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         kuid_t user_id;
  64         kgid_t 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         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         clear_inode(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         sync_filesystem(sb);
 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     = make_kuid(&init_user_ns, attr->uid);
 170         inode->i_gid     = make_kgid(&init_user_ns, 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         struct timespec old_mtime;
 203
 204         spin_lock(&fc->lock);
 205         if ((attr_version != 0 && fi->attr_version > attr_version) ||
 206             test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
 207                 spin_unlock(&fc->lock);
 208                 return;
 209         }
 210
 211         old_mtime = inode->i_mtime;
 212         fuse_change_attributes_common(inode, attr, attr_valid);
 213
 214         oldsize = inode->i_size;
 215         i_size_write(inode, attr->size);
 216         spin_unlock(&fc->lock);
 217
 218         if (S_ISREG(inode->i_mode)) {
 219                 bool inval = false;
 220
 221                 if (oldsize != attr->size) {
 222                         truncate_pagecache(inode, oldsize, attr->size);
 223                         inval = true;
 224                 } else if (fc->auto_inval_data) {
 225                         struct timespec new_mtime = {
 226                                 .tv_sec = attr->mtime,
 227                                 .tv_nsec = attr->mtimensec,
 228                         };
 229
 230                         /*
 231                          * Auto inval mode also checks and invalidates if mtime
 232                          * has changed.
 233                          */
 234                         if (!timespec_equal(&old_mtime, &new_mtime))
 235                                 inval = true;
 236                 }
 237
 238                 if (inval)
 239                         invalidate_inode_pages2(inode->i_mapping);
 240         }
 241 }
 242
 243 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
 244 {
 245         inode->i_mode = attr->mode & S_IFMT;
 246         inode->i_size = attr->size;
 247         if (S_ISREG(inode->i_mode)) {
 248                 fuse_init_common(inode);
 249                 fuse_init_file_inode(inode);
 250         } else if (S_ISDIR(inode->i_mode))
 251                 fuse_init_dir(inode);
 252         else if (S_ISLNK(inode->i_mode))
 253                 fuse_init_symlink(inode);
 254         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
 255                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
 256                 fuse_init_common(inode);
 257                 init_special_inode(inode, inode->i_mode,
 258                                    new_decode_dev(attr->rdev));
 259         } else
 260                 BUG();
 261 }
 262
 263 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
 264 {
 265         u64 nodeid = *(u64 *) _nodeidp;
 266         if (get_node_id(inode) == nodeid)
 267                 return 1;
 268         else
 269                 return 0;
 270 }
 271
 272 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
 273 {
 274         u64 nodeid = *(u64 *) _nodeidp;
 275         get_fuse_inode(inode)->nodeid = nodeid;
 276         return 0;
 277 }
 278
 279 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
 280                         int generation, struct fuse_attr *attr,
 281                         u64 attr_valid, u64 attr_version)
 282 {
 283         struct inode *inode;
 284         struct fuse_inode *fi;
 285         struct fuse_conn *fc = get_fuse_conn_super(sb);
 286
 287  retry:
 288         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
 289         if (!inode)
 290                 return NULL;
 291
 292         if ((inode->i_state & I_NEW)) {
 293                 inode->i_flags |= S_NOATIME|S_NOCMTIME;
 294                 inode->i_generation = generation;
 295                 inode->i_data.backing_dev_info = &fc->bdi;
 296                 fuse_init_inode(inode, attr);
 297                 unlock_new_inode(inode);
 298         } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
 299                 /* Inode has changed type, any I/O on the old should fail */
 300                 make_bad_inode(inode);
 301                 iput(inode);
 302                 goto retry;
 303         }
 304
 305         fi = get_fuse_inode(inode);
 306         spin_lock(&fc->lock);
 307         fi->nlookup++;
 308         spin_unlock(&fc->lock);
 309         fuse_change_attributes(inode, attr, attr_valid, attr_version);
 310
 311         return inode;
 312 }
 313
 314 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
 315                              loff_t offset, loff_t len)
