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 #include <uapi/linux/magic.h>
  27
  28 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  29 MODULE_DESCRIPTION("Filesystem in Userspace");
  30 MODULE_LICENSE("GPL");
  31
  32 static struct kmem_cache *fuse_inode_cachep;
  33 struct list_head fuse_conn_list;
  34 DEFINE_MUTEX(fuse_mutex);
  35
  36 static int set_global_limit(const char *val, const struct kernel_param *kp);
  37
  38 unsigned max_user_bgreq;
  39 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
  40                   &max_user_bgreq, 0644);
  41 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
  42 MODULE_PARM_DESC(max_user_bgreq,
  43  "Global limit for the maximum number of backgrounded requests an "
  44  "unprivileged user can set");
  45
  46 unsigned max_user_congthresh;
  47 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
  48                   &max_user_congthresh, 0644);
  49 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
  50 MODULE_PARM_DESC(max_user_congthresh,
  51  "Global limit for the maximum congestion threshold an "
  52  "unprivileged user can set");
  53
  54 #define FUSE_DEFAULT_BLKSIZE 512
  55
  56 /** Maximum number of outstanding background requests */
  57 #define FUSE_DEFAULT_MAX_BACKGROUND 12
  58
  59 /** Congestion starts at 75% of maximum */
  60 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
  61
  62 #ifdef CONFIG_BLOCK
  63 static struct file_system_type fuseblk_fs_type;
  64 #endif
  65
  66 struct fuse_forget_link *fuse_alloc_forget(void)
  67 {
  68         return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL_ACCOUNT);
  69 }
  70
  71 static struct inode *fuse_alloc_inode(struct super_block *sb)
  72 {
  73         struct fuse_inode *fi;
  74
  75         fi = alloc_inode_sb(sb, fuse_inode_cachep, GFP_KERNEL);
  76         if (!fi)
  77                 return NULL;
  78
  79         fi->i_time = 0;
  80         fi->inval_mask = 0;
  81         fi->nodeid = 0;
  82         fi->nlookup = 0;
  83         fi->attr_version = 0;
  84         fi->orig_ino = 0;
  85         fi->state = 0;
  86         mutex_init(&fi->mutex);
  87         spin_lock_init(&fi->lock);
  88         fi->forget = fuse_alloc_forget();
  89         if (!fi->forget)
  90                 goto out_free;
  91
  92         if (IS_ENABLED(CONFIG_FUSE_DAX) && !fuse_dax_inode_alloc(sb, fi))
  93                 goto out_free_forget;
  94
  95         return &fi->inode;
  96
  97 out_free_forget:
  98         kfree(fi->forget);
  99 out_free:
 100         kmem_cache_free(fuse_inode_cachep, fi);
 101         return NULL;
 102 }
 103
 104 static void fuse_free_inode(struct inode *inode)
 105 {
 106         struct fuse_inode *fi = get_fuse_inode(inode);
 107
 108         mutex_destroy(&fi->mutex);
 109         kfree(fi->forget);
 110 #ifdef CONFIG_FUSE_DAX
 111         kfree(fi->dax);
 112 #endif
 113         kmem_cache_free(fuse_inode_cachep, fi);
 114 }
 115
 116 static void fuse_evict_inode(struct inode *inode)
 117 {
 118         struct fuse_inode *fi = get_fuse_inode(inode);
 119
 120         /* Will write inode on close/munmap and in all other dirtiers */
 121         WARN_ON(inode->i_state & I_DIRTY_INODE);
 122
 123         truncate_inode_pages_final(&inode->i_data);
 124         clear_inode(inode);
 125         if (inode->i_sb->s_flags & SB_ACTIVE) {
 126                 struct fuse_conn *fc = get_fuse_conn(inode);
 127
 128                 if (FUSE_IS_DAX(inode))
 129                         fuse_dax_inode_cleanup(inode);
 130                 if (fi->nlookup) {
 131                         fuse_queue_forget(fc, fi->forget, fi->nodeid,
 132                                           fi->nlookup);
 133                         fi->forget = NULL;
 134                 }
 135         }
 136         if (S_ISREG(inode->i_mode) && !fuse_is_bad(inode)) {
 137                 WARN_ON(!list_empty(&fi->write_files));
 138                 WARN_ON(!list_empty(&fi->queued_writes));
 139         }
 140 }
 141
 142 static int fuse_reconfigure(struct fs_context *fsc)
 143 {
 144         struct super_block *sb = fsc->root->d_sb;
 145
 146         sync_filesystem(sb);
 147         if (fsc->sb_flags & SB_MANDLOCK)
 148                 return -EINVAL;
 149
 150         return 0;
 151 }
 152
 153 /*
 154  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
 155  * so that it will fit.
 156  */
 157 static ino_t fuse_squash_ino(u64 ino64)
 158 {
 159         ino_t ino = (ino_t) ino64;
 160         if (sizeof(ino_t) < sizeof(u64))
 161                 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
 162         return ino;
 163 }
 164
 165 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
 166                                    u64 attr_valid, u32 cache_mask)
 167 {
 168         struct fuse_conn *fc = get_fuse_conn(inode);
 169         struct fuse_inode *fi = get_fuse_inode(inode);
 170
 171         lockdep_assert_held(&fi->lock);
 172
 173         fi->attr_version = atomic64_inc_return(&fc->attr_version);
 174         fi->i_time = attr_valid;
 175         WRITE_ONCE(fi->inval_mask, 0);
 176
 177         inode->i_ino     = fuse_squash_ino(attr->ino);
 178         inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
 179         set_nlink(inode, attr->nlink);
 180         inode->i_uid     = make_kuid(fc->user_ns, attr->uid);
 181         inode->i_gid     = make_kgid(fc->user_ns, attr->gid);
 182         inode->i_blocks  = attr->blocks;
 183
 184         /* Sanitize nsecs */
 185         attr->atimensec = min_t(u32, attr->atimensec, NSEC_PER_SEC - 1);
 186         attr->mtimensec = min_t(u32, attr->mtimensec, NSEC_PER_SEC - 1);
 187         attr->ctimensec = min_t(u32, attr->ctimensec, NSEC_PER_SEC - 1);
 188
 189         inode->i_atime.tv_sec   = attr->atime;
 190         inode->i_atime.tv_nsec  = attr->atimensec;
 191         /* mtime from server may be stale due to local buffered write */
 192         if (!(cache_mask & STATX_MTIME)) {
 193                 inode->i_mtime.tv_sec   = attr->mtime;
 194                 inode->i_mtime.tv_nsec  = attr->mtimensec;
 195         }
 196         if (!(cache_mask & STATX_CTIME)) {
 197                 inode_set_ctime(inode, attr->ctime, attr->ctimensec);
 198         }
 199
 200         if (attr->blksize != 0)
 201                 inode->i_blkbits = ilog2(attr->blksize);
 202         else
 203                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
 204
 205         /*
 206          * Don't set the sticky bit in i_mode, unless we want the VFS
 207          * to check permissions.  This prevents failures due to the
 208          * check in may_delete().
 209          */
 210         fi->orig_i_mode = inode->i_mode;
 211         if (!fc->default_permissions)
 212                 inode->i_mode &= ~S_ISVTX;
 213
 214         fi->orig_ino = attr->ino;
 215
 216         /*
 217          * We are refreshing inode data and it is possible that another
 218          * client set suid/sgid or security.capability xattr. So clear
 219          * S_NOSEC. Ideally, we could have cleared it only if suid/sgid
 220          * was set or if security.capability xattr was set. But we don't
 221          * know if security.capability has been set or not. So clear it
 222          * anyway. Its less efficient but should be safe.
 223          */
 224         inode->i_flags &= ~S_NOSEC;
 225 }
 226
 227 u32 fuse_get_cache_mask(struct inode *inode)
 228 {
 229         struct fuse_conn *fc = get_fuse_conn(inode);
 230
 231         if (!fc->writeback_cache || !S_ISREG(inode->i_mode))
 232                 return 0;
 233
 234         return STATX_MTIME | STATX_CTIME | STATX_SIZE;
 235 }
 236
 237 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
 238                             u64 attr_valid, u64 attr_version)
 239 {
 240         struct fuse_conn *fc = get_fuse_conn(inode);
 241         struct fuse_inode *fi = get_fuse_inode(inode);
 242         u32 cache_mask;
 243         loff_t oldsize;
 244         struct timespec64 old_mtime;
 245
 246         spin_lock(&fi->lock);
 247         /*
 248          * In case of writeback_cache enabled, writes update mtime, ctime and
 249          * may update i_size.  In these cases trust the cached value in the
 250          * inode.
