git.oblomov.eu Git - linux-2.6/blob - fs/fuse/inode.c

   1 /*
   2   FUSE: Filesystem in Userspace
   3   Copyright (C) 2001-2006  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         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                 request_send(fc, req);
 273                 fuse_put_request(fc, req);
 274         }
 275 }
 276
 277 static void fuse_put_super(struct super_block *sb)
 278 {
 279         struct fuse_conn *fc = get_fuse_conn_super(sb);
 280
 281         fuse_send_destroy(fc);
 282         spin_lock(&fc->lock);
 283         fc->connected = 0;
 284         fc->blocked = 0;
 285         spin_unlock(&fc->lock);
 286         /* Flush all readers on this fs */
 287         kill_fasync(&fc->fasync, SIGIO, POLL_IN);
 288         wake_up_all(&fc->waitq);
 289         wake_up_all(&fc->blocked_waitq);
 290         wake_up_all(&fc->reserved_req_waitq);
 291         mutex_lock(&fuse_mutex);
 292         list_del(&fc->entry);
 293         fuse_ctl_remove_conn(fc);
 294         mutex_unlock(&fuse_mutex);
 295         fuse_conn_put(fc);
 296 }
 297
 298 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
 299 {
 300         stbuf->f_type    = FUSE_SUPER_MAGIC;
 301         stbuf->f_bsize   = attr->bsize;
 302         stbuf->f_frsize  = attr->frsize;
 303         stbuf->f_blocks  = attr->blocks;
 304         stbuf->f_bfree   = attr->bfree;
 305         stbuf->f_bavail  = attr->bavail;
 306         stbuf->f_files   = attr->files;
 307         stbuf->f_ffree   = attr->ffree;
 308         stbuf->f_namelen = attr->namelen;
 309         /* fsid is left zero */
 310 }
 311
 312 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
 313 {
 314         struct super_block *sb = dentry->d_sb;
 315         struct fuse_conn *fc = get_fuse_conn_super(sb);
 316         struct fuse_req *req;
 317         struct fuse_statfs_out outarg;
 318         int err;
 319
 320         if (!fuse_allow_task(fc, current)) {
 321                 buf->f_type = FUSE_SUPER_MAGIC;
 322                 return 0;
 323         }
 324
 325         req = fuse_get_req(fc);
 326         if (IS_ERR(req))
 327                 return PTR_ERR(req);
 328
 329         memset(&outarg, 0, sizeof(outarg));
 330         req->in.numargs = 0;
 331         req->in.h.opcode = FUSE_STATFS;
 332         req->in.h.nodeid = get_node_id(dentry->d_inode);
 333         req->out.numargs = 1;
 334         req->out.args[0].size =
 335                 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
 336         req->out.args[0].value = &outarg;
 337         request_send(fc, req);
 338         err = req->out.h.error;
 339         if (!err)
 340                 convert_fuse_statfs(buf, &outarg.st);
 341         fuse_put_request(fc, req);
 342         return err;
 343 }
 344
 345 enum {
 346         OPT_FD,
 347         OPT_ROOTMODE,
 348         OPT_USER_ID,
 349         OPT_GROUP_ID,
 350         OPT_DEFAULT_PERMISSIONS,
 351         OPT_ALLOW_OTHER,
 352         OPT_MAX_READ,
 353         OPT_BLKSIZE,
 354         OPT_ERR
 355 };
 356
 357 static const match_table_t tokens = {
 358         {OPT_FD,                        "fd=%u"},
 359         {OPT_ROOTMODE,                  "rootmode=%o"},
 360         {OPT_USER_ID,                   "user_id=%u"},
 361         {OPT_GROUP_ID,                  "group_id=%u"},
 362         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 363         {OPT_ALLOW_OTHER,               "allow_other"},
 364         {OPT_MAX_READ,                  "max_read=%u"},
 365         {OPT_BLKSIZE,                   "blksize=%u"},
 366         {OPT_ERR,                       NULL}
 367 };
 368
 369 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
 370 {
 371         char *p;
 372         memset(d, 0, sizeof(struct fuse_mount_data));
 373         d->max_read = ~0;
 374         d->blksize = FUSE_DEFAULT_BLKSIZE;
 375
 376         while ((p = strsep(&opt, ",")) != NULL) {
 377                 int token;
 378                 int value;
 379                 substring_t args[MAX_OPT_ARGS];
 380                 if (!*p)
 381                         continue;
 382
 383                 token = match_token(p, tokens, args);
 384                 switch (token) {
 385                 case OPT_FD:
 386                         if (match_int(&args[0], &value))
 387                                 return 0;
 388                         d->fd = value;
 389                         d->fd_present = 1;
 390                         break;
 391
 392                 case OPT_ROOTMODE:
 393                         if (match_octal(&args[0], &value))
 394                                 return 0;
