fs/fuse/inode.c

   1 /*
   2   FUSE: Filesystem in Userspace
   3   Copyright (C) 2001-2008  Miklos Szeredi <miklos@szeredi.hu>
   4
   5   This program can be distributed under the terms of the GNU GPL.
   6   See the file COPYING.
   7 */
   8
   9 #include "fuse_i.h"
  10
  11 #include <linux/pagemap.h>
  12 #include <linux/slab.h>
  13 #include <linux/file.h>
  14 #include <linux/seq_file.h>
  15 #include <linux/init.h>
  16 #include <linux/module.h>
  17 #include <linux/moduleparam.h>
  18 #include <linux/parser.h>
  19 #include <linux/statfs.h>
  20 #include <linux/random.h>
  21 #include <linux/sched.h>
  22 #include <linux/exportfs.h>
  23
  24 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  25 MODULE_DESCRIPTION("Filesystem in Userspace");
  26 MODULE_LICENSE("GPL");
  27
  28 static struct kmem_cache *fuse_inode_cachep;
  29 struct list_head fuse_conn_list;
  30 DEFINE_MUTEX(fuse_mutex);
  31
  32 static int set_global_limit(const char *val, struct kernel_param *kp);
  33
  34 unsigned max_user_bgreq;
  35 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
  36                   &max_user_bgreq, 0644);
  37 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
  38 MODULE_PARM_DESC(max_user_bgreq,
  39  "Global limit for the maximum number of backgrounded requests an "
  40  "unprivileged user can set");
  41
  42 unsigned max_user_congthresh;
  43 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
  44                   &max_user_congthresh, 0644);
  45 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
  46 MODULE_PARM_DESC(max_user_congthresh,
  47  "Global limit for the maximum congestion threshold an "
  48  "unprivileged user can set");
  49
  50 #define FUSE_SUPER_MAGIC 0x65735546
  51
  52 #define FUSE_DEFAULT_BLKSIZE 512
  53
  54 /** Maximum number of outstanding background requests */
  55 #define FUSE_DEFAULT_MAX_BACKGROUND 12
  56
  57 /** Congestion starts at 75% of maximum */
  58 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
  59
  60 struct fuse_mount_data {
  61         int fd;
  62         unsigned rootmode;
  63         unsigned user_id;
  64         unsigned group_id;
  65         unsigned fd_present:1;
  66         unsigned rootmode_present:1;
  67         unsigned user_id_present:1;
  68         unsigned group_id_present:1;
  69         unsigned flags;
  70         unsigned max_read;
  71         unsigned blksize;
  72 };
  73
  74 struct fuse_forget_link *fuse_alloc_forget(void)
  75 {
  76         return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL);
  77 }
  78
  79 static struct inode *fuse_alloc_inode(struct super_block *sb)
  80 {
  81         struct inode *inode;
  82         struct fuse_inode *fi;
  83
  84         inode = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
  85         if (!inode)
  86                 return NULL;
  87
  88         fi = get_fuse_inode(inode);
  89         fi->i_time = 0;
  90         fi->nodeid = 0;
  91         fi->nlookup = 0;
  92         fi->attr_version = 0;
  93         fi->writectr = 0;
  94         INIT_LIST_HEAD(&fi->write_files);
  95         INIT_LIST_HEAD(&fi->queued_writes);
  96         INIT_LIST_HEAD(&fi->writepages);
  97         init_waitqueue_head(&fi->page_waitq);
  98         fi->forget = fuse_alloc_forget();
  99         if (!fi->forget) {
 100                 kmem_cache_free(fuse_inode_cachep, inode);
 101                 return NULL;
 102         }
 103
 104         return inode;
 105 }
 106
 107 static void fuse_i_callback(struct rcu_head *head)
 108 {
 109         struct inode *inode = container_of(head, struct inode, i_rcu);
 110         INIT_LIST_HEAD(&inode->i_dentry);
 111         kmem_cache_free(fuse_inode_cachep, inode);
 112 }
 113
 114 static void fuse_destroy_inode(struct inode *inode)
 115 {
 116         struct fuse_inode *fi = get_fuse_inode(inode);
 117         BUG_ON(!list_empty(&fi->write_files));
 118         BUG_ON(!list_empty(&fi->queued_writes));
 119         kfree(fi->forget);
 120         call_rcu(&inode->i_rcu, fuse_i_callback);
 121 }
 122
 123 static void fuse_evict_inode(struct inode *inode)
 124 {
 125         truncate_inode_pages(&inode->i_data, 0);
 126         end_writeback(inode);
 127         if (inode->i_sb->s_flags & MS_ACTIVE) {
 128                 struct fuse_conn *fc = get_fuse_conn(inode);
 129                 struct fuse_inode *fi = get_fuse_inode(inode);
 130                 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
 131                 fi->forget = NULL;
 132         }
 133 }
 134
 135 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
 136 {
 137         if (*flags & MS_MANDLOCK)
 138                 return -EINVAL;
 139
 140         return 0;
 141 }
 142
 143 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
 144                                    u64 attr_valid)
 145 {
 146         struct fuse_conn *fc = get_fuse_conn(inode);
 147         struct fuse_inode *fi = get_fuse_inode(inode);
 148
 149         fi->attr_version = ++fc->attr_version;
 150         fi->i_time = attr_valid;
 151
 152         inode->i_ino     = attr->ino;
 153         inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
 154         set_nlink(inode, attr->nlink);
 155         inode->i_uid     = attr->uid;
 156         inode->i_gid     = attr->gid;
 157         inode->i_blocks  = attr->blocks;
 158         inode->i_atime.tv_sec   = attr->atime;
 159         inode->i_atime.tv_nsec  = attr->atimensec;
 160         inode->i_mtime.tv_sec   = attr->mtime;
 161         inode->i_mtime.tv_nsec  = attr->mtimensec;
 162         inode->i_ctime.tv_sec   = attr->ctime;
 163         inode->i_ctime.tv_nsec  = attr->ctimensec;
 164
 165         if (attr->blksize != 0)
 166                 inode->i_blkbits = ilog2(attr->blksize);
 167         else
 168                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
 169
 170         /*
 171          * Don't set the sticky bit in i_mode, unless we want the VFS
 172          * to check permissions.  This prevents failures due to the
 173          * check in may_delete().
