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         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         kmem_cache_free(fuse_inode_cachep, inode);
 112 }
 113
 114 static void fuse_destroy_inode(struct inode *inode)
 115 {
 116         struct fuse_inode *fi = get_fuse_inode(inode);
 117         BUG_ON(!list_empty(&fi->write_files));
 118         BUG_ON(!list_empty(&fi->queued_writes));
 119         kfree(fi->forget);
 120         call_rcu(&inode->i_rcu, fuse_i_callback);
 121 }
 122
 123 static void fuse_evict_inode(struct inode *inode)
 124 {
 125         truncate_inode_pages(&inode->i_data, 0);
 126         clear_inode(inode);
 127         if (inode->i_sb->s_flags & MS_ACTIVE) {
 128                 struct fuse_conn *fc = get_fuse_conn(inode);
 129                 struct fuse_inode *fi = get_fuse_inode(inode);
 130                 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
 131                 fi->forget = NULL;
 132         }
 133 }
 134
 135 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
 136 {
 137         if (*flags & MS_MANDLOCK)
 138                 return -EINVAL;
 139
 140         return 0;
 141 }
 142
 143 /*
 144  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
 145  * so that it will fit.
 146  */
 147 static ino_t fuse_squash_ino(u64 ino64)
 148 {
 149         ino_t ino = (ino_t) ino64;
 150         if (sizeof(ino_t) < sizeof(u64))
 151                 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
 152         return ino;
 153 }
 154
 155 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
 156                                    u64 attr_valid)
 157 {
 158         struct fuse_conn *fc = get_fuse_conn(inode);
 159         struct fuse_inode *fi = get_fuse_inode(inode);
 160
 161         fi->attr_version = ++fc->attr_version;
 162         fi->i_time = attr_valid;
 163
 164         inode->i_ino     = fuse_squash_ino(attr->ino);
 165         inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
 166         set_nlink(inode, attr->nlink);
 167         inode->i_uid     = attr->uid;
 168         inode->i_gid     = attr->gid;
 169         inode->i_blocks  = attr->blocks;
 170         inode->i_atime.tv_sec   = attr->atime;
 171         inode->i_atime.tv_nsec  = attr->atimensec;
 172         inode->i_mtime.tv_sec   = attr->mtime;
 173         inode->i_mtime.tv_nsec  = attr->mtimensec;
 174         inode->i_ctime.tv_sec   = attr->ctime;
 175         inode->i_ctime.tv_nsec  = attr->ctimensec;
 176
 177         if (attr->blksize != 0)
 178                 inode->i_blkbits = ilog2(attr->blksize);
 179         else
 180                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
 181
 182         /*
 183          * Don't set the sticky bit in i_mode, unless we want the VFS
 184          * to check permissions.  This prevents failures due to the
 185          * check in may_delete().
 186          */
 187         fi->orig_i_mode = inode->i_mode;
 188         if (!(fc->flags & FUSE_DEFAULT_PERMISSIONS))
 189                 inode->i_mode &= ~S_ISVTX;
 190
 191         fi->orig_ino = attr->ino;
 192 }
 193
 194 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
 195                             u64 attr_valid, u64 attr_version)
 196 {
 197         struct fuse_conn *fc = get_fuse_conn(inode);
 198         struct fuse_inode *fi = get_fuse_inode(inode);
 199         loff_t oldsize;
 200         struct timespec old_mtime;
 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         old_mtime = inode->i_mtime;
 209         fuse_change_attributes_common(inode, attr, attr_valid);
 210
 211         oldsize = inode->i_size;
 212         i_size_write(inode, attr->size);
 213         spin_unlock(&fc->lock);
 214
 215         if (S_ISREG(inode->i_mode)) {
 216                 bool inval = false;
 217
 218                 if (oldsize != attr->size) {
 219                         truncate_pagecache(inode, oldsize, attr->size);
 220                         inval = true;
 221                 } else if (fc->auto_inval_data) {
 222                         struct timespec new_mtime = {
 223                                 .tv_sec = attr->mtime,
 224                                 .tv_nsec = attr->mtimensec,
 225                         };
 226
 227                         /*
 228                          * Auto inval mode also checks and invalidates if mtime
 229                          * has changed.
