1 /* 2 FUSE: Filesystem in Userspace 3 Copyright (C) 2001-2008 Miklos Szeredi <miklos@szeredi.hu> 4 5 This program can be distributed under the terms of the GNU GPL. 6 See the file COPYING. 7 */ 8 9 #include "fuse_i.h" 10 11 #include <linux/pagemap.h> 12 #include <linux/slab.h> 13 #include <linux/file.h> 14 #include <linux/seq_file.h> 15 #include <linux/init.h> 16 #include <linux/module.h> 17 #include <linux/moduleparam.h> 18 #include <linux/fs_context.h> 19 #include <linux/fs_parser.h> 20 #include <linux/statfs.h> 21 #include <linux/random.h> 22 #include <linux/sched.h> 23 #include <linux/exportfs.h> 24 #include <linux/posix_acl.h> 25 #include <linux/pid_namespace.h> 26 #include <uapi/linux/magic.h> 27 28 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>"); 29 MODULE_DESCRIPTION("Filesystem in Userspace"); 30 MODULE_LICENSE("GPL"); 31 32 static struct kmem_cache *fuse_inode_cachep; 33 struct list_head fuse_conn_list; 34 DEFINE_MUTEX(fuse_mutex); 35 36 static int set_global_limit(const char *val, const struct kernel_param *kp); 37 38 unsigned max_user_bgreq; 39 module_param_call(max_user_bgreq, set_global_limit, param_get_uint, 40 &max_user_bgreq, 0644); 41 __MODULE_PARM_TYPE(max_user_bgreq, "uint"); 42 MODULE_PARM_DESC(max_user_bgreq, 43 "Global limit for the maximum number of backgrounded requests an " 44 "unprivileged user can set"); 45 46 unsigned max_user_congthresh; 47 module_param_call(max_user_congthresh, set_global_limit, param_get_uint, 48 &max_user_congthresh, 0644); 49 __MODULE_PARM_TYPE(max_user_congthresh, "uint"); 50 MODULE_PARM_DESC(max_user_congthresh, 51 "Global limit for the maximum congestion threshold an " 52 "unprivileged user can set"); 53 54 #define FUSE_DEFAULT_BLKSIZE 512 55 56 /** Maximum number of outstanding background requests */ 57 #define FUSE_DEFAULT_MAX_BACKGROUND 12 58 59 /** Congestion starts at 75% of maximum */ 60 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4) 61 62 #ifdef CONFIG_BLOCK 63 static struct file_system_type fuseblk_fs_type; 64 #endif 65 66 struct fuse_forget_link *fuse_alloc_forget(void) 67 { 68 return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL_ACCOUNT); 69 } 70 71 static struct inode *fuse_alloc_inode(struct super_block *sb) 72 { 73 struct fuse_inode *fi; 74 75 fi = alloc_inode_sb(sb, fuse_inode_cachep, GFP_KERNEL); 76 if (!fi) 77 return NULL; 78 79 fi->i_time = 0; 80 fi->inval_mask = 0; 81 fi->nodeid = 0; 82 fi->nlookup = 0; 83 fi->attr_version = 0; 84 fi->orig_ino = 0; 85 fi->state = 0; 86 mutex_init(&fi->mutex); 87 spin_lock_init(&fi->lock); 88 fi->forget = fuse_alloc_forget(); 89 if (!fi->forget) 90 goto out_free; 91 92 if (IS_ENABLED(CONFIG_FUSE_DAX) && !fuse_dax_inode_alloc(sb, fi)) 93 goto out_free_forget; 94 95 return &fi->inode; 96 97 out_free_forget: 98 kfree(fi->forget); 99 out_free: 100 kmem_cache_free(fuse_inode_cachep, fi); 101 return NULL; 102 } 103 104 static void fuse_free_inode(struct inode *inode) 105 { 106 struct fuse_inode *fi = get_fuse_inode(inode); 107 108 mutex_destroy(&fi->mutex); 109 kfree(fi->forget); 110 #ifdef CONFIG_FUSE_DAX 111 kfree(fi->dax); 112 #endif 113 kmem_cache_free(fuse_inode_cachep, fi); 114 } 115 116 static void fuse_evict_inode(struct inode *inode) 117 { 118 struct fuse_inode *fi = get_fuse_inode(inode); 119 120 /* Will write inode on close/munmap and in all other dirtiers */ 121 WARN_ON(inode->i_state & I_DIRTY_INODE); 122 123 truncate_inode_pages_final(&inode->i_data); 124 clear_inode(inode); 125 if (inode->i_sb->s_flags & SB_ACTIVE) { 126 struct fuse_conn *fc = get_fuse_conn(inode); 127 128 if (FUSE_IS_DAX(inode)) 129 fuse_dax_inode_cleanup(inode); 130 if (fi->nlookup) { 131 fuse_queue_forget(fc, fi->forget, fi->nodeid, 132 fi->nlookup); 133 fi->forget = NULL; 134 } 135 } 136 if (S_ISREG(inode->i_mode) && !fuse_is_bad(inode)) { 137 WARN_ON(!list_empty(&fi->write_files)); 138 WARN_ON(!list_empty(&fi->queued_writes)); 139 } 140 } 141 142 static int fuse_reconfigure(struct fs_context *fsc) 143 { 144 struct super_block *sb = fsc->root->d_sb; 145 146 sync_filesystem(sb); 147 if (fsc->sb_flags & SB_MANDLOCK) 148 return -EINVAL; 149 150 return 0; 151 } 152 153 /* 154 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down 155 * so that it will fit. 156 */ 157 static ino_t fuse_squash_ino(u64 ino64) 158 { 159 ino_t ino = (ino_t) ino64; 160 if (sizeof(ino_t) < sizeof(u64)) 161 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8; 162 return ino; 163 } 164 165 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr, 166 u64 attr_valid, u32 cache_mask) 167 { 168 struct fuse_conn *fc = get_fuse_conn(inode); 169 struct fuse_inode *fi = get_fuse_inode(inode); 170 171 lockdep_assert_held(&fi->lock); 172 173 fi->attr_version = atomic64_inc_return(&fc->attr_version); 174 fi->i_time = attr_valid; 175 WRITE_ONCE(fi->inval_mask, 0); 176 177 inode->i_ino = fuse_squash_ino(attr->ino); 178 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777); 179 set_nlink(inode, attr->nlink); 180 inode->i_uid = make_kuid(fc->user_ns, attr->uid); 181 inode->i_gid = make_kgid(fc->user_ns, attr->gid); 182 inode->i_blocks = attr->blocks; 183 184 /* Sanitize nsecs */ 185 attr->atimensec = min_t(u32, attr->atimensec, NSEC_PER_SEC - 1); 186 attr->mtimensec = min_t(u32, attr->mtimensec, NSEC_PER_SEC - 1); 187 attr->ctimensec = min_t(u32, attr->ctimensec, NSEC_PER_SEC - 1); 188 189 inode->i_atime.tv_sec = attr->atime; 190 inode->i_atime.tv_nsec = attr->atimensec; 191 /* mtime from server may be stale due to local buffered write */ 192 if (!(cache_mask & STATX_MTIME)) { 193 inode->i_mtime.tv_sec = attr->mtime; 194 inode->i_mtime.tv_nsec = attr->mtimensec; 195 } 196 if (!(cache_mask & STATX_CTIME)) { 197 inode->i_ctime.tv_sec = attr->ctime; 198 inode->i_ctime.tv_nsec = attr->ctimensec; 199 } 200 201 if (attr->blksize != 0) 202 inode->i_blkbits = ilog2(attr->blksize); 203 else 204 inode->i_blkbits = inode->i_sb->s_blocksize_bits; 205 206 /* 207 * Don't set the sticky bit in i_mode, unless we want the VFS 208 * to check permissions. This prevents failures due to the 209 * check in may_delete(). 210 */ 211 fi->orig_i_mode = inode->i_mode; 212 if (!fc->default_permissions) 213 inode->i_mode &= ~S_ISVTX; 214 215 fi->orig_ino = attr->ino; 216 217 /* 218 * We are refreshing inode data and it is possible that another 219 * client set suid/sgid or security.capability xattr. So clear 220 * S_NOSEC. Ideally, we could have cleared it only if suid/sgid 221 * was set or if security.capability xattr was set. But we don't 222 * know if security.capability has been set or not. So clear it 223 * anyway. Its less efficient but should be safe. 224 */ 225 inode->i_flags &= ~S_NOSEC; 226 } 227 228 u32 fuse_get_cache_mask(struct inode *inode) 229 { 230 struct fuse_conn *fc = get_fuse_conn(inode); 231 232 if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) 233 return 0; 234 235 return STATX_MTIME | STATX_CTIME | STATX_SIZE; 236 } 237 238 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr, 239 u64 attr_valid, u64 attr_version) 240 { 241 struct fuse_conn *fc = get_fuse_conn(inode); 242 struct fuse_inode *fi = get_fuse_inode(inode); 243 u32 cache_mask; 244 loff_t oldsize; 245 struct timespec64 old_mtime; 246 247 spin_lock(&fi->lock); 248 /* 249 * In case of writeback_cache enabled, writes update mtime, ctime and 250 * may update i_size. In these cases trust the cached value in the 251 * inode. 