fs/fuse/inode.c

   1 /*
   2   FUSE: Filesystem in Userspace
   3   Copyright (C) 2001-2008  Miklos Szeredi <miklos@szeredi.hu>
   4
   5   This program can be distributed under the terms of the GNU GPL.
   6   See the file COPYING.
   7 */
   8
   9 #include "fuse_i.h"
  10
  11 #include <linux/pagemap.h>
  12 #include <linux/slab.h>
  13 #include <linux/file.h>
  14 #include <linux/seq_file.h>
  15 #include <linux/init.h>
  16 #include <linux/module.h>
  17 #include <linux/moduleparam.h>
  18 #include <linux/parser.h>
  19 #include <linux/statfs.h>
  20 #include <linux/random.h>
  21 #include <linux/sched.h>
  22 #include <linux/exportfs.h>
  23 #include <linux/posix_acl.h>
  24
  25 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  26 MODULE_DESCRIPTION("Filesystem in Userspace");
  27 MODULE_LICENSE("GPL");
  28
  29 static struct kmem_cache *fuse_inode_cachep;
  30 struct list_head fuse_conn_list;
  31 DEFINE_MUTEX(fuse_mutex);
  32
  33 static int set_global_limit(const char *val, struct kernel_param *kp);
  34
  35 unsigned max_user_bgreq;
  36 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
  37                   &max_user_bgreq, 0644);
  38 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
  39 MODULE_PARM_DESC(max_user_bgreq,
  40  "Global limit for the maximum number of backgrounded requests an "
  41  "unprivileged user can set");
  42
  43 unsigned max_user_congthresh;
  44 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
  45                   &max_user_congthresh, 0644);
  46 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
  47 MODULE_PARM_DESC(max_user_congthresh,
  48  "Global limit for the maximum congestion threshold an "
  49  "unprivileged user can set");
  50
  51 #define FUSE_SUPER_MAGIC 0x65735546
  52
  53 #define FUSE_DEFAULT_BLKSIZE 512
  54
  55 /** Maximum number of outstanding background requests */
  56 #define FUSE_DEFAULT_MAX_BACKGROUND 12
  57
  58 /** Congestion starts at 75% of maximum */
  59 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
  60
  61 struct fuse_mount_data {
  62         int fd;
  63         unsigned rootmode;
  64         kuid_t user_id;
  65         kgid_t group_id;
  66         unsigned fd_present:1;
  67         unsigned rootmode_present:1;
  68         unsigned user_id_present:1;
  69         unsigned group_id_present:1;
  70         unsigned default_permissions:1;
  71         unsigned allow_other:1;
  72         unsigned max_read;
  73         unsigned blksize;
  74 };
  75
  76 struct fuse_forget_link *fuse_alloc_forget(void)
  77 {
  78         return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL);
  79 }
  80
  81 static struct inode *fuse_alloc_inode(struct super_block *sb)
  82 {
  83         struct inode *inode;
  84         struct fuse_inode *fi;
  85
  86         inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
  87         if (!inode)
  88                 return NULL;
  89
  90         fi = get_fuse_inode(inode);
  91         fi->i_time = 0;
  92         fi->nodeid = 0;
  93         fi->nlookup = 0;
  94         fi->attr_version = 0;
  95         fi->writectr = 0;
  96         fi->orig_ino = 0;
  97         fi->state = 0;
  98         INIT_LIST_HEAD(&fi->write_files);
  99         INIT_LIST_HEAD(&fi->queued_writes);
 100         INIT_LIST_HEAD(&fi->writepages);
 101         init_waitqueue_head(&fi->page_waitq);
 102         mutex_init(&fi->mutex);
 103         fi->forget = fuse_alloc_forget();
 104         if (!fi->forget) {
 105                 kmem_cache_free(fuse_inode_cachep, inode);
 106                 return NULL;
 107         }
 108
 109         return inode;
 110 }
 111
 112 static void fuse_i_callback(struct rcu_head *head)
 113 {
 114         struct inode *inode = container_of(head, struct inode, i_rcu);
 115         kmem_cache_free(fuse_inode_cachep, inode);
 116 }
 117
 118 static void fuse_destroy_inode(struct inode *inode)
 119 {
 120         struct fuse_inode *fi = get_fuse_inode(inode);
 121         BUG_ON(!list_empty(&fi->write_files));
 122         BUG_ON(!list_empty(&fi->queued_writes));
 123         mutex_destroy(&fi->mutex);
 124         kfree(fi->forget);
 125         call_rcu(&inode->i_rcu, fuse_i_callback);
 126 }
 127
 128 static void fuse_evict_inode(struct inode *inode)
 129 {
 130         truncate_inode_pages_final(&inode->i_data);
 131         clear_inode(inode);
 132         if (inode->i_sb->s_flags & MS_ACTIVE) {
 133                 struct fuse_conn *fc = get_fuse_conn(inode);
 134                 struct fuse_inode *fi = get_fuse_inode(inode);
 135                 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
 136                 fi->forget = NULL;
 137         }
 138 }
 139
 140 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
 141 {
 142         sync_filesystem(sb);
 143         if (*flags & MS_MANDLOCK)
 144                 return -EINVAL;
 145
 146         return 0;
 147 }
 148
 149 /*
 150  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
 151  * so that it will fit.
