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         fi->orig_ino = 0;
  95         fi->state = 0;
  96         INIT_LIST_HEAD(&fi->write_files);
  97         INIT_LIST_HEAD(&fi->queued_writes);
  98         INIT_LIST_HEAD(&fi->writepages);
  99         init_waitqueue_head(&fi->page_waitq);
 100         fi->forget = fuse_alloc_forget();
 101         if (!fi->forget) {
 102                 kmem_cache_free(fuse_inode_cachep, inode);
 103                 return NULL;
 104         }
 105
 106         return inode;
 107 }
 108
 109 static void fuse_i_callback(struct rcu_head *head)
 110 {
 111         struct inode *inode = container_of(head, struct inode, i_rcu);
 112         INIT_LIST_HEAD(&inode->i_dentry);
 113         kmem_cache_free(fuse_inode_cachep, inode);
 114 }
 115
 116 static void fuse_destroy_inode(struct inode *inode)
 117 {
 118         struct fuse_inode *fi = get_fuse_inode(inode);
 119         BUG_ON(!list_empty(&fi->write_files));
 120         BUG_ON(!list_empty(&fi->queued_writes));
 121         kfree(fi->forget);
 122         call_rcu(&inode->i_rcu, fuse_i_callback);
 123 }
 124
 125 static void fuse_evict_inode(struct inode *inode)
 126 {
 127         truncate_inode_pages(&inode->i_data, 0);
 128         end_writeback(inode);
 129         if (inode->i_sb->s_flags & MS_ACTIVE) {
 130                 struct fuse_conn *fc = get_fuse_conn(inode);
 131                 struct fuse_inode *fi = get_fuse_inode(inode);
 132                 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
 133                 fi->forget = NULL;
 134         }
 135 }
 136
 137 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
 138 {
 139         if (*flags & MS_MANDLOCK)
 140                 return -EINVAL;
 141
 142         return 0;
 143 }
 144
 145 /*
 146  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
 147  * so that it will fit.
 148  */
 149 static ino_t fuse_squash_ino(u64 ino64)
 150 {
 151         ino_t ino = (ino_t) ino64;
 152         if (sizeof(ino_t) < sizeof(u64))
 153                 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
 154         return ino;
 155 }
 156
 157 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
 158                                    u64 attr_valid)
 159 {
 160         struct fuse_conn *fc = get_fuse_conn(inode);
 161         struct fuse_inode *fi = get_fuse_inode(inode);
 162
 163         fi->attr_version = ++fc->attr_version;
 164         fi->i_time = attr_valid;
 165
 166         inode->i_ino     = fuse_squash_ino(attr->ino);
 167         inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
 168         set_nlink(inode, attr->nlink);
 169         inode->i_uid     = attr->uid;
 170         inode->i_gid     = attr->gid;
 171         inode->i_blocks  = attr->blocks;
 172         inode->i_atime.tv_sec   = attr->atime;
 173         inode->i_atime.tv_nsec  = attr->atimensec;
 174         inode->i_mtime.tv_sec   = attr->mtime;
 175         inode->i_mtime.tv_nsec  = attr->mtimensec;
 176         inode->i_ctime.tv_sec   = attr->ctime;
 177         inode->i_ctime.tv_nsec  = attr->ctimensec;
 178
 179         if (attr->blksize != 0)
 180                 inode->i_blkbits = ilog2(attr->blksize);
 181         else
 182                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
 183
 184         /*
 185          * Don't set the sticky bit in i_mode, unless we want the VFS
 186          * to check permissions.  This prevents failures due to the
 187          * check in may_delete().
