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/fs_context.h>
  19 #include <linux/fs_parser.h>
  20 #include <linux/statfs.h>
  21 #include <linux/random.h>
  22 #include <linux/sched.h>
  23 #include <linux/exportfs.h>
  24 #include <linux/posix_acl.h>
  25 #include <linux/pid_namespace.h>
  26
  27 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  28 MODULE_DESCRIPTION("Filesystem in Userspace");
  29 MODULE_LICENSE("GPL");
  30
  31 static struct kmem_cache *fuse_inode_cachep;
  32 struct list_head fuse_conn_list;
  33 DEFINE_MUTEX(fuse_mutex);
  34
  35 static int set_global_limit(const char *val, const struct kernel_param *kp);
  36
  37 unsigned max_user_bgreq;
  38 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
  39                   &max_user_bgreq, 0644);
  40 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
  41 MODULE_PARM_DESC(max_user_bgreq,
  42  "Global limit for the maximum number of backgrounded requests an "
  43  "unprivileged user can set");
  44
  45 unsigned max_user_congthresh;
  46 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
  47                   &max_user_congthresh, 0644);
  48 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
  49 MODULE_PARM_DESC(max_user_congthresh,
  50  "Global limit for the maximum congestion threshold an "
  51  "unprivileged user can set");
  52
  53 #define FUSE_SUPER_MAGIC 0x65735546
  54
  55 #define FUSE_DEFAULT_BLKSIZE 512
  56
  57 /** Maximum number of outstanding background requests */
  58 #define FUSE_DEFAULT_MAX_BACKGROUND 12
  59
  60 /** Congestion starts at 75% of maximum */
  61 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
  62
  63 #ifdef CONFIG_BLOCK
  64 static struct file_system_type fuseblk_fs_type;
  65 #endif
  66
  67 struct fuse_forget_link *fuse_alloc_forget(void)
  68 {
  69         return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL_ACCOUNT);
  70 }
  71
  72 static struct inode *fuse_alloc_inode(struct super_block *sb)
  73 {
  74         struct fuse_inode *fi;
  75
  76         fi = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
  77         if (!fi)
  78                 return NULL;
  79
  80         fi->i_time = 0;
  81         fi->inval_mask = 0;
  82         fi->nodeid = 0;
  83         fi->nlookup = 0;
  84         fi->attr_version = 0;
  85         fi->orig_ino = 0;
  86         fi->state = 0;
  87         mutex_init(&fi->mutex);
  88         spin_lock_init(&fi->lock);
  89         fi->forget = fuse_alloc_forget();
  90         if (!fi->forget) {
  91                 kmem_cache_free(fuse_inode_cachep, fi);
  92                 return NULL;
  93         }
  94
  95         return &fi->inode;
  96 }
  97
  98 static void fuse_free_inode(struct inode *inode)
  99 {
 100         struct fuse_inode *fi = get_fuse_inode(inode);
 101
 102         mutex_destroy(&fi->mutex);
 103         kfree(fi->forget);
 104         kmem_cache_free(fuse_inode_cachep, fi);
 105 }
 106
 107 static void fuse_evict_inode(struct inode *inode)
 108 {
 109         struct fuse_inode *fi = get_fuse_inode(inode);
 110
 111         truncate_inode_pages_final(&inode->i_data);
 112         clear_inode(inode);
 113         if (inode->i_sb->s_flags & SB_ACTIVE) {
 114                 struct fuse_conn *fc = get_fuse_conn(inode);
 115                 fuse_queue_forget(fc, fi->forget, fi->nodeid, fi->nlookup);
 116                 fi->forget = NULL;
 117         }
 118         if (S_ISREG(inode->i_mode) && !is_bad_inode(inode)) {
 119                 WARN_ON(!list_empty(&fi->write_files));
 120                 WARN_ON(!list_empty(&fi->queued_writes));
 121         }
 122 }
 123
 124 static int fuse_remount_fs(struct super_block *sb, int *flags, char *data)
 125 {
 126         sync_filesystem(sb);
 127         if (*flags & SB_MANDLOCK)
 128                 return -EINVAL;
 129
 130         return 0;
 131 }
 132
 133 /*
 134  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
 135  * so that it will fit.
 136  */
 137 static ino_t fuse_squash_ino(u64 ino64)
 138 {
 139         ino_t ino = (ino_t) ino64;
 140         if (sizeof(ino_t) < sizeof(u64))
 141                 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
 142         return ino;
 143 }
 144
 145 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
 146                                    u64 attr_valid)
 147 {
 148         struct fuse_conn *fc = get_fuse_conn(inode);
 149         struct fuse_inode *fi = get_fuse_inode(inode);
 150
 151         lockdep_assert_held(&fi->lock);
 152
 153         fi->attr_version = atomic64_inc_return(&fc->attr_version);
 154         fi->i_time = attr_valid;
 155         WRITE_ONCE(fi->inval_mask, 0);
 156
 157         inode->i_ino     = fuse_squash_ino(attr->ino);
 158         inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
 159         set_nlink(inode, attr->nlink);
 160         inode->i_uid     = make_kuid(fc->user_ns, attr->uid);
 161         inode->i_gid     = make_kgid(fc->user_ns, attr->gid);
 162         inode->i_blocks  = attr->blocks;
 163         inode->i_atime.tv_sec   = attr->atime;
 164         inode->i_atime.tv_nsec  = attr->atimensec;
 165         /* mtime from server may be stale due to local buffered write */
 166         if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
 167                 inode->i_mtime.tv_sec   = attr->mtime;
 168                 inode->i_mtime.tv_nsec  = attr->mtimensec;
 169                 inode->i_ctime.tv_sec   = attr->ctime;
 170                 inode->i_ctime.tv_nsec  = attr->ctimensec;
 171         }
 172
 173         if (attr->blksize != 0)
 174                 inode->i_blkbits = ilog2(attr->blksize);
 175         else
 176                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
 177
 178         /*
 179          * Don't set the sticky bit in i_mode, unless we want the VFS
 180          * to check permissions.  This prevents failures due to the
 181          * check in may_delete().