 251          */
 252         cache_mask = fuse_get_cache_mask(inode);
 253         if (cache_mask & STATX_SIZE)
 254                 attr->size = i_size_read(inode);
 255
 256         if (cache_mask & STATX_MTIME) {
 257                 attr->mtime = inode->i_mtime.tv_sec;
 258                 attr->mtimensec = inode->i_mtime.tv_nsec;
 259         }
 260         if (cache_mask & STATX_CTIME) {
 261                 attr->ctime = inode_get_ctime(inode).tv_sec;
 262                 attr->ctimensec = inode_get_ctime(inode).tv_nsec;
 263         }
 264
 265         if ((attr_version != 0 && fi->attr_version > attr_version) ||
 266             test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
 267                 spin_unlock(&fi->lock);
 268                 return;
 269         }
 270
 271         old_mtime = inode->i_mtime;
 272         fuse_change_attributes_common(inode, attr, attr_valid, cache_mask);
 273
 274         oldsize = inode->i_size;
 275         /*
 276          * In case of writeback_cache enabled, the cached writes beyond EOF
 277          * extend local i_size without keeping userspace server in sync. So,
 278          * attr->size coming from server can be stale. We cannot trust it.
 279          */
 280         if (!(cache_mask & STATX_SIZE))
 281                 i_size_write(inode, attr->size);
 282         spin_unlock(&fi->lock);
 283
 284         if (!cache_mask && S_ISREG(inode->i_mode)) {
 285                 bool inval = false;
 286
 287                 if (oldsize != attr->size) {
 288                         truncate_pagecache(inode, attr->size);
 289                         if (!fc->explicit_inval_data)
 290                                 inval = true;
 291                 } else if (fc->auto_inval_data) {
 292                         struct timespec64 new_mtime = {
 293                                 .tv_sec = attr->mtime,
 294                                 .tv_nsec = attr->mtimensec,
 295                         };
 296
 297                         /*
 298                          * Auto inval mode also checks and invalidates if mtime
 299                          * has changed.
 300                          */
 301                         if (!timespec64_equal(&old_mtime, &new_mtime))
 302                                 inval = true;
 303                 }
 304
 305                 if (inval)
 306                         invalidate_inode_pages2(inode->i_mapping);
 307         }
 308
 309         if (IS_ENABLED(CONFIG_FUSE_DAX))
 310                 fuse_dax_dontcache(inode, attr->flags);
 311 }
 312
 313 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr,
 314                             struct fuse_conn *fc)
 315 {
 316         inode->i_mode = attr->mode & S_IFMT;
 317         inode->i_size = attr->size;
 318         inode->i_mtime.tv_sec  = attr->mtime;
 319         inode->i_mtime.tv_nsec = attr->mtimensec;
 320         inode_set_ctime(inode, attr->ctime, attr->ctimensec);
 321         if (S_ISREG(inode->i_mode)) {
 322                 fuse_init_common(inode);
 323                 fuse_init_file_inode(inode, attr->flags);
 324         } else if (S_ISDIR(inode->i_mode))
 325                 fuse_init_dir(inode);
 326         else if (S_ISLNK(inode->i_mode))
 327                 fuse_init_symlink(inode);
 328         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
 329                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
 330                 fuse_init_common(inode);
 331                 init_special_inode(inode, inode->i_mode,
 332                                    new_decode_dev(attr->rdev));
 333         } else
 334                 BUG();
 335         /*
 336          * Ensure that we don't cache acls for daemons without FUSE_POSIX_ACL
 337          * so they see the exact same behavior as before.
 338          */
 339         if (!fc->posix_acl)
 340                 inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
 341 }
 342
 343 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
 344 {
 345         u64 nodeid = *(u64 *) _nodeidp;
 346         if (get_node_id(inode) == nodeid)
 347                 return 1;
 348         else
 349                 return 0;
 350 }
 351
 352 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
 353 {
 354         u64 nodeid = *(u64 *) _nodeidp;
 355         get_fuse_inode(inode)->nodeid = nodeid;
 356         return 0;
 357 }
 358
 359 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
 360                         int generation, struct fuse_attr *attr,
 361                         u64 attr_valid, u64 attr_version)
 362 {
 363         struct inode *inode;
 364         struct fuse_inode *fi;
 365         struct fuse_conn *fc = get_fuse_conn_super(sb);
 366
 367         /*
 368          * Auto mount points get their node id from the submount root, which is
 369          * not a unique identifier within this filesystem.
 370          *
 371          * To avoid conflicts, do not place submount points into the inode hash
 372          * table.
 373          */
 374         if (fc->auto_submounts && (attr->flags & FUSE_ATTR_SUBMOUNT) &&
 375             S_ISDIR(attr->mode)) {
 376                 inode = new_inode(sb);
 377                 if (!inode)
 378                         return NULL;
 379
 380                 fuse_init_inode(inode, attr, fc);
 381                 get_fuse_inode(inode)->nodeid = nodeid;
 382                 inode->i_flags |= S_AUTOMOUNT;
 383                 goto done;
 384         }
 385
 386 retry:
 387         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
 388         if (!inode)
 389                 return NULL;
 390
 391         if ((inode->i_state & I_NEW)) {
 392                 inode->i_flags |= S_NOATIME;
 393                 if (!fc->writeback_cache || !S_ISREG(attr->mode))
 394                         inode->i_flags |= S_NOCMTIME;
 395                 inode->i_generation = generation;
 396                 fuse_init_inode(inode, attr, fc);
 397                 unlock_new_inode(inode);
 398         } else if (fuse_stale_inode(inode, generation, attr)) {
 399                 /* nodeid was reused, any I/O on the old inode should fail */
 400                 fuse_make_bad(inode);
 401                 iput(inode);
 402                 goto retry;
 403         }
 404 done:
 405         fi = get_fuse_inode(inode);
 406         spin_lock(&fi->lock);
 407         fi->nlookup++;
 408         spin_unlock(&fi->lock);
 409         fuse_change_attributes(inode, attr, attr_valid, attr_version);
 410
 411         return inode;
 412 }
 413
 414 struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid,
 415                            struct fuse_mount **fm)
 416 {
 417         struct fuse_mount *fm_iter;
 418         struct inode *inode;
 419
 420         WARN_ON(!rwsem_is_locked(&fc->killsb));
 421         list_for_each_entry(fm_iter, &fc->mounts, fc_entry) {
 422                 if (!fm_iter->sb)
 423                         continue;
 424
 425                 inode = ilookup5(fm_iter->sb, nodeid, fuse_inode_eq, &nodeid);
 426                 if (inode) {
 427                         if (fm)
 428                                 *fm = fm_iter;
 429                         return inode;
 430                 }
 431         }
 432
 433         return NULL;
 434 }
 435
 436 int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
 437                              loff_t offset, loff_t len)
 438 {
 439         struct fuse_inode *fi;
 440         struct inode *inode;
 441         pgoff_t pg_start;
 442         pgoff_t pg_end;
 443
 444         inode = fuse_ilookup(fc, nodeid, NULL);
 445         if (!inode)
 446                 return -ENOENT;
 447
 448         fi = get_fuse_inode(inode);
 449         spin_lock(&fi->lock);
 450         fi->attr_version = atomic64_inc_return(&fc->attr_version);
 451         spin_unlock(&fi->lock);
 452
 453         fuse_invalidate_attr(inode);
 454         forget_all_cached_acls(inode);
 455         if (offset >= 0) {
 456                 pg_start = offset >> PAGE_SHIFT;
 457                 if (len <= 0)
 458                         pg_end = -1;
 459                 else
 460                         pg_end = (offset + len - 1) >> PAGE_SHIFT;
 461                 invalidate_inode_pages2_range(inode->i_mapping,
 462                                               pg_start, pg_end);
 463         }
 464         iput(inode);
 465         return 0;
 466 }
 467
 468 bool fuse_lock_inode(struct inode *inode)
 469 {
 470         bool locked = false;
 471
 472         if (!get_fuse_conn(inode)->parallel_dirops) {
 473                 mutex_lock(&get_fuse_inode(inode)->mutex);
 474                 locked = true;
 475         }
 476
 477         return locked;
 478 }
 479
 480 void fuse_unlock_inode(struct inode *inode, bool locked)
 481 {
 482         if (locked)
 483                 mutex_unlock(&get_fuse_inode(inode)->mutex);
 484 }
 485
 486 static void fuse_umount_begin(struct super_block *sb)
 487 {
 488         struct fuse_conn *fc = get_fuse_conn_super(sb);
 489
 490         if (fc->no_force_umount)
 491                 return;
 492
 493         fuse_abort_conn(fc);
 494
 495         // Only retire block-device-based superblocks.