 174          */
 175         fi->orig_i_mode = inode->i_mode;
 176         if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
 177                 inode->i_mode &= ~S_ISVTX;
 178 }
 179
 180 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
 181                             u64 attr_valid, u64 attr_version)
 182 {
 183         struct fuse_conn *fc = get_fuse_conn(inode);
 184         struct fuse_inode *fi = get_fuse_inode(inode);
 185         loff_t oldsize;
 186
 187         spin_lock(&fc->lock);
 188         if (attr_version != 0 && fi->attr_version > attr_version) {
 189                 spin_unlock(&fc->lock);
 190                 return;
 191         }
 192
 193         fuse_change_attributes_common(inode, attr, attr_valid);
 194
 195         oldsize = inode->i_size;
 196         i_size_write(inode, attr->size);
 197         spin_unlock(&fc->lock);
 198
 199         if (S_ISREG(inode->i_mode) && oldsize != attr->size) {
 200                 truncate_pagecache(inode, oldsize, attr->size);
 201                 invalidate_inode_pages2(inode->i_mapping);
 202         }
 203 }
 204
 205 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
 206 {
 207         inode->i_mode = attr->mode & S_IFMT;
 208         inode->i_size = attr->size;
 209         if (S_ISREG(inode->i_mode)) {
 210                 fuse_init_common(inode);
 211                 fuse_init_file_inode(inode);
 212         } else if (S_ISDIR(inode->i_mode))
 213                 fuse_init_dir(inode);
 214         else if (S_ISLNK(inode->i_mode))
 215                 fuse_init_symlink(inode);
 216         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
 217                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
 218                 fuse_init_common(inode);
 219                 init_special_inode(inode, inode->i_mode,
 220                                    new_decode_dev(attr->rdev));
 221         } else
 222                 BUG();
 223 }
 224
 225 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
 226 {
 227         u64 nodeid = *(u64 *) _nodeidp;
 228         if (get_node_id(inode) == nodeid)
 229                 return 1;
 230         else
 231                 return 0;
 232 }
 233
 234 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
 235 {
 236         u64 nodeid = *(u64 *) _nodeidp;
 237         get_fuse_inode(inode)->nodeid = nodeid;
 238         return 0;
 239 }
 240
 241 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
 242                         int generation, struct fuse_attr *attr,
 243                         u64 attr_valid, u64 attr_version)
 244 {
 245         struct inode *inode;
 246         struct fuse_inode *fi;
 247         struct fuse_conn *fc = get_fuse_conn_super(sb);
 248
 249  retry:
 250         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
 251         if (!inode)
 252                 return NULL;
 253
 254         if ((inode->i_state & I_NEW)) {
 255                 inode->i_flags |= S_NOATIME|S_NOCMTIME;
 256                 inode->i_generation = generation;
 257                 inode->i_data.backing_dev_info = &fc->bdi;
 258                 fuse_init_inode(inode, attr);
 259                 unlock_new_inode(inode);
 260         } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
 261                 /* Inode has changed type, any I/O on the old should fail */
 262                 make_bad_inode(inode);
 263                 iput(inode);
 264                 goto retry;
 265         }
 266
 267         fi = get_fuse_inode(inode);
 268         spin_lock(&fc->lock);
 269         fi->nlookup++;
 270         spin_unlock(&fc->lock);
 271         fuse_change_attributes(inode, attr, attr_valid, attr_version);
 272
 273         return inode;
 274 }
 275
 276 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
 277                              loff_t offset, loff_t len)
 278 {
 279         struct inode *inode;
 280         pgoff_t pg_start;
 281         pgoff_t pg_end;
 282
 283         inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
 284         if (!inode)
 285                 return -ENOENT;
 286
 287         fuse_invalidate_attr(inode);
 288         if (offset >= 0) {
 289                 pg_start = offset >> PAGE_CACHE_SHIFT;
 290                 if (len <= 0)
 291                         pg_end = -1;
 292                 else
 293                         pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
 294                 invalidate_inode_pages2_range(inode->i_mapping,
 295                                               pg_start, pg_end);
 296         }
 297         iput(inode);
 298         return 0;
 299 }
 300
 301 static void fuse_umount_begin(struct super_block *sb)
 302 {
 303         fuse_abort_conn(get_fuse_conn_super(sb));
 304 }
 305
 306 static void fuse_send_destroy(struct fuse_conn *fc)
 307 {
 308         struct fuse_req *req = fc->destroy_req;
 309         if (req && fc->conn_init) {
 310                 fc->destroy_req = NULL;
 311                 req->in.h.opcode = FUSE_DESTROY;
 312                 req->force = 1;
 313                 fuse_request_send(fc, req);
 314                 fuse_put_request(fc, req);
 315         }
 316 }
 317
 318 static void fuse_bdi_destroy(struct fuse_conn *fc)
 319 {
 320         if (fc->bdi_initialized)
 321                 bdi_destroy(&fc->bdi);
 322 }
 323
