fs/fuse/inode.c

   1 /*
   2   FUSE: Filesystem in Userspace
   3   Copyright (C) 2001-2008  Miklos Szeredi <miklos@szeredi.hu>
   4
   5   This program can be distributed under the terms of the GNU GPL.
   6   See the file COPYING.
   7 */
   8
   9 #include "fuse_i.h"
  10
  11 #include <linux/pagemap.h>
  12 #include <linux/slab.h>
  13 #include <linux/file.h>
  14 #include <linux/seq_file.h>
  15 #include <linux/init.h>
  16 #include <linux/module.h>
  17 #include <linux/moduleparam.h>
  18 #include <linux/fs_context.h>
  19 #include <linux/fs_parser.h>
  20 #include <linux/statfs.h>
  21 #include <linux/random.h>
  22 #include <linux/sched.h>
  23 #include <linux/exportfs.h>
  24 #include <linux/posix_acl.h>
  25 #include <linux/pid_namespace.h>
  26
  27 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>");
  28 MODULE_DESCRIPTION("Filesystem in Userspace");
  29 MODULE_LICENSE("GPL");
  30
  31 static struct kmem_cache *fuse_inode_cachep;
  32 struct list_head fuse_conn_list;
  33 DEFINE_MUTEX(fuse_mutex);
  34
  35 static int set_global_limit(const char *val, const struct kernel_param *kp);
  36
  37 unsigned max_user_bgreq;
  38 module_param_call(max_user_bgreq, set_global_limit, param_get_uint,
  39                   &max_user_bgreq, 0644);
  40 __MODULE_PARM_TYPE(max_user_bgreq, "uint");
  41 MODULE_PARM_DESC(max_user_bgreq,
  42  "Global limit for the maximum number of backgrounded requests an "
  43  "unprivileged user can set");
  44
  45 unsigned max_user_congthresh;
  46 module_param_call(max_user_congthresh, set_global_limit, param_get_uint,
  47                   &max_user_congthresh, 0644);
  48 __MODULE_PARM_TYPE(max_user_congthresh, "uint");
  49 MODULE_PARM_DESC(max_user_congthresh,
  50  "Global limit for the maximum congestion threshold an "
  51  "unprivileged user can set");
  52
  53 #define FUSE_SUPER_MAGIC 0x65735546
  54
  55 #define FUSE_DEFAULT_BLKSIZE 512
  56
  57 /** Maximum number of outstanding background requests */
  58 #define FUSE_DEFAULT_MAX_BACKGROUND 12
  59
  60 /** Congestion starts at 75% of maximum */
  61 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4)
  62
  63 #ifdef CONFIG_BLOCK
  64 static struct file_system_type fuseblk_fs_type;
  65 #endif
  66
  67 struct fuse_forget_link *fuse_alloc_forget(void)
  68 {
  69         return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL_ACCOUNT);
  70 }
  71
  72 static struct inode *fuse_alloc_inode(struct super_block *sb)
  73 {
  74         struct fuse_inode *fi;
  75
  76         fi = kmem_cache_alloc(fuse_inode_cachep, GFP_KERNEL);
  77         if (!fi)
  78                 return NULL;
  79
  80         fi->i_time = 0;
  81         fi->inval_mask = 0;
  82         fi->nodeid = 0;
  83         fi->nlookup = 0;
  84         fi->attr_version = 0;
  85         fi->orig_ino = 0;
  86         fi->state = 0;
  87         mutex_init(&fi->mutex);
  88         init_rwsem(&fi->i_mmap_sem);
  89         spin_lock_init(&fi->lock);
  90         fi->forget = fuse_alloc_forget();
  91         if (!fi->forget)
  92                 goto out_free;
  93
  94         if (IS_ENABLED(CONFIG_FUSE_DAX) && !fuse_dax_inode_alloc(sb, fi))
  95                 goto out_free_forget;
  96
  97         return &fi->inode;
  98
  99 out_free_forget:
 100         kfree(fi->forget);
 101 out_free:
 102         kmem_cache_free(fuse_inode_cachep, fi);
 103         return NULL;
 104 }
 105
 106 static void fuse_free_inode(struct inode *inode)
 107 {
 108         struct fuse_inode *fi = get_fuse_inode(inode);
 109
 110         mutex_destroy(&fi->mutex);
 111         kfree(fi->forget);
 112 #ifdef CONFIG_FUSE_DAX
 113         kfree(fi->dax);
 114 #endif
 115         kmem_cache_free(fuse_inode_cachep, fi);
 116 }
 117
 118 static void fuse_evict_inode(struct inode *inode)
 119 {
 120         struct fuse_inode *fi = get_fuse_inode(inode);
 121
 122         truncate_inode_pages_final(&inode->i_data);
 123         clear_inode(inode);
 124         if (inode->i_sb->s_flags & SB_ACTIVE) {
 125                 struct fuse_conn *fc = get_fuse_conn(inode);
 126
 127                 if (FUSE_IS_DAX(inode))
 128                         fuse_dax_inode_cleanup(inode);
 129                 if (fi->nlookup) {
 130                         fuse_queue_forget(fc, fi->forget, fi->nodeid,
 131                                           fi->nlookup);
 132                         fi->forget = NULL;
 133                 }
 134         }
 135         if (S_ISREG(inode->i_mode) && !is_bad_inode(inode)) {
 136                 WARN_ON(!list_empty(&fi->write_files));
 137                 WARN_ON(!list_empty(&fi->queued_writes));
 138         }
 139 }
 140
 141 static int fuse_reconfigure(struct fs_context *fc)
 142 {
 143         struct super_block *sb = fc->root->d_sb;
 144
 145         sync_filesystem(sb);
 146         if (fc->sb_flags & SB_MANDLOCK)
 147                 return -EINVAL;
 148
 149         return 0;
 150 }
 151
 152 /*
 153  * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down
 154  * so that it will fit.
 155  */
 156 static ino_t fuse_squash_ino(u64 ino64)
 157 {
 158         ino_t ino = (ino_t) ino64;
 159         if (sizeof(ino_t) < sizeof(u64))
 160                 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8;
 161         return ino;
 162 }
 163
 164 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr,
 165                                    u64 attr_valid)
 166 {
 167         struct fuse_conn *fc = get_fuse_conn(inode);
 168         struct fuse_inode *fi = get_fuse_inode(inode);
 169
 170         lockdep_assert_held(&fi->lock);
 171
 172         fi->attr_version = atomic64_inc_return(&fc->attr_version);
 173         fi->i_time = attr_valid;
 174         WRITE_ONCE(fi->inval_mask, 0);
 175
 176         inode->i_ino     = fuse_squash_ino(attr->ino);
 177         inode->i_mode    = (inode->i_mode & S_IFMT) | (attr->mode & 07777);
 178         set_nlink(inode, attr->nlink);
 179         inode->i_uid     = make_kuid(fc->user_ns, attr->uid);
 180         inode->i_gid     = make_kgid(fc->user_ns, attr->gid);
 181         inode->i_blocks  = attr->blocks;
 182         inode->i_atime.tv_sec   = attr->atime;
 183         inode->i_atime.tv_nsec  = attr->atimensec;
 184         /* mtime from server may be stale due to local buffered write */
 185         if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) {
 186                 inode->i_mtime.tv_sec   = attr->mtime;
 187                 inode->i_mtime.tv_nsec  = attr->mtimensec;
 188                 inode->i_ctime.tv_sec   = attr->ctime;
 189                 inode->i_ctime.tv_nsec  = attr->ctimensec;
 190         }
 191
 192         if (attr->blksize != 0)
 193                 inode->i_blkbits = ilog2(attr->blksize);
 194         else
 195                 inode->i_blkbits = inode->i_sb->s_blocksize_bits;
 196
 197         /*
 198          * Don't set the sticky bit in i_mode, unless we want the VFS
 199          * to check permissions.  This prevents failures due to the
 200          * check in may_delete().
