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()
  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         fi->orig_ino = 0;
  95         INIT_LIST_HEAD(&fi->write_files);
  96         INIT_LIST_HEAD(&fi->queued_writes);
  97         INIT_LIST_HEAD(&fi->writepages);
  98         init_waitqueue_head(&fi->page_waitq);
  99         fi->forget = fuse_alloc_forget();
 100         if (!fi->forget) {
 101                 kmem_cache_free(fuse_inode_cachep, inode);
 102                 return NULL;
 103         }
 104
 105         return inode;
 106 }
 107
 108 static void fuse_i_callback(struct rcu_head *head)
 109 {
 110         struct inode *inode = container_of(head, struct inode, i_rcu);
 111         INIT_LIST_HEAD(&inode->i_dentry);
 112         kmem_cache_free(fuse_inode_cachep, inode);
 113 }
 114
 115 static void fuse_destroy_inode(struct inode *inode)
 116 {
 117         struct fuse_inode *fi = get_fuse_inode(inode);
 118         BUG_ON(!list_empty(&fi->write_files));
 119         BUG_ON(!list_empty(&fi->queued_writes));
 120         kfree(fi->forget);
 121         call_rcu(&inode->i_rcu, fuse_i_callback);
 122 }
 123
 124 static void fuse_evict_inode(struct inode *inode)
 125 {
 126         truncate_inode_pages(&inode->i_data, 0);
 127         end_writeback(inode);
 128         if (inode->i_sb->s_flags & MS_ACTIVE) {
 129                 struct fuse_conn *fc = get_fuse_conn(inode);
 130                 struct fuse_inode *fi = get_fuse_inode(inode);
 131                 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
 132                 fi->forget = NULL;
 133         }
 134 }
 135
 136 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
 137 {
 138         if (*flags & MS_MANDLOCK)
 139                 return -EINVAL;
 140
 141         return 0;
 142 }
 143
 144 /*
 145  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
 146  * so that it will fit.
 147  */
 148 static ino_t fuse_squash_ino(u64 ino64)
 149 {
 150         ino_t ino = (ino_t) ino64;
 151         if (sizeof(ino_t) < sizeof(u64))
 152                 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
 153         return ino;
 154 }
 155
 156 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
 157                                    u64 attr_valid)
 158 {
 159         struct fuse_conn *fc = get_fuse_conn(inode);
 160         struct fuse_inode *fi = get_fuse_inode(inode);
 161
 162         fi->attr_version = ++fc->attr_version;
 163         fi->i_time = attr_valid;
 164
 165         inode->i_ino     = fuse_squash_ino(attr->ino);
 166         inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
 167         inode->i_nlink   = attr->nlink;
 168         inode->i_uid     = attr->uid;
 169         inode->i_gid     = attr->gid;
 170         inode->i_blocks  = attr->blocks;
 171         inode->i_atime.tv_sec   = attr->atime;
 172         inode->i_atime.tv_nsec  = attr->atimensec;
 173         inode->i_mtime.tv_sec   = attr->mtime;
 174         inode->i_mtime.tv_nsec  = attr->mtimensec;
 175         inode->i_ctime.tv_sec   = attr->ctime;
 176         inode->i_ctime.tv_nsec  = attr->ctimensec;
 177
 178         if (attr->blksize != 0)
 179                 inode->i_blkbits = ilog2(attr->blksize);
 180         else
 181                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
 182
 183         /*
 184          * Don't set the sticky bit in i_mode, unless we want the VFS
 185          * to check permissions.  This prevents failures due to the
 186          * check in may_delete().
