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) && !fuse_is_bad(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 * We are refreshing inode data and it is possible that another 210 * client set suid/sgid or security.capability xattr. So clear 211 * S_NOSEC. Ideally, we could have cleared it only if suid/sgid 212 * was set or if security.capability xattr was set. But we don't 213 * know if security.capability has been set or not. So clear it 214 * anyway. Its less efficient but should be safe. 215 */ 216 inode->i_flags &= ~S_NOSEC; 217 } 218 219 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr, 220 u64 attr_valid, u64 attr_version) 221 { 222 struct fuse_conn *fc = get_fuse_conn(inode); 223 struct fuse_inode *fi = get_fuse_inode(inode); 224 bool is_wb = fc->writeback_cache; 225 loff_t oldsize; 226 struct timespec64 old_mtime; 227 228 spin_lock(&fi->lock); 229 if ((attr_version != 0 && fi->attr_version > attr_version) || 230 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) { 231 spin_unlock(&fi->lock); 232 return; 233 } 234 235 old_mtime = inode->i_mtime; 236 fuse_change_attributes_common(inode, attr, attr_valid); 237 238 oldsize = inode->i_size; 239 /* 240 * In case of writeback_cache enabled, the cached writes beyond EOF 241 * extend local i_size without keeping userspace server in sync. So, 242 * attr->size coming from server can be stale. We cannot trust it. 243 */ 244 if (!is_wb || !S_ISREG(inode->i_mode)) 245 i_size_write(inode, attr->size); 246 spin_unlock(&fi->lock); 247 248 if (!is_wb && S_ISREG(inode->i_mode)) { 249 bool inval = false; 250 251 if (oldsize != attr->size) { 252 truncate_pagecache(inode, attr->size); 253 if (!fc->explicit_inval_data) 254 inval = true; 255 } else if (fc->auto_inval_data) { 256 struct timespec64 new_mtime = { 257 .tv_sec = attr->mtime, 258 .tv_nsec = attr->mtimensec, 259 }; 260 261 /* 262 * Auto inval mode also checks and invalidates if mtime 263 * has changed. 264 */ 265 if (!timespec64_equal(&old_mtime, &new_mtime)) 266 inval = true; 267 } 268 269 if (inval) 270 invalidate_inode_pages2(inode->i_mapping); 271 } 272 } 273 274 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr) 275 { 276 inode->i_mode = attr->mode & S_IFMT; 277 inode->i_size = attr->size; 278 inode->i_mtime.tv_sec = attr->mtime; 279 inode->i_mtime.tv_nsec = attr->mtimensec; 280 inode->i_ctime.tv_sec = attr->ctime; 281 inode->i_ctime.tv_nsec = attr->ctimensec; 282 if (S_ISREG(inode->i_mode)) { 283 fuse_init_common(inode); 284 fuse_init_file_inode(inode); 285 } else if (S_ISDIR(inode->i_mode)) 286 fuse_init_dir(inode); 287 else if (S_ISLNK(inode->i_mode)) 288 fuse_init_symlink(inode); 289 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || 290 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { 291 fuse_init_common(inode); 292 init_special_inode(inode, inode->i_mode, 293 new_decode_dev(attr->rdev)); 294 } else 295 BUG(); 296 } 297 298 static int fuse_inode_eq(struct inode *inode, void *_nodeidp) 299 { 300 u64 nodeid = *(u64 *) _nodeidp; 301 if (get_node_id(inode) == nodeid) 302 return 1; 303 else 304 return 0; 305 } 306 307 static int fuse_inode_set(struct inode *inode, void *_nodeidp) 308 { 309 u64 nodeid = *(u64 *) _nodeidp; 310 get_fuse_inode(inode)->nodeid = nodeid; 311 return 0; 312 } 313 314 struct inode *fuse_iget(struct super_block *sb, u64 nodeid, 315 int generation, struct fuse_attr *attr, 316 u64 attr_valid, u64 attr_version) 317 { 318 struct inode *inode; 319 struct fuse_inode *fi; 320 struct fuse_conn *fc = get_fuse_conn_super(sb); 321 322 /* 323 * Auto mount points get their node id from the submount root, which is 324 * not a unique identifier within this filesystem. 325 * 326 * To avoid conflicts, do not place submount points into the inode hash 327 * table. 328 */ 329 if (fc->auto_submounts && (attr->flags & FUSE_ATTR_SUBMOUNT) && 330 S_ISDIR(attr->mode)) { 331 inode = new_inode(sb); 332 if (!inode) 333 return NULL; 334 335 fuse_init_inode(inode, attr); 336 get_fuse_inode(inode)->nodeid = nodeid; 337 inode->i_flags |= S_AUTOMOUNT; 338 goto done; 339 } 340 341 retry: 342 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid); 343 if (!inode) 344 return NULL; 345 346 if ((inode->i_state & I_NEW)) { 347 inode->i_flags |= S_NOATIME; 348 if (!fc->writeback_cache || !S_ISREG(attr->mode)) 349 inode->i_flags |= S_NOCMTIME; 350 inode->i_generation = generation; 351 fuse_init_inode(inode, attr); 352 unlock_new_inode(inode); 353 } else if ((inode->i_mode ^ attr->mode) & S_IFMT) { 354 /* Inode has changed type, any I/O on the old should fail */ 355 fuse_make_bad(inode); 356 iput(inode); 357 goto retry; 358 } 359 done: 360 fi = get_fuse_inode(inode); 361 spin_lock(&fi->lock); 362 fi->nlookup++; 363 spin_unlock(&fi->lock); 364 fuse_change_attributes(inode, attr, attr_valid, attr_version); 365 366 return inode; 367 } 368 369 struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid, 370 struct fuse_mount **fm) 371 { 372 struct fuse_mount *fm_iter; 373 struct inode *inode; 374 375 WARN_ON(!rwsem_is_locked(&fc->killsb)); 376 list_for_each_entry(fm_iter, &fc->mounts, fc_entry) { 377 if (!fm_iter->sb) 378 continue; 