252 */ 253 cache_mask = fuse_get_cache_mask(inode); 254 if (cache_mask & STATX_SIZE) 255 attr->size = i_size_read(inode); 256 257 if (cache_mask & STATX_MTIME) { 258 attr->mtime = inode->i_mtime.tv_sec; 259 attr->mtimensec = inode->i_mtime.tv_nsec; 260 } 261 if (cache_mask & STATX_CTIME) { 262 attr->ctime = inode->i_ctime.tv_sec; 263 attr->ctimensec = inode->i_ctime.tv_nsec; 264 } 265 266 if ((attr_version != 0 && fi->attr_version > attr_version) || 267 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) { 268 spin_unlock(&fi->lock); 269 return; 270 } 271 272 old_mtime = inode->i_mtime; 273 fuse_change_attributes_common(inode, attr, attr_valid, cache_mask); 274 275 oldsize = inode->i_size; 276 /* 277 * In case of writeback_cache enabled, the cached writes beyond EOF 278 * extend local i_size without keeping userspace server in sync. So, 279 * attr->size coming from server can be stale. We cannot trust it. 280 */ 281 if (!(cache_mask & STATX_SIZE)) 282 i_size_write(inode, attr->size); 283 spin_unlock(&fi->lock); 284 285 if (!cache_mask && S_ISREG(inode->i_mode)) { 286 bool inval = false; 287 288 if (oldsize != attr->size) { 289 truncate_pagecache(inode, attr->size); 290 if (!fc->explicit_inval_data) 291 inval = true; 292 } else if (fc->auto_inval_data) { 293 struct timespec64 new_mtime = { 294 .tv_sec = attr->mtime, 295 .tv_nsec = attr->mtimensec, 296 }; 297 298 /* 299 * Auto inval mode also checks and invalidates if mtime 300 * has changed. 301 */ 302 if (!timespec64_equal(&old_mtime, &new_mtime)) 303 inval = true; 304 } 305 306 if (inval) 307 invalidate_inode_pages2(inode->i_mapping); 308 } 309 310 if (IS_ENABLED(CONFIG_FUSE_DAX)) 311 fuse_dax_dontcache(inode, attr->flags); 312 } 313 314 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr) 315 { 316 inode->i_mode = attr->mode & S_IFMT; 317 inode->i_size = attr->size; 318 inode->i_mtime.tv_sec = attr->mtime; 319 inode->i_mtime.tv_nsec = attr->mtimensec; 320 inode->i_ctime.tv_sec = attr->ctime; 321 inode->i_ctime.tv_nsec = attr->ctimensec; 322 if (S_ISREG(inode->i_mode)) { 323 fuse_init_common(inode); 324 fuse_init_file_inode(inode, attr->flags); 325 } else if (S_ISDIR(inode->i_mode)) 326 fuse_init_dir(inode); 327 else if (S_ISLNK(inode->i_mode)) 328 fuse_init_symlink(inode); 329 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || 330 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { 331 fuse_init_common(inode); 332 init_special_inode(inode, inode->i_mode, 333 new_decode_dev(attr->rdev)); 334 } else 335 BUG(); 336 } 337 338 static int fuse_inode_eq(struct inode *inode, void *_nodeidp) 339 { 340 u64 nodeid = *(u64 *) _nodeidp; 341 if (get_node_id(inode) == nodeid) 342 return 1; 343 else 344 return 0; 345 } 346 347 static int fuse_inode_set(struct inode *inode, void *_nodeidp) 348 { 349 u64 nodeid = *(u64 *) _nodeidp; 350 get_fuse_inode(inode)->nodeid = nodeid; 351 return 0; 352 } 353 354 struct inode *fuse_iget(struct super_block *sb, u64 nodeid, 355 int generation, struct fuse_attr *attr, 356 u64 attr_valid, u64 attr_version) 357 { 358 struct inode *inode; 359 struct fuse_inode *fi; 360 struct fuse_conn *fc = get_fuse_conn_super(sb); 361 362 /* 363 * Auto mount points get their node id from the submount root, which is 364 * not a unique identifier within this filesystem. 365 * 366 * To avoid conflicts, do not place submount points into the inode hash 367 * table. 368 */ 369 if (fc->auto_submounts && (attr->flags & FUSE_ATTR_SUBMOUNT) && 370 S_ISDIR(attr->mode)) { 371 inode = new_inode(sb); 372 if (!inode) 373 return NULL; 374 375 fuse_init_inode(inode, attr); 376 get_fuse_inode(inode)->nodeid = nodeid; 377 inode->i_flags |= S_AUTOMOUNT; 378 goto done; 379 } 380 381 retry: 382 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid); 383 if (!inode) 384 return NULL; 385 386 if ((inode->i_state & I_NEW)) { 387 inode->i_flags |= S_NOATIME; 388 if (!fc->writeback_cache || !S_ISREG(attr->mode)) 389 inode->i_flags |= S_NOCMTIME; 390 inode->i_generation = generation; 391 fuse_init_inode(inode, attr); 392 unlock_new_inode(inode); 393 } else if (fuse_stale_inode(inode, generation, attr)) { 394 /* nodeid was reused, any I/O on the old inode should fail */ 395 fuse_make_bad(inode); 396 iput(inode); 397 goto retry; 398 } 399 done: 400 fi = get_fuse_inode(inode); 401 spin_lock(&fi->lock); 402 fi->nlookup++; 403 spin_unlock(&fi->lock); 404 fuse_change_attributes(inode, attr, attr_valid, attr_version); 405 406 return inode; 407 } 408 409 struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid, 410 struct fuse_mount **fm) 411 { 412 struct fuse_mount *fm_iter; 413 struct inode *inode; 414 415 WARN_ON(!rwsem_is_locked(&fc->killsb)); 416 list_for_each_entry(fm_iter, &fc->mounts, fc_entry) { 417 if (!fm_iter->sb) 418 continue; 419 420 inode = ilookup5(fm_iter->sb, nodeid, fuse_inode_eq, &nodeid); 421 if (inode) { 422 if (fm) 423 *fm = fm_iter; 424 return inode; 425 } 426 } 427 428 return NULL; 429 } 430 431 int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid, 432 loff_t offset, loff_t len) 433 { 434 struct fuse_inode *fi; 435 struct inode *inode; 436 pgoff_t pg_start; 437 pgoff_t pg_end; 438 439 inode = fuse_ilookup(fc, nodeid, NULL); 440 if (!inode) 441 return -ENOENT; 442 443 fi = get_fuse_inode(inode); 444 spin_lock(&fi->lock); 445 fi->attr_version = atomic64_inc_return(&fc->attr_version); 446 spin_unlock(&fi->lock); 447 448 fuse_invalidate_attr(inode); 449 forget_all_cached_acls(inode); 450 if (offset >= 0) { 451 pg_start = offset >> PAGE_SHIFT; 452 if (len <= 0) 453 pg_end = -1; 454 else 455 pg_end = (offset + len - 1) >> PAGE_SHIFT; 456 invalidate_inode_pages2_range(inode->i_mapping, 457 pg_start, pg_end); 458 } 459 iput(inode); 460 return 0; 461 } 462 463 bool fuse_lock_inode(struct inode *inode) 464 { 465 bool locked = false; 466 467 if (!get_fuse_conn(inode)->parallel_dirops) { 468 mutex_lock(&get_fuse_inode(inode)->mutex); 469 locked = true; 470 } 471 472 return locked; 473 } 474 475 void fuse_unlock_inode(struct inode *inode, bool locked) 476 { 477 if (locked) 478 mutex_unlock(&get_fuse_inode(inode)->mutex); 479 } 480 481 static void fuse_umount_begin(struct super_block *sb) 482 { 483 struct fuse_conn *fc = get_fuse_conn_super(sb); 484 485 if (fc->no_force_umount) 486 return; 487 488 fuse_abort_conn(fc); 489 490 // Only retire block-device-based superblocks. 