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/parser.h>
  18 #include <linux/statfs.h>
  19 #include <linux/random.h>
  20 #include <linux/sched.h>
  21 #include <linux/exportfs.h>
  22
  23 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  24 MODULE_DESCRIPTION("Filesystem in Userspace");
  25 MODULE_LICENSE("GPL");
  26
  27 static struct kmem_cache *fuse_inode_cachep;
  28 struct list_head fuse_conn_list;
  29 DEFINE_MUTEX(fuse_mutex);
  30
  31 #define FUSE_SUPER_MAGIC 0x65735546
  32
  33 #define FUSE_DEFAULT_BLKSIZE 512
  34
  35 struct fuse_mount_data {
  36         int fd;
  37         unsigned rootmode;
  38         unsigned user_id;
  39         unsigned group_id;
  40         unsigned fd_present:1;
  41         unsigned rootmode_present:1;
  42         unsigned user_id_present:1;
  43         unsigned group_id_present:1;
  44         unsigned flags;
  45         unsigned max_read;
  46         unsigned blksize;
  47 };
  48
  49 static struct inode *fuse_alloc_inode(struct super_block *sb)
  50 {
  51         struct inode *inode;
  52         struct fuse_inode *fi;
  53
  54         inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
  55         if (!inode)
  56                 return NULL;
  57
  58         fi = get_fuse_inode(inode);
  59         fi->i_time = 0;
  60         fi->nodeid = 0;
  61         fi->nlookup = 0;
  62         fi->attr_version = 0;
  63         fi->writectr = 0;
  64         INIT_LIST_HEAD(&fi->write_files);
  65         INIT_LIST_HEAD(&fi->queued_writes);
  66         INIT_LIST_HEAD(&fi->writepages);
  67         init_waitqueue_head(&fi->page_waitq);
  68         fi->forget_req = fuse_request_alloc();
  69         if (!fi->forget_req) {
  70                 kmem_cache_free(fuse_inode_cachep, inode);
  71                 return NULL;
  72         }
  73
  74         return inode;
  75 }
  76
  77 static void fuse_destroy_inode(struct inode *inode)
  78 {
  79         struct fuse_inode *fi = get_fuse_inode(inode);
  80         BUG_ON(!list_empty(&fi->write_files));
  81         BUG_ON(!list_empty(&fi->queued_writes));
  82         if (fi->forget_req)
  83                 fuse_request_free(fi->forget_req);
  84         kmem_cache_free(fuse_inode_cachep, inode);
  85 }
  86
  87 void fuse_send_forget(struct fuse_conn *fc, struct fuse_req *req,
  88                       u64 nodeid, u64 nlookup)
  89 {
  90         struct fuse_forget_in *inarg = &req->misc.forget_in;
  91         inarg->nlookup = nlookup;
  92         req->in.h.opcode = FUSE_FORGET;
  93         req->in.h.nodeid = nodeid;
  94         req->in.numargs = 1;
  95         req->in.args[0].size = sizeof(struct fuse_forget_in);
  96         req->in.args[0].value = inarg;
  97         fuse_request_send_noreply(fc, req);
  98 }
  99
 100 static void fuse_clear_inode(struct inode *inode)
 101 {
 102         if (inode->i_sb->s_flags & MS_ACTIVE) {
 103                 struct fuse_conn *fc = get_fuse_conn(inode);
 104                 struct fuse_inode *fi = get_fuse_inode(inode);
 105                 fuse_send_forget(fc, fi->forget_req, fi->nodeid, fi->nlookup);
 106                 fi->forget_req = NULL;
 107         }
 108 }
 109
 110 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
 111 {
 112         if (*flags & MS_MANDLOCK)
 113                 return -EINVAL;
 114
 115         return 0;
 116 }
 117
 118 void fuse_truncate(struct address_space *mapping, loff_t offset)
 119 {
 120         /* See vmtruncate() */
 121         unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
 122         truncate_inode_pages(mapping, offset);
 123         unmap_mapping_range(mapping, offset + PAGE_SIZE - 1, 0, 1);
 124 }
 125
 126 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
 127                                    u64 attr_valid)
 128 {
 129         struct fuse_conn *fc = get_fuse_conn(inode);
 130         struct fuse_inode *fi = get_fuse_inode(inode);
 131
 132         fi->attr_version = ++fc->attr_version;
 133         fi->i_time = attr_valid;
 134
 135         inode->i_ino     = attr->ino;
 136         inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
 137         inode->i_nlink   = attr->nlink;
 138         inode->i_uid     = attr->uid;
 139         inode->i_gid     = attr->gid;
 140         inode->i_blocks  = attr->blocks;
 141         inode->i_atime.tv_sec   = attr->atime;
 142         inode->i_atime.tv_nsec  = attr->atimensec;
 143         inode->i_mtime.tv_sec   = attr->mtime;
 144         inode->i_mtime.tv_nsec  = attr->mtimensec;
 145         inode->i_ctime.tv_sec   = attr->ctime;
 146         inode->i_ctime.tv_nsec  = attr->ctimensec;
 147
 148         if (attr->blksize != 0)
 149                 inode->i_blkbits = ilog2(attr->blksize);
 150         else
 151                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
 152
 153         /*
 154          * Don't set the sticky bit in i_mode, unless we want the VFS
 155          * to check permissions.  This prevents failures due to the
 156          * check in may_delete().