379 380 inode = ilookup5(fm_iter->sb, nodeid, fuse_inode_eq, &nodeid); 381 if (inode) { 382 if (fm) 383 *fm = fm_iter; 384 return inode; 385 } 386 } 387 388 return NULL; 389 } 390 391 int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid, 392 loff_t offset, loff_t len) 393 { 394 struct fuse_inode *fi; 395 struct inode *inode; 396 pgoff_t pg_start; 397 pgoff_t pg_end; 398 399 inode = fuse_ilookup(fc, nodeid, NULL); 400 if (!inode) 401 return -ENOENT; 402 403 fi = get_fuse_inode(inode); 404 spin_lock(&fi->lock); 405 fi->attr_version = atomic64_inc_return(&fc->attr_version); 406 spin_unlock(&fi->lock); 407 408 fuse_invalidate_attr(inode); 409 forget_all_cached_acls(inode); 410 if (offset >= 0) { 411 pg_start = offset >> PAGE_SHIFT; 412 if (len <= 0) 413 pg_end = -1; 414 else 415 pg_end = (offset + len - 1) >> PAGE_SHIFT; 416 invalidate_inode_pages2_range(inode->i_mapping, 417 pg_start, pg_end); 418 } 419 iput(inode); 420 return 0; 421 } 422 423 bool fuse_lock_inode(struct inode *inode) 424 { 425 bool locked = false; 426 427 if (!get_fuse_conn(inode)->parallel_dirops) { 428 mutex_lock(&get_fuse_inode(inode)->mutex); 429 locked = true; 430 } 431 432 return locked; 433 } 434 435 void fuse_unlock_inode(struct inode *inode, bool locked) 436 { 437 if (locked) 438 mutex_unlock(&get_fuse_inode(inode)->mutex); 439 } 440 441 static void fuse_umount_begin(struct super_block *sb) 442 { 443 struct fuse_conn *fc = get_fuse_conn_super(sb); 444 445 if (!fc->no_force_umount) 446 fuse_abort_conn(fc); 447 } 448 449 static void fuse_send_destroy(struct fuse_mount *fm) 450 { 451 if (fm->fc->conn_init) { 452 FUSE_ARGS(args); 453 454 args.opcode = FUSE_DESTROY; 455 args.force = true; 456 args.nocreds = true; 457 fuse_simple_request(fm, &args); 458 } 459 } 460 461 static void fuse_put_super(struct super_block *sb) 462 { 463 struct fuse_mount *fm = get_fuse_mount_super(sb); 464 465 fuse_conn_put(fm->fc); 466 kfree(fm); 467 } 468 469 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr) 470 { 471 stbuf->f_type = FUSE_SUPER_MAGIC; 472 stbuf->f_bsize = attr->bsize; 473 stbuf->f_frsize = attr->frsize; 474 stbuf->f_blocks = attr->blocks; 475 stbuf->f_bfree = attr->bfree; 476 stbuf->f_bavail = attr->bavail; 477 stbuf->f_files = attr->files; 478 stbuf->f_ffree = attr->ffree; 479 stbuf->f_namelen = attr->namelen; 480 /* fsid is left zero */ 481 } 482 483 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf) 484 { 485 struct super_block *sb = dentry->d_sb; 486 struct fuse_mount *fm = get_fuse_mount_super(sb); 487 FUSE_ARGS(args); 488 struct fuse_statfs_out outarg; 489 int err; 490 491 if (!fuse_allow_current_process(fm->fc)) { 492 buf->f_type = FUSE_SUPER_MAGIC; 493 return 0; 494 } 495 496 memset(&outarg, 0, sizeof(outarg)); 497 args.in_numargs = 0; 498 args.opcode = FUSE_STATFS; 499 args.nodeid = get_node_id(d_inode(dentry)); 500 args.out_numargs = 1; 501 args.out_args[0].size = sizeof(outarg); 502 args.out_args[0].value = &outarg; 503 err = fuse_simple_request(fm, &args); 504 if (!err) 505 convert_fuse_statfs(buf, &outarg.st); 506 return err; 507 } 508 509 enum { 510 OPT_SOURCE, 511 OPT_SUBTYPE, 512 OPT_FD, 513 OPT_ROOTMODE, 514 OPT_USER_ID, 515 OPT_GROUP_ID, 516 OPT_DEFAULT_PERMISSIONS, 517 OPT_ALLOW_OTHER, 518 OPT_MAX_READ, 519 OPT_BLKSIZE, 520 OPT_ERR 521 }; 522 523 static const struct fs_parameter_spec fuse_fs_parameters[] = { 524 fsparam_string ("source", OPT_SOURCE), 525 fsparam_u32 ("fd", OPT_FD), 526 fsparam_u32oct ("rootmode", OPT_ROOTMODE), 527 fsparam_u32 ("user_id", OPT_USER_ID), 528 fsparam_u32 ("group_id", OPT_GROUP_ID), 529 fsparam_flag ("default_permissions", OPT_DEFAULT_PERMISSIONS), 530 fsparam_flag ("allow_other", OPT_ALLOW_OTHER), 531 fsparam_u32 ("max_read", OPT_MAX_READ), 532 fsparam_u32 ("blksize", OPT_BLKSIZE), 533 fsparam_string ("subtype", OPT_SUBTYPE), 534 {} 535 }; 536 537 static int fuse_parse_param(struct fs_context *fc, struct fs_parameter *param) 538 { 539 struct fs_parse_result result; 540 struct fuse_fs_context *ctx = fc->fs_private; 541 int opt; 542 543 if (fc->purpose == FS_CONTEXT_FOR_RECONFIGURE) { 544 /* 545 * Ignore options coming from mount(MS_REMOUNT) for backward 546 * compatibility. 547 */ 548 if (fc->oldapi) 549 return 0; 550 551 return invalfc(fc, "No changes allowed in reconfigure"); 552 } 553 554 opt = fs_parse(fc, fuse_fs_parameters, param, &result); 555 if (opt < 0) 556 return opt; 557 558 switch (opt) { 559 case OPT_SOURCE: 560 if (fc->source) 561 return invalfc(fc, "Multiple sources specified"); 562 fc->source = param->string; 563 param->string = NULL; 564 break; 565 566 case OPT_SUBTYPE: 567 if (ctx->subtype) 568 return invalfc(fc, "Multiple subtypes specified"); 569 ctx->subtype = param->string; 570 param->string = NULL; 571 return 0; 572 573 case OPT_FD: 574 ctx->fd = result.uint_32; 575 ctx->fd_present = true; 576 break; 577 578 case OPT_ROOTMODE: 579 if (!fuse_valid_type(result.uint_32)) 580 return invalfc(fc, "Invalid rootmode"); 581 ctx->rootmode = result.uint_32; 582 ctx->rootmode_present = true; 583 break; 584 585 case OPT_USER_ID: 586 ctx->user_id = make_kuid(fc->user_ns, result.uint_32); 587 if (!uid_valid(ctx->user_id)) 588 return invalfc(fc, "Invalid user_id"); 589 ctx->user_id_present = true; 590 break; 591 592 case OPT_GROUP_ID: 593 ctx->group_id = make_kgid(fc->user_ns, result.uint_32); 594 if (!gid_valid(ctx->group_id)) 595 return invalfc(fc, "Invalid group_id"); 596 ctx->group_id_present = true; 597 break; 598 599 case OPT_DEFAULT_PERMISSIONS: 600 ctx->default_permissions = true; 601 break; 602 603 case OPT_ALLOW_OTHER: 604 ctx->allow_other = true; 605 break; 606 607 case OPT_MAX_READ: 608 ctx->max_read = result.uint_32; 609 break; 610 611 case OPT_BLKSIZE: 612 if (!ctx->is_bdev) 613 return invalfc(fc, "blksize only supported for fuseblk"); 614 ctx->blksize = result.uint_32; 615 break; 616 617 default: 618 return -EINVAL; 619 } 620 621 return 0; 622 } 623 624 static void fuse_free_fc(struct fs_context *fc) 625 { 626 struct fuse_fs_context *ctx = fc->fs_private; 627 628 if (ctx) { 629 kfree(ctx->subtype); 630 kfree(ctx); 631 } 632 } 633 634 static int fuse_show_options(struct seq_file *m, struct dentry *root) 635 { 636 struct super_block *sb = root->d_sb; 637 struct fuse_conn *fc = get_fuse_conn_super(sb); 638 639 if (fc->legacy_opts_show) { 640 seq_printf(m, ",user_id=%u", 641 from_kuid_munged(fc->user_ns, fc->user_id)); 642 seq_printf(m, ",group_id=%u", 643 from_kgid_munged(fc->user_ns, fc->group_id)); 644 if (fc->default_permissions) 645 seq_puts(m, ",default_permissions"); 646 if (fc->allow_other) 647 seq_puts(m, ",allow_other"); 648 if (fc->max_read != ~0) 649 seq_printf(m, ",max_read=%u", fc->max_read); 650 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE) 651 seq_printf(m, ",blksize=%lu", sb->s_blocksize); 652 } 653 #ifdef CONFIG_FUSE_DAX 654 if (fc->dax) 655 seq_puts(m, ",dax"); 656 #endif 657 658 return 0; 659 } 660 661 static void fuse_iqueue_init(struct fuse_iqueue *fiq, 662 const struct fuse_iqueue_ops *ops, 663 void *priv) 664 { 665 memset(fiq, 0, sizeof(struct fuse_iqueue)); 666 spin_lock_init(&fiq->lock); 667 init_waitqueue_head(&fiq->waitq); 668 INIT_LIST_HEAD(&fiq->pending); 669 INIT_LIST_HEAD(&fiq->interrupts); 670 fiq->forget_list_tail = &fiq->forget_list_head; 671 fiq->connected = 1; 672 fiq->ops = ops; 673 fiq->priv = priv; 674 } 675 676 static void fuse_pqueue_init(struct fuse_pqueue *fpq) 677 { 678 unsigned int i; 679 680 spin_lock_init(&fpq->lock); 681 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++) 682 INIT_LIST_HEAD(&fpq->processing[i]); 683 INIT_LIST_HEAD(&fpq->io); 684 fpq->connected = 1; 685 } 686 687 void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm, 688 struct user_namespace *user_ns, 689 const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv) 690 { 691 memset(fc, 0, sizeof(*fc)); 692 spin_lock_init(&fc->lock); 693 spin_lock_init(&fc->bg_lock); 694 init_rwsem(&fc->killsb); 695 refcount_set(&fc->count, 1); 696 atomic_set(&fc->dev_count, 1); 697 init_waitqueue_head(&fc->blocked_waitq); 698 fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv); 699 INIT_LIST_HEAD(&fc->bg_queue); 700 INIT_LIST_HEAD(&fc->entry); 701 INIT_LIST_HEAD(&fc->devices); 702 atomic_set(&fc->num_waiting, 0); 703 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND; 704 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD; 705 atomic64_set(&fc->khctr, 0); 706 fc->polled_files = RB_ROOT; 707 fc->blocked = 0; 708 fc->initialized = 0; 709 fc->connected = 1; 710 atomic64_set(&fc->attr_version, 1); 711 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key)); 712 fc->pid_ns = get_pid_ns(task_active_pid_ns(current)); 713 fc->user_ns = get_user_ns(user_ns); 714 fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ; 715 716 INIT_LIST_HEAD(&fc->mounts); 717 list_add(&fm->fc_entry, &fc->mounts); 718 fm->fc = fc; 719 } 720 EXPORT_SYMBOL_GPL(fuse_conn_init); 721 722 void fuse_conn_put(struct fuse_conn *fc) 723 { 724 if (refcount_dec_and_test(&fc->count)) { 725 struct fuse_iqueue *fiq = &fc->iq; 726 727 if (IS_ENABLED(CONFIG_FUSE_DAX)) 728 fuse_dax_conn_free(fc); 729 if (fiq->ops->release) 730 fiq->ops->release(fiq); 731 put_pid_ns(fc->pid_ns); 732 put_user_ns(fc->user_ns); 733 fc->release(fc); 734 } 735 } 736 EXPORT_SYMBOL_GPL(fuse_conn_put); 737 738 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc) 739 { 740 refcount_inc(&fc->count); 741 return fc; 742 } 743 EXPORT_SYMBOL_GPL(fuse_conn_get); 744 745 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode) 746 { 747 struct fuse_attr attr; 748 memset(&attr, 0, sizeof(attr)); 749 750 attr.mode = mode; 751 attr.ino = FUSE_ROOT_ID; 752 attr.nlink = 1; 753 return fuse_iget(sb, 1, 0, &attr, 0, 0); 754 } 755 756 struct fuse_inode_handle { 757 u64 nodeid; 758 u32 generation; 759 }; 760 761 static struct dentry *fuse_get_dentry(struct super_block *sb, 762 struct fuse_inode_handle *handle) 763 { 764 struct fuse_conn *fc = get_fuse_conn_super(sb); 765 struct inode *inode; 766 struct dentry *entry; 767 int err = -ESTALE; 768 769 if (handle->nodeid == 0) 770 goto out_err; 771 772 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid); 773 if (!inode) { 774 struct fuse_entry_out outarg; 775 const struct qstr name = QSTR_INIT(".", 1); 776 777 if (!fc->export_support) 778 goto out_err; 779 780 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg, 781 &inode); 782 if (err && err != -ENOENT) 783 goto out_err; 784 if (err || !inode) { 785 err = -ESTALE; 786 goto out_err; 787 } 788 err = -EIO; 789 if (get_node_id(inode) != handle->nodeid) 790 goto out_iput; 791 } 792 err = -ESTALE; 793 if (inode->i_generation != handle->generation) 794 goto out_iput; 795 796 entry = d_obtain_alias(inode); 797 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID) 798 fuse_invalidate_entry_cache(entry); 799 800 return entry; 801 802 out_iput: 803 iput(inode); 804 out_err: 805 return ERR_PTR(err); 806 } 807 808 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len, 809 struct inode *parent) 810 { 811 int len = parent ? 