 187          */
 188         fi->orig_i_mode = inode->i_mode;
 189         if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
 190                 inode->i_mode &= ~S_ISVTX;
 191
 192         fi->orig_ino = attr->ino;
 193 }
 194
 195 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
 196                             u64 attr_valid, u64 attr_version)
 197 {
 198         struct fuse_conn *fc = get_fuse_conn(inode);
 199         struct fuse_inode *fi = get_fuse_inode(inode);
 200         loff_t oldsize;
 201
 202         spin_lock(&fc->lock);
 203         if (attr_version != 0 && fi->attr_version > attr_version) {
 204                 spin_unlock(&fc->lock);
 205                 return;
 206         }
 207
 208         fuse_change_attributes_common(inode, attr, attr_valid);
 209
 210         oldsize = inode->i_size;
 211         i_size_write(inode, attr->size);
 212         spin_unlock(&fc->lock);
 213
 214         if (S_ISREG(inode->i_mode) && oldsize != attr->size) {
 215                 truncate_pagecache(inode, oldsize, attr->size);
 216                 invalidate_inode_pages2(inode->i_mapping);
 217         }
 218 }
 219
 220 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
 221 {
 222         inode->i_mode = attr->mode & S_IFMT;
 223         inode->i_size = attr->size;
 224         if (S_ISREG(inode->i_mode)) {
 225                 fuse_init_common(inode);
 226                 fuse_init_file_inode(inode);
 227         } else if (S_ISDIR(inode->i_mode))
 228                 fuse_init_dir(inode);
 229         else if (S_ISLNK(inode->i_mode))
 230                 fuse_init_symlink(inode);
 231         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
 232                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
 233                 fuse_init_common(inode);
 234                 init_special_inode(inode, inode->i_mode,
 235                                    new_decode_dev(attr->rdev));
 236         } else
 237                 BUG();
 238 }
 239
 240 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
 241 {
 242         u64 nodeid = *(u64 *) _nodeidp;
 243         if (get_node_id(inode) == nodeid)
 244                 return 1;
 245         else
 246                 return 0;
 247 }
 248
 249 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
 250 {
 251         u64 nodeid = *(u64 *) _nodeidp;
 252         get_fuse_inode(inode)->nodeid = nodeid;
 253         return 0;
 254 }
 255
 256 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
 257                         int generation, struct fuse_attr *attr,
 258                         u64 attr_valid, u64 attr_version)
 259 {
 260         struct inode *inode;
 261         struct fuse_inode *fi;
 262         struct fuse_conn *fc = get_fuse_conn_super(sb);
 263
 264  retry:
 265         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
 266         if (!inode)
 267                 return NULL;
 268
 269         if ((inode->i_state & I_NEW)) {
 270                 inode->i_flags |= S_NOATIME|S_NOCMTIME;
 271                 inode->i_generation = generation;
 272                 inode->i_data.backing_dev_info = &fc->bdi;
 273                 fuse_init_inode(inode, attr);
 274                 unlock_new_inode(inode);
 275         } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
 276                 /* Inode has changed type, any I/O on the old should fail */
 277                 make_bad_inode(inode);
 278                 iput(inode);
 279                 goto retry;
 280         }
 281
 282         fi = get_fuse_inode(inode);
 283         spin_lock(&fc->lock);
 284         fi->nlookup++;
 285         spin_unlock(&fc->lock);
 286         fuse_change_attributes(inode, attr, attr_valid, attr_version);
 287
 288         return inode;
 289 }
 290
 291 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
 292                              loff_t offset, loff_t len)
 293 {
 294         struct inode *inode;
 295         pgoff_t pg_start;
 296         pgoff_t pg_end;
 297
 298         inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
 299         if (!inode)
 300                 return -ENOENT;
 301
 302         fuse_invalidate_attr(inode);
 303         if (offset >= 0) {
 304                 pg_start = offset >> PAGE_CACHE_SHIFT;
 305                 if (len <= 0)
 306                         pg_end = -1;
 307                 else
 308                         pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
 309                 invalidate_inode_pages2_range(inode->i_mapping,
 310                                               pg_start, pg_end);
 311         }
 312         iput(inode);
 313         return 0;
 314 }
 315
 316 static void fuse_umount_begin(struct super_block *sb)
 317 {
 318         fuse_abort_conn(get_fuse_conn_super(sb));
 319 }
 320
 321 static void fuse_send_destroy(struct fuse_conn *fc)
 322 {
 323         struct fuse_req *req = fc->destroy_req;
 324         if (req && fc->conn_init) {
 325                 fc->destroy_req = NULL;
 326                 req->in.h.opcode = FUSE_DESTROY;
 327                 req->force = 1;
 328                 fuse_request_send(fc, req);
 329                 fuse_put_request(fc, req);
 330         }
 331 }
 332
 333 static void fuse_bdi_destroy(struct fuse_conn *fc)
 334 {