 201          */
 202         fi->orig_i_mode = inode->i_mode;
 203         if (!fc->default_permissions)
 204                 inode->i_mode &= ~S_ISVTX;
 205
 206         fi->orig_ino = attr->ino;
 207 }
 208
 209 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr,
 210                             u64 attr_valid, u64 attr_version)
 211 {
 212         struct fuse_conn *fc = get_fuse_conn(inode);
 213         struct fuse_inode *fi = get_fuse_inode(inode);
 214         bool is_wb = fc->writeback_cache;
 215         loff_t oldsize;
 216         struct timespec64 old_mtime;
 217
 218         spin_lock(&fi->lock);
 219         if ((attr_version != 0 && fi->attr_version > attr_version) ||
 220             test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) {
 221                 spin_unlock(&fi->lock);
 222                 return;
 223         }
 224
 225         old_mtime = inode->i_mtime;
 226         fuse_change_attributes_common(inode, attr, attr_valid);
 227
 228         oldsize = inode->i_size;
 229         /*
 230          * In case of writeback_cache enabled, the cached writes beyond EOF
 231          * extend local i_size without keeping userspace server in sync. So,
 232          * attr->size coming from server can be stale. We cannot trust it.
 233          */
 234         if (!is_wb || !S_ISREG(inode->i_mode))
 235                 i_size_write(inode, attr->size);
 236         spin_unlock(&fi->lock);
 237
 238         if (!is_wb && S_ISREG(inode->i_mode)) {
 239                 bool inval = false;
 240
 241                 if (oldsize != attr->size) {
 242                         truncate_pagecache(inode, attr->size);
 243                         if (!fc->explicit_inval_data)
 244                                 inval = true;
 245                 } else if (fc->auto_inval_data) {
 246                         struct timespec64 new_mtime = {
 247                                 .tv_sec = attr->mtime,
 248                                 .tv_nsec = attr->mtimensec,
 249                         };
 250
 251                         /*
 252                          * Auto inval mode also checks and invalidates if mtime
 253                          * has changed.
 254                          */
 255                         if (!timespec64_equal(&old_mtime, &new_mtime))
 256                                 inval = true;
 257                 }
 258
 259                 if (inval)
 260                         invalidate_inode_pages2(inode->i_mapping);
 261         }
 262 }
 263
 264 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr)
 265 {
 266         inode->i_mode = attr->mode & S_IFMT;
 267         inode->i_size = attr->size;
 268         inode->i_mtime.tv_sec  = attr->mtime;
 269         inode->i_mtime.tv_nsec = attr->mtimensec;
 270         inode->i_ctime.tv_sec  = attr->ctime;
 271         inode->i_ctime.tv_nsec = attr->ctimensec;
 272         if (S_ISREG(inode->i_mode)) {
 273                 fuse_init_common(inode);
 274                 fuse_init_file_inode(inode);
 275         } else if (S_ISDIR(inode->i_mode))
 276                 fuse_init_dir(inode);
 277         else if (S_ISLNK(inode->i_mode))
 278                 fuse_init_symlink(inode);
 279         else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) ||
 280                  S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) {
 281                 fuse_init_common(inode);
 282                 init_special_inode(inode, inode->i_mode,
 283                                    new_decode_dev(attr->rdev));
 284         } else
 285                 BUG();
 286 }
 287
 288 static int fuse_inode_eq(struct inode *inode, void *_nodeidp)
 289 {
 290         u64 nodeid = *(u64 *) _nodeidp;
 291         if (get_node_id(inode) == nodeid)
 292                 return 1;
 293         else
 294                 return 0;
 295 }
 296
 297 static int fuse_inode_set(struct inode *inode, void *_nodeidp)
 298 {
 299         u64 nodeid = *(u64 *) _nodeidp;
 300         get_fuse_inode(inode)->nodeid = nodeid;
 301         return 0;
 302 }
 303
 304 struct inode *fuse_iget(struct super_block *sb, u64 nodeid,
 305                         int generation, struct fuse_attr *attr,
 306                         u64 attr_valid, u64 attr_version)
 307 {
 308         struct inode *inode;
 309         struct fuse_inode *fi;
 310         struct fuse_conn *fc = get_fuse_conn_super(sb);
 311
 312         /*
 313          * Auto mount points get their node id from the submount root, which is
 314          * not a unique identifier within this filesystem.
 315          *
 316          * To avoid conflicts, do not place submount points into the inode hash
 317          * table.
 318          */
 319         if (fc->auto_submounts && (attr->flags & FUSE_ATTR_SUBMOUNT) &&
 320             S_ISDIR(attr->mode)) {
 321                 inode = new_inode(sb);
 322                 if (!inode)
 323                         return NULL;
 324
 325                 fuse_init_inode(inode, attr);
 326                 get_fuse_inode(inode)->nodeid = nodeid;
 327                 inode->i_flags |= S_AUTOMOUNT;
 328                 goto done;
 329         }
 330
 331 retry:
 332         inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid);
 333         if (!inode)
 334                 return NULL;
 335
 336         if ((inode->i_state & I_NEW)) {
 337                 inode->i_flags |= S_NOATIME;
 338                 if (!fc->writeback_cache || !S_ISREG(attr->mode))
 339                         inode->i_flags |= S_NOCMTIME;
 340                 inode->i_generation = generation;
 341                 fuse_init_inode(inode, attr);
 342                 unlock_new_inode(inode);
 343         } else if ((inode->i_mode ^ attr->mode) & S_IFMT) {
 344                 /* Inode has changed type, any I/O on the old should fail */
 345                 make_bad_inode(inode);
 346                 iput(inode);
 347                 goto retry;
 348         }
 349 done:
 350         fi = get_fuse_inode(inode);
 351         spin_lock(&fi->lock);
 352         fi->nlookup++;
 353         spin_unlock(&fi->lock);
 354         fuse_change_attributes(inode, attr, attr_valid, attr_version);
 355
 356         return inode;
 357 }
 358
 359 struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid,
 360                            struct fuse_mount **fm)
 361 {
 362         struct fuse_mount *fm_iter;
 363         struct inode *inode;
 364
 365         WARN_ON(!rwsem_is_locked(&fc->killsb));
 366         list_for_each_entry(fm_iter, &fc->mounts, fc_entry) {
 367                 if (!fm_iter->sb)
 368                         continue;
 369
 370                 inode = ilookup5(fm_iter->sb, nodeid, fuse_inode_eq, &nodeid);
 371                 if (inode) {
 372                         if (fm)
 373                                 *fm = fm_iter;
 374                         return inode;
 375                 }
 376         }
 377
 378         return NULL;
 379 }
 380
 381 int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid,
 382                              loff_t offset, loff_t len)
 383 {
 384         struct fuse_inode *fi;
 385         struct inode *inode;
 386         pgoff_t pg_start;
 387         pgoff_t pg_end;
 388
 389         inode = fuse_ilookup(fc, nodeid, NULL);
 390         if (!inode)
 391                 return -ENOENT;
 392
 393         fi = get_fuse_inode(inode);
 394         spin_lock(&fi->lock);
 395         fi->attr_version = atomic64_inc_return(&fc->attr_version);
 396         spin_unlock(&fi->lock);
 397
 398         fuse_invalidate_attr(inode);
 399         forget_all_cached_acls(inode);
 400         if (offset >= 0) {
 401                 pg_start = offset >> PAGE_SHIFT;
 402                 if (len <= 0)
 403                         pg_end = -1;
 404                 else
 405                         pg_end = (offset + len - 1) >> PAGE_SHIFT;
 406                 invalidate_inode_pages2_range(inode->i_mapping,
 407                                               pg_start, pg_end);
 408         }
 409         iput(inode);
 410         return 0;
 411 }
 412
 413 bool fuse_lock_inode(struct inode *inode)
 414 {
 415         bool locked = false;
 416
 417         if (!get_fuse_conn(inode)->parallel_dirops) {
 418                 mutex_lock(&get_fuse_inode(inode)->mutex);
 419                 locked = true;
 420         }
 421
 422         return locked;
 423 }
 424
 425 void fuse_unlock_inode(struct inode *inode, bool locked)
 426 {
 427         if (locked)
 428                 mutex_unlock(&get_fuse_inode(inode)->mutex);
 429 }
 430
 431 static void fuse_umount_begin(struct super_block *sb)
 432 {
 433         struct fuse_conn *fc = get_fuse_conn_super(sb);
 434
 435         if (!fc->no_force_umount)