 188          */
 189         fi->orig_i_mode = inode->i_mode;
 190         if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
 191                 inode->i_mode &= ~S_ISVTX;
 192
 193         fi->orig_ino = attr->ino;
 194 }
 195
 196 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
 197                             u64 attr_valid, u64 attr_version)
 198 {
 199         struct fuse_conn *fc = get_fuse_conn(inode);
 200         struct fuse_inode *fi = get_fuse_inode(inode);
 201         loff_t oldsize;
 202
 203         spin_lock(&fc->lock);
 204         if ((attr_version != 0 && fi->attr_version > attr_version) ||
 205             test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
 206                 spin_unlock(&fc->lock);
 207                 return;
 208         }
 209
 210         fuse_change_attributes_common(inode, attr, attr_valid);
 211
 212         oldsize = inode->i_size;
 213         i_size_write(inode, attr->size);
 214         spin_unlock(&fc->lock);
 215
 216         if (S_ISREG(inode->i_mode) && oldsize != attr->size) {
 217                 truncate_pagecache(inode, oldsize, attr->size);
 218                 invalidate_inode_pages2(inode->i_mapping);
 219         }
 220 }
 221
 222 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
 223 {
 224         inode->i_mode = attr->mode & S_IFMT;
 225         inode->i_size = attr->size;
 226         if (S_ISREG(inode->i_mode)) {
 227                 fuse_init_common(inode);
 228                 fuse_init_file_inode(inode);
 229         } else if (S_ISDIR(inode->i_mode))
 230                 fuse_init_dir(inode);
 231         else if (S_ISLNK(inode->i_mode))
 232                 fuse_init_symlink(inode);
 233         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
 234                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
 235                 fuse_init_common(inode);
 236                 init_special_inode(inode, inode->i_mode,
 237                                    new_decode_dev(attr->rdev));
 238         } else
 239                 BUG();
 240 }
 241
 242 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
 243 {
 244         u64 nodeid = *(u64 *) _nodeidp;
 245         if (get_node_id(inode) == nodeid)
 246                 return 1;
 247         else
 248                 return 0;
 249 }
 250
 251 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
 252 {
 253         u64 nodeid = *(u64 *) _nodeidp;
 254         get_fuse_inode(inode)->nodeid = nodeid;
 255         return 0;
 256 }
 257
 258 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
 259                         int generation, struct fuse_attr *attr,
 260                         u64 attr_valid, u64 attr_version)
 261 {
 262         struct inode *inode;
 263         struct fuse_inode *fi;
 264         struct fuse_conn *fc = get_fuse_conn_super(sb);
 265
 266  retry:
 267         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
 268         if (!inode)
 269                 return NULL;
 270
 271         if ((inode->i_state & I_NEW)) {
 272                 inode->i_flags |= S_NOATIME|S_NOCMTIME;
 273                 inode->i_generation = generation;
 274                 inode->i_data.backing_dev_info = &fc->bdi;
 275                 fuse_init_inode(inode, attr);
 276                 unlock_new_inode(inode);
 277         } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
 278                 /* Inode has changed type, any I/O on the old should fail */
 279                 make_bad_inode(inode);
 280                 iput(inode);
 281                 goto retry;
 282         }
 283
 284         fi = get_fuse_inode(inode);
 285         spin_lock(&fc->lock);
 286         fi->nlookup++;
 287         spin_unlock(&fc->lock);
 288         fuse_change_attributes(inode, attr, attr_valid, attr_version);
 289
 290         return inode;
 291 }
 292
 293 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
 294                              loff_t offset, loff_t len)
 295 {
 296         struct inode *inode;
 297         pgoff_t pg_start;
 298         pgoff_t pg_end;
 299
 300         inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
 301         if (!inode)
 302                 return -ENOENT;
 303
 304         fuse_invalidate_attr(inode);
 305         if (offset >= 0) {
 306                 pg_start = offset >> PAGE_CACHE_SHIFT;
 307                 if (len <= 0)
 308                         pg_end = -1;
 309                 else
 310                         pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
 311                 invalidate_inode_pages2_range(inode->i_mapping,
 312                                               pg_start, pg_end);
 313         }
 314         iput(inode);
 315         return 0;
 316 }
 317
 318 static void fuse_umount_begin(struct super_block *sb)
 319 {
 320         fuse_abort_conn(get_fuse_conn_super(sb));
 321 }
 322
 323 static void fuse_send_destroy(struct fuse_conn *fc)
 324 {
 325         struct fuse_req *req = fc->destroy_req;
 326         if (req && fc->conn_init) {
 327                 fc->destroy_req = NULL;
 328                 req->in.h.opcode = FUSE_DESTROY;
 329                 req->force = 1;
 330                 fuse_request_send(fc, req);
 331                 fuse_put_request(fc, req);
 332         }
 333 }
 334
 335 static void fuse_bdi_destroy(struct fuse_conn *fc)
 336 {
 337         if (fc->bdi_initialized)
