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