 335         if (fc->bdi_initialized)
 336                 bdi_destroy(&fc->bdi);
 337 }
 338
 339 void fuse_conn_kill(struct fuse_conn *fc)
 340 {
 341         spin_lock(&fc->lock);
 342         fc->connected = 0;
 343         fc->blocked = 0;
 344         spin_unlock(&fc->lock);
 345         /* Flush all readers on this fs */
 346         kill_fasync(&fc->fasync, SIGIO, POLL_IN);
 347         wake_up_all(&fc->waitq);
 348         wake_up_all(&fc->blocked_waitq);
 349         wake_up_all(&fc->reserved_req_waitq);
 350         mutex_lock(&fuse_mutex);
 351         list_del(&fc->entry);
 352         fuse_ctl_remove_conn(fc);
 353         mutex_unlock(&fuse_mutex);
 354         fuse_bdi_destroy(fc);
 355 }
 356 EXPORT_SYMBOL_GPL(fuse_conn_kill);
 357
 358 static void fuse_put_super(struct super_block *sb)
 359 {
 360         struct fuse_conn *fc = get_fuse_conn_super(sb);
 361
 362         fuse_send_destroy(fc);
 363         fuse_conn_kill(fc);
 364         fuse_conn_put(fc);
 365 }
 366
 367 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
 368 {
 369         stbuf->f_type    = FUSE_SUPER_MAGIC;
 370         stbuf->f_bsize   = attr->bsize;
 371         stbuf->f_frsize  = attr->frsize;
 372         stbuf->f_blocks  = attr->blocks;
 373         stbuf->f_bfree   = attr->bfree;
 374         stbuf->f_bavail  = attr->bavail;
 375         stbuf->f_files   = attr->files;
 376         stbuf->f_ffree   = attr->ffree;
 377         stbuf->f_namelen = attr->namelen;
 378         /* fsid is left zero */
 379 }
 380
 381 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
 382 {
 383         struct super_block *sb = dentry->d_sb;
 384         struct fuse_conn *fc = get_fuse_conn_super(sb);
 385         struct fuse_req *req;
 386         struct fuse_statfs_out outarg;
 387         int err;
 388
 389         if (!fuse_allow_task(fc, current)) {
 390                 buf->f_type = FUSE_SUPER_MAGIC;
 391                 return 0;
 392         }
 393
 394         req = fuse_get_req(fc);
 395         if (IS_ERR(req))
 396                 return PTR_ERR(req);
 397
 398         memset(&outarg, 0, sizeof(outarg));
 399         req->in.numargs = 0;
 400         req->in.h.opcode = FUSE_STATFS;
 401         req->in.h.nodeid = get_node_id(dentry->d_inode);
 402         req->out.numargs = 1;
 403         req->out.args[0].size =
 404                 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
 405         req->out.args[0].value = &outarg;
 406         fuse_request_send(fc, req);
 407         err = req->out.h.error;
 408         if (!err)
 409                 convert_fuse_statfs(buf, &outarg.st);
 410         fuse_put_request(fc, req);
 411         return err;
 412 }
 413
 414 enum {
 415         OPT_FD,
 416         OPT_ROOTMODE,
 417         OPT_USER_ID,
 418         OPT_GROUP_ID,
 419         OPT_DEFAULT_PERMISSIONS,
 420         OPT_ALLOW_OTHER,
 421         OPT_MAX_READ,
 422         OPT_BLKSIZE,
 423         OPT_ERR
 424 };
 425
 426 static const match_table_t tokens = {
 427         {OPT_FD,                        "fd=%u"},
 428         {OPT_ROOTMODE,                  "rootmode=%o"},
 429         {OPT_USER_ID,                   "user_id=%u"},
 430         {OPT_GROUP_ID,                  "group_id=%u"},
 431         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 432         {OPT_ALLOW_OTHER,               "allow_other"},
 433         {OPT_MAX_READ,                  "max_read=%u"},
 434         {OPT_BLKSIZE,                   "blksize=%u"},
 435         {OPT_ERR,                       NULL}
 436 };
 437
 438 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
 439 {
 440         char *p;
 441         memset(d, 0, sizeof(struct fuse_mount_data));
 442         d->max_read = ~0;
 443         d->blksize = FUSE_DEFAULT_BLKSIZE;
 444
 445         while ((p = strsep(&opt, ",")) != NULL) {
 446                 int token;
 447                 int value;
 448                 substring_t args[MAX_OPT_ARGS];
 449                 if (!*p)
 450                         continue;
 451
 452                 token = match_token(p, tokens, args);
 453                 switch (token) {
 454                 case OPT_FD:
 455                         if (match_int(&args[0], &value))
 456                                 return 0;
 457                         d->fd = value;
 458                         d->fd_present = 1;
 459                         break;
 460
 461                 case OPT_ROOTMODE:
 462                         if (match_octal(&args[0], &value))
 463                                 return 0;
 464                         if (!fuse_valid_type(value))
 465                                 return 0;
 466                         d->rootmode = value;
 467                         d->rootmode_present = 1;
 468                         break;
 469
 470                 case OPT_USER_ID:
 471                         if (match_int(&args[0], &value))
 472                                 return 0;
 473                         d->user_id = value;
 474                         d->user_id_present = 1;
 475                         break;
 476
 477                 case OPT_GROUP_ID:
 478                         if (match_int(&args[0], &value))
 479                                 return 0;
 480                         d->group_id = value;
