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