fs/fuse/inode.c

   1 /*
   2   FUSE: Filesystem in Userspace
   3   Copyright (C) 2001-2008  Miklos Szeredi <miklos@szeredi.hu>
   4
   5   This program can be distributed under the terms of the GNU GPL.
   6   See the file COPYING.
   7 */
   8
   9 #include "fuse_i.h"
  10
  11 #include <linux/pagemap.h>
  12 #include <linux/slab.h>
  13 #include <linux/file.h>
  14 #include <linux/seq_file.h>
  15 #include <linux/init.h>
  16 #include <linux/module.h>
  17 #include <linux/moduleparam.h>
  18 #include <linux/parser.h>
  19 #include <linux/statfs.h>
  20 #include <linux/random.h>
  21 #include <linux/sched.h>
  22 #include <linux/exportfs.h>
  23 #include <linux/posix_acl.h>
  24 #include <linux/pid_namespace.h>
  25
  26 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  27 MODULE_DESCRIPTION("Filesystem in Userspace");
  28 MODULE_LICENSE("GPL");
  29
  30 static struct kmem_cache *fuse_inode_cachep;
  31 struct list_head fuse_conn_list;
  32 DEFINE_MUTEX(fuse_mutex);
  33
  34 static int set_global_limit(const char *val, struct kernel_param *kp);
  35
  36 unsigned max_user_bgreq;
  37 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
  38                   &max_user_bgreq, 0644);
  39 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
  40 MODULE_PARM_DESC(max_user_bgreq,
  41  "Global limit for the maximum number of backgrounded requests an "
  42  "unprivileged user can set");
  43
  44 unsigned max_user_congthresh;
  45 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
  46                   &max_user_congthresh, 0644);
  47 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
  48 MODULE_PARM_DESC(max_user_congthresh,
  49  "Global limit for the maximum congestion threshold an "
  50  "unprivileged user can set");
  51
  52 #define FUSE_SUPER_MAGIC 0x65735546
  53
  54 #define FUSE_DEFAULT_BLKSIZE 512
  55
  56 /** Maximum number of outstanding background requests */
  57 #define FUSE_DEFAULT_MAX_BACKGROUND 12
  58
  59 /** Congestion starts at 75% of maximum */
  60 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
  61
  62 struct fuse_mount_data {
  63         int fd;
  64         unsigned rootmode;
  65         kuid_t user_id;
  66         kgid_t group_id;
  67         unsigned fd_present:1;
  68         unsigned rootmode_present:1;
  69         unsigned user_id_present:1;
  70         unsigned group_id_present:1;
  71         unsigned default_permissions:1;
  72         unsigned allow_other:1;
  73         unsigned max_read;
  74         unsigned blksize;
  75 };
  76
  77 struct fuse_forget_link *fuse_alloc_forget(void)
  78 {
  79         return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL);
  80 }
  81
  82 static struct inode *fuse_alloc_inode(struct super_block *sb)
  83 {
  84         struct inode *inode;
  85         struct fuse_inode *fi;
  86
  87         inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
  88         if (!inode)
  89                 return NULL;
  90
  91         fi = get_fuse_inode(inode);
  92         fi->i_time = 0;
  93         fi->nodeid = 0;
  94         fi->nlookup = 0;
  95         fi->attr_version = 0;
  96         fi->writectr = 0;
  97         fi->orig_ino = 0;
  98         fi->state = 0;
  99         INIT_LIST_HEAD(&fi->write_files);
 100         INIT_LIST_HEAD(&fi->queued_writes);
 101         INIT_LIST_HEAD(&fi->writepages);
 102         init_waitqueue_head(&fi->page_waitq);
 103         mutex_init(&fi->mutex);
 104         fi->forget = fuse_alloc_forget();
 105         if (!fi->forget) {
 106                 kmem_cache_free(fuse_inode_cachep, inode);
 107                 return NULL;
 108         }
 109
 110         return inode;
 111 }
 112
 113 static void fuse_i_callback(struct rcu_head *head)
 114 {
 115         struct inode *inode = container_of(head, struct inode, i_rcu);
 116         kmem_cache_free(fuse_inode_cachep, inode);
 117 }
 118
 119 static void fuse_destroy_inode(struct inode *inode)
 120 {
 121         struct fuse_inode *fi = get_fuse_inode(inode);
 122         BUG_ON(!list_empty(&fi->write_files));
 123         BUG_ON(!list_empty(&fi->queued_writes));
 124         mutex_destroy(&fi->mutex);
 125         kfree(fi->forget);
 126         call_rcu(&inode->i_rcu, fuse_i_callback);
 127 }
 128
 129 static void fuse_evict_inode(struct inode *inode)
 130 {
 131         truncate_inode_pages_final(&inode->i_data);
 132         clear_inode(inode);
 133         if (inode->i_sb->s_flags & MS_ACTIVE) {
 134                 struct fuse_conn *fc = get_fuse_conn(inode);
 135                 struct fuse_inode *fi = get_fuse_inode(inode);
 136                 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
 137                 fi->forget = NULL;
 138         }
 139 }
 140
 141 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
 142 {
 143         sync_filesystem(sb);
 144         if (*flags & MS_MANDLOCK)
 145                 return -EINVAL;
 146
 147         return 0;
 148 }
 149
 150 /*
 151  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
 152  * so that it will fit.
