fs/fuse/inode.c

   1 /*
   2   FUSE: Filesystem in Userspace
   3   Copyright (C) 2001-2008  Miklos Szeredi <[email protected]>
   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 <[email protected]>");
  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 fuse_submount_lookup *fuse_alloc_submount_lookup(void)
  72 {
  73         struct fuse_submount_lookup *sl;
  74
  75         sl = kzalloc(sizeof(struct fuse_submount_lookup), GFP_KERNEL_ACCOUNT);
  76         if (!sl)
  77                 return NULL;
  78         sl->forget = fuse_alloc_forget();
  79         if (!sl->forget)
  80                 goto out_free;
  81
  82         return sl;
  83
  84 out_free:
  85         kfree(sl);
  86         return NULL;
  87 }
  88
  89 static struct inode *fuse_alloc_inode(struct super_block *sb)
  90 {
  91         struct fuse_inode *fi;
  92
  93         fi = alloc_inode_sb(sb, 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         fi->submount_lookup = NULL;
 105         mutex_init(&fi->mutex);
 106         spin_lock_init(&fi->lock);
 107         fi->forget = fuse_alloc_forget();
 108         if (!fi->forget)
 109                 goto out_free;
 110
 111         if (IS_ENABLED(CONFIG_FUSE_DAX) && !fuse_dax_inode_alloc(sb, fi))
 112                 goto out_free_forget;
 113
 114         if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
 115                 fuse_inode_backing_set(fi, NULL);
 116
 117         return &fi->inode;
 118
 119 out_free_forget:
 120         kfree(fi->forget);
 121 out_free:
 122         kmem_cache_free(fuse_inode_cachep, fi);
 123         return NULL;
 124 }
 125
 126 static void fuse_free_inode(struct inode *inode)
 127 {
 128         struct fuse_inode *fi = get_fuse_inode(inode);
 129
 130         mutex_destroy(&fi->mutex);
 131         kfree(fi->forget);
 132 #ifdef CONFIG_FUSE_DAX
 133         kfree(fi->dax);
 134 #endif
 135         if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
 136                 fuse_backing_put(fuse_inode_backing(fi));
 137
 138         kmem_cache_free(fuse_inode_cachep, fi);
 139 }
 140
 141 static void fuse_cleanup_submount_lookup(struct fuse_conn *fc,
 142                                          struct fuse_submount_lookup *sl)
 143 {
 144         if (!refcount_dec_and_test(&sl->count))
 145                 return;
 146
 147         fuse_queue_forget(fc, sl->forget, sl->nodeid, 1);
 148         sl->forget = NULL;
 149         kfree(sl);
 150 }
 151
 152 static void fuse_evict_inode(struct inode *inode)
 153 {
 154         struct fuse_inode *fi = get_fuse_inode(inode);
 155
 156         /* Will write inode on close/munmap and in all other dirtiers */
 157         WARN_ON(inode->i_state & I_DIRTY_INODE);
 158
 159         truncate_inode_pages_final(&inode->i_data);
 160         clear_inode(inode);
 161         if (inode->i_sb->s_flags & SB_ACTIVE) {
 162                 struct fuse_conn *fc = get_fuse_conn(inode);
 163
 164                 if (FUSE_IS_DAX(inode))
 165                         fuse_dax_inode_cleanup(inode);
 166                 if (fi->nlookup) {
 167                         fuse_queue_forget(fc, fi->forget, fi->nodeid,
 168                                           fi->nlookup);
 169                         fi->forget = NULL;
 170                 }
 171
 172                 if (fi->submount_lookup) {
 173                         fuse_cleanup_submount_lookup(fc, fi->submount_lookup);
 174                         fi->submount_lookup = NULL;
 175                 }
 176         }
 177         if (S_ISREG(inode->i_mode) && !fuse_is_bad(inode)) {
 178                 WARN_ON(fi->iocachectr != 0);
 179                 WARN_ON(!list_empty(&fi->write_files));
 180                 WARN_ON(!list_empty(&fi->queued_writes));
 181         }
 182 }
 183
 184 static int fuse_reconfigure(struct fs_context *fsc)
 185 {
 186         struct super_block *sb = fsc->root->d_sb;
 187
 188         sync_filesystem(sb);
 189         if (fsc->sb_flags & SB_MANDLOCK)
 190                 return -EINVAL;
 191
 192         return 0;
 193 }
 194
 195 /*
 196  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
 197  * so that it will fit.
 198  */
 199 static ino_t fuse_squash_ino(u64 ino64)
 200 {
 201         ino_t ino = (ino_t) ino64;
 202         if (sizeof(ino_t) < sizeof(u64))
 203                 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
 204         return ino;
 205 }
 206
 207 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
 208                                    struct fuse_statx *sx,
 209                                    u64 attr_valid, u32 cache_mask)
 210 {
 211         struct fuse_conn *fc = get_fuse_conn(inode);
 212         struct fuse_inode *fi = get_fuse_inode(inode);
 213
 214         lockdep_assert_held(&fi->lock);
 215
 216         fi->attr_version = atomic64_inc_return(&fc->attr_version);
 217         fi->i_time = attr_valid;
 218         /* Clear basic stats from invalid mask */
 219         set_mask_bits(&fi->inval_mask, STATX_BASIC_STATS, 0);
 220
 221         inode->i_ino     = fuse_squash_ino(attr->ino);
 222         inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
 223         set_nlink(inode, attr->nlink);
 224         inode->i_uid     = make_kuid(fc->user_ns, attr->uid);
 225         inode->i_gid     = make_kgid(fc->user_ns, attr->gid);
 226         inode->i_blocks  = attr->blocks;
 227
 228         /* Sanitize nsecs */
 229         attr->atimensec = min_t(u32, attr->atimensec, NSEC_PER_SEC - 1);
 230         attr->mtimensec = min_t(u32, attr->mtimensec, NSEC_PER_SEC - 1);
 231         attr->ctimensec = min_t(u32, attr->ctimensec, NSEC_PER_SEC - 1);
 232
 233         inode_set_atime(inode, attr->atime, attr->atimensec);
 234         /* mtime from server may be stale due to local buffered write */
 235         if (!(cache_mask & STATX_MTIME)) {
 236                 inode_set_mtime(inode, attr->mtime, attr->mtimensec);
 237         }
 238         if (!(cache_mask & STATX_CTIME)) {
 239                 inode_set_ctime(inode, attr->ctime, attr->ctimensec);
 240         }
 241         if (sx) {
 242                 /* Sanitize nsecs */
 243                 sx->btime.tv_nsec =
 244                         min_t(u32, sx->btime.tv_nsec, NSEC_PER_SEC - 1);
 245
 246                 /*
 247                  * Btime has been queried, cache is valid (whether or not btime
 248                  * is available or not) so clear STATX_BTIME from inval_mask.
 249                  *
 250                  * Availability of the btime attribute is indicated in
 251                  * FUSE_I_BTIME
 252                  */
 253                 set_mask_bits(&fi->inval_mask, STATX_BTIME, 0);
 254                 if (sx->mask & STATX_BTIME) {
 255                         set_bit(FUSE_I_BTIME, &fi->state);
 256                         fi->i_btime.tv_sec = sx->btime.tv_sec;
 257                         fi->i_btime.tv_nsec = sx->btime.tv_nsec;
 258                 }
 259         }
 260
 261         if (attr->blksize != 0)
 262                 inode->i_blkbits = ilog2(attr->blksize);
 263         else
 264                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
 265
 266         /*
 267          * Don't set the sticky bit in i_mode, unless we want the VFS
 268          * to check permissions.  This prevents failures due to the
 269          * check in may_delete().
 270          */
 271         fi->orig_i_mode = inode->i_mode;
 272         if (!fc->default_permissions)
 273                 inode->i_mode &= ~S_ISVTX;
 274
 275         fi->orig_ino = attr->ino;
 276
 277         /*
 278          * We are refreshing inode data and it is possible that another
 279          * client set suid/sgid or security.capability xattr. So clear
 280          * S_NOSEC. Ideally, we could have cleared it only if suid/sgid
 281          * was set or if security.capability xattr was set. But we don't
 282          * know if security.capability has been set or not. So clear it
 283          * anyway. Its less efficient but should be safe.
 284          */
 285         inode->i_flags &= ~S_NOSEC;
 286 }
 287
 288 u32 fuse_get_cache_mask(struct inode *inode)
 289 {
 290         struct fuse_conn *fc = get_fuse_conn(inode);
 291
 292         if (!fc->writeback_cache || !S_ISREG(inode->i_mode))
 293                 return 0;
 294
 295         return STATX_MTIME | STATX_CTIME | STATX_SIZE;
 296 }
 297
 298 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
 299                             struct fuse_statx *sx,
 300                             u64 attr_valid, u64 attr_version)
 301 {
 302         struct fuse_conn *fc = get_fuse_conn(inode);
 303         struct fuse_inode *fi = get_fuse_inode(inode);
 304         u32 cache_mask;
 305         loff_t oldsize;
 306         struct timespec64 old_mtime;
 307
 308         spin_lock(&fi->lock);
 309         /*
 310          * In case of writeback_cache enabled, writes update mtime, ctime and
 311          * may update i_size.  In these cases trust the cached value in the
 312          * inode.