 157          */
 158         fi->orig_i_mode = inode->i_mode;
 159         if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
 160                 inode->i_mode &= ~S_ISVTX;
 161 }
 162
 163 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
 164                             u64 attr_valid, u64 attr_version)
 165 {
 166         struct fuse_conn *fc = get_fuse_conn(inode);
 167         struct fuse_inode *fi = get_fuse_inode(inode);
 168         loff_t oldsize;
 169
 170         spin_lock(&fc->lock);
 171         if (attr_version != 0 && fi->attr_version > attr_version) {
 172                 spin_unlock(&fc->lock);
 173                 return;
 174         }
 175
 176         fuse_change_attributes_common(inode, attr, attr_valid);
 177
 178         oldsize = inode->i_size;
 179         i_size_write(inode, attr->size);
 180         spin_unlock(&fc->lock);
 181
 182         if (S_ISREG(inode->i_mode) && oldsize != attr->size) {
 183                 if (attr->size < oldsize)
 184                         fuse_truncate(inode->i_mapping, attr->size);
 185                 invalidate_inode_pages2(inode->i_mapping);
 186         }
 187 }
 188
 189 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
 190 {
 191         inode->i_mode = attr->mode & S_IFMT;
 192         inode->i_size = attr->size;
 193         if (S_ISREG(inode->i_mode)) {
 194                 fuse_init_common(inode);
 195                 fuse_init_file_inode(inode);
 196         } else if (S_ISDIR(inode->i_mode))
 197                 fuse_init_dir(inode);
 198         else if (S_ISLNK(inode->i_mode))
 199                 fuse_init_symlink(inode);
 200         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
 201                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
 202                 fuse_init_common(inode);
 203                 init_special_inode(inode, inode->i_mode,
 204                                    new_decode_dev(attr->rdev));
 205         } else
 206                 BUG();
 207 }
 208
 209 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
 210 {
 211         u64 nodeid = *(u64 *) _nodeidp;
 212         if (get_node_id(inode) == nodeid)
 213                 return 1;
 214         else
 215                 return 0;
 216 }
 217
 218 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
 219 {
 220         u64 nodeid = *(u64 *) _nodeidp;
 221         get_fuse_inode(inode)->nodeid = nodeid;
 222         return 0;
 223 }
 224
 225 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
 226                         int generation, struct fuse_attr *attr,
 227                         u64 attr_valid, u64 attr_version)
 228 {
 229         struct inode *inode;
 230         struct fuse_inode *fi;
 231         struct fuse_conn *fc = get_fuse_conn_super(sb);
 232
 233  retry:
 234         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
 235         if (!inode)
 236                 return NULL;
 237
 238         if ((inode->i_state & I_NEW)) {
 239                 inode->i_flags |= S_NOATIME|S_NOCMTIME;
 240                 inode->i_generation = generation;
 241                 inode->i_data.backing_dev_info = &fc->bdi;
 242                 fuse_init_inode(inode, attr);
 243                 unlock_new_inode(inode);
 244         } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
 245                 /* Inode has changed type, any I/O on the old should fail */
 246                 make_bad_inode(inode);
 247                 iput(inode);
 248                 goto retry;
 249         }
 250
 251         fi = get_fuse_inode(inode);
 252         spin_lock(&fc->lock);
 253         fi->nlookup++;
 254         spin_unlock(&fc->lock);
 255         fuse_change_attributes(inode, attr, attr_valid, attr_version);
 256
 257         return inode;
 258 }
 259
 260 static void fuse_umount_begin(struct super_block *sb)
 261 {
 262         fuse_abort_conn(get_fuse_conn_super(sb));
 263 }
 264
 265 static void fuse_send_destroy(struct fuse_conn *fc)
 266 {
 267         struct fuse_req *req = fc->destroy_req;
 268         if (req && fc->conn_init) {
 269                 fc->destroy_req = NULL;
 270                 req->in.h.opcode = FUSE_DESTROY;
 271                 req->force = 1;
 272                 fuse_request_send(fc, req);
 273                 fuse_put_request(fc, req);
 274         }
 275 }
 276
 277 static void fuse_bdi_destroy(struct fuse_conn *fc)
 278 {
 279         if (fc->bdi_initialized)
 280                 bdi_destroy(&fc->bdi);
 281 }
 282
 283 void fuse_conn_kill(struct fuse_conn *fc)
 284 {
 285         spin_lock(&fc->lock);
 286         fc->connected = 0;
 287         fc->blocked = 0;
 288         spin_unlock(&fc->lock);
 289         /* Flush all readers on this fs */