 153  */
 154 static ino_t fuse_squash_ino(u64 ino64)
 155 {
 156         ino_t ino = (ino_t) ino64;
 157         if (sizeof(ino_t) < sizeof(u64))
 158                 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
 159         return ino;
 160 }
 161
 162 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
 163                                    u64 attr_valid)
 164 {
 165         struct fuse_conn *fc = get_fuse_conn(inode);
 166         struct fuse_inode *fi = get_fuse_inode(inode);
 167
 168         fi->attr_version = ++fc->attr_version;
 169         fi->i_time = attr_valid;
 170
 171         inode->i_ino     = fuse_squash_ino(attr->ino);
 172         inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
 173         set_nlink(inode, attr->nlink);
 174         inode->i_uid     = make_kuid(fc->user_ns, attr->uid);
 175         inode->i_gid     = make_kgid(fc->user_ns, attr->gid);
 176         inode->i_blocks  = attr->blocks;
 177         inode->i_atime.tv_sec   = attr->atime;
 178         inode->i_atime.tv_nsec  = attr->atimensec;
 179         /* mtime from server may be stale due to local buffered write */
 180         if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
 181                 inode->i_mtime.tv_sec   = attr->mtime;
 182                 inode->i_mtime.tv_nsec  = attr->mtimensec;
 183                 inode->i_ctime.tv_sec   = attr->ctime;
 184                 inode->i_ctime.tv_nsec  = attr->ctimensec;
 185         }
 186
 187         if (attr->blksize != 0)
 188                 inode->i_blkbits = ilog2(attr->blksize);
 189         else
 190                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
 191
 192         /*
 193          * Don't set the sticky bit in i_mode, unless we want the VFS
 194          * to check permissions.  This prevents failures due to the
 195          * check in may_delete().
 196          */
 197         fi->orig_i_mode = inode->i_mode;
 198         if (!fc->default_permissions)
 199                 inode->i_mode &= ~S_ISVTX;
 200
 201         fi->orig_ino = attr->ino;
 202 }
 203
 204 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
 205                             u64 attr_valid, u64 attr_version)
 206 {
 207         struct fuse_conn *fc = get_fuse_conn(inode);
 208         struct fuse_inode *fi = get_fuse_inode(inode);
 209         bool is_wb = fc->writeback_cache;
 210         loff_t oldsize;
 211         struct timespec old_mtime;
 212
 213         spin_lock(&fc->lock);
 214         if ((attr_version != 0 && fi->attr_version > attr_version) ||
 215             test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
 216                 spin_unlock(&fc->lock);
 217                 return;
 218         }
 219
 220         old_mtime = inode->i_mtime;
 221         fuse_change_attributes_common(inode, attr, attr_valid);
 222
 223         oldsize = inode->i_size;
 224         /*
 225          * In case of writeback_cache enabled, the cached writes beyond EOF
 226          * extend local i_size without keeping userspace server in sync. So,
 227          * attr->size coming from server can be stale. We cannot trust it.
 228          */
 229         if (!is_wb || !S_ISREG(inode->i_mode))
 230                 i_size_write(inode, attr->size);
 231         spin_unlock(&fc->lock);
 232
 233         if (!is_wb && S_ISREG(inode->i_mode)) {
 234                 bool inval = false;
 235
 236                 if (oldsize != attr->size) {
 237                         truncate_pagecache(inode, attr->size);
 238                         inval = true;
 239                 } else if (fc->auto_inval_data) {
 240                         struct timespec new_mtime = {
 241                                 .tv_sec = attr->mtime,
 242                                 .tv_nsec = attr->mtimensec,
 243                         };
 244
 245                         /*
 246                          * Auto inval mode also checks and invalidates if mtime
 247                          * has changed.
 248                          */
 249                         if (!timespec_equal(&old_mtime, &new_mtime))
 250                                 inval = true;
 251                 }
 252
 253                 if (inval)
 254                         invalidate_inode_pages2(inode->i_mapping);
 255         }
 256 }
 257
 258 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
 259 {
 260         inode->i_mode = attr->mode & S_IFMT;
 261         inode->i_size = attr->size;
 262         inode->i_mtime.tv_sec  = attr->mtime;
 263         inode->i_mtime.tv_nsec = attr->mtimensec;
 264         inode->i_ctime.tv_sec  = attr->ctime;
 265         inode->i_ctime.tv_nsec = attr->ctimensec;
 266         if (S_ISREG(inode->i_mode)) {
 267                 fuse_init_common(inode);
 268                 fuse_init_file_inode(inode);
 269         } else if (S_ISDIR(inode->i_mode))
 270                 fuse_init_dir(inode);
 271         else if (S_ISLNK(inode->i_mode))
 272                 fuse_init_symlink(inode);
 273         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
 274                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
 275                 fuse_init_common(inode);
 276                 init_special_inode(inode, inode->i_mode,
 277                                    new_decode_dev(attr->rdev));
 278         } else
 279                 BUG();
 280 }
 281
 282 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
 283 {
 284         u64 nodeid = *(u64 *) _nodeidp;
 285         if (get_node_id(inode) == nodeid)
 286                 return 1;
 287         else
 288                 return 0;
 289 }
 290
 291 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
 292 {
 293         u64 nodeid = *(u64 *) _nodeidp;
 294         get_fuse_inode(inode)->nodeid = nodeid;
 295         return 0;
 296 }
 297
 298 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
 299                         int generation, struct fuse_attr *attr,
 300                         u64 attr_valid, u64 attr_version)
 301 {
 302         struct inode *inode;
 303         struct fuse_inode *fi;
 304         struct fuse_conn *fc = get_fuse_conn_super(sb);
 305
 306  retry:
 307         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
 308         if (!inode)
 309                 return NULL;
 310
 311         if ((inode->i_state & I_NEW)) {
 312                 inode->i_flags |= S_NOATIME;
 313                 if (!fc->writeback_cache || !S_ISREG(attr->mode))
 314                         inode->i_flags |= S_NOCMTIME;
 315                 inode->i_generation = generation;