 230                          */
 231                         if (!timespec_equal(&old_mtime, &new_mtime))
 232                                 inval = true;
 233                 }
 234
 235                 if (inval)
 236                         invalidate_inode_pages2(inode->i_mapping);
 237         }
 238 }
 239
 240 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
 241 {
 242         inode->i_mode = attr->mode & S_IFMT;
 243         inode->i_size = attr->size;
 244         if (S_ISREG(inode->i_mode)) {
 245                 fuse_init_common(inode);
 246                 fuse_init_file_inode(inode);
 247         } else if (S_ISDIR(inode->i_mode))
 248                 fuse_init_dir(inode);
 249         else if (S_ISLNK(inode->i_mode))
 250                 fuse_init_symlink(inode);
 251         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
 252                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
 253                 fuse_init_common(inode);
 254                 init_special_inode(inode, inode->i_mode,
 255                                    new_decode_dev(attr->rdev));
 256         } else
 257                 BUG();
 258 }
 259
 260 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
 261 {
 262         u64 nodeid = *(u64 *) _nodeidp;
 263         if (get_node_id(inode) == nodeid)
 264                 return 1;
 265         else
 266                 return 0;
 267 }
 268
 269 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
 270 {
 271         u64 nodeid = *(u64 *) _nodeidp;
 272         get_fuse_inode(inode)->nodeid = nodeid;
 273         return 0;
 274 }
 275
 276 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
 277                         int generation, struct fuse_attr *attr,
 278                         u64 attr_valid, u64 attr_version)
 279 {
 280         struct inode *inode;
 281         struct fuse_inode *fi;
 282         struct fuse_conn *fc = get_fuse_conn_super(sb);
 283
 284  retry:
 285         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
 286         if (!inode)
 287                 return NULL;
 288
 289         if ((inode->i_state & I_NEW)) {
 290                 inode->i_flags |= S_NOATIME|S_NOCMTIME;
 291                 inode->i_generation = generation;
 292                 inode->i_data.backing_dev_info = &fc->bdi;
 293                 fuse_init_inode(inode, attr);
 294                 unlock_new_inode(inode);
 295         } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
 296                 /* Inode has changed type, any I/O on the old should fail */
 297                 make_bad_inode(inode);
 298                 iput(inode);
 299                 goto retry;
 300         }
 301
 302         fi = get_fuse_inode(inode);
 303         spin_lock(&fc->lock);
 304         fi->nlookup++;
 305         spin_unlock(&fc->lock);
 306         fuse_change_attributes(inode, attr, attr_valid, attr_version);
 307
 308         return inode;
 309 }
 310
 311 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
 312                              loff_t offset, loff_t len)
 313 {
 314         struct inode *inode;
 315         pgoff_t pg_start;
 316         pgoff_t pg_end;
 317
 318         inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
 319         if (!inode)
 320                 return -ENOENT;
 321
 322         fuse_invalidate_attr(inode);
 323         if (offset >= 0) {
 324                 pg_start = offset >> PAGE_CACHE_SHIFT;
 325                 if (len <= 0)
 326                         pg_end = -1;
 327                 else
 328                         pg_end = (offset + len - 1) >> PAGE_CACHE_SHIFT;
 329                 invalidate_inode_pages2_range(inode->i_mapping,
 330                                               pg_start, pg_end);
 331         }
 332         iput(inode);
 333         return 0;
 334 }
 335
 336 static void fuse_umount_begin(struct super_block *sb)
 337 {
 338         fuse_abort_conn(get_fuse_conn_super(sb));
 339 }
 340
 341 static void fuse_send_destroy(struct fuse_conn *fc)
 342 {
 343         struct fuse_req *req = fc->destroy_req;
 344         if (req && fc->conn_init) {
 345                 fc->destroy_req = NULL;
 346                 req->in.h.opcode = FUSE_DESTROY;
 347                 req->force = 1;
 348                 fuse_request_send(fc, req);
 349                 fuse_put_request(fc, req);
 350         }
 351 }
 352
 353 static void fuse_bdi_destroy(struct fuse_conn *fc)
 354 {
 355         if (fc->bdi_initialized)
 356                 bdi_destroy(&fc->bdi);
 357 }
 358
 359 void fuse_conn_kill(struct fuse_conn *fc)
 360 {
 361         spin_lock(&fc->lock);
 362         fc->connected = 0;
 363         fc->blocked = 0;
 364         spin_unlock(&fc->lock);
 365         /* Flush all readers on this fs */
 366         kill_fasync(&fc->fasync, SIGIO, POLL_IN);
 367         wake_up_all(&fc->waitq);
 368         wake_up_all(&fc->blocked_waitq);
 369         wake_up_all(&fc->reserved_req_waitq);
 370         mutex_lock(&fuse_mutex);
 371         list_del(&fc->entry);
 372         fuse_ctl_remove_conn(fc);
 373         mutex_unlock(&fuse_mutex);
 374         fuse_bdi_destroy(fc);