491 if (sb->s_bdev != NULL) 492 retire_super(sb); 493 } 494 495 static void fuse_send_destroy(struct fuse_mount *fm) 496 { 497 if (fm->fc->conn_init) { 498 FUSE_ARGS(args); 499 500 args.opcode = FUSE_DESTROY; 501 args.force = true; 502 args.nocreds = true; 503 fuse_simple_request(fm, &args); 504 } 505 } 506 507 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr) 508 { 509 stbuf->f_type = FUSE_SUPER_MAGIC; 510 stbuf->f_bsize = attr->bsize; 511 stbuf->f_frsize = attr->frsize; 512 stbuf->f_blocks = attr->blocks; 513 stbuf->f_bfree = attr->bfree; 514 stbuf->f_bavail = attr->bavail; 515 stbuf->f_files = attr->files; 516 stbuf->f_ffree = attr->ffree; 517 stbuf->f_namelen = attr->namelen; 518 /* fsid is left zero */ 519 } 520 521 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf) 522 { 523 struct super_block *sb = dentry->d_sb; 524 struct fuse_mount *fm = get_fuse_mount_super(sb); 525 FUSE_ARGS(args); 526 struct fuse_statfs_out outarg; 527 int err; 528 529 if (!fuse_allow_current_process(fm->fc)) { 530 buf->f_type = FUSE_SUPER_MAGIC; 531 return 0; 532 } 533 534 memset(&outarg, 0, sizeof(outarg)); 535 args.in_numargs = 0; 536 args.opcode = FUSE_STATFS; 537 args.nodeid = get_node_id(d_inode(dentry)); 538 args.out_numargs = 1; 539 args.out_args[0].size = sizeof(outarg); 540 args.out_args[0].value = &outarg; 541 err = fuse_simple_request(fm, &args); 542 if (!err) 543 convert_fuse_statfs(buf, &outarg.st); 544 return err; 545 } 546 547 static struct fuse_sync_bucket *fuse_sync_bucket_alloc(void) 548 { 549 struct fuse_sync_bucket *bucket; 550 551 bucket = kzalloc(sizeof(*bucket), GFP_KERNEL | __GFP_NOFAIL); 552 if (bucket) { 553 init_waitqueue_head(&bucket->waitq); 554 /* Initial active count */ 555 atomic_set(&bucket->count, 1); 556 } 557 return bucket; 558 } 559 560 static void fuse_sync_fs_writes(struct fuse_conn *fc) 561 { 562 struct fuse_sync_bucket *bucket, *new_bucket; 563 int count; 564 565 new_bucket = fuse_sync_bucket_alloc(); 566 spin_lock(&fc->lock); 567 bucket = rcu_dereference_protected(fc->curr_bucket, 1); 568 count = atomic_read(&bucket->count); 569 WARN_ON(count < 1); 570 /* No outstanding writes? */ 571 if (count == 1) { 572 spin_unlock(&fc->lock); 573 kfree(new_bucket); 574 return; 575 } 576 577 /* 578 * Completion of new bucket depends on completion of this bucket, so add 579 * one more count. 580 */ 581 atomic_inc(&new_bucket->count); 582 rcu_assign_pointer(fc->curr_bucket, new_bucket); 583 spin_unlock(&fc->lock); 584 /* 585 * Drop initial active count. At this point if all writes in this and 586 * ancestor buckets complete, the count will go to zero and this task 587 * will be woken up. 588 */ 589 atomic_dec(&bucket->count); 590 591 wait_event(bucket->waitq, atomic_read(&bucket->count) == 0); 592 593 /* Drop temp count on descendant bucket */ 594 fuse_sync_bucket_dec(new_bucket); 595 kfree_rcu(bucket, rcu); 596 } 597 598 static int fuse_sync_fs(struct super_block *sb, int wait) 599 { 600 struct fuse_mount *fm = get_fuse_mount_super(sb); 601 struct fuse_conn *fc = fm->fc; 602 struct fuse_syncfs_in inarg; 603 FUSE_ARGS(args); 604 int err; 605 606 /* 607 * Userspace cannot handle the wait == 0 case. Avoid a 608 * gratuitous roundtrip. 609 */ 610 if (!wait) 611 return 0; 612 613 /* The filesystem is being unmounted. Nothing to do. */ 614 if (!sb->s_root) 615 return 0; 616 617 if (!fc->sync_fs) 618 return 0; 619 620 fuse_sync_fs_writes(fc); 621 622 memset(&inarg, 0, sizeof(inarg)); 623 args.in_numargs = 1; 624 args.in_args[0].size = sizeof(inarg); 625 args.in_args[0].value = &inarg; 626 args.opcode = FUSE_SYNCFS; 627 args.nodeid = get_node_id(sb->s_root->d_inode); 628 args.out_numargs = 0; 629 630 err = fuse_simple_request(fm, &args); 631 if (err == -ENOSYS) { 632 fc->sync_fs = 0; 633 err = 0; 634 } 635 636 return err; 637 } 638 639 enum { 640 OPT_SOURCE, 641 OPT_SUBTYPE, 642 OPT_FD, 643 OPT_ROOTMODE, 644 OPT_USER_ID, 645 OPT_GROUP_ID, 646 OPT_DEFAULT_PERMISSIONS, 647 OPT_ALLOW_OTHER, 648 OPT_MAX_READ, 649 OPT_BLKSIZE, 650 OPT_ERR 651 }; 652 653 static const struct fs_parameter_spec fuse_fs_parameters[] = { 654 fsparam_string ("source", OPT_SOURCE), 655 fsparam_u32 ("fd", OPT_FD), 656 fsparam_u32oct ("rootmode", OPT_ROOTMODE), 657 fsparam_u32 ("user_id", OPT_USER_ID), 658 fsparam_u32 ("group_id", OPT_GROUP_ID), 659 fsparam_flag ("default_permissions", OPT_DEFAULT_PERMISSIONS), 660 fsparam_flag ("allow_other", OPT_ALLOW_OTHER), 661 fsparam_u32 ("max_read", OPT_MAX_READ), 662 fsparam_u32 ("blksize", OPT_BLKSIZE), 663 fsparam_string ("subtype", OPT_SUBTYPE), 664 {} 665 }; 666 667 static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param) 668 { 669 struct fs_parse_result result; 670 struct fuse_fs_context *ctx = fsc->fs_private; 671 int opt; 672 673 if (fsc->purpose == FS_CONTEXT_FOR_RECONFIGURE) { 674 /* 675 * Ignore options coming from mount(MS_REMOUNT) for backward 676 * compatibility. 677 */ 678 if (fsc->oldapi) 679 return 0; 680 681 return invalfc(fsc, "No changes allowed in reconfigure"); 682 } 683 684 opt = fs_parse(fsc, fuse_fs_parameters, param, &result); 685 if (opt < 0) 686 return opt; 687 688 switch (opt) { 689 case OPT_SOURCE: 690 if (fsc->source) 691 return invalfc(fsc, "Multiple sources specified"); 692 fsc->source = param->string; 693 param->string = NULL; 694 break; 695 696 case OPT_SUBTYPE: 697 if (ctx->subtype) 698 return invalfc(fsc, "Multiple subtypes specified"); 699 ctx->subtype = param->string; 700 param->string = NULL; 701 return 0; 702 703 case OPT_FD: 704 ctx->fd = result.uint_32; 705 ctx->fd_present = true; 706 break; 707 708 case OPT_ROOTMODE: 709 if (!fuse_valid_type(result.uint_32)) 710 return invalfc(fsc, "Invalid rootmode"); 711 ctx->rootmode = result.uint_32; 712 ctx->rootmode_present = true; 713 break; 714 715 case OPT_USER_ID: 716 ctx->user_id = make_kuid(fsc->user_ns, result.uint_32); 717 if (!uid_valid(ctx->user_id)) 718 return invalfc(fsc, "Invalid user_id"); 719 ctx->user_id_present = true; 720 break; 721 722 case OPT_GROUP_ID: 723 ctx->group_id = make_kgid(fsc->user_ns, result.uint_32); 724 if (!gid_valid(ctx->group_id)) 725 return invalfc(fsc, "Invalid group_id"); 726 ctx->group_id_present = true; 727 break; 728 729 case OPT_DEFAULT_PERMISSIONS: 730 ctx->default_permissions = true; 731 break; 732 733 case OPT_ALLOW_OTHER: 734 ctx->allow_other = true; 735 break; 736 737 case OPT_MAX_READ: 738 ctx->max_read = result.uint_32; 739 break; 740 741 case OPT_BLKSIZE: 742 if (!ctx->is_bdev) 743 return invalfc(fsc, "blksize only supported for fuseblk"); 