 152  */
 153 static ino_t fuse_squash_ino(u64 ino64)
 154 {
 155         ino_t ino = (ino_t) ino64;
 156         if (sizeof(ino_t) < sizeof(u64))
 157                 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
 158         return ino;
 159 }
 160
 161 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
 162                                    u64 attr_valid)
 163 {
 164         struct fuse_conn *fc = get_fuse_conn(inode);
 165         struct fuse_inode *fi = get_fuse_inode(inode);
 166
 167         fi->attr_version = ++fc->attr_version;
 168         fi->i_time = attr_valid;
 169
 170         inode->i_ino     = fuse_squash_ino(attr->ino);
 171         inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
 172         set_nlink(inode, attr->nlink);
 173         inode->i_uid     = make_kuid(&init_user_ns, attr->uid);
 174         inode->i_gid     = make_kgid(&init_user_ns, attr->gid);
 175         inode->i_blocks  = attr->blocks;
 176
 177         /* Sanitize nsecs */
 178         attr->atimensec = min_t(u32, attr->atimensec, NSEC_PER_SEC - 1);
 179         attr->mtimensec = min_t(u32, attr->mtimensec, NSEC_PER_SEC - 1);
 180         attr->ctimensec = min_t(u32, attr->ctimensec, NSEC_PER_SEC - 1);
 181
 182         inode->i_atime.tv_sec   = attr->atime;
 183         inode->i_atime.tv_nsec  = attr->atimensec;
 184         /* mtime from server may be stale due to local buffered write */
 185         if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
 186                 inode->i_mtime.tv_sec   = attr->mtime;
 187                 inode->i_mtime.tv_nsec  = attr->mtimensec;
 188                 inode->i_ctime.tv_sec   = attr->ctime;
 189                 inode->i_ctime.tv_nsec  = attr->ctimensec;
 190         }
 191
 192         if (attr->blksize != 0)
 193                 inode->i_blkbits = ilog2(attr->blksize);
 194         else
 195                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
 196
 197         /*
 198          * Don't set the sticky bit in i_mode, unless we want the VFS
 199          * to check permissions.  This prevents failures due to the
 200          * check in may_delete().
 201          */
 202         fi->orig_i_mode = inode->i_mode;
 203         if (!fc->default_permissions)
 204                 inode->i_mode &= ~S_ISVTX;
 205
 206         fi->orig_ino = attr->ino;
 207 }
 208
 209 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
 210                             u64 attr_valid, u64 attr_version)
 211 {
 212         struct fuse_conn *fc = get_fuse_conn(inode);
 213         struct fuse_inode *fi = get_fuse_inode(inode);
 214         bool is_wb = fc->writeback_cache;
 215         loff_t oldsize;
 216         struct timespec old_mtime;
 217
 218         spin_lock(&fc->lock);
 219         if ((attr_version != 0 && fi->attr_version > attr_version) ||
 220             test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
 221                 spin_unlock(&fc->lock);
 222                 return;
 223         }
 224
 225         old_mtime = inode->i_mtime;
 226         fuse_change_attributes_common(inode, attr, attr_valid);
 227
 228         oldsize = inode->i_size;
 229         /*
 230          * In case of writeback_cache enabled, the cached writes beyond EOF
 231          * extend local i_size without keeping userspace server in sync. So,
 232          * attr->size coming from server can be stale. We cannot trust it.
 233          */
 234         if (!is_wb || !S_ISREG(inode->i_mode))
 235                 i_size_write(inode, attr->size);
 236         spin_unlock(&fc->lock);
 237
 238         if (!is_wb && S_ISREG(inode->i_mode)) {
 239                 bool inval = false;
 240
 241                 if (oldsize != attr->size) {
 242                         truncate_pagecache(inode, attr->size);
 243                         inval = true;
 244                 } else if (fc->auto_inval_data) {
 245                         struct timespec new_mtime = {
 246                                 .tv_sec = attr->mtime,
 247                                 .tv_nsec = attr->mtimensec,
 248                         };
 249
 250                         /*
 251                          * Auto inval mode also checks and invalidates if mtime
 252                          * has changed.
 253                          */
 254                         if (!timespec_equal(&old_mtime, &new_mtime))
 255                                 inval = true;
 256                 }
 257
 258                 if (inval)
 259                         invalidate_inode_pages2(inode->i_mapping);
 260         }
 261 }
 262
 263 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
 264 {
 265         inode->i_mode = attr->mode & S_IFMT;
 266         inode->i_size = attr->size;
 267         inode->i_mtime.tv_sec  = attr->mtime;
 268         inode->i_mtime.tv_nsec = attr->mtimensec;
 269         inode->i_ctime.tv_sec  = attr->ctime;
 270         inode->i_ctime.tv_nsec = attr->ctimensec;
 271         if (S_ISREG(inode->i_mode)) {
 272                 fuse_init_common(inode);
 273                 fuse_init_file_inode(inode);
 274         } else if (S_ISDIR(inode->i_mode))
 275                 fuse_init_dir(inode);
 276         else if (S_ISLNK(inode->i_mode))
 277                 fuse_init_symlink(inode);
 278         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
 279                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
 280                 fuse_init_common(inode);
 281                 init_special_inode(inode, inode->i_mode,
 282                                    new_decode_dev(attr->rdev));
 283         } else
 284                 BUG();
 285 }
 286
 287 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
 288 {
 289         u64 nodeid = *(u64 *) _nodeidp;
 290         if (get_node_id(inode) == nodeid)
 291                 return 1;
 292         else
 293                 return 0;
 294 }
 295
 296 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
 297 {
 298         u64 nodeid = *(u64 *) _nodeidp;
 299         get_fuse_inode(inode)->nodeid = nodeid;
 300         return 0;
 301 }
 302
 303 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
 304                         int generation, struct fuse_attr *attr,
 305                         u64 attr_valid, u64 attr_version)
 306 {
 307         struct inode *inode;
 308         struct fuse_inode *fi;
 309         struct fuse_conn *fc = get_fuse_conn_super(sb);
 310
 311  retry:
 312         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
 313         if (!inode)
 314                 return NULL;
 315
 316         if ((inode->i_state & I_NEW)) {
 317                 inode->i_flags |= S_NOATIME;
 318                 if (!fc->writeback_cache || !S_ISREG(attr->mode))
 319                         inode->i_flags |= S_NOCMTIME;
 320                 inode->i_generation = generation;
 321                 fuse_init_inode(inode, attr);
 322                 unlock_new_inode(inode);
 323         } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
 324                 /* Inode has changed type, any I/O on the old should fail */
 325                 fuse_make_bad(inode);
 326                 iput(inode);
 327                 goto retry;
 328         }
 329
 330         fi = get_fuse_inode(inode);
 331         spin_lock(&fc->lock);
 332         fi->nlookup++;
 333         spin_unlock(&fc->lock);
 334         fuse_change_attributes(inode, attr, attr_valid, attr_version);
 335
 336         return inode;
 337 }
 338
 339 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
 340                              loff_t offset, loff_t len)
 341 {
 342         struct inode *inode;
 343         pgoff_t pg_start;
 344         pgoff_t pg_end;
 345
 346         inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
 347         if (!inode)
 348                 return -ENOENT;
 349
 350         fuse_invalidate_attr(inode);
 351         forget_all_cached_acls(inode);
 352         if (offset >= 0) {
 353                 pg_start = offset >> PAGE_SHIFT;
 354                 if (len <= 0)
 355                         pg_end = -1;
 356                 else
 357                         pg_end = (offset + len - 1) >> PAGE_SHIFT;
 358                 invalidate_inode_pages2_range(inode->i_mapping,
 359                                               pg_start, pg_end);
 360         }
 361         iput(inode);
 362         return 0;
 363 }
 364
 365 bool fuse_lock_inode(struct inode *inode)
 366 {
 367         bool locked = false;
 368
 369         if (!get_fuse_conn(inode)->parallel_dirops) {
 370                 mutex_lock(&get_fuse_inode(inode)->mutex);
 371                 locked = true;
 372         }
 373
 374         return locked;
 375 }
 376
 377 void fuse_unlock_inode(struct inode *inode, bool locked)
 378 {
 379         if (locked)
 380                 mutex_unlock(&get_fuse_inode(inode)->mutex);
 381 }
 382
 383 static void fuse_umount_begin(struct super_block *sb)
 384 {
 385         fuse_abort_conn(get_fuse_conn_super(sb));
 386 }
 387
 388 static void fuse_send_destroy(struct fuse_conn *fc)
 389 {
 390         struct fuse_req *req = fc->destroy_req;
 391         if (req && fc->conn_init) {
 392                 fc->destroy_req = NULL;
 393                 req->in.h.opcode = FUSE_DESTROY;
 394                 __set_bit(FR_FORCE, &req->flags);
 395                 __clear_bit(FR_BACKGROUND, &req->flags);
 396                 fuse_request_send(fc, req);
 397                 fuse_put_request(fc, req);
 398         }
 399 }
 400
 401 static void fuse_bdi_destroy(struct fuse_conn *fc)
 402 {
 403         if (fc->bdi_initialized)
 404                 bdi_destroy(&fc->bdi);
 405 }
 406
 407 static void fuse_put_super(struct super_block *sb)
 408 {
 409         struct fuse_conn *fc = get_fuse_conn_super(sb);
 410
 411         mutex_lock(&fuse_mutex);
 412         list_del(&fc->entry);
 413         fuse_ctl_remove_conn(fc);
 414         mutex_unlock(&fuse_mutex);
 415         fuse_bdi_destroy(fc);
 416
 417         fuse_conn_put(fc);
 418 }
 419
 420 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
 421 {
 422         stbuf->f_type    = FUSE_SUPER_MAGIC;
 423         stbuf->f_bsize   = attr->bsize;
 424         stbuf->f_frsize  = attr->frsize;
 425         stbuf->f_blocks  = attr->blocks;
 426         stbuf->f_bfree   = attr->bfree;
 427         stbuf->f_bavail  = attr->bavail;
 428         stbuf->f_files   = attr->files;
 429         stbuf->f_ffree   = attr->ffree;
 430         stbuf->f_namelen = attr->namelen;
 431         /* fsid is left zero */
 432 }
 433
 434 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
 435 {
 436         struct super_block *sb = dentry->d_sb;
 437         struct fuse_conn *fc = get_fuse_conn_super(sb);
 438         FUSE_ARGS(args);
 439         struct fuse_statfs_out outarg;
 440         int err;
 441
 442         if (!fuse_allow_current_process(fc)) {
 443                 buf->f_type = FUSE_SUPER_MAGIC;
 444                 return 0;
 445         }
 446
 447         memset(&outarg, 0, sizeof(outarg));
 448         args.in.numargs = 0;
 449         args.in.h.opcode = FUSE_STATFS;
 450         args.in.h.nodeid = get_node_id(d_inode(dentry));
 451         args.out.numargs = 1;
 452         args.out.args[0].size = sizeof(outarg);
 453         args.out.args[0].value = &outarg;
 454         err = fuse_simple_request(fc, &args);
 455         if (!err)
 456                 convert_fuse_statfs(buf, &outarg.st);
 457         return err;
 458 }
 459
 460 enum {
 461         OPT_FD,
 462         OPT_ROOTMODE,
 463         OPT_USER_ID,
 464         OPT_GROUP_ID,
 465         OPT_DEFAULT_PERMISSIONS,
 466         OPT_ALLOW_OTHER,
 467         OPT_MAX_READ,
 468         OPT_BLKSIZE,
 469         OPT_ERR
 470 };
 471
 472 static const match_table_t tokens = {
 473         {OPT_FD,                        "fd=%u"},
 474         {OPT_ROOTMODE,                  "rootmode=%o"},
 475         {OPT_USER_ID,                   "user_id=%u"},
 476         {OPT_GROUP_ID,                  "group_id=%u"},
 477         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 478         {OPT_ALLOW_OTHER,               "allow_other"},
 479         {OPT_MAX_READ,                  "max_read=%u"},
 480         {OPT_BLKSIZE,                   "blksize=%u"},
 481         {OPT_ERR,                       NULL}
 482 };
 483
 484 static int fuse_match_uint(substring_t *s, unsigned int *res)
 485 {
 486         int err = -ENOMEM;
 487         char *buf = match_strdup(s);
 488         if (buf) {
 489                 err = kstrtouint(buf, 10, res);
 490                 kfree(buf);
 491         }
 492         return err;
 493 }
 494
 495 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
 496 {
 497         char *p;
 498         memset(d, 0, sizeof(struct fuse_mount_data));
 499         d->max_read = ~0;
 500         d->blksize = FUSE_DEFAULT_BLKSIZE;
 501
 502         while ((p = strsep(&opt, ",")) != NULL) {