 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         fuse_conn_kill(fc);
 384         fuse_conn_put(fc);
 385 }
 386
 387 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
 388 {
 389         stbuf->f_type    = FUSE_SUPER_MAGIC;
 390         stbuf->f_bsize   = attr->bsize;
 391         stbuf->f_frsize  = attr->frsize;
 392         stbuf->f_blocks  = attr->blocks;
 393         stbuf->f_bfree   = attr->bfree;
 394         stbuf->f_bavail  = attr->bavail;
 395         stbuf->f_files   = attr->files;
 396         stbuf->f_ffree   = attr->ffree;
 397         stbuf->f_namelen = attr->namelen;
 398         /* fsid is left zero */
 399 }
 400
 401 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
 402 {
 403         struct super_block *sb = dentry->d_sb;
 404         struct fuse_conn *fc = get_fuse_conn_super(sb);
 405         struct fuse_req *req;
 406         struct fuse_statfs_out outarg;
 407         int err;
 408
 409         if (!fuse_allow_task(fc, current)) {
 410                 buf->f_type = FUSE_SUPER_MAGIC;
 411                 return 0;
 412         }
 413
 414         req = fuse_get_req(fc);
 415         if (IS_ERR(req))
 416                 return PTR_ERR(req);
 417
 418         memset(&outarg, 0, sizeof(outarg));
 419         req->in.numargs = 0;
 420         req->in.h.opcode = FUSE_STATFS;
 421         req->in.h.nodeid = get_node_id(dentry->d_inode);
 422         req->out.numargs = 1;
 423         req->out.args[0].size =
 424                 fc->minor < 4 ? FUSE_COMPAT_STATFS_SIZE : sizeof(outarg);
 425         req->out.args[0].value = &outarg;
 426         fuse_request_send(fc, req);
 427         err = req->out.h.error;
 428         if (!err)
 429                 convert_fuse_statfs(buf, &outarg.st);
 430         fuse_put_request(fc, req);
 431         return err;
 432 }
 433
 434 enum {
 435         OPT_FD,
 436         OPT_ROOTMODE,
 437         OPT_USER_ID,
 438         OPT_GROUP_ID,
 439         OPT_DEFAULT_PERMISSIONS,
 440         OPT_ALLOW_OTHER,
 441         OPT_MAX_READ,
 442         OPT_BLKSIZE,
 443         OPT_ERR
 444 };
 445
 446 static const match_table_t tokens = {
 447         {OPT_FD,                        "fd=%u"},
 448         {OPT_ROOTMODE,                  "rootmode=%o"},
 449         {OPT_USER_ID,                   "user_id=%u"},
 450         {OPT_GROUP_ID,                  "group_id=%u"},
 451         {OPT_DEFAULT_PERMISSIONS,       "default_permissions"},
 452         {OPT_ALLOW_OTHER,               "allow_other"},
 453         {OPT_MAX_READ,                  "max_read=%u"},
 454         {OPT_BLKSIZE,                   "blksize=%u"},
 455         {OPT_ERR,                       NULL}
 456 };
 457
 458 static int parse_fuse_opt(char *opt, struct fuse_mount_data *d, int is_bdev)
 459 {
 460         char *p;
 461         memset(d, 0, sizeof(struct fuse_mount_data));
 462         d->max_read = ~0;
 463         d->blksize = FUSE_DEFAULT_BLKSIZE;
 464
 465         while ((p = strsep(&opt, ",")) != NULL) {
 466                 int token;
 467                 int value;
 468                 substring_t args[MAX_OPT_ARGS];
 469                 if (!*p)
 470                         continue;
 471
 472                 token = match_token(p, tokens, args);
 473                 switch (token) {
 474                 case OPT_FD:
 475                         if (match_int(&args[0], &value))
 476                                 return 0;
 477                         d->fd = value;
 478                         d->fd_present = 1;
 479                         break;
 480
 481                 case OPT_ROOTMODE:
 482                         if (match_octal(&args[0], &value))
 483                                 return 0;
 484                         if (!fuse_valid_type(value))
 485                                 return 0;
 486                         d->rootmode = value;
 487                         d->rootmode_present = 1;
 488                         break;
 489
 490                 case OPT_USER_ID:
 491                         if (match_int(&args[0], &value))
 492                                 return 0;
 493                         d->user_id = value;
 494                         d->user_id_present = 1;
 495                         break;
 496
 497                 case OPT_GROUP_ID:
 498                         if (match_int(&args[0], &value))
 499                                 return 0;
 500                         d->group_id = value;
 501                         d->group_id_present = 1;
 502                         break;
 503
 504                 case OPT_DEFAULT_PERMISSIONS:
 505                         d->flags |= FUSE_DEFAULT_PERMISSIONS;
 506                         break;
 507
 508                 case OPT_ALLOW_OTHER:
 509                         d->flags |= FUSE_ALLOW_OTHER;
 510                         break;
 511
 512                 case OPT_MAX_READ:
 513                         if (match_int(&args[0], &value))
 514                                 return 0;
 515                         d->max_read = value;
 516                         break;
 517
 518                 case OPT_BLKSIZE:
 519                         if (!is_bdev || match_int(&args[0], &value))
