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 #include "dev_uring_i.h" 11 12 #include <linux/dax.h> 13 #include <linux/pagemap.h> 14 #include <linux/slab.h> 15 #include <linux/file.h> 16 #include <linux/seq_file.h> 17 #include <linux/init.h> 18 #include <linux/module.h> 19 #include <linux/moduleparam.h> 20 #include <linux/fs_context.h> 21 #include <linux/fs_parser.h> 22 #include <linux/statfs.h> 23 #include <linux/random.h> 24 #include <linux/sched.h> 25 #include <linux/exportfs.h> 26 #include <linux/posix_acl.h> 27 #include <linux/pid_namespace.h> 28 #include <uapi/linux/magic.h> 29 30 MODULE_AUTHOR("Miklos Szeredi <miklos@szeredi.hu>"); 31 MODULE_DESCRIPTION("Filesystem in Userspace"); 32 MODULE_LICENSE("GPL"); 33 34 static struct kmem_cache *fuse_inode_cachep; 35 struct list_head fuse_conn_list; 36 DEFINE_MUTEX(fuse_mutex); 37 38 static int set_global_limit(const char *val, const struct kernel_param *kp); 39 40 unsigned int fuse_max_pages_limit = 256; 41 /* default is no timeout */ 42 unsigned int fuse_default_req_timeout; 43 unsigned int fuse_max_req_timeout; 44 45 unsigned int max_user_bgreq; 46 module_param_call(max_user_bgreq, set_global_limit, param_get_uint, 47 &max_user_bgreq, 0644); 48 __MODULE_PARM_TYPE(max_user_bgreq, "uint"); 49 MODULE_PARM_DESC(max_user_bgreq, 50 "Global limit for the maximum number of backgrounded requests an " 51 "unprivileged user can set"); 52 53 unsigned int max_user_congthresh; 54 module_param_call(max_user_congthresh, set_global_limit, param_get_uint, 55 &max_user_congthresh, 0644); 56 __MODULE_PARM_TYPE(max_user_congthresh, "uint"); 57 MODULE_PARM_DESC(max_user_congthresh, 58 "Global limit for the maximum congestion threshold an " 59 "unprivileged user can set"); 60 61 #define FUSE_DEFAULT_BLKSIZE 512 62 63 /** Maximum number of outstanding background requests */ 64 #define FUSE_DEFAULT_MAX_BACKGROUND 12 65 66 /** Congestion starts at 75% of maximum */ 67 #define FUSE_DEFAULT_CONGESTION_THRESHOLD (FUSE_DEFAULT_MAX_BACKGROUND * 3 / 4) 68 69 #ifdef CONFIG_BLOCK 70 static struct file_system_type fuseblk_fs_type; 71 #endif 72 73 struct fuse_forget_link *fuse_alloc_forget(void) 74 { 75 return kzalloc(sizeof(struct fuse_forget_link), GFP_KERNEL_ACCOUNT); 76 } 77 78 static struct fuse_submount_lookup *fuse_alloc_submount_lookup(void) 79 { 80 struct fuse_submount_lookup *sl; 81 82 sl = kzalloc(sizeof(struct fuse_submount_lookup), GFP_KERNEL_ACCOUNT); 83 if (!sl) 84 return NULL; 85 sl->forget = fuse_alloc_forget(); 86 if (!sl->forget) 87 goto out_free; 88 89 return sl; 90 91 out_free: 92 kfree(sl); 93 return NULL; 94 } 95 96 static struct inode *fuse_alloc_inode(struct super_block *sb) 97 { 98 struct fuse_inode *fi; 99 100 fi = alloc_inode_sb(sb, fuse_inode_cachep, GFP_KERNEL); 101 if (!fi) 102 return NULL; 103 104 fi->i_time = 0; 105 fi->inval_mask = ~0; 106 fi->nodeid = 0; 107 fi->nlookup = 0; 108 fi->attr_version = 0; 109 fi->orig_ino = 0; 110 fi->state = 0; 111 fi->submount_lookup = NULL; 112 mutex_init(&fi->mutex); 113 spin_lock_init(&fi->lock); 114 fi->forget = fuse_alloc_forget(); 115 if (!fi->forget) 116 goto out_free; 117 118 if (IS_ENABLED(CONFIG_FUSE_DAX) && !fuse_dax_inode_alloc(sb, fi)) 119 goto out_free_forget; 120 121 if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH)) 122 fuse_inode_backing_set(fi, NULL); 123 124 return &fi->inode; 125 126 out_free_forget: 127 kfree(fi->forget); 128 out_free: 129 kmem_cache_free(fuse_inode_cachep, fi); 130 return NULL; 131 } 132 133 static void