 316                 fuse_init_inode(inode, attr);
 317                 unlock_new_inode(inode);
 318         } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
 319                 /* Inode has changed type, any I/O on the old should fail */
 320                 make_bad_inode(inode);
 321                 iput(inode);
 322                 goto retry;
 323         }
 324
 325         fi = get_fuse_inode(inode);
 326         spin_lock(&fc->lock);
 327         fi->nlookup++;
 328         spin_unlock(&fc->lock);
 329         fuse_change_attributes(inode, attr, attr_valid, attr_version);
 330
 331         return inode;
 332 }
 333
 334 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
 335                              loff_t offset, loff_t len)
 336 {
 337         struct inode *inode;
 338         pgoff_t pg_start;
 339         pgoff_t pg_end;
 340
 341         inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
 342         if (!inode)
 343                 return -ENOENT;
 344
 345         fuse_invalidate_attr(inode);
 346         forget_all_cached_acls(inode);
 347         if (offset >= 0) {
 348                 pg_start = offset >> PAGE_SHIFT;
 349                 if (len <= 0)
 350                         pg_end = -1;
 351                 else
 352                         pg_end = (offset + len - 1) >> PAGE_SHIFT;
 353                 invalidate_inode_pages2_range(inode->i_mapping,
 354                                               pg_start, pg_end);
 355         }
 356         iput(inode);
 357         return 0;
 358 }
 359
 360 void fuse_lock_inode(struct inode *inode)
 361 {
 362         if (!get_fuse_conn(inode)->parallel_dirops)
 363                 mutex_lock(&get_fuse_inode(inode)->mutex);
 364 }
 365
 366 void fuse_unlock_inode(struct inode *inode)
 367 {
 368         if (!get_fuse_conn(inode)->parallel_dirops)
 369                 mutex_unlock(&get_fuse_inode(inode)->mutex);
 370 }
 371
 372 static void fuse_umount_begin(struct super_block *sb)
 373 {
 374         fuse_abort_conn(get_fuse_conn_super(sb));
 375 }
 376
 377 static void fuse_send_destroy(struct fuse_conn *fc)
 378 {
 379         struct fuse_req *req = fc->destroy_req;
 380         if (req && fc->conn_init) {
 381                 fc->destroy_req = NULL;
 382                 req->in.h.opcode = FUSE_DESTROY;
 383                 __set_bit(FR_FORCE, &req->flags);
 384                 __clear_bit(FR_BACKGROUND, &req->flags);
 385                 fuse_request_send(fc, req);
 386                 fuse_put_request(fc, req);
 387         }
 388 }
 389
 390 static void fuse_put_super(struct super_block *sb)
 391 {
 392         struct fuse_conn *fc = get_fuse_conn_super(sb);
 393
 394         fuse_send_destroy(fc);
 395
 396         fuse_abort_conn(fc);
 397         mutex_lock(&fuse_mutex);
 398         list_del(&fc->entry);
 399         fuse_ctl_remove_conn(fc);
 400         mutex_unlock(&fuse_mutex);
 401
 402         fuse_conn_put(fc);
 403 }
 404
 405 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
 406 {
 407         stbuf->f_type    = FUSE_SUPER_MAGIC;
 408         stbuf->f_bsize   = attr->bsize;
 409         stbuf->f_frsize  = attr->frsize;
 410         stbuf->f_blocks  = attr->blocks;
 411         stbuf->f_bfree   = attr->bfree;
 412         stbuf->f_bavail  = attr->bavail;
 413         stbuf->f_files   = attr->files;
 414         stbuf->f_ffree   = attr->ffree;
 415         stbuf->f_namelen = attr->namelen;
 416         /* fsid is left zero */
 417 }
 418
 419 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
 420 {
 421         struct super_block *sb = dentry->d_sb;
 422         struct fuse_conn *fc = get_fuse_conn_super(sb);
 423         FUSE_ARGS(args);
 424         struct fuse_statfs_out outarg;
 425         int err;
 426
 427         if (!fuse_allow_current_process(fc)) {
 428                 buf->f_type = FUSE_SUPER_MAGIC;
 429                 return 0;
 430         }
 431
 432         memset(&outarg, 0, sizeof(outarg));
 433         args.in.numargs = 0;
 434         args.in.h.opcode = FUSE_STATFS;
 435         args.in.h.nodeid = get_node_id(d_inode(dentry));
 436         args.out.numargs = 1;
 437         args.out.args[0].size = sizeof(outarg);
 438         args.out.args[0].value = &outarg;
 439         err = fuse_simple_request(fc, &args);
 440         if (!err)
 441                 convert_fuse_statfs(buf, &outarg.st);
 442         return err;
 443 }
 444
 445 enum {
 446         OPT_FD,
 447         OPT_ROOTMODE,
 448         OPT_USER_ID,
 449         OPT_GROUP_ID,
 450         OPT_DEFAULT_PERMISSIONS,
 451         OPT_ALLOW_OTHER,
 452         OPT_MAX_READ,
 453         OPT_BLKSIZE,
 454         OPT_ERR
 455 };
 456
 457 static const match_table_t tokens = {
 458         {OPT_FD,                        "fd=%u"},
 459         {OPT_ROOTMODE,                  "rootmode=%o"},
 460         {OPT_USER_ID,                   "user_id=%u"},
 461         {OPT_GROUP_ID,                  "group_id=%u"},
 462         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 463         {OPT_ALLOW_OTHER,               "allow_other"},
 464         {OPT_MAX_READ,                  "max_read=%u"},
 465         {OPT_BLKSIZE,                   "blksize=%u"},
 466         {OPT_ERR,                       NULL}
 467 };
 468
 469 static int fuse_match_uint(substring_t *s, unsigned int *res)
 470 {
 471         int err = -ENOMEM;
 472         char *buf = match_strdup(s);
 473         if (buf) {
 474                 err = kstrtouint(buf, 10, res);
 475                 kfree(buf);
 476         }
 477         return err;
 478 }
 479
 480 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev,
 481                           struct user_namespace *user_ns)
 482 {
 483         char *p;
 484         memset(d, 0, sizeof(struct fuse_mount_data));
 485         d->max_read = ~0;
 486         d->blksize = FUSE_DEFAULT_BLKSIZE;
 487
 488         while ((p = strsep(&opt, ",")) != NULL) {
 489                 int token;
 490                 int value;
 491                 unsigned uv;
 492                 substring_t args[MAX_OPT_ARGS];
 493                 if (!*p)
 494                         continue;
 495
 496                 token = match_token(p, tokens, args);
 497                 switch (token) {
 498                 case OPT_FD:
 499                         if (match_int(&args[0], &value))
 500                                 return 0;
 501                         d->fd = value;
 502                         d->fd_present = 1;
 503                         break;
 504
 505                 case OPT_ROOTMODE:
 506                         if (match_octal(&args[0], &value))
 507                                 return 0;
 508                         if (!fuse_valid_type(value))
 509                                 return 0;
 510                         d->rootmode = value;
 511                         d->rootmode_present = 1;
 512                         break;
 513
 514                 case OPT_USER_ID:
 515                         if (fuse_match_uint(&args[0], &uv))
 516                                 return 0;
 517                         d->user_id = make_kuid(user_ns, uv);
 518                         if (!uid_valid(d->user_id))
 519                                 return 0;
 520                         d->user_id_present = 1;
 521                         break;
 522
 523                 case OPT_GROUP_ID:
 524                         if (fuse_match_uint(&args[0], &uv))
 525                                 return 0;
 526                         d->group_id = make_kgid(user_ns, uv);
 527                         if (!gid_valid(d->group_id))
 528                                 return 0;
 529                         d->group_id_present = 1;
 530                         break;
 531
 532                 case OPT_DEFAULT_PERMISSIONS:
 533                         d->default_permissions = 1;
 534                         break;
 535
 536                 case OPT_ALLOW_OTHER:
 537                         d->allow_other = 1;
 538                         break;
 539
 540                 case OPT_MAX_READ:
 541                         if (match_int(&args[0], &value))
 542                                 return 0;
 543                         d->max_read = value;
 544                         break;
 545
 546                 case OPT_BLKSIZE:
 547                         if (!is_bdev || match_int(&args[0], &value))
 548                                 return 0;
 549                         d->blksize = value;
 550                         break;
 551
 552                 default:
 553                         return 0;
 554                 }
 555         }
 556
 557         if (!d->fd_present || !d->rootmode_present ||
 558             !d->user_id_present || !d->group_id_present)
 559                 return 0;
 560
 561         return 1;
 562 }
 563
 564 static int fuse_show_options(struct seq_file *m, struct dentry *root)
 565 {
 566         struct super_block *sb = root->d_sb;
 567         struct fuse_conn *fc = get_fuse_conn_super(sb);
 568
 569         seq_printf(m, ",user_id=%u", from_kuid_munged(fc->user_ns, fc->user_id));
 570         seq_printf(m, ",group_id=%u", from_kgid_munged(fc->user_ns, fc->group_id));
 571         if (fc->default_permissions)
 572                 seq_puts(m, ",default_permissions");
 573         if (fc->allow_other)
 574                 seq_puts(m, ",allow_other");
 575         if (fc->max_read != ~0)
 576                 seq_printf(m, ",max_read=%u", fc->max_read);
 577         if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
 578                 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
 579         return 0;
 580 }
 581
 582 static void fuse_iqueue_init(struct fuse_iqueue *fiq)
 583 {
 584         memset(fiq, 0, sizeof(struct fuse_iqueue));
 585         init_waitqueue_head(&fiq->waitq);
 586         INIT_LIST_HEAD(&fiq->pending);
 587         INIT_LIST_HEAD(&fiq->interrupts);
 588         fiq->forget_list_tail = &fiq->forget_list_head;
 589         fiq->connected = 1;
 590 }
 591
 592 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
 593 {
 594         memset(fpq, 0, sizeof(struct fuse_pqueue));
 595         spin_lock_init(&fpq->lock);
 596         INIT_LIST_HEAD(&fpq->processing);
 597         INIT_LIST_HEAD(&fpq->io);
 598         fpq->connected = 1;
 599 }
 600
 601 void fuse_conn_init(struct fuse_conn *fc, struct user_namespace *user_ns)
 602 {
 603         memset(fc, 0, sizeof(*fc));
 604         spin_lock_init(&fc->lock);
 605         init_rwsem(&fc->killsb);
 606         refcount_set(&fc->count, 1);
 607         atomic_set(&fc->dev_count, 1);
 608         init_waitqueue_head(&fc->blocked_waitq);
 609         init_waitqueue_head(&fc->reserved_req_waitq);
 610         fuse_iqueue_init(&fc->iq);
 611         INIT_LIST_HEAD(&fc->bg_queue);
 612         INIT_LIST_HEAD(&fc->entry);
 613         INIT_LIST_HEAD(&fc->devices);
 614         atomic_set(&fc->num_waiting, 0);
 615         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
 616         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
 617         fc->khctr = 0;
 618         fc->polled_files = RB_ROOT;
 619         fc->blocked = 0;
 620         fc->initialized = 0;
 621         fc->connected = 1;
 622         fc->attr_version = 1;
 623         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
 624         fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
 625         fc->user_ns = get_user_ns(user_ns);
 626 }
 627 EXPORT_SYMBOL_GPL(fuse_conn_init);
 628
 629 void fuse_conn_put(struct fuse_conn *fc)
 630 {
 631         if (refcount_dec_and_test(&fc->count)) {
 632                 if (fc->destroy_req)
 633                         fuse_request_free(fc->destroy_req);
 634                 put_pid_ns(fc->pid_ns);
 635                 put_user_ns(fc->user_ns);
 636                 fc->release(fc);
 637         }
 638 }
 639 EXPORT_SYMBOL_GPL(fuse_conn_put);
 640
 641 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
 642 {
 643         refcount_inc(&fc->count);
 644         return fc;
 645 }
 646 EXPORT_SYMBOL_GPL(fuse_conn_get);
 647
 648 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
 649 {
 650         struct fuse_attr attr;
 651         memset(&attr, 0, sizeof(attr));
 652
 653         attr.mode = mode;
 654         attr.ino = FUSE_ROOT_ID;
 655         attr.nlink = 1;
 656         return fuse_iget(sb, 1, 0, &attr, 0, 0);
 657 }
 658
 659 struct fuse_inode_handle {
 660         u64 nodeid;
 661         u32 generation;
 662 };
 663
 664 static struct dentry *fuse_get_dentry(struct super_block *sb,
 665                                       struct fuse_inode_handle *handle)
 666 {
 667         struct fuse_conn *fc = get_fuse_conn_super(sb);
 668         struct inode *inode;
 669         struct dentry *entry;
 670         int err = -ESTALE;
 671
 672         if (handle->nodeid == 0)
 673                 goto out_err;
 674
 675         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
 676         if (!inode) {
 677                 struct fuse_entry_out outarg;
 678                 const struct qstr name = QSTR_INIT(".", 1);
 679