 481                         d->group_id_present = 1;
 482                         break;
 483
 484                 case OPT_DEFAULT_PERMISSIONS:
 485                         d->flags |= FUSE_DEFAULT_PERMISSIONS;
 486                         break;
 487
 488                 case OPT_ALLOW_OTHER:
 489                         d->flags |= FUSE_ALLOW_OTHER;
 490                         break;
 491
 492                 case OPT_MAX_READ:
 493                         if (match_int(&args[0], &value))
 494                                 return 0;
 495                         d->max_read = value;
 496                         break;
 497
 498                 case OPT_BLKSIZE:
 499                         if (!is_bdev || match_int(&args[0], &value))
 500                                 return 0;
 501                         d->blksize = value;
 502                         break;
 503
 504                 default:
 505                         return 0;
 506                 }
 507         }
 508
 509         if (!d->fd_present || !d->rootmode_present ||
 510             !d->user_id_present || !d->group_id_present)
 511                 return 0;
 512
 513         return 1;
 514 }
 515
 516 static int fuse_show_options(struct seq_file *m, struct vfsmount *mnt)
 517 {
 518         struct fuse_conn *fc = get_fuse_conn_super(mnt->mnt_sb);
 519
 520         seq_printf(m, ",user_id=%u", fc->user_id);
 521         seq_printf(m, ",group_id=%u", fc->group_id);
 522         if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
 523                 seq_puts(m, ",default_permissions");
 524         if (fc->flags & FUSE_ALLOW_OTHER)
 525                 seq_puts(m, ",allow_other");
 526         if (fc->max_read != ~0)
 527                 seq_printf(m, ",max_read=%u", fc->max_read);
 528         if (mnt->mnt_sb->s_bdev &&
 529             mnt->mnt_sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
 530                 seq_printf(m, ",blksize=%lu", mnt->mnt_sb->s_blocksize);
 531         return 0;
 532 }
 533
 534 void fuse_conn_init(struct fuse_conn *fc)
 535 {
 536         memset(fc, 0, sizeof(*fc));
 537         spin_lock_init(&fc->lock);
 538         mutex_init(&fc->inst_mutex);
 539         init_rwsem(&fc->killsb);
 540         atomic_set(&fc->count, 1);
 541         init_waitqueue_head(&fc->waitq);
 542         init_waitqueue_head(&fc->blocked_waitq);
 543         init_waitqueue_head(&fc->reserved_req_waitq);
 544         INIT_LIST_HEAD(&fc->pending);
 545         INIT_LIST_HEAD(&fc->processing);
 546         INIT_LIST_HEAD(&fc->io);
 547         INIT_LIST_HEAD(&fc->interrupts);
 548         INIT_LIST_HEAD(&fc->bg_queue);
 549         INIT_LIST_HEAD(&fc->entry);
 550         fc->forget_list_tail = &fc->forget_list_head;
 551         atomic_set(&fc->num_waiting, 0);
 552         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
 553         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
 554         fc->khctr = 0;
 555         fc->polled_files = RB_ROOT;
 556         fc->reqctr = 0;
 557         fc->blocked = 1;
 558         fc->attr_version = 1;
 559         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
 560 }
 561 EXPORT_SYMBOL_GPL(fuse_conn_init);
 562
 563 void fuse_conn_put(struct fuse_conn *fc)
 564 {
 565         if (atomic_dec_and_test(&fc->count)) {
 566                 if (fc->destroy_req)
 567                         fuse_request_free(fc->destroy_req);
 568                 mutex_destroy(&fc->inst_mutex);
 569                 fc->release(fc);
 570         }
 571 }
 572 EXPORT_SYMBOL_GPL(fuse_conn_put);
 573
 574 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
 575 {
 576         atomic_inc(&fc->count);
 577         return fc;
 578 }
 579 EXPORT_SYMBOL_GPL(fuse_conn_get);
 580
 581 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
 582 {
 583         struct fuse_attr attr;
 584         memset(&attr, 0, sizeof(attr));
 585
 586         attr.mode = mode;
 587         attr.ino = FUSE_ROOT_ID;
 588         attr.nlink = 1;
 589         return fuse_iget(sb, 1, 0, &attr, 0, 0);
 590 }
 591
 592 struct fuse_inode_handle {
 593         u64 nodeid;
 594         u32 generation;
 595 };
 596
 597 static struct dentry *fuse_get_dentry(struct super_block *sb,
 598                                       struct fuse_inode_handle *handle)
 599 {
 600         struct fuse_conn *fc = get_fuse_conn_super(sb);
 601         struct inode *inode;
 602         struct dentry *entry;
 603         int err = -ESTALE;
 604
 605         if (handle->nodeid == 0)
 606                 goto out_err;
 607
 608         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
 609         if (!inode) {
 610                 struct fuse_entry_out outarg;
 611                 struct qstr name;
 612
 613                 if (!fc->export_support)
 614                         goto out_err;
 615
 616                 name.len = 1;
 617                 name.name = ".";
 618                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
 619                                        &inode);
 620                 if (err && err != -ENOENT)
 621                         goto out_err;
 622                 if (err || !inode) {
 623                         err = -ESTALE;
 624                         goto out_err;
 625                 }
 626                 err = -EIO;