 496         if (sb->s_bdev != NULL)
 497                 retire_super(sb);
 498 }
 499
 500 static void fuse_send_destroy(struct fuse_mount *fm)
 501 {
 502         if (fm->fc->conn_init) {
 503                 FUSE_ARGS(args);
 504
 505                 args.opcode = FUSE_DESTROY;
 506                 args.force = true;
 507                 args.nocreds = true;
 508                 fuse_simple_request(fm, &args);
 509         }
 510 }
 511
 512 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
 513 {
 514         stbuf->f_type    = FUSE_SUPER_MAGIC;
 515         stbuf->f_bsize   = attr->bsize;
 516         stbuf->f_frsize  = attr->frsize;
 517         stbuf->f_blocks  = attr->blocks;
 518         stbuf->f_bfree   = attr->bfree;
 519         stbuf->f_bavail  = attr->bavail;
 520         stbuf->f_files   = attr->files;
 521         stbuf->f_ffree   = attr->ffree;
 522         stbuf->f_namelen = attr->namelen;
 523         /* fsid is left zero */
 524 }
 525
 526 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
 527 {
 528         struct super_block *sb = dentry->d_sb;
 529         struct fuse_mount *fm = get_fuse_mount_super(sb);
 530         FUSE_ARGS(args);
 531         struct fuse_statfs_out outarg;
 532         int err;
 533
 534         if (!fuse_allow_current_process(fm->fc)) {
 535                 buf->f_type = FUSE_SUPER_MAGIC;
 536                 return 0;
 537         }
 538
 539         memset(&outarg, 0, sizeof(outarg));
 540         args.in_numargs = 0;
 541         args.opcode = FUSE_STATFS;
 542         args.nodeid = get_node_id(d_inode(dentry));
 543         args.out_numargs = 1;
 544         args.out_args[0].size = sizeof(outarg);
 545         args.out_args[0].value = &outarg;
 546         err = fuse_simple_request(fm, &args);
 547         if (!err)
 548                 convert_fuse_statfs(buf, &outarg.st);
 549         return err;
 550 }
 551
 552 static struct fuse_sync_bucket *fuse_sync_bucket_alloc(void)
 553 {
 554         struct fuse_sync_bucket *bucket;
 555
 556         bucket = kzalloc(sizeof(*bucket), GFP_KERNEL | __GFP_NOFAIL);
 557         if (bucket) {
 558                 init_waitqueue_head(&bucket->waitq);
 559                 /* Initial active count */
 560                 atomic_set(&bucket->count, 1);
 561         }
 562         return bucket;
 563 }
 564
 565 static void fuse_sync_fs_writes(struct fuse_conn *fc)
 566 {
 567         struct fuse_sync_bucket *bucket, *new_bucket;
 568         int count;
 569
 570         new_bucket = fuse_sync_bucket_alloc();
 571         spin_lock(&fc->lock);
 572         bucket = rcu_dereference_protected(fc->curr_bucket, 1);
 573         count = atomic_read(&bucket->count);
 574         WARN_ON(count < 1);
 575         /* No outstanding writes? */
 576         if (count == 1) {
 577                 spin_unlock(&fc->lock);
 578                 kfree(new_bucket);
 579                 return;
 580         }
 581
 582         /*
 583          * Completion of new bucket depends on completion of this bucket, so add
 584          * one more count.
 585          */
 586         atomic_inc(&new_bucket->count);
 587         rcu_assign_pointer(fc->curr_bucket, new_bucket);
 588         spin_unlock(&fc->lock);
 589         /*
 590          * Drop initial active count.  At this point if all writes in this and
 591          * ancestor buckets complete, the count will go to zero and this task
 592          * will be woken up.
 593          */
 594         atomic_dec(&bucket->count);
 595
 596         wait_event(bucket->waitq, atomic_read(&bucket->count) == 0);
 597
 598         /* Drop temp count on descendant bucket */
 599         fuse_sync_bucket_dec(new_bucket);
 600         kfree_rcu(bucket, rcu);
 601 }
 602
 603 static int fuse_sync_fs(struct super_block *sb, int wait)
 604 {
 605         struct fuse_mount *fm = get_fuse_mount_super(sb);
 606         struct fuse_conn *fc = fm->fc;
 607         struct fuse_syncfs_in inarg;
 608         FUSE_ARGS(args);
 609         int err;
 610
 611         /*
 612          * Userspace cannot handle the wait == 0 case.  Avoid a
 613          * gratuitous roundtrip.
 614          */
 615         if (!wait)
 616                 return 0;
 617
 618         /* The filesystem is being unmounted.  Nothing to do. */
 619         if (!sb->s_root)
 620                 return 0;
 621
 622         if (!fc->sync_fs)
 623                 return 0;
 624
 625         fuse_sync_fs_writes(fc);
 626
 627         memset(&inarg, 0, sizeof(inarg));
 628         args.in_numargs = 1;
 629         args.in_args[0].size = sizeof(inarg);
 630         args.in_args[0].value = &inarg;
 631         args.opcode = FUSE_SYNCFS;
 632         args.nodeid = get_node_id(sb->s_root->d_inode);
 633         args.out_numargs = 0;
 634
 635         err = fuse_simple_request(fm, &args);
 636         if (err == -ENOSYS) {
 637                 fc->sync_fs = 0;
 638                 err = 0;
 639         }
 640
 641         return err;
 642 }
 643
 644 enum {
 645         OPT_SOURCE,
 646         OPT_SUBTYPE,
 647         OPT_FD,
 648         OPT_ROOTMODE,
 649         OPT_USER_ID,
 650         OPT_GROUP_ID,
 651         OPT_DEFAULT_PERMISSIONS,
 652         OPT_ALLOW_OTHER,
 653         OPT_MAX_READ,
 654         OPT_BLKSIZE,
 655         OPT_ERR
 656 };
 657
 658 static const struct fs_parameter_spec fuse_fs_parameters[] = {
 659         fsparam_string  ("source",              OPT_SOURCE),
 660         fsparam_u32     ("fd",                  OPT_FD),
 661         fsparam_u32oct  ("rootmode",            OPT_ROOTMODE),
 662         fsparam_u32     ("user_id",             OPT_USER_ID),
 663         fsparam_u32     ("group_id",            OPT_GROUP_ID),
 664         fsparam_flag    ("default_permissions", OPT_DEFAULT_PERMISSIONS),
 665         fsparam_flag    ("allow_other",         OPT_ALLOW_OTHER),
 666         fsparam_u32     ("max_read",            OPT_MAX_READ),
 667         fsparam_u32     ("blksize",             OPT_BLKSIZE),
 668         fsparam_string  ("subtype",             OPT_SUBTYPE),
 669         {}
 670 };
 671
 672 static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param)
 673 {
 674         struct fs_parse_result result;
 675         struct fuse_fs_context *ctx = fsc->fs_private;
 676         int opt;
 677
 678         if (fsc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
 679                 /*
 680                  * Ignore options coming from mount(MS_REMOUNT) for backward
 681                  * compatibility.