 313          */
 314         cache_mask = fuse_get_cache_mask(inode);
 315         if (cache_mask & STATX_SIZE)
 316                 attr->size = i_size_read(inode);
 317
 318         if (cache_mask & STATX_MTIME) {
 319                 attr->mtime = inode_get_mtime_sec(inode);
 320                 attr->mtimensec = inode_get_mtime_nsec(inode);
 321         }
 322         if (cache_mask & STATX_CTIME) {
 323                 attr->ctime = inode_get_ctime_sec(inode);
 324                 attr->ctimensec = inode_get_ctime_nsec(inode);
 325         }
 326
 327         if ((attr_version != 0 && fi->attr_version > attr_version) ||
 328             test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
 329                 spin_unlock(&fi->lock);
 330                 return;
 331         }
 332
 333         old_mtime = inode_get_mtime(inode);
 334         fuse_change_attributes_common(inode, attr, sx, attr_valid, cache_mask);
 335
 336         oldsize = inode->i_size;
 337         /*
 338          * In case of writeback_cache enabled, the cached writes beyond EOF
 339          * extend local i_size without keeping userspace server in sync. So,
 340          * attr->size coming from server can be stale. We cannot trust it.
 341          */
 342         if (!(cache_mask & STATX_SIZE))
 343                 i_size_write(inode, attr->size);
 344         spin_unlock(&fi->lock);
 345
 346         if (!cache_mask && S_ISREG(inode->i_mode)) {
 347                 bool inval = false;
 348
 349                 if (oldsize != attr->size) {
 350                         truncate_pagecache(inode, attr->size);
 351                         if (!fc->explicit_inval_data)
 352                                 inval = true;
 353                 } else if (fc->auto_inval_data) {
 354                         struct timespec64 new_mtime = {
 355                                 .tv_sec = attr->mtime,
 356                                 .tv_nsec = attr->mtimensec,
 357                         };
 358
 359                         /*
 360                          * Auto inval mode also checks and invalidates if mtime
 361                          * has changed.
 362                          */
 363                         if (!timespec64_equal(&old_mtime, &new_mtime))
 364                                 inval = true;
 365                 }
 366
 367                 if (inval)
 368                         invalidate_inode_pages2(inode->i_mapping);
 369         }
 370
 371         if (IS_ENABLED(CONFIG_FUSE_DAX))
 372                 fuse_dax_dontcache(inode, attr->flags);
 373 }
 374
 375 static void fuse_init_submount_lookup(struct fuse_submount_lookup *sl,
 376                                       u64 nodeid)
 377 {
 378         sl->nodeid = nodeid;
 379         refcount_set(&sl->count, 1);
 380 }
 381
 382 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr,
 383                             struct fuse_conn *fc)
 384 {
 385         inode->i_mode = attr->mode & S_IFMT;
 386         inode->i_size = attr->size;
 387         inode_set_mtime(inode, attr->mtime, attr->mtimensec);
 388         inode_set_ctime(inode, attr->ctime, attr->ctimensec);
 389         if (S_ISREG(inode->i_mode)) {
 390                 fuse_init_common(inode);
 391                 fuse_init_file_inode(inode, attr->flags);
 392         } else if (S_ISDIR(inode->i_mode))
 393                 fuse_init_dir(inode);
 394         else if (S_ISLNK(inode->i_mode))
 395                 fuse_init_symlink(inode);
 396         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
 397                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
 398                 fuse_init_common(inode);
 399                 init_special_inode(inode, inode->i_mode,
 400                                    new_decode_dev(attr->rdev));
 401         } else
 402                 BUG();
 403         /*
 404          * Ensure that we don't cache acls for daemons without FUSE_POSIX_ACL
 405          * so they see the exact same behavior as before.
 406          */
 407         if (!fc->posix_acl)
 408                 inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE;
 409 }
 410
 411 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
 412 {
 413         u64 nodeid = *(u64 *) _nodeidp;
 414         if (get_node_id(inode) == nodeid)
 415                 return 1;
 416         else
 417                 return 0;
 418 }
 419
 420 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
 421 {
 422         u64 nodeid = *(u64 *) _nodeidp;
 423         get_fuse_inode(inode)->nodeid = nodeid;
 424         return 0;
 425 }
 426
 427 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
 428                         int generation, struct fuse_attr *attr,
 429                         u64 attr_valid, u64 attr_version)
 430 {
 431         struct inode *inode;
 432         struct fuse_inode *fi;
 433         struct fuse_conn *fc = get_fuse_conn_super(sb);
 434
 435         /*
 436          * Auto mount points get their node id from the submount root, which is
 437          * not a unique identifier within this filesystem.
 438          *
 439          * To avoid conflicts, do not place submount points into the inode hash
 440          * table.
 441          */
 442         if (fc->auto_submounts && (attr->flags & FUSE_ATTR_SUBMOUNT) &&
 443             S_ISDIR(attr->mode)) {
 444                 struct fuse_inode *fi;
 445
 446                 inode = new_inode(sb);
 447                 if (!inode)
 448                         return NULL;
 449
 450                 fuse_init_inode(inode, attr, fc);
 451                 fi = get_fuse_inode(inode);
 452                 fi->nodeid = nodeid;
 453                 fi->submount_lookup = fuse_alloc_submount_lookup();
 454                 if (!fi->submount_lookup) {
 455                         iput(inode);
 456                         return NULL;
 457                 }
 458                 /* Sets nlookup = 1 on fi->submount_lookup->nlookup */
 459                 fuse_init_submount_lookup(fi->submount_lookup, nodeid);
 460                 inode->i_flags |= S_AUTOMOUNT;
 461                 goto done;
 462         }
 463
 464 retry:
 465         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
 466         if (!inode)
 467                 return NULL;
 468
 469         if ((inode->i_state & I_NEW)) {
 470                 inode->i_flags |= S_NOATIME;
 471                 if (!fc->writeback_cache || !S_ISREG(attr->mode))
 472                         inode->i_flags |= S_NOCMTIME;
 473                 inode->i_generation = generation;
 474                 fuse_init_inode(inode, attr, fc);
 475                 unlock_new_inode(inode);
 476         } else if (fuse_stale_inode(inode, generation, attr)) {
 477                 /* nodeid was reused, any I/O on the old inode should fail */
 478                 fuse_make_bad(inode);
 479                 if (inode != d_inode(sb->s_root)) {
 480                         remove_inode_hash(inode);
 481                         iput(inode);
 482                         goto retry;
 483                 }
 484         }
 485         fi = get_fuse_inode(inode);
 486         spin_lock(&fi->lock);
 487         fi->nlookup++;
 488         spin_unlock(&fi->lock);
 489 done:
 490         fuse_change_attributes(inode, attr, NULL, attr_valid, attr_version);
 491
 492         return inode;
 493 }
 494
 495 struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid,
 496                            struct fuse_mount **fm)
 497 {
 498         struct fuse_mount *fm_iter;
 499         struct inode *inode;
 500
 501         WARN_ON(!rwsem_is_locked(&fc->killsb));
 502         list_for_each_entry(fm_iter, &fc->mounts, fc_entry) {
 503                 if (!fm_iter->sb)
 504                         continue;
 505
 506                 inode = ilookup5(fm_iter->sb, nodeid, fuse_inode_eq, &nodeid);
 507                 if (inode) {
 508                         if (fm)
 509                                 *fm = fm_iter;
 510                         return inode;
 511                 }
 512         }
 513
 514         return NULL;
 515 }
 516
 517 int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
 518                              loff_t offset, loff_t len)
 519 {
 520         struct fuse_inode *fi;
 521         struct inode *inode;
 522         pgoff_t pg_start;
 523         pgoff_t pg_end;
 524
 525         inode = fuse_ilookup(fc, nodeid, NULL);
 526         if (!inode)
 527                 return -ENOENT;
 528
 529         fi = get_fuse_inode(inode);
 530         spin_lock(&fi->lock);
 531         fi->attr_version = atomic64_inc_return(&fc->attr_version);
 532         spin_unlock(&fi->lock);
 533
 534         fuse_invalidate_attr(inode);
 535         forget_all_cached_acls(inode);
 536         if (offset >= 0) {
 537                 pg_start = offset >> PAGE_SHIFT;
 538                 if (len <= 0)
 539                         pg_end = -1;
 540                 else
 541                         pg_end = (offset + len - 1) >> PAGE_SHIFT;
 542                 invalidate_inode_pages2_range(inode->i_mapping,
 543                                               pg_start, pg_end);
 544         }
 545         iput(inode);
 546         return 0;
 547 }
 548
 549 bool fuse_lock_inode(struct inode *inode)
 550 {
 551         bool locked = false;
 552
 553         if (!get_fuse_conn(inode)->parallel_dirops) {
 554                 mutex_lock(&get_fuse_inode(inode)->mutex);
 555                 locked = true;
 556         }
 557
 558         return locked;
 559 }
 560
 561 void fuse_unlock_inode(struct inode *inode, bool locked)
 562 {
 563         if (locked)
 564                 mutex_unlock(&get_fuse_inode(inode)->mutex);
 565 }
 566
 567 static void fuse_umount_begin(struct super_block *sb)
 568 {
 569         struct fuse_conn *fc = get_fuse_conn_super(sb);
 570
 571         if (fc->no_force_umount)
 572                 return;
 573
 574         fuse_abort_conn(fc);
 575
 576         // Only retire block-device-based superblocks.