 182          */
 183         fi->orig_i_mode = inode->i_mode;
 184         if (!fc->default_permissions)
 185                 inode->i_mode &= ~S_ISVTX;
 186
 187         fi->orig_ino = attr->ino;
 188 }
 189
 190 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
 191                             u64 attr_valid, u64 attr_version)
 192 {
 193         struct fuse_conn *fc = get_fuse_conn(inode);
 194         struct fuse_inode *fi = get_fuse_inode(inode);
 195         bool is_wb = fc->writeback_cache;
 196         loff_t oldsize;
 197         struct timespec64 old_mtime;
 198
 199         spin_lock(&fi->lock);
 200         if ((attr_version != 0 && fi->attr_version > attr_version) ||
 201             test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
 202                 spin_unlock(&fi->lock);
 203                 return;
 204         }
 205
 206         old_mtime = inode->i_mtime;
 207         fuse_change_attributes_common(inode, attr, attr_valid);
 208
 209         oldsize = inode->i_size;
 210         /*
 211          * In case of writeback_cache enabled, the cached writes beyond EOF
 212          * extend local i_size without keeping userspace server in sync. So,
 213          * attr->size coming from server can be stale. We cannot trust it.
 214          */
 215         if (!is_wb || !S_ISREG(inode->i_mode))
 216                 i_size_write(inode, attr->size);
 217         spin_unlock(&fi->lock);
 218
 219         if (!is_wb && S_ISREG(inode->i_mode)) {
 220                 bool inval = false;
 221
 222                 if (oldsize != attr->size) {
 223                         truncate_pagecache(inode, attr->size);
 224                         if (!fc->explicit_inval_data)
 225                                 inval = true;
 226                 } else if (fc->auto_inval_data) {
 227                         struct timespec64 new_mtime = {
 228                                 .tv_sec = attr->mtime,
 229                                 .tv_nsec = attr->mtimensec,
 230                         };
 231
 232                         /*
 233                          * Auto inval mode also checks and invalidates if mtime
 234                          * has changed.
 235                          */
 236                         if (!timespec64_equal(&old_mtime, &new_mtime))
 237                                 inval = true;
 238                 }
 239
 240                 if (inval)
 241                         invalidate_inode_pages2(inode->i_mapping);
 242         }
 243 }
 244
 245 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
 246 {
 247         inode->i_mode = attr->mode & S_IFMT;
 248         inode->i_size = attr->size;
 249         inode->i_mtime.tv_sec  = attr->mtime;
 250         inode->i_mtime.tv_nsec = attr->mtimensec;
 251         inode->i_ctime.tv_sec  = attr->ctime;
 252         inode->i_ctime.tv_nsec = attr->ctimensec;
 253         if (S_ISREG(inode->i_mode)) {
 254                 fuse_init_common(inode);
 255                 fuse_init_file_inode(inode);
 256         } else if (S_ISDIR(inode->i_mode))
 257                 fuse_init_dir(inode);
 258         else if (S_ISLNK(inode->i_mode))
 259                 fuse_init_symlink(inode);
 260         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
 261                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
 262                 fuse_init_common(inode);
 263                 init_special_inode(inode, inode->i_mode,
 264                                    new_decode_dev(attr->rdev));
 265         } else
 266                 BUG();
 267 }
 268
 269 int fuse_inode_eq(struct inode *inode, void *_nodeidp)
 270 {
 271         u64 nodeid = *(u64 *) _nodeidp;
 272         if (get_node_id(inode) == nodeid)
 273                 return 1;
 274         else
 275                 return 0;
 276 }
 277
 278 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
 279 {
 280         u64 nodeid = *(u64 *) _nodeidp;
 281         get_fuse_inode(inode)->nodeid = nodeid;
 282         return 0;
 283 }
 284
 285 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
 286                         int generation, struct fuse_attr *attr,
 287                         u64 attr_valid, u64 attr_version)
 288 {
 289         struct inode *inode;
 290         struct fuse_inode *fi;
 291         struct fuse_conn *fc = get_fuse_conn_super(sb);
 292
 293  retry:
 294         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
 295         if (!inode)
 296                 return NULL;
 297
 298         if ((inode->i_state & I_NEW)) {
 299                 inode->i_flags |= S_NOATIME;
 300                 if (!fc->writeback_cache || !S_ISREG(attr->mode))
 301                         inode->i_flags |= S_NOCMTIME;
 302                 inode->i_generation = generation;
 303                 fuse_init_inode(inode, attr);
 304                 unlock_new_inode(inode);
 305         } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
 306                 /* Inode has changed type, any I/O on the old should fail */
 307                 make_bad_inode(inode);
 308                 iput(inode);
 309                 goto retry;
 310         }
 311
 312         fi = get_fuse_inode(inode);
 313         spin_lock(&fi->lock);
 314         fi->nlookup++;
 315         spin_unlock(&fi->lock);
 316         fuse_change_attributes(inode, attr, attr_valid, attr_version);
 317
 318         return inode;
 319 }
 320
 321 int fuse_reverse_inval_inode(struct super_block *sb, u64 nodeid,
 322                              loff_t offset, loff_t len)
 323 {
 324         struct inode *inode;
 325         pgoff_t pg_start;
 326         pgoff_t pg_end;
 327
 328         inode = ilookup5(sb, nodeid, fuse_inode_eq, &nodeid);
 329         if (!inode)
 330                 return -ENOENT;
 331
 332         fuse_invalidate_attr(inode);
 333         forget_all_cached_acls(inode);
 334         if (offset >= 0) {
 335                 pg_start = offset >> PAGE_SHIFT;
 336                 if (len <= 0)
 337                         pg_end = -1;
 338                 else
 339                         pg_end = (offset + len - 1) >> PAGE_SHIFT;
 340                 invalidate_inode_pages2_range(inode->i_mapping,
 341                                               pg_start, pg_end);
 342         }
 343         iput(inode);
 344         return 0;
 345 }
 346
 347 bool fuse_lock_inode(struct inode *inode)
 348 {
 349         bool locked = false;
 350
 351         if (!get_fuse_conn(inode)->parallel_dirops) {
 352                 mutex_lock(&get_fuse_inode(inode)->mutex);
 353                 locked = true;