 503                 int token;
 504                 int value;
 505                 unsigned uv;
 506                 substring_t args[MAX_OPT_ARGS];
 507                 if (!*p)
 508                         continue;
 509
 510                 token = match_token(p, tokens, args);
 511                 switch (token) {
 512                 case OPT_FD:
 513                         if (match_int(&args[0], &value))
 514                                 return 0;
 515                         d->fd = value;
 516                         d->fd_present = 1;
 517                         break;
 518
 519                 case OPT_ROOTMODE:
 520                         if (match_octal(&args[0], &value))
 521                                 return 0;
 522                         if (!fuse_valid_type(value))
 523                                 return 0;
 524                         d->rootmode = value;
 525                         d->rootmode_present = 1;
 526                         break;
 527
 528                 case OPT_USER_ID:
 529                         if (fuse_match_uint(&args[0], &uv))
 530                                 return 0;
 531                         d->user_id = make_kuid(current_user_ns(), uv);
 532                         if (!uid_valid(d->user_id))
 533                                 return 0;
 534                         d->user_id_present = 1;
 535                         break;
 536
 537                 case OPT_GROUP_ID:
 538                         if (fuse_match_uint(&args[0], &uv))
 539                                 return 0;
 540                         d->group_id = make_kgid(current_user_ns(), uv);
 541                         if (!gid_valid(d->group_id))
 542                                 return 0;
 543                         d->group_id_present = 1;
 544                         break;
 545
 546                 case OPT_DEFAULT_PERMISSIONS:
 547                         d->default_permissions = 1;
 548                         break;
 549
 550                 case OPT_ALLOW_OTHER:
 551                         d->allow_other = 1;
 552                         break;
 553
 554                 case OPT_MAX_READ:
 555                         if (match_int(&args[0], &value))
 556                                 return 0;
 557                         d->max_read = value;
 558                         break;
 559
 560                 case OPT_BLKSIZE:
 561                         if (!is_bdev || match_int(&args[0], &value))
 562                                 return 0;
 563                         d->blksize = value;
 564                         break;
 565
 566                 default:
 567                         return 0;
 568                 }
 569         }
 570
 571         if (!d->fd_present || !d->rootmode_present ||
 572             !d->user_id_present || !d->group_id_present)
 573                 return 0;
 574
 575         return 1;
 576 }
 577
 578 static int fuse_show_options(struct seq_file *m, struct dentry *root)
 579 {
 580         struct super_block *sb = root->d_sb;
 581         struct fuse_conn *fc = get_fuse_conn_super(sb);
 582
 583         seq_printf(m, ",user_id=%u", from_kuid_munged(&init_user_ns, fc->user_id));
 584         seq_printf(m, ",group_id=%u", from_kgid_munged(&init_user_ns, fc->group_id));
 585         if (fc->default_permissions)
 586                 seq_puts(m, ",default_permissions");
 587         if (fc->allow_other)
 588                 seq_puts(m, ",allow_other");
 589         if (fc->max_read != ~0)
 590                 seq_printf(m, ",max_read=%u", fc->max_read);
 591         if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
 592                 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
 593         return 0;
 594 }
 595
 596 static void fuse_iqueue_init(struct fuse_iqueue *fiq)
 597 {
 598         memset(fiq, 0, sizeof(struct fuse_iqueue));
 599         init_waitqueue_head(&fiq->waitq);
 600         INIT_LIST_HEAD(&fiq->pending);
 601         INIT_LIST_HEAD(&fiq->interrupts);
 602         fiq->forget_list_tail = &fiq->forget_list_head;
 603         fiq->connected = 1;
 604 }
 605
 606 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
 607 {
 608         memset(fpq, 0, sizeof(struct fuse_pqueue));
 609         spin_lock_init(&fpq->lock);
 610         INIT_LIST_HEAD(&fpq->processing);
 611         INIT_LIST_HEAD(&fpq->io);
 612         fpq->connected = 1;
 613 }
 614
 615 void fuse_conn_init(struct fuse_conn *fc)
 616 {
 617         memset(fc, 0, sizeof(*fc));
 618         spin_lock_init(&fc->lock);
 619         init_rwsem(&fc->killsb);
 620         atomic_set(&fc->count, 1);
 621         atomic_set(&fc->dev_count, 1);
 622         init_waitqueue_head(&fc->blocked_waitq);
 623         init_waitqueue_head(&fc->reserved_req_waitq);
 624         fuse_iqueue_init(&fc->iq);
 625         INIT_LIST_HEAD(&fc->bg_queue);
 626         INIT_LIST_HEAD(&fc->entry);
 627         INIT_LIST_HEAD(&fc->devices);
 628         atomic_set(&fc->num_waiting, 0);
 629         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
 630         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
 631         fc->khctr = 0;
 632         fc->polled_files = RB_ROOT;
 633         fc->blocked = 0;
 634         fc->initialized = 0;
 635         fc->connected = 1;
 636         fc->attr_version = 1;
 637         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
 638 }
 639 EXPORT_SYMBOL_GPL(fuse_conn_init);
 640
 641 void fuse_conn_put(struct fuse_conn *fc)
 642 {
 643         if (atomic_dec_and_test(&fc->count)) {
 644                 if (fc->destroy_req)
 645                         fuse_request_free(fc->destroy_req);
 646                 fc->release(fc);
 647         }
 648 }
 649 EXPORT_SYMBOL_GPL(fuse_conn_put);
 650
 651 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
 652 {
 653         atomic_inc(&fc->count);
 654         return fc;
 655 }
 656 EXPORT_SYMBOL_GPL(fuse_conn_get);
 657
 658 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
 659 {
 660         struct fuse_attr attr;
 661         memset(&attr, 0, sizeof(attr));
 662
 663         attr.mode = mode;
 664         attr.ino = FUSE_ROOT_ID;
 665         attr.nlink = 1;
 666         return fuse_iget(sb, 1, 0, &attr, 0, 0);
 667 }
 668
 669 struct fuse_inode_handle {
 670         u64 nodeid;
 671         u32 generation;
 672 };
 673
 674 static struct dentry *fuse_get_dentry(struct super_block *sb,
 675                                       struct fuse_inode_handle *handle)