744 ctx->blksize = result.uint_32; 745 break; 746 747 default: 748 return -EINVAL; 749 } 750 751 return 0; 752 } 753 754 static void fuse_free_fsc(struct fs_context *fsc) 755 { 756 struct fuse_fs_context *ctx = fsc->fs_private; 757 758 if (ctx) { 759 kfree(ctx->subtype); 760 kfree(ctx); 761 } 762 } 763 764 static int fuse_show_options(struct seq_file *m, struct dentry *root) 765 { 766 struct super_block *sb = root->d_sb; 767 struct fuse_conn *fc = get_fuse_conn_super(sb); 768 769 if (fc->legacy_opts_show) { 770 seq_printf(m, ",user_id=%u", 771 from_kuid_munged(fc->user_ns, fc->user_id)); 772 seq_printf(m, ",group_id=%u", 773 from_kgid_munged(fc->user_ns, fc->group_id)); 774 if (fc->default_permissions) 775 seq_puts(m, ",default_permissions"); 776 if (fc->allow_other) 777 seq_puts(m, ",allow_other"); 778 if (fc->max_read != ~0) 779 seq_printf(m, ",max_read=%u", fc->max_read); 780 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE) 781 seq_printf(m, ",blksize=%lu", sb->s_blocksize); 782 } 783 #ifdef CONFIG_FUSE_DAX 784 if (fc->dax_mode == FUSE_DAX_ALWAYS) 785 seq_puts(m, ",dax=always"); 786 else if (fc->dax_mode == FUSE_DAX_NEVER) 787 seq_puts(m, ",dax=never"); 788 else if (fc->dax_mode == FUSE_DAX_INODE_USER) 789 seq_puts(m, ",dax=inode"); 790 #endif 791 792 return 0; 793 } 794 795 static void fuse_iqueue_init(struct fuse_iqueue *fiq, 796 const struct fuse_iqueue_ops *ops, 797 void *priv) 798 { 799 memset(fiq, 0, sizeof(struct fuse_iqueue)); 800 spin_lock_init(&fiq->lock); 801 init_waitqueue_head(&fiq->waitq); 802 INIT_LIST_HEAD(&fiq->pending); 803 INIT_LIST_HEAD(&fiq->interrupts); 804 fiq->forget_list_tail = &fiq->forget_list_head; 805 fiq->connected = 1; 806 fiq->ops = ops; 807 fiq->priv = priv; 808 } 809 810 static void fuse_pqueue_init(struct fuse_pqueue *fpq) 811 { 812 unsigned int i; 813 814 spin_lock_init(&fpq->lock); 815 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++) 816 INIT_LIST_HEAD(&fpq->processing[i]); 817 INIT_LIST_HEAD(&fpq->io); 818 fpq->connected = 1; 819 } 820 821 void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm, 822 struct user_namespace *user_ns, 823 const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv) 824 { 825 memset(fc, 0, sizeof(*fc)); 826 spin_lock_init(&fc->lock); 827 spin_lock_init(&fc->bg_lock); 828 init_rwsem(&fc->killsb); 829 refcount_set(&fc->count, 1); 830 atomic_set(&fc->dev_count, 1); 831 init_waitqueue_head(&fc->blocked_waitq); 832 fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv); 833 INIT_LIST_HEAD(&fc->bg_queue); 834 INIT_LIST_HEAD(&fc->entry); 835 INIT_LIST_HEAD(&fc->devices); 836 atomic_set(&fc->num_waiting, 0); 837 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND; 838 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD; 839 atomic64_set(&fc->khctr, 0); 840 fc->polled_files = RB_ROOT; 841 fc->blocked = 0; 842 fc->initialized = 0; 843 fc->connected = 1; 844 atomic64_set(&fc->attr_version, 1); 845 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key)); 846 fc->pid_ns = get_pid_ns(task_active_pid_ns(current)); 847 fc->user_ns = get_user_ns(user_ns); 848 fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ; 849 fc->max_pages_limit = FUSE_MAX_MAX_PAGES; 850 851 INIT_LIST_HEAD(&fc->mounts); 852 list_add(&fm->fc_entry, &fc->mounts); 853 fm->fc = fc; 854 } 855 EXPORT_SYMBOL_GPL(fuse_conn_init); 856 857 void fuse_conn_put(struct fuse_conn *fc) 858 { 859 if (refcount_dec_and_test(&fc->count)) { 860 struct fuse_iqueue *fiq = &fc->iq; 861 struct fuse_sync_bucket *bucket; 862 863 if (IS_ENABLED(CONFIG_FUSE_DAX)) 864 fuse_dax_conn_free(fc); 865 if (fiq->ops->release) 866 fiq->ops->release(fiq); 867 put_pid_ns(fc->pid_ns); 868 put_user_ns(fc->user_ns); 869 bucket = rcu_dereference_protected(fc->curr_bucket, 1); 870 if (bucket) { 871 WARN_ON(atomic_read(&bucket->count) != 1); 872 kfree(bucket); 873 } 874 fc->release(fc); 875 } 876 } 877 EXPORT_SYMBOL_GPL(fuse_conn_put); 878 879 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc) 880 { 881 refcount_inc(&fc->count); 882 return fc; 883 } 884 EXPORT_SYMBOL_GPL(fuse_conn_get); 885 886 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned mode) 887 { 888 struct fuse_attr attr; 889 memset(&attr, 0, sizeof(attr)); 890 891 attr.mode = mode; 892 attr.ino = FUSE_ROOT_ID; 893 attr.nlink = 1; 894 return fuse_iget(sb, 1, 0, &attr, 0, 0); 895 } 896 897 struct fuse_inode_handle { 898 u64 nodeid; 899 u32 generation; 900 }; 901 902 static struct dentry *fuse_get_dentry(struct super_block *sb, 903 struct fuse_inode_handle *handle) 904 { 905 struct fuse_conn *fc = get_fuse_conn_super(sb); 906 struct inode *inode; 907 struct dentry *entry; 908 int err = -ESTALE; 909 910 if (handle->nodeid == 0) 911 goto out_err; 912 913 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid); 914 if (!inode) { 915 struct fuse_entry_out outarg; 916 const struct qstr name = QSTR_INIT(".", 1); 917 918 if (!fc->export_support) 919 goto out_err; 920 921 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg, 922 &inode); 923 if (err && err != -ENOENT) 924 goto out_err; 925 if (err || !inode) { 926 err = -ESTALE; 927 goto out_err; 928 } 929 err = -EIO; 930 if (get_node_id(inode) != handle->nodeid) 931 goto out_iput; 932 } 933 err = -ESTALE; 934 if (inode->i_generation != handle->generation) 935 goto out_iput; 936 937 entry = d_obtain_alias(inode); 938 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID) 939 fuse_invalidate_entry_cache(entry); 940 941 return entry; 942 943 out_iput: 944 iput(inode); 945 out_err: 946 return ERR_PTR(err); 947 } 948 949 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len, 950 struct inode *parent) 951 { 952 int len = parent ? 6 : 3; 953 u64 nodeid; 954 u32 generation; 955 956 if (*max_len < len) { 957 *max_len = len; 958 return FILEID_INVALID; 959 } 960 961 nodeid = get_fuse_inode(inode)->nodeid; 962 generation = inode->i_generation; 963 964 fh[0] = (u32)(nodeid >> 32); 965 fh[1] = (u32)(nodeid & 0xffffffff); 966 fh[2] = generation; 967 968 if (parent) { 969 nodeid = get_fuse_inode(parent)->nodeid; 970 generation = parent->i_generation; 971 972 fh[3] = (u32)(nodeid >> 32); 973 fh[4] = (u32)(nodeid & 0xffffffff); 974 fh[5] = generation; 975 } 976 977 *max_len = len; 978 return parent ? 