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