 338                 bdi_destroy(&fc->bdi);
 339 }
 340
 341 void fuse_conn_kill(struct fuse_conn *fc)
 342 {
 343         spin_lock(&fc->lock);
 344         fc->connected = 0;
 345         fc->blocked = 0;
 346         spin_unlock(&fc->lock);
 347         /* Flush all readers on this fs */
 348         kill_fasync(&fc->fasync, SIGIO, POLL_IN);
 349         wake_up_all(&fc->waitq);
 350         wake_up_all(&fc->blocked_waitq);
 351         wake_up_all(&fc->reserved_req_waitq);
 352         mutex_lock(&fuse_mutex);
 353         list_del(&fc->entry);
 354         fuse_ctl_remove_conn(fc);
 355         mutex_unlock(&fuse_mutex);
 356         fuse_bdi_destroy(fc);
 357 }
 358 EXPORT_SYMBOL_GPL(fuse_conn_kill);
 359
 360 static void fuse_put_super(struct super_block *sb)
 361 {
 362         struct fuse_conn *fc = get_fuse_conn_super(sb);
 363
 364         fuse_send_destroy(fc);
 365         fuse_conn_kill(fc);
 366         fuse_conn_put(fc);
 367 }
 368
 369 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
 370 {
 371         stbuf->f_type    = FUSE_SUPER_MAGIC;
 372         stbuf->f_bsize   = attr->bsize;
 373         stbuf->f_frsize  = attr->frsize;
 374         stbuf->f_blocks  = attr->blocks;
 375         stbuf->f_bfree   = attr->bfree;
 376         stbuf->f_bavail  = attr->bavail;
 377         stbuf->f_files   = attr->files;
 378         stbuf->f_ffree   = attr->ffree;
 379         stbuf->f_namelen = attr->namelen;
 380         /* fsid is left zero */
 381 }
 382
 383 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
 384 {
 385         struct super_block *sb = dentry->d_sb;
 386         struct fuse_conn *fc = get_fuse_conn_super(sb);
 387         struct fuse_req *req;
 388         struct fuse_statfs_out outarg;
 389         int err;
 390
 391         if (!fuse_allow_task(fc, current)) {
 392                 buf->f_type = FUSE_SUPER_MAGIC;
 393                 return 0;
 394         }
 395
 396         req = fuse_get_req(fc);
 397         if (IS_ERR(req))
 398                 return PTR_ERR(req);
 399
 400         memset(&outarg, 0, sizeof(outarg));
 401         req->in.numargs = 0;
 402         req->in.h.opcode = FUSE_STATFS;
 403         req->in.h.nodeid = get_node_id(dentry->d_inode);
 404         req->out.numargs = 1;
 405         req->out.args[0].size =
 406                 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
 407         req->out.args[0].value = &outarg;
 408         fuse_request_send(fc, req);
 409         err = req->out.h.error;
 410         if (!err)
 411                 convert_fuse_statfs(buf, &outarg.st);
 412         fuse_put_request(fc, req);
 413         return err;
 414 }
 415
 416 enum {
 417         OPT_FD,
 418         OPT_ROOTMODE,
 419         OPT_USER_ID,
 420         OPT_GROUP_ID,
 421         OPT_DEFAULT_PERMISSIONS,
 422         OPT_ALLOW_OTHER,
 423         OPT_MAX_READ,
 424         OPT_BLKSIZE,
 425         OPT_ERR
 426 };
 427
 428 static const match_table_t tokens = {
 429         {OPT_FD,                        "fd=%u"},
 430         {OPT_ROOTMODE,                  "rootmode=%o"},
 431         {OPT_USER_ID,                   "user_id=%u"},
 432         {OPT_GROUP_ID,                  "group_id=%u"},
 433         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 434         {OPT_ALLOW_OTHER,               "allow_other"},
 435         {OPT_MAX_READ,                  "max_read=%u"},
 436         {OPT_BLKSIZE,                   "blksize=%u"},
 437         {OPT_ERR,                       NULL}
 438 };
 439
 440 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
 441 {
 442         char *p;
 443         memset(d, 0, sizeof(struct fuse_mount_data));
 444         d->max_read = ~0;
 445         d->blksize = FUSE_DEFAULT_BLKSIZE;
 446
 447         while ((p = strsep(&opt, ",")) != NULL) {
 448                 int token;
 449                 int value;
 450                 substring_t args[MAX_OPT_ARGS];
 451                 if (!*p)
 452                         continue;
 453
 454                 token = match_token(p, tokens, args);
 455                 switch (token) {
 456                 case OPT_FD:
 457                         if (match_int(&args[0], &value))
 458                                 return 0;
 459                         d->fd = value;
 460                         d->fd_present = 1;
 461                         break;
 462
 463                 case OPT_ROOTMODE:
 464                         if (match_octal(&args[0], &value))
 465                                 return 0;
 466                         if (!fuse_valid_type(value))
 467                                 return 0;
 468                         d->rootmode = value;
 469                         d->rootmode_present = 1;
 470                         break;
 471
 472                 case OPT_USER_ID:
 473                         if (match_int(&args[0], &value))
 474                                 return 0;
 475                         d->user_id = value;
 476                         d->user_id_present = 1;
 477                         break;
 478
 479                 case OPT_GROUP_ID:
 480                         if (match_int(&args[0], &value))
 481                                 return 0;
 482                         d->group_id = value;
 483                         d->group_id_present = 1;
 484                         break;
 485
 486                 case OPT_DEFAULT_PERMISSIONS:
 487                         d->flags |= FUSE_DEFAULT_PERMISSIONS;
 488                         break;
 489
 490                 case OPT_ALLOW_OTHER:
 491                         d->flags |= FUSE_ALLOW_OTHER;
 492                         break;
 493
 494                 case OPT_MAX_READ:
 495                         if (match_int(&args[0], &value))
 496                                 return 0;
 497                         d->max_read = value;
 498                         break;
 499
 500                 case OPT_BLKSIZE:
 501                         if (!is_bdev || match_int(&args[0], &value))
 502                                 return 0;
 503                         d->blksize = value;
 504                         break;
 505
 506                 default:
 507                         return 0;
 508                 }
 509         }
 510
 511         if (!d->fd_present || !d->rootmode_present ||
 512             !d->user_id_present || !d->group_id_present)
 513                 return 0;
 514
 515         return 1;
 516 }
 517
 518 static int fuse_show_options(struct seq_file *m, struct vfsmount *mnt)
 519 {
 520         struct fuse_conn *fc = get_fuse_conn_super(mnt->mnt_sb);
 521
 522         seq_printf(m, ",user_id=%u", fc->user_id);
 523         seq_printf(m, ",group_id=%u", fc->group_id);
 524         if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
 525                 seq_puts(m, ",default_permissions");
 526         if (fc->flags & FUSE_ALLOW_OTHER)
 527                 seq_puts(m, ",allow_other");
 528         if (fc->max_read != ~0)
 529                 seq_printf(m, ",max_read=%u", fc->max_read);
 530         if (mnt->mnt_sb->s_bdev &&
 531             mnt->mnt_sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
 532                 seq_printf(m, ",blksize=%lu", mnt->mnt_sb->s_blocksize);
 533         return 0;
 534 }
 535
 536 void fuse_conn_init(struct fuse_conn *fc)
 537 {
 538         memset(fc, 0, sizeof(*fc));
 539         spin_lock_init(&fc->lock);
 540         mutex_init(&fc->inst_mutex);
 541         init_rwsem(&fc->killsb);
 542         atomic_set(&fc->count, 1);
 543         init_waitqueue_head(&fc->waitq);
 544         init_waitqueue_head(&fc->blocked_waitq);
 545         init_waitqueue_head(&fc->reserved_req_waitq);
 546         INIT_LIST_HEAD(&fc->pending);
 547         INIT_LIST_HEAD(&fc->processing);
 548         INIT_LIST_HEAD(&fc->io);
 549         INIT_LIST_HEAD(&fc->interrupts);
 550         INIT_LIST_HEAD(&fc->bg_queue);
 551         INIT_LIST_HEAD(&fc->entry);
 552         fc->forget_list_tail = &fc->forget_list_head;
 553         atomic_set(&fc->num_waiting, 0);
 554         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
 555         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
 556         fc->khctr = 0;
 557         fc->polled_files = RB_ROOT;
 558         fc->reqctr = 0;
 559         fc->blocked = 1;
 560         fc->attr_version = 1;
 561         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
 562 }
 563 EXPORT_SYMBOL_GPL(fuse_conn_init);
 564
 565 void fuse_conn_put(struct fuse_conn *fc)
 566 {
 567         if (atomic_dec_and_test(&fc->count)) {
 568                 if (fc->destroy_req)
 569                         fuse_request_free(fc->destroy_req);
 570                 mutex_destroy(&fc->inst_mutex);
 571                 fc->release(fc);
 572         }
 573 }
 574 EXPORT_SYMBOL_GPL(fuse_conn_put);
 575
 576 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
 577 {
 578         atomic_inc(&fc->count);
 579         return fc;
 580 }
 581 EXPORT_SYMBOL_GPL(fuse_conn_get);
 582
 583 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
 584 {
 585         struct fuse_attr attr;
 586         memset(&attr, 0, sizeof(attr));
 587
 588         attr.mode = mode;
 589         attr.ino = FUSE_ROOT_ID;
 590         attr.nlink = 1;
 591         return fuse_iget(sb, 1, 0, &attr, 0, 0);
 592 }
 593
 594 struct fuse_inode_handle {
 595         u64 nodeid;
 596         u32 generation;
 597 };
 598
 599 static struct dentry *fuse_get_dentry(struct super_block *sb,
 600                                       struct fuse_inode_handle *handle)
 601 {
 602         struct fuse_conn *fc = get_fuse_conn_super(sb);
 603         struct inode *inode;
 604         struct dentry *entry;
 605         int err = -ESTALE;
 606
 607         if (handle->nodeid == 0)
 608                 goto out_err;
 609
 610         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
 611         if (!inode) {
 612                 struct fuse_entry_out outarg;
 613                 struct qstr name;
 614
 615                 if (!fc->export_support)
 616                         goto out_err;
 617
 618                 name.len = 1;
 619                 name.name = ".";
 620                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
 621                                        &inode);
 622                 if (err && err != -ENOENT)
 623                         goto out_err;
 624                 if (err || !inode) {
 625                         err = -ESTALE;
 626                         goto out_err;
 627                 }
 628                 err = -EIO;
 629                 if (get_node_id(inode) != handle->nodeid)
 630                         goto out_iput;
 631         }
 632         err = -ESTALE;
 633         if (inode->i_generation != handle->generation)
 634                 goto out_iput;
 635