 354         }
 355
 356         return locked;
 357 }
 358
 359 void fuse_unlock_inode(struct inode *inode, bool locked)
 360 {
 361         if (locked)
 362                 mutex_unlock(&get_fuse_inode(inode)->mutex);
 363 }
 364
 365 static void fuse_umount_begin(struct super_block *sb)
 366 {
 367         struct fuse_conn *fc = get_fuse_conn_super(sb);
 368
 369         if (!fc->no_force_umount)
 370                 fuse_abort_conn(fc);
 371 }
 372
 373 static void fuse_send_destroy(struct fuse_conn *fc)
 374 {
 375         if (fc->conn_init) {
 376                 FUSE_ARGS(args);
 377
 378                 args.opcode = FUSE_DESTROY;
 379                 args.force = true;
 380                 args.nocreds = true;
 381                 fuse_simple_request(fc, &args);
 382         }
 383 }
 384
 385 static void fuse_put_super(struct super_block *sb)
 386 {
 387         struct fuse_conn *fc = get_fuse_conn_super(sb);
 388
 389         mutex_lock(&fuse_mutex);
 390         list_del(&fc->entry);
 391         fuse_ctl_remove_conn(fc);
 392         mutex_unlock(&fuse_mutex);
 393
 394         fuse_conn_put(fc);
 395 }
 396
 397 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
 398 {
 399         stbuf->f_type    = FUSE_SUPER_MAGIC;
 400         stbuf->f_bsize   = attr->bsize;
 401         stbuf->f_frsize  = attr->frsize;
 402         stbuf->f_blocks  = attr->blocks;
 403         stbuf->f_bfree   = attr->bfree;
 404         stbuf->f_bavail  = attr->bavail;
 405         stbuf->f_files   = attr->files;
 406         stbuf->f_ffree   = attr->ffree;
 407         stbuf->f_namelen = attr->namelen;
 408         /* fsid is left zero */
 409 }
 410
 411 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
 412 {
 413         struct super_block *sb = dentry->d_sb;
 414         struct fuse_conn *fc = get_fuse_conn_super(sb);
 415         FUSE_ARGS(args);
 416         struct fuse_statfs_out outarg;
 417         int err;
 418
 419         if (!fuse_allow_current_process(fc)) {
 420                 buf->f_type = FUSE_SUPER_MAGIC;
 421                 return 0;
 422         }
 423
 424         memset(&outarg, 0, sizeof(outarg));
 425         args.in_numargs = 0;
 426         args.opcode = FUSE_STATFS;
 427         args.nodeid = get_node_id(d_inode(dentry));
 428         args.out_numargs = 1;
 429         args.out_args[0].size = sizeof(outarg);
 430         args.out_args[0].value = &outarg;
 431         err = fuse_simple_request(fc, &args);
 432         if (!err)
 433                 convert_fuse_statfs(buf, &outarg.st);
 434         return err;
 435 }
 436
 437 enum {
 438         OPT_SOURCE,
 439         OPT_SUBTYPE,
 440         OPT_FD,
 441         OPT_ROOTMODE,
 442         OPT_USER_ID,
 443         OPT_GROUP_ID,
 444         OPT_DEFAULT_PERMISSIONS,
 445         OPT_ALLOW_OTHER,
 446         OPT_MAX_READ,
 447         OPT_BLKSIZE,
 448         OPT_ERR
 449 };
 450
 451 static const struct fs_parameter_spec fuse_fs_parameters[] = {
 452         fsparam_string  ("source",              OPT_SOURCE),
 453         fsparam_u32     ("fd",                  OPT_FD),
 454         fsparam_u32oct  ("rootmode",            OPT_ROOTMODE),
 455         fsparam_u32     ("user_id",             OPT_USER_ID),
 456         fsparam_u32     ("group_id",            OPT_GROUP_ID),
 457         fsparam_flag    ("default_permissions", OPT_DEFAULT_PERMISSIONS),
 458         fsparam_flag    ("allow_other",         OPT_ALLOW_OTHER),
 459         fsparam_u32     ("max_read",            OPT_MAX_READ),
 460         fsparam_u32     ("blksize",             OPT_BLKSIZE),
 461         fsparam_string  ("subtype",             OPT_SUBTYPE),
 462         {}
 463 };
 464
 465 static int fuse_parse_param(struct fs_context *fc, struct fs_parameter *param)
 466 {
 467         struct fs_parse_result result;
 468         struct fuse_fs_context *ctx = fc->fs_private;
 469         int opt;
 470
 471         opt = fs_parse(fc, fuse_fs_parameters, param, &result);
 472         if (opt < 0)
 473                 return opt;
 474
 475         switch (opt) {
 476         case OPT_SOURCE:
 477                 if (fc->source)
 478                         return invalfc(fc, "Multiple sources specified");
 479                 fc->source = param->string;
 480                 param->string = NULL;
 481                 break;
 482
 483         case OPT_SUBTYPE:
 484                 if (ctx->subtype)
 485                         return invalfc(fc, "Multiple subtypes specified");
 486                 ctx->subtype = param->string;
 487                 param->string = NULL;
 488                 return 0;
 489
 490         case OPT_FD:
 491                 ctx->fd = result.uint_32;
 492                 ctx->fd_present = true;
 493                 break;
 494
 495         case OPT_ROOTMODE:
 496                 if (!fuse_valid_type(result.uint_32))
 497                         return invalfc(fc, "Invalid rootmode");
 498                 ctx->rootmode = result.uint_32;
 499                 ctx->rootmode_present = true;
 500                 break;
 501
 502         case OPT_USER_ID:
 503                 ctx->user_id = make_kuid(fc->user_ns, result.uint_32);
 504                 if (!uid_valid(ctx->user_id))
 505                         return invalfc(fc, "Invalid user_id");
 506                 ctx->user_id_present = true;
 507                 break;
 508
 509         case OPT_GROUP_ID:
 510                 ctx->group_id = make_kgid(fc->user_ns, result.uint_32);
 511                 if (!gid_valid(ctx->group_id))
 512                         return invalfc(fc, "Invalid group_id");
 513                 ctx->group_id_present = true;
 514                 break;
 515
 516         case OPT_DEFAULT_PERMISSIONS:
 517                 ctx->default_permissions = true;
 518                 break;
 519
 520         case OPT_ALLOW_OTHER:
 521                 ctx->allow_other = true;
 522                 break;
 523
 524         case OPT_MAX_READ:
 525                 ctx->max_read = result.uint_32;
 526                 break;
 527
 528         case OPT_BLKSIZE:
 529                 if (!ctx->is_bdev)
 530                         return invalfc(fc, "blksize only supported for fuseblk");
 531                 ctx->blksize = result.uint_32;
 532                 break;
 533
 534         default:
 535                 return -EINVAL;
 536         }
 537
 538         return 0;
 539 }
 540
 541 static void fuse_free_fc(struct fs_context *fc)
 542 {
 543         struct fuse_fs_context *ctx = fc->fs_private;
 544
 545         if (ctx) {
 546                 kfree(ctx->subtype);