 324 void fuse_conn_kill(struct fuse_conn *fc)
 325 {
 326         spin_lock(&fc->lock);
 327         fc->connected = 0;
 328         fc->blocked = 0;
 329         spin_unlock(&fc->lock);
 330         /* Flush all readers on this fs */
 331         kill_fasync(&fc->fasync, SIGIO, POLL_IN);
 332         wake_up_all(&fc->waitq);
 333         wake_up_all(&fc->blocked_waitq);
 334         wake_up_all(&fc->reserved_req_waitq);
 335         mutex_lock(&fuse_mutex);
 336         list_del(&fc->entry);
 337         fuse_ctl_remove_conn(fc);
 338         mutex_unlock(&fuse_mutex);
 339         fuse_bdi_destroy(fc);
 340 }
 341 EXPORT_SYMBOL_GPL(fuse_conn_kill);
 342
 343 static void fuse_put_super(struct super_block *sb)
 344 {
 345         struct fuse_conn *fc = get_fuse_conn_super(sb);
 346
 347         fuse_send_destroy(fc);
 348         fuse_conn_kill(fc);
 349         fuse_conn_put(fc);
 350 }
 351
 352 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
 353 {
 354         stbuf->f_type    = FUSE_SUPER_MAGIC;
 355         stbuf->f_bsize   = attr->bsize;
 356         stbuf->f_frsize  = attr->frsize;
 357         stbuf->f_blocks  = attr->blocks;
 358         stbuf->f_bfree   = attr->bfree;
 359         stbuf->f_bavail  = attr->bavail;
 360         stbuf->f_files   = attr->files;
 361         stbuf->f_ffree   = attr->ffree;
 362         stbuf->f_namelen = attr->namelen;
 363         /* fsid is left zero */
 364 }
 365
 366 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
 367 {
 368         struct super_block *sb = dentry->d_sb;
 369         struct fuse_conn *fc = get_fuse_conn_super(sb);
 370         struct fuse_req *req;
 371         struct fuse_statfs_out outarg;
 372         int err;
 373
 374         if (!fuse_allow_task(fc, current)) {
 375                 buf->f_type = FUSE_SUPER_MAGIC;
 376                 return 0;
 377         }
 378
 379         req = fuse_get_req(fc);
 380         if (IS_ERR(req))
 381                 return PTR_ERR(req);
 382
 383         memset(&outarg, 0, sizeof(outarg));
 384         req->in.numargs = 0;
 385         req->in.h.opcode = FUSE_STATFS;
 386         req->in.h.nodeid = get_node_id(dentry->d_inode);
 387         req->out.numargs = 1;
 388         req->out.args[0].size =
 389                 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
 390         req->out.args[0].value = &outarg;
 391         fuse_request_send(fc, req);
 392         err = req->out.h.error;
 393         if (!err)
 394                 convert_fuse_statfs(buf, &outarg.st);
 395         fuse_put_request(fc, req);
 396         return err;
 397 }
 398
 399 enum {
 400         OPT_FD,
 401         OPT_ROOTMODE,
 402         OPT_USER_ID,
 403         OPT_GROUP_ID,
 404         OPT_DEFAULT_PERMISSIONS,
 405         OPT_ALLOW_OTHER,
 406         OPT_MAX_READ,
 407         OPT_BLKSIZE,
 408         OPT_ERR
 409 };
 410
 411 static const match_table_t tokens = {
 412         {OPT_FD,                        "fd=%u"},
 413         {OPT_ROOTMODE,                  "rootmode=%o"},
 414         {OPT_USER_ID,                   "user_id=%u"},
 415         {OPT_GROUP_ID,                  "group_id=%u"},
 416         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 417         {OPT_ALLOW_OTHER,               "allow_other"},
 418         {OPT_MAX_READ,                  "max_read=%u"},
 419         {OPT_BLKSIZE,                   "blksize=%u"},
 420         {OPT_ERR,                       NULL}
 421 };
 422
 423 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
 424 {
 425         char *p;
 426         memset(d, 0, sizeof(struct fuse_mount_data));
 427         d->max_read = ~0;
 428         d->blksize = FUSE_DEFAULT_BLKSIZE;
 429
 430         while ((p = strsep(&opt, ",")) != NULL) {
 431                 int token;
 432                 int value;
 433                 substring_t args[MAX_OPT_ARGS];
 434                 if (!*p)
 435                         continue;
 436
 437                 token = match_token(p, tokens, args);
 438                 switch (token) {
 439                 case OPT_FD:
 440                         if (match_int(&args[0], &value))
 441                                 return 0;
 442                         d->fd = value;
 443                         d->fd_present = 1;
 444                         break;
 445
 446                 case OPT_ROOTMODE:
 447                         if (match_octal(&args[0], &value))
 448                                 return 0;
 449                         if (!fuse_valid_type(value))
 450                                 return 0;
 451                         d->rootmode = value;
 452                         d->rootmode_present = 1;
 453                         break;
 454
 455                 case OPT_USER_ID:
 456                         if (match_int(&args[0], &value))
 457                                 return 0;
 458                         d->user_id = value;
 459                         d->user_id_present = 1;
 460                         break;
 461
 462                 case OPT_GROUP_ID:
 463                         if (match_int(&args[0], &value))
 464                                 return 0;
 465                         d->group_id = value;
 466                         d->group_id_present = 1;
 467                         break;
 468
 469                 case OPT_DEFAULT_PERMISSIONS:
 470                         d->flags |= FUSE_DEFAULT_PERMISSIONS;
 471                         break;
 472
 473                 case OPT_ALLOW_OTHER:
 474                         d->flags |= FUSE_ALLOW_OTHER;
 475                         break;
 476
 477                 case OPT_MAX_READ:
 478                         if (match_int(&args[0], &value))
 479                                 return 0;
 480                         d->max_read = value;
 481                         break;
 482
 483                 case OPT_BLKSIZE:
 484                         if (!is_bdev || match_int(&args[0], &value))
 485                                 return 0;
 486                         d->blksize = value;
 487                         break;
 488
 489                 default:
 490                         return 0;
 491                 }
 492         }
 493
 494         if (!d->fd_present || !d->rootmode_present ||
 495             !d->user_id_present || !d->group_id_present)
 496                 return 0;
 497
 498         return 1;
 499 }
 500
 501 static int fuse_show_options(struct seq_file *m, struct vfsmount *mnt)
 502 {
 503         struct fuse_conn *fc = get_fuse_conn_super(mnt->mnt_sb);
 504
 505         seq_printf(m, ",user_id=%u", fc->user_id);
 506         seq_printf(m, ",group_id=%u", fc->group_id);
 507         if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
 508                 seq_puts(m, ",default_permissions");
 509         if (fc->flags & FUSE_ALLOW_OTHER)
 510                 seq_puts(m, ",allow_other");
 511         if (fc->max_read != ~0)
 512                 seq_printf(m, ",max_read=%u", fc->max_read);
 513         if (mnt->mnt_sb->s_bdev &&
 514             mnt->mnt_sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
 515                 seq_printf(m, ",blksize=%lu", mnt->mnt_sb->s_blocksize);
 516         return 0;
 517 }
 518
 519 void fuse_conn_init(struct fuse_conn *fc)
 520 {
 521         memset(fc, 0, sizeof(*fc));
 522         spin_lock_init(&fc->lock);
 523         mutex_init(&fc->inst_mutex);
 524         init_rwsem(&fc->killsb);
 525         atomic_set(&fc->count, 1);
 526         init_waitqueue_head(&fc->waitq);
 527         init_waitqueue_head(&fc->blocked_waitq);
 528         init_waitqueue_head(&fc->reserved_req_waitq);
 529         INIT_LIST_HEAD(&fc->pending);
 530         INIT_LIST_HEAD(&fc->processing);
 531         INIT_LIST_HEAD(&fc->io);
 532         INIT_LIST_HEAD(&fc->interrupts);
 533         INIT_LIST_HEAD(&fc->bg_queue);
 534         INIT_LIST_HEAD(&fc->entry);
 535         fc->forget_list_tail = &fc->forget_list_head;
 536         atomic_set(&fc->num_waiting, 0);
 537         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
 538         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
 539         fc->khctr = 0;
 540         fc->polled_files = RB_ROOT;
 541         fc->reqctr = 0;
 542         fc->blocked = 1;
 543         fc->attr_version = 1;
 544         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
 545 }
 546 EXPORT_SYMBOL_GPL(fuse_conn_init);
 547
 548 void fuse_conn_put(struct fuse_conn *fc)
 549 {
 550         if (atomic_dec_and_test(&fc->count)) {
 551                 if (fc->destroy_req)
 552                         fuse_request_free(fc->destroy_req);
 553                 mutex_destroy(&fc->inst_mutex);
 554                 fc->release(fc);
 555         }
 556 }
 557 EXPORT_SYMBOL_GPL(fuse_conn_put);
 558
 559 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
 560 {
 561         atomic_inc(&fc->count);
 562         return fc;
 563 }
 564 EXPORT_SYMBOL_GPL(fuse_conn_get);
 565
 566 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
 567 {
 568         struct fuse_attr attr;
 569         memset(&attr, 0, sizeof(attr));
 570
 571         attr.mode = mode;
 572         attr.ino = FUSE_ROOT_ID;
 573         attr.nlink = 1;
 574         return fuse_iget(sb, 1, 0, &attr, 0, 0);
 575 }
 576
 577 struct fuse_inode_handle {
 578         u64 nodeid;
 579         u32 generation;
 580 };
 581
 582 static struct dentry *fuse_get_dentry(struct super_block *sb,
 583                                       struct fuse_inode_handle *handle)
 584 {
 585         struct fuse_conn *fc = get_fuse_conn_super(sb);
 586         struct inode *inode;
 587         struct dentry *entry;
 588         int err = -ESTALE;
 589
 590         if (handle->nodeid == 0)
 591                 goto out_err;
 592
 593         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
 594         if (!inode) {
 595                 struct fuse_entry_out outarg;
 596                 struct qstr name;
 597
 598                 if (!fc->export_support)
 599                         goto out_err;
 600
 601                 name.len = 1;
 602                 name.name = ".";
 603                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
 604                                        &inode);
 605                 if (err && err != -ENOENT)
 606                         goto out_err;
 607                 if (err || !inode) {
 608                         err = -ESTALE;
 609                         goto out_err;
 610                 }
 611                 err = -EIO;
 612                 if (get_node_id(inode) != handle->nodeid)
 613                         goto out_iput;
 614         }
 615         err = -ESTALE;
 616         if (inode->i_generation != handle->generation)