 290         kill_fasync(&fc->fasync, SIGIO, POLL_IN);
 291         wake_up_all(&fc->waitq);
 292         wake_up_all(&fc->blocked_waitq);
 293         wake_up_all(&fc->reserved_req_waitq);
 294         mutex_lock(&fuse_mutex);
 295         list_del(&fc->entry);
 296         fuse_ctl_remove_conn(fc);
 297         mutex_unlock(&fuse_mutex);
 298         fuse_bdi_destroy(fc);
 299 }
 300 EXPORT_SYMBOL_GPL(fuse_conn_kill);
 301
 302 static void fuse_put_super(struct super_block *sb)
 303 {
 304         struct fuse_conn *fc = get_fuse_conn_super(sb);
 305
 306         fuse_send_destroy(fc);
 307         fuse_conn_kill(fc);
 308         fuse_conn_put(fc);
 309 }
 310
 311 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
 312 {
 313         stbuf->f_type    = FUSE_SUPER_MAGIC;
 314         stbuf->f_bsize   = attr->bsize;
 315         stbuf->f_frsize  = attr->frsize;
 316         stbuf->f_blocks  = attr->blocks;
 317         stbuf->f_bfree   = attr->bfree;
 318         stbuf->f_bavail  = attr->bavail;
 319         stbuf->f_files   = attr->files;
 320         stbuf->f_ffree   = attr->ffree;
 321         stbuf->f_namelen = attr->namelen;
 322         /* fsid is left zero */
 323 }
 324
 325 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
 326 {
 327         struct super_block *sb = dentry->d_sb;
 328         struct fuse_conn *fc = get_fuse_conn_super(sb);
 329         struct fuse_req *req;
 330         struct fuse_statfs_out outarg;
 331         int err;
 332
 333         if (!fuse_allow_task(fc, current)) {
 334                 buf->f_type = FUSE_SUPER_MAGIC;
 335                 return 0;
 336         }
 337
 338         req = fuse_get_req(fc);
 339         if (IS_ERR(req))
 340                 return PTR_ERR(req);
 341
 342         memset(&outarg, 0, sizeof(outarg));
 343         req->in.numargs = 0;
 344         req->in.h.opcode = FUSE_STATFS;
 345         req->in.h.nodeid = get_node_id(dentry->d_inode);
 346         req->out.numargs = 1;
 347         req->out.args[0].size =
 348                 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
 349         req->out.args[0].value = &outarg;
 350         fuse_request_send(fc, req);
 351         err = req->out.h.error;
 352         if (!err)
 353                 convert_fuse_statfs(buf, &outarg.st);
 354         fuse_put_request(fc, req);
 355         return err;
 356 }
 357
 358 enum {
 359         OPT_FD,
 360         OPT_ROOTMODE,
 361         OPT_USER_ID,
 362         OPT_GROUP_ID,
 363         OPT_DEFAULT_PERMISSIONS,
 364         OPT_ALLOW_OTHER,
 365         OPT_MAX_READ,
 366         OPT_BLKSIZE,
 367         OPT_ERR
 368 };
 369
 370 static const match_table_t tokens = {
 371         {OPT_FD,                        "fd=%u"},
 372         {OPT_ROOTMODE,                  "rootmode=%o"},
 373         {OPT_USER_ID,                   "user_id=%u"},
 374         {OPT_GROUP_ID,                  "group_id=%u"},
 375         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 376         {OPT_ALLOW_OTHER,               "allow_other"},
 377         {OPT_MAX_READ,                  "max_read=%u"},
 378         {OPT_BLKSIZE,                   "blksize=%u"},
 379         {OPT_ERR,                       NULL}
 380 };
 381
 382 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
 383 {
 384         char *p;
 385         memset(d, 0, sizeof(struct fuse_mount_data));
 386         d->max_read = ~0;
 387         d->blksize = FUSE_DEFAULT_BLKSIZE;
 388
 389         while ((p = strsep(&opt, ",")) != NULL) {
 390                 int token;
 391                 int value;
 392                 substring_t args[MAX_OPT_ARGS];
 393                 if (!*p)
 394                         continue;
 395
 396                 token = match_token(p, tokens, args);
 397                 switch (token) {
 398                 case OPT_FD:
 399                         if (match_int(&args[0], &value))
 400                                 return 0;
 401                         d->fd = value;
 402                         d->fd_present = 1;
 403                         break;
 404
 405                 case OPT_ROOTMODE:
 406                         if (match_octal(&args[0], &value))
 407                                 return 0;
 408                         if (!fuse_valid_type(value))
 409                                 return 0;
 410                         d->rootmode = value;
 411                         d->rootmode_present = 1;
 412                         break;
 413
 414                 case OPT_USER_ID:
 415                         if (match_int(&args[0], &value))
 416                                 return 0;
 417                         d->user_id = value;
 418                         d->user_id_present = 1;
 419                         break;
 420
 421                 case OPT_GROUP_ID:
 422                         if (match_int(&args[0], &value))
 423                                 return 0;
 424                         d->group_id = value;
 425                         d->group_id_present = 1;
 426                         break;
 427
 428                 case OPT_DEFAULT_PERMISSIONS:
 429                         d->flags |= FUSE_DEFAULT_PERMISSIONS;
 430                         break;
 431
 432                 case OPT_ALLOW_OTHER:
 433                         d->flags |= FUSE_ALLOW_OTHER;
 434                         break;
 435
 436                 case OPT_MAX_READ:
 437                         if (match_int(&args[0], &value))
 438                                 return 0;
 439                         d->max_read = value;
 440                         break;
 441
 442                 case OPT_BLKSIZE:
 443                         if (!is_bdev || match_int(&args[0], &value))
 444                                 return 0;
 445                         d->blksize = value;
 446                         break;
 447
 448                 default:
 449                         return 0;
 450                 }
 451         }
 452
 453         if (!d->fd_present || !d->rootmode_present ||
 454             !d->user_id_present || !d->group_id_present)
 455                 return 0;
 456
 457         return 1;
 458 }
 459
 460 static int fuse_show_options(struct seq_file *m, struct vfsmount *mnt)
 461 {
 462         struct fuse_conn *fc = get_fuse_conn_super(mnt->mnt_sb);
 463
 464         seq_printf(m, ",user_id=%u", fc->user_id);
 465         seq_printf(m, ",group_id=%u", fc->group_id);
 466         if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
 467                 seq_puts(m, ",default_permissions");
 468         if (fc->flags & FUSE_ALLOW_OTHER)
 469                 seq_puts(m, ",allow_other");
 470         if (fc->max_read != ~0)
 471                 seq_printf(m, ",max_read=%u", fc->max_read);
 472         if (mnt->mnt_sb->s_bdev &&
 473             mnt->mnt_sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
 474                 seq_printf(m, ",blksize=%lu", mnt->mnt_sb->s_blocksize);
 475         return 0;
 476 }
 477
 478 void fuse_conn_init(struct fuse_conn *fc)
 479 {
 480         memset(fc, 0, sizeof(*fc));
 481         spin_lock_init(&fc->lock);
 482         mutex_init(&fc->inst_mutex);
 483         atomic_set(&fc->count, 1);
 484         init_waitqueue_head(&fc->waitq);
 485         init_waitqueue_head(&fc->blocked_waitq);
 486         init_waitqueue_head(&fc->reserved_req_waitq);
 487         INIT_LIST_HEAD(&fc->pending);
 488         INIT_LIST_HEAD(&fc->processing);
 489         INIT_LIST_HEAD(&fc->io);
 490         INIT_LIST_HEAD(&fc->interrupts);
 491         INIT_LIST_HEAD(&fc->bg_queue);
 492         INIT_LIST_HEAD(&fc->entry);
 493         atomic_set(&fc->num_waiting, 0);
 494         fc->khctr = 0;
 495         fc->polled_files = RB_ROOT;
 496         fc->reqctr = 0;
 497         fc->blocked = 1;
 498         fc->attr_version = 1;
 499         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
 500 }
 501 EXPORT_SYMBOL_GPL(fuse_conn_init);
 502
 503 void fuse_conn_put(struct fuse_conn *fc)
 504 {
 505         if (atomic_dec_and_test(&fc->count)) {
 506                 if (fc->destroy_req)
 507                         fuse_request_free(fc->destroy_req);
 508                 mutex_destroy(&fc->inst_mutex);
 509                 fc->release(fc);
 510         }
 511 }
 512 EXPORT_SYMBOL_GPL(fuse_conn_put);
 513
 514 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
 515 {
 516         atomic_inc(&fc->count);
 517         return fc;
 518 }
 519 EXPORT_SYMBOL_GPL(fuse_conn_get);
 520
 521 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
 522 {
 523         struct fuse_attr attr;
 524         memset(&attr, 0, sizeof(attr));
 525
 526         attr.mode = mode;
 527         attr.ino = FUSE_ROOT_ID;
 528         attr.nlink = 1;
 529         return fuse_iget(sb, 1, 0, &attr, 0, 0);
 530 }
 531
 532 struct fuse_inode_handle {
 533         u64 nodeid;
 534         u32 generation;
 535 };
 536
 537 static struct dentry *fuse_get_dentry(struct super_block *sb,
 538                                       struct fuse_inode_handle *handle)
 539 {
 540         struct fuse_conn *fc = get_fuse_conn_super(sb);
 541         struct inode *inode;
 542         struct dentry *entry;
 543         int err = -ESTALE;
 544
 545         if (handle->nodeid == 0)
 546                 goto out_err;
 547
 548         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
 549         if (!inode) {
 550                 struct fuse_entry_out outarg;
 551                 struct qstr name;
 552
 553                 if (!fc->export_support)