 547                 kfree(ctx);
 548         }
 549 }
 550
 551 static int fuse_show_options(struct seq_file *m, struct dentry *root)
 552 {
 553         struct super_block *sb = root->d_sb;
 554         struct fuse_conn *fc = get_fuse_conn_super(sb);
 555
 556         if (fc->no_mount_options)
 557                 return 0;
 558
 559         seq_printf(m, ",user_id=%u", from_kuid_munged(fc->user_ns, fc->user_id));
 560         seq_printf(m, ",group_id=%u", from_kgid_munged(fc->user_ns, fc->group_id));
 561         if (fc->default_permissions)
 562                 seq_puts(m, ",default_permissions");
 563         if (fc->allow_other)
 564                 seq_puts(m, ",allow_other");
 565         if (fc->max_read != ~0)
 566                 seq_printf(m, ",max_read=%u", fc->max_read);
 567         if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
 568                 seq_printf(m, ",blksize=%lu", sb->s_blocksize);
 569         return 0;
 570 }
 571
 572 static void fuse_iqueue_init(struct fuse_iqueue *fiq,
 573                              const struct fuse_iqueue_ops *ops,
 574                              void *priv)
 575 {
 576         memset(fiq, 0, sizeof(struct fuse_iqueue));
 577         spin_lock_init(&fiq->lock);
 578         init_waitqueue_head(&fiq->waitq);
 579         INIT_LIST_HEAD(&fiq->pending);
 580         INIT_LIST_HEAD(&fiq->interrupts);
 581         fiq->forget_list_tail = &fiq->forget_list_head;
 582         fiq->connected = 1;
 583         fiq->ops = ops;
 584         fiq->priv = priv;
 585 }
 586
 587 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
 588 {
 589         unsigned int i;
 590
 591         spin_lock_init(&fpq->lock);
 592         for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
 593                 INIT_LIST_HEAD(&fpq->processing[i]);
 594         INIT_LIST_HEAD(&fpq->io);
 595         fpq->connected = 1;
 596 }
 597
 598 void fuse_conn_init(struct fuse_conn *fc, struct user_namespace *user_ns,
 599                     const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
 600 {
 601         memset(fc, 0, sizeof(*fc));
 602         spin_lock_init(&fc->lock);
 603         spin_lock_init(&fc->bg_lock);
 604         init_rwsem(&fc->killsb);
 605         refcount_set(&fc->count, 1);
 606         atomic_set(&fc->dev_count, 1);
 607         init_waitqueue_head(&fc->blocked_waitq);
 608         fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
 609         INIT_LIST_HEAD(&fc->bg_queue);
 610         INIT_LIST_HEAD(&fc->entry);
 611         INIT_LIST_HEAD(&fc->devices);
 612         atomic_set(&fc->num_waiting, 0);
 613         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
 614         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
 615         atomic64_set(&fc->khctr, 0);
 616         fc->polled_files = RB_ROOT;
 617         fc->blocked = 0;
 618         fc->initialized = 0;
 619         fc->connected = 1;
 620         atomic64_set(&fc->attr_version, 1);
 621         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
 622         fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
 623         fc->user_ns = get_user_ns(user_ns);
 624         fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
 625 }
 626 EXPORT_SYMBOL_GPL(fuse_conn_init);
 627
 628 void fuse_conn_put(struct fuse_conn *fc)
 629 {
 630         if (refcount_dec_and_test(&fc->count)) {
 631                 struct fuse_iqueue *fiq = &fc->iq;
 632
 633                 if (fiq->ops->release)
 634                         fiq->ops->release(fiq);
 635                 put_pid_ns(fc->pid_ns);
 636                 put_user_ns(fc->user_ns);
 637                 fc->release(fc);
 638         }
 639 }
 640 EXPORT_SYMBOL_GPL(fuse_conn_put);
 641
 642 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
 643 {
 644         refcount_inc(&fc->count);
 645         return fc;
 646 }
 647 EXPORT_SYMBOL_GPL(fuse_conn_get);
 648
 649 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
 650 {
 651         struct fuse_attr attr;
 652         memset(&attr, 0, sizeof(attr));
 653
 654         attr.mode = mode;
 655         attr.ino = FUSE_ROOT_ID;
 656         attr.nlink = 1;
 657         return fuse_iget(sb, 1, 0, &attr, 0, 0);
 658 }
 659
 660 struct fuse_inode_handle {
 661         u64 nodeid;
 662         u32 generation;
 663 };
 664
 665 static struct dentry *fuse_get_dentry(struct super_block *sb,
 666                                       struct fuse_inode_handle *handle)
 667 {
 668         struct fuse_conn *fc = get_fuse_conn_super(sb);
 669         struct inode *inode;
 670         struct dentry *entry;
 671         int err = -ESTALE;
 672
 673         if (handle->nodeid == 0)
 674                 goto out_err;
 675
 676         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
 677         if (!inode) {
 678                 struct fuse_entry_out outarg;
 679                 const struct qstr name = QSTR_INIT(".", 1);
 680
 681                 if (!fc->export_support)
 682                         goto out_err;
 683
 684                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
 685                                        &inode);
 686                 if (err && err != -ENOENT)
 687                         goto out_err;
 688                 if (err || !inode) {
 689                         err = -ESTALE;
 690                         goto out_err;
 691                 }
 692                 err = -EIO;
 693                 if (get_node_id(inode) != handle->nodeid)
 694                         goto out_iput;
 695         }
 696         err = -ESTALE;
 697         if (inode->i_generation != handle->generation)
 698                 goto out_iput;
 699
 700         entry = d_obtain_alias(inode);
 701         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
 702                 fuse_invalidate_entry_cache(entry);
 703
 704         return entry;
 705
 706  out_iput:
 707         iput(inode);
 708  out_err:
 709         return ERR_PTR(err);
 710 }
 711
 712 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
 713                            struct inode *parent)
 714 {
 715         int len = parent ? 6 : 3;
 716         u64 nodeid;
 717         u32 generation;
 718
 719         if (*max_len < len) {
 720                 *max_len = len;
 721                 return  FILEID_INVALID;
 722         }
 723
 724         nodeid = get_fuse_inode(inode)->nodeid;
 725         generation = inode->i_generation;
 726
 727         fh[0] = (u32)(nodeid >> 32);
 728         fh[1] = (u32)(nodeid & 0xffffffff);
 729         fh[2] = generation;
 730
 731         if (parent) {
