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_bdi_destroy(struct fuse_conn *fc)
 281 {
 282         if (fc->bdi_initialized)
 283                 bdi_destroy(&fc->bdi);
 284 }
 285
 286 void fuse_conn_kill(struct fuse_conn *fc)
 287 {
 288         spin_lock(&fc->lock);
 289         fc->connected = 0;
 290         fc->blocked = 0;
 291         spin_unlock(&fc->lock);
 292         /* Flush all readers on this fs */
 293         kill_fasync(&fc->fasync, SIGIO, POLL_IN);
 294         wake_up_all(&fc->waitq);
 295         wake_up_all(&fc->blocked_waitq);
 296         wake_up_all(&fc->reserved_req_waitq);
 297         mutex_lock(&fuse_mutex);
 298         list_del(&fc->entry);
 299         fuse_ctl_remove_conn(fc);
 300         mutex_unlock(&fuse_mutex);
 301         fuse_bdi_destroy(fc);
 302 }
 303 EXPORT_SYMBOL_GPL(fuse_conn_kill);
 304
 305 static void fuse_put_super(struct super_block *sb)
 306 {
 307         struct fuse_conn *fc = get_fuse_conn_super(sb);
 308
 309         fuse_send_destroy(fc);
 310         fuse_conn_kill(fc);
 311         fuse_conn_put(fc);
 312 }
 313
 314 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
 315 {
 316         stbuf->f_type    = FUSE_SUPER_MAGIC;
 317         stbuf->f_bsize   = attr->bsize;
 318         stbuf->f_frsize  = attr->frsize;
 319         stbuf->f_blocks  = attr->blocks;
 320         stbuf->f_bfree   = attr->bfree;
 321         stbuf->f_bavail  = attr->bavail;
 322         stbuf->f_files   = attr->files;
 323         stbuf->f_ffree   = attr->ffree;
 324         stbuf->f_namelen = attr->namelen;
 325         /* fsid is left zero */
 326 }
 327
 328 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
 329 {
 330         struct super_block *sb = dentry->d_sb;
 331         struct fuse_conn *fc = get_fuse_conn_super(sb);
 332         struct fuse_req *req;
 333         struct fuse_statfs_out outarg;
 334         int err;
 335
 336         if (!fuse_allow_task(fc, current)) {
 337                 buf->f_type = FUSE_SUPER_MAGIC;
 338                 return 0;
 339         }
 340
 341         req = fuse_get_req(fc);
 342         if (IS_ERR(req))
 343                 return PTR_ERR(req);
 344
 345         memset(&outarg, 0, sizeof(outarg));
 346         req->in.numargs = 0;
 347         req->in.h.opcode = FUSE_STATFS;
 348         req->in.h.nodeid = get_node_id(dentry->d_inode);
 349         req->out.numargs = 1;
 350         req->out.args[0].size =
 351                 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
 352         req->out.args[0].value = &outarg;
 353         fuse_request_send(fc, req);
 354         err = req->out.h.error;
 355         if (!err)
 356                 convert_fuse_statfs(buf, &outarg.st);
 357         fuse_put_request(fc, req);
 358         return err;
 359 }
 360
 361 enum {
 362         OPT_FD,
 363         OPT_ROOTMODE,
 364         OPT_USER_ID,
 365         OPT_GROUP_ID,
 366         OPT_DEFAULT_PERMISSIONS,
 367         OPT_ALLOW_OTHER,
 368         OPT_MAX_READ,
 369         OPT_BLKSIZE,
 370         OPT_ERR
 371 };
 372
 373 static const match_table_t tokens = {
 374         {OPT_FD,                        "fd=%u"},
 375         {OPT_ROOTMODE,                  "rootmode=%o"},
 376         {OPT_USER_ID,                   "user_id=%u"},
 377         {OPT_GROUP_ID,                  "group_id=%u"},
 378         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 379         {OPT_ALLOW_OTHER,               "allow_other"},
 380         {OPT_MAX_READ,                  "max_read=%u"},
 381         {OPT_BLKSIZE,                   "blksize=%u"},
 382         {OPT_ERR,                       NULL}
 383 };
 384
 385 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
 386 {
 387         char *p;
 388         memset(d, 0, sizeof(struct fuse_mount_data));
 389         d->max_read = ~0;
 390         d->blksize = FUSE_DEFAULT_BLKSIZE;
 391
 392         while ((p = strsep(&opt, ",")) != NULL) {
 393                 int token;
 394                 int value;
 395                 substring_t args[MAX_OPT_ARGS];
 396                 if (!*p)