fuse_free_inode(struct inode *inode) 134 { 135 struct fuse_inode *fi = get_fuse_inode(inode); 136 137 mutex_destroy(&fi->mutex); 138 kfree(fi->forget); 139 #ifdef CONFIG_FUSE_DAX 140 kfree(fi->dax); 141 #endif 142 if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH)) 143 fuse_backing_put(fuse_inode_backing(fi)); 144 145 kmem_cache_free(fuse_inode_cachep, fi); 146 } 147 148 static void fuse_cleanup_submount_lookup(struct fuse_conn *fc, 149 struct fuse_submount_lookup *sl) 150 { 151 if (!refcount_dec_and_test(&sl->count)) 152 return; 153 154 fuse_queue_forget(fc, sl->forget, sl->nodeid, 1); 155 sl->forget = NULL; 156 kfree(sl); 157 } 158 159 static void fuse_evict_inode(struct inode *inode) 160 { 161 struct fuse_inode *fi = get_fuse_inode(inode); 162 163 /* Will write inode on close/munmap and in all other dirtiers */ 164 WARN_ON(inode->i_state & I_DIRTY_INODE); 165 166 if (FUSE_IS_DAX(inode)) 167 dax_break_layout_final(inode); 168 169 truncate_inode_pages_final(&inode->i_data); 170 clear_inode(inode); 171 if (inode->i_sb->s_flags & SB_ACTIVE) { 172 struct fuse_conn *fc = get_fuse_conn(inode); 173 174 if (FUSE_IS_DAX(inode)) 175 fuse_dax_inode_cleanup(inode); 176 if (fi->nlookup) { 177 fuse_queue_forget(fc, fi->forget, fi->nodeid, 178 fi->nlookup); 179 fi->forget = NULL; 180 } 181 182 if (fi->submount_lookup) { 183 fuse_cleanup_submount_lookup(fc, fi->submount_lookup); 184 fi->submount_lookup = NULL; 185 } 186 /* 187 * Evict of non-deleted inode may race with outstanding 188 * LOOKUP/READDIRPLUS requests and result in inconsistency when 189 * the request finishes. Deal with that here by bumping a 190 * counter that can be compared to the starting value. 191 */ 192 if (inode->i_nlink > 0) 193 atomic64_inc(&fc->evict_ctr); 194 } 195 if (S_ISREG(inode->i_mode) && !fuse_is_bad(inode)) { 196 WARN_ON(fi->iocachectr != 0); 197 WARN_ON(!list_empty(&fi->write_files)); 198 WARN_ON(!list_empty(&fi->queued_writes)); 199 } 200 } 201 202 static int fuse_reconfigure(struct fs_context *fsc) 203 { 204 struct super_block *sb = fsc->root->d_sb; 205 206 sync_filesystem(sb); 207 if (fsc->sb_flags & SB_MANDLOCK) 208 return -EINVAL; 209 210 return 0; 211 } 212 213 /* 214 * ino_t is 32-bits on 32-bit arch. We have to squash the 64-bit value down 215 * so that it will fit. 216 */ 217 static ino_t fuse_squash_ino(u64 ino64) 218 { 219 ino_t ino = (ino_t) ino64; 220 if (sizeof(ino_t) < sizeof(u64)) 221 ino ^= ino64 >> (sizeof(u64) - sizeof(ino_t)) * 8; 222 return ino; 223 } 224 225 void fuse_change_attributes_common(struct inode *inode, struct fuse_attr *attr, 226 struct fuse_statx *sx, 227 u64 attr_valid, u32 cache_mask, 228 u64 evict_ctr) 229 { 230 struct fuse_conn *fc = get_fuse_conn(inode); 231 struct fuse_inode *fi = get_fuse_inode(inode); 232 233 lockdep_assert_held(&fi->lock); 234 235 /* 236 * Clear basic stats from invalid mask. 237 * 238 * Don't do this if this is coming from a fuse_iget() call and there 239 * might have been a racing evict which would've invalidated the result 240 * if the attr_version would've been preserved. 