 617                 goto out_iput;
 618
 619         entry = d_obtain_alias(inode);
 620         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
 621                 fuse_invalidate_entry_cache(entry);
 622
 623         return entry;
 624
 625  out_iput:
 626         iput(inode);
 627  out_err:
 628         return ERR_PTR(err);
 629 }
 630
 631 static int fuse_encode_fh(struct dentry *dentry, u32 *fh, int *max_len,
 632                            int connectable)
 633 {
 634         struct inode *inode = dentry->d_inode;
 635         bool encode_parent = connectable && !S_ISDIR(inode->i_mode);
 636         int len = encode_parent ? 6 : 3;
 637         u64 nodeid;
 638         u32 generation;
 639
 640         if (*max_len < len) {
 641                 *max_len = len;
 642                 return  255;
 643         }
 644
 645         nodeid = get_fuse_inode(inode)->nodeid;
 646         generation = inode->i_generation;
 647
 648         fh[0] = (u32)(nodeid >> 32);
 649         fh[1] = (u32)(nodeid & 0xffffffff);
 650         fh[2] = generation;
 651
 652         if (encode_parent) {
 653                 struct inode *parent;
 654
 655                 spin_lock(&dentry->d_lock);
 656                 parent = dentry->d_parent->d_inode;
 657                 nodeid = get_fuse_inode(parent)->nodeid;
 658                 generation = parent->i_generation;
 659                 spin_unlock(&dentry->d_lock);
 660
 661                 fh[3] = (u32)(nodeid >> 32);
 662                 fh[4] = (u32)(nodeid & 0xffffffff);
 663                 fh[5] = generation;
 664         }
 665
 666         *max_len = len;
 667         return encode_parent ? 0x82 : 0x81;
 668 }
 669
 670 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
 671                 struct fid *fid, int fh_len, int fh_type)
 672 {
 673         struct fuse_inode_handle handle;
 674
 675         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
 676                 return NULL;
 677
 678         handle.nodeid = (u64) fid->raw[0] << 32;
 679         handle.nodeid |= (u64) fid->raw[1];
 680         handle.generation = fid->raw[2];
 681         return fuse_get_dentry(sb, &handle);
 682 }
 683
 684 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
 685                 struct fid *fid, int fh_len, int fh_type)
 686 {
 687         struct fuse_inode_handle parent;
 688
 689         if (fh_type != 0x82 || fh_len < 6)
 690                 return NULL;
 691
 692         parent.nodeid = (u64) fid->raw[3] << 32;
 693         parent.nodeid |= (u64) fid->raw[4];
 694         parent.generation = fid->raw[5];
 695         return fuse_get_dentry(sb, &parent);
 696 }
 697
 698 static struct dentry *fuse_get_parent(struct dentry *child)
 699 {
 700         struct inode *child_inode = child->d_inode;
 701         struct fuse_conn *fc = get_fuse_conn(child_inode);
 702         struct inode *inode;
 703         struct dentry *parent;
 704         struct fuse_entry_out outarg;
 705         struct qstr name;
 706         int err;
 707
 708         if (!fc->export_support)
 709                 return ERR_PTR(-ESTALE);
 710
 711         name.len = 2;
 712         name.name = "..";
 713         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
 714                                &name, &outarg, &inode);
 715         if (err) {
 716                 if (err == -ENOENT)
 717                         return ERR_PTR(-ESTALE);
 718                 return ERR_PTR(err);
 719         }
 720
 721         parent = d_obtain_alias(inode);
 722         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
 723                 fuse_invalidate_entry_cache(parent);
 724
 725         return parent;
 726 }
 727
 728 static const struct export_operations fuse_export_operations = {
 729         .fh_to_dentry   = fuse_fh_to_dentry,
 730         .fh_to_parent   = fuse_fh_to_parent,
 731         .encode_fh      = fuse_encode_fh,
 732         .get_parent     = fuse_get_parent,
 733 };
 734
 735 static const struct super_operations fuse_super_operations = {
 736         .alloc_inode    = fuse_alloc_inode,
 737         .destroy_inode  = fuse_destroy_inode,
 738         .evict_inode    = fuse_evict_inode,
 739         .drop_inode     = generic_delete_inode,
 740         .remount_fs     = fuse_remount_fs,
 741         .put_super      = fuse_put_super,
 742         .umount_begin   = fuse_umount_begin,
 743         .statfs         = fuse_statfs,
 744         .show_options   = fuse_show_options,
 745 };
 746
 747 static void sanitize_global_limit(unsigned *limit)
 748 {
 749         if (*limit == 0)
 750                 *limit = ((num_physpages << PAGE_SHIFT) >> 13) /
 751                          sizeof(struct fuse_req);
 752
 753         if (*limit >= 1 << 16)
 754                 *limit = (1 << 16) - 1;
 755 }
 756
 757 static int set_global_limit(const char *val, struct kernel_param *kp)
 758 {
 759         int rv;
 760
 761         rv = param_set_uint(val, kp);
 762         if (rv)
 763                 return rv;
 764
 765         sanitize_global_limit((unsigned *)kp->arg);
 766
 767         return 0;
 768 }
 769
 770 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
 771 {
 772         int cap_sys_admin = capable(CAP_SYS_ADMIN);
 773
 774         if (arg->minor < 13)
 775                 return;
 776