 520                                 return 0;
 521                         d->blksize = value;
 522                         break;
 523
 524                 default:
 525                         return 0;
 526                 }
 527         }
 528
 529         if (!d->fd_present || !d->rootmode_present ||
 530             !d->user_id_present || !d->group_id_present)
 531                 return 0;
 532
 533         return 1;
 534 }
 535
 536 static int fuse_show_options(struct seq_file *m, struct dentry *root)
 537 {
 538         struct super_block *sb = root->d_sb;
 539         struct fuse_conn *fc = get_fuse_conn_super(sb);
 540
 541         seq_printf(m, ",user_id=%u", fc->user_id);
 542         seq_printf(m, ",group_id=%u", fc->group_id);
 543         if (fc->flags & FUSE_DEFAULT_PERMISSIONS)
 544                 seq_puts(m, ",default_permissions");
 545         if (fc->flags & FUSE_ALLOW_OTHER)
 546                 seq_puts(m, ",allow_other");
 547         if (fc->max_read != ~0)
 548                 seq_printf(m, ",max_read=%u", fc->max_read);
 549         if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
 550                 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
 551         return 0;
 552 }
 553
 554 void fuse_conn_init(struct fuse_conn *fc)
 555 {
 556         memset(fc, 0, sizeof(*fc));
 557         spin_lock_init(&fc->lock);
 558         mutex_init(&fc->inst_mutex);
 559         init_rwsem(&fc->killsb);
 560         atomic_set(&fc->count, 1);
 561         init_waitqueue_head(&fc->waitq);
 562         init_waitqueue_head(&fc->blocked_waitq);
 563         init_waitqueue_head(&fc->reserved_req_waitq);
 564         INIT_LIST_HEAD(&fc->pending);
 565         INIT_LIST_HEAD(&fc->processing);
 566         INIT_LIST_HEAD(&fc->io);
 567         INIT_LIST_HEAD(&fc->interrupts);
 568         INIT_LIST_HEAD(&fc->bg_queue);
 569         INIT_LIST_HEAD(&fc->entry);
 570         fc->forget_list_tail = &fc->forget_list_head;
 571         atomic_set(&fc->num_waiting, 0);
 572         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
 573         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
 574         fc->khctr = 0;
 575         fc->polled_files = RB_ROOT;
 576         fc->reqctr = 0;
 577         fc->blocked = 1;
 578         fc->attr_version = 1;
 579         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
 580 }
 581 EXPORT_SYMBOL_GPL(fuse_conn_init);
 582
 583 void fuse_conn_put(struct fuse_conn *fc)
 584 {
 585         if (atomic_dec_and_test(&fc->count)) {
 586                 if (fc->destroy_req)
 587                         fuse_request_free(fc->destroy_req);
 588                 mutex_destroy(&fc->inst_mutex);
 589                 fc->release(fc);
 590         }
 591 }
 592 EXPORT_SYMBOL_GPL(fuse_conn_put);
 593
 594 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
 595 {
 596         atomic_inc(&fc->count);
 597         return fc;
 598 }
 599 EXPORT_SYMBOL_GPL(fuse_conn_get);
 600
 601 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
 602 {
 603         struct fuse_attr attr;
 604         memset(&attr, 0, sizeof(attr));
 605
 606         attr.mode = mode;
 607         attr.ino = FUSE_ROOT_ID;
 608         attr.nlink = 1;
 609         return fuse_iget(sb, 1, 0, &attr, 0, 0);
 610 }
 611
 612 struct fuse_inode_handle {
 613         u64 nodeid;
 614         u32 generation;
 615 };
 616
 617 static struct dentry *fuse_get_dentry(struct super_block *sb,
 618                                       struct fuse_inode_handle *handle)
 619 {
 620         struct fuse_conn *fc = get_fuse_conn_super(sb);
 621         struct inode *inode;
 622         struct dentry *entry;
 623         int err = -ESTALE;
 624
 625         if (handle->nodeid == 0)
 626                 goto out_err;
 627
 628         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
 629         if (!inode) {
 630                 struct fuse_entry_out outarg;
 631                 struct qstr name;
 632
 633                 if (!fc->export_support)
 634                         goto out_err;
 635
 636                 name.len = 1;
 637                 name.name = ".";
 638                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
 639                                        &inode);
 640                 if (err && err != -ENOENT)
 641                         goto out_err;
 642                 if (err || !inode) {
 643                         err = -ESTALE;
 644                         goto out_err;
 645                 }
 646                 err = -EIO;
 647                 if (get_node_id(inode) != handle->nodeid)
 648                         goto out_iput;
 649         }
 650         err = -ESTALE;
 651         if (inode->i_generation != handle->generation)
 652                 goto out_iput;
 653
 654         entry = d_obtain_alias(inode);
 655         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
 656                 fuse_invalidate_entry_cache(entry);
 657
 658         return entry;
 659
 660  out_iput:
 661         iput(inode);
 662  out_err:
 663         return ERR_PTR(err);
 664 }
 665
 666 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