241 * 242 * !evict_ctr -> this is create 243 * fi->attr_version != 0 -> this is not a new inode 244 * evict_ctr == fuse_get_evict_ctr() -> no evicts while during request 245 */ 246 if (!evict_ctr || fi->attr_version || evict_ctr == fuse_get_evict_ctr(fc)) 247 set_mask_bits(&fi->inval_mask, STATX_BASIC_STATS, 0); 248 249 fi->attr_version = atomic64_inc_return(&fc->attr_version); 250 fi->i_time = attr_valid; 251 252 inode->i_ino = fuse_squash_ino(attr->ino); 253 inode->i_mode = (inode->i_mode & S_IFMT) | (attr->mode & 07777); 254 set_nlink(inode, attr->nlink); 255 inode->i_uid = make_kuid(fc->user_ns, attr->uid); 256 inode->i_gid = make_kgid(fc->user_ns, attr->gid); 257 inode->i_blocks = attr->blocks; 258 259 /* Sanitize nsecs */ 260 attr->atimensec = min_t(u32, attr->atimensec, NSEC_PER_SEC - 1); 261 attr->mtimensec = min_t(u32, attr->mtimensec, NSEC_PER_SEC - 1); 262 attr->ctimensec = min_t(u32, attr->ctimensec, NSEC_PER_SEC - 1); 263 264 inode_set_atime(inode, attr->atime, attr->atimensec); 265 /* mtime from server may be stale due to local buffered write */ 266 if (!(cache_mask & STATX_MTIME)) { 267 inode_set_mtime(inode, attr->mtime, attr->mtimensec); 268 } 269 if (!(cache_mask & STATX_CTIME)) { 270 inode_set_ctime(inode, attr->ctime, attr->ctimensec); 271 } 272 if (sx) { 273 /* Sanitize nsecs */ 274 sx->btime.tv_nsec = 275 min_t(u32, sx->btime.tv_nsec, NSEC_PER_SEC - 1); 276 277 /* 278 * Btime has been queried, cache is valid (whether or not btime 279 * is available or not) so clear STATX_BTIME from inval_mask. 280 * 281 * Availability of the btime attribute is indicated in 282 * FUSE_I_BTIME 283 */ 284 set_mask_bits(&fi->inval_mask, STATX_BTIME, 0); 285 if (sx->mask & STATX_BTIME) { 286 set_bit(FUSE_I_BTIME, &fi->state); 287 fi->i_btime.tv_sec = sx->btime.tv_sec; 288 fi->i_btime.tv_nsec = sx->btime.tv_nsec; 289 } 290 } 291 292 if (attr->blksize != 0) 293 inode->i_blkbits = ilog2(attr->blksize); 294 else 295 inode->i_blkbits = inode->i_sb->s_blocksize_bits; 296 297 /* 298 * Don't set the sticky bit in i_mode, unless we want the VFS 299 * to check permissions. This prevents failures due to the 300 * check in may_delete(). 301 */ 302 fi->orig_i_mode = inode->i_mode; 303 if (!fc->default_permissions) 304 inode->i_mode &= ~S_ISVTX; 305 306 fi->orig_ino = attr->ino; 307 308 /* 309 * We are refreshing inode data and it is possible that another 310 * client set suid/sgid or security.capability xattr. So clear 311 * S_NOSEC. Ideally, we could have cleared it only if suid/sgid 312 * was set or if security.capability xattr was set. But we don't 313 * know if security.capability has been set or not. So clear it 314 * anyway. Its less efficient but should be safe. 315 */ 316 inode->i_flags &= ~S_NOSEC; 317 } 318 319 u32 fuse_get_cache_mask(struct inode *inode) 320 { 321 struct fuse_conn *fc = get_fuse_conn(inode); 322 323 if (!fc->writeback_cache || !S_ISREG(inode->i_mode)) 324 return 0; 325 326 return STATX_MTIME | STATX_CTIME | STATX_SIZE; 327 } 328 329 static void fuse_change_attributes_i(struct inode *inode, struct fuse_attr *attr, 330 struct fuse_statx *sx, u64 attr_valid, 331 u64 attr_version, u64 evict_ctr) 332 { 333 struct fuse_conn *fc = get_fuse_conn(inode); 334 struct fuse_inode *fi = get_fuse_inode(inode); 335 u32 cache_mask; 336 loff_t oldsize; 337 struct timespec64 old_mtime; 338 339 spin_lock(&fi->lock); 340 /* 341 * In case of writeback_cache enabled, writes update mtime, ctime and 342 * may update i_size. In these cases trust the cached value in the 343 * inode. 