 397                         continue;
 398
 399                 token = match_token(p, tokens, args);
 400                 switch (token) {
 401                 case OPT_FD:
 402                         if (match_int(&args[0], &value))
 403                                 return 0;
 404                         d->fd = value;
 405                         d->fd_present = 1;
 406                         break;
 407
 408                 case OPT_ROOTMODE:
 409                         if (match_octal(&args[0], &value))
 410                                 return 0;
 411                         if (!fuse_valid_type(value))
 412                                 return 0;
 413                         d->rootmode = value;
 414                         d->rootmode_present = 1;
 415                         break;
 416
 417                 case OPT_USER_ID:
 418                         if (match_int(&args[0], &value))
 419                                 return 0;
 420                         d->user_id = value;
 421                         d->user_id_present = 1;
 422                         break;
 423
 424                 case OPT_GROUP_ID:
 425                         if (match_int(&args[0], &value))
 426                                 return 0;
 427                         d->group_id = value;
 428                         d->group_id_present = 1;
 429                         break;
 430
 431                 case OPT_DEFAULT_PERMISSIONS:
 432                         d->flags |= FUSE_DEFAULT_PERMISSIONS;
 433                         break;
 434
 435                 case OPT_ALLOW_OTHER:
 436                         d->flags |= FUSE_ALLOW_OTHER;
 437                         break;
 438
 439                 case OPT_MAX_READ:
 440                         if (match_int(&args[0], &value))
 441                                 return 0;
 442                         d->max_read = value;
 443                         break;
 444
 445                 case OPT_BLKSIZE:
 446                         if (!is_bdev || match_int(&args[0], &value))
 447                                 return 0;
 448                         d->blksize = value;
 449                         break;
 450
 451                 default:
 452                         return 0;
 453                 }
 454         }
 455
 456         if (!d->fd_present || !d->rootmode_present ||
 457             !d->user_id_present || !d->group_id_present)
 458                 return 0;
 459
 460         return 1;
 461 }
 462
 463 static int fuse_show_options(struct seq_file *m, struct vfsmount *mnt)
 464 {
 465         struct fuse_conn *fc = get_fuse_conn_super(mnt->mnt_sb);
 466
 467         seq_printf(m, ",user_id=%u", fc->user_id);
 468         seq_printf(m, ",group_id=%u", fc->group_id);
 469         if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
 470                 seq_puts(m, ",default_permissions");
 471         if (fc->flags & FUSE_ALLOW_OTHER)
 472                 seq_puts(m, ",allow_other");
 473         if (fc->max_read != ~0)
 474                 seq_printf(m, ",max_read=%u", fc->max_read);
 475         if (mnt->mnt_sb->s_bdev &&
 476             mnt->mnt_sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
 477                 seq_printf(m, ",blksize=%lu", mnt->mnt_sb->s_blocksize);
 478         return 0;
 479 }
 480
 481 void fuse_conn_init(struct fuse_conn *fc)
 482 {
 483         memset(fc, 0, sizeof(*fc));
 484         spin_lock_init(&fc->lock);
 485         mutex_init(&fc->inst_mutex);
 486         atomic_set(&fc->count, 1);
 487         init_waitqueue_head(&fc->waitq);
 488         init_waitqueue_head(&fc->blocked_waitq);
 489         init_waitqueue_head(&fc->reserved_req_waitq);
 490         INIT_LIST_HEAD(&fc->pending);
 491         INIT_LIST_HEAD(&fc->processing);
 492         INIT_LIST_HEAD(&fc->io);
 493         INIT_LIST_HEAD(&fc->interrupts);
 494         INIT_LIST_HEAD(&fc->bg_queue);
 495         INIT_LIST_HEAD(&fc->entry);
 496         atomic_set(&fc->num_waiting, 0);
 497         fc->khctr = 0;
 498         fc->polled_files = RB_ROOT;