 680                 if (!fc->export_support)
 681                         goto out_err;
 682
 683                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
 684                                        &inode);
 685                 if (err && err != -ENOENT)
 686                         goto out_err;
 687                 if (err || !inode) {
 688                         err = -ESTALE;
 689                         goto out_err;
 690                 }
 691                 err = -EIO;
 692                 if (get_node_id(inode) != handle->nodeid)
 693                         goto out_iput;
 694         }
 695         err = -ESTALE;
 696         if (inode->i_generation != handle->generation)
 697                 goto out_iput;
 698
 699         entry = d_obtain_alias(inode);
 700         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
 701                 fuse_invalidate_entry_cache(entry);
 702
 703         return entry;
 704
 705  out_iput:
 706         iput(inode);
 707  out_err:
 708         return ERR_PTR(err);
 709 }
 710
 711 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
 712                            struct inode *parent)
 713 {
 714         int len = parent ? 6 : 3;
 715         u64 nodeid;
 716         u32 generation;
 717
 718         if (*max_len < len) {
 719                 *max_len = len;
 720                 return  FILEID_INVALID;
 721         }
 722
 723         nodeid = get_fuse_inode(inode)->nodeid;
 724         generation = inode->i_generation;
 725
 726         fh[0] = (u32)(nodeid >> 32);
 727         fh[1] = (u32)(nodeid & 0xffffffff);
 728         fh[2] = generation;
 729
 730         if (parent) {
 731                 nodeid = get_fuse_inode(parent)->nodeid;
 732                 generation = parent->i_generation;
 733
 734                 fh[3] = (u32)(nodeid >> 32);
 735                 fh[4] = (u32)(nodeid & 0xffffffff);
 736                 fh[5] = generation;
 737         }
 738
 739         *max_len = len;
 740         return parent ? 0x82 : 0x81;
 741 }
 742
 743 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
 744                 struct fid *fid, int fh_len, int fh_type)
 745 {
 746         struct fuse_inode_handle handle;
 747
 748         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
 749                 return NULL;
 750
 751         handle.nodeid = (u64) fid->raw[0] << 32;
 752         handle.nodeid |= (u64) fid->raw[1];
 753         handle.generation = fid->raw[2];
 754         return fuse_get_dentry(sb, &handle);
 755 }
 756
 757 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
 758                 struct fid *fid, int fh_len, int fh_type)
 759 {
 760         struct fuse_inode_handle parent;
 761
 762         if (fh_type != 0x82 || fh_len < 6)
 763                 return NULL;
 764
 765         parent.nodeid = (u64) fid->raw[3] << 32;
 766         parent.nodeid |= (u64) fid->raw[4];
 767         parent.generation = fid->raw[5];
 768         return fuse_get_dentry(sb, &parent);
 769 }
 770
 771 static struct dentry *fuse_get_parent(struct dentry *child)
 772 {
 773         struct inode *child_inode = d_inode(child);
 774         struct fuse_conn *fc = get_fuse_conn(child_inode);
 775         struct inode *inode;
 776         struct dentry *parent;
 777         struct fuse_entry_out outarg;
 778         const struct qstr name = QSTR_INIT("..", 2);
 779         int err;
 780
 781         if (!fc->export_support)
 782                 return ERR_PTR(-ESTALE);
 783
 784         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
 785                                &name, &outarg, &inode);
 786         if (err) {
 787                 if (err == -ENOENT)
 788                         return ERR_PTR(-ESTALE);
 789                 return ERR_PTR(err);
 790         }
 791
 792         parent = d_obtain_alias(inode);
 793         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
 794                 fuse_invalidate_entry_cache(parent);
 795
 796         return parent;
 797 }
 798
 799 static const struct export_operations fuse_export_operations = {
 800         .fh_to_dentry   = fuse_fh_to_dentry,
 801         .fh_to_parent   = fuse_fh_to_parent,
 802         .encode_fh      = fuse_encode_fh,
 803         .get_parent     = fuse_get_parent,
 804 };
 805
 806 static const struct super_operations fuse_super_operations = {
 807         .alloc_inode    = fuse_alloc_inode,
 808         .destroy_inode  = fuse_destroy_inode,
 809         .evict_inode    = fuse_evict_inode,
 810         .write_inode    = fuse_write_inode,
 811         .drop_inode     = generic_delete_inode,
 812         .remount_fs     = fuse_remount_fs,
 813         .put_super      = fuse_put_super,
 814         .umount_begin   = fuse_umount_begin,
 815         .statfs         = fuse_statfs,
 816         .show_options   = fuse_show_options,
 817 };
 818
 819 static void sanitize_global_limit(unsigned *limit)
 820 {
 821         if (*limit == 0)
 822                 *limit = ((totalram_pages << PAGE_SHIFT) >> 13) /
 823                          sizeof(struct fuse_req);
 824
 825         if (*limit >= 1 << 16)
 826                 *limit = (1 << 16) - 1;
 827 }
 828
 829 static int set_global_limit(const char *val, struct kernel_param *kp)
 830 {
 831         int rv;
 832
 833         rv = param_set_uint(val, kp);
 834         if (rv)
 835                 return rv;
 836
 837         sanitize_global_limit((unsigned *)kp->arg);
 838
 839         return 0;
 840 }
 841
 842 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
 843 {
 844         int cap_sys_admin = capable(CAP_SYS_ADMIN);
 845
 846         if (arg->minor < 13)
 847                 return;
 848
 849         sanitize_global_limit(&max_user_bgreq);
 850         sanitize_global_limit(&max_user_congthresh);
 851
 852         if (arg->max_background) {
 853                 fc->max_background = arg->max_background;
 854
 855                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
 856                         fc->max_background = max_user_bgreq;
 857         }
 858         if (arg->congestion_threshold) {
 859                 fc->congestion_threshold = arg->congestion_threshold;
 860
 861                 if (!cap_sys_admin &&
 862                     fc->congestion_threshold > max_user_congthresh)
 863                         fc->congestion_threshold = max_user_congthresh;
 864         }
 865 }
 866
 867 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
 868 {
 869         struct fuse_init_out *arg = &req->misc.init_out;
 870
 871         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