 627                 if (get_node_id(inode) != handle->nodeid)
 628                         goto out_iput;
 629         }
 630         err = -ESTALE;
 631         if (inode->i_generation != handle->generation)
 632                 goto out_iput;
 633
 634         entry = d_obtain_alias(inode);
 635         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
 636                 fuse_invalidate_entry_cache(entry);
 637
 638         return entry;
 639
 640  out_iput:
 641         iput(inode);
 642  out_err:
 643         return ERR_PTR(err);
 644 }
 645
 646 static int fuse_encode_fh(struct dentry *dentry, u32 *fh, int *max_len,
 647                            int connectable)
 648 {
 649         struct inode *inode = dentry->d_inode;
 650         bool encode_parent = connectable && !S_ISDIR(inode->i_mode);
 651         int len = encode_parent ? 6 : 3;
 652         u64 nodeid;
 653         u32 generation;
 654
 655         if (*max_len < len) {
 656                 *max_len = len;
 657                 return  255;
 658         }
 659
 660         nodeid = get_fuse_inode(inode)->nodeid;
 661         generation = inode->i_generation;
 662
 663         fh[0] = (u32)(nodeid >> 32);
 664         fh[1] = (u32)(nodeid & 0xffffffff);
 665         fh[2] = generation;
 666
 667         if (encode_parent) {
 668                 struct inode *parent;
 669
 670                 spin_lock(&dentry->d_lock);
 671                 parent = dentry->d_parent->d_inode;
 672                 nodeid = get_fuse_inode(parent)->nodeid;
 673                 generation = parent->i_generation;
 674                 spin_unlock(&dentry->d_lock);
 675
 676                 fh[3] = (u32)(nodeid >> 32);
 677                 fh[4] = (u32)(nodeid & 0xffffffff);
 678                 fh[5] = generation;
 679         }
 680
 681         *max_len = len;
 682         return encode_parent ? 0x82 : 0x81;
 683 }
 684
 685 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
 686                 struct fid *fid, int fh_len, int fh_type)
 687 {
 688         struct fuse_inode_handle handle;
 689
 690         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
 691                 return NULL;
 692
 693         handle.nodeid = (u64) fid->raw[0] << 32;
 694         handle.nodeid |= (u64) fid->raw[1];
 695         handle.generation = fid->raw[2];
 696         return fuse_get_dentry(sb, &handle);
 697 }
 698
 699 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
 700                 struct fid *fid, int fh_len, int fh_type)
 701 {
 702         struct fuse_inode_handle parent;
 703
 704         if (fh_type != 0x82 || fh_len < 6)
 705                 return NULL;
 706
 707         parent.nodeid = (u64) fid->raw[3] << 32;
 708         parent.nodeid |= (u64) fid->raw[4];
 709         parent.generation = fid->raw[5];
 710         return fuse_get_dentry(sb, &parent);
 711 }
 712
 713 static struct dentry *fuse_get_parent(struct dentry *child)
 714 {
 715         struct inode *child_inode = child->d_inode;
 716         struct fuse_conn *fc = get_fuse_conn(child_inode);
 717         struct inode *inode;
 718         struct dentry *parent;
 719         struct fuse_entry_out outarg;
 720         struct qstr name;
 721         int err;
 722
 723         if (!fc->export_support)
 724                 return ERR_PTR(-ESTALE);
 725
 726         name.len = 2;
 727         name.name = "..";
 728         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
 729                                &name, &outarg, &inode);
 730         if (err) {
 731                 if (err == -ENOENT)
 732                         return ERR_PTR(-ESTALE);
 733                 return ERR_PTR(err);
 734         }
 735
 736         parent = d_obtain_alias(inode);
 737         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
 738                 fuse_invalidate_entry_cache(parent);
 739
 740         return parent;
 741 }
 742
 743 static const struct export_operations fuse_export_operations = {
 744         .fh_to_dentry   = fuse_fh_to_dentry,
 745         .fh_to_parent   = fuse_fh_to_parent,
 746         .encode_fh      = fuse_encode_fh,
 747         .get_parent     = fuse_get_parent,
 748 };
 749
 750 static const struct super_operations fuse_super_operations = {
 751         .alloc_inode    = fuse_alloc_inode,
 752         .destroy_inode  = fuse_destroy_inode,
 753         .evict_inode    = fuse_evict_inode,
 754         .drop_inode     = generic_delete_inode,
 755         .remount_fs     = fuse_remount_fs,
 756         .put_super      = fuse_put_super,
 757         .umount_begin   = fuse_umount_begin,
 758         .statfs         = fuse_statfs,
 759         .show_options   = fuse_show_options,
 760 };
 761
 762 static void sanitize_global_limit(unsigned *limit)
 763 {
 764         if (*limit == 0)
 765                 *limit = ((num_physpages << PAGE_SHIFT) >> 13) /
 766                          sizeof(struct fuse_req);
 767
 768         if (*limit >= 1 << 16)
 769                 *limit = (1 << 16) - 1;
 770 }
 771
 772 static int set_global_limit(const char *val, struct kernel_param *kp)
 773 {
 774         int rv;
 775
 776         rv = param_set_uint(val, kp);
 777         if (rv)
 778                 return rv;
 779
 780         sanitize_global_limit((unsigned *)kp->arg);
 781
 782         return 0;