 316 {
 317         struct inode *inode;
 318         pgoff_t pg_start;
 319         pgoff_t pg_end;
 320
 321         inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
 322         if (!inode)
 323                 return -ENOENT;
 324
 325         fuse_invalidate_attr(inode);
 326         if (offset >= 0) {
 327                 pg_start = offset >> PAGE_CACHE_SHIFT;
 328                 if (len <= 0)
 329                         pg_end = -1;
 330                 else
 331                         pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
 332                 invalidate_inode_pages2_range(inode->i_mapping,
 333                                               pg_start, pg_end);
 334         }
 335         iput(inode);
 336         return 0;
 337 }
 338
 339 static void fuse_umount_begin(struct super_block *sb)
 340 {
 341         fuse_abort_conn(get_fuse_conn_super(sb));
 342 }
 343
 344 static void fuse_send_destroy(struct fuse_conn *fc)
 345 {
 346         struct fuse_req *req = fc->destroy_req;
 347         if (req && fc->conn_init) {
 348                 fc->destroy_req = NULL;
 349                 req->in.h.opcode = FUSE_DESTROY;
 350                 req->force = 1;
 351                 req->background = 0;
 352                 fuse_request_send(fc, req);
 353                 fuse_put_request(fc, req);
 354         }
 355 }
 356
 357 static void fuse_bdi_destroy(struct fuse_conn *fc)
 358 {
 359         if (fc->bdi_initialized)
 360                 bdi_destroy(&fc->bdi);
 361 }
 362
 363 void fuse_conn_kill(struct fuse_conn *fc)
 364 {
 365         spin_lock(&fc->lock);
 366         fc->connected = 0;
 367         fc->blocked = 0;
 368         fc->initialized = 1;
 369         spin_unlock(&fc->lock);
 370         /* Flush all readers on this fs */
 371         kill_fasync(&fc->fasync, SIGIO, POLL_IN);
 372         wake_up_all(&fc->waitq);
 373         wake_up_all(&fc->blocked_waitq);
 374         wake_up_all(&fc->reserved_req_waitq);
 375 }
 376 EXPORT_SYMBOL_GPL(fuse_conn_kill);
 377
 378 static void fuse_put_super(struct super_block *sb)
 379 {
 380         struct fuse_conn *fc = get_fuse_conn_super(sb);
 381
 382         fuse_send_destroy(fc);
 383
 384         fuse_conn_kill(fc);
 385         mutex_lock(&fuse_mutex);
 386         list_del(&fc->entry);
 387         fuse_ctl_remove_conn(fc);
 388         mutex_unlock(&fuse_mutex);
 389         fuse_bdi_destroy(fc);
 390
 391         fuse_conn_put(fc);
 392 }
 393
 394 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
 395 {
 396         stbuf->f_type    = FUSE_SUPER_MAGIC;
 397         stbuf->f_bsize   = attr->bsize;
 398         stbuf->f_frsize  = attr->frsize;
 399         stbuf->f_blocks  = attr->blocks;
 400         stbuf->f_bfree   = attr->bfree;
 401         stbuf->f_bavail  = attr->bavail;
 402         stbuf->f_files   = attr->files;
 403         stbuf->f_ffree   = attr->ffree;
 404         stbuf->f_namelen = attr->namelen;
 405         /* fsid is left zero */
 406 }
 407
 408 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
 409 {
 410         struct super_block *sb = dentry->d_sb;
 411         struct fuse_conn *fc = get_fuse_conn_super(sb);
 412         struct fuse_req *req;
 413         struct fuse_statfs_out outarg;
 414         int err;
 415
 416         if (!fuse_allow_current_process(fc)) {
 417                 buf->f_type = FUSE_SUPER_MAGIC;
 418                 return 0;
 419         }
 420
 421         req = fuse_get_req_nopages(fc);
 422         if (IS_ERR(req))
 423                 return PTR_ERR(req);
 424
 425         memset(&outarg, 0, sizeof(outarg));
 426         req->in.numargs = 0;
 427         req->in.h.opcode = FUSE_STATFS;
 428         req->in.h.nodeid = get_node_id(dentry->d_inode);
 429         req->out.numargs = 1;
 430         req->out.args[0].size =
 431                 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
 432         req->out.args[0].value = &outarg;
 433         fuse_request_send(fc, req);
 434         err = req->out.h.error;
 435         if (!err)
 436                 convert_fuse_statfs(buf, &outarg.st);
 437         fuse_put_request(fc, req);
 438         return err;
 439 }
 440
 441 enum {
 442         OPT_FD,
 443         OPT_ROOTMODE,
 444         OPT_USER_ID,
 445         OPT_GROUP_ID,
 446         OPT_DEFAULT_PERMISSIONS,
 447         OPT_ALLOW_OTHER,
 448         OPT_MAX_READ,
 449         OPT_BLKSIZE,
 450         OPT_ERR
 451 };
 452
 453 static const match_table_t tokens = {
 454         {OPT_FD,                        "fd=%u"},
 455         {OPT_ROOTMODE,                  "rootmode=%o"},
 456         {OPT_USER_ID,                   "user_id=%u"},
 457         {OPT_GROUP_ID,                  "group_id=%u"},
 458         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 459         {OPT_ALLOW_OTHER,               "allow_other"},