 777         sanitize_global_limit(&max_user_bgreq);
 778         sanitize_global_limit(&max_user_congthresh);
 779
 780         if (arg->max_background) {
 781                 fc->max_background = arg->max_background;
 782
 783                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
 784                         fc->max_background = max_user_bgreq;
 785         }
 786         if (arg->congestion_threshold) {
 787                 fc->congestion_threshold = arg->congestion_threshold;
 788
 789                 if (!cap_sys_admin &&
 790                     fc->congestion_threshold > max_user_congthresh)
 791                         fc->congestion_threshold = max_user_congthresh;
 792         }
 793 }
 794
 795 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
 796 {
 797         struct fuse_init_out *arg = &req->misc.init_out;
 798
 799         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
 800                 fc->conn_error = 1;
 801         else {
 802                 unsigned long ra_pages;
 803
 804                 process_init_limits(fc, arg);
 805
 806                 if (arg->minor >= 6) {
 807                         ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
 808                         if (arg->flags & FUSE_ASYNC_READ)
 809                                 fc->async_read = 1;
 810                         if (!(arg->flags & FUSE_POSIX_LOCKS))
 811                                 fc->no_lock = 1;
 812                         if (arg->minor >= 17) {
 813                                 if (!(arg->flags & FUSE_FLOCK_LOCKS))
 814                                         fc->no_flock = 1;
 815                         } else {
 816                                 if (!(arg->flags & FUSE_POSIX_LOCKS))
 817                                         fc->no_flock = 1;
 818                         }
 819                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
 820                                 fc->atomic_o_trunc = 1;
 821                         if (arg->minor >= 9) {
 822                                 /* LOOKUP has dependency on proto version */
 823                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
 824                                         fc->export_support = 1;
 825                         }
 826                         if (arg->flags & FUSE_BIG_WRITES)
 827                                 fc->big_writes = 1;
 828                         if (arg->flags & FUSE_DONT_MASK)
 829                                 fc->dont_mask = 1;
 830                 } else {
 831                         ra_pages = fc->max_read / PAGE_CACHE_SIZE;
 832                         fc->no_lock = 1;
 833                         fc->no_flock = 1;
 834                 }
 835
 836                 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
 837                 fc->minor = arg->minor;
 838                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
 839                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
 840                 fc->conn_init = 1;
 841         }
 842         fc->blocked = 0;
 843         wake_up_all(&fc->blocked_waitq);
 844 }
 845
 846 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
 847 {
 848         struct fuse_init_in *arg = &req->misc.init_in;
 849
 850         arg->major = FUSE_KERNEL_VERSION;
 851         arg->minor = FUSE_KERNEL_MINOR_VERSION;
 852         arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
 853         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
 854                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
 855                 FUSE_FLOCK_LOCKS;
 856         req->in.h.opcode = FUSE_INIT;
 857         req->in.numargs = 1;
 858         req->in.args[0].size = sizeof(*arg);
 859         req->in.args[0].value = arg;
 860         req->out.numargs = 1;
 861         /* Variable length argument used for backward compatibility
 862            with interface version < 7.5.  Rest of init_out is zeroed
 863            by do_get_request(), so a short reply is not a problem */
 864         req->out.argvar = 1;
 865         req->out.args[0].size = sizeof(struct fuse_init_out);
 866         req->out.args[0].value = &req->misc.init_out;
 867         req->end = process_init_reply;
 868         fuse_request_send_background(fc, req);
 869 }
 870
 871 static void fuse_free_conn(struct fuse_conn *fc)
 872 {
 873         kfree(fc);
 874 }
 875
 876 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
 877 {
 878         int err;
 879
 880         fc->bdi.name = "fuse";
 881         fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
 882         /* fuse does it's own writeback accounting */
 883         fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB;
 884
 885         err = bdi_init(&fc->bdi);
 886         if (err)
 887                 return err;
 888
 889         fc->bdi_initialized = 1;
 890
 891         if (sb->s_bdev) {
 892                 err =  bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
 893                                     MAJOR(fc->dev), MINOR(fc->dev));
 894         } else {
 895                 err = bdi_register_dev(&fc->bdi, fc->dev);
 896         }
 897
 898         if (err)
 899                 return err;
 900
 901         /*
 902          * For a single fuse filesystem use max 1% of dirty +
 903          * writeback threshold.