 682                  */
 683                 if (fsc->oldapi)
 684                         return 0;
 685
 686                 return invalfc(fsc, "No changes allowed in reconfigure");
 687         }
 688
 689         opt = fs_parse(fsc, fuse_fs_parameters, param, &result);
 690         if (opt < 0)
 691                 return opt;
 692
 693         switch (opt) {
 694         case OPT_SOURCE:
 695                 if (fsc->source)
 696                         return invalfc(fsc, "Multiple sources specified");
 697                 fsc->source = param->string;
 698                 param->string = NULL;
 699                 break;
 700
 701         case OPT_SUBTYPE:
 702                 if (ctx->subtype)
 703                         return invalfc(fsc, "Multiple subtypes specified");
 704                 ctx->subtype = param->string;
 705                 param->string = NULL;
 706                 return 0;
 707
 708         case OPT_FD:
 709                 ctx->fd = result.uint_32;
 710                 ctx->fd_present = true;
 711                 break;
 712
 713         case OPT_ROOTMODE:
 714                 if (!fuse_valid_type(result.uint_32))
 715                         return invalfc(fsc, "Invalid rootmode");
 716                 ctx->rootmode = result.uint_32;
 717                 ctx->rootmode_present = true;
 718                 break;
 719
 720         case OPT_USER_ID:
 721                 ctx->user_id = make_kuid(fsc->user_ns, result.uint_32);
 722                 if (!uid_valid(ctx->user_id))
 723                         return invalfc(fsc, "Invalid user_id");
 724                 ctx->user_id_present = true;
 725                 break;
 726
 727         case OPT_GROUP_ID:
 728                 ctx->group_id = make_kgid(fsc->user_ns, result.uint_32);
 729                 if (!gid_valid(ctx->group_id))
 730                         return invalfc(fsc, "Invalid group_id");
 731                 ctx->group_id_present = true;
 732                 break;
 733
 734         case OPT_DEFAULT_PERMISSIONS:
 735                 ctx->default_permissions = true;
 736                 break;
 737
 738         case OPT_ALLOW_OTHER:
 739                 ctx->allow_other = true;
 740                 break;
 741
 742         case OPT_MAX_READ:
 743                 ctx->max_read = result.uint_32;
 744                 break;
 745
 746         case OPT_BLKSIZE:
 747                 if (!ctx->is_bdev)
 748                         return invalfc(fsc, "blksize only supported for fuseblk");
 749                 ctx->blksize = result.uint_32;
 750                 break;
 751
 752         default:
 753                 return -EINVAL;
 754         }
 755
 756         return 0;
 757 }
 758
 759 static void fuse_free_fsc(struct fs_context *fsc)
 760 {
 761         struct fuse_fs_context *ctx = fsc->fs_private;
 762
 763         if (ctx) {
 764                 kfree(ctx->subtype);
 765                 kfree(ctx);
 766         }
 767 }
 768
 769 static int fuse_show_options(struct seq_file *m, struct dentry *root)
 770 {
 771         struct super_block *sb = root->d_sb;
 772         struct fuse_conn *fc = get_fuse_conn_super(sb);
 773
 774         if (fc->legacy_opts_show) {
 775                 seq_printf(m, ",user_id=%u",
 776                            from_kuid_munged(fc->user_ns, fc->user_id));
 777                 seq_printf(m, ",group_id=%u",
 778                            from_kgid_munged(fc->user_ns, fc->group_id));
 779                 if (fc->default_permissions)
 780                         seq_puts(m, ",default_permissions");
 781                 if (fc->allow_other)
 782                         seq_puts(m, ",allow_other");
 783                 if (fc->max_read != ~0)
 784                         seq_printf(m, ",max_read=%u", fc->max_read);
 785                 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
 786                         seq_printf(m, ",blksize=%lu", sb->s_blocksize);
 787         }
 788 #ifdef CONFIG_FUSE_DAX
 789         if (fc->dax_mode == FUSE_DAX_ALWAYS)
 790                 seq_puts(m, ",dax=always");
 791         else if (fc->dax_mode == FUSE_DAX_NEVER)
 792                 seq_puts(m, ",dax=never");
 793         else if (fc->dax_mode == FUSE_DAX_INODE_USER)
 794                 seq_puts(m, ",dax=inode");
 795 #endif
 796
 797         return 0;
 798 }
 799
 800 static void fuse_iqueue_init(struct fuse_iqueue *fiq,
 801                              const struct fuse_iqueue_ops *ops,
 802                              void *priv)
 803 {
 804         memset(fiq, 0, sizeof(struct fuse_iqueue));
 805         spin_lock_init(&fiq->lock);
 806         init_waitqueue_head(&fiq->waitq);
 807         INIT_LIST_HEAD(&fiq->pending);
 808         INIT_LIST_HEAD(&fiq->interrupts);
 809         fiq->forget_list_tail = &fiq->forget_list_head;
 810         fiq->connected = 1;
 811         fiq->ops = ops;
 812         fiq->priv = priv;
 813 }
 814
 815 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
 816 {
 817         unsigned int i;
 818
 819         spin_lock_init(&fpq->lock);
 820         for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
 821                 INIT_LIST_HEAD(&fpq->processing[i]);
 822         INIT_LIST_HEAD(&fpq->io);
 823         fpq->connected = 1;
 824 }
 825
 826 void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm,
 827                     struct user_namespace *user_ns,
 828                     const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
 829 {
 830         memset(fc, 0, sizeof(*fc));
 831         spin_lock_init(&fc->lock);
 832         spin_lock_init(&fc->bg_lock);
 833         init_rwsem(&fc->killsb);
 834         refcount_set(&fc->count, 1);
 835         atomic_set(&fc->dev_count, 1);
 836         init_waitqueue_head(&fc->blocked_waitq);
 837         fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
 838         INIT_LIST_HEAD(&fc->bg_queue);
 839         INIT_LIST_HEAD(&fc->entry);
 840         INIT_LIST_HEAD(&fc->devices);
 841         atomic_set(&fc->num_waiting, 0);
 842         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
 843         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
 844         atomic64_set(&fc->khctr, 0);
 845         fc->polled_files = RB_ROOT;
 846         fc->blocked = 0;
 847         fc->initialized = 0;
 848         fc->connected = 1;
 849         atomic64_set(&fc->attr_version, 1);
 850         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
 851         fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
 852         fc->user_ns = get_user_ns(user_ns);
 853         fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
 854         fc->max_pages_limit = FUSE_MAX_MAX_PAGES;
 855
 856         INIT_LIST_HEAD(&fc->mounts);
 857         list_add(&fm->fc_entry, &fc->mounts);
 858         fm->fc = fc;
 859 }
 860 EXPORT_SYMBOL_GPL(fuse_conn_init);
 861
 862 void fuse_conn_put(struct fuse_conn *fc)
 863 {
 864         if (refcount_dec_and_test(&fc->count)) {
 865                 struct fuse_iqueue *fiq = &fc->iq;
 866                 struct fuse_sync_bucket *bucket;
 867
 868                 if (IS_ENABLED(CONFIG_FUSE_DAX))
 869                         fuse_dax_conn_free(fc);
 870                 if (fiq->ops->release)
 871                         fiq->ops->release(fiq);
 872                 put_pid_ns(fc->pid_ns);
 873                 put_user_ns(fc->user_ns);
 874                 bucket = rcu_dereference_protected(fc->curr_bucket, 1);
 875                 if (bucket) {
 876                         WARN_ON(atomic_read(&bucket->count) != 1);
 877                         kfree(bucket);
 878                 }
 879                 fc->release(fc);
 880         }
 881 }
 882 EXPORT_SYMBOL_GPL(fuse_conn_put);
 883
 884 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
 885 {
 886         refcount_inc(&fc->count);
 887         return fc;
 888 }
 889 EXPORT_SYMBOL_GPL(fuse_conn_get);
 890
 891 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
 892 {
 893         struct fuse_attr attr;
 894         memset(&attr, 0, sizeof(attr));
 895
 896         attr.mode = mode;
 897         attr.ino = FUSE_ROOT_ID;
 898         attr.nlink = 1;
 899         return fuse_iget(sb, 1, 0, &attr, 0, 0);
 900 }
 901
 902 struct fuse_inode_handle {
 903         u64 nodeid;
 904         u32 generation;
 905 };
 906
 907 static struct dentry *fuse_get_dentry(struct super_block *sb,
 908                                       struct fuse_inode_handle *handle)
 909 {
 910         struct fuse_conn *fc = get_fuse_conn_super(sb);
 911         struct inode *inode;
 912         struct dentry *entry;
 913         int err = -ESTALE;
 914
 915         if (handle->nodeid == 0)
 916                 goto out_err;
 917
 918         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
 919         if (!inode) {
 920                 struct fuse_entry_out outarg;
 921                 const struct qstr name = QSTR_INIT(".", 1);
 922
 923                 if (!fc->export_support)
 924                         goto out_err;
 925
 926                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
 927                                        &inode);
 928                 if (err && err != -ENOENT)
 929                         goto out_err;
 930                 if (err || !inode) {
 931                         err = -ESTALE;
 932                         goto out_err;
 933                 }
 934                 err = -EIO;
 935                 if (get_node_id(inode) != handle->nodeid)
 936                         goto out_iput;
 937         }
 938         err = -ESTALE;
 939         if (inode->i_generation != handle->generation)
 940                 goto out_iput;
 941
 942         entry = d_obtain_alias(inode);
 943         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
 944                 fuse_invalidate_entry_cache(entry);
 945
 946         return entry;
 947
 948  out_iput:
 949         iput(inode);
 950  out_err:
 951         return ERR_PTR(err);
 952 }
 953
 954 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
 955                            struct inode *parent)
 956 {
 957         int len = parent ? 6 : 3;
 958         u64 nodeid;
 959         u32 generation;
 960
 961         if (*max_len < len) {
 962                 *max_len = len;
 963                 return  FILEID_INVALID;
 964         }
 965
 966         nodeid = get_fuse_inode(inode)->nodeid;
 967         generation = inode->i_generation;
 968
 969         fh[0] = (u32)(nodeid >> 32);
 970         fh[1] = (u32)(nodeid & 0xffffffff);
 971         fh[2] = generation;
 972
 973         if (parent) {
 974                 nodeid = get_fuse_inode(parent)->nodeid;
 975                 generation = parent->i_generation;
 976
 977                 fh[3] = (u32)(nodeid >> 32);
 978                 fh[4] = (u32)(nodeid & 0xffffffff);
 979                 fh[5] = generation;
 980         }
 981
 982         *max_len = len;
 983         return parent ? 0x82 : 0x81;
 984 }
 985
 986 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
 987                 struct fid *fid, int fh_len, int fh_type)
 988 {
 989         struct fuse_inode_handle handle;
 990
 991         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
 992                 return NULL;
 993
 994         handle.nodeid = (u64) fid->raw[0] << 32;
 995         handle.nodeid |= (u64) fid->raw[1];
 996         handle.generation = fid->raw[2];
 997         return fuse_get_dentry(sb, &handle);
 998 }
 999
1000 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
1001                 struct fid *fid, int fh_len, int fh_type)
1002 {
1003         struct fuse_inode_handle parent;
1004
1005         if (fh_type != 0x82 || fh_len < 6)
1006                 return NULL;
1007
1008         parent.nodeid = (u64) fid->raw[3] << 32;
1009         parent.nodeid |= (u64) fid->raw[4];
1010         parent.generation = fid->raw[5];
1011         return fuse_get_dentry(sb, &parent);
1012 }
1013
1014 static struct dentry *fuse_get_parent(struct dentry *child)
1015 {
1016         struct inode *child_inode = d_inode(child);
1017         struct fuse_conn *fc = get_fuse_conn(child_inode);
1018         struct inode *inode;
1019         struct dentry *parent;
1020         struct fuse_entry_out outarg;
1021         int err;
1022
1023         if (!fc->export_support)
1024                 return ERR_PTR(-ESTALE);
1025
1026         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
1027                                &dotdot_name, &outarg, &inode);
1028         if (err) {
1029                 if (err == -ENOENT)
1030                         return ERR_PTR(-ESTALE);
1031                 return ERR_PTR(err);
1032         }
1033
1034         parent = d_obtain_alias(inode);
1035         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
1036                 fuse_invalidate_entry_cache(parent);
1037
1038         return parent;
1039 }
1040
1041 static const struct export_operations fuse_export_operations = {
1042         .fh_to_dentry   = fuse_fh_to_dentry,
1043         .fh_to_parent   = fuse_fh_to_parent,
1044         .encode_fh      = fuse_encode_fh,
1045         .get_parent     = fuse_get_parent,
1046 };
1047
1048 static const struct super_operations fuse_super_operations = {
1049         .alloc_inode    = fuse_alloc_inode,
1050         .free_inode     = fuse_free_inode,
1051         .evict_inode    = fuse_evict_inode,
1052         .write_inode    = fuse_write_inode,
1053         .drop_inode     = generic_delete_inode,
1054         .umount_begin   = fuse_umount_begin,
1055         .statfs         = fuse_statfs,
1056         .sync_fs        = fuse_sync_fs,
1057         .show_options   = fuse_show_options,
1058 };
1059
1060 static void sanitize_global_limit(unsigned *limit)
1061 {
1062         /*
1063          * The default maximum number of async requests is calculated to consume
1064          * 1/2^13 of the total memory, assuming 392 bytes per request.
