc, init_req);
1215
1216         return 0;
1217
1218  err_unlock:
1219         mutex_unlock(&fuse_mutex);
1220  err_free_init_req:
1221         fuse_request_free(init_req);
1222  err_put_root:
1223         dput(root_dentry);
1224  err_dev_free:
1225         fuse_dev_free(fud);
1226  err_put_conn:
1227         fuse_conn_put(fc);
1228         sb->s_fs_info = NULL;
1229  err_fput:
1230         fput(file);
1231  err:
1232         return err;
1233 }
1234
1235 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1236                        int flags, const char *dev_name,
1237                        void *raw_data)
1238 {
1239         return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1240 }
1241
1242 static void fuse_sb_destroy(struct super_block *sb)
1243 {
1244         struct fuse_conn *fc = get_fuse_conn_super(sb);
1245
1246         if (fc) {
1247                 fuse_send_destroy(fc);
1248
1249                 fuse_abort_conn(fc);
1250                 fuse_wait_aborted(fc);
1251
1252                 down_write(&fc->killsb);
1253                 fc->sb = NULL;
1254                 up_write(&fc->killsb);
1255         }
1256 }
1257
1258 static void fuse_kill_sb_anon(struct super_block *sb)
1259 {
1260         fuse_sb_destroy(sb);
1261         kill_anon_super(sb);
1262 }
1263
1264 static struct file_system_type fuse_fs_type = {
1265         .owner          = THIS_MODULE,
1266         .name           = "fuse",
1267         .fs_flags       = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1268         .mount          = fuse_mount,
1269         .kill_sb        = fuse_kill_sb_anon,
1270 };
1271 MODULE_ALIAS_FS("fuse");
1272
1273 #ifdef CONFIG_BLOCK
1274 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1275                            int flags, const char *dev_name,
1276                            void *raw_data)
1277 {
1278         return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1279 }
1280
1281 static void fuse_kill_sb_blk(struct super_block *sb)
1282 {
1283         fuse_sb_destroy(sb);
1284         kill_block_super(sb);
1285 }
1286
1287 static struct file_system_type fuseblk_fs_type = {
1288         .owner          = THIS_MODULE,
1289         .name           = "fuseblk",
1290         .mount          = fuse_mount_blk,
1291         .kill_sb        = fuse_kill_sb_blk,
1292         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1293 };
1294 MODULE_ALIAS_FS("fuseblk");
1295
1296 static inline int register_fuseblk(void)
1297 {
1298         return register_filesystem(&fuseblk_fs_type);
1299 }
1300
1301 static inline void unregister_fuseblk(void)
1302 {
1303         unregister_filesystem(&fuseblk_fs_type);
1304 }
1305 #else
1306 static inline int register_fuseblk(void)
1307 {
1308         return 0;
1309 }
1310
1311 static inline void unregister_fuseblk(void)
1312 {
1313 }
1314 #endif
1315
1316 static void fuse_inode_init_once(void *foo)
1317 {
1318         struct inode *inode = foo;
1319
1320         inode_init_once(inode);
1321 }
1322
1323 static int __init fuse_fs_init(void)
1324 {
1325         int err;
1326
1327         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1328                         sizeof(struct fuse_inode), 0,
1329                         SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1330                         fuse_inode_init_once);
1331         err = -ENOMEM;
1332         if (!fuse_inode_cachep)
1333                 goto out;
1334
1335         err = register_fuseblk();
1336         if (err)
1337                 goto out2;
1338
1339         err = register_filesystem(&fuse_fs_type);
1340         if (err)
1341                 goto out3;
1342
1343         return 0;
1344
1345  out3:
1346         unregister_fuseblk();
1347  out2:
1348         kmem_cache_destroy(fuse_inode_cachep);
1349  out:
1350         return err;
1351 }
1352
1353 static void fuse_fs_cleanup(void)
1354 {
1355         unregister_filesystem(&fuse_fs_type);
1356         unregister_fuseblk();
1357
1358         /*
1359          * Make sure all delayed rcu free inodes are flushed before we
1360          * destroy cache.
1361          */
1362         rcu_barrier();
1363         kmem_cache_destroy(fuse_inode_cachep);
1364 }
1365
1366 static struct kobject *fuse_kobj;
1367
1368 static int fuse_sysfs_init(void)
1369 {
1370         int err;
1371
1372         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1373         if (!fuse_kobj) {
1374                 err = -ENOMEM;
1375                 goto out_err;
1376         }
1377
1378         err = sysfs_create_mount_point(fuse_kobj, "connections");
1379         if (err)
1380                 goto out_fuse_unregister;
1381
1382         return 0;
1383
1384  out_fuse_unregister:
1385         kobject_put(fuse_kobj);
1386  out_err:
1387         return err;
1388 }
1389
1390 static void fuse_sysfs_cleanup(void)
1391 {
1392         sysfs_remove_mount_point(fuse_kobj, "connections");
1393         kobject_put(fuse_kobj);
1394 }
1395
1396 static int __init fuse_init(void)
1397 {
1398         int res;
1399
1400         printk(KERN_INFO "fuse init (API version %i.%i)\n",
1401                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1402
1403         INIT_LIST_HEAD(&fuse_conn_list);
1404         res = fuse_fs_init();
1405         if (res)
1406                 goto err;
1407
1408         res = fuse_dev_init();
1409         if (res)
1410                 goto err_fs_cleanup;
1411
1412         res = fuse_sysfs_init();
1413         if (res)
1414                 goto err_dev_cleanup;
1415
1416         res = fuse_ctl_init();
1417         if (res)
1418                 goto err_sysfs_cleanup;
1419
1420         sanitize_global_limit(&max_user_bgreq);
1421         sanitize_global_limit(&max_user_congthresh);
1422
1423         return 0;
1424
1425  err_sysfs_cleanup:
1426         fuse_sysfs_cleanup();
1427  err_dev_cleanup:
1428         fuse_dev_cleanup();
1429  err_fs_cleanup:
1430         fuse_fs_cleanup();
1431  err:
1432         return res;
1433 }
1434
1435 static void __exit fuse_exit(void)
1436 {
1437         printk(KERN_DEBUG "fuse exit\n");
1438
1439         fuse_ctl_cleanup();
1440         fuse_sysfs_cleanup();
1441         fuse_fs_cleanup();
1442         fuse_dev_cleanup();
1443 }
1444
1445 module_init(fuse_init);
1446 module_exit(fuse_exit);