 577         if (sb->s_bdev != NULL)
 578                 retire_super(sb);
 579 }
 580
 581 static void fuse_send_destroy(struct fuse_mount *fm)
 582 {
 583         if (fm->fc->conn_init) {
 584                 FUSE_ARGS(args);
 585
 586                 args.opcode = FUSE_DESTROY;
 587                 args.force = true;
 588                 args.nocreds = true;
 589                 fuse_simple_request(fm, &args);
 590         }
 591 }
 592
 593 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
 594 {
 595         stbuf->f_type    = FUSE_SUPER_MAGIC;
 596         stbuf->f_bsize   = attr->bsize;
 597         stbuf->f_frsize  = attr->frsize;
 598         stbuf->f_blocks  = attr->blocks;
 599         stbuf->f_bfree   = attr->bfree;
 600         stbuf->f_bavail  = attr->bavail;
 601         stbuf->f_files   = attr->files;
 602         stbuf->f_ffree   = attr->ffree;
 603         stbuf->f_namelen = attr->namelen;
 604         /* fsid is left zero */
 605 }
 606
 607 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
 608 {
 609         struct super_block *sb = dentry->d_sb;
 610         struct fuse_mount *fm = get_fuse_mount_super(sb);
 611         FUSE_ARGS(args);
 612         struct fuse_statfs_out outarg;
 613         int err;
 614
 615         if (!fuse_allow_current_process(fm->fc)) {
 616                 buf->f_type = FUSE_SUPER_MAGIC;
 617                 return 0;
 618         }
 619
 620         memset(&outarg, 0, sizeof(outarg));
 621         args.in_numargs = 0;
 622         args.opcode = FUSE_STATFS;
 623         args.nodeid = get_node_id(d_inode(dentry));
 624         args.out_numargs = 1;
 625         args.out_args[0].size = sizeof(outarg);
 626         args.out_args[0].value = &outarg;
 627         err = fuse_simple_request(fm, &args);
 628         if (!err)
 629                 convert_fuse_statfs(buf, &outarg.st);
 630         return err;
 631 }
 632
 633 static struct fuse_sync_bucket *fuse_sync_bucket_alloc(void)
 634 {
 635         struct fuse_sync_bucket *bucket;
 636
 637         bucket = kzalloc(sizeof(*bucket), GFP_KERNEL | __GFP_NOFAIL);
 638         if (bucket) {
 639                 init_waitqueue_head(&bucket->waitq);
 640                 /* Initial active count */
 641                 atomic_set(&bucket->count, 1);
 642         }
 643         return bucket;
 644 }
 645
 646 static void fuse_sync_fs_writes(struct fuse_conn *fc)
 647 {
 648         struct fuse_sync_bucket *bucket, *new_bucket;
 649         int count;
 650
 651         new_bucket = fuse_sync_bucket_alloc();
 652         spin_lock(&fc->lock);
 653         bucket = rcu_dereference_protected(fc->curr_bucket, 1);
 654         count = atomic_read(&bucket->count);
 655         WARN_ON(count < 1);
 656         /* No outstanding writes? */
 657         if (count == 1) {
 658                 spin_unlock(&fc->lock);
 659                 kfree(new_bucket);
 660                 return;
 661         }
 662
 663         /*
 664          * Completion of new bucket depends on completion of this bucket, so add
 665          * one more count.
 666          */
 667         atomic_inc(&new_bucket->count);
 668         rcu_assign_pointer(fc->curr_bucket, new_bucket);
 669         spin_unlock(&fc->lock);
 670         /*
 671          * Drop initial active count.  At this point if all writes in this and
 672          * ancestor buckets complete, the count will go to zero and this task
 673          * will be woken up.
 674          */
 675         atomic_dec(&bucket->count);
 676
 677         wait_event(bucket->waitq, atomic_read(&bucket->count) == 0);
 678
 679         /* Drop temp count on descendant bucket */
 680         fuse_sync_bucket_dec(new_bucket);
 681         kfree_rcu(bucket, rcu);
 682 }
 683
 684 static int fuse_sync_fs(struct super_block *sb, int wait)
 685 {
 686         struct fuse_mount *fm = get_fuse_mount_super(sb);
 687         struct fuse_conn *fc = fm->fc;
 688         struct fuse_syncfs_in inarg;
 689         FUSE_ARGS(args);
 690         int err;
 691
 692         /*
 693          * Userspace cannot handle the wait == 0 case.  Avoid a
 694          * gratuitous roundtrip.
 695          */
 696         if (!wait)
 697                 return 0;
 698
 699         /* The filesystem is being unmounted.  Nothing to do. */
 700         if (!sb->s_root)
 701                 return 0;
 702
 703         if (!fc->sync_fs)
 704                 return 0;
 705
 706         fuse_sync_fs_writes(fc);
 707
 708         memset(&inarg, 0, sizeof(inarg));
 709         args.in_numargs = 1;
 710         args.in_args[0].size = sizeof(inarg);
 711         args.in_args[0].value = &inarg;
 712         args.opcode = FUSE_SYNCFS;
 713         args.nodeid = get_node_id(sb->s_root->d_inode);
 714         args.out_numargs = 0;
 715
 716         err = fuse_simple_request(fm, &args);
 717         if (err == -ENOSYS) {
 718                 fc->sync_fs = 0;
 719                 err = 0;
 720         }
 721
 722         return err;
 723 }
 724
 725 enum {
 726         OPT_SOURCE,
 727         OPT_SUBTYPE,
 728         OPT_FD,
 729         OPT_ROOTMODE,
 730         OPT_USER_ID,
 731         OPT_GROUP_ID,
 732         OPT_DEFAULT_PERMISSIONS,
 733         OPT_ALLOW_OTHER,
 734         OPT_MAX_READ,
 735         OPT_BLKSIZE,
 736         OPT_ERR
 737 };
 738
 739 static const struct fs_parameter_spec fuse_fs_parameters[] = {
 740         fsparam_string  ("source",              OPT_SOURCE),
 741         fsparam_u32     ("fd",                  OPT_FD),
 742         fsparam_u32oct  ("rootmode",            OPT_ROOTMODE),
 743         fsparam_u32     ("user_id",             OPT_USER_ID),
 744         fsparam_u32     ("group_id",            OPT_GROUP_ID),
 745         fsparam_flag    ("default_permissions", OPT_DEFAULT_PERMISSIONS),
 746         fsparam_flag    ("allow_other",         OPT_ALLOW_OTHER),
 747         fsparam_u32     ("max_read",            OPT_MAX_READ),
 748         fsparam_u32     ("blksize",             OPT_BLKSIZE),
 749         fsparam_string  ("subtype",             OPT_SUBTYPE),
 750         {}
 751 };
 752
 753 static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param)
 754 {
 755         struct fs_parse_result result;
 756         struct fuse_fs_context *ctx = fsc->fs_private;
 757         int opt;
 758
 759         if (fsc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
 760                 /*
 761                  * Ignore options coming from mount(MS_REMOUNT) for backward
 762                  * compatibility.