 636         entry = d_obtain_alias(inode);
 637         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
 638                 fuse_invalidate_entry_cache(entry);
 639
 640         return entry;
 641
 642  out_iput:
 643         iput(inode);
 644  out_err:
 645         return ERR_PTR(err);
 646 }
 647
 648 static int fuse_encode_fh(struct dentry *dentry, u32 *fh, int *max_len,
 649                            int connectable)
 650 {
 651         struct inode *inode = dentry->d_inode;
 652         bool encode_parent = connectable && !S_ISDIR(inode->i_mode);
 653         int len = encode_parent ? 6 : 3;
 654         u64 nodeid;
 655         u32 generation;
 656
 657         if (*max_len < len) {
 658                 *max_len = len;
 659                 return  255;
 660         }
 661
 662         nodeid = get_fuse_inode(inode)->nodeid;
 663         generation = inode->i_generation;
 664
 665         fh[0] = (u32)(nodeid >> 32);
 666         fh[1] = (u32)(nodeid & 0xffffffff);
 667         fh[2] = generation;
 668
 669         if (encode_parent) {
 670                 struct inode *parent;
 671
 672                 spin_lock(&dentry->d_lock);
 673                 parent = dentry->d_parent->d_inode;
 674                 nodeid = get_fuse_inode(parent)->nodeid;
 675                 generation = parent->i_generation;
 676                 spin_unlock(&dentry->d_lock);
 677
 678                 fh[3] = (u32)(nodeid >> 32);
 679                 fh[4] = (u32)(nodeid & 0xffffffff);
 680                 fh[5] = generation;
 681         }
 682
 683         *max_len = len;
 684         return encode_parent ? 0x82 : 0x81;
 685 }
 686
 687 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
 688                 struct fid *fid, int fh_len, int fh_type)
 689 {
 690         struct fuse_inode_handle handle;
 691
 692         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
 693                 return NULL;
 694
 695         handle.nodeid = (u64) fid->raw[0] << 32;
 696         handle.nodeid |= (u64) fid->raw[1];
 697         handle.generation = fid->raw[2];
 698         return fuse_get_dentry(sb, &handle);
 699 }
 700
 701 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
 702                 struct fid *fid, int fh_len, int fh_type)
 703 {
 704         struct fuse_inode_handle parent;
 705
 706         if (fh_type != 0x82 || fh_len < 6)
 707                 return NULL;
 708
 709         parent.nodeid = (u64) fid->raw[3] << 32;
 710         parent.nodeid |= (u64) fid->raw[4];
 711         parent.generation = fid->raw[5];
 712         return fuse_get_dentry(sb, &parent);
 713 }
 714
 715 static struct dentry *fuse_get_parent(struct dentry *child)
 716 {
 717         struct inode *child_inode = child->d_inode;
 718         struct fuse_conn *fc = get_fuse_conn(child_inode);
 719         struct inode *inode;
 720         struct dentry *parent;
 721         struct fuse_entry_out outarg;
 722         struct qstr name;
 723         int err;
 724
 725         if (!fc->export_support)
 726                 return ERR_PTR(-ESTALE);
 727
 728         name.len = 2;
 729         name.name = "..";
 730         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
 731                                &name, &outarg, &inode);
 732         if (err) {
 733                 if (err == -ENOENT)
 734                         return ERR_PTR(-ESTALE);
 735                 return ERR_PTR(err);
 736         }
 737
 738         parent = d_obtain_alias(inode);
 739         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
 740                 fuse_invalidate_entry_cache(parent);
 741
 742         return parent;
 743 }
 744
 745 static const struct export_operations fuse_export_operations = {
 746         .fh_to_dentry   = fuse_fh_to_dentry,
 747         .fh_to_parent   = fuse_fh_to_parent,
 748         .encode_fh      = fuse_encode_fh,
 749         .get_parent     = fuse_get_parent,
 750 };
 751
 752 static const struct super_operations fuse_super_operations = {
 753         .alloc_inode    = fuse_alloc_inode,
 754         .destroy_inode  = fuse_destroy_inode,
 755         .evict_inode    = fuse_evict_inode,
 756         .drop_inode     = generic_delete_inode,
 757         .remount_fs     = fuse_remount_fs,
 758         .put_super      = fuse_put_super,
 759         .umount_begin   = fuse_umount_begin,
 760         .statfs         = fuse_statfs,
 761         .show_options   = fuse_show_options,
 762 };
 763
 764 static void sanitize_global_limit(unsigned *limit)
 765 {
 766         if (*limit == 0)
 767                 *limit = ((num_physpages << PAGE_SHIFT) >> 13) /
 768                          sizeof(struct fuse_req);
 769
 770         if (*limit >= 1 << 16)
 771                 *limit = (1 << 16) - 1;
 772 }
 773
 774 static int set_global_limit(const char *val, struct kernel_param *kp)
 775 {
 776         int rv;
 777
 778         rv = param_set_uint(val, kp);
 779         if (rv)
 780                 return rv;
 781
 782         sanitize_global_limit((unsigned *)kp->arg);
 783
 784         return 0;
 785 }
 786
 787 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
 788 {
 789         int cap_sys_admin = capable(CAP_SYS_ADMIN);
 790
 791         if (arg->minor < 13)
 792                 return;
 793
 794         sanitize_global_limit(&max_user_bgreq);
 795         sanitize_global_limit(&max_user_congthresh);
 796
 797         if (arg->max_background) {