 783 }
 784
 785 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
 786 {
 787         int cap_sys_admin = capable(CAP_SYS_ADMIN);
 788
 789         if (arg->minor < 13)
 790                 return;
 791
 792         sanitize_global_limit(&max_user_bgreq);
 793         sanitize_global_limit(&max_user_congthresh);
 794
 795         if (arg->max_background) {
 796                 fc->max_background = arg->max_background;
 797
 798                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
 799                         fc->max_background = max_user_bgreq;
 800         }
 801         if (arg->congestion_threshold) {
 802                 fc->congestion_threshold = arg->congestion_threshold;
 803
 804                 if (!cap_sys_admin &&
 805                     fc->congestion_threshold > max_user_congthresh)
 806                         fc->congestion_threshold = max_user_congthresh;
 807         }
 808 }
 809
 810 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
 811 {
 812         struct fuse_init_out *arg = &req->misc.init_out;
 813
 814         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
 815                 fc->conn_error = 1;
 816         else {
 817                 unsigned long ra_pages;
 818
 819                 process_init_limits(fc, arg);
 820
 821                 if (arg->minor >= 6) {
 822                         ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
 823                         if (arg->flags & FUSE_ASYNC_READ)
 824                                 fc->async_read = 1;
 825                         if (!(arg->flags & FUSE_POSIX_LOCKS))
 826                                 fc->no_lock = 1;
 827                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
 828                                 fc->atomic_o_trunc = 1;
 829                         if (arg->minor >= 9) {
 830                                 /* LOOKUP has dependency on proto version */
 831                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
 832                                         fc->export_support = 1;
 833                         }
 834                         if (arg->flags & FUSE_BIG_WRITES)
 835                                 fc->big_writes = 1;
 836                         if (arg->flags & FUSE_DONT_MASK)
 837                                 fc->dont_mask = 1;
 838                 } else {
 839                         ra_pages = fc->max_read / PAGE_CACHE_SIZE;
 840                         fc->no_lock = 1;
 841                 }
 842
 843                 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
 844                 fc->minor = arg->minor;
 845                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
 846                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
 847                 fc->conn_init = 1;
 848         }
 849         fc->blocked = 0;
 850         wake_up_all(&fc->blocked_waitq);
 851 }
 852
 853 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
 854 {
 855         struct fuse_init_in *arg = &req->misc.init_in;
 856
 857         arg->major = FUSE_KERNEL_VERSION;
 858         arg->minor = FUSE_KERNEL_MINOR_VERSION;
 859         arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
 860         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
 861                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK;
 862         req->in.h.opcode = FUSE_INIT;
 863         req->in.numargs = 1;
 864         req->in.args[0].size = sizeof(*arg);
 865         req->in.args[0].value = arg;
 866         req->out.numargs = 1;
 867         /* Variable length argument used for backward compatibility
 868            with interface version < 7.5.  Rest of init_out is zeroed
 869            by do_get_request(), so a short reply is not a problem */
 870         req->out.argvar = 1;
 871         req->out.args[0].size = sizeof(struct fuse_init_out);
 872         req->out.args[0].value = &req->misc.init_out;
 873         req->end = process_init_reply;
 874         fuse_request_send_background(fc, req);
 875 }
 876
 877 static void fuse_free_conn(struct fuse_conn *fc)
 878 {
 879         kfree(fc);
 880 }
 881
 882 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
 883 {
 884         int err;
 885
 886         fc->bdi.name = "fuse";
 887         fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
 888         /* fuse does it's own writeback accounting */
 889         fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB;
 890
 891         err = bdi_init(&fc->bdi);
 892         if (err)
 893                 return err;
 894
 895         fc->bdi_initialized = 1;
 896
 897         if (sb->s_bdev) {
 898                 err =  bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
 899                                     MAJOR(fc->dev), MINOR(fc->dev));
 900         } else {
 901                 err = bdi_register_dev(&fc->bdi, fc->dev);
 902         }
 903
 904         if (err)
 905                 return err;
 906
 907         /*
 908          * For a single fuse filesystem use max 1% of dirty +
 909          * writeback threshold.
 910          *
 911          * This gives about 1M of write buffer for memory maps on a
 912          * machine with 1G and 10% dirty_ratio, which should be more
 913          * than enough.