0x82 : 0x81; 979 } 980 981 static struct dentry *fuse_fh_to_dentry(struct super_block *sb, 982 struct fid *fid, int fh_len, int fh_type) 983 { 984 struct fuse_inode_handle handle; 985 986 if ((fh_type != 0x81 && fh_type != 0x82) || fh_len < 3) 987 return NULL; 988 989 handle.nodeid = (u64) fid->raw[0] << 32; 990 handle.nodeid |= (u64) fid->raw[1]; 991 handle.generation = fid->raw[2]; 992 return fuse_get_dentry(sb, &handle); 993 } 994 995 static struct dentry *fuse_fh_to_parent(struct super_block *sb, 996 struct fid *fid, int fh_len, int fh_type) 997 { 998 struct fuse_inode_handle parent; 999 1000 if (fh_type != 0x82 || fh_len < 6) 1001 return NULL; 1002 1003 parent.nodeid = (u64) fid->raw[3] << 32; 1004 parent.nodeid |= (u64) fid->raw[4]; 1005 parent.generation = fid->raw[5]; 1006 return fuse_get_dentry(sb, &parent); 1007 } 1008 1009 static struct dentry *fuse_get_parent(struct dentry *child) 1010 { 1011 struct inode *child_inode = d_inode(child); 1012 struct fuse_conn *fc = get_fuse_conn(child_inode); 1013 struct inode *inode; 1014 struct dentry *parent; 1015 struct fuse_entry_out outarg; 1016 int err; 1017 1018 if (!fc->export_support) 1019 return ERR_PTR(-ESTALE); 1020 1021 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode), 1022 &dotdot_name, &outarg, &inode); 1023 if (err) { 1024 if (err == -ENOENT) 1025 return ERR_PTR(-ESTALE); 1026 return ERR_PTR(err); 1027 } 1028 1029 parent = d_obtain_alias(inode); 1030 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID) 1031 fuse_invalidate_entry_cache(parent); 1032 1033 return parent; 1034 } 1035 1036 static const struct export_operations fuse_export_operations = { 1037 .fh_to_dentry = fuse_fh_to_dentry, 1038 .fh_to_parent = fuse_fh_to_parent, 1039 .encode_fh = fuse_encode_fh, 1040 .get_parent = fuse_get_parent, 1041 }; 1042 1043 static const struct super_operations fuse_super_operations = { 1044 .alloc_inode = fuse_alloc_inode, 1045 .free_inode = fuse_free_inode, 1046 .evict_inode = fuse_evict_inode, 1047 .write_inode = fuse_write_inode, 1048 .drop_inode = generic_delete_inode, 1049 .umount_begin = fuse_umount_begin, 1050 .statfs = fuse_statfs, 1051 .sync_fs = fuse_sync_fs, 1052 .show_options = fuse_show_options, 1053 }; 1054 1055 static void sanitize_global_limit(unsigned *limit) 1056 { 1057 /* 1058 * The default maximum number of async requests is calculated to consume 1059 * 1/2^13 of the total memory, assuming 392 bytes per request. 1060 */ 1061 if (*limit == 0) 1062 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392; 1063 1064 if (*limit >= 1 << 16) 1065 *limit = (1 << 16) - 1; 1066 } 1067 1068 static int set_global_limit(const char *val, const struct kernel_param *kp) 1069 { 1070 int rv; 1071 1072 rv = param_set_uint(val, kp); 1073 if (rv) 1074 return rv; 1075 1076 sanitize_global_limit((unsigned *)kp->arg); 1077 1078 return 0; 1079 } 1080 1081 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg) 1082 { 1083 int cap_sys_admin = capable(CAP_SYS_ADMIN); 1084 1085 if (arg->minor < 13) 1086 return; 1087 1088 sanitize_global_limit(&max_user_bgreq); 1089 sanitize_global_limit(&max_user_congthresh); 1090 1091 spin_lock(&fc->bg_lock); 1092 if (arg->max_background) { 1093 fc->max_background = arg->max_background; 1094 1095 if (!cap_sys_admin && fc->max_background > max_user_bgreq) 1096 fc->max_background = max_user_bgreq; 1097 } 1098 if (arg->congestion_threshold) { 1099 fc->congestion_threshold = arg->congestion_threshold; 1100 1101 if (!cap_sys_admin && 1102 fc->congestion_threshold > max_user_congthresh) 1103 fc->congestion_threshold = max_user_congthresh; 1104 } 1105 spin_unlock(&fc->bg_lock); 1106 } 1107 1108 struct fuse_init_args { 1109 struct fuse_args args; 1110 struct fuse_init_in in; 1111 struct fuse_init_out out; 1112 }; 1113 1114 static void process_init_reply(struct fuse_mount *fm, struct fuse_args *args, 1115 int error) 1116 { 1117 struct fuse_conn *fc = fm->fc; 1118 struct fuse_init_args *ia = container_of(args, typeof(*ia), args); 1119 struct fuse_init_out *arg = &ia->out; 1120 bool ok = true; 1121 1122 if (error || arg->major != FUSE_KERNEL_VERSION) 1123 ok = false; 1124 else { 1125 unsigned long ra_pages; 1126 1127 process_init_limits(fc, arg); 1128 1129 if (arg->minor >= 6) { 1130 u64 flags = arg->flags | (u64) arg->flags2 << 32; 1131 1132 ra_pages = arg->max_readahead / PAGE_SIZE; 1133 if (flags & FUSE_ASYNC_READ) 1134 fc->async_read = 1; 1135 if (!(flags & FUSE_POSIX_LOCKS)) 1136 fc->no_lock = 1; 1137 if (arg->minor >= 17) { 1138 if (!(flags & FUSE_FLOCK_LOCKS)) 1139 fc->no_flock = 1; 1140 } else { 1141 if (!(flags & FUSE_POSIX_LOCKS)) 1142 fc->no_flock = 1; 1143 } 1144 if (flags & FUSE_ATOMIC_O_TRUNC) 1145 fc->atomic_o_trunc = 1; 1146 if (arg->minor >= 9) { 1147 /* LOOKUP has dependency on proto version */ 1148 if (flags & FUSE_EXPORT_SUPPORT) 1149 fc->export_support = 1; 1150 } 1151 if (flags & FUSE_BIG_WRITES) 1152 fc->big_writes = 1; 1153 if (flags & FUSE_DONT_MASK) 1154 fc->dont_mask = 1; 1155 if (flags & FUSE_AUTO_INVAL_DATA) 1156 fc->auto_inval_data = 1; 1157 else if (flags & FUSE_EXPLICIT_INVAL_DATA) 1158 fc->explicit_inval_data = 1; 1159 if (flags & FUSE_DO_READDIRPLUS) { 1160 fc->do_readdirplus = 1; 1161 if (flags & FUSE_READDIRPLUS_AUTO) 1162 fc->readdirplus_auto = 1; 1163 } 1164 if (flags & FUSE_ASYNC_DIO) 1165 fc->async_dio = 1; 1166 if (flags & FUSE_WRITEBACK_CACHE) 1167 fc->writeback_cache = 1; 1168 if (flags & FUSE_PARALLEL_DIROPS) 1169 fc->parallel_dirops = 1; 1170 if (flags & FUSE_HANDLE_KILLPRIV) 1171 fc->handle_killpriv = 1; 1172 if (arg->time_gran && arg->time_gran <= 1000000000) 1173 fm->sb->s_time_gran = arg->time_gran; 1174 if ((flags & FUSE_POSIX_ACL)) { 1175 fc->default_permissions = 1; 1176 fc->posix_acl = 1; 1177 fm->sb->s_xattr = fuse_acl_xattr_handlers; 1178 } 1179 if (flags & FUSE_CACHE_SYMLINKS) 1180 fc->cache_symlinks = 1; 1181 if (flags & FUSE_ABORT_ERROR) 1182 fc->abort_err = 1; 1183 if (flags & FUSE_MAX_PAGES) { 1184 fc->max_pages = 1185 min_t(unsigned int, fc->max_pages_limit, 1186 max_t(unsigned int, arg->max_pages, 1)); 1187 } 1188 if (IS_ENABLED(CONFIG_FUSE_DAX)) { 1189 if (flags & FUSE_MAP_ALIGNMENT && 1190 !fuse_dax_check_alignment(fc, arg->map_alignment)) { 1191 ok = false; 1192 } 1193 if (flags & FUSE_HAS_INODE_DAX) 1194 fc->inode_dax = 1; 1195 } 1196 if (flags & FUSE_HANDLE_KILLPRIV_V2) { 1197 fc->handle_killpriv_v2 = 1; 1198 fm->sb->s_flags |= SB_NOSEC; 1199 } 1200 if (flags & FUSE_SETXATTR_EXT) 1201 fc->setxattr_ext = 1; 1202 if (flags & FUSE_SECURITY_CTX) 1203 fc->init_security = 1; 1204 } else { 1205 ra_pages = fc->max_read / PAGE_SIZE; 1206 fc->no_lock = 1; 1207 fc->no_flock = 1; 1208 } 1209 1210 fm->sb->s_bdi->ra_pages = 1211 min(fm->sb->s_bdi->ra_pages, ra_pages); 1212 fc->minor = arg->minor; 1213 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write; 1214 fc->max_write = max_t(unsigned, 4096, fc->max_write); 1215 fc->conn_init = 1; 1216 } 1217 kfree(ia); 1218 1219 if (!ok) { 1220 fc->conn_init = 0; 1221 fc->conn_error = 1; 1222 } 1223 1224 fuse_set_initialized(fc); 1225 wake_up_all(&fc->blocked_waitq); 1226 } 1227 1228 void fuse_send_init(struct fuse_mount *fm) 1229 { 1230 struct fuse_init_args *ia; 1231 u64 flags; 1232 1233 ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL); 1234 1235 