 460         {OPT_MAX_READ,                  "max_read=%u"},
 461         {OPT_BLKSIZE,                   "blksize=%u"},
 462         {OPT_ERR,                       NULL}
 463 };
 464
 465 static int fuse_match_uint(substring_t *s, unsigned int *res)
 466 {
 467         int err = -ENOMEM;
 468         char *buf = match_strdup(s);
 469         if (buf) {
 470                 err = kstrtouint(buf, 10, res);
 471                 kfree(buf);
 472         }
 473         return err;
 474 }
 475
 476 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
 477 {
 478         char *p;
 479         memset(d, 0, sizeof(struct fuse_mount_data));
 480         d->max_read = ~0;
 481         d->blksize = FUSE_DEFAULT_BLKSIZE;
 482
 483         while ((p = strsep(&opt, ",")) != NULL) {
 484                 int token;
 485                 int value;
 486                 unsigned uv;
 487                 substring_t args[MAX_OPT_ARGS];
 488                 if (!*p)
 489                         continue;
 490
 491                 token = match_token(p, tokens, args);
 492                 switch (token) {
 493                 case OPT_FD:
 494                         if (match_int(&args[0], &value))
 495                                 return 0;
 496                         d->fd = value;
 497                         d->fd_present = 1;
 498                         break;
 499
 500                 case OPT_ROOTMODE:
 501                         if (match_octal(&args[0], &value))
 502                                 return 0;
 503                         if (!fuse_valid_type(value))
 504                                 return 0;
 505                         d->rootmode = value;
 506                         d->rootmode_present = 1;
 507                         break;
 508
 509                 case OPT_USER_ID:
 510                         if (fuse_match_uint(&args[0], &uv))
 511                                 return 0;
 512                         d->user_id = make_kuid(current_user_ns(), uv);
 513                         if (!uid_valid(d->user_id))
 514                                 return 0;
 515                         d->user_id_present = 1;
 516                         break;
 517
 518                 case OPT_GROUP_ID:
 519                         if (fuse_match_uint(&args[0], &uv))
 520                                 return 0;
 521                         d->group_id = make_kgid(current_user_ns(), uv);
 522                         if (!gid_valid(d->group_id))
 523                                 return 0;
 524                         d->group_id_present = 1;
 525                         break;
 526
 527                 case OPT_DEFAULT_PERMISSIONS:
 528                         d->flags |= FUSE_DEFAULT_PERMISSIONS;
 529                         break;
 530
 531                 case OPT_ALLOW_OTHER:
 532                         d->flags |= FUSE_ALLOW_OTHER;
 533                         break;
 534
 535                 case OPT_MAX_READ:
 536                         if (match_int(&args[0], &value))
 537                                 return 0;
 538                         d->max_read = value;
 539                         break;
 540
 541                 case OPT_BLKSIZE:
 542                         if (!is_bdev || match_int(&args[0], &value))
 543                                 return 0;
 544                         d->blksize = value;
 545                         break;
 546
 547                 default:
 548                         return 0;
 549                 }
 550         }
 551
 552         if (!d->fd_present || !d->rootmode_present ||
 553             !d->user_id_present || !d->group_id_present)
 554                 return 0;
 555
 556         return 1;
 557 }
 558
 559 static int fuse_show_options(struct seq_file *m, struct dentry *root)
 560 {
 561         struct super_block *sb = root->d_sb;
 562         struct fuse_conn *fc = get_fuse_conn_super(sb);
 563
 564         seq_printf(m, ",user_id=%u", from_kuid_munged(&init_user_ns, fc->user_id));
 565         seq_printf(m, ",group_id=%u", from_kgid_munged(&init_user_ns, fc->group_id));
 566         if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
 567                 seq_puts(m, ",default_permissions");
 568         if (fc->flags & FUSE_ALLOW_OTHER)
 569                 seq_puts(m, ",allow_other");
 570         if (fc->max_read != ~0)
 571                 seq_printf(m, ",max_read=%u", fc->max_read);
 572         if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
 573                 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
 574         return 0;
 575 }
 576
 577 void fuse_conn_init(struct fuse_conn *fc)
 578 {
 579         memset(fc, 0, sizeof(*fc));
 580         spin_lock_init(&fc->lock);
 581         mutex_init(&fc->inst_mutex);
 582         init_rwsem(&fc->killsb);
 583         atomic_set(&fc->count, 1);