 763                  */
 764                 if (fsc->oldapi)
 765                         return 0;
 766
 767                 return invalfc(fsc, "No changes allowed in reconfigure");
 768         }
 769
 770         opt = fs_parse(fsc, fuse_fs_parameters, param, &result);
 771         if (opt < 0)
 772                 return opt;
 773
 774         switch (opt) {
 775         case OPT_SOURCE:
 776                 if (fsc->source)
 777                         return invalfc(fsc, "Multiple sources specified");
 778                 fsc->source = param->string;
 779                 param->string = NULL;
 780                 break;
 781
 782         case OPT_SUBTYPE:
 783                 if (ctx->subtype)
 784                         return invalfc(fsc, "Multiple subtypes specified");
 785                 ctx->subtype = param->string;
 786                 param->string = NULL;
 787                 return 0;
 788
 789         case OPT_FD:
 790                 ctx->fd = result.uint_32;
 791                 ctx->fd_present = true;
 792                 break;
 793
 794         case OPT_ROOTMODE:
 795                 if (!fuse_valid_type(result.uint_32))
 796                         return invalfc(fsc, "Invalid rootmode");
 797                 ctx->rootmode = result.uint_32;
 798                 ctx->rootmode_present = true;
 799                 break;
 800
 801         case OPT_USER_ID:
 802                 ctx->user_id = make_kuid(fsc->user_ns, result.uint_32);
 803                 if (!uid_valid(ctx->user_id))
 804                         return invalfc(fsc, "Invalid user_id");
 805                 ctx->user_id_present = true;
 806                 break;
 807
 808         case OPT_GROUP_ID:
 809                 ctx->group_id = make_kgid(fsc->user_ns, result.uint_32);
 810                 if (!gid_valid(ctx->group_id))
 811                         return invalfc(fsc, "Invalid group_id");
 812                 ctx->group_id_present = true;
 813                 break;
 814
 815         case OPT_DEFAULT_PERMISSIONS:
 816                 ctx->default_permissions = true;
 817                 break;
 818
 819         case OPT_ALLOW_OTHER:
 820                 ctx->allow_other = true;
 821                 break;
 822
 823         case OPT_MAX_READ:
 824                 ctx->max_read = result.uint_32;
 825                 break;
 826
 827         case OPT_BLKSIZE:
 828                 if (!ctx->is_bdev)
 829                         return invalfc(fsc, "blksize only supported for fuseblk");
 830                 ctx->blksize = result.uint_32;
 831                 break;
 832
 833         default:
 834                 return -EINVAL;
 835         }
 836
 837         return 0;
 838 }
 839
 840 static void fuse_free_fsc(struct fs_context *fsc)
 841 {
 842         struct fuse_fs_context *ctx = fsc->fs_private;
 843
 844         if (ctx) {
 845                 kfree(ctx->subtype);
 846                 kfree(ctx);
 847         }
 848 }
 849
 850 static int fuse_show_options(struct seq_file *m, struct dentry *root)
 851 {
 852         struct super_block *sb = root->d_sb;
 853         struct fuse_conn *fc = get_fuse_conn_super(sb);
 854
 855         if (fc->legacy_opts_show) {
 856                 seq_printf(m, ",user_id=%u",
 857                            from_kuid_munged(fc->user_ns, fc->user_id));
 858                 seq_printf(m, ",group_id=%u",
 859                            from_kgid_munged(fc->user_ns, fc->group_id));
 860                 if (fc->default_permissions)
 861                         seq_puts(m, ",default_permissions");
 862                 if (fc->allow_other)
 863                         seq_puts(m, ",allow_other");
 864                 if (fc->max_read != ~0)
 865                         seq_printf(m, ",max_read=%u", fc->max_read);
 866                 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
 867                         seq_printf(m, ",blksize=%lu", sb->s_blocksize);
 868         }
 869 #ifdef CONFIG_FUSE_DAX
 870         if (fc->dax_mode == FUSE_DAX_ALWAYS)
 871                 seq_puts(m, ",dax=always");
 872         else if (fc->dax_mode == FUSE_DAX_NEVER)
 873                 seq_puts(m, ",dax=never");
 874         else if (fc->dax_mode == FUSE_DAX_INODE_USER)
 875                 seq_puts(m, ",dax=inode");
 876 #endif
 877
 878         return 0;
 879 }
 880
 881 static void fuse_iqueue_init(struct fuse_iqueue *fiq,
 882                              const struct fuse_iqueue_ops *ops,
 883                              void *priv)
 884 {
 885         memset(fiq, 0, sizeof(struct fuse_iqueue));
 886         spin_lock_init(&fiq->lock);
 887         init_waitqueue_head(&fiq->waitq);
 888         INIT_LIST_HEAD(&fiq->pending);
 889         INIT_LIST_HEAD(&fiq->interrupts);
 890         fiq->forget_list_tail = &fiq->forget_list_head;
 891         fiq->connected = 1;
 892         fiq->ops = ops;
 893         fiq->priv = priv;
 894 }
 895
 896 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
 897 {
 898         unsigned int i;
 899
 900         spin_lock_init(&fpq->lock);
 901         for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
 902                 INIT_LIST_HEAD(&fpq->processing[i]);
 903         INIT_LIST_HEAD(&fpq->io);
 904         fpq->connected = 1;
 905 }
 906
 907 void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm,
 908                     struct user_namespace *user_ns,
 909                     const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
 910 {
 911         memset(fc, 0, sizeof(*fc));
 912         spin_lock_init(&fc->lock);
 913         spin_lock_init(&fc->bg_lock);
 914         init_rwsem(&fc->killsb);
 915         refcount_set(&fc->count, 1);
 916         atomic_set(&fc->dev_count, 1);
 917         init_waitqueue_head(&fc->blocked_waitq);
 918         fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
 919         INIT_LIST_HEAD(&fc->bg_queue);
 920         INIT_LIST_HEAD(&fc->entry);
 921         INIT_LIST_HEAD(&fc->devices);
 922         atomic_set(&fc->num_waiting, 0);
 923         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
 924         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
 925         atomic64_set(&fc->khctr, 0);
 926         fc->polled_files = RB_ROOT;
 927         fc->blocked = 0;
 928         fc->initialized = 0;
 929         fc->connected = 1;
 930         atomic64_set(&fc->attr_version, 1);
 931         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
 932         fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
 933         fc->user_ns = get_user_ns(user_ns);
 934         fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
 935         fc->max_pages_limit = FUSE_MAX_MAX_PAGES;
 936
 937         if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
 938                 fuse_backing_files_init(fc);
 939
 940         INIT_LIST_HEAD(&fc->mounts);
 941         list_add(&fm->fc_entry, &fc->mounts);
 942         fm->fc = fc;
 943 }
 944 EXPORT_SYMBOL_GPL(fuse_conn_init);
 945
 946 static void delayed_release(struct rcu_head *p)
 947 {
 948         struct fuse_conn *fc = container_of(p, struct fuse_conn, rcu);
 949
 950         put_user_ns(fc->user_ns);
 951         fc->release(fc);
 952 }
 953
 954 void fuse_conn_put(struct fuse_conn *fc)
 955 {
 956         if (refcount_dec_and_test(&fc->count)) {
 957                 struct fuse_iqueue *fiq = &fc->iq;
 958                 struct fuse_sync_bucket *bucket;
 959
 960                 if (IS_ENABLED(CONFIG_FUSE_DAX))
 961                         fuse_dax_conn_free(fc);
 962                 if (fiq->ops->release)
 963                         fiq->ops->release(fiq);
 964                 put_pid_ns(fc->pid_ns);
 965                 bucket = rcu_dereference_protected(fc->curr_bucket, 1);
 966                 if (bucket) {
 967                         WARN_ON(atomic_read(&bucket->count) != 1);
 968                         kfree(bucket);
 969                 }
 970                 if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
 971                         fuse_backing_files_free(fc);
 972                 call_rcu(&fc->rcu, delayed_release);
 973         }
 974 }
 975 EXPORT_SYMBOL_GPL(fuse_conn_put);
 976
 977 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
 978 {
 979         refcount_inc(&fc->count);
 980         return fc;
 981 }
 982 EXPORT_SYMBOL_GPL(fuse_conn_get);
 983
 984 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
 985 {
 986         struct fuse_attr attr;
 987         memset(&attr, 0, sizeof(attr));
 988
 989         attr.mode = mode;
 990         attr.ino = FUSE_ROOT_ID;
 991         attr.nlink = 1;
 992         return fuse_iget(sb, FUSE_ROOT_ID, 0, &attr, 0, 0);
 993 }
 994
 995 struct fuse_inode_handle {
 996         u64 nodeid;
 997         u32 generation;
 998 };
 999
1000 static struct dentry *fuse_get_dentry(struct super_block *sb,
1001                                       struct fuse_inode_handle *handle)
1002 {
1003         struct fuse_conn *fc = get_fuse_conn_super(sb);
1004         struct inode *inode;
1005         struct dentry *entry;
1006         int err = -ESTALE;
1007
1008         if (handle->nodeid == 0)
1009                 goto out_err;
1010
1011         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
1012         if (!inode) {
1013                 struct fuse_entry_out outarg;
1014                 const struct qstr name = QSTR_INIT(".", 1);
1015
1016                 if (!fc->export_support)
1017                         goto out_err;
1018
1019                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
1020                                        &inode);
1021                 if (err && err != -ENOENT)
1022                         goto out_err;
1023                 if (err || !inode) {
1024                         err = -ESTALE;
1025                         goto out_err;
1026                 }
1027                 err = -EIO;
1028                 if (get_node_id(inode) != handle->nodeid)
1029                         goto out_iput;
1030         }
1031         err = -ESTALE;
1032         if (inode->i_generation != handle->generation)
1033                 goto out_iput;
1034
1035         entry = d_obtain_alias(inode);
1036         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
1037                 fuse_invalidate_entry_cache(entry);
1038
1039         return entry;
1040
1041  out_iput:
1042         iput(inode);
1043  out_err:
1044         return ERR_PTR(err);
1045 }
1046
1047 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
1048                            struct inode *parent)
1049 {
1050         int len = parent ? 6 : 3;
1051         u64 nodeid;
1052         u32 generation;
1053
1054         if (*max_len < len) {
1055                 *max_len = len;
1056                 return  FILEID_INVALID;
1057         }
1058
1059         nodeid = get_fuse_inode(inode)->nodeid;
1060         generation = inode->i_generation;
1061
1062         fh[0] = (u32)(nodeid >> 32);
1063         fh[1] = (u32)(nodeid & 0xffffffff);
1064         fh[2] = generation;
1065
1066         if (parent) {
1067                 nodeid = get_fuse_inode(parent)->nodeid;
1068                 generation = parent->i_generation;
1069
1070                 fh[3] = (u32)(nodeid >> 32);
1071                 fh[4] = (u32)(nodeid & 0xffffffff);
1072                 fh[5] = generation;
1073         }
1074
1075         *max_len = len;
1076         return parent ? FILEID_INO64_GEN_PARENT : FILEID_INO64_GEN;
1077 }
1078
1079 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
1080                 struct fid *fid, int fh_len, int fh_type)
1081 {
1082         struct fuse_inode_handle handle;
1083
1084         if ((fh_type != FILEID_INO64_GEN &&
1085              fh_type != FILEID_INO64_GEN_PARENT) || fh_len < 3)
1086                 return NULL;
1087
1088         handle.nodeid = (u64) fid->raw[0] << 32;
1089         handle.nodeid |= (u64) fid->raw[1];
1090         handle.generation = fid->raw[2];
1091         return fuse_get_dentry(sb, &handle);
1092 }
1093
1094 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
1095                 struct fid *fid, int fh_len, int fh_type)
1096 {
1097         struct fuse_inode_handle parent;
1098
1099         if (fh_type != FILEID_INO64_GEN_PARENT || fh_len < 6)
1100                 return NULL;
1101
1102         parent.nodeid = (u64) fid->raw[3] << 32;
1103         parent.nodeid |= (u64) fid->raw[4];
1104         parent.generation = fid->raw[5];
1105         return fuse_get_dentry(sb, &parent);
1106 }
1107
1108 static struct dentry *fuse_get_parent(struct dentry *child)
1109 {
1110         struct inode *child_inode = d_inode(child);
1111         struct fuse_conn *fc = get_fuse_conn(child_inode);
1112         struct inode *inode;
1113         struct dentry *parent;
1114         struct fuse_entry_out outarg;
1115         int err;
1116
1117         if (!fc->export_support)
1118                 return ERR_PTR(-ESTALE);
1119
1120         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
1121                                &dotdot_name, &outarg, &inode);
1122         if (err) {
1123                 if (err == -ENOENT)
1124                         return ERR_PTR(-ESTALE);
1125                 return ERR_PTR(err);
1126         }
1127
1128         parent = d_obtain_alias(inode);
1129         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
1130                 fuse_invalidate_entry_cache(parent);
1131
1132         return parent;
1133 }
1134
1135 /* only for fid encoding; no support for file handle */
1136 static const struct export_operations fuse_export_fid_operations = {
1137         .encode_fh      = fuse_encode_fh,
1138 };
1139
1140 static const struct export_operations fuse_export_operations = {
1141         .fh_to_dentry   = fuse_fh_to_dentry,
1142         .fh_to_parent   = fuse_fh_to_parent,
1143         .encode_fh      = fuse_encode_fh,
1144         .get_parent     = fuse_get_parent,
1145 };
1146
1147 static const struct super_operations fuse_super_operations = {
1148         .alloc_inode    = fuse_alloc_inode,
1149         .free_inode     = fuse_free_inode,
1150         .evict_inode    = fuse_evict_inode,
1151         .write_inode    = fuse_write_inode,
1152         .drop_inode     = generic_delete_inode,
1153         .umount_begin   = fuse_umount_begin,
1154         .statfs         = fuse_statfs,
1155         .sync_fs        = fuse_sync_fs,
1156         .show_options   = fuse_show_options,
1157 };
1158
1159 static void sanitize_global_limit(unsigned *limit)
1160 {
1161         /*
1162          * The default maximum number of async requests is calculated to consume
1163          * 1/2^13 of the total memory, assuming 392 bytes per request.