 872                 fc->conn_error = 1;
 873         else {
 874                 unsigned long ra_pages;
 875
 876                 process_init_limits(fc, arg);
 877
 878                 if (arg->minor >= 6) {
 879                         ra_pages = arg->max_readahead / PAGE_SIZE;
 880                         if (arg->flags & FUSE_ASYNC_READ)
 881                                 fc->async_read = 1;
 882                         if (!(arg->flags & FUSE_POSIX_LOCKS))
 883                                 fc->no_lock = 1;
 884                         if (arg->minor >= 17) {
 885                                 if (!(arg->flags & FUSE_FLOCK_LOCKS))
 886                                         fc->no_flock = 1;
 887                         } else {
 888                                 if (!(arg->flags & FUSE_POSIX_LOCKS))
 889                                         fc->no_flock = 1;
 890                         }
 891                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
 892                                 fc->atomic_o_trunc = 1;
 893                         if (arg->minor >= 9) {
 894                                 /* LOOKUP has dependency on proto version */
 895                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
 896                                         fc->export_support = 1;
 897                         }
 898                         if (arg->flags & FUSE_BIG_WRITES)
 899                                 fc->big_writes = 1;
 900                         if (arg->flags & FUSE_DONT_MASK)
 901                                 fc->dont_mask = 1;
 902                         if (arg->flags & FUSE_AUTO_INVAL_DATA)
 903                                 fc->auto_inval_data = 1;
 904                         if (arg->flags & FUSE_DO_READDIRPLUS) {
 905                                 fc->do_readdirplus = 1;
 906                                 if (arg->flags & FUSE_READDIRPLUS_AUTO)
 907                                         fc->readdirplus_auto = 1;
 908                         }
 909                         if (arg->flags & FUSE_ASYNC_DIO)
 910                                 fc->async_dio = 1;
 911                         if (arg->flags & FUSE_WRITEBACK_CACHE)
 912                                 fc->writeback_cache = 1;
 913                         if (arg->flags & FUSE_PARALLEL_DIROPS)
 914                                 fc->parallel_dirops = 1;
 915                         if (arg->flags & FUSE_HANDLE_KILLPRIV)
 916                                 fc->handle_killpriv = 1;
 917                         if (arg->time_gran && arg->time_gran <= 1000000000)
 918                                 fc->sb->s_time_gran = arg->time_gran;
 919                         if ((arg->flags & FUSE_POSIX_ACL)) {
 920                                 fc->default_permissions = 1;
 921                                 fc->posix_acl = 1;
 922                                 fc->sb->s_xattr = fuse_acl_xattr_handlers;
 923                         }
 924                 } else {
 925                         ra_pages = fc->max_read / PAGE_SIZE;
 926                         fc->no_lock = 1;
 927                         fc->no_flock = 1;
 928                 }
 929
 930                 fc->sb->s_bdi->ra_pages =
 931                                 min(fc->sb->s_bdi->ra_pages, ra_pages);
 932                 fc->minor = arg->minor;
 933                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
 934                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
 935                 fc->conn_init = 1;
 936         }
 937         fuse_set_initialized(fc);
 938         wake_up_all(&fc->blocked_waitq);
 939 }
 940
 941 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
 942 {
 943         struct fuse_init_in *arg = &req->misc.init_in;
 944
 945         arg->major = FUSE_KERNEL_VERSION;
 946         arg->minor = FUSE_KERNEL_MINOR_VERSION;
 947         arg->max_readahead = fc->sb->s_bdi->ra_pages * PAGE_SIZE;
 948         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
 949                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
 950                 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
 951                 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
 952                 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
 953                 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
 954                 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL;
 955         req->in.h.opcode = FUSE_INIT;
 956         req->in.numargs = 1;
 957         req->in.args[0].size = sizeof(*arg);
 958         req->in.args[0].value = arg;
 959         req->out.numargs = 1;
 960         /* Variable length argument used for backward compatibility
 961            with interface version < 7.5.  Rest of init_out is zeroed
 962            by do_get_request(), so a short reply is not a problem */
 963         req->out.argvar = 1;
 964         req->out.args[0].size = sizeof(struct fuse_init_out);
 965         req->out.args[0].value = &req->misc.init_out;
 966         req->end = process_init_reply;
 967         fuse_request_send_background(fc, req);
 968 }
 969
 970 static void fuse_free_conn(struct fuse_conn *fc)
 971 {
 972         WARN_ON(!list_empty(&fc->devices));
 973         kfree_rcu(fc, rcu);
 974 }
 975
 976 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
 977 {
 978         int err;
 979         char *suffix = "";
 980
 981         if (sb->s_bdev) {
 982                 suffix = "-fuseblk";
 983                 /*
 984                  * sb->s_bdi points to blkdev's bdi however we want to redirect
 985                  * it to our private bdi...
 986                  */
 987                 bdi_put(sb->s_bdi);
 988                 sb->s_bdi = &noop_backing_dev_info;
 989         }
 990         err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
 991                                    MINOR(fc->dev), suffix);
 992         if (err)
 993                 return err;
 994
 995         sb->s_bdi->ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_SIZE;
 996         /* fuse does it's own writeback accounting */
 997         sb->s_bdi->capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
 998
 999         /*
1000          * For a single fuse filesystem use max 1% of dirty +
1001          * writeback threshold.
1002          *
1003          * This gives about 1M of write buffer for memory maps on a
1004          * machine with 1G and 10% dirty_ratio, which should be more
1005          * than enough.