 395                         if (!fuse_valid_type(value))
 396                                 return 0;
 397                         d->rootmode = value;
 398                         d->rootmode_present = 1;
 399                         break;
 400
 401                 case OPT_USER_ID:
 402                         if (match_int(&args[0], &value))
 403                                 return 0;
 404                         d->user_id = value;
 405                         d->user_id_present = 1;
 406                         break;
 407
 408                 case OPT_GROUP_ID:
 409                         if (match_int(&args[0], &value))
 410                                 return 0;
 411                         d->group_id = value;
 412                         d->group_id_present = 1;
 413                         break;
 414
 415                 case OPT_DEFAULT_PERMISSIONS:
 416                         d->flags |= FUSE_DEFAULT_PERMISSIONS;
 417                         break;
 418
 419                 case OPT_ALLOW_OTHER:
 420                         d->flags |= FUSE_ALLOW_OTHER;
 421                         break;
 422
 423                 case OPT_MAX_READ:
 424                         if (match_int(&args[0], &value))
 425                                 return 0;
 426                         d->max_read = value;
 427                         break;
 428
 429                 case OPT_BLKSIZE:
 430                         if (!is_bdev || match_int(&args[0], &value))
 431                                 return 0;
 432                         d->blksize = value;
 433                         break;
 434
 435                 default:
 436                         return 0;
 437                 }
 438         }
 439
 440         if (!d->fd_present || !d->rootmode_present ||
 441             !d->user_id_present || !d->group_id_present)
 442                 return 0;
 443
 444         return 1;
 445 }
 446
 447 static int fuse_show_options(struct seq_file *m, struct vfsmount *mnt)
 448 {
 449         struct fuse_conn *fc = get_fuse_conn_super(mnt->mnt_sb);
 450
 451         seq_printf(m, ",user_id=%u", fc->user_id);
 452         seq_printf(m, ",group_id=%u", fc->group_id);
 453         if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
 454                 seq_puts(m, ",default_permissions");
 455         if (fc->flags & FUSE_ALLOW_OTHER)
 456                 seq_puts(m, ",allow_other");
 457         if (fc->max_read != ~0)
 458                 seq_printf(m, ",max_read=%u", fc->max_read);
 459         if (mnt->mnt_sb->s_bdev &&
 460             mnt->mnt_sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
 461                 seq_printf(m, ",blksize=%lu", mnt->mnt_sb->s_blocksize);
 462         return 0;
 463 }
 464
 465 static struct fuse_conn *new_conn(struct super_block *sb)
 466 {
 467         struct fuse_conn *fc;
 468         int err;
 469
 470         fc = kzalloc(sizeof(*fc), GFP_KERNEL);
 471         if (fc) {
 472                 spin_lock_init(&fc->lock);
 473                 mutex_init(&fc->inst_mutex);
 474                 atomic_set(&fc->count, 1);
 475                 init_waitqueue_head(&fc->waitq);
 476                 init_waitqueue_head(&fc->blocked_waitq);
 477                 init_waitqueue_head(&fc->reserved_req_waitq);
 478                 INIT_LIST_HEAD(&fc->pending);
 479                 INIT_LIST_HEAD(&fc->processing);
 480                 INIT_LIST_HEAD(&fc->io);
 481                 INIT_LIST_HEAD(&fc->interrupts);
 482                 INIT_LIST_HEAD(&fc->bg_queue);
 483                 atomic_set(&fc->num_waiting, 0);
 484                 fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
 485                 fc->bdi.unplug_io_fn = default_unplug_io_fn;
 486                 /* fuse does it's own writeback accounting */
 487                 fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB;
 488                 fc->dev = sb->s_dev;
 489                 err = bdi_init(&fc->bdi);
 490                 if (err)
 491                         goto error_kfree;
 492                 if (sb->s_bdev) {
 493                         err = bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
 494                                            MAJOR(fc->dev), MINOR(fc->dev));
 495                 } else {
 496                         err = bdi_register_dev(&fc->bdi, fc->dev);
 497                 }
 498                 if (err)
 499                         goto error_bdi_destroy;
 500                 /*
 501                  * For a single fuse filesystem use max 1% of dirty +
 502                  * writeback threshold.