 499         fc->reqctr = 0;
 500         fc->blocked = 1;
 501         fc->attr_version = 1;
 502         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
 503 }
 504 EXPORT_SYMBOL_GPL(fuse_conn_init);
 505
 506 void fuse_conn_put(struct fuse_conn *fc)
 507 {
 508         if (atomic_dec_and_test(&fc->count)) {
 509                 if (fc->destroy_req)
 510                         fuse_request_free(fc->destroy_req);
 511                 mutex_destroy(&fc->inst_mutex);
 512                 fc->release(fc);
 513         }
 514 }
 515 EXPORT_SYMBOL_GPL(fuse_conn_put);
 516
 517 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
 518 {
 519         atomic_inc(&fc->count);
 520         return fc;
 521 }
 522 EXPORT_SYMBOL_GPL(fuse_conn_get);
 523
 524 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
 525 {
 526         struct fuse_attr attr;
 527         memset(&attr, 0, sizeof(attr));
 528
 529         attr.mode = mode;
 530         attr.ino = FUSE_ROOT_ID;
 531         attr.nlink = 1;
 532         return fuse_iget(sb, 1, 0, &attr, 0, 0);
 533 }
 534
 535 struct fuse_inode_handle {
 536         u64 nodeid;
 537         u32 generation;
 538 };
 539
 540 static struct dentry *fuse_get_dentry(struct super_block *sb,
 541                                       struct fuse_inode_handle *handle)
 542 {
 543         struct fuse_conn *fc = get_fuse_conn_super(sb);
 544         struct inode *inode;
 545         struct dentry *entry;
 546         int err = -ESTALE;
 547
 548         if (handle->nodeid == 0)
 549                 goto out_err;
 550
 551         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
 552         if (!inode) {
 553                 struct fuse_entry_out outarg;
 554                 struct qstr name;
 555
 556                 if (!fc->export_support)
 557                         goto out_err;
 558
 559                 name.len = 1;
 560                 name.name = ".";
 561                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
 562                                        &inode);
 563                 if (err && err != -ENOENT)
 564                         goto out_err;
 565                 if (err || !inode) {
 566                         err = -ESTALE;
 567                         goto out_err;
 568                 }
 569                 err = -EIO;
 570                 if (get_node_id(inode) != handle->nodeid)
 571                         goto out_iput;
 572         }
 573         err = -ESTALE;
 574         if (inode->i_generation != handle->generation)
 575                 goto out_iput;
 576
 577         entry = d_obtain_alias(inode);
 578         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID) {
 579                 entry->d_op = &fuse_dentry_operations;
 580                 fuse_invalidate_entry_cache(entry);
 581         }
 582
 583         return entry;
 584
 585  out_iput:
 586         iput(inode);
 587  out_err:
 588         return ERR_PTR(err);
 589 }
 590
 591 static int fuse_encode_fh(struct dentry *dentry, u32 *fh, int *max_len,
 592                            int connectable)
 593 {
 594         struct inode *inode = dentry->d_inode;
 595         bool encode_parent = connectable && !S_ISDIR(inode->i_mode);
 596         int len = encode_parent ? 6 : 3;
 597         u64 nodeid;
 598         u32 generation;
 599
 600         if (*max_len < len)
 601                 return  255;
 602
 603         nodeid = get_fuse_inode(inode)->nodeid;
 604         generation = inode->i_generation;
 605
 606         fh[0] = (u32)(nodeid >> 32);
 607         fh[1] = (u32)(nodeid & 0xffffffff);
 608         fh[2] = generation;
 609
 610         if (encode_parent) {
 611                 struct inode *parent;
 612
 613                 spin_lock(&dentry->d_lock);
 614                 parent = dentry->d_parent->d_inode;
 615                 nodeid = get_fuse_inode(parent)->nodeid;
 616                 generation = parent->i_generation;
 617                 spin_unlock(&dentry->d_lock);
 618
 619                 fh[3] = (u32)(nodeid >> 32);