344 */ 345 cache_mask = fuse_get_cache_mask(inode); 346 if (cache_mask & STATX_SIZE) 347 attr->size = i_size_read(inode); 348 349 if (cache_mask & STATX_MTIME) { 350 attr->mtime = inode_get_mtime_sec(inode); 351 attr->mtimensec = inode_get_mtime_nsec(inode); 352 } 353 if (cache_mask & STATX_CTIME) { 354 attr->ctime = inode_get_ctime_sec(inode); 355 attr->ctimensec = inode_get_ctime_nsec(inode); 356 } 357 358 if ((attr_version != 0 && fi->attr_version > attr_version) || 359 test_bit(FUSE_I_SIZE_UNSTABLE, &fi->state)) { 360 spin_unlock(&fi->lock); 361 return; 362 } 363 364 old_mtime = inode_get_mtime(inode); 365 fuse_change_attributes_common(inode, attr, sx, attr_valid, cache_mask, 366 evict_ctr); 367 368 oldsize = inode->i_size; 369 /* 370 * In case of writeback_cache enabled, the cached writes beyond EOF 371 * extend local i_size without keeping userspace server in sync. So, 372 * attr->size coming from server can be stale. We cannot trust it. 373 */ 374 if (!(cache_mask & STATX_SIZE)) 375 i_size_write(inode, attr->size); 376 spin_unlock(&fi->lock); 377 378 if (!cache_mask && S_ISREG(inode->i_mode)) { 379 bool inval = false; 380 381 if (oldsize != attr->size) { 382 truncate_pagecache(inode, attr->size); 383 if (!fc->explicit_inval_data) 384 inval = true; 385 } else if (fc->auto_inval_data) { 386 struct timespec64 new_mtime = { 387 .tv_sec = attr->mtime, 388 .tv_nsec = attr->mtimensec, 389 }; 390 391 /* 392 * Auto inval mode also checks and invalidates if mtime 393 * has changed. 394 */ 395 if (!timespec64_equal(&old_mtime, &new_mtime)) 396 inval = true; 397 } 398 399 if (inval) 400 invalidate_inode_pages2(inode->i_mapping); 401 } 402 403 if (IS_ENABLED(CONFIG_FUSE_DAX)) 404 fuse_dax_dontcache(inode, attr->flags); 405 } 406 407 void fuse_change_attributes(struct inode *inode, struct fuse_attr *attr, 408 struct fuse_statx *sx, u64 attr_valid, 409 u64 attr_version) 410 { 411 fuse_change_attributes_i(inode, attr, sx, attr_valid, attr_version, 0); 412 } 413 414 static void fuse_init_submount_lookup(struct fuse_submount_lookup *sl, 415 u64 nodeid) 416 { 417 sl->nodeid = nodeid; 418 refcount_set(&sl->count, 1); 419 } 420 421 static void fuse_init_inode(struct inode *inode, struct fuse_attr *attr, 422 struct fuse_conn *fc) 423 { 424 inode->i_mode = attr->mode & S_IFMT; 425 inode->i_size = attr->size; 426 inode_set_mtime(inode, attr->mtime, attr->mtimensec); 427 inode_set_ctime(inode, attr->ctime, attr->ctimensec); 428 if (S_ISREG(inode->i_mode)) { 429 fuse_init_common(inode); 430 fuse_init_file_inode(inode, attr->flags); 431 } else if (S_ISDIR(inode->i_mode)) 432 fuse_init_dir(inode); 433 else if (S_ISLNK(inode->i_mode)) 434 fuse_init_symlink(inode); 435 else if (S_ISCHR(inode->i_mode) || S_ISBLK(inode->i_mode) || 436 S_ISFIFO(inode->i_mode) || S_ISSOCK(inode->i_mode)) { 437 fuse_init_common(inode); 438 init_special_inode(inode, inode->i_mode, 439 new_decode_dev(attr->rdev)); 440 } else 441 BUG(); 442 /* 443 * Ensure that we don't cache acls for daemons without FUSE_POSIX_ACL 444 * so they see the exact same behavior as before. 445 */ 446 if (!fc->posix_acl) 447 inode->i_acl = inode->i_default_acl = ACL_DONT_CACHE; 448 } 449 450 static int fuse_inode_eq(struct inode *inode, void *_nodeidp) 451 { 452 u64 nodeid = *(u64 *) _nodeidp; 453 if (get_node_id(inode) == nodeid) 454 return 1; 455 else 456 return 0; 457 } 458 459 static int fuse_inode_set(struct inode *inode, void *_nodeidp) 460 { 461 u64 nodeid = *(u64 *) _nodeidp; 462 get_fuse_inode(inode)->nodeid = nodeid; 463 return 0; 464 } 465 466 struct inode *fuse_iget(struct super_block *sb, u64 nodeid, 467 int generation, struct fuse_attr *attr, 468 u64 attr_valid, u64 attr_version, 469 u64 evict_ctr) 470 { 471 struct inode *inode; 472 struct fuse_inode *fi; 473 struct fuse_conn *fc = get_fuse_conn_super(sb); 474 475 /* 476 * Auto mount points get their node id from the submount root, which is 477 * not a unique identifier within this filesystem. 