 667                            struct inode *parent)
 668 {
 669         int len = parent ? 6 : 3;
 670         u64 nodeid;
 671         u32 generation;
 672
 673         if (*max_len < len) {
 674                 *max_len = len;
 675                 return  255;
 676         }
 677
 678         nodeid = get_fuse_inode(inode)->nodeid;
 679         generation = inode->i_generation;
 680
 681         fh[0] = (u32)(nodeid >> 32);
 682         fh[1] = (u32)(nodeid & 0xffffffff);
 683         fh[2] = generation;
 684
 685         if (parent) {
 686                 nodeid = get_fuse_inode(parent)->nodeid;
 687                 generation = parent->i_generation;
 688
 689                 fh[3] = (u32)(nodeid >> 32);
 690                 fh[4] = (u32)(nodeid & 0xffffffff);
 691                 fh[5] = generation;
 692         }
 693
 694         *max_len = len;
 695         return parent ? 0x82 : 0x81;
 696 }
 697
 698 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
 699                 struct fid *fid, int fh_len, int fh_type)
 700 {
 701         struct fuse_inode_handle handle;
 702
 703         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
 704                 return NULL;
 705
 706         handle.nodeid = (u64) fid->raw[0] << 32;
 707         handle.nodeid |= (u64) fid->raw[1];
 708         handle.generation = fid->raw[2];
 709         return fuse_get_dentry(sb, &handle);
 710 }
 711
 712 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
 713                 struct fid *fid, int fh_len, int fh_type)
 714 {
 715         struct fuse_inode_handle parent;
 716
 717         if (fh_type != 0x82 || fh_len < 6)
 718                 return NULL;
 719
 720         parent.nodeid = (u64) fid->raw[3] << 32;
 721         parent.nodeid |= (u64) fid->raw[4];
 722         parent.generation = fid->raw[5];
 723         return fuse_get_dentry(sb, &parent);
 724 }
 725
 726 static struct dentry *fuse_get_parent(struct dentry *child)
 727 {
 728         struct inode *child_inode = child->d_inode;
 729         struct fuse_conn *fc = get_fuse_conn(child_inode);
 730         struct inode *inode;
 731         struct dentry *parent;
 732         struct fuse_entry_out outarg;
 733         struct qstr name;
 734         int err;
 735
 736         if (!fc->export_support)
 737                 return ERR_PTR(-ESTALE);
 738
 739         name.len = 2;
 740         name.name = "..";
 741         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
 742                                &name, &outarg, &inode);
 743         if (err) {
 744                 if (err == -ENOENT)
 745                         return ERR_PTR(-ESTALE);
 746                 return ERR_PTR(err);
 747         }
 748
 749         parent = d_obtain_alias(inode);
 750         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
 751                 fuse_invalidate_entry_cache(parent);
 752
 753         return parent;
 754 }
 755
 756 static const struct export_operations fuse_export_operations = {
 757         .fh_to_dentry   = fuse_fh_to_dentry,
 758         .fh_to_parent   = fuse_fh_to_parent,
 759         .encode_fh      = fuse_encode_fh,
 760         .get_parent     = fuse_get_parent,
 761 };
 762
 763 static const struct super_operations fuse_super_operations = {
 764         .alloc_inode    = fuse_alloc_inode,
 765         .destroy_inode  = fuse_destroy_inode,
 766         .evict_inode    = fuse_evict_inode,
 767         .drop_inode     = generic_delete_inode,
 768         .remount_fs     = fuse_remount_fs,
 769         .put_super      = fuse_put_super,
 770         .umount_begin   = fuse_umount_begin,
 771         .statfs         = fuse_statfs,
 772         .show_options   = fuse_show_options,
 773 };
 774
 775 static void sanitize_global_limit(unsigned *limit)
 776 {
 777         if (*limit == 0)
 778                 *limit = ((num_physpages << PAGE_SHIFT) >> 13) /
 779                          sizeof(struct fuse_req);
 780
 781         if (*limit >= 1 << 16)
 782                 *limit = (1 << 16) - 1;
 783 }
 784
 785 static int set_global_limit(const char *val, struct kernel_param *kp)
 786 {
 787         int rv;
 788
 789         rv = param_set_uint(val, kp);
 790         if (rv)
 791                 return rv;
 792
 793         sanitize_global_limit((unsigned *)kp->arg);
 794
 795         return 0;
 796 }
 797
 798 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
 799 {
 800         int cap_sys_admin = capable(CAP_SYS_ADMIN);
 801
 802         if (arg->minor < 13)
 803                 return;
 804
 805         sanitize_global_limit(&max_user_bgreq);
 806         sanitize_global_limit(&max_user_congthresh);
 807
 808         if (arg->max_background) {
 809                 fc->max_background = arg->max_background;
 810
 811                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
 812                         fc->max_background = max_user_bgreq;
 813         }
 814         if (arg->congestion_threshold) {
 815                 fc->congestion_threshold = arg->congestion_threshold;
 816
 817                 if (!cap_sys_admin &&
 818                     fc->congestion_threshold > max_user_congthresh)
 819                         fc->congestion_threshold = max_user_congthresh;