6 : 3; 812 u64 nodeid; 813 u32 generation; 814 815 if (*max_len < len) { 816 *max_len = len; 817 return FILEID_INVALID; 818 } 819 820 nodeid = get_fuse_inode(inode)->nodeid; 821 generation = inode->i_generation; 822 823 fh[0] = (u32)(nodeid >> 32); 824 fh[1] = (u32)(nodeid & 0xffffffff); 825 fh[2] = generation; 826 827 if (parent) { 828 nodeid = get_fuse_inode(parent)->nodeid; 829 generation = parent->i_generation; 830 831 fh[3] = (u32)(nodeid >> 32); 832 fh[4] = (u32)(nodeid & 0xffffffff); 833 fh[5] = generation; 834 } 835 836 *max_len = len; 837 return parent ? 0x82 : 0x81; 838 } 839 840 static struct dentry *fuse_fh_to_dentry(struct super_block *sb, 841 struct fid *fid, int fh_len, int fh_type) 842 { 843 struct fuse_inode_handle handle; 844 845 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3) 846 return NULL; 847 848 handle.nodeid = (u64) fid->raw[0] << 32; 849 handle.nodeid |= (u64) fid->raw[1]; 850 handle.generation = fid->raw[2]; 851 return fuse_get_dentry(sb, &handle); 852 } 853 854 static struct dentry *fuse_fh_to_parent(struct super_block *sb, 855 struct fid *fid, int fh_len, int fh_type) 856 { 857 struct fuse_inode_handle parent; 858 859 if (fh_type != 0x82 || fh_len < 6) 860 return NULL; 861 862 parent.nodeid = (u64) fid->raw[3] << 32; 863 parent.nodeid |= (u64) fid->raw[4]; 864 parent.generation = fid->raw[5]; 865 return fuse_get_dentry(sb, &parent); 866 } 867 868 static struct dentry *fuse_get_parent(struct dentry *child) 869 { 870 struct inode *child_inode = d_inode(child); 871 struct fuse_conn *fc = get_fuse_conn(child_inode); 872 struct inode *inode; 873 struct dentry *parent; 874 struct fuse_entry_out outarg; 875 const struct qstr name = QSTR_INIT("..", 2); 876 int err; 877 878 if (!fc->export_support) 879 return ERR_PTR(-ESTALE); 880 881 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode), 882 &name, &outarg, &inode); 883 if (err) { 884 if (err == -ENOENT) 885 return ERR_PTR(-ESTALE); 886 return ERR_PTR(err); 887 } 888 889 parent = d_obtain_alias(inode); 890 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID) 891 fuse_invalidate_entry_cache(parent); 892 893 return parent; 894 } 895 896 static const struct export_operations fuse_export_operations = { 897 .fh_to_dentry = fuse_fh_to_dentry, 898 .fh_to_parent = fuse_fh_to_parent, 899 .encode_fh = fuse_encode_fh, 900 .get_parent = fuse_get_parent, 901 }; 902 903 static const struct super_operations fuse_super_operations = { 904 .alloc_inode = fuse_alloc_inode, 905 .free_inode = fuse_free_inode, 906 .evict_inode = fuse_evict_inode, 907 .write_inode = fuse_write_inode, 908 .drop_inode = generic_delete_inode, 909 .put_super = fuse_put_super, 910 .umount_begin = fuse_umount_begin, 911 .statfs = fuse_statfs, 912 .show_options = fuse_show_options, 913 }; 914 915 static void sanitize_global_limit(unsigned *limit) 916 { 917 /* 918 * The default maximum number of async requests is calculated to consume 919 * 1/2^13 of the total memory, assuming 392 bytes per request. 920 */ 921 if (*limit == 0) 922 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392; 923 924 if (*limit >= 1 << 16) 925 *limit = (1 << 16) - 1; 926 } 927 928 static int set_global_limit(const char *val, const struct kernel_param *kp) 929 { 930 int rv; 931 932 rv = param_set_uint(val, kp); 933 if (rv) 934 return rv; 935 936 sanitize_global_limit((unsigned *)kp->arg); 937 938 return 0; 939 } 940 941 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg) 942 { 943 int cap_sys_admin = capable(CAP_SYS_ADMIN); 944 945 if (arg->minor < 13) 946 return; 947 948 sanitize_global_limit(&max_user_bgreq); 949 sanitize_global_limit(&max_user_congthresh); 950 951 spin_lock(&fc->bg_lock); 952 if (arg->max_background) { 953 fc->max_background = arg->max_background; 954 955 if (!cap_sys_admin && fc->max_background > max_user_bgreq) 956 fc->max_background = max_user_bgreq; 957 } 958 if (arg->congestion_threshold) { 959 fc->congestion_threshold = arg->congestion_threshold; 960 961 if (!cap_sys_admin && 962 fc->congestion_threshold > max_user_congthresh) 963 fc->congestion_threshold = max_user_congthresh; 964 } 965 spin_unlock(&fc->bg_lock); 966 } 967 968 struct fuse_init_args { 969 struct fuse_args args; 970 struct fuse_init_in in; 971 struct fuse_init_out out; 972 }; 973 974 static void process_init_reply(struct fuse_mount *fm, struct fuse_args *args, 975 int error) 976 { 977 struct fuse_conn *fc = fm->fc; 978 struct fuse_init_args *ia = container_of(args, typeof(*ia), args); 979 struct fuse_init_out *arg = &ia->out; 980 bool ok = true; 981 982 if (error || arg->major != FUSE_KERNEL_VERSION) 983 ok = false; 984 else { 985 unsigned long ra_pages; 986 987 process_init_limits(fc, arg); 988 989 if (arg->minor >= 6) { 990 ra_pages = arg->max_readahead / PAGE_SIZE; 991 if (arg->flags & FUSE_ASYNC_READ) 992 fc->async_read = 1; 993 if (!