 554                         goto out_err;
 555
 556                 name.len = 1;
 557                 name.name = ".";
 558                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
 559                                        &inode);
 560                 if (err && err != -ENOENT)
 561                         goto out_err;
 562                 if (err || !inode) {
 563                         err = -ESTALE;
 564                         goto out_err;
 565                 }
 566                 err = -EIO;
 567                 if (get_node_id(inode) != handle->nodeid)
 568                         goto out_iput;
 569         }
 570         err = -ESTALE;
 571         if (inode->i_generation != handle->generation)
 572                 goto out_iput;
 573
 574         entry = d_obtain_alias(inode);
 575         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID) {
 576                 entry->d_op = &fuse_dentry_operations;
 577                 fuse_invalidate_entry_cache(entry);
 578         }
 579
 580         return entry;
 581
 582  out_iput:
 583         iput(inode);
 584  out_err:
 585         return ERR_PTR(err);
 586 }
 587
 588 static int fuse_encode_fh(struct dentry *dentry, u32 *fh, int *max_len,
 589                            int connectable)
 590 {
 591         struct inode *inode = dentry->d_inode;
 592         bool encode_parent = connectable && !S_ISDIR(inode->i_mode);
 593         int len = encode_parent ? 6 : 3;
 594         u64 nodeid;
 595         u32 generation;
 596
 597         if (*max_len < len)
 598                 return  255;
 599
 600         nodeid = get_fuse_inode(inode)->nodeid;
 601         generation = inode->i_generation;
 602
 603         fh[0] = (u32)(nodeid >> 32);
 604         fh[1] = (u32)(nodeid & 0xffffffff);
 605         fh[2] = generation;
 606
 607         if (encode_parent) {
 608                 struct inode *parent;
 609
 610                 spin_lock(&dentry->d_lock);
 611                 parent = dentry->d_parent->d_inode;
 612                 nodeid = get_fuse_inode(parent)->nodeid;
 613                 generation = parent->i_generation;
 614                 spin_unlock(&dentry->d_lock);
 615
 616                 fh[3] = (u32)(nodeid >> 32);
 617                 fh[4] = (u32)(nodeid & 0xffffffff);
 618                 fh[5] = generation;
 619         }
 620
 621         *max_len = len;
 622         return encode_parent ? 0x82 : 0x81;
 623 }
 624
 625 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
 626                 struct fid *fid, int fh_len, int fh_type)
 627 {
 628         struct fuse_inode_handle handle;
 629
 630         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
 631                 return NULL;
 632
 633         handle.nodeid = (u64) fid->raw[0] << 32;
 634         handle.nodeid |= (u64) fid->raw[1];
 635         handle.generation = fid->raw[2];
 636         return fuse_get_dentry(sb, &handle);
 637 }
 638
 639 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
 640                 struct fid *fid, int fh_len, int fh_type)
 641 {
 642         struct fuse_inode_handle parent;
 643
 644         if (fh_type != 0x82 || fh_len < 6)
 645                 return NULL;
 646
 647         parent.nodeid = (u64) fid->raw[3] << 32;
 648         parent.nodeid |= (u64) fid->raw[4];
 649         parent.generation = fid->raw[5];
 650         return fuse_get_dentry(sb, &parent);
 651 }
 652
 653 static struct dentry *fuse_get_parent(struct dentry *child)
 654 {
 655         struct inode *child_inode = child->d_inode;
 656         struct fuse_conn *fc = get_fuse_conn(child_inode);
 657         struct inode *inode;
 658         struct dentry *parent;
 659         struct fuse_entry_out outarg;
 660         struct qstr name;
 661         int err;
 662
 663         if (!fc->export_support)
 664                 return ERR_PTR(-ESTALE);
 665
 666         name.len = 2;
 667         name.name = "..";
 668         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
 669                                &name, &outarg, &inode);
 670         if (err) {
 671                 if (err == -ENOENT)
 672                         return ERR_PTR(-ESTALE);
 673                 return ERR_PTR(err);
 674         }
 675
 676         parent = d_obtain_alias(inode);
 677         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID) {
 678                 parent->d_op = &fuse_dentry_operations;
 679                 fuse_invalidate_entry_cache(parent);
 680         }
 681
 682         return parent;
 683 }
 684
 685 static const struct export_operations fuse_export_operations = {
 686         .fh_to_dentry   = fuse_fh_to_dentry,
 687         .fh_to_parent   = fuse_fh_to_parent,
 688         .encode_fh      = fuse_encode_fh,
 689         .get_parent     = fuse_get_parent,