 732                 nodeid = get_fuse_inode(parent)->nodeid;
 733                 generation = parent->i_generation;
 734
 735                 fh[3] = (u32)(nodeid >> 32);
 736                 fh[4] = (u32)(nodeid & 0xffffffff);
 737                 fh[5] = generation;
 738         }
 739
 740         *max_len = len;
 741         return parent ? 0x82 : 0x81;
 742 }
 743
 744 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
 745                 struct fid *fid, int fh_len, int fh_type)
 746 {
 747         struct fuse_inode_handle handle;
 748
 749         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
 750                 return NULL;
 751
 752         handle.nodeid = (u64) fid->raw[0] << 32;
 753         handle.nodeid |= (u64) fid->raw[1];
 754         handle.generation = fid->raw[2];
 755         return fuse_get_dentry(sb, &handle);
 756 }
 757
 758 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
 759                 struct fid *fid, int fh_len, int fh_type)
 760 {
 761         struct fuse_inode_handle parent;
 762
 763         if (fh_type != 0x82 || fh_len < 6)
 764                 return NULL;
 765
 766         parent.nodeid = (u64) fid->raw[3] << 32;
 767         parent.nodeid |= (u64) fid->raw[4];
 768         parent.generation = fid->raw[5];
 769         return fuse_get_dentry(sb, &parent);
 770 }
 771
 772 static struct dentry *fuse_get_parent(struct dentry *child)
 773 {
 774         struct inode *child_inode = d_inode(child);
 775         struct fuse_conn *fc = get_fuse_conn(child_inode);
 776         struct inode *inode;
 777         struct dentry *parent;
 778         struct fuse_entry_out outarg;
 779         const struct qstr name = QSTR_INIT("..", 2);
 780         int err;
 781
 782         if (!fc->export_support)
 783                 return ERR_PTR(-ESTALE);
 784
 785         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
 786                                &name, &outarg, &inode);
 787         if (err) {
 788                 if (err == -ENOENT)
 789                         return ERR_PTR(-ESTALE);
 790                 return ERR_PTR(err);
 791         }
 792
 793         parent = d_obtain_alias(inode);
 794         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
 795                 fuse_invalidate_entry_cache(parent);
 796
 797         return parent;
 798 }
 799
 800 static const struct export_operations fuse_export_operations = {
 801         .fh_to_dentry   = fuse_fh_to_dentry,
 802         .fh_to_parent   = fuse_fh_to_parent,
 803         .encode_fh      = fuse_encode_fh,
 804         .get_parent     = fuse_get_parent,
 805 };
 806
 807 static const struct super_operations fuse_super_operations = {
 808         .alloc_inode    = fuse_alloc_inode,
 809         .free_inode     = fuse_free_inode,
 810         .evict_inode    = fuse_evict_inode,
 811         .write_inode    = fuse_write_inode,
 812         .drop_inode     = generic_delete_inode,
 813         .remount_fs     = fuse_remount_fs,
 814         .put_super      = fuse_put_super,
 815         .umount_begin   = fuse_umount_begin,
 816         .statfs         = fuse_statfs,
 817         .show_options   = fuse_show_options,
 818 };
 819
 820 static void sanitize_global_limit(unsigned *limit)
 821 {
 822         /*
 823          * The default maximum number of async requests is calculated to consume
 824          * 1/2^13 of the total memory, assuming 392 bytes per request.
 825          */
 826         if (*limit == 0)
 827                 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
 828
 829         if (*limit >= 1 << 16)
 830                 *limit = (1 << 16) - 1;
 831 }
 832
 833 static int set_global_limit(const char *val, const struct kernel_param *kp)
 834 {
 835         int rv;
 836
 837         rv = param_set_uint(val, kp);
 838         if (rv)
 839                 return rv;
 840
 841         sanitize_global_limit((unsigned *)kp->arg);
 842
 843         return 0;
 844 }
 845
 846 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
 847 {
 848         int cap_sys_admin = capable(CAP_SYS_ADMIN);
 849
 850         if (arg->minor < 13)
 851                 return;
 852
 853         sanitize_global_limit(&max_user_bgreq);
 854         sanitize_global_limit(&max_user_congthresh);
 855
 856         spin_lock(&fc->bg_lock);
 857         if (arg->max_background) {
 858                 fc->max_background = arg->max_background;
 859
 860                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
 861                         fc->max_background = max_user_bgreq;
 862         }
 863         if (arg->congestion_threshold) {
 864                 fc->congestion_threshold = arg->congestion_threshold;
 865
 866                 if (!cap_sys_admin &&
 867                     fc->congestion_threshold > max_user_congthresh)
 868                         fc->congestion_threshold = max_user_congthresh;
 869         }
 870         spin_unlock(&fc->bg_lock);
 871 }
 872
 873 struct fuse_init_args {
 874         struct fuse_args args;
 875         struct fuse_init_in in;
 876         struct fuse_init_out out;
 877 };
 878
 879 static void process_init_reply(struct fuse_conn *fc, struct fuse_args *args,
 880                                int error)
 881 {
 882         struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
 883         struct fuse_init_out *arg = &ia->out;
 884
 885         if (error || arg->major != FUSE_KERNEL_VERSION)
 886                 fc->conn_error = 1;
 887         else {
 888                 unsigned long ra_pages;
 889
 890                 process_init_limits(fc, arg);
 891
 892                 if (arg->minor >= 6) {
 893                         ra_pages = arg->max_readahead / PAGE_SIZE;
 894                         if (arg->flags & FUSE_ASYNC_READ)
 895                                 fc->async_read = 1;
 896                         if (!(arg->flags & FUSE_POSIX_LOCKS))
 897                                 fc->no_lock = 1;
 898                         if (arg->minor >= 17) {
 899                                 if (!(arg->flags & FUSE_FLOCK_LOCKS))
 900                                         fc->no_flock = 1;
 901                         } else {
 902                                 if (!(arg->flags & FUSE_POSIX_LOCKS))
 903                                         fc->no_flock = 1;
 904                         }
 905                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
 906                                 fc->atomic_o_trunc = 1;
 907                         if (arg->minor >= 9) {
 908                                 /* LOOKUP has dependency on proto version */