 676 {
 677         struct fuse_conn *fc = get_fuse_conn_super(sb);
 678         struct inode *inode;
 679         struct dentry *entry;
 680         int err = -ESTALE;
 681
 682         if (handle->nodeid == 0)
 683                 goto out_err;
 684
 685         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
 686         if (!inode) {
 687                 struct fuse_entry_out outarg;
 688                 const struct qstr name = QSTR_INIT(".", 1);
 689
 690                 if (!fc->export_support)
 691                         goto out_err;
 692
 693                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
 694                                        &inode);
 695                 if (err && err != -ENOENT)
 696                         goto out_err;
 697                 if (err || !inode) {
 698                         err = -ESTALE;
 699                         goto out_err;
 700                 }
 701                 err = -EIO;
 702                 if (get_node_id(inode) != handle->nodeid)
 703                         goto out_iput;
 704         }
 705         err = -ESTALE;
 706         if (inode->i_generation != handle->generation)
 707                 goto out_iput;
 708
 709         entry = d_obtain_alias(inode);
 710         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
 711                 fuse_invalidate_entry_cache(entry);
 712
 713         return entry;
 714
 715  out_iput:
 716         iput(inode);
 717  out_err:
 718         return ERR_PTR(err);
 719 }
 720
 721 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
 722                            struct inode *parent)
 723 {
 724         int len = parent ? 6 : 3;
 725         u64 nodeid;
 726         u32 generation;
 727
 728         if (*max_len < len) {
 729                 *max_len = len;
 730                 return  FILEID_INVALID;
 731         }
 732
 733         nodeid = get_fuse_inode(inode)->nodeid;
 734         generation = inode->i_generation;
 735
 736         fh[0] = (u32)(nodeid >> 32);
 737         fh[1] = (u32)(nodeid & 0xffffffff);
 738         fh[2] = generation;
 739
 740         if (parent) {
 741                 nodeid = get_fuse_inode(parent)->nodeid;
 742                 generation = parent->i_generation;
 743
 744                 fh[3] = (u32)(nodeid >> 32);
 745                 fh[4] = (u32)(nodeid & 0xffffffff);
 746                 fh[5] = generation;
 747         }
 748
 749         *max_len = len;
 750         return parent ? 0x82 : 0x81;
 751 }
 752
 753 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
 754                 struct fid *fid, int fh_len, int fh_type)
 755 {
 756         struct fuse_inode_handle handle;
 757
 758         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
 759                 return NULL;
 760
 761         handle.nodeid = (u64) fid->raw[0] << 32;
 762         handle.nodeid |= (u64) fid->raw[1];
 763         handle.generation = fid->raw[2];
 764         return fuse_get_dentry(sb, &handle);
 765 }
 766
 767 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
 768                 struct fid *fid, int fh_len, int fh_type)
 769 {
 770         struct fuse_inode_handle parent;
 771
 772         if (fh_type != 0x82 || fh_len < 6)
 773                 return NULL;
 774
 775         parent.nodeid = (u64) fid->raw[3] << 32;
 776         parent.nodeid |= (u64) fid->raw[4];
 777         parent.generation = fid->raw[5];
 778         return fuse_get_dentry(sb, &parent);
 779 }
 780
 781 static struct dentry *fuse_get_parent(struct dentry *child)
 782 {
 783         struct inode *child_inode = d_inode(child);
 784         struct fuse_conn *fc = get_fuse_conn(child_inode);
 785         struct inode *inode;
 786         struct dentry *parent;
 787         struct fuse_entry_out outarg;
 788         const struct qstr name = QSTR_INIT("..", 2);
 789         int err;
 790
 791         if (!fc->export_support)
 792                 return ERR_PTR(-ESTALE);
 793
 794         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
 795                                &name, &outarg, &inode);
 796         if (err) {
 797                 if (err == -ENOENT)
 798                         return ERR_PTR(-ESTALE);
 799                 return ERR_PTR(err);
 800         }
 801
 802         parent = d_obtain_alias(inode);
 803         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
 804                 fuse_invalidate_entry_cache(parent);
 805
 806         return parent;
 807 }
 808
 809 static const struct export_operations fuse_export_operations = {
 810         .fh_to_dentry   = fuse_fh_to_dentry,
 811         .fh_to_parent   = fuse_fh_to_parent,
 812         .encode_fh      = fuse_encode_fh,
 813         .get_parent     = fuse_get_parent,
 814 };
 815
 816 static const struct super_operations fuse_super_operations = {
 817         .alloc_inode    = fuse_alloc_inode,
 818         .destroy_inode  = fuse_destroy_inode,
 819         .evict_inode    = fuse_evict_inode,
 820         .write_inode    = fuse_write_inode,
 821         .drop_inode     = generic_delete_inode,
 822         .remount_fs     = fuse_remount_fs,
 823         .put_super      = fuse_put_super,
 824         .umount_begin   = fuse_umount_begin,
 825         .statfs         = fuse_statfs,
 826         .show_options   = fuse_show_options,
 827 };
 828
 829 static void sanitize_global_limit(unsigned *limit)
 830 {
 831         if (*limit == 0)
 832                 *limit = ((totalram_pages << PAGE_SHIFT) >> 13) /
 833                          sizeof(struct fuse_req);
 834
 835         if (*limit >= 1 << 16)
 836                 *limit = (1 << 16) - 1;
 837 }
 838
 839 static int set_global_limit(const char *val, struct kernel_param *kp)
 840 {
 841         int rv;
 842
 843         rv = param_set_uint(val, kp);
 844         if (rv)
 845                 return rv;
 846
 847         sanitize_global_limit((unsigned *)kp->arg);
 848
 849         return 0;
 850 }
 851
 852 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
 853 {
 854         int cap_sys_admin = capable(CAP_SYS_ADMIN);
 855
 856         if (arg->minor < 13)
 857                 return;
 858
 859         sanitize_global_limit(&max_user_bgreq);
 860         sanitize_global_limit(&max_user_congthresh);
 861
 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 }
 876
 877 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
 878 {
 879         struct fuse_init_out *arg = &req->misc.init_out;
 880