1006          *
1007          * Privileged users can raise it by writing to
1008          *
1009          *    /sys/class/bdi/<bdi>/max_ratio
1010          */
1011         bdi_set_max_ratio(sb->s_bdi, 1);
1012
1013         return 0;
1014 }
1015
1016 struct fuse_dev *fuse_dev_alloc(struct fuse_conn *fc)
1017 {
1018         struct fuse_dev *fud;
1019
1020         fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1021         if (fud) {
1022                 fud->fc = fuse_conn_get(fc);
1023                 fuse_pqueue_init(&fud->pq);
1024
1025                 spin_lock(&fc->lock);
1026                 list_add_tail(&fud->entry, &fc->devices);
1027                 spin_unlock(&fc->lock);
1028         }
1029
1030         return fud;
1031 }
1032 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1033
1034 void fuse_dev_free(struct fuse_dev *fud)
1035 {
1036         struct fuse_conn *fc = fud->fc;
1037
1038         if (fc) {
1039                 spin_lock(&fc->lock);
1040                 list_del(&fud->entry);
1041                 spin_unlock(&fc->lock);
1042
1043                 fuse_conn_put(fc);
1044         }
1045         kfree(fud);
1046 }
1047 EXPORT_SYMBOL_GPL(fuse_dev_free);
1048
1049 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
1050 {
1051         struct fuse_dev *fud;
1052         struct fuse_conn *fc;
1053         struct inode *root;
1054         struct fuse_mount_data d;
1055         struct file *file;
1056         struct dentry *root_dentry;
1057         struct fuse_req *init_req;
1058         int err;
1059         int is_bdev = sb->s_bdev != NULL;
1060
1061         err = -EINVAL;
1062         if (sb->s_flags & MS_MANDLOCK)
1063                 goto err;
1064
1065         sb->s_flags &= ~(MS_NOSEC | MS_I_VERSION);
1066
1067         if (!parse_fuse_opt(data, &d, is_bdev, sb->s_user_ns))
1068                 goto err;
1069
1070         if (is_bdev) {
1071 #ifdef CONFIG_BLOCK
1072                 err = -EINVAL;
1073                 if (!sb_set_blocksize(sb, d.blksize))
1074                         goto err;
1075 #endif
1076         } else {
1077                 sb->s_blocksize = PAGE_SIZE;
1078                 sb->s_blocksize_bits = PAGE_SHIFT;
1079         }
1080         sb->s_magic = FUSE_SUPER_MAGIC;
1081         sb->s_op = &fuse_super_operations;
1082         sb->s_xattr = fuse_xattr_handlers;
1083         sb->s_maxbytes = MAX_LFS_FILESIZE;
1084         sb->s_time_gran = 1;
1085         sb->s_export_op = &fuse_export_operations;
1086
1087         file = fget(d.fd);
1088         err = -EINVAL;
1089         if (!file)
1090                 goto err;
1091
1092         /*
1093          * Require mount to happen from the same user namespace which
1094          * opened /dev/fuse to prevent potential attacks.
1095          */
1096         if (file->f_op != &fuse_dev_operations ||
1097             file->f_cred->user_ns != sb->s_user_ns)
1098                 goto err_fput;
1099
1100         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1101         err = -ENOMEM;
1102         if (!fc)
1103                 goto err_fput;
1104
1105         fuse_conn_init(fc, sb->s_user_ns);
1106         fc->release = fuse_free_conn;
1107
1108         fud = fuse_dev_alloc(fc);
1109         if (!fud)
1110                 goto err_put_conn;
1111
1112         fc->dev = sb->s_dev;
1113         fc->sb = sb;
1114         err = fuse_bdi_init(fc, sb);
1115         if (err)
1116                 goto err_dev_free;
1117
1118         /* Handle umasking inside the fuse code */
1119         if (sb->s_flags & MS_POSIXACL)
1120                 fc->dont_mask = 1;
1121         sb->s_flags |= MS_POSIXACL;
1122
1123         fc->default_permissions = d.default_permissions;
1124         fc->allow_other = d.allow_other;
1125         fc->user_id = d.user_id;
1126         fc->group_id = d.group_id;
1127         fc->max_read = max_t(unsigned, 4096, d.max_read);
1128
1129         /* Used by get_root_inode() */
1130         sb->s_fs_info = fc;
1131
1132         err = -ENOMEM;
1133         root = fuse_get_root_inode(sb, d.rootmode);
1134         sb->s_d_op = &fuse_root_dentry_operations;
1135         root_dentry = d_make_root(root);
1136         if (!root_dentry)
1137                 goto err_dev_free;
1138         /* Root dentry doesn't have .d_revalidate */
1139         sb->s_d_op = &fuse_dentry_operations;
1140
1141         init_req = fuse_request_alloc(0);
1142         if (!init_req)
1143                 goto err_put_root;
1144         __set_bit(FR_BACKGROUND, &init_req->flags);
1145
1146         if (is_bdev) {
1147                 fc->destroy_req = fuse_request_alloc(0);
1148                 if (!fc->destroy_req)
1149                         goto err_free_init_req;
1150         }
1151
1152         mutex_lock(&fuse_mutex);
1153         err = -EINVAL;
1154         if (file->private_data)
1155                 goto err_unlock;
1156
1157         err = fuse_ctl_add_conn(fc);
1158         if (err)
1159                 goto err_unlock;
1160
1161         list_add_tail(&fc->entry, &fuse_conn_list);
1162         sb->s_root = root_dentry;
1163         file->private_data = fud;
1164         mutex_unlock(&fuse_mutex);
1165         /*
1166          * atomic_dec_and_test() in fput() provides the necessary
1167          * memory barrier for file->private_data to be visible on all
1168          * CPUs after this
1169          */
1170         fput(file);
1171
1172         fuse_send_init(fc, init_req);
1173
1174         return 0;
1175
1176  err_unlock:
1177         mutex_unlock(&fuse_mutex);
1178  err_free_init_req:
1179         fuse_request_free(init_req);
1180  err_put_root:
1181         dput(root_dentry);
1182  err_dev_free:
1183         fuse_dev_free(fud);
1184  err_put_conn:
1185         fuse_conn_put(fc);
1186  err_fput:
1187         fput(file);
1188  err:
1189         return err;
1190 }
1191