 909                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
 910                                         fc->export_support = 1;
 911                         }
 912                         if (arg->flags & FUSE_BIG_WRITES)
 913                                 fc->big_writes = 1;
 914                         if (arg->flags & FUSE_DONT_MASK)
 915                                 fc->dont_mask = 1;
 916                         if (arg->flags & FUSE_AUTO_INVAL_DATA)
 917                                 fc->auto_inval_data = 1;
 918                         else if (arg->flags & FUSE_EXPLICIT_INVAL_DATA)
 919                                 fc->explicit_inval_data = 1;
 920                         if (arg->flags & FUSE_DO_READDIRPLUS) {
 921                                 fc->do_readdirplus = 1;
 922                                 if (arg->flags & FUSE_READDIRPLUS_AUTO)
 923                                         fc->readdirplus_auto = 1;
 924                         }
 925                         if (arg->flags & FUSE_ASYNC_DIO)
 926                                 fc->async_dio = 1;
 927                         if (arg->flags & FUSE_WRITEBACK_CACHE)
 928                                 fc->writeback_cache = 1;
 929                         if (arg->flags & FUSE_PARALLEL_DIROPS)
 930                                 fc->parallel_dirops = 1;
 931                         if (arg->flags & FUSE_HANDLE_KILLPRIV)
 932                                 fc->handle_killpriv = 1;
 933                         if (arg->time_gran && arg->time_gran <= 1000000000)
 934                                 fc->sb->s_time_gran = arg->time_gran;
 935                         if ((arg->flags & FUSE_POSIX_ACL)) {
 936                                 fc->default_permissions = 1;
 937                                 fc->posix_acl = 1;
 938                                 fc->sb->s_xattr = fuse_acl_xattr_handlers;
 939                         }
 940                         if (arg->flags & FUSE_CACHE_SYMLINKS)
 941                                 fc->cache_symlinks = 1;
 942                         if (arg->flags & FUSE_ABORT_ERROR)
 943                                 fc->abort_err = 1;
 944                         if (arg->flags & FUSE_MAX_PAGES) {
 945                                 fc->max_pages =
 946                                         min_t(unsigned int, FUSE_MAX_MAX_PAGES,
 947                                         max_t(unsigned int, arg->max_pages, 1));
 948                         }
 949                 } else {
 950                         ra_pages = fc->max_read / PAGE_SIZE;
 951                         fc->no_lock = 1;
 952                         fc->no_flock = 1;
 953                 }
 954
 955                 fc->sb->s_bdi->ra_pages =
 956                                 min(fc->sb->s_bdi->ra_pages, ra_pages);
 957                 fc->minor = arg->minor;
 958                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
 959                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
 960                 fc->conn_init = 1;
 961         }
 962         kfree(ia);
 963
 964         fuse_set_initialized(fc);
 965         wake_up_all(&fc->blocked_waitq);
 966 }
 967
 968 void fuse_send_init(struct fuse_conn *fc)
 969 {
 970         struct fuse_init_args *ia;
 971
 972         ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
 973
 974         ia->in.major = FUSE_KERNEL_VERSION;
 975         ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
 976         ia->in.max_readahead = fc->sb->s_bdi->ra_pages * PAGE_SIZE;
 977         ia->in.flags |=
 978                 FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
 979                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
 980                 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
 981                 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
 982                 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
 983                 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
 984                 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
 985                 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
 986                 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA;
 987         ia->args.opcode = FUSE_INIT;
 988         ia->args.in_numargs = 1;
 989         ia->args.in_args[0].size = sizeof(ia->in);
 990         ia->args.in_args[0].value = &ia->in;
 991         ia->args.out_numargs = 1;
 992         /* Variable length argument used for backward compatibility
 993            with interface version < 7.5.  Rest of init_out is zeroed
 994            by do_get_request(), so a short reply is not a problem */
 995         ia->args.out_argvar = true;
 996         ia->args.out_args[0].size = sizeof(ia->out);
 997         ia->args.out_args[0].value = &ia->out;
 998         ia->args.force = true;
 999         ia->args.nocreds = true;
1000         ia->args.end = process_init_reply;
1001
1002         if (fuse_simple_background(fc, &ia->args, GFP_KERNEL) != 0)
1003                 process_init_reply(fc, &ia->args, -ENOTCONN);
1004 }
1005 EXPORT_SYMBOL_GPL(fuse_send_init);
1006
1007 void fuse_free_conn(struct fuse_conn *fc)
1008 {
1009         WARN_ON(!list_empty(&fc->devices));
1010         kfree_rcu(fc, rcu);
1011 }
1012 EXPORT_SYMBOL_GPL(fuse_free_conn);
1013
1014 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1015 {
1016         int err;
1017         char *suffix = "";
1018
1019         if (sb->s_bdev) {
1020                 suffix = "-fuseblk";
1021                 /*
1022                  * sb->s_bdi points to blkdev's bdi however we want to redirect
1023                  * it to our private bdi...
1024                  */
1025                 bdi_put(sb->s_bdi);
1026                 sb->s_bdi = &noop_backing_dev_info;
1027         }
1028         err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1029                                    MINOR(fc->dev), suffix);
1030         if (err)
1031                 return err;
1032
1033         sb->s_bdi->ra_pages = VM_READAHEAD_PAGES;
1034         /* fuse does it's own writeback accounting */
1035         sb->s_bdi->capabilities = BDI_CAP_NO_ACCT_WB | BDI_CAP_STRICTLIMIT;
1036
1037         /*
1038          * For a single fuse filesystem use max 1% of dirty +
1039          * writeback threshold.
1040          *
1041          * This gives about 1M of write buffer for memory maps on a
1042          * machine with 1G and 10% dirty_ratio, which should be more
1043          * than enough.