 503                  *
 504                  * This gives about 1M of write buffer for memory maps on a
 505                  * machine with 1G and 10% dirty_ratio, which should be more
 506                  * than enough.
 507                  *
 508                  * Privileged users can raise it by writing to
 509                  *
 510                  *    /sys/class/bdi/<bdi>/max_ratio
 511                  */
 512                 bdi_set_max_ratio(&fc->bdi, 1);
 513                 fc->reqctr = 0;
 514                 fc->blocked = 1;
 515                 fc->attr_version = 1;
 516                 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
 517         }
 518         return fc;
 519
 520 error_bdi_destroy:
 521         bdi_destroy(&fc->bdi);
 522 error_kfree:
 523         mutex_destroy(&fc->inst_mutex);
 524         kfree(fc);
 525         return NULL;
 526 }
 527
 528 void fuse_conn_put(struct fuse_conn *fc)
 529 {
 530         if (atomic_dec_and_test(&fc->count)) {
 531                 if (fc->destroy_req)
 532                         fuse_request_free(fc->destroy_req);
 533                 mutex_destroy(&fc->inst_mutex);
 534                 bdi_destroy(&fc->bdi);
 535                 kfree(fc);
 536         }
 537 }
 538
 539 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
 540 {
 541         atomic_inc(&fc->count);
 542         return fc;
 543 }
 544
 545 static struct inode *get_root_inode(struct super_block *sb, unsigned mode)
 546 {
 547         struct fuse_attr attr;
 548         memset(&attr, 0, sizeof(attr));
 549
 550         attr.mode = mode;
 551         attr.ino = FUSE_ROOT_ID;
 552         attr.nlink = 1;
 553         return fuse_iget(sb, 1, 0, &attr, 0, 0);
 554 }
 555
 556 struct fuse_inode_handle
 557 {
 558         u64 nodeid;
 559         u32 generation;
 560 };
 561
 562 static struct dentry *fuse_get_dentry(struct super_block *sb,
 563                                       struct fuse_inode_handle *handle)
 564 {
 565         struct fuse_conn *fc = get_fuse_conn_super(sb);
 566         struct inode *inode;
 567         struct dentry *entry;
 568         int err = -ESTALE;
 569
 570         if (handle->nodeid == 0)
 571                 goto out_err;
 572
 573         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
 574         if (!inode) {
 575                 struct fuse_entry_out outarg;
 576                 struct qstr name;
 577
 578                 if (!fc->export_support)
 579                         goto out_err;
 580
 581                 name.len = 1;
 582                 name.name = ".";
 583                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
 584                                        &inode);
 585                 if (err && err != -ENOENT)
 586                         goto out_err;
 587                 if (err || !inode) {
 588                         err = -ESTALE;
 589                         goto out_err;
 590                 }
 591                 err = -EIO;
 592                 if (get_node_id(inode) != handle->nodeid)
 593                         goto out_iput;
 594         }
 595         err = -ESTALE;
 596         if (inode->i_generation != handle->generation)
 597                 goto out_iput;
 598
 599         entry = d_obtain_alias(inode);
 600         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID) {
 601                 entry->d_op = &fuse_dentry_operations;
 602                 fuse_invalidate_entry_cache(entry);
 603         }
 604
 605         return entry;
 606
 607  out_iput:
 608         iput(inode);
 609  out_err:
 610         return ERR_PTR(err);
 611 }
 612
 613 static int fuse_encode_fh(struct dentry *dentry, u32 *fh, int *max_len,
 614                            int connectable)
 615 {
 616         struct inode *inode = dentry->d_inode;