 798                 fc->max_background = arg->max_background;
 799
 800                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
 801                         fc->max_background = max_user_bgreq;
 802         }
 803         if (arg->congestion_threshold) {
 804                 fc->congestion_threshold = arg->congestion_threshold;
 805
 806                 if (!cap_sys_admin &&
 807                     fc->congestion_threshold > max_user_congthresh)
 808                         fc->congestion_threshold = max_user_congthresh;
 809         }
 810 }
 811
 812 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
 813 {
 814         struct fuse_init_out *arg = &req->misc.init_out;
 815
 816         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
 817                 fc->conn_error = 1;
 818         else {
 819                 unsigned long ra_pages;
 820
 821                 process_init_limits(fc, arg);
 822
 823                 if (arg->minor >= 6) {
 824                         ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
 825                         if (arg->flags & FUSE_ASYNC_READ)
 826                                 fc->async_read = 1;
 827                         if (!(arg->flags & FUSE_POSIX_LOCKS))
 828                                 fc->no_lock = 1;
 829                         if (arg->minor >= 17) {
 830                                 if (!(arg->flags & FUSE_FLOCK_LOCKS))
 831                                         fc->no_flock = 1;
 832                         } else {
 833                                 if (!(arg->flags & FUSE_POSIX_LOCKS))
 834                                         fc->no_flock = 1;
 835                         }
 836                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
 837                                 fc->atomic_o_trunc = 1;
 838                         if (arg->minor >= 9) {
 839                                 /* LOOKUP has dependency on proto version */
 840                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
 841                                         fc->export_support = 1;
 842                         }
 843                         if (arg->flags & FUSE_BIG_WRITES)
 844                                 fc->big_writes = 1;
 845                         if (arg->flags & FUSE_DONT_MASK)
 846                                 fc->dont_mask = 1;
 847                 } else {
 848                         ra_pages = fc->max_read / PAGE_CACHE_SIZE;
 849                         fc->no_lock = 1;
 850                         fc->no_flock = 1;
 851                 }
 852
 853                 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
 854                 fc->minor = arg->minor;
 855                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
 856                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
 857                 fc->conn_init = 1;
 858         }
 859         fc->blocked = 0;
 860         wake_up_all(&fc->blocked_waitq);
 861 }
 862
 863 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
 864 {
 865         struct fuse_init_in *arg = &req->misc.init_in;
 866
 867         arg->major = FUSE_KERNEL_VERSION;
 868         arg->minor = FUSE_KERNEL_MINOR_VERSION;
 869         arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
 870         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
 871                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
 872                 FUSE_FLOCK_LOCKS;
 873         req->in.h.opcode = FUSE_INIT;
 874         req->in.numargs = 1;
 875         req->in.args[0].size = sizeof(*arg);
 876         req->in.args[0].value = arg;
 877         req->out.numargs = 1;
 878         /* Variable length argument used for backward compatibility
 879            with interface version < 7.5.  Rest of init_out is zeroed
 880            by do_get_request(), so a short reply is not a problem */
 881         req->out.argvar = 1;
 882         req->out.args[0].size = sizeof(struct fuse_init_out);
 883         req->out.args[0].value = &req->misc.init_out;
 884         req->end = process_init_reply;
 885         fuse_request_send_background(fc, req);
 886 }
 887
 888 static void fuse_free_conn(struct fuse_conn *fc)
 889 {
 890         kfree(fc);
 891 }
 892
 893 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
 894 {
 895         int err;
 896
 897         fc->bdi.name = "fuse";
 898         fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
 899         /* fuse does it's own writeback accounting */
 900         fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB;
 901
 902         err = bdi_init(&fc->bdi);
 903         if (err)
 904                 return err;
 905
 906         fc->bdi_initialized = 1;
 907
 908         if (sb->s_bdev) {
 909                 err =  bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
 910                                     MAJOR(fc->dev), MINOR(fc->dev));
 911         } else {
 912                 err = bdi_register_dev(&fc->bdi, fc->dev);
 913         }
 914
 915         if (err)
 916                 return err;
 917
 918         /*
 919          * For a single fuse filesystem use max 1% of dirty +
 920          * writeback threshold.
 921          *
 922          * This gives about 1M of write buffer for memory maps on a
 923          * machine with 1G and 10% dirty_ratio, which should be more
 924          * than enough.