 904          *
 905          * This gives about 1M of write buffer for memory maps on a
 906          * machine with 1G and 10% dirty_ratio, which should be more
 907          * than enough.
 908          *
 909          * Privileged users can raise it by writing to
 910          *
 911          *    /sys/class/bdi/<bdi>/max_ratio
 912          */
 913         bdi_set_max_ratio(&fc->bdi, 1);
 914
 915         return 0;
 916 }
 917
 918 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
 919 {
 920         struct fuse_conn *fc;
 921         struct inode *root;
 922         struct fuse_mount_data d;
 923         struct file *file;
 924         struct dentry *root_dentry;
 925         struct fuse_req *init_req;
 926         int err;
 927         int is_bdev = sb->s_bdev != NULL;
 928
 929         err = -EINVAL;
 930         if (sb->s_flags & MS_MANDLOCK)
 931                 goto err;
 932
 933         sb->s_flags &= ~MS_NOSEC;
 934
 935         if (!parse_fuse_opt((char *) data, &d, is_bdev))
 936                 goto err;
 937
 938         if (is_bdev) {
 939 #ifdef CONFIG_BLOCK
 940                 err = -EINVAL;
 941                 if (!sb_set_blocksize(sb, d.blksize))
 942                         goto err;
 943 #endif
 944         } else {
 945                 sb->s_blocksize = PAGE_CACHE_SIZE;
 946                 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 947         }
 948         sb->s_magic = FUSE_SUPER_MAGIC;
 949         sb->s_op = &fuse_super_operations;
 950         sb->s_maxbytes = MAX_LFS_FILESIZE;
 951         sb->s_export_op = &fuse_export_operations;
 952
 953         file = fget(d.fd);
 954         err = -EINVAL;
 955         if (!file)
 956                 goto err;
 957
 958         if (file->f_op != &fuse_dev_operations)
 959                 goto err_fput;
 960
 961         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
 962         err = -ENOMEM;
 963         if (!fc)
 964                 goto err_fput;
 965
 966         fuse_conn_init(fc);
 967
 968         fc->dev = sb->s_dev;
 969         fc->sb = sb;
 970         err = fuse_bdi_init(fc, sb);
 971         if (err)
 972                 goto err_put_conn;
 973
 974         sb->s_bdi = &fc->bdi;
 975
 976         /* Handle umasking inside the fuse code */
 977         if (sb->s_flags & MS_POSIXACL)
 978                 fc->dont_mask = 1;
 979         sb->s_flags |= MS_POSIXACL;
 980
 981         fc->release = fuse_free_conn;
 982         fc->flags = d.flags;
 983         fc->user_id = d.user_id;
 984         fc->group_id = d.group_id;
 985         fc->max_read = max_t(unsigned, 4096, d.max_read);
 986
 987         /* Used by get_root_inode() */
 988         sb->s_fs_info = fc;
 989
 990         err = -ENOMEM;
 991         root = fuse_get_root_inode(sb, d.rootmode);
 992         if (!root)
 993                 goto err_put_conn;
 994
 995         root_dentry = d_alloc_root(root);
 996         if (!root_dentry) {
 997                 iput(root);
 998                 goto err_put_conn;
 999         }
1000         /* only now - we want root dentry with NULL ->d_op */
1001         sb->s_d_op = &fuse_dentry_operations;
1002
1003         init_req = fuse_request_alloc();
1004         if (!init_req)
1005                 goto err_put_root;
1006
1007         if (is_bdev) {
1008                 fc->destroy_req = fuse_request_alloc();
1009                 if (!fc->destroy_req)
1010                         goto err_free_init_req;
1011         }
1012
1013         mutex_lock(&fuse_mutex);
1014         err = -EINVAL;
1015         if (file->private_data)
1016                 goto err_unlock;
1017
1018         err = fuse_ctl_add_conn(fc);
1019         if (err)
1020                 goto err_unlock;
1021
1022         list_add_tail(&fc->entry, &fuse_conn_list);
1023         sb->s_root = root_dentry;
1024         fc->connected = 1;
1025         file->private_data = fuse_conn_get(fc);
1026         mutex_unlock(&fuse_mutex);
1027         /*
1028          * atomic_dec_and_test() in fput() provides the necessary
1029          * memory barrier for file->private_data to be visible on all
1030          * CPUs after this
1031          */
1032         fput(file);
1033
1034         fuse_send_init(fc, init_req);
1035
1036         return 0;
1037
1038  err_unlock:
1039         mutex_unlock(&fuse_mutex);
1040  err_free_init_req:
1041         fuse_request_free(init_req);
1042  err_put_root:
1043         dput(root_dentry);
1044  err_put_conn:
1045         fuse_bdi_destroy(fc);
1046         fuse_conn_put(fc);
1047  err_fput:
1048         fput(file);
1049  err:
1050         return err;
1051 }
1052
1053 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1054                        int flags, const char *dev_name,
1055                        void *raw_data)
1056 {
1057         return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1058 }
1059
1060 static void fuse_kill_sb_anon(struct super_block *sb)
1061 {
1062         struct fuse_conn *fc = get_fuse_conn_super(sb);
1063
1064         if (fc) {
1065                 down_write(&fc->killsb);
1066                 fc->sb = NULL;
1067                 up_write(&fc->killsb);