 617         bool encode_parent = connectable && !S_ISDIR(inode->i_mode);
 618         int len = encode_parent ? 6 : 3;
 619         u64 nodeid;
 620         u32 generation;
 621
 622         if (*max_len < len)
 623                 return  255;
 624
 625         nodeid = get_fuse_inode(inode)->nodeid;
 626         generation = inode->i_generation;
 627
 628         fh[0] = (u32)(nodeid >> 32);
 629         fh[1] = (u32)(nodeid & 0xffffffff);
 630         fh[2] = generation;
 631
 632         if (encode_parent) {
 633                 struct inode *parent;
 634
 635                 spin_lock(&dentry->d_lock);
 636                 parent = dentry->d_parent->d_inode;
 637                 nodeid = get_fuse_inode(parent)->nodeid;
 638                 generation = parent->i_generation;
 639                 spin_unlock(&dentry->d_lock);
 640
 641                 fh[3] = (u32)(nodeid >> 32);
 642                 fh[4] = (u32)(nodeid & 0xffffffff);
 643                 fh[5] = generation;
 644         }
 645
 646         *max_len = len;
 647         return encode_parent ? 0x82 : 0x81;
 648 }
 649
 650 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
 651                 struct fid *fid, int fh_len, int fh_type)
 652 {
 653         struct fuse_inode_handle handle;
 654
 655         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
 656                 return NULL;
 657
 658         handle.nodeid = (u64) fid->raw[0] << 32;
 659         handle.nodeid |= (u64) fid->raw[1];
 660         handle.generation = fid->raw[2];
 661         return fuse_get_dentry(sb, &handle);
 662 }
 663
 664 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
 665                 struct fid *fid, int fh_len, int fh_type)
 666 {
 667         struct fuse_inode_handle parent;
 668
 669         if (fh_type != 0x82 || fh_len < 6)
 670                 return NULL;
 671
 672         parent.nodeid = (u64) fid->raw[3] << 32;
 673         parent.nodeid |= (u64) fid->raw[4];
 674         parent.generation = fid->raw[5];
 675         return fuse_get_dentry(sb, &parent);
 676 }
 677
 678 static struct dentry *fuse_get_parent(struct dentry *child)
 679 {
 680         struct inode *child_inode = child->d_inode;
 681         struct fuse_conn *fc = get_fuse_conn(child_inode);
 682         struct inode *inode;
 683         struct dentry *parent;
 684         struct fuse_entry_out outarg;
 685         struct qstr name;
 686         int err;
 687
 688         if (!fc->export_support)
 689                 return ERR_PTR(-ESTALE);
 690
 691         name.len = 2;
 692         name.name = "..";
 693         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
 694                                &name, &outarg, &inode);
 695         if (err) {
 696                 if (err == -ENOENT)
 697                         return ERR_PTR(-ESTALE);
 698                 return ERR_PTR(err);
 699         }
 700
 701         parent = d_obtain_alias(inode);
 702         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID) {
 703                 parent->d_op = &fuse_dentry_operations;
 704                 fuse_invalidate_entry_cache(parent);
 705         }
 706
 707         return parent;
 708 }
 709
 710 static const struct export_operations fuse_export_operations = {
 711         .fh_to_dentry   = fuse_fh_to_dentry,
 712         .fh_to_parent   = fuse_fh_to_parent,
 713         .encode_fh      = fuse_encode_fh,
 714         .get_parent     = fuse_get_parent,
 715 };
 716
 717 static const struct super_operations fuse_super_operations = {
 718         .alloc_inode    = fuse_alloc_inode,
 719         .destroy_inode  = fuse_destroy_inode,
 720         .clear_inode    = fuse_clear_inode,
 721         .drop_inode     = generic_delete_inode,
 722         .remount_fs     = fuse_remount_fs,
 723         .put_super      = fuse_put_super,