(arg->flags & FUSE_POSIX_LOCKS)) 994 fc->no_lock = 1; 995 if (arg->minor >= 17) { 996 if (!(arg->flags & FUSE_FLOCK_LOCKS)) 997 fc->no_flock = 1; 998 } else { 999 if (!(arg->flags & FUSE_POSIX_LOCKS)) 1000 fc->no_flock = 1; 1001 } 1002 if (arg->flags & FUSE_ATOMIC_O_TRUNC) 1003 fc->atomic_o_trunc = 1; 1004 if (arg->minor >= 9) { 1005 /* LOOKUP has dependency on proto version */ 1006 if (arg->flags & FUSE_EXPORT_SUPPORT) 1007 fc->export_support = 1; 1008 } 1009 if (arg->flags & FUSE_BIG_WRITES) 1010 fc->big_writes = 1; 1011 if (arg->flags & FUSE_DONT_MASK) 1012 fc->dont_mask = 1; 1013 if (arg->flags & FUSE_AUTO_INVAL_DATA) 1014 fc->auto_inval_data = 1; 1015 else if (arg->flags & FUSE_EXPLICIT_INVAL_DATA) 1016 fc->explicit_inval_data = 1; 1017 if (arg->flags & FUSE_DO_READDIRPLUS) { 1018 fc->do_readdirplus = 1; 1019 if (arg->flags & FUSE_READDIRPLUS_AUTO) 1020 fc->readdirplus_auto = 1; 1021 } 1022 if (arg->flags & FUSE_ASYNC_DIO) 1023 fc->async_dio = 1; 1024 if (arg->flags & FUSE_WRITEBACK_CACHE) 1025 fc->writeback_cache = 1; 1026 if (arg->flags & FUSE_PARALLEL_DIROPS) 1027 fc->parallel_dirops = 1; 1028 if (arg->flags & FUSE_HANDLE_KILLPRIV) 1029 fc->handle_killpriv = 1; 1030 if (arg->time_gran && arg->time_gran <= 1000000000) 1031 fm->sb->s_time_gran = arg->time_gran; 1032 if ((arg->flags & FUSE_POSIX_ACL)) { 1033 fc->default_permissions = 1; 1034 fc->posix_acl = 1; 1035 fm->sb->s_xattr = fuse_acl_xattr_handlers; 1036 } 1037 if (arg->flags & FUSE_CACHE_SYMLINKS) 1038 fc->cache_symlinks = 1; 1039 if (arg->flags & FUSE_ABORT_ERROR) 1040 fc->abort_err = 1; 1041 if (arg->flags & FUSE_MAX_PAGES) { 1042 fc->max_pages = 1043 min_t(unsigned int, FUSE_MAX_MAX_PAGES, 1044 max_t(unsigned int, arg->max_pages, 1)); 1045 } 1046 if (IS_ENABLED(CONFIG_FUSE_DAX) && 1047 arg->flags & FUSE_MAP_ALIGNMENT && 1048 !fuse_dax_check_alignment(fc, arg->map_alignment)) { 1049 ok = false; 1050 } 1051 if (arg->flags & FUSE_HANDLE_KILLPRIV_V2) { 1052 fc->handle_killpriv_v2 = 1; 1053 fm->sb->s_flags |= SB_NOSEC; 1054 } 1055 } else { 1056 ra_pages = fc->max_read / PAGE_SIZE; 1057 fc->no_lock = 1; 1058 fc->no_flock = 1; 1059 } 1060 1061 fm->sb->s_bdi->ra_pages = 1062 min(fm->sb->s_bdi->ra_pages, ra_pages); 1063 fc->minor = arg->minor; 1064 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write; 1065 fc->max_write = max_t(unsigned, 4096, fc->max_write); 1066 fc->conn_init = 1; 1067 } 1068 kfree(ia); 1069 1070 if (!ok) { 1071 fc->conn_init = 0; 1072 fc->conn_error = 1; 1073 } 1074 1075 fuse_set_initialized(fc); 1076 wake_up_all(&fc->blocked_waitq); 1077 } 1078 1079 void fuse_send_init(struct fuse_mount *fm) 1080 { 1081 struct fuse_init_args *ia; 1082 1083 ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL); 1084 1085 ia->in.major = FUSE_KERNEL_VERSION; 1086 ia->in.minor = FUSE_KERNEL_MINOR_VERSION; 1087 ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE; 1088 ia->in.flags |= 1089 FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC | 1090 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK | 1091 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ | 1092 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA | 1093 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO | 1094 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT | 1095 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL | 1096 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS | 1097 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA | 1098 FUSE_HANDLE_KILLPRIV_V2; 1099 #ifdef CONFIG_FUSE_DAX 1100 if (fm->fc->dax) 1101 ia->in.flags |= FUSE_MAP_ALIGNMENT; 1102 #endif 1103 if (fm->fc->auto_submounts) 1104 ia->in.flags |= FUSE_SUBMOUNTS; 1105 1106 ia->args.opcode = FUSE_INIT; 1107 ia->args.in_numargs = 1; 1108 ia->args.in_args[0].size = sizeof(ia->in); 1109 ia->args.in_args[0].value = &ia->in; 1110 ia->args.out_numargs = 1; 1111 /* Variable length argument used for backward compatibility 1112 with interface version < 7.5. Rest of init_out is zeroed 1113 by do_get_request(), so a short reply is not a problem */ 1114 ia->args.out_argvar = true; 1115 ia->args.out_args[0].size = sizeof(ia->out); 1116 ia->args.out_args[0].value = &ia->out; 1117 ia->args.force = true; 1118 ia->args.nocreds = true; 1119 ia->args.end = process_init_reply; 1120 1121 if (fuse_simple_background(fm, &ia->args, GFP_KERNEL) != 0) 1122 process_init_reply(fm, &ia->args, -ENOTCONN); 1123 } 1124 EXPORT_SYMBOL_GPL(fuse_send_init); 1125 1126 void fuse_free_conn(struct fuse_conn *fc) 1127 { 1128 WARN_ON(!list_empty(&fc->devices)); 1129 kfree_rcu(fc, rcu); 1130 } 1131 EXPORT_SYMBOL_GPL(fuse_free_conn); 1132 1133 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb) 1134 { 1135 int err; 1136 char *suffix = ""; 1137 1138 if (sb->s_bdev) { 1139 suffix = "-fuseblk"; 1140 /* 1141 * sb->s_bdi points to blkdev's bdi however we want to redirect 1142 * it to our private bdi... 1143 */ 1144 bdi_put(sb->s_bdi); 1145 sb->s_bdi = &noop_backing_dev_info; 1146 } 1147 err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev), 1148 MINOR(fc->dev), suffix); 1149 if (err) 1150 return err; 1151 1152 /* fuse does it's own writeback accounting */ 1153 sb->s_bdi->capabilities &= ~BDI_CAP_WRITEBACK_ACCT; 1154 sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT; 1155 1156 /* 1157 * For a single fuse filesystem use max 1% of dirty + 1158 * writeback threshold. 