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