 620                 fh[4] = (u32)(nodeid & 0xffffffff);
 621                 fh[5] = generation;
 622         }
 623
 624         *max_len = len;
 625         return encode_parent ? 0x82 : 0x81;
 626 }
 627
 628 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
 629                 struct fid *fid, int fh_len, int fh_type)
 630 {
 631         struct fuse_inode_handle handle;
 632
 633         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
 634                 return NULL;
 635
 636         handle.nodeid = (u64) fid->raw[0] << 32;
 637         handle.nodeid |= (u64) fid->raw[1];
 638         handle.generation = fid->raw[2];
 639         return fuse_get_dentry(sb, &handle);
 640 }
 641
 642 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
 643                 struct fid *fid, int fh_len, int fh_type)
 644 {
 645         struct fuse_inode_handle parent;
 646
 647         if (fh_type != 0x82 || fh_len < 6)
 648                 return NULL;
 649
 650         parent.nodeid = (u64) fid->raw[3] << 32;
 651         parent.nodeid |= (u64) fid->raw[4];
 652         parent.generation = fid->raw[5];
 653         return fuse_get_dentry(sb, &parent);
 654 }
 655
 656 static struct dentry *fuse_get_parent(struct dentry *child)
 657 {
 658         struct inode *child_inode = child->d_inode;
 659         struct fuse_conn *fc = get_fuse_conn(child_inode);
 660         struct inode *inode;
 661         struct dentry *parent;
 662         struct fuse_entry_out outarg;
 663         struct qstr name;
 664         int err;
 665
 666         if (!fc->export_support)
 667                 return ERR_PTR(-ESTALE);
 668
 669         name.len = 2;
 670         name.name = "..";
 671         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
 672                                &name, &outarg, &inode);
 673         if (err) {
 674                 if (err == -ENOENT)
 675                         return ERR_PTR(-ESTALE);
 676                 return ERR_PTR(err);
 677         }
 678
 679         parent = d_obtain_alias(inode);
 680         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID) {
 681                 parent->d_op = &fuse_dentry_operations;
 682                 fuse_invalidate_entry_cache(parent);
 683         }
 684
 685         return parent;
 686 }
 687
 688 static const struct export_operations fuse_export_operations = {
 689         .fh_to_dentry   = fuse_fh_to_dentry,
 690         .fh_to_parent   = fuse_fh_to_parent,
 691         .encode_fh      = fuse_encode_fh,
 692         .get_parent     = fuse_get_parent,
 693 };
 694
 695 static const struct super_operations fuse_super_operations = {
 696         .alloc_inode    = fuse_alloc_inode,
 697         .destroy_inode  = fuse_destroy_inode,
 698         .clear_inode    = fuse_clear_inode,
 699         .drop_inode     = generic_delete_inode,
 700         .remount_fs     = fuse_remount_fs,
 701         .put_super      = fuse_put_super,
 702         .umount_begin   = fuse_umount_begin,
 703         .statfs         = fuse_statfs,
 704         .show_options   = fuse_show_options,
 705 };
 706
 707 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
 708 {
 709         struct fuse_init_out *arg = &req->misc.init_out;
 710
 711         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
 712                 fc->conn_error = 1;
 713         else {
 714                 unsigned long ra_pages;
 715
 716                 if (arg->minor >= 6) {
 717                         ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
 718                         if (arg->flags & FUSE_ASYNC_READ)
 719                                 fc->async_read = 1;
 720                         if (!(arg->flags & FUSE_POSIX_LOCKS))
 721                                 fc->no_lock = 1;
 722                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
 723                                 fc->atomic_o_trunc = 1;
 724                         if (arg->minor >= 9) {
 725                                 /* LOOKUP has dependency on proto version */
 726                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