478 * 479 * To avoid conflicts, do not place submount points into the inode hash 480 * table. 481 */ 482 if (fc->auto_submounts && (attr->flags & FUSE_ATTR_SUBMOUNT) && 483 S_ISDIR(attr->mode)) { 484 struct fuse_inode *fi; 485 486 inode = new_inode(sb); 487 if (!inode) 488 return NULL; 489 490 fuse_init_inode(inode, attr, fc); 491 fi = get_fuse_inode(inode); 492 fi->nodeid = nodeid; 493 fi->submount_lookup = fuse_alloc_submount_lookup(); 494 if (!fi->submount_lookup) { 495 iput(inode); 496 return NULL; 497 } 498 /* Sets nlookup = 1 on fi->submount_lookup->nlookup */ 499 fuse_init_submount_lookup(fi->submount_lookup, nodeid); 500 inode->i_flags |= S_AUTOMOUNT; 501 goto done; 502 } 503 504 retry: 505 inode = iget5_locked(sb, nodeid, fuse_inode_eq, fuse_inode_set, &nodeid); 506 if (!inode) 507 return NULL; 508 509 if ((inode->i_state & I_NEW)) { 510 inode->i_flags |= S_NOATIME; 511 if (!fc->writeback_cache || !S_ISREG(attr->mode)) 512 inode->i_flags |= S_NOCMTIME; 513 inode->i_generation = generation; 514 fuse_init_inode(inode, attr, fc); 515 unlock_new_inode(inode); 516 } else if (fuse_stale_inode(inode, generation, attr)) { 517 /* nodeid was reused, any I/O on the old inode should fail */ 518 fuse_make_bad(inode); 519 if (inode != d_inode(sb->s_root)) { 520 remove_inode_hash(inode); 521 iput(inode); 522 goto retry; 523 } 524 } 525 fi = get_fuse_inode(inode); 526 spin_lock(&fi->lock); 527 fi->nlookup++; 528 spin_unlock(&fi->lock); 529 done: 530 fuse_change_attributes_i(inode, attr, NULL, attr_valid, attr_version, 531 evict_ctr); 532 return inode; 533 } 534 535 struct inode *fuse_ilookup(struct fuse_conn *fc, u64 nodeid, 536 struct fuse_mount **fm) 537 { 538 struct fuse_mount *fm_iter; 539 struct inode *inode; 540 541 WARN_ON(!rwsem_is_locked(&fc->killsb)); 542 list_for_each_entry(fm_iter, &fc->mounts, fc_entry) { 543 if (!fm_iter->sb) 544 continue; 545 546 inode = ilookup5(fm_iter->sb, nodeid, fuse_inode_eq, &nodeid); 547 if (inode) { 548 if (fm) 549 *fm = fm_iter; 550 return inode; 551 } 552 } 553 554 return NULL; 555 } 556 557 int fuse_reverse_inval_inode(struct fuse_conn *fc, u64 nodeid, 558 loff_t offset, loff_t len) 559 { 560 struct fuse_inode *fi; 561 struct inode *inode; 562 pgoff_t pg_start; 563 pgoff_t pg_end; 564 565 inode = fuse_ilookup(fc, nodeid, NULL); 566 if (!inode) 567 return -ENOENT; 568 569 fi = get_fuse_inode(inode); 570 spin_lock(&fi->lock); 571 fi->attr_version = atomic64_inc_return(&fc->attr_version); 572 spin_unlock(&fi->lock); 573 574 fuse_invalidate_attr(inode); 575 forget_all_cached_acls(inode); 576 if (offset >= 0) { 577 pg_start = offset >> PAGE_SHIFT; 578 if (len <= 0) 579 pg_end = -1; 580 else 581 pg_end = (offset + len - 1) >> PAGE_SHIFT; 582 invalidate_inode_pages2_range(inode->i_mapping, 583 pg_start, pg_end); 584 } 585 iput(inode); 586 return 0; 587 } 588 589 bool fuse_lock_inode(struct inode *inode) 590 { 591 bool locked = false; 592 593 if (!get_fuse_conn(inode)->parallel_dirops) { 594 mutex_lock(&get_fuse_inode(inode)->mutex); 595 locked = true; 596 } 597 598 return locked; 599 } 600 601 void fuse_unlock_inode(struct inode *inode, bool locked) 602 { 603 if (locked) 604 mutex_unlock(&get_fuse_inode(inode)->mutex); 605 } 606 607 static void fuse_umount_begin(struct super_block *sb) 608 { 609 struct fuse_conn *fc = get_fuse_conn_super(sb); 610 611 if (fc->no_force_umount) 612 return; 613 614 fuse_abort_conn(fc); 615 616 // Only retire block-device-based superblocks. 