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