 727                                         fc->export_support = 1;
 728                         }
 729                         if (arg->flags & FUSE_BIG_WRITES)
 730                                 fc->big_writes = 1;
 731                 } else {
 732                         ra_pages = fc->max_read / PAGE_CACHE_SIZE;
 733                         fc->no_lock = 1;
 734                 }
 735
 736                 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
 737                 fc->minor = arg->minor;
 738                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
 739                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
 740                 fc->conn_init = 1;
 741         }
 742         fc->blocked = 0;
 743         wake_up_all(&fc->blocked_waitq);
 744 }
 745
 746 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
 747 {
 748         struct fuse_init_in *arg = &req->misc.init_in;
 749
 750         arg->major = FUSE_KERNEL_VERSION;
 751         arg->minor = FUSE_KERNEL_MINOR_VERSION;
 752         arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
 753         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
 754                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES;
 755         req->in.h.opcode = FUSE_INIT;
 756         req->in.numargs = 1;
 757         req->in.args[0].size = sizeof(*arg);
 758         req->in.args[0].value = arg;
 759         req->out.numargs = 1;
 760         /* Variable length arguement used for backward compatibility
 761            with interface version < 7.5.  Rest of init_out is zeroed
 762            by do_get_request(), so a short reply is not a problem */
 763         req->out.argvar = 1;
 764         req->out.args[0].size = sizeof(struct fuse_init_out);
 765         req->out.args[0].value = &req->misc.init_out;
 766         req->end = process_init_reply;
 767         fuse_request_send_background(fc, req);
 768 }
 769
 770 static void fuse_free_conn(struct fuse_conn *fc)
 771 {
 772         kfree(fc);
 773 }
 774
 775 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
 776 {
 777         int err;
 778
 779         fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
 780         fc->bdi.unplug_io_fn = default_unplug_io_fn;
 781         /* fuse does it's own writeback accounting */
 782         fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB;
 783
 784         err = bdi_init(&fc->bdi);
 785         if (err)
 786                 return err;
 787
 788         fc->bdi_initialized = 1;
 789
 790         if (sb->s_bdev) {
 791                 err =  bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
 792                                     MAJOR(fc->dev), MINOR(fc->dev));
 793         } else {
 794                 err = bdi_register_dev(&fc->bdi, fc->dev);
 795         }
 796
 797         if (err)
 798                 return err;
 799
 800         /*
 801          * For a single fuse filesystem use max 1% of dirty +
 802          * writeback threshold.
 803          *
 804          * This gives about 1M of write buffer for memory maps on a
 805          * machine with 1G and 10% dirty_ratio, which should be more
 806          * than enough.
 807          *
 808          * Privileged users can raise it by writing to
 809          *
 810          *    /sys/class/bdi/<bdi>/max_ratio
 811          */
 812         bdi_set_max_ratio(&fc->bdi, 1);
 813
 814         return 0;
 815 }
 816
 817 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
 818 {
 819         struct fuse_conn *fc;
 820         struct inode *root;
 821         struct fuse_mount_data d;
 822         struct file *file;
 823         struct dentry *root_dentry;
 824         struct fuse_req *init_req;
 825         int err;
 826         int is_bdev = sb->s_bdev != NULL;
 827
 828         err = -EINVAL;
 829         if (sb->s_flags & MS_MANDLOCK)
 830                 goto err;
 831
 832         if (!parse_fuse_opt((char *) data, &d, is_bdev))
 833                 goto err;
 834
 835         if (is_bdev) {
 836 #ifdef CONFIG_BLOCK