617 if (sb->s_bdev != NULL) 618 retire_super(sb); 619 } 620 621 static void fuse_send_destroy(struct fuse_mount *fm) 622 { 623 if (fm->fc->conn_init) { 624 FUSE_ARGS(args); 625 626 args.opcode = FUSE_DESTROY; 627 args.force = true; 628 args.nocreds = true; 629 fuse_simple_request(fm, &args); 630 } 631 } 632 633 static void convert_fuse_statfs(struct kstatfs *stbuf, struct fuse_kstatfs *attr) 634 { 635 stbuf->f_type = FUSE_SUPER_MAGIC; 636 stbuf->f_bsize = attr->bsize; 637 stbuf->f_frsize = attr->frsize; 638 stbuf->f_blocks = attr->blocks; 639 stbuf->f_bfree = attr->bfree; 640 stbuf->f_bavail = attr->bavail; 641 stbuf->f_files = attr->files; 642 stbuf->f_ffree = attr->ffree; 643 stbuf->f_namelen = attr->namelen; 644 /* fsid is left zero */ 645 } 646 647 static int fuse_statfs(struct dentry *dentry, struct kstatfs *buf) 648 { 649 struct super_block *sb = dentry->d_sb; 650 struct fuse_mount *fm = get_fuse_mount_super(sb); 651 FUSE_ARGS(args); 652 struct fuse_statfs_out outarg; 653 int err; 654 655 if (!fuse_allow_current_process(fm->fc)) { 656 buf->f_type = FUSE_SUPER_MAGIC; 657 return 0; 658 } 659 660 memset(&outarg, 0, sizeof(outarg)); 661 args.in_numargs = 0; 662 args.opcode = FUSE_STATFS; 663 args.nodeid = get_node_id(d_inode(dentry)); 664 args.out_numargs = 1; 665 args.out_args[0].size = sizeof(outarg); 666 args.out_args[0].value = &outarg; 667 err = fuse_simple_request(fm, &args); 668 if (!err) 669 convert_fuse_statfs(buf, &outarg.st); 670 return err; 671 } 672 673 static struct fuse_sync_bucket *fuse_sync_bucket_alloc(void) 674 { 675 struct fuse_sync_bucket *bucket; 676 677 bucket = kzalloc(sizeof(*bucket), GFP_KERNEL | __GFP_NOFAIL); 678 if (bucket) { 679 init_waitqueue_head(&bucket->waitq); 680 /* Initial active count */ 681 atomic_set(&bucket->count, 1); 682 } 683 return bucket; 684 } 685 686 static void fuse_sync_fs_writes(struct fuse_conn *fc) 687 { 688 struct fuse_sync_bucket *bucket, *new_bucket; 689 int count; 690 691 new_bucket = fuse_sync_bucket_alloc(); 692 spin_lock(&fc->lock); 693 bucket = rcu_dereference_protected(fc->curr_bucket, 1); 694 count = atomic_read(&bucket->count); 695 WARN_ON(count < 1); 696 /* No outstanding writes? */ 697 if (count == 1) { 698 spin_unlock(&fc->lock); 699 kfree(new_bucket); 700 return; 701 } 702 703 /* 704 * Completion of new bucket depends on completion of this bucket, so add 705 * one more count. 706 */ 707 atomic_inc(&new_bucket->count); 708 rcu_assign_pointer(fc->curr_bucket, new_bucket); 709 spin_unlock(&fc->lock); 710 /* 711 * Drop initial active count. At this point if all writes in this and 712 * ancestor buckets complete, the count will go to zero and this task 713 * will be woken up. 714 */ 715 atomic_dec(&bucket->count); 716 717 wait_event(bucket->waitq, atomic_read(&bucket->count) == 0); 718 719 /* Drop temp count on descendant bucket */ 720 fuse_sync_bucket_dec(new_bucket); 721 kfree_rcu(bucket, rcu); 722 } 723 724 static int fuse_sync_fs(struct super_block *sb, int wait) 725 { 726 struct fuse_mount *fm = get_fuse_mount_super(sb); 727 struct fuse_conn *fc = fm->fc; 728 struct fuse_syncfs_in inarg; 729 FUSE_ARGS(args); 730 int err; 731 732 /* 733 * Userspace cannot handle the wait == 0 case. Avoid a 734 * gratuitous roundtrip. 