 584         init_waitqueue_head(&fc->waitq);
 585         init_waitqueue_head(&fc->blocked_waitq);
 586         init_waitqueue_head(&fc->reserved_req_waitq);
 587         INIT_LIST_HEAD(&fc->pending);
 588         INIT_LIST_HEAD(&fc->processing);
 589         INIT_LIST_HEAD(&fc->io);
 590         INIT_LIST_HEAD(&fc->interrupts);
 591         INIT_LIST_HEAD(&fc->bg_queue);
 592         INIT_LIST_HEAD(&fc->entry);
 593         fc->forget_list_tail = &fc->forget_list_head;
 594         atomic_set(&fc->num_waiting, 0);
 595         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
 596         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
 597         fc->khctr = 0;
 598         fc->polled_files = RB_ROOT;
 599         fc->reqctr = 0;
 600         fc->blocked = 0;
 601         fc->initialized = 0;
 602         fc->attr_version = 1;
 603         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
 604 }
 605 EXPORT_SYMBOL_GPL(fuse_conn_init);
 606
 607 void fuse_conn_put(struct fuse_conn *fc)
 608 {
 609         if (atomic_dec_and_test(&fc->count)) {
 610                 if (fc->destroy_req)
 611                         fuse_request_free(fc->destroy_req);
 612                 mutex_destroy(&fc->inst_mutex);
 613                 fc->release(fc);
 614         }
 615 }
 616 EXPORT_SYMBOL_GPL(fuse_conn_put);
 617
 618 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
 619 {
 620         atomic_inc(&fc->count);
 621         return fc;
 622 }
 623 EXPORT_SYMBOL_GPL(fuse_conn_get);
 624
 625 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
 626 {
 627         struct fuse_attr attr;
 628         memset(&attr, 0, sizeof(attr));
 629
 630         attr.mode = mode;
 631         attr.ino = FUSE_ROOT_ID;
 632         attr.nlink = 1;
 633         return fuse_iget(sb, 1, 0, &attr, 0, 0);
 634 }
 635
 636 struct fuse_inode_handle {
 637         u64 nodeid;
 638         u32 generation;
 639 };
 640
 641 static struct dentry *fuse_get_dentry(struct super_block *sb,
 642                                       struct fuse_inode_handle *handle)
 643 {
 644         struct fuse_conn *fc = get_fuse_conn_super(sb);
 645         struct inode *inode;
 646         struct dentry *entry;
 647         int err = -ESTALE;
 648
 649         if (handle->nodeid == 0)
 650                 goto out_err;
 651
 652         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
 653         if (!inode) {
 654                 struct fuse_entry_out outarg;
 655                 struct qstr name;
 656
 657                 if (!fc->export_support)
 658                         goto out_err;
 659
 660                 name.len = 1;
 661                 name.name = ".";
 662                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
 663                                        &inode);
 664                 if (err && err != -ENOENT)
 665                         goto out_err;
 666                 if (err || !inode) {
 667                         err = -ESTALE;
 668                         goto out_err;
 669                 }
 670                 err = -EIO;
 671                 if (get_node_id(inode) != handle->nodeid)
 672                         goto out_iput;
 673         }
 674         err = -ESTALE;
 675         if (inode->i_generation != handle->generation)
 676                 goto out_iput;
 677
 678         entry = d_obtain_alias(inode);
 679         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
 680                 fuse_invalidate_entry_cache(entry);
 681
 682         return entry;
 683
 684  out_iput:
 685         iput(inode);
 686  out_err:
 687         return ERR_PTR(err);
 688 }
 689
 690 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
 691                            struct inode *parent)
 692 {
 693         int len = parent ? 6 : 3;
 694         u64 nodeid;
 695         u32 generation;
 696
 697         if (*max_len < len) {
 698                 *max_len = len;
 699                 return  FILEID_INVALID;
 700         }
 701
 702         nodeid = get_fuse_inode(inode)->nodeid;
 703         generation = inode->i_generation;
 704
 705         fh[0] = (u32)(nodeid >> 32);
 706         fh[1] = (u32)(nodeid & 0xffffffff);
 707         fh[2] = generation;
 708
 709         if (parent) {
 710                 nodeid = get_fuse_inode(parent)->nodeid;
 711                 generation = parent->i_generation;
 712
 713                 fh[3] = (u32)(nodeid >> 32);
 714                 fh[4] = (u32)(nodeid & 0xffffffff);
 715                 fh[5] = generation;
 716         }
 717
 718         *max_len = len;
 719         return parent ? 0x82 : 0x81;
 720 }
 721
 722 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
 723                 struct fid *fid, int fh_len, int fh_type)
 724 {
 725         struct fuse_inode_handle handle;
 726
 727         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