 436                 fuse_abort_conn(fc);
 437 }
 438
 439 static void fuse_send_destroy(struct fuse_mount *fm)
 440 {
 441         if (fm->fc->conn_init) {
 442                 FUSE_ARGS(args);
 443
 444                 args.opcode = FUSE_DESTROY;
 445                 args.force = true;
 446                 args.nocreds = true;
 447                 fuse_simple_request(fm, &args);
 448         }
 449 }
 450
 451 static void fuse_put_super(struct super_block *sb)
 452 {
 453         struct fuse_mount *fm = get_fuse_mount_super(sb);
 454
 455         fuse_mount_put(fm);
 456 }
 457
 458 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr)
 459 {
 460         stbuf->f_type    = FUSE_SUPER_MAGIC;
 461         stbuf->f_bsize   = attr->bsize;
 462         stbuf->f_frsize  = attr->frsize;
 463         stbuf->f_blocks  = attr->blocks;
 464         stbuf->f_bfree   = attr->bfree;
 465         stbuf->f_bavail  = attr->bavail;
 466         stbuf->f_files   = attr->files;
 467         stbuf->f_ffree   = attr->ffree;
 468         stbuf->f_namelen = attr->namelen;
 469         /* fsid is left zero */
 470 }
 471
 472 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf)
 473 {
 474         struct super_block *sb = dentry->d_sb;
 475         struct fuse_mount *fm = get_fuse_mount_super(sb);
 476         FUSE_ARGS(args);
 477         struct fuse_statfs_out outarg;
 478         int err;
 479
 480         if (!fuse_allow_current_process(fm->fc)) {
 481                 buf->f_type = FUSE_SUPER_MAGIC;
 482                 return 0;
 483         }
 484
 485         memset(&outarg, 0, sizeof(outarg));
 486         args.in_numargs = 0;
 487         args.opcode = FUSE_STATFS;
 488         args.nodeid = get_node_id(d_inode(dentry));
 489         args.out_numargs = 1;
 490         args.out_args[0].size = sizeof(outarg);
 491         args.out_args[0].value = &outarg;
 492         err = fuse_simple_request(fm, &args);
 493         if (!err)
 494                 convert_fuse_statfs(buf, &outarg.st);
 495         return err;
 496 }
 497
 498 enum {
 499         OPT_SOURCE,
 500         OPT_SUBTYPE,
 501         OPT_FD,
 502         OPT_ROOTMODE,
 503         OPT_USER_ID,
 504         OPT_GROUP_ID,
 505         OPT_DEFAULT_PERMISSIONS,
 506         OPT_ALLOW_OTHER,
 507         OPT_MAX_READ,
 508         OPT_BLKSIZE,
 509         OPT_ERR
 510 };
 511
 512 static const struct fs_parameter_spec fuse_fs_parameters[] = {
 513         fsparam_string  ("source",              OPT_SOURCE),
 514         fsparam_u32     ("fd",                  OPT_FD),
 515         fsparam_u32oct  ("rootmode",            OPT_ROOTMODE),
 516         fsparam_u32     ("user_id",             OPT_USER_ID),
 517         fsparam_u32     ("group_id",            OPT_GROUP_ID),
 518         fsparam_flag    ("default_permissions", OPT_DEFAULT_PERMISSIONS),
 519         fsparam_flag    ("allow_other",         OPT_ALLOW_OTHER),
 520         fsparam_u32     ("max_read",            OPT_MAX_READ),
 521         fsparam_u32     ("blksize",             OPT_BLKSIZE),
 522         fsparam_string  ("subtype",             OPT_SUBTYPE),
 523         {}
 524 };
 525
 526 static int fuse_parse_param(struct fs_context *fc, struct fs_parameter *param)
 527 {
 528         struct fs_parse_result result;
 529         struct fuse_fs_context *ctx = fc->fs_private;
 530         int opt;
 531
 532         if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) {
 533                 /*
 534                  * Ignore options coming from mount(MS_REMOUNT) for backward
 535                  * compatibility.
 536                  */
 537                 if (fc->oldapi)
 538                         return 0;
 539
 540                 return invalfc(fc, "No changes allowed in reconfigure");
 541         }
 542
 543         opt = fs_parse(fc, fuse_fs_parameters, param, &result);
 544         if (opt < 0)
 545                 return opt;
 546
 547         switch (opt) {
 548         case OPT_SOURCE:
 549                 if (fc->source)
 550                         return invalfc(fc, "Multiple sources specified");
 551                 fc->source = param->string;
 552                 param->string = NULL;
 553                 break;
 554
 555         case OPT_SUBTYPE:
 556                 if (ctx->subtype)
 557                         return invalfc(fc, "Multiple subtypes specified");
 558                 ctx->subtype = param->string;
 559                 param->string = NULL;
 560                 return 0;
 561
 562         case OPT_FD:
 563                 ctx->fd = result.uint_32;
 564                 ctx->fd_present = true;
 565                 break;
 566
 567         case OPT_ROOTMODE:
 568                 if (!fuse_valid_type(result.uint_32))
 569                         return invalfc(fc, "Invalid rootmode");
 570                 ctx->rootmode = result.uint_32;
 571                 ctx->rootmode_present = true;
 572                 break;
 573
 574         case OPT_USER_ID:
 575                 ctx->user_id = make_kuid(fc->user_ns, result.uint_32);
 576                 if (!uid_valid(ctx->user_id))
 577                         return invalfc(fc, "Invalid user_id");
 578                 ctx->user_id_present = true;
 579                 break;
 580
 581         case OPT_GROUP_ID:
 582                 ctx->group_id = make_kgid(fc->user_ns, result.uint_32);
 583                 if (!gid_valid(ctx->group_id))
 584                         return invalfc(fc, "Invalid group_id");
 585                 ctx->group_id_present = true;
 586                 break;
 587
 588         case OPT_DEFAULT_PERMISSIONS:
 589                 ctx->default_permissions = true;
 590                 break;
 591
 592         case OPT_ALLOW_OTHER:
 593                 ctx->allow_other = true;
 594                 break;
 595
 596         case OPT_MAX_READ:
 597                 ctx->max_read = result.uint_32;
 598                 break;
 599
 600         case OPT_BLKSIZE:
 601                 if (!ctx->is_bdev)
 602                         return invalfc(fc, "blksize only supported for fuseblk");
 603                 ctx->blksize = result.uint_32;
 604                 break;
 605
 606         default:
 607                 return -EINVAL;
 608         }
 609
 610         return 0;
 611 }
 612
 613 static void fuse_free_fc(struct fs_context *fc)
 614 {
 615         struct fuse_fs_context *ctx = fc->fs_private;
 616
 617         if (ctx) {
 618                 kfree(ctx->subtype);
 619                 kfree(ctx);
 620         }
 621 }
 622
 623 static int fuse_show_options(struct seq_file *m, struct dentry *root)
 624 {
 625         struct super_block *sb = root->d_sb;
 626         struct fuse_conn *fc = get_fuse_conn_super(sb);
 627
 628         if (fc->legacy_opts_show) {
 629                 seq_printf(m, ",user_id=%u",
 630                            from_kuid_munged(fc->user_ns, fc->user_id));
 631                 seq_printf(m, ",group_id=%u",
 632                            from_kgid_munged(fc->user_ns, fc->group_id));
 633                 if (fc->default_permissions)
 634                         seq_puts(m, ",default_permissions");
 635                 if (fc->allow_other)
 636                         seq_puts(m, ",allow_other");
 637                 if (fc->max_read != ~0)
 638                         seq_printf(m, ",max_read=%u", fc->max_read);
 639                 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE)
 640                         seq_printf(m, ",blksize=%lu", sb->s_blocksize);
 641         }
 642 #ifdef CONFIG_FUSE_DAX
 643         if (fc->dax)
 644                 seq_puts(m, ",dax");
 645 #endif
 646
 647         return 0;
 648 }
 649
 650 static void fuse_iqueue_init(struct fuse_iqueue *fiq,
 651                              const struct fuse_iqueue_ops *ops,
 652                              void *priv)
 653 {
 654         memset(fiq, 0, sizeof(struct fuse_iqueue));
 655         spin_lock_init(&fiq->lock);
 656         init_waitqueue_head(&fiq->waitq);
 657         INIT_LIST_HEAD(&fiq->pending);
 658         INIT_LIST_HEAD(&fiq->interrupts);
 659         fiq->forget_list_tail = &fiq->forget_list_head;
 660         fiq->connected = 1;
 661         fiq->ops = ops;
 662         fiq->priv = priv;
 663 }
 664
 665 static void fuse_pqueue_init(struct fuse_pqueue *fpq)
 666 {
 667         unsigned int i;
 668
 669         spin_lock_init(&fpq->lock);
 670         for (i = 0; i < FUSE_PQ_HASH_SIZE; i++)
 671                 INIT_LIST_HEAD(&fpq->processing[i]);
 672         INIT_LIST_HEAD(&fpq->io);
 673         fpq->connected = 1;
 674 }
 675
 676 void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm,
 677                     struct user_namespace *user_ns,
 678                     const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv)