ia->in.major = FUSE_KERNEL_VERSION; 1236 ia->in.minor = FUSE_KERNEL_MINOR_VERSION; 1237 ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE; 1238 flags = 1239 FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC | 1240 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK | 1241 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ | 1242 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA | 1243 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO | 1244 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT | 1245 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL | 1246 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS | 1247 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA | 1248 FUSE_HANDLE_KILLPRIV_V2 | FUSE_SETXATTR_EXT | FUSE_INIT_EXT | 1249 FUSE_SECURITY_CTX; 1250 #ifdef CONFIG_FUSE_DAX 1251 if (fm->fc->dax) 1252 flags |= FUSE_MAP_ALIGNMENT; 1253 if (fuse_is_inode_dax_mode(fm->fc->dax_mode)) 1254 flags |= FUSE_HAS_INODE_DAX; 1255 #endif 1256 if (fm->fc->auto_submounts) 1257 flags |= FUSE_SUBMOUNTS; 1258 1259 ia->in.flags = flags; 1260 ia->in.flags2 = flags >> 32; 1261 1262 ia->args.opcode = FUSE_INIT; 1263 ia->args.in_numargs = 1; 1264 ia->args.in_args[0].size = sizeof(ia->in); 1265 ia->args.in_args[0].value = &ia->in; 1266 ia->args.out_numargs = 1; 1267 /* Variable length argument used for backward compatibility 1268 with interface version < 7.5. Rest of init_out is zeroed 1269 by do_get_request(), so a short reply is not a problem */ 1270 ia->args.out_argvar = true; 1271 ia->args.out_args[0].size = sizeof(ia->out); 1272 ia->args.out_args[0].value = &ia->out; 1273 ia->args.force = true; 1274 ia->args.nocreds = true; 1275 ia->args.end = process_init_reply; 1276 1277 if (fuse_simple_background(fm, &ia->args, GFP_KERNEL) != 0) 1278 process_init_reply(fm, &ia->args, -ENOTCONN); 1279 } 1280 EXPORT_SYMBOL_GPL(fuse_send_init); 1281 1282 void fuse_free_conn(struct fuse_conn *fc) 1283 { 1284 WARN_ON(!list_empty(&fc->devices)); 1285 kfree_rcu(fc, rcu); 1286 } 1287 EXPORT_SYMBOL_GPL(fuse_free_conn); 1288 1289 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb) 1290 { 1291 int err; 1292 char *suffix = ""; 1293 1294 if (sb->s_bdev) { 1295 suffix = "-fuseblk"; 1296 /* 1297 * sb->s_bdi points to blkdev's bdi however we want to redirect 1298 * it to our private bdi... 1299 */ 1300 bdi_put(sb->s_bdi); 1301 sb->s_bdi = &noop_backing_dev_info; 1302 } 1303 err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev), 1304 MINOR(fc->dev), suffix); 1305 if (err) 1306 return err; 1307 1308 /* fuse does it's own writeback accounting */ 1309 sb->s_bdi->capabilities &= ~BDI_CAP_WRITEBACK_ACCT; 1310 sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT; 1311 1312 /* 1313 * For a single fuse filesystem use max 1% of dirty + 1314 * writeback threshold. 1315 * 1316 * This gives about 1M of write buffer for memory maps on a 1317 * machine with 1G and 10% dirty_ratio, which should be more 1318 * than enough. 1319 * 1320 * Privileged users can raise it by writing to 1321 * 1322 * /sys/class/bdi/<bdi>/max_ratio 1323 */ 1324 bdi_set_max_ratio(sb->s_bdi, 1); 1325 1326 return 0; 1327 } 1328 1329 struct fuse_dev *fuse_dev_alloc(void) 1330 { 1331 struct fuse_dev *fud; 1332 struct list_head *pq; 1333 1334 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL); 1335 if (!fud) 1336 return NULL; 1337 1338 pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL); 1339 if (!pq) { 1340 kfree(fud); 1341 return NULL; 1342 } 1343 1344 fud->pq.processing = pq; 1345 fuse_pqueue_init(&fud->pq); 1346 1347 return fud; 1348 } 1349 EXPORT_SYMBOL_GPL(fuse_dev_alloc); 1350 1351 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc) 1352 { 1353 fud->fc = fuse_conn_get(fc); 1354 spin_lock(&fc->lock); 1355 list_add_tail(&fud->entry, &fc->devices); 1356 spin_unlock(&fc->lock); 1357 } 1358 EXPORT_SYMBOL_GPL(fuse_dev_install); 1359 1360 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc) 1361 { 1362 struct fuse_dev *fud; 1363 1364 fud = fuse_dev_alloc(); 1365 if (!fud) 1366 return NULL; 1367 1368 fuse_dev_install(fud, fc); 1369 return fud; 1370 } 1371 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install); 1372 1373 void fuse_dev_free(struct fuse_dev *fud) 1374 { 1375 struct fuse_conn *fc = fud->fc; 1376 1377 if (fc) { 1378 spin_lock(&fc->lock); 1379 list_del(&fud->entry); 1380 spin_unlock(&fc->lock); 1381 1382 fuse_conn_put(fc); 1383 } 1384 kfree(fud->pq.processing); 1385 kfree(fud); 1386 } 1387 EXPORT_SYMBOL_GPL(fuse_dev_free); 1388 1389 static void fuse_fill_attr_from_inode(struct fuse_attr *attr, 1390 const struct fuse_inode *fi) 1391 { 1392 *attr = (struct fuse_attr){ 1393 .ino = fi->inode.i_ino, 1394 .size = fi->inode.i_size, 1395 .blocks = fi->inode.i_blocks, 1396 .atime = fi->inode.i_atime.tv_sec, 1397 .mtime = fi->inode.i_mtime.tv_sec, 1398 .ctime = fi->inode.i_ctime.tv_sec, 1399 .atimensec = fi->inode.i_atime.tv_nsec, 1400 .mtimensec = fi->inode.i_mtime.tv_nsec, 1401 .ctimensec = fi->inode.i_ctime.tv_nsec, 1402 .mode = fi->inode.i_mode, 1403 .nlink = fi->inode.i_nlink, 1404 .uid = fi->inode.i_uid.val, 1405 .gid = fi->inode.i_gid.val, 1406 .rdev = fi->inode.i_rdev, 1407 .blksize = 1u << fi->inode.i_blkbits, 1408 }; 1409 } 1410 1411 static void fuse_sb_defaults(struct super_block *sb) 1412 { 1413 sb->s_magic = FUSE_SUPER_MAGIC; 1414 sb->s_op = &fuse_super_operations; 1415 sb->s_xattr = fuse_xattr_handlers; 1416 sb->s_maxbytes = MAX_LFS_FILESIZE; 1417 sb->s_time_gran = 1; 1418 sb->s_export_op = &fuse_export_operations; 1419 sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE; 1420 if (sb->s_user_ns != &init_user_ns) 1421 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER; 1422 sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION); 1423 1424 /* 1425 * If we are not in the initial user namespace posix 1426 * acls must be translated. 