 724         .umount_begin   = fuse_umount_begin,
 725         .statfs         = fuse_statfs,
 726         .show_options   = fuse_show_options,
 727 };
 728
 729 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
 730 {
 731         struct fuse_init_out *arg = &req->misc.init_out;
 732
 733         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
 734                 fc->conn_error = 1;
 735         else {
 736                 unsigned long ra_pages;
 737
 738                 if (arg->minor >= 6) {
 739                         ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
 740                         if (arg->flags & FUSE_ASYNC_READ)
 741                                 fc->async_read = 1;
 742                         if (!(arg->flags & FUSE_POSIX_LOCKS))
 743                                 fc->no_lock = 1;
 744                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
 745                                 fc->atomic_o_trunc = 1;
 746                         if (arg->minor >= 9) {
 747                                 /* LOOKUP has dependency on proto version */
 748                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
 749                                         fc->export_support = 1;
 750                         }
 751                         if (arg->flags & FUSE_BIG_WRITES)
 752                                 fc->big_writes = 1;
 753                 } else {
 754                         ra_pages = fc->max_read / PAGE_CACHE_SIZE;
 755                         fc->no_lock = 1;
 756                 }
 757
 758                 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
 759                 fc->minor = arg->minor;
 760                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
 761                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
 762                 fc->conn_init = 1;
 763         }
 764         fuse_put_request(fc, req);
 765         fc->blocked = 0;
 766         wake_up_all(&fc->blocked_waitq);
 767 }
 768
 769 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
 770 {
 771         struct fuse_init_in *arg = &req->misc.init_in;
 772
 773         arg->major = FUSE_KERNEL_VERSION;
 774         arg->minor = FUSE_KERNEL_MINOR_VERSION;
 775         arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
 776         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
 777                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES;
 778         req->in.h.opcode = FUSE_INIT;
 779         req->in.numargs = 1;
 780         req->in.args[0].size = sizeof(*arg);
 781         req->in.args[0].value = arg;
 782         req->out.numargs = 1;
 783         /* Variable length arguement used for backward compatibility
 784            with interface version < 7.5.  Rest of init_out is zeroed
 785            by do_get_request(), so a short reply is not a problem */
 786         req->out.argvar = 1;
 787         req->out.args[0].size = sizeof(struct fuse_init_out);
 788         req->out.args[0].value = &req->misc.init_out;
 789         req->end = process_init_reply;
 790         request_send_background(fc, req);
 791 }
 792
 793 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
 794 {
 795         struct fuse_conn *fc;
 796         struct inode *root;
 797         struct fuse_mount_data d;
 798         struct file *file;
 799         struct dentry *root_dentry;
 800         struct fuse_req *init_req;
 801         int err;
 802         int is_bdev = sb->s_bdev != NULL;
 803
 804         if (sb->s_flags & MS_MANDLOCK)
 805                 return -EINVAL;
 806
 807         if (!parse_fuse_opt((char *) data, &d, is_bdev))
 808                 return -EINVAL;
 809
 810         if (is_bdev) {
 811 #ifdef CONFIG_BLOCK
 812                 if (!sb_set_blocksize(sb, d.blksize))