1164          */
1165         if (*limit == 0)
1166                 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
1167
1168         if (*limit >= 1 << 16)
1169                 *limit = (1 << 16) - 1;
1170 }
1171
1172 static int set_global_limit(const char *val, const struct kernel_param *kp)
1173 {
1174         int rv;
1175
1176         rv = param_set_uint(val, kp);
1177         if (rv)
1178                 return rv;
1179
1180         sanitize_global_limit((unsigned *)kp->arg);
1181
1182         return 0;
1183 }
1184
1185 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
1186 {
1187         int cap_sys_admin = capable(CAP_SYS_ADMIN);
1188
1189         if (arg->minor < 13)
1190                 return;
1191
1192         sanitize_global_limit(&max_user_bgreq);
1193         sanitize_global_limit(&max_user_congthresh);
1194
1195         spin_lock(&fc->bg_lock);
1196         if (arg->max_background) {
1197                 fc->max_background = arg->max_background;
1198
1199                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
1200                         fc->max_background = max_user_bgreq;
1201         }
1202         if (arg->congestion_threshold) {
1203                 fc->congestion_threshold = arg->congestion_threshold;
1204
1205                 if (!cap_sys_admin &&
1206                     fc->congestion_threshold > max_user_congthresh)
1207                         fc->congestion_threshold = max_user_congthresh;
1208         }
1209         spin_unlock(&fc->bg_lock);
1210 }
1211
1212 struct fuse_init_args {
1213         struct fuse_args args;
1214         struct fuse_init_in in;
1215         struct fuse_init_out out;
1216 };
1217
1218 static void process_init_reply(struct fuse_mount *fm, struct fuse_args *args,
1219                                int error)
1220 {
1221         struct fuse_conn *fc = fm->fc;
1222         struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
1223         struct fuse_init_out *arg = &ia->out;
1224         bool ok = true;
1225
1226         if (error || arg->major != FUSE_KERNEL_VERSION)
1227                 ok = false;
1228         else {
1229                 unsigned long ra_pages;
1230
1231                 process_init_limits(fc, arg);
1232
1233                 if (arg->minor >= 6) {
1234                         u64 flags = arg->flags;
1235
1236                         if (flags & FUSE_INIT_EXT)
1237                                 flags |= (u64) arg->flags2 << 32;
1238
1239                         ra_pages = arg->max_readahead / PAGE_SIZE;
1240                         if (flags & FUSE_ASYNC_READ)
1241                                 fc->async_read = 1;
1242                         if (!(flags & FUSE_POSIX_LOCKS))
1243                                 fc->no_lock = 1;
1244                         if (arg->minor >= 17) {
1245                                 if (!(flags & FUSE_FLOCK_LOCKS))
1246                                         fc->no_flock = 1;
1247                         } else {
1248                                 if (!(flags & FUSE_POSIX_LOCKS))
1249                                         fc->no_flock = 1;
1250                         }
1251                         if (flags & FUSE_ATOMIC_O_TRUNC)
1252                                 fc->atomic_o_trunc = 1;
1253                         if (arg->minor >= 9) {
1254                                 /* LOOKUP has dependency on proto version */
1255                                 if (flags & FUSE_EXPORT_SUPPORT)
1256                                         fc->export_support = 1;
1257                         }
1258                         if (flags & FUSE_BIG_WRITES)
1259                                 fc->big_writes = 1;
1260                         if (flags & FUSE_DONT_MASK)
1261                                 fc->dont_mask = 1;
1262                         if (flags & FUSE_AUTO_INVAL_DATA)
1263                                 fc->auto_inval_data = 1;
1264                         else if (flags & FUSE_EXPLICIT_INVAL_DATA)
1265                                 fc->explicit_inval_data = 1;
1266                         if (flags & FUSE_DO_READDIRPLUS) {
1267                                 fc->do_readdirplus = 1;
1268                                 if (flags & FUSE_READDIRPLUS_AUTO)
1269                                         fc->readdirplus_auto = 1;
1270                         }
1271                         if (flags & FUSE_ASYNC_DIO)
1272                                 fc->async_dio = 1;
1273                         if (flags & FUSE_WRITEBACK_CACHE)
1274                                 fc->writeback_cache = 1;
1275                         if (flags & FUSE_PARALLEL_DIROPS)
1276                                 fc->parallel_dirops = 1;
1277                         if (flags & FUSE_HANDLE_KILLPRIV)
1278                                 fc->handle_killpriv = 1;
1279                         if (arg->time_gran && arg->time_gran <= 1000000000)
1280                                 fm->sb->s_time_gran = arg->time_gran;
1281                         if ((flags & FUSE_POSIX_ACL)) {
1282                                 fc->default_permissions = 1;
1283                                 fc->posix_acl = 1;
1284                         }
1285                         if (flags & FUSE_CACHE_SYMLINKS)
1286                                 fc->cache_symlinks = 1;
1287                         if (flags & FUSE_ABORT_ERROR)
1288                                 fc->abort_err = 1;
1289                         if (flags & FUSE_MAX_PAGES) {
1290                                 fc->max_pages =
1291                                         min_t(unsigned int, fc->max_pages_limit,
1292                                         max_t(unsigned int, arg->max_pages, 1));
1293                         }
1294                         if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1295                                 if (flags & FUSE_MAP_ALIGNMENT &&
1296                                     !fuse_dax_check_alignment(fc, arg->map_alignment)) {
1297                                         ok = false;
1298                                 }
1299                                 if (flags & FUSE_HAS_INODE_DAX)
1300                                         fc->inode_dax = 1;
1301                         }
1302                         if (flags & FUSE_HANDLE_KILLPRIV_V2) {
1303                                 fc->handle_killpriv_v2 = 1;
1304                                 fm->sb->s_flags |= SB_NOSEC;
1305                         }
1306                         if (flags & FUSE_SETXATTR_EXT)
1307                                 fc->setxattr_ext = 1;
1308                         if (flags & FUSE_SECURITY_CTX)
1309                                 fc->init_security = 1;
1310                         if (flags & FUSE_CREATE_SUPP_GROUP)
1311                                 fc->create_supp_group = 1;
1312                         if (flags & FUSE_DIRECT_IO_ALLOW_MMAP)
1313                                 fc->direct_io_allow_mmap = 1;
1314                         /*
1315                          * max_stack_depth is the max stack depth of FUSE fs,
1316                          * so it has to be at least 1 to support passthrough
1317                          * to backing files.
1318                          *
1319                          * with max_stack_depth > 1, the backing files can be
1320                          * on a stacked fs (e.g. overlayfs) themselves and with
1321                          * max_stack_depth == 1, FUSE fs can be stacked as the
1322                          * underlying fs of a stacked fs (e.g. overlayfs).