1427 */ 1428 if (sb->s_user_ns != &init_user_ns) 1429 sb->s_xattr = fuse_no_acl_xattr_handlers; 1430 } 1431 1432 static int fuse_fill_super_submount(struct super_block *sb, 1433 struct fuse_inode *parent_fi) 1434 { 1435 struct fuse_mount *fm = get_fuse_mount_super(sb); 1436 struct super_block *parent_sb = parent_fi->inode.i_sb; 1437 struct fuse_attr root_attr; 1438 struct inode *root; 1439 1440 fuse_sb_defaults(sb); 1441 fm->sb = sb; 1442 1443 WARN_ON(sb->s_bdi != &noop_backing_dev_info); 1444 sb->s_bdi = bdi_get(parent_sb->s_bdi); 1445 1446 sb->s_xattr = parent_sb->s_xattr; 1447 sb->s_time_gran = parent_sb->s_time_gran; 1448 sb->s_blocksize = parent_sb->s_blocksize; 1449 sb->s_blocksize_bits = parent_sb->s_blocksize_bits; 1450 sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL); 1451 if (parent_sb->s_subtype && !sb->s_subtype) 1452 return -ENOMEM; 1453 1454 fuse_fill_attr_from_inode(&root_attr, parent_fi); 1455 root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0); 1456 /* 1457 * This inode is just a duplicate, so it is not looked up and 1458 * its nlookup should not be incremented. fuse_iget() does 1459 * that, though, so undo it here. 1460 */ 1461 get_fuse_inode(root)->nlookup--; 1462 sb->s_d_op = &fuse_dentry_operations; 1463 sb->s_root = d_make_root(root); 1464 if (!sb->s_root) 1465 return -ENOMEM; 1466 1467 return 0; 1468 } 1469 1470 /* Filesystem context private data holds the FUSE inode of the mount point */ 1471 static int fuse_get_tree_submount(struct fs_context *fsc) 1472 { 1473 struct fuse_mount *fm; 1474 struct fuse_inode *mp_fi = fsc->fs_private; 1475 struct fuse_conn *fc = get_fuse_conn(&mp_fi->inode); 1476 struct super_block *sb; 1477 int err; 1478 1479 fm = kzalloc(sizeof(struct fuse_mount), GFP_KERNEL); 1480 if (!fm) 1481 return -ENOMEM; 1482 1483 fm->fc = fuse_conn_get(fc); 1484 fsc->s_fs_info = fm; 1485 sb = sget_fc(fsc, NULL, set_anon_super_fc); 1486 if (fsc->s_fs_info) 1487 fuse_mount_destroy(fm); 1488 if (IS_ERR(sb)) 1489 return PTR_ERR(sb); 1490 1491 /* Initialize superblock, making @mp_fi its root */ 1492 err = fuse_fill_super_submount(sb, mp_fi); 1493 if (err) { 1494 deactivate_locked_super(sb); 1495 return err; 1496 } 1497 1498 down_write(&fc->killsb); 1499 list_add_tail(&fm->fc_entry, &fc->mounts); 1500 up_write(&fc->killsb); 1501 1502 sb->s_flags |= SB_ACTIVE; 1503 fsc->root = dget(sb->s_root); 1504 1505 return 0; 1506 } 1507 1508 static const struct fs_context_operations fuse_context_submount_ops = { 1509 .get_tree = fuse_get_tree_submount, 1510 }; 1511 1512 int fuse_init_fs_context_submount(struct fs_context *fsc) 1513 { 1514 fsc->ops = &fuse_context_submount_ops; 1515 return 0; 1516 } 1517 EXPORT_SYMBOL_GPL(fuse_init_fs_context_submount); 1518 1519 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx) 1520 { 1521 struct fuse_dev *fud = NULL; 1522 struct fuse_mount *fm = get_fuse_mount_super(sb); 1523 struct fuse_conn *fc = fm->fc; 1524 struct inode *root; 1525 struct dentry *root_dentry; 1526 int err; 1527 1528 err = -EINVAL; 1529 if (sb->s_flags & SB_MANDLOCK) 1530 goto err; 1531 1532 rcu_assign_pointer(fc->curr_bucket, fuse_sync_bucket_alloc()); 1533 fuse_sb_defaults(sb); 1534 1535 if (ctx->is_bdev) { 1536 #ifdef CONFIG_BLOCK 1537 err = -EINVAL; 1538 if (!sb_set_blocksize(sb, ctx->blksize)) 1539 goto err; 1540 #endif 1541 } else { 1542 sb->s_blocksize = PAGE_SIZE; 1543 sb->s_blocksize_bits = PAGE_SHIFT; 1544 } 1545 1546 sb->s_subtype = ctx->subtype; 1547 ctx->subtype = NULL; 1548 if (IS_ENABLED(CONFIG_FUSE_DAX)) { 1549 err = fuse_dax_conn_alloc(fc, ctx->dax_mode, ctx->dax_dev); 1550 if (err) 1551 goto err; 1552 } 1553 1554 if (ctx->fudptr) { 1555 err = -ENOMEM; 1556 fud = fuse_dev_alloc_install(fc); 1557 if (!fud) 1558 goto err_free_dax; 1559 } 1560 1561 fc->dev = sb->s_dev; 1562 fm->sb = sb; 1563 err = fuse_bdi_init(fc, sb); 1564 if (err) 1565 goto err_dev_free; 1566 1567 /* Handle umasking inside the fuse code */ 1568 if (sb->s_flags & SB_POSIXACL) 1569 fc->dont_mask = 1; 1570 sb->s_flags |= SB_POSIXACL; 1571 1572 fc->default_permissions = ctx->default_permissions; 1573 fc->allow_other = ctx->allow_other; 1574 fc->user_id = ctx->user_id; 1575 fc->group_id = ctx->group_id; 1576 fc->legacy_opts_show = ctx->legacy_opts_show; 1577 fc->max_read = max_t(unsigned int, 4096, ctx->max_read); 1578 fc->destroy = ctx->destroy; 1579 fc->no_control = ctx->no_control; 1580 fc->no_force_umount = ctx->no_force_umount; 1581 1582 err = -ENOMEM; 1583 root = fuse_get_root_inode(sb, ctx->rootmode); 1584 sb->s_d_op = &fuse_root_dentry_operations; 1585 root_dentry = d_make_root(root); 1586 if (!root_dentry) 1587 goto err_dev_free; 1588 /* Root dentry doesn't have .d_revalidate */ 1589 sb->s_d_op = &fuse_dentry_operations; 1590 1591 mutex_lock(&fuse_mutex); 1592 err = -EINVAL; 1593 if (ctx->fudptr && *ctx->fudptr) 1594 goto err_unlock; 1595 1596 err = fuse_ctl_add_conn(fc); 1597 if (err) 1598 goto err_unlock; 1599 1600 list_add_tail(&fc->entry, &fuse_conn_list); 1601 sb->s_root = root_dentry; 1602 if (ctx->fudptr) 1603 *ctx->fudptr = fud; 1604 mutex_unlock(&fuse_mutex); 1605 return 0; 1606 1607 err_unlock: 1608 mutex_unlock(&fuse_mutex); 1609 dput(root_dentry); 1610 err_dev_free: 1611 if (fud) 1612 fuse_dev_free(fud); 1613 err_free_dax: 1614 if (IS_ENABLED(CONFIG_FUSE_DAX)) 1615 fuse_dax_conn_free(fc); 1616 err: 1617 return err; 1618 } 1619 EXPORT_SYMBOL_GPL(fuse_fill_super_common); 1620 1621 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc) 1622 { 1623 struct fuse_fs_context *ctx = fsc->fs_private; 1624 int err; 1625 1626 if (!ctx->file || !ctx->rootmode_present || 1627 !ctx->user_id_present || !ctx->group_id_present) 1628 return -EINVAL; 1629 1630 /* 1631 * Require mount to happen from the same user namespace which 1632 * opened /dev/fuse to prevent potential attacks. 1633 */ 1634 if ((ctx->file->f_op != &fuse_dev_operations) || 1635 (ctx->file->f_cred->user_ns != sb->s_user_ns)) 1636 return -EINVAL; 1637 ctx->fudptr = &ctx->file->private_data; 1638 1639 err = fuse_fill_super_common(sb, ctx); 1640 if (err) 1641 return err; 1642 /* file->private_data shall be visible on all CPUs after this */ 1643 smp_mb(); 1644 fuse_send_init(get_fuse_mount_super(sb)); 1645 return 0; 1646 } 1647 1648 /* 1649 * This is the path where user supplied an already initialized fuse dev. In 1650 * this case never create a new super if the old one is gone. 1651 */ 1652 static int fuse_set_no_super(struct super_block *sb, struct fs_context *fsc) 1653 { 1654 return -ENOTCONN; 1655 } 1656 1657 static int fuse_test_super(struct super_block *sb, struct fs_context *fsc) 1658 { 1659 1660 return fsc->sget_key == get_fuse_conn_super(sb); 1661 } 1662 1663 static int fuse_get_tree(struct fs_context *fsc) 1664 { 1665 struct fuse_fs_context *ctx = fsc->fs_private; 1666 struct fuse_dev *fud; 1667 struct fuse_conn *fc; 1668 struct fuse_mount *fm; 1669 struct super_block *sb; 1670 int err; 1671 1672 fc = kmalloc(sizeof(*fc), GFP_KERNEL); 1673 if (!fc) 1674 return -ENOMEM; 1675 1676 fm = kzalloc(sizeof(*fm), GFP_KERNEL); 1677 if (!fm) { 1678 kfree(fc); 1679 return -ENOMEM; 1680 } 1681 1682 fuse_conn_init(fc, fm, fsc->user_ns, &fuse_dev_fiq_ops, NULL); 1683 fc->release = fuse_free_conn; 1684 1685 fsc->s_fs_info = fm; 1686 1687 if (ctx->fd_present) 1688 ctx->file = fget(ctx->fd); 1689 1690 if (IS_ENABLED(CONFIG_BLOCK) && ctx->is_bdev) { 1691 err = get_tree_bdev(fsc, fuse_fill_super); 1692 goto out; 1693 } 1694 /* 1695 * While block dev mount can be initialized with a dummy device fd 1696 * (found by device name), normal fuse mounts can't 1697 */ 1698 err = -EINVAL; 1699 if (!ctx->file) 1700 goto out; 1701 1702 /* 1703 * Allow creating a fuse mount with an already initialized fuse 1704 * connection 1705 */ 1706 fud = READ_ONCE(ctx->file->private_data); 1707 if (ctx->file->f_op == &fuse_dev_operations && fud) { 1708 fsc->sget_key = fud->fc; 1709 sb = sget_fc(fsc, fuse_test_super, fuse_set_no_super); 1710 err = PTR_ERR_OR_ZERO(sb); 1711 if (!IS_ERR(sb)) 1712 fsc->root = dget(sb->s_root); 1713 } else { 1714 err = get_tree_nodev(fsc, fuse_fill_super); 1715 } 1716 out: 1717 if (fsc->s_fs_info) 1718 fuse_mount_destroy(fm); 1719 if (ctx->file) 1720 fput(ctx->file); 1721 return err; 1722 } 1723 1724 static const struct fs_context_operations fuse_context_ops = { 1725 .free = fuse_free_fsc, 1726 .parse_param = fuse_parse_param, 1727 .reconfigure = fuse_reconfigure, 1728 .get_tree = fuse_get_tree, 1729 }; 1730 1731 /* 1732 * Set up the filesystem mount context. 