 728                 return NULL;
 729
 730         handle.nodeid = (u64) fid->raw[0] << 32;
 731         handle.nodeid |= (u64) fid->raw[1];
 732         handle.generation = fid->raw[2];
 733         return fuse_get_dentry(sb, &handle);
 734 }
 735
 736 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
 737                 struct fid *fid, int fh_len, int fh_type)
 738 {
 739         struct fuse_inode_handle parent;
 740
 741         if (fh_type != 0x82 || fh_len < 6)
 742                 return NULL;
 743
 744         parent.nodeid = (u64) fid->raw[3] << 32;
 745         parent.nodeid |= (u64) fid->raw[4];
 746         parent.generation = fid->raw[5];
 747         return fuse_get_dentry(sb, &parent);
 748 }
 749
 750 static struct dentry *fuse_get_parent(struct dentry *child)
 751 {
 752         struct inode *child_inode = child->d_inode;
 753         struct fuse_conn *fc = get_fuse_conn(child_inode);
 754         struct inode *inode;
 755         struct dentry *parent;
 756         struct fuse_entry_out outarg;
 757         struct qstr name;
 758         int err;
 759
 760         if (!fc->export_support)
 761                 return ERR_PTR(-ESTALE);
 762
 763         name.len = 2;
 764         name.name = "..";
 765         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
 766                                &name, &outarg, &inode);
 767         if (err) {
 768                 if (err == -ENOENT)
 769                         return ERR_PTR(-ESTALE);
 770                 return ERR_PTR(err);
 771         }
 772
 773         parent = d_obtain_alias(inode);
 774         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
 775                 fuse_invalidate_entry_cache(parent);
 776
 777         return parent;
 778 }
 779
 780 static const struct export_operations fuse_export_operations = {
 781         .fh_to_dentry   = fuse_fh_to_dentry,
 782         .fh_to_parent   = fuse_fh_to_parent,
 783         .encode_fh      = fuse_encode_fh,
 784         .get_parent     = fuse_get_parent,
 785 };
 786
 787 static const struct super_operations fuse_super_operations = {
 788         .alloc_inode    = fuse_alloc_inode,
 789         .destroy_inode  = fuse_destroy_inode,
 790         .evict_inode    = fuse_evict_inode,
 791         .drop_inode     = generic_delete_inode,
 792         .remount_fs     = fuse_remount_fs,
 793         .put_super      = fuse_put_super,
 794         .umount_begin   = fuse_umount_begin,
 795         .statfs         = fuse_statfs,
 796         .show_options   = fuse_show_options,
 797 };
 798
 799 static void sanitize_global_limit(unsigned *limit)
 800 {
 801         if (*limit == 0)
 802                 *limit = ((num_physpages << PAGE_SHIFT) >> 13) /
 803                          sizeof(struct fuse_req);
 804
 805         if (*limit >= 1 << 16)
 806                 *limit = (1 << 16) - 1;
 807 }
 808
 809 static int set_global_limit(const char *val, struct kernel_param *kp)
 810 {
 811         int rv;
 812
 813         rv = param_set_uint(val, kp);
 814         if (rv)
 815                 return rv;
 816
 817         sanitize_global_limit((unsigned *)kp->arg);
 818
 819         return 0;
 820 }
 821
 822 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
 823 {
 824         int cap_sys_admin = capable(CAP_SYS_ADMIN);
 825
 826         if (arg->minor < 13)
 827                 return;
 828
 829         sanitize_global_limit(&max_user_bgreq);
 830         sanitize_global_limit(&max_user_congthresh);
 831
 832         if (arg->max_background) {
 833                 fc->max_background = arg->max_background;
 834
 835                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
 836                         fc->max_background = max_user_bgreq;
 837         }
 838         if (arg->congestion_threshold) {
 839                 fc->congestion_threshold = arg->congestion_threshold;
 840
 841                 if (!cap_sys_admin &&
 842                     fc->congestion_threshold > max_user_congthresh)
 843                         fc->congestion_threshold = max_user_congthresh;
 844         }
 845 }
 846
 847 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
 848 {
 849         struct fuse_init_out *arg = &req->misc.init_out;
 850
 851         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
 852                 fc->conn_error = 1;
 853         else {
 854                 unsigned long ra_pages;
 855
 856                 process_init_limits(fc, arg);
 857
 858                 if (arg->minor >= 6) {
 859                         ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
 860                         if (arg->flags & FUSE_ASYNC_READ)
 861                                 fc->async_read = 1;
 862                         if (!(arg->flags & FUSE_POSIX_LOCKS))
 863                                 fc->no_lock = 1;