 679 {
 680         memset(fc, 0, sizeof(*fc));
 681         spin_lock_init(&fc->lock);
 682         spin_lock_init(&fc->bg_lock);
 683         init_rwsem(&fc->killsb);
 684         refcount_set(&fc->count, 1);
 685         atomic_set(&fc->dev_count, 1);
 686         init_waitqueue_head(&fc->blocked_waitq);
 687         fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv);
 688         INIT_LIST_HEAD(&fc->bg_queue);
 689         INIT_LIST_HEAD(&fc->entry);
 690         INIT_LIST_HEAD(&fc->devices);
 691         atomic_set(&fc->num_waiting, 0);
 692         fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND;
 693         fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD;
 694         atomic64_set(&fc->khctr, 0);
 695         fc->polled_files = RB_ROOT;
 696         fc->blocked = 0;
 697         fc->initialized = 0;
 698         fc->connected = 1;
 699         atomic64_set(&fc->attr_version, 1);
 700         get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key));
 701         fc->pid_ns = get_pid_ns(task_active_pid_ns(current));
 702         fc->user_ns = get_user_ns(user_ns);
 703         fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ;
 704
 705         INIT_LIST_HEAD(&fc->mounts);
 706         list_add(&fm->fc_entry, &fc->mounts);
 707         fm->fc = fc;
 708         refcount_set(&fm->count, 1);
 709 }
 710 EXPORT_SYMBOL_GPL(fuse_conn_init);
 711
 712 void fuse_conn_put(struct fuse_conn *fc)
 713 {
 714         if (refcount_dec_and_test(&fc->count)) {
 715                 struct fuse_iqueue *fiq = &fc->iq;
 716
 717                 if (IS_ENABLED(CONFIG_FUSE_DAX))
 718                         fuse_dax_conn_free(fc);
 719                 if (fiq->ops->release)
 720                         fiq->ops->release(fiq);
 721                 put_pid_ns(fc->pid_ns);
 722                 put_user_ns(fc->user_ns);
 723                 fc->release(fc);
 724         }
 725 }
 726 EXPORT_SYMBOL_GPL(fuse_conn_put);
 727
 728 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc)
 729 {
 730         refcount_inc(&fc->count);
 731         return fc;
 732 }
 733 EXPORT_SYMBOL_GPL(fuse_conn_get);
 734
 735 void fuse_mount_put(struct fuse_mount *fm)
 736 {
 737         if (refcount_dec_and_test(&fm->count)) {
 738                 if (fm->fc)
 739                         fuse_conn_put(fm->fc);
 740                 kfree(fm);
 741         }
 742 }
 743 EXPORT_SYMBOL_GPL(fuse_mount_put);
 744
 745 struct fuse_mount *fuse_mount_get(struct fuse_mount *fm)
 746 {
 747         refcount_inc(&fm->count);
 748         return fm;
 749 }
 750 EXPORT_SYMBOL_GPL(fuse_mount_get);
 751
 752 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode)
 753 {
 754         struct fuse_attr attr;
 755         memset(&attr, 0, sizeof(attr));
 756
 757         attr.mode = mode;
 758         attr.ino = FUSE_ROOT_ID;
 759         attr.nlink = 1;
 760         return fuse_iget(sb, 1, 0, &attr, 0, 0);
 761 }
 762
 763 struct fuse_inode_handle {
 764         u64 nodeid;
 765         u32 generation;
 766 };
 767
 768 static struct dentry *fuse_get_dentry(struct super_block *sb,
 769                                       struct fuse_inode_handle *handle)
 770 {
 771         struct fuse_conn *fc = get_fuse_conn_super(sb);
 772         struct inode *inode;
 773         struct dentry *entry;
 774         int err = -ESTALE;
 775
 776         if (handle->nodeid == 0)
 777                 goto out_err;
 778
 779         inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid);
 780         if (!inode) {
 781                 struct fuse_entry_out outarg;
 782                 const struct qstr name = QSTR_INIT(".", 1);
 783
 784                 if (!fc->export_support)
 785                         goto out_err;
 786
 787                 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg,
 788                                        &inode);
 789                 if (err && err != -ENOENT)
 790                         goto out_err;
 791                 if (err || !inode) {
 792                         err = -ESTALE;
 793                         goto out_err;
 794                 }
 795                 err = -EIO;
 796                 if (get_node_id(inode) != handle->nodeid)
 797                         goto out_iput;
 798         }
 799         err = -ESTALE;
 800         if (inode->i_generation != handle->generation)
 801                 goto out_iput;
 802
 803         entry = d_obtain_alias(inode);
 804         if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID)
 805                 fuse_invalidate_entry_cache(entry);
 806
 807         return entry;
 808
 809  out_iput:
 810         iput(inode);
 811  out_err:
 812         return ERR_PTR(err);
 813 }
 814
 815 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len,
 816                            struct inode *parent)
 817 {
 818         int len = parent ? 6 : 3;
 819         u64 nodeid;
 820         u32 generation;
 821
 822         if (*max_len < len) {
 823                 *max_len = len;
 824                 return  FILEID_INVALID;
 825         }
 826
 827         nodeid = get_fuse_inode(inode)->nodeid;
 828         generation = inode->i_generation;
 829
 830         fh[0] = (u32)(nodeid >> 32);
 831         fh[1] = (u32)(nodeid & 0xffffffff);
 832         fh[2] = generation;
 833
 834         if (parent) {
 835                 nodeid = get_fuse_inode(parent)->nodeid;
 836                 generation = parent->i_generation;
 837
 838                 fh[3] = (u32)(nodeid >> 32);
 839                 fh[4] = (u32)(nodeid & 0xffffffff);
 840                 fh[5] = generation;
 841         }
 842
 843         *max_len = len;
 844         return parent ? 0x82 : 0x81;
 845 }
 846
 847 static struct dentry *fuse_fh_to_dentry(struct super_block *sb,
 848                 struct fid *fid, int fh_len, int fh_type)
 849 {
 850         struct fuse_inode_handle handle;
 851
 852         if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3)
 853                 return NULL;
 854
 855         handle.nodeid = (u64) fid->raw[0] << 32;
 856         handle.nodeid |= (u64) fid->raw[1];
 857         handle.generation = fid->raw[2];
 858         return fuse_get_dentry(sb, &handle);
 859 }
 860
 861 static struct dentry *fuse_fh_to_parent(struct super_block *sb,
 862                 struct fid *fid, int fh_len, int fh_type)
 863 {
 864         struct fuse_inode_handle parent;
 865
 866         if (fh_type != 0x82 || fh_len < 6)
 867                 return NULL;
 868
 869         parent.nodeid = (u64) fid->raw[3] << 32;
 870         parent.nodeid |= (u64) fid->raw[4];
 871         parent.generation = fid->raw[5];
 872         return fuse_get_dentry(sb, &parent);
 873 }
 874
 875 static struct dentry *fuse_get_parent(struct dentry *child)
 876 {
 877         struct inode *child_inode = d_inode(child);
 878         struct fuse_conn *fc = get_fuse_conn(child_inode);
 879         struct inode *inode;
 880         struct dentry *parent;
 881         struct fuse_entry_out outarg;
 882         const struct qstr name = QSTR_INIT("..", 2);
 883         int err;
 884
 885         if (!fc->export_support)
 886                 return ERR_PTR(-ESTALE);
 887
 888         err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode),
 889                                &name, &outarg, &inode);
 890         if (err) {
 891                 if (err == -ENOENT)
 892                         return ERR_PTR(-ESTALE);
 893                 return ERR_PTR(err);
 894         }
 895
 896         parent = d_obtain_alias(inode);
 897         if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID)
 898                 fuse_invalidate_entry_cache(parent);
 899
 900         return parent;
 901 }
 902
 903 static const struct export_operations fuse_export_operations = {
 904         .fh_to_dentry   = fuse_fh_to_dentry,
 905         .fh_to_parent   = fuse_fh_to_parent,
 906         .encode_fh      = fuse_encode_fh,
 907         .get_parent     = fuse_get_parent,
 908 };
 909
 910 static const struct super_operations fuse_super_operations = {
 911         .alloc_inode    = fuse_alloc_inode,
 912         .free_inode     = fuse_free_inode,
 913         .evict_inode    = fuse_evict_inode,
 914         .write_inode    = fuse_write_inode,
 915         .drop_inode     = generic_delete_inode,
 916         .put_super      = fuse_put_super,
 917         .umount_begin   = fuse_umount_begin,
 918         .statfs         = fuse_statfs,
 919         .show_options   = fuse_show_options,
 920 };
 921
 922 static void sanitize_global_limit(unsigned *limit)
 923 {
 924         /*
 925          * The default maximum number of async requests is calculated to consume
 926          * 1/2^13 of the total memory, assuming 392 bytes per request.