 813                         return -EINVAL;
 814 #endif
 815         } else {
 816                 sb->s_blocksize = PAGE_CACHE_SIZE;
 817                 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 818         }
 819         sb->s_magic = FUSE_SUPER_MAGIC;
 820         sb->s_op = &fuse_super_operations;
 821         sb->s_maxbytes = MAX_LFS_FILESIZE;
 822         sb->s_export_op = &fuse_export_operations;
 823
 824         file = fget(d.fd);
 825         if (!file)
 826                 return -EINVAL;
 827
 828         if (file->f_op != &fuse_dev_operations)
 829                 return -EINVAL;
 830
 831         fc = new_conn(sb);
 832         if (!fc)
 833                 return -ENOMEM;
 834
 835         fc->flags = d.flags;
 836         fc->user_id = d.user_id;
 837         fc->group_id = d.group_id;
 838         fc->max_read = max_t(unsigned, 4096, d.max_read);
 839
 840         /* Used by get_root_inode() */
 841         sb->s_fs_info = fc;
 842
 843         err = -ENOMEM;
 844         root = get_root_inode(sb, d.rootmode);
 845         if (!root)
 846                 goto err;
 847
 848         root_dentry = d_alloc_root(root);
 849         if (!root_dentry) {
 850                 iput(root);
 851                 goto err;
 852         }
 853
 854         init_req = fuse_request_alloc();
 855         if (!init_req)
 856                 goto err_put_root;
 857
 858         if (is_bdev) {
 859                 fc->destroy_req = fuse_request_alloc();
 860                 if (!fc->destroy_req)
 861                         goto err_free_init_req;
 862         }
 863
 864         mutex_lock(&fuse_mutex);
 865         err = -EINVAL;
 866         if (file->private_data)
 867                 goto err_unlock;
 868
 869         err = fuse_ctl_add_conn(fc);
 870         if (err)
 871                 goto err_unlock;
 872
 873         list_add_tail(&fc->entry, &fuse_conn_list);
 874         sb->s_root = root_dentry;
 875         fc->connected = 1;
 876         file->private_data = fuse_conn_get(fc);
 877         mutex_unlock(&fuse_mutex);
 878         /*
 879          * atomic_dec_and_test() in fput() provides the necessary
 880          * memory barrier for file->private_data to be visible on all
 881          * CPUs after this
 882          */
 883         fput(file);
 884
 885         fuse_send_init(fc, init_req);
 886
 887         return 0;
 888
 889  err_unlock:
 890         mutex_unlock(&fuse_mutex);
 891  err_free_init_req:
 892         fuse_request_free(init_req);
 893  err_put_root:
 894         dput(root_dentry);
 895  err:
 896         fput(file);
 897         fuse_conn_put(fc);
 898         return err;
 899 }
 900
 901 static int fuse_get_sb(struct file_system_type *fs_type,
 902                        int flags, const char *dev_name,
 903                        void *raw_data, struct vfsmount *mnt)
 904 {
 905         return get_sb_nodev(fs_type, flags, raw_data, fuse_fill_super, mnt);
 906 }
 907
 908 static struct file_system_type fuse_fs_type = {
 909         .owner          = THIS_MODULE,
 910         .name           = "fuse",
 911         .fs_flags       = FS_HAS_SUBTYPE,
 912         .get_sb         = fuse_get_sb,
 913         .kill_sb        = kill_anon_super,
 914 };
 915
 916 #ifdef CONFIG_BLOCK
 917 static int fuse_get_sb_blk(struct file_system_type *fs_type,
 918                            int flags, const char *dev_name,
 919                            void *raw_data, struct vfsmount *mnt)
 920 {
 921         return get_sb_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super,
 922                            mnt);
 923 }
 924
 925 static struct file_system_type fuseblk_fs_type = {
 926         .owner          = THIS_MODULE,
 927         .name           = "fuseblk",
 928         .get_sb         = fuse_get_sb_blk,