 864                         if (arg->minor >= 17) {
 865                                 if (!(arg->flags & FUSE_FLOCK_LOCKS))
 866                                         fc->no_flock = 1;
 867                         } else {
 868                                 if (!(arg->flags & FUSE_POSIX_LOCKS))
 869                                         fc->no_flock = 1;
 870                         }
 871                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
 872                                 fc->atomic_o_trunc = 1;
 873                         if (arg->minor >= 9) {
 874                                 /* LOOKUP has dependency on proto version */
 875                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
 876                                         fc->export_support = 1;
 877                         }
 878                         if (arg->flags & FUSE_BIG_WRITES)
 879                                 fc->big_writes = 1;
 880                         if (arg->flags & FUSE_DONT_MASK)
 881                                 fc->dont_mask = 1;
 882                         if (arg->flags & FUSE_AUTO_INVAL_DATA)
 883                                 fc->auto_inval_data = 1;
 884                         if (arg->flags & FUSE_DO_READDIRPLUS) {
 885                                 fc->do_readdirplus = 1;
 886                                 if (arg->flags & FUSE_READDIRPLUS_AUTO)
 887                                         fc->readdirplus_auto = 1;
 888                         }
 889                         if (arg->flags & FUSE_ASYNC_DIO)
 890                                 fc->async_dio = 1;
 891                 } else {
 892                         ra_pages = fc->max_read / PAGE_CACHE_SIZE;
 893                         fc->no_lock = 1;
 894                         fc->no_flock = 1;
 895                 }
 896
 897                 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
 898                 fc->minor = arg->minor;
 899                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
 900                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
 901                 fc->conn_init = 1;
 902         }
 903         fc->initialized = 1;
 904         wake_up_all(&fc->blocked_waitq);
 905 }
 906
 907 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
 908 {
 909         struct fuse_init_in *arg = &req->misc.init_in;
 910
 911         arg->major = FUSE_KERNEL_VERSION;
 912         arg->minor = FUSE_KERNEL_MINOR_VERSION;
 913         arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
 914         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
 915                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
 916                 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
 917                 FUSE_FLOCK_LOCKS | FUSE_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
 918                 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO;
 919         req->in.h.opcode = FUSE_INIT;
 920         req->in.numargs = 1;
 921         req->in.args[0].size = sizeof(*arg);
 922         req->in.args[0].value = arg;
 923         req->out.numargs = 1;
 924         /* Variable length argument used for backward compatibility
 925            with interface version < 7.5.  Rest of init_out is zeroed
 926            by do_get_request(), so a short reply is not a problem */
 927         req->out.argvar = 1;
 928         req->out.args[0].size = sizeof(struct fuse_init_out);
 929         req->out.args[0].value = &req->misc.init_out;
 930         req->end = process_init_reply;
 931         fuse_request_send_background(fc, req);
 932 }
 933
 934 static void fuse_free_conn(struct fuse_conn *fc)
 935 {
 936         kfree(fc);
 937 }
 938
 939 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
 940 {
 941         int err;
 942
 943         fc->bdi.name = "fuse";
 944         fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
 945         /* fuse does it's own writeback accounting */
 946         fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB;
 947
 948         err = bdi_init(&fc->bdi);
 949         if (err)
 950                 return err;
 951
 952         fc->bdi_initialized = 1;
 953
 954         if (sb->s_bdev) {
 955                 err =  bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
 956                                     MAJOR(fc->dev), MINOR(fc->dev));
 957         } else {
 958                 err = bdi_register_dev(&fc->bdi, fc->dev);
 959         }
 960
 961         if (err)
 962                 return err;
 963
 964         /*
 965          * For a single fuse filesystem use max 1% of dirty +
 966          * writeback threshold.
 967          *
 968          * This gives about 1M of write buffer for memory maps on a
 969          * machine with 1G and 10% dirty_ratio, which should be more
 970          * than enough.
