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