 837                 err = -EINVAL;
 838                 if (!sb_set_blocksize(sb, d.blksize))
 839                         goto err;
 840 #endif
 841         } else {
 842                 sb->s_blocksize = PAGE_CACHE_SIZE;
 843                 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 844         }
 845         sb->s_magic = FUSE_SUPER_MAGIC;
 846         sb->s_op = &fuse_super_operations;
 847         sb->s_maxbytes = MAX_LFS_FILESIZE;
 848         sb->s_export_op = &fuse_export_operations;
 849
 850         file = fget(d.fd);
 851         err = -EINVAL;
 852         if (!file)
 853                 goto err;
 854
 855         if (file->f_op != &fuse_dev_operations)
 856                 goto err_fput;
 857
 858         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
 859         err = -ENOMEM;
 860         if (!fc)
 861                 goto err_fput;
 862
 863         fuse_conn_init(fc);
 864
 865         fc->dev = sb->s_dev;
 866         err = fuse_bdi_init(fc, sb);
 867         if (err)
 868                 goto err_put_conn;
 869
 870         fc->release = fuse_free_conn;
 871         fc->flags = d.flags;
 872         fc->user_id = d.user_id;
 873         fc->group_id = d.group_id;
 874         fc->max_read = max_t(unsigned, 4096, d.max_read);
 875
 876         /* Used by get_root_inode() */
 877         sb->s_fs_info = fc;
 878
 879         err = -ENOMEM;
 880         root = fuse_get_root_inode(sb, d.rootmode);
 881         if (!root)
 882                 goto err_put_conn;
 883
 884         root_dentry = d_alloc_root(root);
 885         if (!root_dentry) {
 886                 iput(root);
 887                 goto err_put_conn;
 888         }
 889
 890         init_req = fuse_request_alloc();
 891         if (!init_req)
 892                 goto err_put_root;
 893
 894         if (is_bdev) {
 895                 fc->destroy_req = fuse_request_alloc();
 896                 if (!fc->destroy_req)
 897                         goto err_free_init_req;
 898         }
 899
 900         mutex_lock(&fuse_mutex);
 901         err = -EINVAL;
 902         if (file->private_data)
 903                 goto err_unlock;
 904
 905         err = fuse_ctl_add_conn(fc);
 906         if (err)
 907                 goto err_unlock;
 908
 909         list_add_tail(&fc->entry, &fuse_conn_list);
 910         sb->s_root = root_dentry;
 911         fc->connected = 1;
 912         file->private_data = fuse_conn_get(fc);
 913         mutex_unlock(&fuse_mutex);
 914         /*
 915          * atomic_dec_and_test() in fput() provides the necessary
 916          * memory barrier for file->private_data to be visible on all
 917          * CPUs after this
 918          */
 919         fput(file);
 920
 921         fuse_send_init(fc, init_req);
 922
 923         return 0;
 924
 925  err_unlock:
 926         mutex_unlock(&fuse_mutex);
 927  err_free_init_req:
 928         fuse_request_free(init_req);
 929  err_put_root:
 930         dput(root_dentry);
 931  err_put_conn:
 932         fuse_bdi_destroy(fc);
 933         fuse_conn_put(fc);
 934  err_fput:
 935         fput(file);
 936  err:
 937         return err;
 938 }
 939
 940 static int fuse_get_sb(struct file_system_type *fs_type,
 941                        int flags, const char *dev_name,
 942                        void *raw_data, struct vfsmount *mnt)
 943 {
 944         return get_sb_nodev(fs_type, flags, raw_data, fuse_fill_super, mnt);
 945 }
 946
 947 static struct file_system_type fuse_fs_type = {
 948         .owner          = THIS_MODULE,
 949         .name           = "fuse",
 950         .fs_flags       = FS_HAS_SUBTYPE,
 951         .get_sb         = fuse_get_sb,
 952         .kill_sb        = kill_anon_super,
 953 };
 954
 955 #ifdef CONFIG_BLOCK
 956 static int fuse_get_sb_blk(struct file_system_type *fs_type,
 957                            int flags, const char *dev_name,
 958                            void *raw_data, struct vfsmount *mnt)
 959 {
 960         return get_sb_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super,