 925          *
 926          * Privileged users can raise it by writing to
 927          *
 928          *    /sys/class/bdi/<bdi>/max_ratio
 929          */
 930         bdi_set_max_ratio(&fc->bdi, 1);
 931
 932         return 0;
 933 }
 934
 935 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
 936 {
 937         struct fuse_conn *fc;
 938         struct inode *root;
 939         struct fuse_mount_data d;
 940         struct file *file;
 941         struct dentry *root_dentry;
 942         struct fuse_req *init_req;
 943         int err;
 944         int is_bdev = sb->s_bdev != NULL;
 945
 946         err = -EINVAL;
 947         if (sb->s_flags & MS_MANDLOCK)
 948                 goto err;
 949
 950         sb->s_flags &= ~MS_NOSEC;
 951
 952         if (!parse_fuse_opt((char *) data, &d, is_bdev))
 953                 goto err;
 954
 955         if (is_bdev) {
 956 #ifdef CONFIG_BLOCK
 957                 err = -EINVAL;
 958                 if (!sb_set_blocksize(sb, d.blksize))
 959                         goto err;
 960 #endif
 961         } else {
 962                 sb->s_blocksize = PAGE_CACHE_SIZE;
 963                 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 964         }
 965         sb->s_magic = FUSE_SUPER_MAGIC;
 966         sb->s_op = &fuse_super_operations;
 967         sb->s_maxbytes = MAX_LFS_FILESIZE;
 968         sb->s_export_op = &fuse_export_operations;
 969
 970         file = fget(d.fd);
 971         err = -EINVAL;
 972         if (!file)
 973                 goto err;
 974
 975         if (file->f_op != &fuse_dev_operations)
 976                 goto err_fput;
 977
 978         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
 979         err = -ENOMEM;
 980         if (!fc)
 981                 goto err_fput;
 982
 983         fuse_conn_init(fc);
 984
 985         fc->dev = sb->s_dev;
 986         fc->sb = sb;
 987         err = fuse_bdi_init(fc, sb);
 988         if (err)
 989                 goto err_put_conn;
 990
 991         sb->s_bdi = &fc->bdi;
 992
 993         /* Handle umasking inside the fuse code */
 994         if (sb->s_flags & MS_POSIXACL)
 995                 fc->dont_mask = 1;
 996         sb->s_flags |= MS_POSIXACL;
 997
 998         fc->release = fuse_free_conn;
 999         fc->flags = d.flags;
1000         fc->user_id = d.user_id;
1001         fc->group_id = d.group_id;
1002         fc->max_read = max_t(unsigned, 4096, d.max_read);
1003
1004         /* Used by get_root_inode() */
1005         sb->s_fs_info = fc;
1006
1007         err = -ENOMEM;
1008         root = fuse_get_root_inode(sb, d.rootmode);
1009         if (!root)
1010                 goto err_put_conn;
1011
1012         root_dentry = d_alloc_root(root);
1013         if (!root_dentry) {
1014                 iput(root);
1015                 goto err_put_conn;
1016         }
1017         /* only now - we want root dentry with NULL ->d_op */
1018         sb->s_d_op = &fuse_dentry_operations;
1019
1020         init_req = fuse_request_alloc();
1021         if (!init_req)
1022                 goto err_put_root;
1023
1024         if (is_bdev) {
1025                 fc->destroy_req = fuse_request_alloc();
1026                 if (!fc->destroy_req)
1027                         goto err_free_init_req;
1028         }
1029
1030         mutex_lock(&fuse_mutex);
1031         err = -EINVAL;
1032         if (file->private_data)
1033                 goto err_unlock;
1034
1035         err = fuse_ctl_add_conn(fc);
1036         if (err)
1037                 goto err_unlock;
1038
1039         list_add_tail(&fc->entry, &fuse_conn_list);
1040         sb->s_root = root_dentry;
1041         fc->connected = 1;
1042         file->private_data = fuse_conn_get(fc);
1043         mutex_unlock(&fuse_mutex);
1044         /*
1045          * atomic_dec_and_test() in fput() provides the necessary
1046          * memory barrier for file->private_data to be visible on all
1047          * CPUs after this
1048          */
1049         fput(file);
1050
1051         fuse_send_init(fc, init_req);
1052
1053         return 0;
1054
1055  err_unlock:
1056         mutex_unlock(&fuse_mutex);
1057  err_free_init_req:
1058         fuse_request_free(init_req);
1059  err_put_root:
1060         dput(root_dentry);
1061  err_put_conn:
1062         fuse_bdi_destroy(fc);
1063         fuse_conn_put(fc);
1064  err_fput:
1065         fput(file);
1066  err:
1067         return err;
1068 }
1069
1070 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1071                        int flags, const char *dev_name,
1072                        void *raw_data)
1073 {
1074         return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1075 }
1076
1077 static void fuse_kill_sb_anon(struct super_block *sb)
1078 {
1079         struct fuse_conn *fc = get_fuse_conn_super(sb);
1080
1081         if (fc) {
1082                 down_write(&fc->killsb);
1083                 fc->sb = NULL;
1084                 up_write(&fc->killsb);
1085         }
1086
1087         kill_anon_super(sb);
1088 }
1089