1323                          */
1324                         if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH) &&
1325                             (flags & FUSE_PASSTHROUGH) &&
1326                             arg->max_stack_depth > 0 &&
1327                             arg->max_stack_depth <= FILESYSTEM_MAX_STACK_DEPTH) {
1328                                 fc->passthrough = 1;
1329                                 fc->max_stack_depth = arg->max_stack_depth;
1330                                 fm->sb->s_stack_depth = arg->max_stack_depth;
1331                         }
1332                         if (flags & FUSE_NO_EXPORT_SUPPORT)
1333                                 fm->sb->s_export_op = &fuse_export_fid_operations;
1334                 } else {
1335                         ra_pages = fc->max_read / PAGE_SIZE;
1336                         fc->no_lock = 1;
1337                         fc->no_flock = 1;
1338                 }
1339
1340                 fm->sb->s_bdi->ra_pages =
1341                                 min(fm->sb->s_bdi->ra_pages, ra_pages);
1342                 fc->minor = arg->minor;
1343                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
1344                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
1345                 fc->conn_init = 1;
1346         }
1347         kfree(ia);
1348
1349         if (!ok) {
1350                 fc->conn_init = 0;
1351                 fc->conn_error = 1;
1352         }
1353
1354         fuse_set_initialized(fc);
1355         wake_up_all(&fc->blocked_waitq);
1356 }
1357
1358 void fuse_send_init(struct fuse_mount *fm)
1359 {
1360         struct fuse_init_args *ia;
1361         u64 flags;
1362
1363         ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
1364
1365         ia->in.major = FUSE_KERNEL_VERSION;
1366         ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
1367         ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE;
1368         flags =
1369                 FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
1370                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
1371                 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1372                 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1373                 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1374                 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1375                 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1376                 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1377                 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA |
1378                 FUSE_HANDLE_KILLPRIV_V2 | FUSE_SETXATTR_EXT | FUSE_INIT_EXT |
1379                 FUSE_SECURITY_CTX | FUSE_CREATE_SUPP_GROUP |
1380                 FUSE_HAS_EXPIRE_ONLY | FUSE_DIRECT_IO_ALLOW_MMAP |
1381                 FUSE_NO_EXPORT_SUPPORT | FUSE_HAS_RESEND;
1382 #ifdef CONFIG_FUSE_DAX
1383         if (fm->fc->dax)
1384                 flags |= FUSE_MAP_ALIGNMENT;
1385         if (fuse_is_inode_dax_mode(fm->fc->dax_mode))
1386                 flags |= FUSE_HAS_INODE_DAX;
1387 #endif
1388         if (fm->fc->auto_submounts)
1389                 flags |= FUSE_SUBMOUNTS;
1390         if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH))
1391                 flags |= FUSE_PASSTHROUGH;
1392
1393         ia->in.flags = flags;
1394         ia->in.flags2 = flags >> 32;
1395
1396         ia->args.opcode = FUSE_INIT;
1397         ia->args.in_numargs = 1;
1398         ia->args.in_args[0].size = sizeof(ia->in);
1399         ia->args.in_args[0].value = &ia->in;
1400         ia->args.out_numargs = 1;
1401         /* Variable length argument used for backward compatibility
1402            with interface version < 7.5.  Rest of init_out is zeroed
1403            by do_get_request(), so a short reply is not a problem */
1404         ia->args.out_argvar = true;
1405         ia->args.out_args[0].size = sizeof(ia->out);
1406         ia->args.out_args[0].value = &ia->out;
1407         ia->args.force = true;
1408         ia->args.nocreds = true;
1409         ia->args.end = process_init_reply;
1410
1411         if (fuse_simple_background(fm, &ia->args, GFP_KERNEL) != 0)
1412                 process_init_reply(fm, &ia->args, -ENOTCONN);
1413 }
1414 EXPORT_SYMBOL_GPL(fuse_send_init);
1415
1416 void fuse_free_conn(struct fuse_conn *fc)
1417 {
1418         WARN_ON(!list_empty(&fc->devices));
1419         kfree(fc);
1420 }
1421 EXPORT_SYMBOL_GPL(fuse_free_conn);
1422
1423 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1424 {
1425         int err;
1426         char *suffix = "";
1427
1428         if (sb->s_bdev) {
1429                 suffix = "-fuseblk";
1430                 /*
1431                  * sb->s_bdi points to blkdev's bdi however we want to redirect
1432                  * it to our private bdi...
1433                  */
1434                 bdi_put(sb->s_bdi);
1435                 sb->s_bdi = &noop_backing_dev_info;
1436         }
1437         err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1438                                    MINOR(fc->dev), suffix);
1439         if (err)
1440                 return err;
1441
1442         /* fuse does it's own writeback accounting */
1443         sb->s_bdi->capabilities &= ~BDI_CAP_WRITEBACK_ACCT;
1444         sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT;
1445
1446         /*
1447          * For a single fuse filesystem use max 1% of dirty +
1448          * writeback threshold.
1449          *
1450          * This gives about 1M of write buffer for memory maps on a
1451          * machine with 1G and 10% dirty_ratio, which should be more
1452          * than enough.
1453          *
1454          * Privileged users can raise it by writing to
1455          *
1456          *    /sys/class/bdi/<bdi>/max_ratio
1457          */
1458         bdi_set_max_ratio(sb->s_bdi, 1);
1459
1460         return 0;
1461 }
1462
1463 struct fuse_dev *fuse_dev_alloc(void)
1464 {
1465         struct fuse_dev *fud;
1466         struct list_head *pq;
1467
1468         fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1469         if (!fud)
1470                 return NULL;
1471
1472         pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1473         if (!pq) {
1474                 kfree(fud);
1475                 return NULL;
1476         }
1477
1478         fud->pq.processing = pq;
1479         fuse_pqueue_init(&fud->pq);
1480
1481         return fud;
1482 }
1483 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1484
1485 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1486 {
1487         fud->fc = fuse_conn_get(fc);
1488         spin_lock(&fc->lock);
1489         list_add_tail(&fud->entry, &fc->devices);
1490         spin_unlock(&fc->lock);
1491 }
1492 EXPORT_SYMBOL_GPL(fuse_dev_install);
1493
1494 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1495 {
1496         struct fuse_dev *fud;
1497
1498         fud = fuse_dev_alloc();
1499         if (!fud)
1500                 return NULL;
1501
1502         fuse_dev_install(fud, fc);
1503         return fud;
1504 }
1505 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1506
1507 void fuse_dev_free(struct fuse_dev *fud)
1508 {
1509         struct fuse_conn *fc = fud->fc;
1510
1511         if (fc) {
1512                 spin_lock(&fc->lock);
1513                 list_del(&fud->entry);
1514                 spin_unlock(&fc->lock);
1515
1516                 fuse_conn_put(fc);
1517         }
1518         kfree(fud->pq.processing);
1519         kfree(fud);
1520 }
1521 EXPORT_SYMBOL_GPL(fuse_dev_free);
1522
1523 static void fuse_fill_attr_from_inode(struct fuse_attr *attr,
1524                                       const struct fuse_inode *fi)
1525 {
1526         struct timespec64 atime = inode_get_atime(&fi->inode);
1527         struct timespec64 mtime = inode_get_mtime(&fi->inode);
1528         struct timespec64 ctime = inode_get_ctime(&fi->inode);
1529
1530         *attr = (struct fuse_attr){
1531                 .ino            = fi->inode.i_ino,
1532                 .size           = fi->inode.i_size,
1533                 .blocks         = fi->inode.i_blocks,
1534                 .atime          = atime.tv_sec,
1535                 .mtime          = mtime.tv_sec,
1536                 .ctime          = ctime.tv_sec,
1537                 .atimensec      = atime.tv_nsec,
1538                 .mtimensec      = mtime.tv_nsec,
1539                 .ctimensec      = ctime.tv_nsec,
1540                 .mode           = fi->inode.i_mode,
1541                 .nlink          = fi->inode.i_nlink,
1542                 .uid            = __kuid_val(fi->inode.i_uid),
1543                 .gid            = __kgid_val(fi->inode.i_gid),
1544                 .rdev           = fi->inode.i_rdev,
1545                 .blksize        = 1u << fi->inode.i_blkbits,
1546         };
1547 }
1548
1549 static void fuse_sb_defaults(struct super_block *sb)
1550 {
1551         sb->s_magic = FUSE_SUPER_MAGIC;
1552         sb->s_op = &fuse_super_operations;
1553         sb->s_xattr = fuse_xattr_handlers;
1554         sb->s_maxbytes = MAX_LFS_FILESIZE;
1555         sb->s_time_gran = 1;
1556         sb->s_export_op = &fuse_export_operations;
1557         sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1558         if (sb->s_user_ns != &init_user_ns)
1559                 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1560         sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1561 }
1562
1563 static int fuse_fill_super_submount(struct super_block *sb,
1564                                     struct fuse_inode *parent_fi)
1565 {
1566         struct fuse_mount *fm = get_fuse_mount_super(sb);
1567         struct super_block *parent_sb = parent_fi->inode.i_sb;
1568         struct fuse_attr root_attr;
1569         struct inode *root;
1570         struct fuse_submount_lookup *sl;
1571         struct fuse_inode *fi;
1572
1573         fuse_sb_defaults(sb);
1574         fm->sb = sb;
1575
1576         WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1577         sb->s_bdi = bdi_get(parent_sb->s_bdi);
1578
1579         sb->s_xattr = parent_sb->s_xattr;
1580         sb->s_export_op = parent_sb->s_export_op;
1581         sb->s_time_gran = parent_sb->s_time_gran;
1582         sb->s_blocksize = parent_sb->s_blocksize;
1583         sb->s_blocksize_bits = parent_sb->s_blocksize_bits;
1584         sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL);
1585         if (parent_sb->s_subtype && !sb->s_subtype)
1586                 return -ENOMEM;
1587
1588         fuse_fill_attr_from_inode(&root_attr, parent_fi);
1589         root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0);
1590         /*
1591          * This inode is just a duplicate, so it is not looked up and
1592          * its nlookup should not be incremented.  fuse_iget() does
1593          * that, though, so undo it here.