 971          *
 972          * Privileged users can raise it by writing to
 973          *
 974          *    /sys/class/bdi/<bdi>/max_ratio
 975          */
 976         bdi_set_max_ratio(&fc->bdi, 1);
 977
 978         return 0;
 979 }
 980
 981 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
 982 {
 983         struct fuse_conn *fc;
 984         struct inode *root;
 985         struct fuse_mount_data d;
 986         struct file *file;
 987         struct dentry *root_dentry;
 988         struct fuse_req *init_req;
 989         int err;
 990         int is_bdev = sb->s_bdev != NULL;
 991
 992         err = -EINVAL;
 993         if (sb->s_flags & MS_MANDLOCK)
 994                 goto err;
 995
 996         sb->s_flags &= ~MS_NOSEC;
 997
 998         if (!parse_fuse_opt((char *) data, &d, is_bdev))
 999                 goto err;
1000
1001         if (is_bdev) {
1002 #ifdef CONFIG_BLOCK
1003                 err = -EINVAL;
1004                 if (!sb_set_blocksize(sb, d.blksize))
1005                         goto err;
1006 #endif
1007         } else {
1008                 sb->s_blocksize = PAGE_CACHE_SIZE;
1009                 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
1010         }
1011         sb->s_magic = FUSE_SUPER_MAGIC;
1012         sb->s_op = &fuse_super_operations;
1013         sb->s_maxbytes = MAX_LFS_FILESIZE;
1014         sb->s_time_gran = 1;
1015         sb->s_export_op = &fuse_export_operations;
1016
1017         file = fget(d.fd);
1018         err = -EINVAL;
1019         if (!file)
1020                 goto err;
1021
1022         if ((file->f_op != &fuse_dev_operations) ||
1023             (file->f_cred->user_ns != &init_user_ns))
1024                 goto err_fput;
1025
1026         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1027         err = -ENOMEM;
1028         if (!fc)
1029                 goto err_fput;
1030
1031         fuse_conn_init(fc);
1032         fc->release = fuse_free_conn;
1033
1034         fc->dev = sb->s_dev;
1035         fc->sb = sb;
1036         err = fuse_bdi_init(fc, sb);
1037         if (err)
1038                 goto err_put_conn;
1039
1040         sb->s_bdi = &fc->bdi;
1041
1042         /* Handle umasking inside the fuse code */
1043         if (sb->s_flags & MS_POSIXACL)
1044                 fc->dont_mask = 1;
1045         sb->s_flags |= MS_POSIXACL;
1046
1047         fc->flags = d.flags;
1048         fc->user_id = d.user_id;
1049         fc->group_id = d.group_id;
1050         fc->max_read = max_t(unsigned, 4096, d.max_read);
1051
1052         /* Used by get_root_inode() */
1053         sb->s_fs_info = fc;
1054
1055         err = -ENOMEM;
1056         root = fuse_get_root_inode(sb, d.rootmode);
1057         root_dentry = d_make_root(root);
1058         if (!root_dentry)
1059                 goto err_put_conn;
1060         /* only now - we want root dentry with NULL ->d_op */
1061         sb->s_d_op = &fuse_dentry_operations;
1062
1063         init_req = fuse_request_alloc(0);
1064         if (!init_req)
1065                 goto err_put_root;
1066         init_req->background = 1;
1067
1068         if (is_bdev) {
1069                 fc->destroy_req = fuse_request_alloc(0);
1070                 if (!fc->destroy_req)
1071                         goto err_free_init_req;
1072         }
1073
1074         mutex_lock(&fuse_mutex);
1075         err = -EINVAL;
1076         if (file->private_data)
1077                 goto err_unlock;
1078
1079         err = fuse_ctl_add_conn(fc);
1080         if (err)
1081                 goto err_unlock;
1082
1083         list_add_tail(&fc->entry, &fuse_conn_list);
1084         sb->s_root = root_dentry;
1085         fc->connected = 1;
1086         file->private_data = fuse_conn_get(fc);
1087         mutex_unlock(&fuse_mutex);
1088         /*
1089          * atomic_dec_and_test() in fput() provides the necessary
1090          * memory barrier for file->private_data to be visible on all
1091          * CPUs after this
1092          */
1093         fput(file);
1094
1095         fuse_send_init(fc, init_req);
1096
1097         return 0;
1098
1099  err_unlock:
1100         mutex_unlock(&fuse_mutex);
1101  err_free_init_req:
1102         fuse_request_free(init_req);
1103  err_put_root:
1104         dput(root_dentry);
1105  err_put_conn:
1106         fuse_bdi_destroy(fc);
1107         fuse_conn_put(fc);
1108  err_fput:
1109         fput(file);
1110  err:
1111         return err;
1112 }
1113
1114 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1115                        int flags, const char *dev_name,
1116                        void *raw_data)
1117 {
1118         return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1119 }
1120
1121 static void fuse_kill_sb_anon(struct super_block *sb)
1122 {
1123         struct fuse_conn *fc = get_fuse_conn_super(sb);
1124
1125         if (fc) {
1126                 down_write(&fc->killsb);
1127                 fc->sb = NULL;
1128                 up_write(&fc->killsb);
1129         }
1130
1131         kill_anon_super(sb);
1132 }
1133
1134 static struct file_system_type fuse_fs_type = {
1135         .owner          = THIS_MODULE,
