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