 690 };
 691
 692 static const struct super_operations fuse_super_operations = {
 693         .alloc_inode    = fuse_alloc_inode,
 694         .destroy_inode  = fuse_destroy_inode,
 695         .clear_inode    = fuse_clear_inode,
 696         .drop_inode     = generic_delete_inode,
 697         .remount_fs     = fuse_remount_fs,
 698         .put_super      = fuse_put_super,
 699         .umount_begin   = fuse_umount_begin,
 700         .statfs         = fuse_statfs,
 701         .show_options   = fuse_show_options,
 702 };
 703
 704 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
 705 {
 706         struct fuse_init_out *arg = &req->misc.init_out;
 707
 708         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
 709                 fc->conn_error = 1;
 710         else {
 711                 unsigned long ra_pages;
 712
 713                 if (arg->minor >= 6) {
 714                         ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
 715                         if (arg->flags & FUSE_ASYNC_READ)
 716                                 fc->async_read = 1;
 717                         if (!(arg->flags & FUSE_POSIX_LOCKS))
 718                                 fc->no_lock = 1;
 719                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
 720                                 fc->atomic_o_trunc = 1;
 721                         if (arg->minor >= 9) {
 722                                 /* LOOKUP has dependency on proto version */
 723                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
 724                                         fc->export_support = 1;
 725                         }
 726                         if (arg->flags & FUSE_BIG_WRITES)
 727                                 fc->big_writes = 1;
 728                 } else {
 729                         ra_pages = fc->max_read / PAGE_CACHE_SIZE;
 730                         fc->no_lock = 1;
 731                 }
 732
 733                 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
 734                 fc->minor = arg->minor;
 735                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
 736                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
 737                 fc->conn_init = 1;
 738         }
 739         fc->blocked = 0;
 740         wake_up_all(&fc->blocked_waitq);
 741 }
 742
 743 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
 744 {
 745         struct fuse_init_in *arg = &req->misc.init_in;
 746
 747         arg->major = FUSE_KERNEL_VERSION;
 748         arg->minor = FUSE_KERNEL_MINOR_VERSION;
 749         arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
 750         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
 751                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES;
 752         req->in.h.opcode = FUSE_INIT;
 753         req->in.numargs = 1;
 754         req->in.args[0].size = sizeof(*arg);
 755         req->in.args[0].value = arg;
 756         req->out.numargs = 1;
 757         /* Variable length arguement used for backward compatibility
 758            with interface version < 7.5.  Rest of init_out is zeroed
 759            by do_get_request(), so a short reply is not a problem */
 760         req->out.argvar = 1;
 761         req->out.args[0].size = sizeof(struct fuse_init_out);
 762         req->out.args[0].value = &req->misc.init_out;
 763         req->end = process_init_reply;
 764         fuse_request_send_background(fc, req);
 765 }
 766
 767 static void fuse_free_conn(struct fuse_conn *fc)
 768 {
 769         kfree(fc);
 770 }
 771
 772 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
 773 {
 774         int err;
 775
 776         fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
 777         fc->bdi.unplug_io_fn = default_unplug_io_fn;
 778         /* fuse does it's own writeback accounting */
 779         fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB;
 780
 781         err = bdi_init(&fc->bdi);
 782         if (err)
 783                 return err;
 784
 785         fc->bdi_initialized = 1;
 786
 787         if (sb->s_bdev) {
 788                 err =  bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
 789                                     MAJOR(fc->dev), MINOR(fc->dev));
 790         } else {
 791                 err = bdi_register_dev(&fc->bdi, fc->dev);
 792         }
 793
 794         if (err)
 795                 return err;
 796
 797         /*
 798          * For a single fuse filesystem use max 1% of dirty +
 799          * writeback threshold.
 800          *
 801          * This gives about 1M of write buffer for memory maps on a
 802          * machine with 1G and 10% dirty_ratio, which should be more
 803          * than enough.