1136         .name           = "fuse",
1137         .fs_flags       = FS_HAS_SUBTYPE,
1138         .mount          = fuse_mount,
1139         .kill_sb        = fuse_kill_sb_anon,
1140 };
1141 MODULE_ALIAS_FS("fuse");
1142
1143 #ifdef CONFIG_BLOCK
1144 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1145                            int flags, const char *dev_name,
1146                            void *raw_data)
1147 {
1148         return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1149 }
1150
1151 static void fuse_kill_sb_blk(struct super_block *sb)
1152 {
1153         struct fuse_conn *fc = get_fuse_conn_super(sb);
1154
1155         if (fc) {
1156                 down_write(&fc->killsb);
1157                 fc->sb = NULL;
1158                 up_write(&fc->killsb);
1159         }
1160
1161         kill_block_super(sb);
1162 }
1163
1164 static struct file_system_type fuseblk_fs_type = {
1165         .owner          = THIS_MODULE,
1166         .name           = "fuseblk",
1167         .mount          = fuse_mount_blk,
1168         .kill_sb        = fuse_kill_sb_blk,
1169         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1170 };
1171 MODULE_ALIAS_FS("fuseblk");
1172
1173 static inline int register_fuseblk(void)
1174 {
1175         return register_filesystem(&fuseblk_fs_type);
1176 }
1177
1178 static inline void unregister_fuseblk(void)
1179 {
1180         unregister_filesystem(&fuseblk_fs_type);
1181 }
1182 #else
1183 static inline int register_fuseblk(void)
1184 {
1185         return 0;
1186 }
1187
1188 static inline void unregister_fuseblk(void)
1189 {
1190 }
1191 #endif
1192
1193 static void fuse_inode_init_once(void *foo)
1194 {
1195         struct inode *inode = foo;
1196
1197         inode_init_once(inode);
1198 }
1199
1200 static int __init fuse_fs_init(void)
1201 {
1202         int err;
1203
1204         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1205                                               sizeof(struct fuse_inode),
1206                                               0, SLAB_HWCACHE_ALIGN,
1207                                               fuse_inode_init_once);
1208         err = -ENOMEM;
1209         if (!fuse_inode_cachep)
1210                 goto out;
1211
1212         err = register_fuseblk();
1213         if (err)
1214                 goto out2;
1215
1216         err = register_filesystem(&fuse_fs_type);
1217         if (err)
1218                 goto out3;
1219
1220         return 0;
1221
1222  out3:
1223         unregister_fuseblk();
1224  out2:
1225         kmem_cache_destroy(fuse_inode_cachep);
1226  out:
1227         return err;
1228 }
1229
1230 static void fuse_fs_cleanup(void)
1231 {
1232         unregister_filesystem(&fuse_fs_type);
1233         unregister_fuseblk();
1234
1235         /*
1236          * Make sure all delayed rcu free inodes are flushed before we
1237          * destroy cache.
1238          */
1239         rcu_barrier();
1240         kmem_cache_destroy(fuse_inode_cachep);
1241 }
1242
1243 static struct kobject *fuse_kobj;
1244 static struct kobject *connections_kobj;
1245
1246 static int fuse_sysfs_init(void)
1247 {
1248         int err;
1249
1250         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1251         if (!fuse_kobj) {
1252                 err = -ENOMEM;
1253                 goto out_err;
1254         }
1255
1256         connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1257         if (!connections_kobj) {
1258                 err = -ENOMEM;
1259                 goto out_fuse_unregister;
1260         }
1261
1262         return 0;
1263
1264  out_fuse_unregister:
1265         kobject_put(fuse_kobj);
1266  out_err:
1267         return err;
1268 }
1269
1270 static void fuse_sysfs_cleanup(void)
1271 {
1272         kobject_put(connections_kobj);
1273         kobject_put(fuse_kobj);
1274 }
1275
1276 static int __init fuse_init(void)
1277 {
1278         int res;
1279
1280         printk(KERN_INFO "fuse init (API version %i.%i)\n",
1281                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1282
1283         INIT_LIST_HEAD(&fuse_conn_list);
1284         res = fuse_fs_init();
1285         if (res)
1286                 goto err;
1287
1288         res = fuse_dev_init();
1289         if (res)
1290                 goto err_fs_cleanup;
1291
1292         res = fuse_sysfs_init();
1293         if (res)
1294                 goto err_dev_cleanup;
1295
1296         res = fuse_ctl_init();
1297         if (res)
1298                 goto err_sysfs_cleanup;
1299
1300         sanitize_global_limit(&max_user_bgreq);
1301         sanitize_global_limit(&max_user_congthresh);
1302
1303         return 0;
1304
1305  err_sysfs_cleanup:
1306         fuse_sysfs_cleanup();
1307  err_dev_cleanup:
1308         fuse_dev_cleanup();
1309  err_fs_cleanup:
1310         fuse_fs_cleanup();
1311  err:
1312         return res;
1313 }
1314
1315 static void __exit fuse_exit(void)
1316 {
1317         printk(KERN_DEBUG "fuse exit\n");
1318
1319         fuse_ctl_cleanup();
1320         fuse_sysfs_cleanup();
1321         fuse_fs_cleanup();
1322         fuse_dev_cleanup();
1323 }
1324
1325 module_init(fuse_init);
1326 module_exit(fuse_exit);