 927          */
 928         if (*limit == 0)
 929                 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392;
 930
 931         if (*limit >= 1 << 16)
 932                 *limit = (1 << 16) - 1;
 933 }
 934
 935 static int set_global_limit(const char *val, const struct kernel_param *kp)
 936 {
 937         int rv;
 938
 939         rv = param_set_uint(val, kp);
 940         if (rv)
 941                 return rv;
 942
 943         sanitize_global_limit((unsigned *)kp->arg);
 944
 945         return 0;
 946 }
 947
 948 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg)
 949 {
 950         int cap_sys_admin = capable(CAP_SYS_ADMIN);
 951
 952         if (arg->minor < 13)
 953                 return;
 954
 955         sanitize_global_limit(&max_user_bgreq);
 956         sanitize_global_limit(&max_user_congthresh);
 957
 958         spin_lock(&fc->bg_lock);
 959         if (arg->max_background) {
 960                 fc->max_background = arg->max_background;
 961
 962                 if (!cap_sys_admin && fc->max_background > max_user_bgreq)
 963                         fc->max_background = max_user_bgreq;
 964         }
 965         if (arg->congestion_threshold) {
 966                 fc->congestion_threshold = arg->congestion_threshold;
 967
 968                 if (!cap_sys_admin &&
 969                     fc->congestion_threshold > max_user_congthresh)
 970                         fc->congestion_threshold = max_user_congthresh;
 971         }
 972         spin_unlock(&fc->bg_lock);
 973 }
 974
 975 struct fuse_init_args {
 976         struct fuse_args args;
 977         struct fuse_init_in in;
 978         struct fuse_init_out out;
 979 };
 980
 981 static void process_init_reply(struct fuse_mount *fm, struct fuse_args *args,
 982                                int error)
 983 {
 984         struct fuse_conn *fc = fm->fc;
 985         struct fuse_init_args *ia = container_of(args, typeof(*ia), args);
 986         struct fuse_init_out *arg = &ia->out;
 987         bool ok = true;
 988
 989         if (error || arg->major != FUSE_KERNEL_VERSION)
 990                 ok = false;
 991         else {
 992                 unsigned long ra_pages;
 993
 994                 process_init_limits(fc, arg);
 995
 996                 if (arg->minor >= 6) {
 997                         ra_pages = arg->max_readahead / PAGE_SIZE;
 998                         if (arg->flags & FUSE_ASYNC_READ)
 999                                 fc->async_read = 1;
1000                         if (!(arg->flags & FUSE_POSIX_LOCKS))
1001                                 fc->no_lock = 1;
1002                         if (arg->minor >= 17) {
1003                                 if (!(arg->flags & FUSE_FLOCK_LOCKS))
1004                                         fc->no_flock = 1;
1005                         } else {
1006                                 if (!(arg->flags & FUSE_POSIX_LOCKS))
1007                                         fc->no_flock = 1;
1008                         }
1009                         if (arg->flags & FUSE_ATOMIC_O_TRUNC)
1010                                 fc->atomic_o_trunc = 1;
1011                         if (arg->minor >= 9) {
1012                                 /* LOOKUP has dependency on proto version */
1013                                 if (arg->flags & FUSE_EXPORT_SUPPORT)
1014                                         fc->export_support = 1;
1015                         }
1016                         if (arg->flags & FUSE_BIG_WRITES)
1017                                 fc->big_writes = 1;
1018                         if (arg->flags & FUSE_DONT_MASK)
1019                                 fc->dont_mask = 1;
1020                         if (arg->flags & FUSE_AUTO_INVAL_DATA)
1021                                 fc->auto_inval_data = 1;
1022                         else if (arg->flags & FUSE_EXPLICIT_INVAL_DATA)
1023                                 fc->explicit_inval_data = 1;
1024                         if (arg->flags & FUSE_DO_READDIRPLUS) {
1025                                 fc->do_readdirplus = 1;
1026                                 if (arg->flags & FUSE_READDIRPLUS_AUTO)
1027                                         fc->readdirplus_auto = 1;
1028                         }
1029                         if (arg->flags & FUSE_ASYNC_DIO)
1030                                 fc->async_dio = 1;
1031                         if (arg->flags & FUSE_WRITEBACK_CACHE)
1032                                 fc->writeback_cache = 1;
1033                         if (arg->flags & FUSE_PARALLEL_DIROPS)
1034                                 fc->parallel_dirops = 1;
1035                         if (arg->flags & FUSE_HANDLE_KILLPRIV)
1036                                 fc->handle_killpriv = 1;
1037                         if (arg->time_gran && arg->time_gran <= 1000000000)
1038                                 fm->sb->s_time_gran = arg->time_gran;
1039                         if ((arg->flags & FUSE_POSIX_ACL)) {
1040                                 fc->default_permissions = 1;
1041                                 fc->posix_acl = 1;
1042                                 fm->sb->s_xattr = fuse_acl_xattr_handlers;
1043                         }
1044                         if (arg->flags & FUSE_CACHE_SYMLINKS)
1045                                 fc->cache_symlinks = 1;
1046                         if (arg->flags & FUSE_ABORT_ERROR)
1047                                 fc->abort_err = 1;
1048                         if (arg->flags & FUSE_MAX_PAGES) {
1049                                 fc->max_pages =
1050                                         min_t(unsigned int, FUSE_MAX_MAX_PAGES,
1051                                         max_t(unsigned int, arg->max_pages, 1));
1052                         }
1053                         if (IS_ENABLED(CONFIG_FUSE_DAX) &&
1054                             arg->flags & FUSE_MAP_ALIGNMENT &&
1055                             !fuse_dax_check_alignment(fc, arg->map_alignment)) {
1056                                 ok = false;
1057                         }
1058                 } else {
1059                         ra_pages = fc->max_read / PAGE_SIZE;
1060                         fc->no_lock = 1;
1061                         fc->no_flock = 1;
1062                 }
1063
1064                 fm->sb->s_bdi->ra_pages =
1065                                 min(fm->sb->s_bdi->ra_pages, ra_pages);
1066                 fc->minor = arg->minor;
1067                 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write;
1068                 fc->max_write = max_t(unsigned, 4096, fc->max_write);
1069                 fc->conn_init = 1;
1070         }
1071         kfree(ia);
1072
1073         if (!ok) {
1074                 fc->conn_init = 0;
1075                 fc->conn_error = 1;
1076         }
1077
1078         fuse_set_initialized(fc);
1079         wake_up_all(&fc->blocked_waitq);
1080 }
1081
1082 void fuse_send_init(struct fuse_mount *fm)
1083 {
1084         struct fuse_init_args *ia;
1085
1086         ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL);
1087
1088         ia->in.major = FUSE_KERNEL_VERSION;