 820         }
 821 }
 822
 823 static void process_init_reply(struct fuse_conn *fc, struct fuse_req *req)
 824 {
 825         struct fuse_init_out *arg = &req->misc.init_out;
 826
 827         if (req->out.h.error || arg->major != FUSE_KERNEL_VERSION)
 828                 fc->conn_error = 1;
 829         else {
 830                 unsigned long ra_pages;
 831
 832                 process_init_limits(fc, arg);
 833
 834                 if (arg->minor >= 6) {
 835                         ra_pages = arg->max_readahead / PAGE_CACHE_SIZE;
 836                         if (arg->flags & FUSE_ASYNC_READ)
 837                                 fc->async_read = 1;
 838                         if (!(arg->flags & FUSE_POSIX_LOCKS))
 839                                 fc->no_lock = 1;
 840                         if (arg->minor >= 17) {
 841                                 if (!(arg->flags & FUSE_FLOCK_LOCKS))
 842                                         fc->no_flock = 1;
 843                         } else {
 844                                 if (!(arg->flags & FUSE_POSIX_LOCKS))
 845                                         fc->no_flock = 1;
 846                         }
 847                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
 848                                 fc->atomic_o_trunc = 1;
 849                         if (arg->minor >= 9) {
 850                                 /* LOOKUP has dependency on proto version */
 851                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
 852                                         fc->export_support = 1;
 853                         }
 854                         if (arg->flags & FUSE_BIG_WRITES)
 855                                 fc->big_writes = 1;
 856                         if (arg->flags & FUSE_DONT_MASK)
 857                                 fc->dont_mask = 1;
 858                         if (arg->flags & FUSE_AUTO_INVAL_DATA)
 859                                 fc->auto_inval_data = 1;
 860                 } else {
 861                         ra_pages = fc->max_read / PAGE_CACHE_SIZE;
 862                         fc->no_lock = 1;
 863                         fc->no_flock = 1;
 864                 }
 865
 866                 fc->bdi.ra_pages = min(fc->bdi.ra_pages, ra_pages);
 867                 fc->minor = arg->minor;
 868                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
 869                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
 870                 fc->conn_init = 1;
 871         }
 872         fc->blocked = 0;
 873         wake_up_all(&fc->blocked_waitq);
 874 }
 875
 876 static void fuse_send_init(struct fuse_conn *fc, struct fuse_req *req)
 877 {
 878         struct fuse_init_in *arg = &req->misc.init_in;
 879
 880         arg->major = FUSE_KERNEL_VERSION;
 881         arg->minor = FUSE_KERNEL_MINOR_VERSION;
 882         arg->max_readahead = fc->bdi.ra_pages * PAGE_CACHE_SIZE;
 883         arg->flags |= FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
 884                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
 885                 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
 886                 FUSE_FLOCK_LOCKS | FUSE_IOCTL_DIR | FUSE_AUTO_INVAL_DATA;
 887         req->in.h.opcode = FUSE_INIT;
 888         req->in.numargs = 1;
 889         req->in.args[0].size = sizeof(*arg);
 890         req->in.args[0].value = arg;
 891         req->out.numargs = 1;
 892         /* Variable length argument used for backward compatibility
 893            with interface version < 7.5.  Rest of init_out is zeroed
 894            by do_get_request(), so a short reply is not a problem */
 895         req->out.argvar = 1;
 896         req->out.args[0].size = sizeof(struct fuse_init_out);
 897         req->out.args[0].value = &req->misc.init_out;
 898         req->end = process_init_reply;
 899         fuse_request_send_background(fc, req);
 900 }
 901
 902 static void fuse_free_conn(struct fuse_conn *fc)
 903 {
 904         kfree(fc);
 905 }
 906
 907 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
 908 {
 909         int err;
 910
 911         fc->bdi.name = "fuse";
 912         fc->bdi.ra_pages = (VM_MAX_READAHEAD * 1024) / PAGE_CACHE_SIZE;
 913         /* fuse does it's own writeback accounting */
 914         fc->bdi.capabilities = BDI_CAP_NO_ACCT_WB;
 915
 916         err = bdi_init(&fc->bdi);
 917         if (err)
 918                 return err;
 919
 920         fc->bdi_initialized = 1;
 921
 922         if (sb->s_bdev) {
 923                 err =  bdi_register(&fc->bdi, NULL, "%u:%u-fuseblk",
 924                                     MAJOR(fc->dev), MINOR(fc->dev));
 925         } else {
 926                 err = bdi_register_dev(&fc->bdi, fc->dev);
 927         }
 928
 929         if (err)
 930                 return err;
 931
 932         /*
 933          * For a single fuse filesystem use max 1% of dirty +
 934          * writeback threshold.
 935          *
 936          * This gives about 1M of write buffer for memory maps on a
 937          * machine with 1G and 10% dirty_ratio, which should be more
 938          * than enough.