735 */ 736 if (!wait) 737 return 0; 738 739 /* The filesystem is being unmounted. Nothing to do. */ 740 if (!sb->s_root) 741 return 0; 742 743 if (!fc->sync_fs) 744 return 0; 745 746 fuse_sync_fs_writes(fc); 747 748 memset(&inarg, 0, sizeof(inarg)); 749 args.in_numargs = 1; 750 args.in_args[0].size = sizeof(inarg); 751 args.in_args[0].value = &inarg; 752 args.opcode = FUSE_SYNCFS; 753 args.nodeid = get_node_id(sb->s_root->d_inode); 754 args.out_numargs = 0; 755 756 err = fuse_simple_request(fm, &args); 757 if (err == -ENOSYS) { 758 fc->sync_fs = 0; 759 err = 0; 760 } 761 762 return err; 763 } 764 765 enum { 766 OPT_SOURCE, 767 OPT_SUBTYPE, 768 OPT_FD, 769 OPT_ROOTMODE, 770 OPT_USER_ID, 771 OPT_GROUP_ID, 772 OPT_DEFAULT_PERMISSIONS, 773 OPT_ALLOW_OTHER, 774 OPT_MAX_READ, 775 OPT_BLKSIZE, 776 OPT_ERR 777 }; 778 779 static const struct fs_parameter_spec fuse_fs_parameters[] = { 780 fsparam_string ("source", OPT_SOURCE), 781 fsparam_u32 ("fd", OPT_FD), 782 fsparam_u32oct ("rootmode", OPT_ROOTMODE), 783 fsparam_uid ("user_id", OPT_USER_ID), 784 fsparam_gid ("group_id", OPT_GROUP_ID), 785 fsparam_flag ("default_permissions", OPT_DEFAULT_PERMISSIONS), 786 fsparam_flag ("allow_other", OPT_ALLOW_OTHER), 787 fsparam_u32 ("max_read", OPT_MAX_READ), 788 fsparam_u32 ("blksize", OPT_BLKSIZE), 789 fsparam_string ("subtype", OPT_SUBTYPE), 790 {} 791 }; 792 793 static int fuse_parse_param(struct fs_context *fsc, struct fs_parameter *param) 794 { 795 struct fs_parse_result result; 796 struct fuse_fs_context *ctx = fsc->fs_private; 797 int opt; 798 kuid_t kuid; 799 kgid_t kgid; 800 801 if (fsc->purpose == FS_CONTEXT_FOR_RECONFIGURE) { 802 /* 803 * Ignore options coming from mount(MS_REMOUNT) for backward 804 * compatibility. 805 */ 806 if (fsc->oldapi) 807 return 0; 808 809 return invalfc(fsc, "No changes allowed in reconfigure"); 810 } 811 812 opt = fs_parse(fsc, fuse_fs_parameters, param, &result); 813 if (opt < 0) 814 return opt; 815 816 switch (opt) { 817 case OPT_SOURCE: 818 if (fsc->source) 819 return invalfc(fsc, "Multiple sources specified"); 820 fsc->source = param->string; 821 param->string = NULL; 822 break; 823 824 case OPT_SUBTYPE: 825 if (ctx->subtype) 826 return invalfc(fsc, "Multiple subtypes specified"); 827 ctx->subtype = param->string; 828 param->string = NULL; 829 return 0; 830 831 case OPT_FD: 832 ctx->fd = result.uint_32; 833 ctx->fd_present = true; 834 break; 835 836 case OPT_ROOTMODE: 837 if (!fuse_valid_type(result.uint_32)) 838 return invalfc(fsc, "Invalid rootmode"); 839 ctx->rootmode = result.uint_32; 840 ctx->rootmode_present = true; 841 break; 842 843 case OPT_USER_ID: 844 kuid = result.uid; 845 /* 846 * The requested uid must be representable in the 847 * filesystem's idmapping. 848 */ 849 if (!kuid_has_mapping(fsc->user_ns, kuid)) 850 return invalfc(fsc, "Invalid user_id"); 851 ctx->user_id = kuid; 852 ctx->user_id_present = true; 853 break; 854 855 case OPT_GROUP_ID: 856 kgid = result.gid; 857 /* 858 * The requested gid must be representable in the 859 * filesystem's idmapping. 860 */ 861 if (!kgid_has_mapping(fsc->user_ns, kgid)) 862 return invalfc(fsc, "Invalid group_id"); 863 ctx->group_id = kgid; 864 ctx->group_id_present = true; 865 break; 866 867 case OPT_DEFAULT_PERMISSIONS: 868 ctx->default_permissions = true; 869 break; 870 871 case OPT_ALLOW_OTHER: 872 ctx->allow_other = true; 873 break; 874 875 case OPT_MAX_READ: 876 ctx->max_read = result.uint_32; 877 break; 878 879 case OPT_BLKSIZE: 880 if (!ctx->is_bdev) 881 return invalfc(fsc, "blksize only supported for fuseblk"); 882 ctx->blksize = result.uint_32; 883 break; 884 885 default: 886 return -EINVAL; 887 } 888 889 return 0; 890 } 891 892 static void fuse_free_fsc(struct fs_context *fsc) 893 { 894 struct fuse_fs_context *ctx = fsc->fs_private; 895 896 if (ctx) { 897 kfree(ctx->subtype); 