 804          *
 805          * Privileged users can raise it by writing to
 806          *
 807          *    /sys/class/bdi/<bdi>/max_ratio
 808          */
 809         bdi_set_max_ratio(&fc->bdi, 1);
 810
 811         return 0;
 812 }
 813
 814 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
 815 {
 816         struct fuse_conn *fc;
 817         struct inode *root;
 818         struct fuse_mount_data d;
 819         struct file *file;
 820         struct dentry *root_dentry;
 821         struct fuse_req *init_req;
 822         int err;
 823         int is_bdev = sb->s_bdev != NULL;
 824
 825         err = -EINVAL;
 826         if (sb->s_flags & MS_MANDLOCK)
 827                 goto err;
 828
 829         if (!parse_fuse_opt((char *) data, &d, is_bdev))
 830                 goto err;
 831
 832         if (is_bdev) {
 833 #ifdef CONFIG_BLOCK
 834                 err = -EINVAL;
 835                 if (!sb_set_blocksize(sb, d.blksize))
 836                         goto err;
 837 #endif
 838         } else {
 839                 sb->s_blocksize = PAGE_CACHE_SIZE;
 840                 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 841         }
 842         sb->s_magic = FUSE_SUPER_MAGIC;
 843         sb->s_op = &fuse_super_operations;
 844         sb->s_maxbytes = MAX_LFS_FILESIZE;
 845         sb->s_export_op = &fuse_export_operations;
 846
 847         file = fget(d.fd);
 848         err = -EINVAL;
 849         if (!file)
 850                 goto err;
 851
 852         if (file->f_op != &fuse_dev_operations)
 853                 goto err_fput;
 854
 855         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
 856         err = -ENOMEM;
 857         if (!fc)
 858                 goto err_fput;
 859
 860         fuse_conn_init(fc);
 861
 862         fc->dev = sb->s_dev;
 863         err = fuse_bdi_init(fc, sb);
 864         if (err)
 865                 goto err_put_conn;
 866
 867         fc->release = fuse_free_conn;
 868         fc->flags = d.flags;
 869         fc->user_id = d.user_id;
 870         fc->group_id = d.group_id;
 871         fc->max_read = max_t(unsigned, 4096, d.max_read);
 872
 873         /* Used by get_root_inode() */
 874         sb->s_fs_info = fc;
 875
 876         err = -ENOMEM;
 877         root = fuse_get_root_inode(sb, d.rootmode);
 878         if (!root)
 879                 goto err_put_conn;
 880
 881         root_dentry = d_alloc_root(root);
 882         if (!root_dentry) {
 883                 iput(root);
 884                 goto err_put_conn;
 885         }
 886
 887         init_req = fuse_request_alloc();
 888         if (!init_req)
 889                 goto err_put_root;
 890
 891         if (is_bdev) {
 892                 fc->destroy_req = fuse_request_alloc();
 893                 if (!fc->destroy_req)
 894                         goto err_free_init_req;
 895         }
 896
 897         mutex_lock(&fuse_mutex);
 898         err = -EINVAL;
 899         if (file->private_data)
 900                 goto err_unlock;
 901
 902         err = fuse_ctl_add_conn(fc);
 903         if (err)
 904                 goto err_unlock;
 905
 906         list_add_tail(&fc->entry, &fuse_conn_list);
 907         sb->s_root = root_dentry;
 908         fc->connected = 1;
 909         file->private_data = fuse_conn_get(fc);
 910         mutex_unlock(&fuse_mutex);
 911         /*
 912          * atomic_dec_and_test() in fput() provides the necessary
 913          * memory barrier for file->private_data to be visible on all
 914          * CPUs after this
 915          */
 916         fput(file);
 917
 918         fuse_send_init(fc, init_req);
 919
 920         return 0;
 921
 922  err_unlock:
 923         mutex_unlock(&fuse_mutex);
 924  err_free_init_req:
 925         fuse_request_free(init_req);
 926  err_put_root:
 927         dput(root_dentry);
 928  err_put_conn:
 929         fuse_bdi_destroy(fc);
 930         fuse_conn_put(fc);
 931  err_fput:
 932         fput(file);
 933  err:
 934         return err;
 935 }
 936
 937 static int fuse_get_sb(struct file_system_type *fs_type,
 938                        int flags, const char *dev_name,
 939                        void *raw_data, struct vfsmount *mnt)
 940 {
 941         return get_sb_nodev(fs_type, flags, raw_data, fuse_fill_super, mnt);
 942 }
 943
 944 static struct file_system_type fuse_fs_type = {
 945         .owner          = THIS_MODULE,
 946         .name           = "fuse",
 947         .fs_flags       = FS_HAS_SUBTYPE,