 939          *
 940          * Privileged users can raise it by writing to
 941          *
 942          *    /sys/class/bdi/<bdi>/max_ratio
 943          */
 944         bdi_set_max_ratio(&fc->bdi, 1);
 945
 946         return 0;
 947 }
 948
 949 static int fuse_fill_super(struct super_block *sb, void *data, int silent)
 950 {
 951         struct fuse_conn *fc;
 952         struct inode *root;
 953         struct fuse_mount_data d;
 954         struct file *file;
 955         struct dentry *root_dentry;
 956         struct fuse_req *init_req;
 957         int err;
 958         int is_bdev = sb->s_bdev != NULL;
 959
 960         err = -EINVAL;
 961         if (sb->s_flags & MS_MANDLOCK)
 962                 goto err;
 963
 964         sb->s_flags &= ~MS_NOSEC;
 965
 966         if (!parse_fuse_opt((char *) data, &d, is_bdev))
 967                 goto err;
 968
 969         if (is_bdev) {
 970 #ifdef CONFIG_BLOCK
 971                 err = -EINVAL;
 972                 if (!sb_set_blocksize(sb, d.blksize))
 973                         goto err;
 974 #endif
 975         } else {
 976                 sb->s_blocksize = PAGE_CACHE_SIZE;
 977                 sb->s_blocksize_bits = PAGE_CACHE_SHIFT;
 978         }
 979         sb->s_magic = FUSE_SUPER_MAGIC;
 980         sb->s_op = &fuse_super_operations;
 981         sb->s_maxbytes = MAX_LFS_FILESIZE;
 982         sb->s_time_gran = 1;
 983         sb->s_export_op = &fuse_export_operations;
 984
 985         file = fget(d.fd);
 986         err = -EINVAL;
 987         if (!file)
 988                 goto err;
 989
 990         if (file->f_op != &fuse_dev_operations)
 991                 goto err_fput;
 992
 993         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
 994         err = -ENOMEM;
 995         if (!fc)
 996                 goto err_fput;
 997
 998         fuse_conn_init(fc);
 999
1000         fc->dev = sb->s_dev;
1001         fc->sb = sb;
1002         err = fuse_bdi_init(fc, sb);
1003         if (err)
1004                 goto err_put_conn;
1005
1006         sb->s_bdi = &fc->bdi;
1007
1008         /* Handle umasking inside the fuse code */
1009         if (sb->s_flags & MS_POSIXACL)
1010                 fc->dont_mask = 1;
1011         sb->s_flags |= MS_POSIXACL;
1012
1013         fc->release = fuse_free_conn;
1014         fc->flags = d.flags;
1015         fc->user_id = d.user_id;
1016         fc->group_id = d.group_id;
1017         fc->max_read = max_t(unsigned, 4096, d.max_read);
1018
1019         /* Used by get_root_inode() */
1020         sb->s_fs_info = fc;
1021
1022         err = -ENOMEM;
1023         root = fuse_get_root_inode(sb, d.rootmode);
1024         root_dentry = d_make_root(root);
1025         if (!root_dentry)
1026                 goto err_put_conn;
1027         /* only now - we want root dentry with NULL ->d_op */
1028         sb->s_d_op = &fuse_dentry_operations;
1029
1030         init_req = fuse_request_alloc();
1031         if (!init_req)
1032                 goto err_put_root;
1033
1034         if (is_bdev) {
1035                 fc->destroy_req = fuse_request_alloc();
1036                 if (!fc->destroy_req)
1037                         goto err_free_init_req;
1038         }
1039
1040         mutex_lock(&fuse_mutex);
1041         err = -EINVAL;
1042         if (file->private_data)
1043                 goto err_unlock;
1044
1045         err = fuse_ctl_add_conn(fc);
1046         if (err)
1047                 goto err_unlock;
1048
1049         list_add_tail(&fc->entry, &fuse_conn_list);
1050         sb->s_root = root_dentry;
1051         fc->connected = 1;
1052         file->private_data = fuse_conn_get(fc);
1053         mutex_unlock(&fuse_mutex);
1054         /*
1055          * atomic_dec_and_test() in fput() provides the necessary
1056          * memory barrier for file->private_data to be visible on all
1057          * CPUs after this
1058          */
1059         fput(file);
1060
1061         fuse_send_init(fc, init_req);
1062
1063         return 0;
1064
1065  err_unlock:
1066         mutex_unlock(&fuse_mutex);
1067  err_free_init_req:
1068         fuse_request_free(init_req);
1069  err_put_root:
1070         dput(root_dentry);
1071  err_put_conn:
1072         fuse_bdi_destroy(fc);
1073         fuse_conn_put(fc);
1074  err_fput:
1075         fput(file);
1076  err:
1077         return err;
1078 }
1079
1080 static struct dentry *fuse_mount(struct file_system_type *fs_type,
1081                        int flags, const char *dev_name,
1082                        void *raw_data)
1083 {
1084         return mount_nodev(fs_type, flags, raw_data, fuse_fill_super);
1085 }
1086
1087 static void fuse_kill_sb_anon(struct super_block *sb)
1088 {
1089         struct fuse_conn *fc = get_fuse_conn_super(sb);
1090
1091         if (fc) {
1092                 down_write(&fc->killsb);
1093                 fc->sb = NULL;
1094                 up_write(&fc->killsb);
1095         }
1096
1097         kill_anon_super(sb);
1098 }
1099
1100 static struct file_system_type fuse_fs_type = {