1192 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1193                        int flags, const char *dev_name,
1194                        void *raw_data)
1195 {
1196         return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1197 }
1198
1199 static void fuse_kill_sb_anon(struct super_block *sb)
1200 {
1201         struct fuse_conn *fc = get_fuse_conn_super(sb);
1202
1203         if (fc) {
1204                 down_write(&fc->killsb);
1205                 fc->sb = NULL;
1206                 up_write(&fc->killsb);
1207         }
1208
1209         kill_anon_super(sb);
1210 }
1211
1212 static struct file_system_type fuse_fs_type = {
1213         .owner          = THIS_MODULE,
1214         .name           = "fuse",
1215         .fs_flags       = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1216         .mount          = fuse_mount,
1217         .kill_sb        = fuse_kill_sb_anon,
1218 };
1219 MODULE_ALIAS_FS("fuse");
1220
1221 #ifdef CONFIG_BLOCK
1222 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1223                            int flags, const char *dev_name,
1224                            void *raw_data)
1225 {
1226         return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1227 }
1228
1229 static void fuse_kill_sb_blk(struct super_block *sb)
1230 {
1231         struct fuse_conn *fc = get_fuse_conn_super(sb);
1232
1233         if (fc) {
1234                 down_write(&fc->killsb);
1235                 fc->sb = NULL;
1236                 up_write(&fc->killsb);
1237         }
1238
1239         kill_block_super(sb);
1240 }
1241
1242 static struct file_system_type fuseblk_fs_type = {
1243         .owner          = THIS_MODULE,
1244         .name           = "fuseblk",
1245         .mount          = fuse_mount_blk,
1246         .kill_sb        = fuse_kill_sb_blk,
1247         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1248 };
1249 MODULE_ALIAS_FS("fuseblk");
1250
1251 static inline int register_fuseblk(void)
1252 {
1253         return register_filesystem(&fuseblk_fs_type);
1254 }
1255
1256 static inline void unregister_fuseblk(void)
1257 {
1258         unregister_filesystem(&fuseblk_fs_type);
1259 }
1260 #else
1261 static inline int register_fuseblk(void)
1262 {
1263         return 0;
1264 }
1265
1266 static inline void unregister_fuseblk(void)
1267 {
1268 }
1269 #endif
1270
1271 static void fuse_inode_init_once(void *foo)
1272 {
1273         struct inode *inode = foo;
1274
1275         inode_init_once(inode);
1276 }
1277
1278 static int __init fuse_fs_init(void)
1279 {
1280         int err;
1281
1282         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1283                                               sizeof(struct fuse_inode), 0,
1284                                               SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT,
1285                                               fuse_inode_init_once);
1286         err = -ENOMEM;
1287         if (!fuse_inode_cachep)
1288                 goto out;
1289
1290         err = register_fuseblk();
1291         if (err)
1292                 goto out2;
1293
1294         err = register_filesystem(&fuse_fs_type);
1295         if (err)
1296                 goto out3;
1297
1298         return 0;
1299
1300  out3:
1301         unregister_fuseblk();
1302  out2:
1303         kmem_cache_destroy(fuse_inode_cachep);
1304  out:
1305         return err;
1306 }
1307
1308 static void fuse_fs_cleanup(void)
1309 {
1310         unregister_filesystem(&fuse_fs_type);
1311         unregister_fuseblk();
1312
1313         /*
1314          * Make sure all delayed rcu free inodes are flushed before we
1315          * destroy cache.
1316          */
1317         rcu_barrier();
1318         kmem_cache_destroy(fuse_inode_cachep);
1319 }
1320
1321 static struct kobject *fuse_kobj;
1322
1323 static int fuse_sysfs_init(void)
1324 {
1325         int err;
1326
1327         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1328         if (!fuse_kobj) {
1329                 err = -ENOMEM;
1330                 goto out_err;
1331         }
1332
1333         err = sysfs_create_mount_point(fuse_kobj, "connections");
1334         if (err)
1335                 goto out_fuse_unregister;
1336
1337         return 0;
1338
1339  out_fuse_unregister:
1340         kobject_put(fuse_kobj);
1341  out_err:
1342         return err;
1343 }
1344
1345 static void fuse_sysfs_cleanup(void)
1346 {
1347         sysfs_remove_mount_point(fuse_kobj, "connections");
1348         kobject_put(fuse_kobj);
1349 }
1350
1351 static int __init fuse_init(void)
1352 {
1353         int res;
1354
1355         printk(KERN_INFO "fuse init (API version %i.%i)\n",
1356                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1357
1358         INIT_LIST_HEAD(&fuse_conn_list);
1359         res = fuse_fs_init();
1360         if (res)
1361                 goto err;
1362
1363         res = fuse_dev_init();
1364         if (res)
1365                 goto err_fs_cleanup;
1366
1367         res = fuse_sysfs_init();
1368         if (res)
1369                 goto err_dev_cleanup;
1370
1371         res = fuse_ctl_init();
1372         if (res)
1373                 goto err_sysfs_cleanup;
1374
1375         sanitize_global_limit(&max_user_bgreq);
1376         sanitize_global_limit(&max_user_congthresh);
1377
1378         return 0;
1379
1380  err_sysfs_cleanup:
1381         fuse_sysfs_cleanup();
1382  err_dev_cleanup:
1383         fuse_dev_cleanup();
1384  err_fs_cleanup:
1385         fuse_fs_cleanup();
1386  err:
1387         return res;
1388 }
1389
1390 static void __exit fuse_exit(void)
1391 {
1392         printk(KERN_DEBUG "fuse exit\n");
1393
1394         fuse_ctl_cleanup();
1395         fuse_sysfs_cleanup();
1396         fuse_fs_cleanup();
1397         fuse_dev_cleanup();
1398 }
1399
1400 module_init(fuse_init);
1401 module_exit(fuse_exit);