 881         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
 882                 fc->conn_error = 1;
 883         else {
 884                 unsigned long ra_pages;
 885
 886                 process_init_limits(fc, arg);
 887
 888                 if (arg->minor >= 6) {
 889                         ra_pages = arg->max_readahead / PAGE_SIZE;
 890                         if (arg->flags & FUSE_ASYNC_READ)
 891                                 fc->async_read = 1;
 892                         if (!(arg->flags & FUSE_POSIX_LOCKS))
 893                                 fc->no_lock = 1;
 894                         if (arg->minor >= 17) {
 895                                 if (!(arg->flags & FUSE_FLOCK_LOCKS))
 896                                         fc->no_flock = 1;
 897                         } else {
 898                                 if (!(arg->flags & FUSE_POSIX_LOCKS))
 899                                         fc->no_flock = 1;
 900                         }
 901                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
 902                                 fc->atomic_o_trunc = 1;
 903                         if (arg->minor >= 9) {
 904                                 /* LOOKUP has dependency on proto version */
 905                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
 906                                         fc->export_support = 1;
 907                         }
 908                         if (arg->flags & FUSE_BIG_WRITES)
 909                                 fc->big_writes = 1;
 910                         if (arg->flags & FUSE_DONT_MASK)
 911                                 fc->dont_mask = 1;
 912                         if (arg->flags & FUSE_AUTO_INVAL_DATA)
 913                                 fc->auto_inval_data = 1;
 914                         if (arg->flags & FUSE_DO_READDIRPLUS) {
 915                                 fc->do_readdirplus = 1;
 916                                 if (arg->flags & FUSE_READDIRPLUS_AUTO)
 917                                         fc->readdirplus_auto = 1;
 918                         }
 919                         if (arg->flags & FUSE_ASYNC_DIO)
 920                                 fc->async_dio = 1;
 921                         if (arg->flags & FUSE_WRITEBACK_CACHE)
 922                                 fc->writeback_cache = 1;
 923                         if (arg->flags & FUSE_PARALLEL_DIROPS)
 924                                 fc->parallel_dirops = 1;
 925                         if (arg->flags & FUSE_HANDLE_KILLPRIV)
 926                                 fc->handle_killpriv = 1;
 927                         if (arg->time_gran && arg->time_gran <= 1000000000)
 928                                 fc->sb->s_time_gran = arg->time_gran;
 929                         if ((arg->flags & FUSE_POSIX_ACL)) {
 930                                 fc->default_permissions = 1;
 931                                 fc->posix_acl = 1;
 932                                 fc->sb->s_xattr = fuse_acl_xattr_handlers;
 933                         }
 934                 } else {
 935                         ra_pages = fc->max_read / PAGE_SIZE;
 936                         fc->no_lock = 1;
 937                         fc->no_flock = 1;
 938                 }
 939
 940                 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
 941                 fc->minor = arg->minor;
 942                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
 943                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
 944                 fc->conn_init = 1;
 945         }
 946         fuse_set_initialized(fc);
 947         wake_up_all(&fc->blocked_waitq);
 948 }
 949
 950 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
 951 {
 952         struct fuse_init_in *arg = &req->misc.init_in;
 953
 954         arg->major = FUSE_KERNEL_VERSION;
 955         arg->minor = FUSE_KERNEL_MINOR_VERSION;
 956         arg->max_readahead = fc->bdi.ra_pages * PAGE_SIZE;
 957         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
 958                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
 959                 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
 960                 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
 961                 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
 962                 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
 963                 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL;
 964         req->in.h.opcode = FUSE_INIT;
 965         req->in.numargs = 1;
 966         req->in.args[0].size = sizeof(*arg);
 967         req->in.args[0].value = arg;
 968         req->out.numargs = 1;
 969         /* Variable length argument used for backward compatibility
 970            with interface version < 7.5.  Rest of init_out is zeroed
 971            by do_get_request(), so a short reply is not a problem */
 972         req->out.argvar = 1;
 973         req->out.args[0].size = sizeof(struct fuse_init_out);
 974         req->out.args[0].value = &req->misc.init_out;
 975         req->end = process_init_reply;
 976         fuse_request_send_background(fc, req);
 977 }
 978
 979 static void fuse_free_conn(struct fuse_conn *fc)
 980 {
 981         WARN_ON(!list_empty(&fc->devices));
 982         kfree_rcu(fc, rcu);
 983 }
 984
 985 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
 986 {
 987         int err;
 988
 989         fc->bdi.name = "fuse";
 990         fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
 991         /* fuse does it's own writeback accounting */
 992         fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
 993
 994         err = bdi_init(&fc->bdi);
 995         if (err)
 996                 return err;
 997
 998         fc->bdi_initialized = 1;
 999
1000         if (sb->s_bdev) {
1001                 err =  bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
1002                                     MAJOR(fc->dev), MINOR(fc->dev));
1003         } else {
1004                 err = bdi_register_dev(&fc->bdi, fc->dev);
1005         }
1006
1007         if (err)
1008                 return err;
1009
1010         /*
1011          * For a single fuse filesystem use max 1% of dirty +
1012          * writeback threshold.
1013          *
1014          * This gives about 1M of write buffer for memory maps on a
1015          * machine with 1G and 10% dirty_ratio, which should be more
1016          * than enough.