1065          */
1066         if (*limit == 0)
1067                 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
1068
1069         if (*limit >= 1 << 16)
1070                 *limit = (1 << 16) - 1;
1071 }
1072
1073 static int set_global_limit(const char *val, const struct kernel_param *kp)
1074 {
1075         int rv;
1076
1077         rv = param_set_uint(val, kp);
1078         if (rv)
1079                 return rv;
1080
1081         sanitize_global_limit((unsigned *)kp->arg);
1082
1083         return 0;
1084 }
1085
1086 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
1087 {
1088         int cap_sys_admin = capable(CAP_SYS_ADMIN);
1089
1090         if (arg->minor < 13)
1091                 return;
1092
1093         sanitize_global_limit(&max_user_bgreq);
1094         sanitize_global_limit(&max_user_congthresh);
1095
1096         spin_lock(&fc->bg_lock);
1097         if (arg->max_background) {
1098                 fc->max_background = arg->max_background;
1099
1100                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
1101                         fc->max_background = max_user_bgreq;
1102         }
1103         if (arg->congestion_threshold) {
1104                 fc->congestion_threshold = arg->congestion_threshold;
1105
1106                 if (!cap_sys_admin &&
1107                     fc->congestion_threshold > max_user_congthresh)
1108                         fc->congestion_threshold = max_user_congthresh;
1109         }
1110         spin_unlock(&fc->bg_lock);
1111 }
1112
1113 struct fuse_init_args {
1114         struct fuse_args args;
1115         struct fuse_init_in in;
1116         struct fuse_init_out out;
1117 };
1118
1119 static void process_init_reply(struct fuse_mount *fm, struct fuse_args *args,
1120                                int error)
1121 {
1122         struct fuse_conn *fc = fm->fc;
1123         struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
1124         struct fuse_init_out *arg = &ia->out;
1125         bool ok = true;
1126
1127         if (error || arg->major != FUSE_KERNEL_VERSION)
1128                 ok = false;
1129         else {
1130                 unsigned long ra_pages;
1131
1132                 process_init_limits(fc, arg);
1133
1134                 if (arg->minor >= 6) {
1135                         u64 flags = arg->flags;
1136
1137                         if (flags & FUSE_INIT_EXT)
1138                                 flags |= (u64) arg->flags2 << 32;
1139
1140                         ra_pages = arg->max_readahead / PAGE_SIZE;
1141                         if (flags & FUSE_ASYNC_READ)
1142                                 fc->async_read = 1;
1143                         if (!(flags & FUSE_POSIX_LOCKS))
1144                                 fc->no_lock = 1;
1145                         if (arg->minor >= 17) {
1146                                 if (!(flags & FUSE_FLOCK_LOCKS))
1147                                         fc->no_flock = 1;
1148                         } else {
1149                                 if (!(flags & FUSE_POSIX_LOCKS))
1150                                         fc->no_flock = 1;
1151                         }
1152                         if (flags & FUSE_ATOMIC_O_TRUNC)
1153                                 fc->atomic_o_trunc = 1;
1154                         if (arg->minor >= 9) {
1155                                 /* LOOKUP has dependency on proto version */
1156                                 if (flags & FUSE_EXPORT_SUPPORT)
1157                                         fc->export_support = 1;
1158                         }
1159                         if (flags & FUSE_BIG_WRITES)
1160                                 fc->big_writes = 1;
1161                         if (flags & FUSE_DONT_MASK)
1162                                 fc->dont_mask = 1;
1163                         if (flags & FUSE_AUTO_INVAL_DATA)
1164                                 fc->auto_inval_data = 1;
1165                         else if (flags & FUSE_EXPLICIT_INVAL_DATA)
1166                                 fc->explicit_inval_data = 1;
1167                         if (flags & FUSE_DO_READDIRPLUS) {
1168                                 fc->do_readdirplus = 1;
1169                                 if (flags & FUSE_READDIRPLUS_AUTO)
1170                                         fc->readdirplus_auto = 1;
1171                         }
1172                         if (flags & FUSE_ASYNC_DIO)
1173                                 fc->async_dio = 1;
1174                         if (flags & FUSE_WRITEBACK_CACHE)
1175                                 fc->writeback_cache = 1;
1176                         if (flags & FUSE_PARALLEL_DIROPS)
1177                                 fc->parallel_dirops = 1;
1178                         if (flags & FUSE_HANDLE_KILLPRIV)
1179                                 fc->handle_killpriv = 1;
1180                         if (arg->time_gran && arg->time_gran <= 1000000000)
1181                                 fm->sb->s_time_gran = arg->time_gran;
1182                         if ((flags & FUSE_POSIX_ACL)) {
1183                                 fc->default_permissions = 1;
1184                                 fc->posix_acl = 1;
1185                         }
1186                         if (flags & FUSE_CACHE_SYMLINKS)
1187                                 fc->cache_symlinks = 1;
1188                         if (flags & FUSE_ABORT_ERROR)
1189                                 fc->abort_err = 1;
1190                         if (flags & FUSE_MAX_PAGES) {
1191                                 fc->max_pages =
1192                                         min_t(unsigned int, fc->max_pages_limit,
1193                                         max_t(unsigned int, arg->max_pages, 1));
1194                         }
1195                         if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1196                                 if (flags & FUSE_MAP_ALIGNMENT &&
1197                                     !fuse_dax_check_alignment(fc, arg->map_alignment)) {
1198                                         ok = false;
1199                                 }
1200                                 if (flags & FUSE_HAS_INODE_DAX)
1201                                         fc->inode_dax = 1;
1202                         }
1203                         if (flags & FUSE_HANDLE_KILLPRIV_V2) {
1204                                 fc->handle_killpriv_v2 = 1;
1205                                 fm->sb->s_flags |= SB_NOSEC;
1206                         }
1207                         if (flags & FUSE_SETXATTR_EXT)
1208                                 fc->setxattr_ext = 1;
1209                         if (flags & FUSE_SECURITY_CTX)
1210                                 fc->init_security = 1;
1211                         if (flags & FUSE_CREATE_SUPP_GROUP)
1212                                 fc->create_supp_group = 1;
1213                 } else {
1214                         ra_pages = fc->max_read / PAGE_SIZE;
1215                         fc->no_lock = 1;
1216                         fc->no_flock = 1;
1217                 }
1218
1219                 fm->sb->s_bdi->ra_pages =
1220                                 min(fm->sb->s_bdi->ra_pages, ra_pages);
1221                 fc->minor = arg->minor;
1222                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
1223                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
1224                 fc->conn_init = 1;
1225         }
1226         kfree(ia);
1227
1228         if (!ok) {
1229                 fc->conn_init = 0;
1230                 fc->conn_error = 1;
1231         }
1232
1233         fuse_set_initialized(fc);
1234         wake_up_all(&fc->blocked_waitq);
1235 }
1236
1237 void fuse_send_init(struct fuse_mount *fm)
1238 {
1239         struct fuse_init_args *ia;
1240         u64 flags;
1241
1242         ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
1243
1244         ia->in.major = FUSE_KERNEL_VERSION;
1245         ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
1246         ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE;
1247         flags =
1248                 FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
1249                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