1068         }
1069
1070         kill_anon_super(sb);
1071 }
1072
1073 static struct file_system_type fuse_fs_type = {
1074         .owner          = THIS_MODULE,
1075         .name           = "fuse",
1076         .fs_flags       = FS_HAS_SUBTYPE,
1077         .mount          = fuse_mount,
1078         .kill_sb        = fuse_kill_sb_anon,
1079 };
1080
1081 #ifdef CONFIG_BLOCK
1082 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1083                            int flags, const char *dev_name,
1084                            void *raw_data)
1085 {
1086         return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1087 }
1088
1089 static void fuse_kill_sb_blk(struct super_block *sb)
1090 {
1091         struct fuse_conn *fc = get_fuse_conn_super(sb);
1092
1093         if (fc) {
1094                 down_write(&fc->killsb);
1095                 fc->sb = NULL;
1096                 up_write(&fc->killsb);
1097         }
1098
1099         kill_block_super(sb);
1100 }
1101
1102 static struct file_system_type fuseblk_fs_type = {
1103         .owner          = THIS_MODULE,
1104         .name           = "fuseblk",
1105         .mount          = fuse_mount_blk,
1106         .kill_sb        = fuse_kill_sb_blk,
1107         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1108 };
1109
1110 static inline int register_fuseblk(void)
1111 {
1112         return register_filesystem(&fuseblk_fs_type);
1113 }
1114
1115 static inline void unregister_fuseblk(void)
1116 {
1117         unregister_filesystem(&fuseblk_fs_type);
1118 }
1119 #else
1120 static inline int register_fuseblk(void)
1121 {
1122         return 0;
1123 }
1124
1125 static inline void unregister_fuseblk(void)
1126 {
1127 }
1128 #endif
1129
1130 static void fuse_inode_init_once(void *foo)
1131 {
1132         struct inode *inode = foo;
1133
1134         inode_init_once(inode);
1135 }
1136
1137 static int __init fuse_fs_init(void)
1138 {
1139         int err;
1140
1141         err = register_filesystem(&fuse_fs_type);
1142         if (err)
1143                 goto out;
1144
1145         err = register_fuseblk();
1146         if (err)
1147                 goto out_unreg;
1148
1149         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1150                                               sizeof(struct fuse_inode),
1151                                               0, SLAB_HWCACHE_ALIGN,
1152                                               fuse_inode_init_once);
1153         err = -ENOMEM;
1154         if (!fuse_inode_cachep)
1155                 goto out_unreg2;
1156
1157         return 0;
1158
1159  out_unreg2:
1160         unregister_fuseblk();
1161  out_unreg:
1162         unregister_filesystem(&fuse_fs_type);
1163  out:
1164         return err;
1165 }
1166
1167 static void fuse_fs_cleanup(void)
1168 {
1169         unregister_filesystem(&fuse_fs_type);
1170         unregister_fuseblk();
1171         kmem_cache_destroy(fuse_inode_cachep);
1172 }
1173
1174 static struct kobject *fuse_kobj;
1175 static struct kobject *connections_kobj;
1176
1177 static int fuse_sysfs_init(void)
1178 {
1179         int err;
1180
1181         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1182         if (!fuse_kobj) {
1183                 err = -ENOMEM;
1184                 goto out_err;
1185         }
1186
1187         connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1188         if (!connections_kobj) {
1189                 err = -ENOMEM;
1190                 goto out_fuse_unregister;
1191         }
1192
1193         return 0;
1194
1195  out_fuse_unregister:
1196         kobject_put(fuse_kobj);
1197  out_err:
1198         return err;
1199 }
1200
1201 static void fuse_sysfs_cleanup(void)
1202 {
1203         kobject_put(connections_kobj);
1204         kobject_put(fuse_kobj);
1205 }
1206
1207 static int __init fuse_init(void)
1208 {
1209         int res;
1210
1211         printk(KERN_INFO "fuse init (API version %i.%i)\n",
1212                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1213
1214         INIT_LIST_HEAD(&fuse_conn_list);
1215         res = fuse_fs_init();
1216         if (res)
1217                 goto err;
1218
1219         res = fuse_dev_init();
1220         if (res)
1221                 goto err_fs_cleanup;
1222
1223         res = fuse_sysfs_init();
1224         if (res)
1225                 goto err_dev_cleanup;
1226
1227         res = fuse_ctl_init();
1228         if (res)
1229                 goto err_sysfs_cleanup;
1230
1231         sanitize_global_limit(&max_user_bgreq);
1232         sanitize_global_limit(&max_user_congthresh);
1233
1234         return 0;
1235
1236  err_sysfs_cleanup:
1237         fuse_sysfs_cleanup();
1238  err_dev_cleanup:
1239         fuse_dev_cleanup();
1240  err_fs_cleanup:
1241         fuse_fs_cleanup();
1242  err:
1243         return res;
1244 }
1245
1246 static void __exit fuse_exit(void)
1247 {
1248         printk(KERN_DEBUG "fuse exit\n");
1249
1250         fuse_ctl_cleanup();
1251         fuse_sysfs_cleanup();
1252         fuse_fs_cleanup();
1253         fuse_dev_cleanup();
1254 }
1255
1256 module_init(fuse_init);
1257 module_exit(fuse_exit);