 961                            mnt);
 962 }
 963
 964 static struct file_system_type fuseblk_fs_type = {
 965         .owner          = THIS_MODULE,
 966         .name           = "fuseblk",
 967         .get_sb         = fuse_get_sb_blk,
 968         .kill_sb        = kill_block_super,
 969         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
 970 };
 971
 972 static inline int register_fuseblk(void)
 973 {
 974         return register_filesystem(&fuseblk_fs_type);
 975 }
 976
 977 static inline void unregister_fuseblk(void)
 978 {
 979         unregister_filesystem(&fuseblk_fs_type);
 980 }
 981 #else
 982 static inline int register_fuseblk(void)
 983 {
 984         return 0;
 985 }
 986
 987 static inline void unregister_fuseblk(void)
 988 {
 989 }
 990 #endif
 991
 992 static void fuse_inode_init_once(void *foo)
 993 {
 994         struct inode *inode = foo;
 995
 996         inode_init_once(inode);
 997 }
 998
 999 static int __init fuse_fs_init(void)
1000 {
1001         int err;
1002
1003         err = register_filesystem(&fuse_fs_type);
1004         if (err)
1005                 goto out;
1006
1007         err = register_fuseblk();
1008         if (err)
1009                 goto out_unreg;
1010
1011         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1012                                               sizeof(struct fuse_inode),
1013                                               0, SLAB_HWCACHE_ALIGN,
1014                                               fuse_inode_init_once);
1015         err = -ENOMEM;
1016         if (!fuse_inode_cachep)
1017                 goto out_unreg2;
1018
1019         return 0;
1020
1021  out_unreg2:
1022         unregister_fuseblk();
1023  out_unreg:
1024         unregister_filesystem(&fuse_fs_type);
1025  out:
1026         return err;
1027 }
1028
1029 static void fuse_fs_cleanup(void)
1030 {
1031         unregister_filesystem(&fuse_fs_type);
1032         unregister_fuseblk();
1033         kmem_cache_destroy(fuse_inode_cachep);
1034 }
1035
1036 static struct kobject *fuse_kobj;
1037 static struct kobject *connections_kobj;
1038
1039 static int fuse_sysfs_init(void)
1040 {
1041         int err;
1042
1043         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1044         if (!fuse_kobj) {
1045                 err = -ENOMEM;
1046                 goto out_err;
1047         }
1048
1049         connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1050         if (!connections_kobj) {
1051                 err = -ENOMEM;
1052                 goto out_fuse_unregister;
1053         }
1054
1055         return 0;
1056
1057  out_fuse_unregister:
1058         kobject_put(fuse_kobj);
1059  out_err:
1060         return err;
1061 }
1062
1063 static void fuse_sysfs_cleanup(void)
1064 {
1065         kobject_put(connections_kobj);
1066         kobject_put(fuse_kobj);
1067 }
1068
1069 static int __init fuse_init(void)
1070 {
1071         int res;
1072
1073         printk(KERN_INFO "fuse init (API version %i.%i)\n",
1074                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1075
1076         INIT_LIST_HEAD(&fuse_conn_list);
1077         res = fuse_fs_init();
1078         if (res)
1079                 goto err;
1080
1081         res = fuse_dev_init();
1082         if (res)
1083                 goto err_fs_cleanup;
1084
1085         res = fuse_sysfs_init();
1086         if (res)
1087                 goto err_dev_cleanup;
1088
1089         res = fuse_ctl_init();
1090         if (res)
1091                 goto err_sysfs_cleanup;
1092
1093         return 0;
1094
1095  err_sysfs_cleanup:
1096         fuse_sysfs_cleanup();
1097  err_dev_cleanup:
1098         fuse_dev_cleanup();
1099  err_fs_cleanup:
1100         fuse_fs_cleanup();
1101  err:
1102         return res;
1103 }
1104
1105 static void __exit fuse_exit(void)
1106 {
1107         printk(KERN_DEBUG "fuse exit\n");
1108
1109         fuse_ctl_cleanup();
1110         fuse_sysfs_cleanup();
1111         fuse_fs_cleanup();
1112         fuse_dev_cleanup();
1113 }
1114
1115 module_init(fuse_init);
1116 module_exit(fuse_exit);