 929         .kill_sb        = kill_block_super,
 930         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
 931 };
 932
 933 static inline int register_fuseblk(void)
 934 {
 935         return register_filesystem(&fuseblk_fs_type);
 936 }
 937
 938 static inline void unregister_fuseblk(void)
 939 {
 940         unregister_filesystem(&fuseblk_fs_type);
 941 }
 942 #else
 943 static inline int register_fuseblk(void)
 944 {
 945         return 0;
 946 }
 947
 948 static inline void unregister_fuseblk(void)
 949 {
 950 }
 951 #endif
 952
 953 static void fuse_inode_init_once(void *foo)
 954 {
 955         struct inode * inode = foo;
 956
 957         inode_init_once(inode);
 958 }
 959
 960 static int __init fuse_fs_init(void)
 961 {
 962         int err;
 963
 964         err = register_filesystem(&fuse_fs_type);
 965         if (err)
 966                 goto out;
 967
 968         err = register_fuseblk();
 969         if (err)
 970                 goto out_unreg;
 971
 972         fuse_inode_cachep = kmem_cache_create("fuse_inode",
 973                                               sizeof(struct fuse_inode),
 974                                               0, SLAB_HWCACHE_ALIGN,
 975                                               fuse_inode_init_once);
 976         err = -ENOMEM;
 977         if (!fuse_inode_cachep)
 978                 goto out_unreg2;
 979
 980         return 0;
 981
 982  out_unreg2:
 983         unregister_fuseblk();
 984  out_unreg:
 985         unregister_filesystem(&fuse_fs_type);
 986  out:
 987         return err;
 988 }
 989
 990 static void fuse_fs_cleanup(void)
 991 {
 992         unregister_filesystem(&fuse_fs_type);
 993         unregister_fuseblk();
 994         kmem_cache_destroy(fuse_inode_cachep);
 995 }
 996
 997 static struct kobject *fuse_kobj;
 998 static struct kobject *connections_kobj;
 999
1000 static int fuse_sysfs_init(void)
1001 {
1002         int err;
1003
1004         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1005         if (!fuse_kobj) {
1006                 err = -ENOMEM;
1007                 goto out_err;
1008         }
1009
1010         connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1011         if (!connections_kobj) {
1012                 err = -ENOMEM;
1013                 goto out_fuse_unregister;
1014         }
1015
1016         return 0;
1017
1018  out_fuse_unregister:
1019         kobject_put(fuse_kobj);
1020  out_err:
1021         return err;
1022 }
1023
1024 static void fuse_sysfs_cleanup(void)
1025 {
1026         kobject_put(connections_kobj);
1027         kobject_put(fuse_kobj);
1028 }
1029
1030 static int __init fuse_init(void)
1031 {
1032         int res;
1033
1034         printk("fuse init (API version %i.%i)\n",
1035                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1036
1037         INIT_LIST_HEAD(&fuse_conn_list);
1038         res = fuse_fs_init();
1039         if (res)
1040                 goto err;
1041
1042         res = fuse_dev_init();
1043         if (res)
1044                 goto err_fs_cleanup;
1045
1046         res = fuse_sysfs_init();
1047         if (res)
1048                 goto err_dev_cleanup;
1049
1050         res = fuse_ctl_init();
1051         if (res)
1052                 goto err_sysfs_cleanup;
1053
1054         return 0;
1055
1056  err_sysfs_cleanup:
1057         fuse_sysfs_cleanup();
1058  err_dev_cleanup:
1059         fuse_dev_cleanup();
1060  err_fs_cleanup:
1061         fuse_fs_cleanup();
1062  err:
1063         return res;
1064 }
1065
1066 static void __exit fuse_exit(void)
1067 {
1068         printk(KERN_DEBUG "fuse exit\n");
1069
1070         fuse_ctl_cleanup();
1071         fuse_sysfs_cleanup();
1072         fuse_fs_cleanup();
1073         fuse_dev_cleanup();
1074 }
1075
1076 module_init(fuse_init);
1077 module_exit(fuse_exit);