1250                 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1251                 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1252                 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1253                 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1254                 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1255                 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1256                 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA |
1257                 FUSE_HANDLE_KILLPRIV_V2 | FUSE_SETXATTR_EXT | FUSE_INIT_EXT |
1258                 FUSE_SECURITY_CTX | FUSE_CREATE_SUPP_GROUP |
1259                 FUSE_HAS_EXPIRE_ONLY;
1260 #ifdef CONFIG_FUSE_DAX
1261         if (fm->fc->dax)
1262                 flags |= FUSE_MAP_ALIGNMENT;
1263         if (fuse_is_inode_dax_mode(fm->fc->dax_mode))
1264                 flags |= FUSE_HAS_INODE_DAX;
1265 #endif
1266         if (fm->fc->auto_submounts)
1267                 flags |= FUSE_SUBMOUNTS;
1268
1269         ia->in.flags = flags;
1270         ia->in.flags2 = flags >> 32;
1271
1272         ia->args.opcode = FUSE_INIT;
1273         ia->args.in_numargs = 1;
1274         ia->args.in_args[0].size = sizeof(ia->in);
1275         ia->args.in_args[0].value = &ia->in;
1276         ia->args.out_numargs = 1;
1277         /* Variable length argument used for backward compatibility
1278            with interface version < 7.5.  Rest of init_out is zeroed
1279            by do_get_request(), so a short reply is not a problem */
1280         ia->args.out_argvar = true;
1281         ia->args.out_args[0].size = sizeof(ia->out);
1282         ia->args.out_args[0].value = &ia->out;
1283         ia->args.force = true;
1284         ia->args.nocreds = true;
1285         ia->args.end = process_init_reply;
1286
1287         if (fuse_simple_background(fm, &ia->args, GFP_KERNEL) != 0)
1288                 process_init_reply(fm, &ia->args, -ENOTCONN);
1289 }
1290 EXPORT_SYMBOL_GPL(fuse_send_init);
1291
1292 void fuse_free_conn(struct fuse_conn *fc)
1293 {
1294         WARN_ON(!list_empty(&fc->devices));
1295         kfree_rcu(fc, rcu);
1296 }
1297 EXPORT_SYMBOL_GPL(fuse_free_conn);
1298
1299 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1300 {
1301         int err;
1302         char *suffix = "";
1303
1304         if (sb->s_bdev) {
1305                 suffix = "-fuseblk";
1306                 /*
1307                  * sb->s_bdi points to blkdev's bdi however we want to redirect
1308                  * it to our private bdi...
1309                  */
1310                 bdi_put(sb->s_bdi);
1311                 sb->s_bdi = &noop_backing_dev_info;
1312         }
1313         err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1314                                    MINOR(fc->dev), suffix);
1315         if (err)
1316                 return err;
1317
1318         /* fuse does it's own writeback accounting */
1319         sb->s_bdi->capabilities &= ~BDI_CAP_WRITEBACK_ACCT;
1320         sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT;
1321
1322         /*
1323          * For a single fuse filesystem use max 1% of dirty +
1324          * writeback threshold.
1325          *
1326          * This gives about 1M of write buffer for memory maps on a
1327          * machine with 1G and 10% dirty_ratio, which should be more
1328          * than enough.
1329          *
1330          * Privileged users can raise it by writing to
1331          *
1332          *    /sys/class/bdi/<bdi>/max_ratio
1333          */
1334         bdi_set_max_ratio(sb->s_bdi, 1);
1335
1336         return 0;
1337 }
1338
1339 struct fuse_dev *fuse_dev_alloc(void)
1340 {
1341         struct fuse_dev *fud;
1342         struct list_head *pq;
1343
1344         fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1345         if (!fud)
1346                 return NULL;
1347
1348         pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1349         if (!pq) {
1350                 kfree(fud);
1351                 return NULL;
1352         }
1353
1354         fud->pq.processing = pq;
1355         fuse_pqueue_init(&fud->pq);
1356
1357         return fud;
1358 }
1359 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1360
1361 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1362 {
1363         fud->fc = fuse_conn_get(fc);
1364         spin_lock(&fc->lock);
1365         list_add_tail(&fud->entry, &fc->devices);
1366         spin_unlock(&fc->lock);
1367 }
1368 EXPORT_SYMBOL_GPL(fuse_dev_install);
1369
1370 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1371 {
1372         struct fuse_dev *fud;
1373
1374         fud = fuse_dev_alloc();
1375         if (!fud)
1376                 return NULL;
1377
1378         fuse_dev_install(fud, fc);
1379         return fud;
1380 }
1381 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1382
1383 void fuse_dev_free(struct fuse_dev *fud)
1384 {
1385         struct fuse_conn *fc = fud->fc;
1386
1387         if (fc) {
1388                 spin_lock(&fc->lock);
1389                 list_del(&fud->entry);
1390                 spin_unlock(&fc->lock);
1391
1392                 fuse_conn_put(fc);
1393         }
1394         kfree(fud->pq.processing);
1395         kfree(fud);
1396 }
1397 EXPORT_SYMBOL_GPL(fuse_dev_free);
1398
1399 static void fuse_fill_attr_from_inode(struct fuse_attr *attr,
1400                                       const struct fuse_inode *fi)
1401 {
1402         struct timespec64 ctime = inode_get_ctime(&fi->inode);
1403
1404         *attr = (struct fuse_attr){
1405                 .ino            = fi->inode.i_ino,
1406                 .size           = fi->inode.i_size,
1407                 .blocks         = fi->inode.i_blocks,
1408                 .atime          = fi->inode.i_atime.tv_sec,
1409                 .mtime          = fi->inode.i_mtime.tv_sec,
1410                 .ctime          = ctime.tv_sec,
1411                 .atimensec      = fi->inode.i_atime.tv_nsec,
1412                 .mtimensec      = fi->inode.i_mtime.tv_nsec,
1413                 .ctimensec      = ctime.tv_nsec,
1414                 .mode           = fi->inode.i_mode,
1415                 .nlink          = fi->inode.i_nlink,
1416                 .uid            = fi->inode.i_uid.val,
1417                 .gid            = fi->inode.i_gid.val,
1418                 .rdev           = fi->inode.i_rdev,
1419                 .blksize        = 1u << fi->inode.i_blkbits,
1420         };
1421 }
1422
1423 static void fuse_sb_defaults(struct super_block *sb)
1424 {
1425         sb->s_magic = FUSE_SUPER_MAGIC;
1426         sb->s_op = &fuse_super_operations;
1427         sb->s_xattr = fuse_xattr_handlers;
1428         sb->s_maxbytes = MAX_LFS_FILESIZE;
1429         sb->s_time_gran = 1;
1430         sb->s_export_op = &fuse_export_operations;
1431         sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1432         if (sb->s_user_ns != &init_user_ns)
1433                 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1434         sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1435 }
1436
1437 static int fuse_fill_super_submount(struct super_block *sb,
1438                                     struct fuse_inode *parent_fi)
1439 {
1440         struct fuse_mount *fm = get_fuse_mount_super(sb);
1441         struct super_block *parent_sb = parent_fi->inode.i_sb;
1442         struct fuse_attr root_attr;
1443         struct inode *root;
1444
1445         fuse_sb_defaults(sb);
1446         fm->sb = sb;
1447
1448         WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1449         sb->s_bdi = bdi_get(parent_sb->s_bdi);
1450
1451         sb->s_xattr = parent_sb->s_xattr;
1452         sb->s_time_gran = parent_sb->s_time_gran;
1453         sb->s_blocksize = parent_sb->s_blocksize;
1454         sb->s_blocksize_bits = parent_sb->s_blocksize_bits;
1455         sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL);
1456         if (parent_sb->s_subtype && !sb->s_subtype)
1457                 return -ENOMEM;
1458
1459         fuse_fill_attr_from_inode(&root_attr, parent_fi);
1460         root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0);
1461         /*
1462          * This inode is just a duplicate, so it is not looked up and
1463          * its nlookup should not be incremented.  fuse_iget() does
1464          * that, though, so undo it here.