 914          *
 915          * Privileged users can raise it by writing to
 916          *
 917          *    /sys/class/bdi/<bdi>/max_ratio
 918          */
 919         bdi_set_max_ratio(&fc->bdi, 1);
 920
 921         return 0;
 922 }
 923
 924 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
 925 {
 926         struct fuse_conn *fc;
 927         struct inode *root;
 928         struct fuse_mount_data d;
 929         struct file *file;
 930         struct dentry *root_dentry;
 931         struct fuse_req *init_req;
 932         int err;
 933         int is_bdev = sb->s_bdev != NULL;
 934
 935         err = -EINVAL;
 936         if (sb->s_flags & MS_MANDLOCK)
 937                 goto err;
 938
 939         sb->s_flags &= ~MS_NOSEC;
 940
 941         if (!parse_fuse_opt((char *) data, &d, is_bdev))
 942                 goto err;
 943
 944         if (is_bdev) {
 945 #ifdef CONFIG_BLOCK
 946                 err = -EINVAL;
 947                 if (!sb_set_blocksize(sb, d.blksize))
 948                         goto err;
 949 #endif
 950         } else {
 951                 sb->s_blocksize = PAGE_CACHE_SIZE;
 952                 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 953         }
 954         sb->s_magic = FUSE_SUPER_MAGIC;
 955         sb->s_op = &fuse_super_operations;
 956         sb->s_maxbytes = MAX_LFS_FILESIZE;
 957         sb->s_export_op = &fuse_export_operations;
 958
 959         file = fget(d.fd);
 960         err = -EINVAL;
 961         if (!file)
 962                 goto err;
 963
 964         if (file->f_op != &fuse_dev_operations)
 965                 goto err_fput;
 966
 967         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
 968         err = -ENOMEM;
 969         if (!fc)
 970                 goto err_fput;
 971
 972         fuse_conn_init(fc);
 973
 974         fc->dev = sb->s_dev;
 975         fc->sb = sb;
 976         err = fuse_bdi_init(fc, sb);
 977         if (err)
 978                 goto err_put_conn;
 979
 980         sb->s_bdi = &fc->bdi;
 981
 982         /* Handle umasking inside the fuse code */
 983         if (sb->s_flags & MS_POSIXACL)
 984                 fc->dont_mask = 1;
 985         sb->s_flags |= MS_POSIXACL;
 986
 987         fc->release = fuse_free_conn;
 988         fc->flags = d.flags;
 989         fc->user_id = d.user_id;
 990         fc->group_id = d.group_id;
 991         fc->max_read = max_t(unsigned, 4096, d.max_read);
 992
 993         /* Used by get_root_inode() */
 994         sb->s_fs_info = fc;
 995
 996         err = -ENOMEM;
 997         root = fuse_get_root_inode(sb, d.rootmode);
 998         if (!root)
 999                 goto err_put_conn;
1000
1001         root_dentry = d_alloc_root(root);
1002         if (!root_dentry) {
1003                 iput(root);
1004                 goto err_put_conn;
1005         }
1006         /* only now - we want root dentry with NULL ->d_op */
1007         sb->s_d_op = &fuse_dentry_operations;
1008
1009         init_req = fuse_request_alloc();
1010         if (!init_req)
1011                 goto err_put_root;
1012
1013         if (is_bdev) {
1014                 fc->destroy_req = fuse_request_alloc();
1015                 if (!fc->destroy_req)
1016                         goto err_free_init_req;
1017         }
1018
1019         mutex_lock(&fuse_mutex);
1020         err = -EINVAL;
1021         if (file->private_data)
1022                 goto err_unlock;
1023
1024         err = fuse_ctl_add_conn(fc);
1025         if (err)
1026                 goto err_unlock;
1027
1028         list_add_tail(&fc->entry, &fuse_conn_list);
1029         sb->s_root = root_dentry;
1030         fc->connected = 1;
1031         file->private_data = fuse_conn_get(fc);
1032         mutex_unlock(&fuse_mutex);
1033         /*
1034          * atomic_dec_and_test() in fput() provides the necessary
1035          * memory barrier for file->private_data to be visible on all
1036          * CPUs after this
1037          */
1038         fput(file);
1039
1040         fuse_send_init(fc, init_req);
1041
1042         return 0;
1043
1044  err_unlock:
1045         mutex_unlock(&fuse_mutex);
1046  err_free_init_req:
1047         fuse_request_free(init_req);
1048  err_put_root:
1049         dput(root_dentry);
1050  err_put_conn:
1051         fuse_bdi_destroy(fc);
1052         fuse_conn_put(fc);
1053  err_fput:
1054         fput(file);
1055  err:
1056         return err;
1057 }
1058
1059 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1060                        int flags, const char *dev_name,
1061                        void *raw_data)
1062 {
1063         return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1064 }
1065
1066 static void fuse_kill_sb_anon(struct super_block *sb)
1067 {
1068         struct fuse_conn *fc = get_fuse_conn_super(sb);
1069
1070         if (fc) {
1071                 down_write(&fc->killsb);
1072                 fc->sb = NULL;
1073                 up_write(&fc->killsb);
1074         }
1075
1076         kill_anon_super(sb);
1077 }
1078