1044          *
1045          * Privileged users can raise it by writing to
1046          *
1047          *    /sys/class/bdi/<bdi>/max_ratio
1048          */
1049         bdi_set_max_ratio(sb->s_bdi, 1);
1050
1051         return 0;
1052 }
1053
1054 struct fuse_dev *fuse_dev_alloc(void)
1055 {
1056         struct fuse_dev *fud;
1057         struct list_head *pq;
1058
1059         fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1060         if (!fud)
1061                 return NULL;
1062
1063         pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1064         if (!pq) {
1065                 kfree(fud);
1066                 return NULL;
1067         }
1068
1069         fud->pq.processing = pq;
1070         fuse_pqueue_init(&fud->pq);
1071
1072         return fud;
1073 }
1074 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1075
1076 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1077 {
1078         fud->fc = fuse_conn_get(fc);
1079         spin_lock(&fc->lock);
1080         list_add_tail(&fud->entry, &fc->devices);
1081         spin_unlock(&fc->lock);
1082 }
1083 EXPORT_SYMBOL_GPL(fuse_dev_install);
1084
1085 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1086 {
1087         struct fuse_dev *fud;
1088
1089         fud = fuse_dev_alloc();
1090         if (!fud)
1091                 return NULL;
1092
1093         fuse_dev_install(fud, fc);
1094         return fud;
1095 }
1096 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1097
1098 void fuse_dev_free(struct fuse_dev *fud)
1099 {
1100         struct fuse_conn *fc = fud->fc;
1101
1102         if (fc) {
1103                 spin_lock(&fc->lock);
1104                 list_del(&fud->entry);
1105                 spin_unlock(&fc->lock);
1106
1107                 fuse_conn_put(fc);
1108         }
1109         kfree(fud->pq.processing);
1110         kfree(fud);
1111 }
1112 EXPORT_SYMBOL_GPL(fuse_dev_free);
1113
1114 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1115 {
1116         struct fuse_dev *fud;
1117         struct fuse_conn *fc = get_fuse_conn_super(sb);
1118         struct inode *root;
1119         struct dentry *root_dentry;
1120         int err;
1121
1122         err = -EINVAL;
1123         if (sb->s_flags & SB_MANDLOCK)
1124                 goto err;
1125
1126         sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1127
1128         if (ctx->is_bdev) {
1129 #ifdef CONFIG_BLOCK
1130                 err = -EINVAL;
1131                 if (!sb_set_blocksize(sb, ctx->blksize))
1132                         goto err;
1133 #endif
1134         } else {
1135                 sb->s_blocksize = PAGE_SIZE;
1136                 sb->s_blocksize_bits = PAGE_SHIFT;
1137         }
1138
1139         sb->s_subtype = ctx->subtype;
1140         ctx->subtype = NULL;
1141         sb->s_magic = FUSE_SUPER_MAGIC;
1142         sb->s_op = &fuse_super_operations;
1143         sb->s_xattr = fuse_xattr_handlers;
1144         sb->s_maxbytes = MAX_LFS_FILESIZE;
1145         sb->s_time_gran = 1;
1146         sb->s_export_op = &fuse_export_operations;
1147         sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1148         if (sb->s_user_ns != &init_user_ns)
1149                 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1150
1151         /*
1152          * If we are not in the initial user namespace posix
1153          * acls must be translated.
1154          */
1155         if (sb->s_user_ns != &init_user_ns)
1156                 sb->s_xattr = fuse_no_acl_xattr_handlers;
1157
1158         fud = fuse_dev_alloc_install(fc);
1159         if (!fud)
1160                 goto err;
1161
1162         fc->dev = sb->s_dev;
1163         fc->sb = sb;
1164         err = fuse_bdi_init(fc, sb);
1165         if (err)
1166                 goto err_dev_free;
1167
1168         /* Handle umasking inside the fuse code */
1169         if (sb->s_flags & SB_POSIXACL)
1170                 fc->dont_mask = 1;
1171         sb->s_flags |= SB_POSIXACL;
1172
1173         fc->default_permissions = ctx->default_permissions;
1174         fc->allow_other = ctx->allow_other;
1175         fc->user_id = ctx->user_id;
1176         fc->group_id = ctx->group_id;
1177         fc->max_read = max_t(unsigned, 4096, ctx->max_read);
1178         fc->destroy = ctx->destroy;
1179         fc->no_control = ctx->no_control;
1180         fc->no_force_umount = ctx->no_force_umount;
1181         fc->no_mount_options = ctx->no_mount_options;
1182
1183         err = -ENOMEM;
1184         root = fuse_get_root_inode(sb, ctx->rootmode);
1185         sb->s_d_op = &fuse_root_dentry_operations;
1186         root_dentry = d_make_root(root);
1187         if (!root_dentry)
1188                 goto err_dev_free;
1189         /* Root dentry doesn't have .d_revalidate */
1190         sb->s_d_op = &fuse_dentry_operations;
1191
1192         mutex_lock(&fuse_mutex);
1193         err = -EINVAL;
1194         if (*ctx->fudptr)
1195                 goto err_unlock;
1196
1197         err = fuse_ctl_add_conn(fc);
1198         if (err)
1199                 goto err_unlock;
1200
1201         list_add_tail(&fc->entry, &fuse_conn_list);
1202         sb->s_root = root_dentry;
1203         *ctx->fudptr = fud;
1204         mutex_unlock(&fuse_mutex);
1205         return 0;
1206
1207  err_unlock:
1208         mutex_unlock(&fuse_mutex);
1209         dput(root_dentry);
1210  err_dev_free:
1211         fuse_dev_free(fud);
1212  err:
1213         return err;
1214 }
1215 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1216
1217 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1218 {
1219         struct fuse_fs_context *ctx = fsc->fs_private;
1220         struct file *file;
1221         int err;
1222         struct fuse_conn *fc;
1223
1224         err = -EINVAL;
1225         file = fget(ctx->fd);
1226         if (!file)
1227                 goto err;
1228
1229         /*
1230          * Require mount to happen from the same user namespace which
1231          * opened /dev/fuse to prevent potential attacks.
1232          */
1233         if ((file->f_op != &fuse_dev_operations) ||
1234             (file->f_cred->user_ns != sb->s_user_ns))
1235                 goto err_fput;
1236         ctx->fudptr = &file->private_data;
1237
1238         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1239         err = -ENOMEM;
1240         if (!fc)
1241                 goto err_fput;
1242
1243         fuse_conn_init(fc, sb->s_user_ns, &fuse_dev_fiq_ops, NULL);
1244         fc->release = fuse_free_conn;
1245         sb->s_fs_info = fc;
1246
1247         err = fuse_fill_super_common(sb, ctx);
1248         if (err)
1249                 goto err_put_conn;
1250         /*
1251          * atomic_dec_and_test() in fput() provides the necessary
1252          * memory barrier for file->private_data to be visible on all
1253          * CPUs after this
1254          */
1255         fput(file);
1256         fuse_send_init(get_fuse_conn_super(sb));
1257         return 0;
1258
1259  err_put_conn:
1260         fuse_conn_put(fc);
1261         sb->s_fs_info = NULL;
1262  err_fput:
1263         fput(file);
1264  err:
1265         return err;
1266 }
1267
1268 static int fuse_get_tree(struct fs_context *fc)
1269 {
1270         struct fuse_fs_context *ctx = fc->fs_private;
1271
1272         if (!ctx->fd_present || !ctx->rootmode_present ||
1273             !ctx->user_id_present || !ctx->group_id_present)
1274                 return -EINVAL;
1275
1276 #ifdef CONFIG_BLOCK
1277         if (ctx->is_bdev)
1278                 return get_tree_bdev(fc, fuse_fill_super);
1279 #endif
1280
1281         return get_tree_nodev(fc, fuse_fill_super);
1282 }
1283
1284 static const struct fs_context_operations fuse_context_ops = {
1285         .free           = fuse_free_fc,
1286         .parse_param    = fuse_parse_param,
1287         .get_tree       = fuse_get_tree,
1288 };
1289
1290 /*
1291  * Set up the filesystem mount context.