 948         .get_sb         = fuse_get_sb,
 949         .kill_sb        = kill_anon_super,
 950 };
 951
 952 #ifdef CONFIG_BLOCK
 953 static int fuse_get_sb_blk(struct file_system_type *fs_type,
 954                            int flags, const char *dev_name,
 955                            void *raw_data, struct vfsmount *mnt)
 956 {
 957         return get_sb_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super,
 958                            mnt);
 959 }
 960
 961 static struct file_system_type fuseblk_fs_type = {
 962         .owner          = THIS_MODULE,
 963         .name           = "fuseblk",
 964         .get_sb         = fuse_get_sb_blk,
 965         .kill_sb        = kill_block_super,
 966         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
 967 };
 968
 969 static inline int register_fuseblk(void)
 970 {
 971         return register_filesystem(&fuseblk_fs_type);
 972 }
 973
 974 static inline void unregister_fuseblk(void)
 975 {
 976         unregister_filesystem(&fuseblk_fs_type);
 977 }
 978 #else
 979 static inline int register_fuseblk(void)
 980 {
 981         return 0;
 982 }
 983
 984 static inline void unregister_fuseblk(void)
 985 {
 986 }
 987 #endif
 988
 989 static void fuse_inode_init_once(void *foo)
 990 {
 991         struct inode *inode = foo;
 992
 993         inode_init_once(inode);
 994 }
 995
 996 static int __init fuse_fs_init(void)
 997 {
 998         int err;
 999
1000         err = register_filesystem(&fuse_fs_type);
1001         if (err)
1002                 goto out;
1003
1004         err = register_fuseblk();
1005         if (err)
1006                 goto out_unreg;
1007
1008         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1009                                               sizeof(struct fuse_inode),
1010                                               0, SLAB_HWCACHE_ALIGN,
1011                                               fuse_inode_init_once);
1012         err = -ENOMEM;
1013         if (!fuse_inode_cachep)
1014                 goto out_unreg2;
1015
1016         return 0;
1017
1018  out_unreg2:
1019         unregister_fuseblk();
1020  out_unreg:
1021         unregister_filesystem(&fuse_fs_type);
1022  out:
1023         return err;
1024 }
1025
1026 static void fuse_fs_cleanup(void)
1027 {
1028         unregister_filesystem(&fuse_fs_type);
1029         unregister_fuseblk();
1030         kmem_cache_destroy(fuse_inode_cachep);
1031 }
1032
1033 static struct kobject *fuse_kobj;
1034 static struct kobject *connections_kobj;
1035
1036 static int fuse_sysfs_init(void)
1037 {
1038         int err;
1039
1040         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1041         if (!fuse_kobj) {
1042                 err = -ENOMEM;
1043                 goto out_err;
1044         }
1045
1046         connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1047         if (!connections_kobj) {
1048                 err = -ENOMEM;
1049                 goto out_fuse_unregister;
1050         }
1051
1052         return 0;
1053
1054  out_fuse_unregister:
1055         kobject_put(fuse_kobj);
1056  out_err:
1057         return err;
1058 }
1059
1060 static void fuse_sysfs_cleanup(void)
1061 {
1062         kobject_put(connections_kobj);
1063         kobject_put(fuse_kobj);
1064 }
1065
1066 static int __init fuse_init(void)
1067 {
1068         int res;
1069
1070         printk(KERN_INFO "fuse init (API version %i.%i)\n",
1071                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1072
1073         INIT_LIST_HEAD(&fuse_conn_list);
1074         res = fuse_fs_init();
1075         if (res)
1076                 goto err;
1077
1078         res = fuse_dev_init();
1079         if (res)
1080                 goto err_fs_cleanup;
1081
1082         res = fuse_sysfs_init();
1083         if (res)
1084                 goto err_dev_cleanup;
1085
1086         res = fuse_ctl_init();
1087         if (res)
1088                 goto err_sysfs_cleanup;
1089
1090         return 0;
1091
1092  err_sysfs_cleanup:
1093         fuse_sysfs_cleanup();
1094  err_dev_cleanup:
1095         fuse_dev_cleanup();
1096  err_fs_cleanup:
1097         fuse_fs_cleanup();
1098  err:
1099         return res;
1100 }
1101
1102 static void __exit fuse_exit(void)
1103 {
1104         printk(KERN_DEBUG "fuse exit\n");
1105
1106         fuse_ctl_cleanup();
1107         fuse_sysfs_cleanup();
1108         fuse_fs_cleanup();
1109         fuse_dev_cleanup();
1110 }
1111
1112 module_init(fuse_init);
1113 module_exit(fuse_exit);