1159 * 1160 * This gives about 1M of write buffer for memory maps on a 1161 * machine with 1G and 10% dirty_ratio, which should be more 1162 * than enough. 1163 * 1164 * Privileged users can raise it by writing to 1165 * 1166 * /sys/class/bdi/<bdi>/max_ratio 1167 */ 1168 bdi_set_max_ratio(sb->s_bdi, 1); 1169 1170 return 0; 1171 } 1172 1173 struct fuse_dev *fuse_dev_alloc(void) 1174 { 1175 struct fuse_dev *fud; 1176 struct list_head *pq; 1177 1178 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL); 1179 if (!fud) 1180 return NULL; 1181 1182 pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL); 1183 if (!pq) { 1184 kfree(fud); 1185 return NULL; 1186 } 1187 1188 fud->pq.processing = pq; 1189 fuse_pqueue_init(&fud->pq); 1190 1191 return fud; 1192 } 1193 EXPORT_SYMBOL_GPL(fuse_dev_alloc); 1194 1195 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc) 1196 { 1197 fud->fc = fuse_conn_get(fc); 1198 spin_lock(&fc->lock); 1199 list_add_tail(&fud->entry, &fc->devices); 1200 spin_unlock(&fc->lock); 1201 } 1202 EXPORT_SYMBOL_GPL(fuse_dev_install); 1203 1204 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc) 1205 { 1206 struct fuse_dev *fud; 1207 1208 fud = fuse_dev_alloc(); 1209 if (!fud) 1210 return NULL; 1211 1212 fuse_dev_install(fud, fc); 1213 return fud; 1214 } 1215 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install); 1216 1217 void fuse_dev_free(struct fuse_dev *fud) 1218 { 1219 struct fuse_conn *fc = fud->fc; 1220 1221 if (fc) { 1222 spin_lock(&fc->lock); 1223 list_del(&fud->entry); 1224 spin_unlock(&fc->lock); 1225 1226 fuse_conn_put(fc); 1227 } 1228 kfree(fud->pq.processing); 1229 kfree(fud); 1230 } 1231 EXPORT_SYMBOL_GPL(fuse_dev_free); 1232 1233 static void fuse_fill_attr_from_inode(struct fuse_attr *attr, 1234 const struct fuse_inode *fi) 1235 { 1236 *attr = (struct fuse_attr){ 1237 .ino = fi->inode.i_ino, 1238 .size = fi->inode.i_size, 1239 .blocks = fi->inode.i_blocks, 1240 .atime = fi->inode.i_atime.tv_sec, 1241 .mtime = fi->inode.i_mtime.tv_sec, 1242 .ctime = fi->inode.i_ctime.tv_sec, 1243 .atimensec = fi->inode.i_atime.tv_nsec, 1244 .mtimensec = fi->inode.i_mtime.tv_nsec, 1245 .ctimensec = fi->inode.i_ctime.tv_nsec, 1246 .mode = fi->inode.i_mode, 1247 .nlink = fi->inode.i_nlink, 1248 .uid = fi->inode.i_uid.val, 1249 .gid = fi->inode.i_gid.val, 1250 .rdev = fi->inode.i_rdev, 1251 .blksize = 1u << fi->inode.i_blkbits, 1252 }; 1253 } 1254 1255 static void fuse_sb_defaults(struct super_block *sb) 1256 { 1257 sb->s_magic = FUSE_SUPER_MAGIC; 1258 sb->s_op = &fuse_super_operations; 1259 sb->s_xattr = fuse_xattr_handlers; 1260 sb->s_maxbytes = MAX_LFS_FILESIZE; 1261 sb->s_time_gran = 1; 1262 sb->s_export_op = &fuse_export_operations; 1263 sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE; 1264 if (sb->s_user_ns != &init_user_ns) 1265 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER; 1266 sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION); 1267 1268 /* 1269 * If we are not in the initial user namespace posix 1270 * acls must be translated. 1271 */ 1272 if (sb->s_user_ns != &init_user_ns) 1273 sb->s_xattr = fuse_no_acl_xattr_handlers; 1274 } 1275 1276 int fuse_fill_super_submount(struct super_block *sb, 1277 struct fuse_inode *parent_fi) 1278 { 1279 struct fuse_mount *fm = get_fuse_mount_super(sb); 1280 struct super_block *parent_sb = parent_fi->inode.i_sb; 1281 struct fuse_attr root_attr; 1282 struct inode *root; 1283 1284 fuse_sb_defaults(sb); 1285 fm->sb = sb; 1286 1287 WARN_ON(sb->s_bdi != &noop_backing_dev_info); 1288 sb->s_bdi = bdi_get(parent_sb->s_bdi); 1289 1290 sb->s_xattr = parent_sb->s_xattr; 1291 sb->s_time_gran = parent_sb->s_time_gran; 1292 sb->s_blocksize = parent_sb->s_blocksize; 1293 sb->s_blocksize_bits = parent_sb->s_blocksize_bits; 1294 sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL); 1295 if (parent_sb->s_subtype && !sb->s_subtype) 1296 return -ENOMEM; 1297 1298 fuse_fill_attr_from_inode(&root_attr, parent_fi); 1299 root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0); 1300 /* 1301 * This inode is just a duplicate, so it is not looked up and 1302 * its nlookup should not be incremented. fuse_iget() does 1303 * that, though, so undo it here. 1304 */ 1305 get_fuse_inode(root)->nlookup--; 1306 sb->s_d_op = &fuse_dentry_operations; 1307 sb->s_root = d_make_root(root); 1308 if (!sb->s_root) 1309 return -ENOMEM; 1310 1311 return 0; 1312 } 1313 1314 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx) 1315 { 1316 struct