1090 static struct file_system_type fuse_fs_type = {
1091         .owner          = THIS_MODULE,
1092         .name           = "fuse",
1093         .fs_flags       = FS_HAS_SUBTYPE,
1094         .mount          = fuse_mount,
1095         .kill_sb        = fuse_kill_sb_anon,
1096 };
1097
1098 #ifdef CONFIG_BLOCK
1099 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1100                            int flags, const char *dev_name,
1101                            void *raw_data)
1102 {
1103         return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1104 }
1105
1106 static void fuse_kill_sb_blk(struct super_block *sb)
1107 {
1108         struct fuse_conn *fc = get_fuse_conn_super(sb);
1109
1110         if (fc) {
1111                 down_write(&fc->killsb);
1112                 fc->sb = NULL;
1113                 up_write(&fc->killsb);
1114         }
1115
1116         kill_block_super(sb);
1117 }
1118
1119 static struct file_system_type fuseblk_fs_type = {
1120         .owner          = THIS_MODULE,
1121         .name           = "fuseblk",
1122         .mount          = fuse_mount_blk,
1123         .kill_sb        = fuse_kill_sb_blk,
1124         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1125 };
1126
1127 static inline int register_fuseblk(void)
1128 {
1129         return register_filesystem(&fuseblk_fs_type);
1130 }
1131
1132 static inline void unregister_fuseblk(void)
1133 {
1134         unregister_filesystem(&fuseblk_fs_type);
1135 }
1136 #else
1137 static inline int register_fuseblk(void)
1138 {
1139         return 0;
1140 }
1141
1142 static inline void unregister_fuseblk(void)
1143 {
1144 }
1145 #endif
1146
1147 static void fuse_inode_init_once(void *foo)
1148 {
1149         struct inode *inode = foo;
1150
1151         inode_init_once(inode);
1152 }
1153
1154 static int __init fuse_fs_init(void)
1155 {
1156         int err;
1157
1158         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1159                                               sizeof(struct fuse_inode),
1160                                               0, SLAB_HWCACHE_ALIGN,
1161                                               fuse_inode_init_once);
1162         err = -ENOMEM;
1163         if (!fuse_inode_cachep)
1164                 goto out;
1165
1166         err = register_fuseblk();
1167         if (err)
1168                 goto out2;
1169
1170         err = register_filesystem(&fuse_fs_type);
1171         if (err)
1172                 goto out3;
1173
1174         return 0;
1175
1176  out3:
1177         unregister_fuseblk();
1178  out2:
1179         kmem_cache_destroy(fuse_inode_cachep);
1180  out:
1181         return err;
1182 }
1183
1184 static void fuse_fs_cleanup(void)
1185 {
1186         unregister_filesystem(&fuse_fs_type);
1187         unregister_fuseblk();
1188         kmem_cache_destroy(fuse_inode_cachep);
1189 }
1190
1191 static struct kobject *fuse_kobj;
1192 static struct kobject *connections_kobj;
1193
1194 static int fuse_sysfs_init(void)
1195 {
1196         int err;
1197
1198         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1199         if (!fuse_kobj) {
1200                 err = -ENOMEM;
1201                 goto out_err;
1202         }
1203
1204         connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1205         if (!connections_kobj) {
1206                 err = -ENOMEM;
1207                 goto out_fuse_unregister;
1208         }
1209
1210         return 0;
1211
1212  out_fuse_unregister:
1213         kobject_put(fuse_kobj);
1214  out_err:
1215         return err;
1216 }
1217
1218 static void fuse_sysfs_cleanup(void)
1219 {
1220         kobject_put(connections_kobj);
1221         kobject_put(fuse_kobj);
1222 }
1223
1224 static int __init fuse_init(void)
1225 {
1226         int res;
1227
1228         printk(KERN_INFO "fuse init (API version %i.%i)\n",
1229                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1230
1231         INIT_LIST_HEAD(&fuse_conn_list);
1232         res = fuse_fs_init();
1233         if (res)
1234                 goto err;
1235
1236         res = fuse_dev_init();
1237         if (res)
1238                 goto err_fs_cleanup;
1239
1240         res = fuse_sysfs_init();
1241         if (res)
1242                 goto err_dev_cleanup;
1243
1244         res = fuse_ctl_init();
1245         if (res)
1246                 goto err_sysfs_cleanup;
1247
1248         sanitize_global_limit(&max_user_bgreq);
1249         sanitize_global_limit(&max_user_congthresh);
1250
1251         return 0;
1252
1253  err_sysfs_cleanup:
1254         fuse_sysfs_cleanup();
1255  err_dev_cleanup:
1256         fuse_dev_cleanup();
1257  err_fs_cleanup:
1258         fuse_fs_cleanup();
1259  err:
1260         return res;
1261 }
1262
1263 static void __exit fuse_exit(void)
1264 {
1265         printk(KERN_DEBUG "fuse exit\n");
1266
1267         fuse_ctl_cleanup();
1268         fuse_sysfs_cleanup();
1269         fuse_fs_cleanup();
1270         fuse_dev_cleanup();
1271 }
1272
1273 module_init(fuse_init);
1274 module_exit(fuse_exit);