1594          */
1595         fi = get_fuse_inode(root);
1596         fi->nlookup--;
1597
1598         sb->s_d_op = &fuse_dentry_operations;
1599         sb->s_root = d_make_root(root);
1600         if (!sb->s_root)
1601                 return -ENOMEM;
1602
1603         /*
1604          * Grab the parent's submount_lookup pointer and take a
1605          * reference on the shared nlookup from the parent.  This is to
1606          * prevent the last forget for this nodeid from getting
1607          * triggered until all users have finished with it.
1608          */
1609         sl = parent_fi->submount_lookup;
1610         WARN_ON(!sl);
1611         if (sl) {
1612                 refcount_inc(&sl->count);
1613                 fi->submount_lookup = sl;
1614         }
1615
1616         return 0;
1617 }
1618
1619 /* Filesystem context private data holds the FUSE inode of the mount point */
1620 static int fuse_get_tree_submount(struct fs_context *fsc)
1621 {
1622         struct fuse_mount *fm;
1623         struct fuse_inode *mp_fi = fsc->fs_private;
1624         struct fuse_conn *fc = get_fuse_conn(&mp_fi->inode);
1625         struct super_block *sb;
1626         int err;
1627
1628         fm = kzalloc(sizeof(struct fuse_mount), GFP_KERNEL);
1629         if (!fm)
1630                 return -ENOMEM;
1631
1632         fm->fc = fuse_conn_get(fc);
1633         fsc->s_fs_info = fm;
1634         sb = sget_fc(fsc, NULL, set_anon_super_fc);
1635         if (fsc->s_fs_info)
1636                 fuse_mount_destroy(fm);
1637         if (IS_ERR(sb))
1638                 return PTR_ERR(sb);
1639
1640         /* Initialize superblock, making @mp_fi its root */
1641         err = fuse_fill_super_submount(sb, mp_fi);
1642         if (err) {
1643                 deactivate_locked_super(sb);
1644                 return err;
1645         }
1646
1647         down_write(&fc->killsb);
1648         list_add_tail(&fm->fc_entry, &fc->mounts);
1649         up_write(&fc->killsb);
1650
1651         sb->s_flags |= SB_ACTIVE;
1652         fsc->root = dget(sb->s_root);
1653
1654         return 0;
1655 }
1656
1657 static const struct fs_context_operations fuse_context_submount_ops = {
1658         .get_tree       = fuse_get_tree_submount,
1659 };
1660
1661 int fuse_init_fs_context_submount(struct fs_context *fsc)
1662 {
1663         fsc->ops = &fuse_context_submount_ops;
1664         return 0;
1665 }
1666 EXPORT_SYMBOL_GPL(fuse_init_fs_context_submount);
1667
1668 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1669 {
1670         struct fuse_dev *fud = NULL;
1671         struct fuse_mount *fm = get_fuse_mount_super(sb);
1672         struct fuse_conn *fc = fm->fc;
1673         struct inode *root;
1674         struct dentry *root_dentry;
1675         int err;
1676
1677         err = -EINVAL;
1678         if (sb->s_flags & SB_MANDLOCK)
1679                 goto err;
1680
1681         rcu_assign_pointer(fc->curr_bucket, fuse_sync_bucket_alloc());
1682         fuse_sb_defaults(sb);
1683
1684         if (ctx->is_bdev) {
1685 #ifdef CONFIG_BLOCK
1686                 err = -EINVAL;
1687                 if (!sb_set_blocksize(sb, ctx->blksize))
1688                         goto err;
1689 #endif
1690         } else {
1691                 sb->s_blocksize = PAGE_SIZE;
1692                 sb->s_blocksize_bits = PAGE_SHIFT;
1693         }
1694
1695         sb->s_subtype = ctx->subtype;
1696         ctx->subtype = NULL;
1697         if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1698                 err = fuse_dax_conn_alloc(fc, ctx->dax_mode, ctx->dax_dev);
1699                 if (err)
1700                         goto err;
1701         }
1702
1703         if (ctx->fudptr) {
1704                 err = -ENOMEM;
1705                 fud = fuse_dev_alloc_install(fc);
1706                 if (!fud)
1707                         goto err_free_dax;
1708         }
1709
1710         fc->dev = sb->s_dev;
1711         fm->sb = sb;
1712         err = fuse_bdi_init(fc, sb);
1713         if (err)
1714                 goto err_dev_free;
1715
1716         /* Handle umasking inside the fuse code */
1717         if (sb->s_flags & SB_POSIXACL)
1718                 fc->dont_mask = 1;
1719         sb->s_flags |= SB_POSIXACL;
1720
1721         fc->default_permissions = ctx->default_permissions;
1722         fc->allow_other = ctx->allow_other;
1723         fc->user_id = ctx->user_id;
1724         fc->group_id = ctx->group_id;
1725         fc->legacy_opts_show = ctx->legacy_opts_show;
1726         fc->max_read = max_t(unsigned int, 4096, ctx->max_read);
1727         fc->destroy = ctx->destroy;
1728         fc->no_control = ctx->no_control;
1729         fc->no_force_umount = ctx->no_force_umount;
1730
1731         err = -ENOMEM;
1732         root = fuse_get_root_inode(sb, ctx->rootmode);
1733         sb->s_d_op = &fuse_root_dentry_operations;
1734         root_dentry = d_make_root(root);
1735         if (!root_dentry)
1736                 goto err_dev_free;
1737         /* Root dentry doesn't have .d_revalidate */
1738         sb->s_d_op = &fuse_dentry_operations;
1739
1740         mutex_lock(&fuse_mutex);
1741         err = -EINVAL;
1742         if (ctx->fudptr && *ctx->fudptr)
1743                 goto err_unlock;
1744
1745         err = fuse_ctl_add_conn(fc);
1746         if (err)
1747                 goto err_unlock;
1748
1749         list_add_tail(&fc->entry, &fuse_conn_list);
1750         sb->s_root = root_dentry;
1751         if (ctx->fudptr)
1752                 *ctx->fudptr = fud;
1753         mutex_unlock(&fuse_mutex);
1754         return 0;
1755
1756  err_unlock:
1757         mutex_unlock(&fuse_mutex);
1758         dput(root_dentry);
1759  err_dev_free:
1760         if (fud)
1761                 fuse_dev_free(fud);
1762  err_free_dax:
1763         if (IS_ENABLED(CONFIG_FUSE_DAX))
1764                 fuse_dax_conn_free(fc);
1765  err:
1766         return err;
1767 }
1768 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1769
1770 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1771 {
1772         struct fuse_fs_context *ctx = fsc->fs_private;
1773         int err;
1774
1775         if (!ctx->file || !ctx->rootmode_present ||
1776             !ctx->user_id_present || !ctx->group_id_present)
1777                 return -EINVAL;
1778
1779         /*
1780          * Require mount to happen from the same user namespace which
1781          * opened /dev/fuse to prevent potential attacks.
1782          */
1783         if ((ctx->file->f_op != &fuse_dev_operations) ||
1784             (ctx->file->f_cred->user_ns != sb->s_user_ns))
1785                 return -EINVAL;
1786         ctx->fudptr = &ctx->file->private_data;
1787
1788         err = fuse_fill_super_common(sb, ctx);
1789         if (err)
1790                 return err;
1791         /* file->private_data shall be visible on all CPUs after this */
1792         smp_mb();
1793         fuse_send_init(get_fuse_mount_super(sb));
1794         return 0;
1795 }
1796
1797 /*
1798  * This is the path where user supplied an already initialized fuse dev.  In
1799  * this case never create a new super if the old one is gone.