1101         .owner          = THIS_MODULE,
1102         .name           = "fuse",
1103         .fs_flags       = FS_HAS_SUBTYPE,
1104         .mount          = fuse_mount,
1105         .kill_sb        = fuse_kill_sb_anon,
1106 };
1107
1108 #ifdef CONFIG_BLOCK
1109 static struct dentry *fuse_mount_blk(struct file_system_type *fs_type,
1110                            int flags, const char *dev_name,
1111                            void *raw_data)
1112 {
1113         return mount_bdev(fs_type, flags, dev_name, raw_data, fuse_fill_super);
1114 }
1115
1116 static void fuse_kill_sb_blk(struct super_block *sb)
1117 {
1118         struct fuse_conn *fc = get_fuse_conn_super(sb);
1119
1120         if (fc) {
1121                 down_write(&fc->killsb);
1122                 fc->sb = NULL;
1123                 up_write(&fc->killsb);
1124         }
1125
1126         kill_block_super(sb);
1127 }
1128
1129 static struct file_system_type fuseblk_fs_type = {
1130         .owner          = THIS_MODULE,
1131         .name           = "fuseblk",
1132         .mount          = fuse_mount_blk,
1133         .kill_sb        = fuse_kill_sb_blk,
1134         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1135 };
1136
1137 static inline int register_fuseblk(void)
1138 {
1139         return register_filesystem(&fuseblk_fs_type);
1140 }
1141
1142 static inline void unregister_fuseblk(void)
1143 {
1144         unregister_filesystem(&fuseblk_fs_type);
1145 }
1146 #else
1147 static inline int register_fuseblk(void)
1148 {
1149         return 0;
1150 }
1151
1152 static inline void unregister_fuseblk(void)
1153 {
1154 }
1155 #endif
1156
1157 static void fuse_inode_init_once(void *foo)
1158 {
1159         struct inode *inode = foo;
1160
1161         inode_init_once(inode);
1162 }
1163
1164 static int __init fuse_fs_init(void)
1165 {
1166         int err;
1167
1168         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1169                                               sizeof(struct fuse_inode),
1170                                               0, SLAB_HWCACHE_ALIGN,
1171                                               fuse_inode_init_once);
1172         err = -ENOMEM;
1173         if (!fuse_inode_cachep)
1174                 goto out;
1175
1176         err = register_fuseblk();
1177         if (err)
1178                 goto out2;
1179
1180         err = register_filesystem(&fuse_fs_type);
1181         if (err)
1182                 goto out3;
1183
1184         return 0;
1185
1186  out3:
1187         unregister_fuseblk();
1188  out2:
1189         kmem_cache_destroy(fuse_inode_cachep);
1190  out:
1191         return err;
1192 }
1193
1194 static void fuse_fs_cleanup(void)
1195 {
1196         unregister_filesystem(&fuse_fs_type);
1197         unregister_fuseblk();
1198         kmem_cache_destroy(fuse_inode_cachep);
1199 }
1200
1201 static struct kobject *fuse_kobj;
1202 static struct kobject *connections_kobj;
1203
1204 static int fuse_sysfs_init(void)
1205 {
1206         int err;
1207
1208         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1209         if (!fuse_kobj) {
1210                 err = -ENOMEM;
1211                 goto out_err;
1212         }
1213
1214         connections_kobj = kobject_create_and_add("connections", fuse_kobj);
1215         if (!connections_kobj) {
1216                 err = -ENOMEM;
1217                 goto out_fuse_unregister;
1218         }
1219
1220         return 0;
1221
1222  out_fuse_unregister:
1223         kobject_put(fuse_kobj);
1224  out_err:
1225         return err;
1226 }
1227
1228 static void fuse_sysfs_cleanup(void)
1229 {
1230         kobject_put(connections_kobj);
1231         kobject_put(fuse_kobj);
1232 }
1233
1234 static int __init fuse_init(void)
1235 {
1236         int res;
1237
1238         printk(KERN_INFO "fuse init (API version %i.%i)\n",
1239                FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1240
1241         INIT_LIST_HEAD(&fuse_conn_list);
1242         res = fuse_fs_init();
1243         if (res)
1244                 goto err;
1245
1246         res = fuse_dev_init();
1247         if (res)
1248                 goto err_fs_cleanup;
1249
1250         res = fuse_sysfs_init();
1251         if (res)
1252                 goto err_dev_cleanup;
1253
1254         res = fuse_ctl_init();
1255         if (res)
1256                 goto err_sysfs_cleanup;
1257
1258         sanitize_global_limit(&max_user_bgreq);
1259         sanitize_global_limit(&max_user_congthresh);
1260
1261         return 0;
1262
1263  err_sysfs_cleanup:
1264         fuse_sysfs_cleanup();
1265  err_dev_cleanup:
1266         fuse_dev_cleanup();
1267  err_fs_cleanup:
1268         fuse_fs_cleanup();
1269  err:
1270         return res;
1271 }
1272
1273 static void __exit fuse_exit(void)
1274 {
1275         printk(KERN_DEBUG "fuse exit\n");
1276
1277         fuse_ctl_cleanup();
1278         fuse_sysfs_cleanup();
1279         fuse_fs_cleanup();
1280         fuse_dev_cleanup();
1281 }
1282
1283 module_init(fuse_init);
1284 module_exit(fuse_exit);