fuse_dev *fud = NULL; 1317 struct fuse_mount *fm = get_fuse_mount_super(sb); 1318 struct fuse_conn *fc = fm->fc; 1319 struct inode *root; 1320 struct dentry *root_dentry; 1321 int err; 1322 1323 err = -EINVAL; 1324 if (sb->s_flags & SB_MANDLOCK) 1325 goto err; 1326 1327 fuse_sb_defaults(sb); 1328 1329 if (ctx->is_bdev) { 1330 #ifdef CONFIG_BLOCK 1331 err = -EINVAL; 1332 if (!sb_set_blocksize(sb, ctx->blksize)) 1333 goto err; 1334 #endif 1335 } else { 1336 sb->s_blocksize = PAGE_SIZE; 1337 sb->s_blocksize_bits = PAGE_SHIFT; 1338 } 1339 1340 sb->s_subtype = ctx->subtype; 1341 ctx->subtype = NULL; 1342 if (IS_ENABLED(CONFIG_FUSE_DAX)) { 1343 err = fuse_dax_conn_alloc(fc, ctx->dax_dev); 1344 if (err) 1345 goto err; 1346 } 1347 1348 if (ctx->fudptr) { 1349 err = -ENOMEM; 1350 fud = fuse_dev_alloc_install(fc); 1351 if (!fud) 1352 goto err_free_dax; 1353 } 1354 1355 fc->dev = sb->s_dev; 1356 fm->sb = sb; 1357 err = fuse_bdi_init(fc, sb); 1358 if (err) 1359 goto err_dev_free; 1360 1361 /* Handle umasking inside the fuse code */ 1362 if (sb->s_flags & SB_POSIXACL) 1363 fc->dont_mask = 1; 1364 sb->s_flags |= SB_POSIXACL; 1365 1366 fc->default_permissions = ctx->default_permissions; 1367 fc->allow_other = ctx->allow_other; 1368 fc->user_id = ctx->user_id; 1369 fc->group_id = ctx->group_id; 1370 fc->legacy_opts_show = ctx->legacy_opts_show; 1371 fc->max_read = max_t(unsigned int, 4096, ctx->max_read); 1372 fc->destroy = ctx->destroy; 1373 fc->no_control = ctx->no_control; 1374 fc->no_force_umount = ctx->no_force_umount; 1375 1376 err = -ENOMEM; 1377 root = fuse_get_root_inode(sb, ctx->rootmode); 1378 sb->s_d_op = &fuse_root_dentry_operations; 1379 root_dentry = d_make_root(root); 1380 if (!root_dentry) 1381 goto err_dev_free; 1382 /* Root dentry doesn't have .d_revalidate */ 1383 sb->s_d_op = &fuse_dentry_operations; 1384 1385 mutex_lock(&fuse_mutex); 1386 err = -EINVAL; 1387 if (ctx->fudptr && *ctx->fudptr) 1388 goto err_unlock; 1389 1390 err = fuse_ctl_add_conn(fc); 1391 if (err) 1392 goto err_unlock; 1393 1394 list_add_tail(&fc->entry, &fuse_conn_list); 1395 sb->s_root = root_dentry; 1396 if (ctx->fudptr) 1397 *ctx->fudptr = fud; 1398 mutex_unlock(&fuse_mutex); 1399 return 0; 1400 1401 err_unlock: 1402 mutex_unlock(&fuse_mutex); 1403 dput(root_dentry); 1404 err_dev_free: 1405 if (fud) 1406 fuse_dev_free(fud); 1407 err_free_dax: 1408 if (IS_ENABLED(CONFIG_FUSE_DAX)) 1409 fuse_dax_conn_free(fc); 1410 err: 1411 return err; 1412 } 1413 EXPORT_SYMBOL_GPL(fuse_fill_super_common); 1414 1415 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc) 1416 { 1417 struct fuse_fs_context *ctx = fsc->fs_private; 1418 struct file *file; 1419 int err; 1420 struct fuse_conn *fc; 1421 struct fuse_mount *fm; 1422 1423 err = -EINVAL; 1424 file = fget(ctx->fd); 1425 if (!file) 1426 goto err; 1427 1428 /* 1429 * Require mount to happen from the same user namespace which 1430 * opened /dev/fuse to prevent potential attacks. 1431 */ 1432 if ((file->f_op != &fuse_dev_operations) || 1433 (file->f_cred->user_ns != sb->s_user_ns)) 1434 goto err_fput; 1435 ctx->fudptr = &file->private_data; 1436 1437 fc = kmalloc(sizeof(*fc), GFP_KERNEL); 1438 err = -ENOMEM; 1439 if (!fc) 1440 goto err_fput; 1441 1442 fm = kzalloc(sizeof(*fm), GFP_KERNEL); 1443 if (!fm) { 1444 kfree(fc); 1445 goto err_fput; 1446 } 1447 1448 fuse_conn_init(fc, fm, sb->s_user_ns, &fuse_dev_fiq_ops, NULL); 1449 fc->release = fuse_free_conn; 1450 1451 sb->s_fs_info = fm; 1452 1453 err = fuse_fill_super_common(sb, ctx); 1454 if (err) 1455 goto err_put_conn; 1456 /* 1457 * atomic_dec_and_test() in fput() provides the necessary 1458 * memory barrier for file->private_data to be visible on all 1459 * CPUs after this 1460 */ 1461 fput(file); 1462 fuse_send_init(get_fuse_mount_super(sb)); 1463 return 0; 1464 1465 err_put_conn: 1466 fuse_conn_put(fc); 1467 kfree(fm); 1468 sb->s_fs_info = NULL; 1469 err_fput: 1470 fput(file); 1471 err: 1472 return err; 1473 } 1474 1475 static int fuse_get_tree(struct fs_context *fc) 1476 { 1477 struct fuse_fs_context *ctx = fc->fs_private; 1478 1479 if (!ctx->fd_present || !ctx->rootmode_present || 1480 !ctx->user_id_present || !ctx->group_id_present) 1481 return -EINVAL; 1482 1483 #ifdef CONFIG_BLOCK 1484 if (ctx->is_bdev) 1485 return get_tree_bdev(fc, fuse_fill_super); 1486 #endif 1487 1488 return get_tree_nodev(fc, fuse_fill_super); 1489 } 1490 1491 static const struct fs_context_operations fuse_context_ops = { 1492 .free = fuse_free_fc, 1493 .parse_param = fuse_parse_param, 1494 .reconfigure = fuse_reconfigure, 1495 .get_tree = fuse_get_tree, 1496 }; 1497 1498 /* 1499 * Set up the filesystem mount context. 