1465          */
1466         get_fuse_inode(root)->nlookup--;
1467         sb->s_d_op = &fuse_dentry_operations;
1468         sb->s_root = d_make_root(root);
1469         if (!sb->s_root)
1470                 return -ENOMEM;
1471
1472         return 0;
1473 }
1474
1475 /* Filesystem context private data holds the FUSE inode of the mount point */
1476 static int fuse_get_tree_submount(struct fs_context *fsc)
1477 {
1478         struct fuse_mount *fm;
1479         struct fuse_inode *mp_fi = fsc->fs_private;
1480         struct fuse_conn *fc = get_fuse_conn(&mp_fi->inode);
1481         struct super_block *sb;
1482         int err;
1483
1484         fm = kzalloc(sizeof(struct fuse_mount), GFP_KERNEL);
1485         if (!fm)
1486                 return -ENOMEM;
1487
1488         fm->fc = fuse_conn_get(fc);
1489         fsc->s_fs_info = fm;
1490         sb = sget_fc(fsc, NULL, set_anon_super_fc);
1491         if (fsc->s_fs_info)
1492                 fuse_mount_destroy(fm);
1493         if (IS_ERR(sb))
1494                 return PTR_ERR(sb);
1495
1496         /* Initialize superblock, making @mp_fi its root */
1497         err = fuse_fill_super_submount(sb, mp_fi);
1498         if (err) {
1499                 deactivate_locked_super(sb);
1500                 return err;
1501         }
1502
1503         down_write(&fc->killsb);
1504         list_add_tail(&fm->fc_entry, &fc->mounts);
1505         up_write(&fc->killsb);
1506
1507         sb->s_flags |= SB_ACTIVE;
1508         fsc->root = dget(sb->s_root);
1509
1510         return 0;
1511 }
1512
1513 static const struct fs_context_operations fuse_context_submount_ops = {
1514         .get_tree       = fuse_get_tree_submount,
1515 };
1516
1517 int fuse_init_fs_context_submount(struct fs_context *fsc)
1518 {
1519         fsc->ops = &fuse_context_submount_ops;
1520         return 0;
1521 }
1522 EXPORT_SYMBOL_GPL(fuse_init_fs_context_submount);
1523
1524 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1525 {
1526         struct fuse_dev *fud = NULL;
1527         struct fuse_mount *fm = get_fuse_mount_super(sb);
1528         struct fuse_conn *fc = fm->fc;
1529         struct inode *root;
1530         struct dentry *root_dentry;
1531         int err;
1532
1533         err = -EINVAL;
1534         if (sb->s_flags & SB_MANDLOCK)
1535                 goto err;
1536
1537         rcu_assign_pointer(fc->curr_bucket, fuse_sync_bucket_alloc());
1538         fuse_sb_defaults(sb);
1539
1540         if (ctx->is_bdev) {
1541 #ifdef CONFIG_BLOCK
1542                 err = -EINVAL;
1543                 if (!sb_set_blocksize(sb, ctx->blksize))
1544                         goto err;
1545 #endif
1546         } else {
1547                 sb->s_blocksize = PAGE_SIZE;
1548                 sb->s_blocksize_bits = PAGE_SHIFT;
1549         }
1550
1551         sb->s_subtype = ctx->subtype;
1552         ctx->subtype = NULL;
1553         if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1554                 err = fuse_dax_conn_alloc(fc, ctx->dax_mode, ctx->dax_dev);
1555                 if (err)
1556                         goto err;
1557         }
1558
1559         if (ctx->fudptr) {
1560                 err = -ENOMEM;
1561                 fud = fuse_dev_alloc_install(fc);
1562                 if (!fud)
1563                         goto err_free_dax;
1564         }
1565
1566         fc->dev = sb->s_dev;
1567         fm->sb = sb;
1568         err = fuse_bdi_init(fc, sb);
1569         if (err)
1570                 goto err_dev_free;
1571
1572         /* Handle umasking inside the fuse code */
1573         if (sb->s_flags & SB_POSIXACL)
1574                 fc->dont_mask = 1;
1575         sb->s_flags |= SB_POSIXACL;
1576
1577         fc->default_permissions = ctx->default_permissions;
1578         fc->allow_other = ctx->allow_other;
1579         fc->user_id = ctx->user_id;
1580         fc->group_id = ctx->group_id;
1581         fc->legacy_opts_show = ctx->legacy_opts_show;
1582         fc->max_read = max_t(unsigned int, 4096, ctx->max_read);
1583         fc->destroy = ctx->destroy;
1584         fc->no_control = ctx->no_control;
1585         fc->no_force_umount = ctx->no_force_umount;
1586
1587         err = -ENOMEM;
1588         root = fuse_get_root_inode(sb, ctx->rootmode);
1589         sb->s_d_op = &fuse_root_dentry_operations;
1590         root_dentry = d_make_root(root);
1591         if (!root_dentry)
1592                 goto err_dev_free;
1593         /* Root dentry doesn't have .d_revalidate */
1594         sb->s_d_op = &fuse_dentry_operations;
1595
1596         mutex_lock(&fuse_mutex);
1597         err = -EINVAL;
1598         if (ctx->fudptr && *ctx->fudptr)
1599                 goto err_unlock;
1600
1601         err = fuse_ctl_add_conn(fc);
1602         if (err)
1603                 goto err_unlock;
1604
1605         list_add_tail(&fc->entry, &fuse_conn_list);
1606         sb->s_root = root_dentry;
1607         if (ctx->fudptr)
1608                 *ctx->fudptr = fud;
1609         mutex_unlock(&fuse_mutex);
1610         return 0;
1611
1612  err_unlock:
1613         mutex_unlock(&fuse_mutex);
1614         dput(root_dentry);
1615  err_dev_free:
1616         if (fud)
1617                 fuse_dev_free(fud);
1618  err_free_dax:
1619         if (IS_ENABLED(CONFIG_FUSE_DAX))
1620                 fuse_dax_conn_free(fc);
1621  err:
1622         return err;
1623 }
1624 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1625
1626 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1627 {
1628         struct fuse_fs_context *ctx = fsc->fs_private;
1629         int err;
1630
1631         if (!ctx->file || !ctx->rootmode_present ||
1632             !ctx->user_id_present || !ctx->group_id_present)
1633                 return -EINVAL;
1634
1635         /*
1636          * Require mount to happen from the same user namespace which
1637          * opened /dev/fuse to prevent potential attacks.
1638          */
1639         if ((ctx->file->f_op != &fuse_dev_operations) ||
1640             (ctx->file->f_cred->user_ns != sb->s_user_ns))
1641                 return -EINVAL;
1642         ctx->fudptr = &ctx->file->private_data;
1643
1644         err = fuse_fill_super_common(sb, ctx);
1645         if (err)
1646                 return err;
1647         /* file->private_data shall be visible on all CPUs after this */
1648         smp_mb();
1649         fuse_send_init(get_fuse_mount_super(sb));
1650         return 0;
1651 }
1652
1653 /*
1654  * This is the path where user supplied an already initialized fuse dev.  In
1655  * this case never create a new super if the old one is gone.
1656  */
1657 static int fuse_set_no_super(struct super_block *sb, struct fs_context *fsc)
1658 {
1659         return -ENOTCONN;
1660 }
1661
1662 static int fuse_test_super(struct super_block *sb, struct fs_context *fsc)
1663 {
1664
1665         return fsc->sget_key == get_fuse_conn_super(sb);
1666 }
1667
1668 static int fuse_get_tree(struct fs_context *fsc)
1669 {
1670         struct fuse_fs_context *ctx = fsc->fs_private;
1671         struct fuse_dev *fud;
1672         struct fuse_conn *fc;
1673         struct fuse_mount *fm;
1674         struct super_block *sb;
1675         int err;
1676
1677         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1678         if (!fc)
1679                 return -ENOMEM;
1680
1681         fm = kzalloc(sizeof(*fm), GFP_KERNEL);
1682         if (!fm) {
1683                 kfree(fc);
1684                 return -ENOMEM;
1685         }
1686
1687         fuse_conn_init(fc, fm, fsc->user_ns, &fuse_dev_fiq_ops, NULL);
1688         fc->release = fuse_free_conn;
1689
1690         fsc->s_fs_info = fm;
1691
1692         if (ctx->fd_present)
1693                 ctx->file = fget(ctx->fd);
1694
1695         if (IS_ENABLED(CONFIG_BLOCK) && ctx->is_bdev) {
1696                 err = get_tree_bdev(fsc, fuse_fill_super);
1697                 goto out;
1698         }
1699         /*
1700          * While block dev mount can be initialized with a dummy device fd
1701          * (found by device name), normal fuse mounts can't
1702          */
1703         err = -EINVAL;
1704         if (!ctx->file)
1705                 goto out;
1706
1707         /*
1708          * Allow creating a fuse mount with an already initialized fuse
1709          * connection
1710          */
1711         fud = READ_ONCE(ctx->file->private_data);
1712         if (ctx->file->f_op == &fuse_dev_operations && fud) {
1713                 fsc->sget_key = fud->fc;
1714                 sb = sget_fc(fsc, fuse_test_super, fuse_set_no_super);
1715                 err = PTR_ERR_OR_ZERO(sb);
1716                 if (!IS_ERR(sb))
1717                         fsc->root = dget(sb->s_root);
1718         } else {
1719                 err = get_tree_nodev(fsc, fuse_fill_super);
1720         }
1721 out:
1722         if (fsc->s_fs_info)
1723                 fuse_mount_destroy(fm);
1724         if (ctx->file)
1725                 fput(ctx->file);
1726         return err;
1727 }
1728
1729 static const struct fs_context_operations fuse_context_ops = {
1730         .free           = fuse_free_fsc,
1731         .parse_param    = fuse_parse_param,
1732         .reconfigure    = fuse_reconfigure,
1733         .get_tree       = fuse_get_tree,
1734 };
1735
1736 /*
1737  * Set up the filesystem mount context.