1089         ia->in.minor = FUSE_KERNEL_MINOR_VERSION;
1090         ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE;
1091         ia->in.flags |=
1092                 FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC |
1093                 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK |
1094                 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ |
1095                 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA |
1096                 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO |
1097                 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT |
1098                 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL |
1099                 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS |
1100                 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA;
1101 #ifdef CONFIG_FUSE_DAX
1102         if (fm->fc->dax)
1103                 ia->in.flags |= FUSE_MAP_ALIGNMENT;
1104 #endif
1105         if (fm->fc->auto_submounts)
1106                 ia->in.flags |= FUSE_SUBMOUNTS;
1107
1108         ia->args.opcode = FUSE_INIT;
1109         ia->args.in_numargs = 1;
1110         ia->args.in_args[0].size = sizeof(ia->in);
1111         ia->args.in_args[0].value = &ia->in;
1112         ia->args.out_numargs = 1;
1113         /* Variable length argument used for backward compatibility
1114            with interface version < 7.5.  Rest of init_out is zeroed
1115            by do_get_request(), so a short reply is not a problem */
1116         ia->args.out_argvar = true;
1117         ia->args.out_args[0].size = sizeof(ia->out);
1118         ia->args.out_args[0].value = &ia->out;
1119         ia->args.force = true;
1120         ia->args.nocreds = true;
1121         ia->args.end = process_init_reply;
1122
1123         if (fuse_simple_background(fm, &ia->args, GFP_KERNEL) != 0)
1124                 process_init_reply(fm, &ia->args, -ENOTCONN);
1125 }
1126 EXPORT_SYMBOL_GPL(fuse_send_init);
1127
1128 void fuse_free_conn(struct fuse_conn *fc)
1129 {
1130         WARN_ON(!list_empty(&fc->devices));
1131         kfree_rcu(fc, rcu);
1132 }
1133 EXPORT_SYMBOL_GPL(fuse_free_conn);
1134
1135 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb)
1136 {
1137         int err;
1138         char *suffix = "";
1139
1140         if (sb->s_bdev) {
1141                 suffix = "-fuseblk";
1142                 /*
1143                  * sb->s_bdi points to blkdev's bdi however we want to redirect
1144                  * it to our private bdi...
1145                  */
1146                 bdi_put(sb->s_bdi);
1147                 sb->s_bdi = &noop_backing_dev_info;
1148         }
1149         err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev),
1150                                    MINOR(fc->dev), suffix);
1151         if (err)
1152                 return err;
1153
1154         /* fuse does it's own writeback accounting */
1155         sb->s_bdi->capabilities &= ~BDI_CAP_WRITEBACK_ACCT;
1156         sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT;
1157
1158         /*
1159          * For a single fuse filesystem use max 1% of dirty +
1160          * writeback threshold.
1161          *
1162          * This gives about 1M of write buffer for memory maps on a
1163          * machine with 1G and 10% dirty_ratio, which should be more
1164          * than enough.
1165          *
1166          * Privileged users can raise it by writing to
1167          *
1168          *    /sys/class/bdi/<bdi>/max_ratio
1169          */
1170         bdi_set_max_ratio(sb->s_bdi, 1);
1171
1172         return 0;
1173 }
1174
1175 struct fuse_dev *fuse_dev_alloc(void)
1176 {
1177         struct fuse_dev *fud;
1178         struct list_head *pq;
1179
1180         fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL);
1181         if (!fud)
1182                 return NULL;
1183
1184         pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL);
1185         if (!pq) {
1186                 kfree(fud);
1187                 return NULL;
1188         }
1189
1190         fud->pq.processing = pq;
1191         fuse_pqueue_init(&fud->pq);
1192
1193         return fud;
1194 }
1195 EXPORT_SYMBOL_GPL(fuse_dev_alloc);
1196
1197 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc)
1198 {
1199         fud->fc = fuse_conn_get(fc);
1200         spin_lock(&fc->lock);
1201         list_add_tail(&fud->entry, &fc->devices);
1202         spin_unlock(&fc->lock);
1203 }
1204 EXPORT_SYMBOL_GPL(fuse_dev_install);
1205
1206 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc)
1207 {
1208         struct fuse_dev *fud;
1209
1210         fud = fuse_dev_alloc();
1211         if (!fud)
1212                 return NULL;
1213
1214         fuse_dev_install(fud, fc);
1215         return fud;
1216 }
1217 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install);
1218
1219 void fuse_dev_free(struct fuse_dev *fud)
1220 {
1221         struct fuse_conn *fc = fud->fc;
1222
1223         if (fc) {
1224                 spin_lock(&fc->lock);
1225                 list_del(&fud->entry);
1226                 spin_unlock(&fc->lock);
1227
1228                 fuse_conn_put(fc);
1229         }
1230         kfree(fud->pq.processing);
1231         kfree(fud);
1232 }
1233 EXPORT_SYMBOL_GPL(fuse_dev_free);
1234
1235 static void fuse_fill_attr_from_inode(struct fuse_attr *attr,
1236                                       const struct fuse_inode *fi)
1237 {
1238         *attr = (struct fuse_attr){
1239                 .ino            = fi->inode.i_ino,
1240                 .size           = fi->inode.i_size,
1241                 .blocks         = fi->inode.i_blocks,
1242                 .atime          = fi->inode.i_atime.tv_sec,
1243                 .mtime          = fi->inode.i_mtime.tv_sec,
1244                 .ctime          = fi->inode.i_ctime.tv_sec,
1245                 .atimensec      = fi->inode.i_atime.tv_nsec,
1246                 .mtimensec      = fi->inode.i_mtime.tv_nsec,
1247                 .ctimensec      = fi->inode.i_ctime.tv_nsec,
1248                 .mode           = fi->inode.i_mode,
1249                 .nlink          = fi->inode.i_nlink,
1250                 .uid            = fi->inode.i_uid.val,
1251                 .gid            = fi->inode.i_gid.val,
1252                 .rdev           = fi->inode.i_rdev,
1253                 .blksize        = 1u << fi->inode.i_blkbits,
1254         };
1255 }
1256
1257 static void fuse_sb_defaults(struct super_block *sb)
1258 {
1259         sb->s_magic = FUSE_SUPER_MAGIC;
1260         sb->s_op = &fuse_super_operations;
1261         sb->s_xattr = fuse_xattr_handlers;
1262         sb->s_maxbytes = MAX_LFS_FILESIZE;
1263         sb->s_time_gran = 1;
1264         sb->s_export_op = &fuse_export_operations;
1265         sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE;
1266         if (sb->s_user_ns != &init_user_ns)
1267                 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER;
1268         sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION);
1269
1270         /*
1271          * If we are not in the initial user namespace posix
1272          * acls must be translated.