1017          *
1018          * Privileged users can raise it by writing to
1019          *
1020          *    /sys/class/bdi/<bdi>/max_ratio
1021          */
1022         bdi_set_max_ratio(&fc->bdi, 1);
1023
1024         return 0;
1025 }
1026
1027 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc)
1028 {
1029         struct fuse_dev *fud;
1030
1031         fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1032         if (fud) {
1033                 fud->fc = fuse_conn_get(fc);
1034                 fuse_pqueue_init(&fud->pq);
1035
1036                 spin_lock(&fc->lock);
1037                 list_add_tail(&fud->entry, &fc->devices);
1038                 spin_unlock(&fc->lock);
1039         }
1040
1041         return fud;
1042 }
1043 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1044
1045 void fuse_dev_free(struct fuse_dev *fud)
1046 {
1047         struct fuse_conn *fc = fud->fc;
1048
1049         if (fc) {
1050                 spin_lock(&fc->lock);
1051                 list_del(&fud->entry);
1052                 spin_unlock(&fc->lock);
1053
1054                 fuse_conn_put(fc);
1055         }
1056         kfree(fud);
1057 }
1058 EXPORT_SYMBOL_GPL(fuse_dev_free);
1059
1060 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
1061 {
1062         struct fuse_dev *fud;
1063         struct fuse_conn *fc;
1064         struct inode *root;
1065         struct fuse_mount_data d;
1066         struct file *file;
1067         struct dentry *root_dentry;
1068         struct fuse_req *init_req;
1069         int err;
1070         int is_bdev = sb->s_bdev != NULL;
1071
1072         err = -EINVAL;
1073         if (sb->s_flags & MS_MANDLOCK)
1074                 goto err;
1075
1076         sb->s_flags &= ~(MS_NOSEC | MS_I_VERSION);
1077
1078         if (!parse_fuse_opt(data, &d, is_bdev))
1079                 goto err;
1080
1081         if (is_bdev) {
1082 #ifdef CONFIG_BLOCK
1083                 err = -EINVAL;
1084                 if (!sb_set_blocksize(sb, d.blksize))
1085                         goto err;
1086 #endif
1087         } else {
1088                 sb->s_blocksize = PAGE_SIZE;
1089                 sb->s_blocksize_bits = PAGE_SHIFT;
1090         }
1091         sb->s_magic = FUSE_SUPER_MAGIC;
1092         sb->s_op = &fuse_super_operations;
1093         sb->s_xattr = fuse_xattr_handlers;
1094         sb->s_maxbytes = MAX_LFS_FILESIZE;
1095         sb->s_time_gran = 1;
1096         sb->s_export_op = &fuse_export_operations;
1097
1098         file = fget(d.fd);
1099         err = -EINVAL;
1100         if (!file)
1101                 goto err;
1102
1103         if ((file->f_op != &fuse_dev_operations) ||
1104             (file->f_cred->user_ns != &init_user_ns))
1105                 goto err_fput;
1106
1107         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1108         err = -ENOMEM;
1109         if (!fc)
1110                 goto err_fput;
1111
1112         fuse_conn_init(fc);
1113         fc->release = fuse_free_conn;
1114
1115         fud = fuse_dev_alloc(fc);
1116         if (!fud)
1117                 goto err_put_conn;
1118
1119         fc->dev = sb->s_dev;
1120         fc->sb = sb;
1121         err = fuse_bdi_init(fc, sb);
1122         if (err)
1123                 goto err_dev_free;
1124
1125         sb->s_bdi = &fc->bdi;
1126
1127         /* Handle umasking inside the fuse code */
1128         if (sb->s_flags & MS_POSIXACL)
1129                 fc->dont_mask = 1;
1130         sb->s_flags |= MS_POSIXACL;
1131
1132         fc->default_permissions = d.default_permissions;
1133         fc->allow_other = d.allow_other;
1134         fc->user_id = d.user_id;
1135         fc->group_id = d.group_id;
1136         fc->max_read = max_t(unsigned, 4096, d.max_read);
1137
1138         /* Used by get_root_inode() */
1139         sb->s_fs_info = fc;
1140
1141         err = -ENOMEM;
1142         root = fuse_get_root_inode(sb, d.rootmode);
1143         sb->s_d_op = &fuse_root_dentry_operations;
1144         root_dentry = d_make_root(root);
1145         if (!root_dentry)
1146                 goto err_dev_free;
1147         /* Root dentry doesn't have .d_revalidate */
1148         sb->s_d_op = &fuse_dentry_operations;
1149
1150         init_req = fuse_request_alloc(0);
1151         if (!init_req)
1152                 goto err_put_root;
1153         __set_bit(FR_BACKGROUND, &init_req->flags);
1154
1155         if (is_bdev) {
1156                 fc->destroy_req = fuse_request_alloc(0);
1157                 if (!fc->destroy_req)
1158                         goto err_free_init_req;
1159         }
1160
1161         mutex_lock(&fuse_mutex);
1162         err = -EINVAL;
1163         if (file->private_data)
1164                 goto err_unlock;
1165
1166         err = fuse_ctl_add_conn(fc);
1167         if (err)
1168                 goto err_unlock;
1169
1170         list_add_tail(&fc->entry, &fuse_conn_list);
1171         sb->s_root = root_dentry;
1172         file->private_data = fud;
1173         mutex_unlock(&fuse_mutex);
1174         /*
1175          * atomic_dec_and_test() in fput() provides the necessary
1176          * memory barrier for file->private_data to be visible on all
1177          * CPUs after this
1178          */
1179         fput(file);
1180
1181         fuse_send_init(fc, init_req);
1182
1183         return 0;
1184
1185  err_unlock:
1186         mutex_unlock(&fuse_mutex);
1187  err_free_init_req:
1188         fuse_request_free(init_req);
1189  err_put_root:
1190         dput(root_dentry);
1191  err_dev_free:
1192         fuse_dev_free(fud);
1193  err_put_conn:
1194         fuse_bdi_destroy(fc);
1195         fuse_conn_put(fc);
1196         sb->s_fs_info = NULL;
1197  err_fput:
1198         fput(file);
1199  err:
1200         return err;
1201 }
1202