1738  */
1739 static int fuse_init_fs_context(struct fs_context *fsc)
1740 {
1741         struct fuse_fs_context *ctx;
1742
1743         ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1744         if (!ctx)
1745                 return -ENOMEM;
1746
1747         ctx->max_read = ~0;
1748         ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1749         ctx->legacy_opts_show = true;
1750
1751 #ifdef CONFIG_BLOCK
1752         if (fsc->fs_type == &fuseblk_fs_type) {
1753                 ctx->is_bdev = true;
1754                 ctx->destroy = true;
1755         }
1756 #endif
1757
1758         fsc->fs_private = ctx;
1759         fsc->ops = &fuse_context_ops;
1760         return 0;
1761 }
1762
1763 bool fuse_mount_remove(struct fuse_mount *fm)
1764 {
1765         struct fuse_conn *fc = fm->fc;
1766         bool last = false;
1767
1768         down_write(&fc->killsb);
1769         list_del_init(&fm->fc_entry);
1770         if (list_empty(&fc->mounts))
1771                 last = true;
1772         up_write(&fc->killsb);
1773
1774         return last;
1775 }
1776 EXPORT_SYMBOL_GPL(fuse_mount_remove);
1777
1778 void fuse_conn_destroy(struct fuse_mount *fm)
1779 {
1780         struct fuse_conn *fc = fm->fc;
1781
1782         if (fc->destroy)
1783                 fuse_send_destroy(fm);
1784
1785         fuse_abort_conn(fc);
1786         fuse_wait_aborted(fc);
1787
1788         if (!list_empty(&fc->entry)) {
1789                 mutex_lock(&fuse_mutex);
1790                 list_del(&fc->entry);
1791                 fuse_ctl_remove_conn(fc);
1792                 mutex_unlock(&fuse_mutex);
1793         }
1794 }
1795 EXPORT_SYMBOL_GPL(fuse_conn_destroy);
1796
1797 static void fuse_sb_destroy(struct super_block *sb)
1798 {
1799         struct fuse_mount *fm = get_fuse_mount_super(sb);
1800         bool last;
1801
1802         if (sb->s_root) {
1803                 last = fuse_mount_remove(fm);
1804                 if (last)
1805                         fuse_conn_destroy(fm);
1806         }
1807 }
1808
1809 void fuse_mount_destroy(struct fuse_mount *fm)
1810 {
1811         fuse_conn_put(fm->fc);
1812         kfree(fm);
1813 }
1814 EXPORT_SYMBOL(fuse_mount_destroy);
1815
1816 static void fuse_kill_sb_anon(struct super_block *sb)
1817 {
1818         fuse_sb_destroy(sb);
1819         kill_anon_super(sb);
1820         fuse_mount_destroy(get_fuse_mount_super(sb));
1821 }
1822
1823 static struct file_system_type fuse_fs_type = {
1824         .owner          = THIS_MODULE,
1825         .name           = "fuse",
1826         .fs_flags       = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1827         .init_fs_context = fuse_init_fs_context,
1828         .parameters     = fuse_fs_parameters,
1829         .kill_sb        = fuse_kill_sb_anon,
1830 };
1831 MODULE_ALIAS_FS("fuse");
1832
1833 #ifdef CONFIG_BLOCK
1834 static void fuse_kill_sb_blk(struct super_block *sb)
1835 {
1836         fuse_sb_destroy(sb);
1837         kill_block_super(sb);
1838         fuse_mount_destroy(get_fuse_mount_super(sb));
1839 }
1840
1841 static struct file_system_type fuseblk_fs_type = {
1842         .owner          = THIS_MODULE,
1843         .name           = "fuseblk",
1844         .init_fs_context = fuse_init_fs_context,
1845         .parameters     = fuse_fs_parameters,
1846         .kill_sb        = fuse_kill_sb_blk,
1847         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1848 };
1849 MODULE_ALIAS_FS("fuseblk");
1850
1851 static inline int register_fuseblk(void)
1852 {
1853         return register_filesystem(&fuseblk_fs_type);
1854 }
1855
1856 static inline void unregister_fuseblk(void)
1857 {
1858         unregister_filesystem(&fuseblk_fs_type);
1859 }
1860 #else
1861 static inline int register_fuseblk(void)
1862 {
1863         return 0;
1864 }
1865
1866 static inline void unregister_fuseblk(void)
1867 {
1868 }
1869 #endif
1870
1871 static void fuse_inode_init_once(void *foo)
1872 {
1873         struct inode *inode = foo;
1874
1875         inode_init_once(inode);
1876 }
1877
1878 static int __init fuse_fs_init(void)
1879 {
1880         int err;
1881
1882         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1883                         sizeof(struct fuse_inode), 0,
1884                         SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1885                         fuse_inode_init_once);
1886         err = -ENOMEM;
1887         if (!fuse_inode_cachep)
1888                 goto out;
1889
1890         err = register_fuseblk();
1891         if (err)
1892                 goto out2;
1893
1894         err = register_filesystem(&fuse_fs_type);
1895         if (err)
1896                 goto out3;
1897
1898         return 0;
1899
1900  out3:
1901         unregister_fuseblk();
1902  out2:
1903         kmem_cache_destroy(fuse_inode_cachep);
1904  out:
1905         return err;
1906 }
1907
1908 static void fuse_fs_cleanup(void)
1909 {
1910         unregister_filesystem(&fuse_fs_type);
1911         unregister_fuseblk();
1912
1913         /*
1914          * Make sure all delayed rcu free inodes are flushed before we
1915          * destroy cache.
1916          */
1917         rcu_barrier();
1918         kmem_cache_destroy(fuse_inode_cachep);
1919 }
1920
1921 static struct kobject *fuse_kobj;
1922
1923 static int fuse_sysfs_init(void)
1924 {
1925         int err;
1926
1927         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1928         if (!fuse_kobj) {
1929                 err = -ENOMEM;
1930                 goto out_err;
1931         }
1932
1933         err = sysfs_create_mount_point(fuse_kobj, "connections");
1934         if (err)
1935                 goto out_fuse_unregister;
1936
1937         return 0;
1938
1939  out_fuse_unregister:
1940         kobject_put(fuse_kobj);
1941  out_err:
1942         return err;
1943 }
1944
1945 static void fuse_sysfs_cleanup(void)
1946 {
1947         sysfs_remove_mount_point(fuse_kobj, "connections");
1948         kobject_put(fuse_kobj);
1949 }
1950
1951 static int __init fuse_init(void)
1952 {
1953         int res;
1954
1955         pr_info("init (API version %i.%i)\n",
1956                 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1957
1958         INIT_LIST_HEAD(&fuse_conn_list);
1959         res = fuse_fs_init();
1960         if (res)
1961                 goto err;
1962
1963         res = fuse_dev_init();
1964         if (res)
1965                 goto err_fs_cleanup;
1966
1967         res = fuse_sysfs_init();
1968         if (res)
1969                 goto err_dev_cleanup;
1970
1971         res = fuse_ctl_init();
1972         if (res)
1973                 goto err_sysfs_cleanup;
1974
1975         sanitize_global_limit(&max_user_bgreq);
1976         sanitize_global_limit(&max_user_congthresh);
1977
1978         return 0;
1979
1980  err_sysfs_cleanup:
1981         fuse_sysfs_cleanup();
1982  err_dev_cleanup:
1983         fuse_dev_cleanup();
1984  err_fs_cleanup:
1985         fuse_fs_cleanup();
1986  err:
1987         return res;
1988 }
1989
1990 static void __exit fuse_exit(void)
1991 {
1992         pr_debug("exit\n");
1993
1994         fuse_ctl_cleanup();
1995         fuse_sysfs_cleanup();
1996         fuse_fs_cleanup();
1997         fuse_dev_cleanup();
1998 }
1999
2000 module_init(fuse_init);
2001 module_exit(fuse_exit);