1273          */
1274         if (sb->s_user_ns != &init_user_ns)
1275                 sb->s_xattr = fuse_no_acl_xattr_handlers;
1276 }
1277
1278 int fuse_fill_super_submount(struct super_block *sb,
1279                              struct fuse_inode *parent_fi)
1280 {
1281         struct fuse_mount *fm = get_fuse_mount_super(sb);
1282         struct super_block *parent_sb = parent_fi->inode.i_sb;
1283         struct fuse_attr root_attr;
1284         struct inode *root;
1285
1286         fuse_sb_defaults(sb);
1287         fm->sb = sb;
1288
1289         WARN_ON(sb->s_bdi != &noop_backing_dev_info);
1290         sb->s_bdi = bdi_get(parent_sb->s_bdi);
1291
1292         sb->s_xattr = parent_sb->s_xattr;
1293         sb->s_time_gran = parent_sb->s_time_gran;
1294         sb->s_blocksize = parent_sb->s_blocksize;
1295         sb->s_blocksize_bits = parent_sb->s_blocksize_bits;
1296         sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL);
1297         if (parent_sb->s_subtype && !sb->s_subtype)
1298                 return -ENOMEM;
1299
1300         fuse_fill_attr_from_inode(&root_attr, parent_fi);
1301         root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0);
1302         /*
1303          * This inode is just a duplicate, so it is not looked up and
1304          * its nlookup should not be incremented.  fuse_iget() does
1305          * that, though, so undo it here.
1306          */
1307         get_fuse_inode(root)->nlookup--;
1308         sb->s_d_op = &fuse_dentry_operations;
1309         sb->s_root = d_make_root(root);
1310         if (!sb->s_root)
1311                 return -ENOMEM;
1312
1313         return 0;
1314 }
1315
1316 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx)
1317 {
1318         struct fuse_dev *fud = NULL;
1319         struct fuse_mount *fm = get_fuse_mount_super(sb);
1320         struct fuse_conn *fc = fm->fc;
1321         struct inode *root;
1322         struct dentry *root_dentry;
1323         int err;
1324
1325         err = -EINVAL;
1326         if (sb->s_flags & SB_MANDLOCK)
1327                 goto err;
1328
1329         fuse_sb_defaults(sb);
1330
1331         if (ctx->is_bdev) {
1332 #ifdef CONFIG_BLOCK
1333                 err = -EINVAL;
1334                 if (!sb_set_blocksize(sb, ctx->blksize))
1335                         goto err;
1336 #endif
1337         } else {
1338                 sb->s_blocksize = PAGE_SIZE;
1339                 sb->s_blocksize_bits = PAGE_SHIFT;
1340         }
1341
1342         sb->s_subtype = ctx->subtype;
1343         ctx->subtype = NULL;
1344         if (IS_ENABLED(CONFIG_FUSE_DAX)) {
1345                 err = fuse_dax_conn_alloc(fc, ctx->dax_dev);
1346                 if (err)
1347                         goto err;
1348         }
1349
1350         if (ctx->fudptr) {
1351                 err = -ENOMEM;
1352                 fud = fuse_dev_alloc_install(fc);
1353                 if (!fud)
1354                         goto err_free_dax;
1355         }
1356
1357         fc->dev = sb->s_dev;
1358         fm->sb = sb;
1359         err = fuse_bdi_init(fc, sb);
1360         if (err)
1361                 goto err_dev_free;
1362
1363         /* Handle umasking inside the fuse code */
1364         if (sb->s_flags & SB_POSIXACL)
1365                 fc->dont_mask = 1;
1366         sb->s_flags |= SB_POSIXACL;
1367
1368         fc->default_permissions = ctx->default_permissions;
1369         fc->allow_other = ctx->allow_other;
1370         fc->user_id = ctx->user_id;
1371         fc->group_id = ctx->group_id;
1372         fc->legacy_opts_show = ctx->legacy_opts_show;
1373         fc->max_read = max_t(unsigned int, 4096, ctx->max_read);
1374         fc->destroy = ctx->destroy;
1375         fc->no_control = ctx->no_control;
1376         fc->no_force_umount = ctx->no_force_umount;
1377
1378         err = -ENOMEM;
1379         root = fuse_get_root_inode(sb, ctx->rootmode);
1380         sb->s_d_op = &fuse_root_dentry_operations;
1381         root_dentry = d_make_root(root);
1382         if (!root_dentry)
1383                 goto err_dev_free;
1384         /* Root dentry doesn't have .d_revalidate */
1385         sb->s_d_op = &fuse_dentry_operations;
1386
1387         mutex_lock(&fuse_mutex);
1388         err = -EINVAL;
1389         if (ctx->fudptr && *ctx->fudptr)
1390                 goto err_unlock;
1391
1392         err = fuse_ctl_add_conn(fc);
1393         if (err)
1394                 goto err_unlock;
1395
1396         list_add_tail(&fc->entry, &fuse_conn_list);
1397         sb->s_root = root_dentry;
1398         if (ctx->fudptr)
1399                 *ctx->fudptr = fud;
1400         mutex_unlock(&fuse_mutex);
1401         return 0;
1402
1403  err_unlock:
1404         mutex_unlock(&fuse_mutex);
1405         dput(root_dentry);
1406  err_dev_free:
1407         if (fud)
1408                 fuse_dev_free(fud);
1409  err_free_dax:
1410         if (IS_ENABLED(CONFIG_FUSE_DAX))
1411                 fuse_dax_conn_free(fc);
1412  err:
1413         return err;
1414 }
1415 EXPORT_SYMBOL_GPL(fuse_fill_super_common);
1416
1417 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc)
1418 {
1419         struct fuse_fs_context *ctx = fsc->fs_private;
1420         struct file *file;
1421         int err;
1422         struct fuse_conn *fc;
1423         struct fuse_mount *fm;
1424
1425         err = -EINVAL;
1426         file = fget(ctx->fd);
1427         if (!file)
1428                 goto err;
1429
1430         /*
1431          * Require mount to happen from the same user namespace which
1432          * opened /dev/fuse to prevent potential attacks.
1433          */
1434         if ((file->f_op != &fuse_dev_operations) ||
1435             (file->f_cred->user_ns != sb->s_user_ns))
1436                 goto err_fput;
1437         ctx->fudptr = &file->private_data;
1438
1439         fc = kmalloc(sizeof(*fc), GFP_KERNEL);
1440         err = -ENOMEM;
1441         if (!fc)
1442                 goto err_fput;
1443
1444         fm = kzalloc(sizeof(*fm), GFP_KERNEL);
1445         if (!fm) {
1446                 kfree(fc);
1447                 goto err_fput;
1448         }
1449
1450         fuse_conn_init(fc, fm, sb->s_user_ns, &fuse_dev_fiq_ops, NULL);
1451         fc->release = fuse_free_conn;
1452
1453         sb->s_fs_info = fm;
1454
1455         err = fuse_fill_super_common(sb, ctx);
1456         if (err)
1457                 goto err_put_conn;
1458         /*
1459          * atomic_dec_and_test() in fput() provides the necessary
1460          * memory barrier for file->private_data to be visible on all
1461          * CPUs after this
1462          */
1463         fput(file);
1464         fuse_send_init(get_fuse_mount_super(sb));
1465         return 0;
1466
1467  err_put_conn:
1468         fuse_mount_put(fm);
1469         sb->s_fs_info = NULL;
1470  err_fput:
1471         fput(file);
1472  err:
1473         return err;
1474 }
1475
1476 static int fuse_get_tree(struct fs_context *fc)
1477 {
1478         struct fuse_fs_context *ctx = fc->fs_private;
1479
1480         if (!ctx->fd_present || !ctx->rootmode_present ||
1481             !ctx->user_id_present || !ctx->group_id_present)
1482                 return -EINVAL;
1483
1484 #ifdef CONFIG_BLOCK
1485         if (ctx->is_bdev)
1486                 return get_tree_bdev(fc, fuse_fill_super);
1487 #endif
1488
1489         return get_tree_nodev(fc, fuse_fill_super);
1490 }
1491
1492 static const struct fs_context_operations fuse_context_ops = {
1493         .free           = fuse_free_fc,
1494         .parse_param    = fuse_parse_param,
1495         .reconfigure    = fuse_reconfigure,
1496         .get_tree       = fuse_get_tree,
1497 };
1498
1499 /*
1500  * Set up the filesystem mount context.