1500 */ 1501 static int fuse_init_fs_context(struct fs_context *fc) 1502 { 1503 struct fuse_fs_context *ctx; 1504 1505 ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL); 1506 if (!ctx) 1507 return -ENOMEM; 1508 1509 ctx->max_read = ~0; 1510 ctx->blksize = FUSE_DEFAULT_BLKSIZE; 1511 ctx->legacy_opts_show = true; 1512 1513 #ifdef CONFIG_BLOCK 1514 if (fc->fs_type == &fuseblk_fs_type) { 1515 ctx->is_bdev = true; 1516 ctx->destroy = true; 1517 } 1518 #endif 1519 1520 fc->fs_private = ctx; 1521 fc->ops = &fuse_context_ops; 1522 return 0; 1523 } 1524 1525 bool fuse_mount_remove(struct fuse_mount *fm) 1526 { 1527 struct fuse_conn *fc = fm->fc; 1528 bool last = false; 1529 1530 down_write(&fc->killsb); 1531 list_del_init(&fm->fc_entry); 1532 if (list_empty(&fc->mounts)) 1533 last = true; 1534 up_write(&fc->killsb); 1535 1536 return last; 1537 } 1538 EXPORT_SYMBOL_GPL(fuse_mount_remove); 1539 1540 void fuse_conn_destroy(struct fuse_mount *fm) 1541 { 1542 struct fuse_conn *fc = fm->fc; 1543 1544 if (fc->destroy) 1545 fuse_send_destroy(fm); 1546 1547 fuse_abort_conn(fc); 1548 fuse_wait_aborted(fc); 1549 1550 if (!list_empty(&fc->entry)) { 1551 mutex_lock(&fuse_mutex); 1552 list_del(&fc->entry); 1553 fuse_ctl_remove_conn(fc); 1554 mutex_unlock(&fuse_mutex); 1555 } 1556 } 1557 EXPORT_SYMBOL_GPL(fuse_conn_destroy); 1558 1559 static void fuse_sb_destroy(struct super_block *sb) 1560 { 1561 struct fuse_mount *fm = get_fuse_mount_super(sb); 1562 bool last; 1563 1564 if (fm) { 1565 last = fuse_mount_remove(fm); 1566 if (last) 1567 fuse_conn_destroy(fm); 1568 } 1569 } 1570 1571 static void fuse_kill_sb_anon(struct super_block *sb) 1572 { 1573 fuse_sb_destroy(sb); 1574 kill_anon_super(sb); 1575 } 1576 1577 static struct file_system_type fuse_fs_type = { 1578 .owner = THIS_MODULE, 1579 .name = "fuse", 1580 .fs_flags = FS_HAS_SUBTYPE | FS_USERNS_MOUNT, 1581 .init_fs_context = fuse_init_fs_context, 1582 .parameters = fuse_fs_parameters, 1583 .kill_sb = fuse_kill_sb_anon, 1584 }; 1585 MODULE_ALIAS_FS("fuse"); 1586 1587 #ifdef CONFIG_BLOCK 1588 static void fuse_kill_sb_blk(struct super_block *sb) 1589 { 1590 fuse_sb_destroy(sb); 1591 kill_block_super(sb); 1592 } 1593 1594 static struct file_system_type fuseblk_fs_type = { 1595 .owner = THIS_MODULE, 1596 .name = "fuseblk", 1597 .init_fs_context = fuse_init_fs_context, 1598 .parameters = fuse_fs_parameters, 1599 .kill_sb = fuse_kill_sb_blk, 1600 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE, 1601 }; 1602 MODULE_ALIAS_FS("fuseblk"); 1603 1604 static inline int register_fuseblk(void) 1605 { 1606 return register_filesystem(&fuseblk_fs_type); 1607 } 1608 1609 static inline void unregister_fuseblk(void) 1610 { 1611 unregister_filesystem(&fuseblk_fs_type); 1612 } 1613 #else 1614 static inline int register_fuseblk(void) 1615 { 1616 return 0; 1617 } 1618 1619 static inline void unregister_fuseblk(void) 1620 { 1621 } 1622 #endif 1623 1624 static void fuse_inode_init_once(void *foo) 1625 { 1626 struct inode *inode = foo; 1627 1628 inode_init_once(inode); 1629 } 1630 1631 static int __init fuse_fs_init(void) 1632 { 1633 int err; 1634 1635 fuse_inode_cachep = kmem_cache_create("fuse_inode", 1636 sizeof(struct fuse_inode), 0, 1637 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT, 1638 fuse_inode_init_once); 1639 err = -ENOMEM; 1640 if (!fuse_inode_cachep) 1641 goto out; 1642 1643 err = register_fuseblk(); 1644 if (err) 1645 goto out2; 1646 1647 err = register_filesystem(&fuse_fs_type); 1648 if (err) 1649 goto out3; 1650 1651 return 0; 1652 1653 out3: 1654 unregister_fuseblk(); 1655 out2: 1656 kmem_cache_destroy(fuse_inode_cachep); 1657 out: 1658 return err; 1659 } 1660 1661 static void fuse_fs_cleanup(void) 1662 { 1663 unregister_filesystem(&fuse_fs_type); 1664 unregister_fuseblk(); 1665 1666 /* 1667 * Make sure all delayed rcu free inodes are flushed before we 1668 * destroy cache. 1669 */ 1670 rcu_barrier(); 1671 kmem_cache_destroy(fuse_inode_cachep); 1672 } 1673 1674 static struct kobject *fuse_kobj; 1675 1676 static int fuse_sysfs_init(void) 1677 { 1678 int err; 1679 1680 fuse_kobj = kobject_create_and_add("fuse", fs_kobj); 1681 if (!fuse_kobj) { 1682 err = -ENOMEM; 1683 goto out_err; 1684 } 1685 1686 err = sysfs_create_mount_point(fuse_kobj, "connections"); 1687 if (err) 1688 goto out_fuse_unregister; 1689 1690 return 0; 1691 1692 out_fuse_unregister: 1693 kobject_put(fuse_kobj); 1694 out_err: 1695 return err; 1696 } 1697 1698 static void fuse_sysfs_cleanup(void) 1699 { 1700 sysfs_remove_mount_point(fuse_kobj, "connections"); 1701 kobject_put(fuse_kobj); 1702 } 1703 1704 static int __init fuse_init(void) 1705 { 1706 int res; 1707 1708 pr_info("init (API version %i.%i)\n", 1709 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION); 1710 1711 INIT_LIST_HEAD(&fuse_conn_list); 1712 res = fuse_fs_init(); 1713 if (res) 1714 goto err; 1715 1716 res = fuse_dev_init(); 1717 if (res) 1718 goto err_fs_cleanup; 1719 1720 res = fuse_sysfs_init(); 1721 if (res) 1722 goto err_dev_cleanup; 1723 1724 res = fuse_ctl_init(); 1725 if (res) 1726 goto err_sysfs_cleanup; 1727 1728 sanitize_global_limit(&max_user_bgreq); 1729 sanitize_global_limit(&max_user_congthresh); 1730 1731 return 0; 1732 1733 err_sysfs_cleanup: 1734 fuse_sysfs_cleanup(); 1735 err_dev_cleanup: 1736 fuse_dev_cleanup(); 1737 err_fs_cleanup: 1738 fuse_fs_cleanup(); 1739 err: 1740 return res; 1741 } 1742 1743 static void __exit fuse_exit(void) 1744 { 1745 pr_debug("exit\n"); 1746 1747 fuse_ctl_cleanup(); 1748 fuse_sysfs_cleanup(); 1749 fuse_fs_cleanup(); 1750 fuse_dev_cleanup(); 1751 } 1752 1753 module_init(fuse_init); 1754 module_exit(fuse_exit); 1755