1079 static struct file_system_type fuse_fs_type = {
1080         .owner          = THIS_MODULE,
1081         .name           = "fuse",
1082         .fs_flags       = FS_HAS_SUBTYPE,
1083         .mount          = fuse_mount,
1084         .kill_sb        = fuse_kill_sb_anon,
1085 };
1086
1087 #ifdef CONFIG_BLOCK
1088 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1089                            int flags, const char *dev_name,
1090                            void *raw_data)
1091 {
1092         return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1093 }
1094
1095 static void fuse_kill_sb_blk(struct super_block *sb)
1096 {
1097         struct fuse_conn *fc = get_fuse_conn_super(sb);
1098
1099         if (fc) {
1100                 down_write(&fc->killsb);
1101                 fc->sb = NULL;
1102                 up_write(&fc->killsb);
1103         }
1104
1105         kill_block_super(sb);
1106 }
1107
1108 static struct file_system_type fuseblk_fs_type = {
1109         .owner          = THIS_MODULE,
1110         .name           = "fuseblk",
1111         .mount          = fuse_mount_blk,
1112         .kill_sb        = fuse_kill_sb_blk,
1113         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1114 };
1115
1116 static inline int register_fuseblk(void)
1117 {
1118         return register_filesystem(&fuseblk_fs_type);
1119 }
1120
1121 static inline void unregister_fuseblk(void)
1122 {
1123         unregister_filesystem(&fuseblk_fs_type);
1124 }
1125 #else
1126 static inline int register_fuseblk(void)
1127 {
1128         return 0;
1129 }
1130
1131 static inline void unregister_fuseblk(void)
1132 {
1133 }
1134 #endif
1135
1136 static void fuse_inode_init_once(void *foo)
1137 {
1138         struct inode *inode = foo;
1139
1140         inode_init_once(inode);
1141 }
1142
1143 static int __init fuse_fs_init(void)
1144 {
1145         int err;
1146
1147         err = register_filesystem(&fuse_fs_type);
1148         if (err)
1149                 goto out;
1150
1151         err = register_fuseblk();
1152         if (err)
1153                 goto out_unreg;
1154
1155         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1156                                               sizeof(struct fuse_inode),
1157                                               0, SLAB_HWCACHE_ALIGN,
1158                                               fuse_inode_init_once);
1159         err = -ENOMEM;
1160         if (!fuse_inode_cachep)
1161                 goto out_unreg2;
1162
1163         return 0;
1164
1165  out_unreg2:
1166         unregister_fuseblk();
1167  out_unreg:
1168         unregister_filesystem(&fuse_fs_type);
1169  out:
1170         return err;
1171 }
1172
1173 static void fuse_fs_cleanup(void)
1174 {
1175         unregister_filesystem(&fuse_fs_type);
1176         unregister_fuseblk();
1177         kmem_cache_destroy(fuse_inode_cachep);
1178 }
1179
1180 static struct kobject *fuse_kobj;
1181 static struct kobject *connections_kobj;
1182
1183 static int fuse_sysfs_init(void)
1184 {
1185         int err;
1186
1187         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1188         if (!fuse_kobj) {
1189                 err = -ENOMEM;
1190                 goto out_err;
1191         }
1192
1193         connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1194         if (!connections_kobj) {
1195                 err = -ENOMEM;
1196                 goto out_fuse_unregister;
1197         }
1198
1199         return 0;
1200
1201  out_fuse_unregister:
1202         kobject_put(fuse_kobj);
1203  out_err:
1204         return err;
1205 }
1206
1207 static void fuse_sysfs_cleanup(void)
1208 {
1209         kobject_put(connections_kobj);
1210         kobject_put(fuse_kobj);
1211 }
1212
1213 static int __init fuse_init(void)
1214 {
1215         int res;
1216
1217         printk(KERN_INFO "fuse init (API version %i.%i)\n",
1218                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1219
1220         INIT_LIST_HEAD(&fuse_conn_list);
1221         res = fuse_fs_init();
1222         if (res)
1223                 goto err;
1224
1225         res = fuse_dev_init();
1226         if (res)
1227                 goto err_fs_cleanup;
1228
1229         res = fuse_sysfs_init();
1230         if (res)
1231                 goto err_dev_cleanup;
1232
1233         res = fuse_ctl_init();
1234         if (res)
1235                 goto err_sysfs_cleanup;
1236
1237         sanitize_global_limit(&max_user_bgreq);
1238         sanitize_global_limit(&max_user_congthresh);
1239
1240         return 0;
1241
1242  err_sysfs_cleanup:
1243         fuse_sysfs_cleanup();
1244  err_dev_cleanup:
1245         fuse_dev_cleanup();
1246  err_fs_cleanup:
1247         fuse_fs_cleanup();
1248  err:
1249         return res;
1250 }
1251
1252 static void __exit fuse_exit(void)
1253 {
1254         printk(KERN_DEBUG "fuse exit\n");
1255
1256         fuse_ctl_cleanup();
1257         fuse_sysfs_cleanup();
1258         fuse_fs_cleanup();
1259         fuse_dev_cleanup();
1260 }
1261
1262 module_init(fuse_init);
1263 module_exit(fuse_exit);