1292  */
1293 static int fuse_init_fs_context(struct fs_context *fc)
1294 {
1295         struct fuse_fs_context *ctx;
1296
1297         ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1298         if (!ctx)
1299                 return -ENOMEM;
1300
1301         ctx->max_read = ~0;
1302         ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1303
1304 #ifdef CONFIG_BLOCK
1305         if (fc->fs_type == &fuseblk_fs_type) {
1306                 ctx->is_bdev = true;
1307                 ctx->destroy = true;
1308         }
1309 #endif
1310
1311         fc->fs_private = ctx;
1312         fc->ops = &fuse_context_ops;
1313         return 0;
1314 }
1315
1316 static void fuse_sb_destroy(struct super_block *sb)
1317 {
1318         struct fuse_conn *fc = get_fuse_conn_super(sb);
1319
1320         if (fc) {
1321                 if (fc->destroy)
1322                         fuse_send_destroy(fc);
1323
1324                 fuse_abort_conn(fc);
1325                 fuse_wait_aborted(fc);
1326
1327                 down_write(&fc->killsb);
1328                 fc->sb = NULL;
1329                 up_write(&fc->killsb);
1330         }
1331 }
1332
1333 void fuse_kill_sb_anon(struct super_block *sb)
1334 {
1335         fuse_sb_destroy(sb);
1336         kill_anon_super(sb);
1337 }
1338 EXPORT_SYMBOL_GPL(fuse_kill_sb_anon);
1339
1340 static struct file_system_type fuse_fs_type = {
1341         .owner          = THIS_MODULE,
1342         .name           = "fuse",
1343         .fs_flags       = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1344         .init_fs_context = fuse_init_fs_context,
1345         .parameters     = fuse_fs_parameters,
1346         .kill_sb        = fuse_kill_sb_anon,
1347 };
1348 MODULE_ALIAS_FS("fuse");
1349
1350 #ifdef CONFIG_BLOCK
1351 static void fuse_kill_sb_blk(struct super_block *sb)
1352 {
1353         fuse_sb_destroy(sb);
1354         kill_block_super(sb);
1355 }
1356
1357 static struct file_system_type fuseblk_fs_type = {
1358         .owner          = THIS_MODULE,
1359         .name           = "fuseblk",
1360         .init_fs_context = fuse_init_fs_context,
1361         .parameters     = fuse_fs_parameters,
1362         .kill_sb        = fuse_kill_sb_blk,
1363         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1364 };
1365 MODULE_ALIAS_FS("fuseblk");
1366
1367 static inline int register_fuseblk(void)
1368 {
1369         return register_filesystem(&fuseblk_fs_type);
1370 }
1371
1372 static inline void unregister_fuseblk(void)
1373 {
1374         unregister_filesystem(&fuseblk_fs_type);
1375 }
1376 #else
1377 static inline int register_fuseblk(void)
1378 {
1379         return 0;
1380 }
1381
1382 static inline void unregister_fuseblk(void)
1383 {
1384 }
1385 #endif
1386
1387 static void fuse_inode_init_once(void *foo)
1388 {
1389         struct inode *inode = foo;
1390
1391         inode_init_once(inode);
1392 }
1393
1394 static int __init fuse_fs_init(void)
1395 {
1396         int err;
1397
1398         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1399                         sizeof(struct fuse_inode), 0,
1400                         SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1401                         fuse_inode_init_once);
1402         err = -ENOMEM;
1403         if (!fuse_inode_cachep)
1404                 goto out;
1405
1406         err = register_fuseblk();
1407         if (err)
1408                 goto out2;
1409
1410         err = register_filesystem(&fuse_fs_type);
1411         if (err)
1412                 goto out3;
1413
1414         return 0;
1415
1416  out3:
1417         unregister_fuseblk();
1418  out2:
1419         kmem_cache_destroy(fuse_inode_cachep);
1420  out:
1421         return err;
1422 }
1423
1424 static void fuse_fs_cleanup(void)
1425 {
1426         unregister_filesystem(&fuse_fs_type);
1427         unregister_fuseblk();
1428
1429         /*
1430          * Make sure all delayed rcu free inodes are flushed before we
1431          * destroy cache.
1432          */
1433         rcu_barrier();
1434         kmem_cache_destroy(fuse_inode_cachep);
1435 }
1436
1437 static struct kobject *fuse_kobj;
1438
1439 static int fuse_sysfs_init(void)
1440 {
1441         int err;
1442
1443         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1444         if (!fuse_kobj) {
1445                 err = -ENOMEM;
1446                 goto out_err;
1447         }
1448
1449         err = sysfs_create_mount_point(fuse_kobj, "connections");
1450         if (err)
1451                 goto out_fuse_unregister;
1452
1453         return 0;
1454
1455  out_fuse_unregister:
1456         kobject_put(fuse_kobj);
1457  out_err:
1458         return err;
1459 }
1460
1461 static void fuse_sysfs_cleanup(void)
1462 {
1463         sysfs_remove_mount_point(fuse_kobj, "connections");
1464         kobject_put(fuse_kobj);
1465 }
1466
1467 static int __init fuse_init(void)
1468 {
1469         int res;
1470
1471         pr_info("init (API version %i.%i)\n",
1472                 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1473
1474         INIT_LIST_HEAD(&fuse_conn_list);
1475         res = fuse_fs_init();
1476         if (res)
1477                 goto err;
1478
1479         res = fuse_dev_init();
1480         if (res)
1481                 goto err_fs_cleanup;
1482
1483         res = fuse_sysfs_init();
1484         if (res)
1485                 goto err_dev_cleanup;
1486
1487         res = fuse_ctl_init();
1488         if (res)
1489                 goto err_sysfs_cleanup;
1490
1491         sanitize_global_limit(&max_user_bgreq);
1492         sanitize_global_limit(&max_user_congthresh);
1493
1494         return 0;
1495
1496  err_sysfs_cleanup:
1497         fuse_sysfs_cleanup();
1498  err_dev_cleanup:
1499         fuse_dev_cleanup();
1500  err_fs_cleanup:
1501         fuse_fs_cleanup();
1502  err:
1503         return res;
1504 }
1505
1506 static void __exit fuse_exit(void)
1507 {
1508         pr_debug("exit\n");
1509
1510         fuse_ctl_cleanup();
1511         fuse_sysfs_cleanup();
1512         fuse_fs_cleanup();
1513         fuse_dev_cleanup();
1514 }
1515
1516 module_init(fuse_init);
1517 module_exit(fuse_exit);