1733 */ 1734 static int fuse_init_fs_context(struct fs_context *fsc) 1735 { 1736 struct fuse_fs_context *ctx; 1737 1738 ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL); 1739 if (!ctx) 1740 return -ENOMEM; 1741 1742 ctx->max_read = ~0; 1743 ctx->blksize = FUSE_DEFAULT_BLKSIZE; 1744 ctx->legacy_opts_show = true; 1745 1746 #ifdef CONFIG_BLOCK 1747 if (fsc->fs_type == &fuseblk_fs_type) { 1748 ctx->is_bdev = true; 1749 ctx->destroy = true; 1750 } 1751 #endif 1752 1753 fsc->fs_private = ctx; 1754 fsc->ops = &fuse_context_ops; 1755 return 0; 1756 } 1757 1758 bool fuse_mount_remove(struct fuse_mount *fm) 1759 { 1760 struct fuse_conn *fc = fm->fc; 1761 bool last = false; 1762 1763 down_write(&fc->killsb); 1764 list_del_init(&fm->fc_entry); 1765 if (list_empty(&fc->mounts)) 1766 last = true; 1767 up_write(&fc->killsb); 1768 1769 return last; 1770 } 1771 EXPORT_SYMBOL_GPL(fuse_mount_remove); 1772 1773 void fuse_conn_destroy(struct fuse_mount *fm) 1774 { 1775 struct fuse_conn *fc = fm->fc; 1776 1777 if (fc->destroy) 1778 fuse_send_destroy(fm); 1779 1780 fuse_abort_conn(fc); 1781 fuse_wait_aborted(fc); 1782 1783 if (!list_empty(&fc->entry)) { 1784 mutex_lock(&fuse_mutex); 1785 list_del(&fc->entry); 1786 fuse_ctl_remove_conn(fc); 1787 mutex_unlock(&fuse_mutex); 1788 } 1789 } 1790 EXPORT_SYMBOL_GPL(fuse_conn_destroy); 1791 1792 static void fuse_sb_destroy(struct super_block *sb) 1793 { 1794 struct fuse_mount *fm = get_fuse_mount_super(sb); 1795 bool last; 1796 1797 if (sb->s_root) { 1798 last = fuse_mount_remove(fm); 1799 if (last) 1800 fuse_conn_destroy(fm); 1801 } 1802 } 1803 1804 void fuse_mount_destroy(struct fuse_mount *fm) 1805 { 1806 fuse_conn_put(fm->fc); 1807 kfree(fm); 1808 } 1809 EXPORT_SYMBOL(fuse_mount_destroy); 1810 1811 static void fuse_kill_sb_anon(struct super_block *sb) 1812 { 1813 fuse_sb_destroy(sb); 1814 kill_anon_super(sb); 1815 fuse_mount_destroy(get_fuse_mount_super(sb)); 1816 } 1817 1818 static struct file_system_type fuse_fs_type = { 1819 .owner = THIS_MODULE, 1820 .name = "fuse", 1821 .fs_flags = FS_HAS_SUBTYPE | FS_USERNS_MOUNT, 1822 .init_fs_context = fuse_init_fs_context, 1823 .parameters = fuse_fs_parameters, 1824 .kill_sb = fuse_kill_sb_anon, 1825 }; 1826 MODULE_ALIAS_FS("fuse"); 1827 1828 #ifdef CONFIG_BLOCK 1829 static void fuse_kill_sb_blk(struct super_block *sb) 1830 { 1831 fuse_sb_destroy(sb); 1832 kill_block_super(sb); 1833 fuse_mount_destroy(get_fuse_mount_super(sb)); 1834 } 1835 1836 static struct file_system_type fuseblk_fs_type = { 1837 .owner = THIS_MODULE, 1838 .name = "fuseblk", 1839 .init_fs_context = fuse_init_fs_context, 1840 .parameters = fuse_fs_parameters, 1841 .kill_sb = fuse_kill_sb_blk, 1842 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE, 1843 }; 1844 MODULE_ALIAS_FS("fuseblk"); 1845 1846 static inline int register_fuseblk(void) 1847 { 1848 return register_filesystem(&fuseblk_fs_type); 1849 } 1850 1851 static inline void unregister_fuseblk(void) 1852 { 1853 unregister_filesystem(&fuseblk_fs_type); 1854 } 1855 #else 1856 static inline int register_fuseblk(void) 1857 { 1858 return 0; 1859 } 1860 1861 static inline void unregister_fuseblk(void) 1862 { 1863 } 1864 #endif 1865 1866 static void fuse_inode_init_once(void *foo) 1867 { 1868 struct inode *inode = foo; 1869 1870 inode_init_once(inode); 1871 } 1872 1873 static int __init fuse_fs_init(void) 1874 { 1875 int err; 1876 1877 fuse_inode_cachep = kmem_cache_create("fuse_inode", 1878 sizeof(struct fuse_inode), 0, 1879 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT, 1880 fuse_inode_init_once); 1881 err = -ENOMEM; 1882 if (!fuse_inode_cachep) 1883 goto out; 1884 1885 err = register_fuseblk(); 1886 if (err) 1887 goto out2; 1888 1889 err = register_filesystem(&fuse_fs_type); 1890 if (err) 1891 goto out3; 1892 1893 return 0; 1894 1895 out3: 1896 unregister_fuseblk(); 1897 out2: 1898 kmem_cache_destroy(fuse_inode_cachep); 1899 out: 1900 return err; 1901 } 1902 1903 static void fuse_fs_cleanup(void) 1904 { 1905 unregister_filesystem(&fuse_fs_type); 1906 unregister_fuseblk(); 1907 1908 /* 1909 * Make sure all delayed rcu free inodes are flushed before we 1910 * destroy cache. 1911 */ 1912 rcu_barrier(); 1913 kmem_cache_destroy(fuse_inode_cachep); 1914 } 1915 1916 static struct kobject *fuse_kobj; 1917 1918 static int fuse_sysfs_init(void) 1919 { 1920 int err; 1921 1922 fuse_kobj = kobject_create_and_add("fuse", fs_kobj); 1923 if (!fuse_kobj) { 1924 err = -ENOMEM; 1925 goto out_err; 1926 } 1927 1928 err = sysfs_create_mount_point(fuse_kobj, "connections"); 1929 if (err) 1930 goto out_fuse_unregister; 1931 1932 return 0; 1933 1934 out_fuse_unregister: 1935 kobject_put(fuse_kobj); 1936 out_err: 1937 return err; 1938 } 1939 1940 static void fuse_sysfs_cleanup(void) 1941 { 1942 sysfs_remove_mount_point(fuse_kobj, "connections"); 1943 kobject_put(fuse_kobj); 1944 } 1945 1946 static int __init fuse_init(void) 1947 { 1948 int res; 1949 1950 pr_info("init (API version %i.%i)\n", 1951 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION); 1952 1953 INIT_LIST_HEAD(&fuse_conn_list); 1954 res = fuse_fs_init(); 1955 if (res) 1956 goto err; 1957 1958 res = fuse_dev_init(); 1959 if (res) 1960 goto err_fs_cleanup; 1961 1962 res = fuse_sysfs_init(); 1963 if (res) 1964 goto err_dev_cleanup; 1965 1966 res = fuse_ctl_init(); 1967 if (res) 1968 goto err_sysfs_cleanup; 1969 1970 sanitize_global_limit(&max_user_bgreq); 1971 sanitize_global_limit(&max_user_congthresh); 1972 1973 return 0; 1974 1975 err_sysfs_cleanup: 1976 fuse_sysfs_cleanup(); 1977 err_dev_cleanup: 1978 fuse_dev_cleanup(); 1979 err_fs_cleanup: 1980 fuse_fs_cleanup(); 1981 err: 1982 return res; 1983 } 1984 1985 static void __exit fuse_exit(void) 1986 { 1987 pr_debug("exit\n"); 1988 1989 fuse_ctl_cleanup(); 1990 fuse_sysfs_cleanup(); 1991 fuse_fs_cleanup(); 1992 fuse_dev_cleanup(); 1993 } 1994 1995 module_init(fuse_init); 1996 module_exit(fuse_exit); 1997