1203 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1204                        int flags, const char *dev_name,
1205                        void *raw_data)
1206 {
1207         return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1208 }
1209
1210 static void fuse_sb_destroy(struct super_block *sb)
1211 {
1212         struct fuse_conn *fc = get_fuse_conn_super(sb);
1213
1214         if (fc) {
1215                 fuse_send_destroy(fc);
1216
1217                 fuse_abort_conn(fc);
1218                 fuse_wait_aborted(fc);
1219
1220                 down_write(&fc->killsb);
1221                 fc->sb = NULL;
1222                 up_write(&fc->killsb);
1223         }
1224 }
1225
1226 static void fuse_kill_sb_anon(struct super_block *sb)
1227 {
1228         fuse_sb_destroy(sb);
1229         kill_anon_super(sb);
1230 }
1231
1232 static struct file_system_type fuse_fs_type = {
1233         .owner          = THIS_MODULE,
1234         .name           = "fuse",
1235         .fs_flags       = FS_HAS_SUBTYPE,
1236         .mount          = fuse_mount,
1237         .kill_sb        = fuse_kill_sb_anon,
1238 };
1239 MODULE_ALIAS_FS("fuse");
1240
1241 #ifdef CONFIG_BLOCK
1242 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1243                            int flags, const char *dev_name,
1244                            void *raw_data)
1245 {
1246         return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1247 }
1248
1249 static void fuse_kill_sb_blk(struct super_block *sb)
1250 {
1251         fuse_sb_destroy(sb);
1252         kill_block_super(sb);
1253 }
1254
1255 static struct file_system_type fuseblk_fs_type = {
1256         .owner          = THIS_MODULE,
1257         .name           = "fuseblk",
1258         .mount          = fuse_mount_blk,
1259         .kill_sb        = fuse_kill_sb_blk,
1260         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1261 };
1262 MODULE_ALIAS_FS("fuseblk");
1263
1264 static inline int register_fuseblk(void)
1265 {
1266         return register_filesystem(&fuseblk_fs_type);
1267 }
1268
1269 static inline void unregister_fuseblk(void)
1270 {
1271         unregister_filesystem(&fuseblk_fs_type);
1272 }
1273 #else
1274 static inline int register_fuseblk(void)
1275 {
1276         return 0;
1277 }
1278
1279 static inline void unregister_fuseblk(void)
1280 {
1281 }
1282 #endif
1283
1284 static void fuse_inode_init_once(void *foo)
1285 {
1286         struct inode *inode = foo;
1287
1288         inode_init_once(inode);
1289 }
1290
1291 static int __init fuse_fs_init(void)
1292 {
1293         int err;
1294
1295         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1296                                               sizeof(struct fuse_inode), 0,
1297                                               SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT,
1298                                               fuse_inode_init_once);
1299         err = -ENOMEM;
1300         if (!fuse_inode_cachep)
1301                 goto out;
1302
1303         err = register_fuseblk();
1304         if (err)
1305                 goto out2;
1306
1307         err = register_filesystem(&fuse_fs_type);
1308         if (err)
1309                 goto out3;
1310
1311         return 0;
1312
1313  out3:
1314         unregister_fuseblk();
1315  out2:
1316         kmem_cache_destroy(fuse_inode_cachep);
1317  out:
1318         return err;
1319 }
1320
1321 static void fuse_fs_cleanup(void)
1322 {
1323         unregister_filesystem(&fuse_fs_type);
1324         unregister_fuseblk();
1325
1326         /*
1327          * Make sure all delayed rcu free inodes are flushed before we
1328          * destroy cache.
1329          */
1330         rcu_barrier();
1331         kmem_cache_destroy(fuse_inode_cachep);
1332 }
1333
1334 static struct kobject *fuse_kobj;
1335
1336 static int fuse_sysfs_init(void)
1337 {
1338         int err;
1339
1340         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1341         if (!fuse_kobj) {
1342                 err = -ENOMEM;
1343                 goto out_err;
1344         }
1345
1346         err = sysfs_create_mount_point(fuse_kobj, "connections");
1347         if (err)
1348                 goto out_fuse_unregister;
1349
1350         return 0;
1351
1352  out_fuse_unregister:
1353         kobject_put(fuse_kobj);
1354  out_err:
1355         return err;
1356 }
1357
1358 static void fuse_sysfs_cleanup(void)
1359 {
1360         sysfs_remove_mount_point(fuse_kobj, "connections");
1361         kobject_put(fuse_kobj);
1362 }
1363
1364 static int __init fuse_init(void)
1365 {
1366         int res;
1367
1368         printk(KERN_INFO "fuse init (API version %i.%i)\n",
1369                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1370
1371         INIT_LIST_HEAD(&fuse_conn_list);
1372         res = fuse_fs_init();
1373         if (res)
1374                 goto err;
1375
1376         res = fuse_dev_init();
1377         if (res)
1378                 goto err_fs_cleanup;
1379
1380         res = fuse_sysfs_init();
1381         if (res)
1382                 goto err_dev_cleanup;
1383
1384         res = fuse_ctl_init();
1385         if (res)
1386                 goto err_sysfs_cleanup;
1387
1388         sanitize_global_limit(&max_user_bgreq);
1389         sanitize_global_limit(&max_user_congthresh);
1390
1391         return 0;
1392
1393  err_sysfs_cleanup:
1394         fuse_sysfs_cleanup();
1395  err_dev_cleanup:
1396         fuse_dev_cleanup();
1397  err_fs_cleanup:
1398         fuse_fs_cleanup();
1399  err:
1400         return res;
1401 }
1402
1403 static void __exit fuse_exit(void)
1404 {
1405         printk(KERN_DEBUG "fuse exit\n");
1406
1407         fuse_ctl_cleanup();
1408         fuse_sysfs_cleanup();
1409         fuse_fs_cleanup();
1410         fuse_dev_cleanup();
1411 }
1412
1413 module_init(fuse_init);
1414 module_exit(fuse_exit);