1501  */
1502 static int fuse_init_fs_context(struct fs_context *fc)
1503 {
1504         struct fuse_fs_context *ctx;
1505
1506         ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL);
1507         if (!ctx)
1508                 return -ENOMEM;
1509
1510         ctx->max_read = ~0;
1511         ctx->blksize = FUSE_DEFAULT_BLKSIZE;
1512         ctx->legacy_opts_show = true;
1513
1514 #ifdef CONFIG_BLOCK
1515         if (fc->fs_type == &fuseblk_fs_type) {
1516                 ctx->is_bdev = true;
1517                 ctx->destroy = true;
1518         }
1519 #endif
1520
1521         fc->fs_private = ctx;
1522         fc->ops = &fuse_context_ops;
1523         return 0;
1524 }
1525
1526 bool fuse_mount_remove(struct fuse_mount *fm)
1527 {
1528         struct fuse_conn *fc = fm->fc;
1529         bool last = false;
1530
1531         down_write(&fc->killsb);
1532         list_del_init(&fm->fc_entry);
1533         if (list_empty(&fc->mounts))
1534                 last = true;
1535         up_write(&fc->killsb);
1536
1537         return last;
1538 }
1539 EXPORT_SYMBOL_GPL(fuse_mount_remove);
1540
1541 void fuse_conn_destroy(struct fuse_mount *fm)
1542 {
1543         struct fuse_conn *fc = fm->fc;
1544
1545         if (fc->destroy)
1546                 fuse_send_destroy(fm);
1547
1548         fuse_abort_conn(fc);
1549         fuse_wait_aborted(fc);
1550
1551         if (!list_empty(&fc->entry)) {
1552                 mutex_lock(&fuse_mutex);
1553                 list_del(&fc->entry);
1554                 fuse_ctl_remove_conn(fc);
1555                 mutex_unlock(&fuse_mutex);
1556         }
1557 }
1558 EXPORT_SYMBOL_GPL(fuse_conn_destroy);
1559
1560 static void fuse_kill_sb_anon(struct super_block *sb)
1561 {
1562         struct fuse_mount *fm = get_fuse_mount_super(sb);
1563         bool last;
1564
1565         if (fm) {
1566                 last = fuse_mount_remove(fm);
1567                 if (last)
1568                         fuse_conn_destroy(fm);
1569         }
1570         kill_anon_super(sb);
1571 }
1572
1573 static struct file_system_type fuse_fs_type = {
1574         .owner          = THIS_MODULE,
1575         .name           = "fuse",
1576         .fs_flags       = FS_HAS_SUBTYPE | FS_USERNS_MOUNT,
1577         .init_fs_context = fuse_init_fs_context,
1578         .parameters     = fuse_fs_parameters,
1579         .kill_sb        = fuse_kill_sb_anon,
1580 };
1581 MODULE_ALIAS_FS("fuse");
1582
1583 #ifdef CONFIG_BLOCK
1584 static void fuse_kill_sb_blk(struct super_block *sb)
1585 {
1586         struct fuse_mount *fm = get_fuse_mount_super(sb);
1587         bool last;
1588
1589         if (fm) {
1590                 last = fuse_mount_remove(fm);
1591                 if (last)
1592                         fuse_conn_destroy(fm);
1593         }
1594         kill_block_super(sb);
1595 }
1596
1597 static struct file_system_type fuseblk_fs_type = {
1598         .owner          = THIS_MODULE,
1599         .name           = "fuseblk",
1600         .init_fs_context = fuse_init_fs_context,
1601         .parameters     = fuse_fs_parameters,
1602         .kill_sb        = fuse_kill_sb_blk,
1603         .fs_flags       = FS_REQUIRES_DEV | FS_HAS_SUBTYPE,
1604 };
1605 MODULE_ALIAS_FS("fuseblk");
1606
1607 static inline int register_fuseblk(void)
1608 {
1609         return register_filesystem(&fuseblk_fs_type);
1610 }
1611
1612 static inline void unregister_fuseblk(void)
1613 {
1614         unregister_filesystem(&fuseblk_fs_type);
1615 }
1616 #else
1617 static inline int register_fuseblk(void)
1618 {
1619         return 0;
1620 }
1621
1622 static inline void unregister_fuseblk(void)
1623 {
1624 }
1625 #endif
1626
1627 static void fuse_inode_init_once(void *foo)
1628 {
1629         struct inode *inode = foo;
1630
1631         inode_init_once(inode);
1632 }
1633
1634 static int __init fuse_fs_init(void)
1635 {
1636         int err;
1637
1638         fuse_inode_cachep = kmem_cache_create("fuse_inode",
1639                         sizeof(struct fuse_inode), 0,
1640                         SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT,
1641                         fuse_inode_init_once);
1642         err = -ENOMEM;
1643         if (!fuse_inode_cachep)
1644                 goto out;
1645
1646         err = register_fuseblk();
1647         if (err)
1648                 goto out2;
1649
1650         err = register_filesystem(&fuse_fs_type);
1651         if (err)
1652                 goto out3;
1653
1654         return 0;
1655
1656  out3:
1657         unregister_fuseblk();
1658  out2:
1659         kmem_cache_destroy(fuse_inode_cachep);
1660  out:
1661         return err;
1662 }
1663
1664 static void fuse_fs_cleanup(void)
1665 {
1666         unregister_filesystem(&fuse_fs_type);
1667         unregister_fuseblk();
1668
1669         /*
1670          * Make sure all delayed rcu free inodes are flushed before we
1671          * destroy cache.
1672          */
1673         rcu_barrier();
1674         kmem_cache_destroy(fuse_inode_cachep);
1675 }
1676
1677 static struct kobject *fuse_kobj;
1678
1679 static int fuse_sysfs_init(void)
1680 {
1681         int err;
1682
1683         fuse_kobj = kobject_create_and_add("fuse", fs_kobj);
1684         if (!fuse_kobj) {
1685                 err = -ENOMEM;
1686                 goto out_err;
1687         }
1688
1689         err = sysfs_create_mount_point(fuse_kobj, "connections");
1690         if (err)
1691                 goto out_fuse_unregister;
1692
1693         return 0;
1694
1695  out_fuse_unregister:
1696         kobject_put(fuse_kobj);
1697  out_err:
1698         return err;
1699 }
1700
1701 static void fuse_sysfs_cleanup(void)
1702 {
1703         sysfs_remove_mount_point(fuse_kobj, "connections");
1704         kobject_put(fuse_kobj);
1705 }
1706
1707 static int __init fuse_init(void)
1708 {
1709         int res;
1710
1711         pr_info("init (API version %i.%i)\n",
1712                 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION);
1713
1714         INIT_LIST_HEAD(&fuse_conn_list);
1715         res = fuse_fs_init();
1716         if (res)
1717                 goto err;
1718
1719         res = fuse_dev_init();
1720         if (res)
1721                 goto err_fs_cleanup;
1722
1723         res = fuse_sysfs_init();
1724         if (res)
1725                 goto err_dev_cleanup;
1726
1727         res = fuse_ctl_init();
1728         if (res)
1729                 goto err_sysfs_cleanup;
1730
1731         sanitize_global_limit(&max_user_bgreq);
1732         sanitize_global_limit(&max_user_congthresh);
1733
1734         return 0;
1735
1736  err_sysfs_cleanup:
1737         fuse_sysfs_cleanup();
1738  err_dev_cleanup:
1739         fuse_dev_cleanup();
1740  err_fs_cleanup:
1741         fuse_fs_cleanup();
1742  err:
1743         return res;
1744 }
1745
1746 static void __exit fuse_exit(void)
1747 {
1748         pr_debug("exit\n");
1749
1750         fuse_ctl_cleanup();
1751         fuse_sysfs_cleanup();
1752         fuse_fs_cleanup();
1753         fuse_dev_cleanup();
1754 }
1755
1756 module_init(fuse_init);
1757 module_exit(fuse_exit);