1800  */
1801 static int fuse_set_no_super(struct super_block *sb, struct fs_context *fsc)
1802 {
1803         return -ENOTCONN;
1804 }
1805
1806 static int fuse_test_super(struct super_block *sb, struct fs_context *fsc)
1807 {
1808
1809         return fsc->sget_key == get_fuse_conn_super(sb);
1810 }
1811
1812 static int fuse_get_tree(struct fs_context *fsc)
1813 {
1814         struct fuse_fs_context *ctx = fsc->fs_private;
1815         struct fuse_dev *fud;
1816         struct fuse_conn *fc;
1817         struct fuse_mount *fm;
1818         struct super_block *sb;
1819         int err;
1820
1821         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1822         if (!fc)
1823                 return -ENOMEM;
1824
1825         fm = kzalloc(sizeof(*fm), GFP_KERNEL);
1826         if (!fm) {
1827                 kfree(fc);
1828                 return -ENOMEM;
1829         }
1830
1831         fuse_conn_init(fc, fm, fsc->user_ns, &fuse_dev_fiq_ops, NULL);
1832         fc->release = fuse_free_conn;
1833
1834         fsc->s_fs_info = fm;
1835
1836         if (ctx->fd_present)
1837                 ctx->file = fget(ctx->fd);
1838
1839         if (IS_ENABLED(CONFIG_BLOCK) && ctx->is_bdev) {
1840                 err = get_tree_bdev(fsc, fuse_fill_super);
1841                 goto out;
1842         }
1843         /*
1844          * While block dev mount can be initialized with a dummy device fd
1845          * (found by device name), normal fuse mounts can't
1846          */
1847         err = -EINVAL;
1848         if (!ctx->file)
1849                 goto out;
1850
1851         /*
1852          * Allow creating a fuse mount with an already initialized fuse
1853          * connection
1854          */
1855         fud = READ_ONCE(ctx->file->private_data);
1856         if (ctx->file->f_op == &fuse_dev_operations && fud) {
1857                 fsc->sget_key = fud->fc;
1858                 sb = sget_fc(fsc, fuse_test_super, fuse_set_no_super);
1859                 err = PTR_ERR_OR_ZERO(sb);
1860                 if (!IS_ERR(sb))
1861                         fsc->root = dget(sb->s_root);
1862         } else {
1863                 err = get_tree_nodev(fsc, fuse_fill_super);
1864         }
1865 out:
1866         if (fsc->s_fs_info)
1867                 fuse_mount_destroy(fm);
1868         if (ctx->file)
1869                 fput(ctx->file);
1870         return err;
1871 }
1872
1873 static const struct fs_context_operations fuse_context_ops = {
1874         .free           = fuse_free_fsc,
1875         .parse_param    = fuse_parse_param,
1876         .reconfigure    = fuse_reconfigure,
1877         .get_tree       = fuse_get_tree,
1878 };
1879
1880 /*
1881  * Set up the filesystem mount context.
1882  */
1883 static int fuse_init_fs_context(struct fs_context *fsc)
1884 {
1885         struct fuse_fs_context *ctx;
1886
1887         ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1888         if (!ctx)
1889                 return -ENOMEM;
1890
1891         ctx->max_read = ~0;
1892         ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1893         ctx->legacy_opts_show = true;
1894
1895 #ifdef CONFIG_BLOCK
1896         if (fsc->fs_type == &fuseblk_fs_type) {
1897                 ctx->is_bdev = true;
1898                 ctx->destroy = true;
1899         }
1900 #endif
1901
1902         fsc->fs_private = ctx;
1903         fsc->ops = &fuse_context_ops;
1904         return 0;
1905 }
1906
1907 bool fuse_mount_remove(struct fuse_mount *fm)
1908 {
1909         struct fuse_conn *fc = fm->fc;
1910         bool last = false;
1911
1912         down_write(&fc->killsb);
1913         list_del_init(&fm->fc_entry);
1914         if (list_empty(&fc->mounts))
1915                 last = true;
1916         up_write(&fc->killsb);
1917
1918         return last;
1919 }
1920 EXPORT_SYMBOL_GPL(fuse_mount_remove);
1921
1922 void fuse_conn_destroy(struct fuse_mount *fm)
1923 {
1924         struct fuse_conn *fc = fm->fc;
1925
1926         if (fc->destroy)
1927                 fuse_send_destroy(fm);
1928
1929         fuse_abort_conn(fc);
1930         fuse_wait_aborted(fc);
1931
1932         if (!list_empty(&fc->entry)) {
1933                 mutex_lock(&fuse_mutex);
1934                 list_del(&fc->entry);
1935                 fuse_ctl_remove_conn(fc);
1936                 mutex_unlock(&fuse_mutex);
1937         }
1938 }
1939 EXPORT_SYMBOL_GPL(fuse_conn_destroy);
1940
1941 static void fuse_sb_destroy(struct super_block *sb)
1942 {
1943         struct fuse_mount *fm = get_fuse_mount_super(sb);
1944         bool last;
1945
1946         if (sb->s_root) {
1947                 last = fuse_mount_remove(fm);
1948                 if (last)
1949                         fuse_conn_destroy(fm);
1950         }
1951 }
1952
1953 void fuse_mount_destroy(struct fuse_mount *fm)
1954 {
1955         fuse_conn_put(fm->fc);
1956         kfree_rcu(fm, rcu);
1957 }
1958 EXPORT_SYMBOL(fuse_mount_destroy);
1959
1960 static void fuse_kill_sb_anon(struct super_block *sb)
1961 {
1962         fuse_sb_destroy(sb);
1963         kill_anon_super(sb);
1964         fuse_mount_destroy(get_fuse_mount_super(sb));
1965 }
1966
1967 static struct file_system_type fuse_fs_type = {
1968         .owner          = THIS_MODULE,
1969         .name           = "fuse",
1970         .fs_flags       = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1971         .init_fs_context = fuse_init_fs_context,
1972         .parameters     = fuse_fs_parameters,
1973         .kill_sb        = fuse_kill_sb_anon,
1974 };
1975 MODULE_ALIAS_FS("fuse");
1976
1977 #ifdef CONFIG_BLOCK
1978 static void fuse_kill_sb_blk(struct super_block *sb)
1979 {
1980         fuse_sb_destroy(sb);
1981         kill_block_super(sb);
1982         fuse_mount_destroy(get_fuse_mount_super(sb));
1983 }
1984
1985 static struct file_system_type fuseblk_fs_type = {
1986         .owner          = THIS_MODULE,
1987         .name           = "fuseblk",
1988         .init_fs_context = fuse_init_fs_context,
1989         .parameters     = fuse_fs_parameters,
1990         .kill_sb        = fuse_kill_sb_blk,
1991         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1992 };
1993 MODULE_ALIAS_FS("fuseblk");
1994
1995 static inline int register_fuseblk(void)
1996 {
1997         return register_filesystem(&fuseblk_fs_type);
1998 }
1999
2000 static inline void unregister_fuseblk(void)
2001 {
2002         unregister_filesystem(&fuseblk_fs_type);
2003 }
2004 #else
2005 static inline int register_fuseblk(void)
2006 {
2007         return 0;
2008 }
2009
2010 static inline void unregister_fuseblk(void)
2011 {
2012 }
2013 #endif
2014
2015 static void fuse_inode_init_once(void *foo)
2016 {
2017         struct inode *inode = foo;
2018
2019         inode_init_once(inode);
2020 }
2021
2022 static int __init fuse_fs_init(void)
2023 {
2024         int err;
2025
2026         fuse_inode_cachep = kmem_cache_create("fuse_inode",
2027                         sizeof(struct fuse_inode), 0,
2028                         SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
2029                         fuse_inode_init_once);
2030         err = -ENOMEM;
2031         if (!fuse_inode_cachep)
2032                 goto out;
2033
2034         err = register_fuseblk();
2035         if (err)
2036                 goto out2;
2037
2038         err = register_filesystem(&fuse_fs_type);
2039         if (err)
2040                 goto out3;
2041
2042         return 0;
2043
2044  out3:
2045         unregister_fuseblk();
2046  out2:
2047         kmem_cache_destroy(fuse_inode_cachep);
2048  out:
2049         return err;
2050 }
2051
2052 static void fuse_fs_cleanup(void)
2053 {
2054         unregister_filesystem(&fuse_fs_type);
2055         unregister_fuseblk();
2056
2057         /*
2058          * Make sure all delayed rcu free inodes are flushed before we
2059          * destroy cache.
2060          */
2061         rcu_barrier();
2062         kmem_cache_destroy(fuse_inode_cachep);
2063 }
2064
2065 static struct kobject *fuse_kobj;
2066
2067 static int fuse_sysfs_init(void)
2068 {
2069         int err;
2070
2071         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
2072         if (!fuse_kobj) {
2073                 err = -ENOMEM;
2074                 goto out_err;
2075         }
2076
2077         err = sysfs_create_mount_point(fuse_kobj, "connections");
2078         if (err)
2079                 goto out_fuse_unregister;
2080
2081         return 0;
2082
2083  out_fuse_unregister:
2084         kobject_put(fuse_kobj);
2085  out_err:
2086         return err;
2087 }
2088
2089 static void fuse_sysfs_cleanup(void)
2090 {
2091         sysfs_remove_mount_point(fuse_kobj, "connections");
2092         kobject_put(fuse_kobj);
2093 }
2094
2095 static int __init fuse_init(void)
2096 {
2097         int res;
2098
2099         pr_info("init (API version %i.%i)\n",
2100                 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
2101
2102         INIT_LIST_HEAD(&fuse_conn_list);
2103         res = fuse_fs_init();
2104         if (res)
2105                 goto err;
2106
2107         res = fuse_dev_init();
2108         if (res)
2109                 goto err_fs_cleanup;
2110
2111         res = fuse_sysfs_init();
2112         if (res)
2113                 goto err_dev_cleanup;
2114
2115         res = fuse_ctl_init();
2116         if (res)
2117                 goto err_sysfs_cleanup;
2118
2119         sanitize_global_limit(&max_user_bgreq);
2120         sanitize_global_limit(&max_user_congthresh);
2121
2122         return 0;
2123
2124  err_sysfs_cleanup:
2125         fuse_sysfs_cleanup();
2126  err_dev_cleanup:
2127         fuse_dev_cleanup();
2128  err_fs_cleanup:
2129         fuse_fs_cleanup();
2130  err:
2131         return res;
2132 }
2133
2134 static void __exit fuse_exit(void)
2135 {
2136         pr_debug("exit\n");
2137
2138         fuse_ctl_cleanup();
2139         fuse_sysfs_cleanup();
2140         fuse_fs_cleanup();
2141         fuse_dev_cleanup();
2142 }
2143
2144 module_init(fuse_init);
2145 module_exit(fuse_exit);