898 kfree(ctx); 899 } 900 } 901 902 static int fuse_show_options(struct seq_file *m, struct dentry *root) 903 { 904 struct super_block *sb = root->d_sb; 905 struct fuse_conn *fc = get_fuse_conn_super(sb); 906 907 if (fc->legacy_opts_show) { 908 seq_printf(m, ",user_id=%u", 909 from_kuid_munged(fc->user_ns, fc->user_id)); 910 seq_printf(m, ",group_id=%u", 911 from_kgid_munged(fc->user_ns, fc->group_id)); 912 if (fc->default_permissions) 913 seq_puts(m, ",default_permissions"); 914 if (fc->allow_other) 915 seq_puts(m, ",allow_other"); 916 if (fc->max_read != ~0) 917 seq_printf(m, ",max_read=%u", fc->max_read); 918 if (sb->s_bdev && sb->s_blocksize != FUSE_DEFAULT_BLKSIZE) 919 seq_printf(m, ",blksize=%lu", sb->s_blocksize); 920 } 921 #ifdef CONFIG_FUSE_DAX 922 if (fc->dax_mode == FUSE_DAX_ALWAYS) 923 seq_puts(m, ",dax=always"); 924 else if (fc->dax_mode == FUSE_DAX_NEVER) 925 seq_puts(m, ",dax=never"); 926 else if (fc->dax_mode == FUSE_DAX_INODE_USER) 927 seq_puts(m, ",dax=inode"); 928 #endif 929 930 return 0; 931 } 932 933 static void fuse_iqueue_init(struct fuse_iqueue *fiq, 934 const struct fuse_iqueue_ops *ops, 935 void *priv) 936 { 937 memset(fiq, 0, sizeof(struct fuse_iqueue)); 938 spin_lock_init(&fiq->lock); 939 init_waitqueue_head(&fiq->waitq); 940 INIT_LIST_HEAD(&fiq->pending); 941 INIT_LIST_HEAD(&fiq->interrupts); 942 fiq->forget_list_tail = &fiq->forget_list_head; 943 fiq->connected = 1; 944 fiq->ops = ops; 945 fiq->priv = priv; 946 } 947 948 void fuse_pqueue_init(struct fuse_pqueue *fpq) 949 { 950 unsigned int i; 951 952 spin_lock_init(&fpq->lock); 953 for (i = 0; i < FUSE_PQ_HASH_SIZE; i++) 954 INIT_LIST_HEAD(&fpq->processing[i]); 955 INIT_LIST_HEAD(&fpq->io); 956 fpq->connected = 1; 957 } 958 959 void fuse_conn_init(struct fuse_conn *fc, struct fuse_mount *fm, 960 struct user_namespace *user_ns, 961 const struct fuse_iqueue_ops *fiq_ops, void *fiq_priv) 962 { 963 memset(fc, 0, sizeof(*fc)); 964 spin_lock_init(&fc->lock); 965 spin_lock_init(&fc->bg_lock); 966 init_rwsem(&fc->killsb); 967 refcount_set(&fc->count, 1); 968 atomic_set(&fc->dev_count, 1); 969 atomic_set(&fc->epoch, 1); 970 init_waitqueue_head(&fc->blocked_waitq); 971 fuse_iqueue_init(&fc->iq, fiq_ops, fiq_priv); 972 INIT_LIST_HEAD(&fc->bg_queue); 973 INIT_LIST_HEAD(&fc->entry); 974 INIT_LIST_HEAD(&fc->devices); 975 atomic_set(&fc->num_waiting, 0); 976 fc->max_background = FUSE_DEFAULT_MAX_BACKGROUND; 977 fc->congestion_threshold = FUSE_DEFAULT_CONGESTION_THRESHOLD; 978 atomic64_set(&fc->khctr, 0); 979 fc->polled_files = RB_ROOT; 980 fc->blocked = 0; 981 fc->initialized = 0; 982 fc->connected = 1; 983 atomic64_set(&fc->attr_version, 1); 984 atomic64_set(&fc->evict_ctr, 1); 985 get_random_bytes(&fc->scramble_key, sizeof(fc->scramble_key)); 986 fc->pid_ns = get_pid_ns(task_active_pid_ns(current)); 987 fc->user_ns = get_user_ns(user_ns); 988 fc->max_pages = FUSE_DEFAULT_MAX_PAGES_PER_REQ; 989 fc->max_pages_limit = fuse_max_pages_limit; 990 fc->name_max = FUSE_NAME_LOW_MAX; 991 fc->timeout.req_timeout = 0; 992 993 if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH)) 994 fuse_backing_files_init(fc); 995 996 INIT_LIST_HEAD(&fc->mounts); 997 list_add(&fm->fc_entry, &fc->mounts); 998 fm->fc = fc; 999 } 1000 EXPORT_SYMBOL_GPL(fuse_conn_init); 1001 1002 static void delayed_release(struct rcu_head *p) 1003 { 1004 struct fuse_conn *fc = container_of(p, struct fuse_conn, rcu); 1005 1006 fuse_uring_destruct(fc); 1007 1008 put_user_ns(fc->user_ns); 1009 fc->release(fc); 1010 } 1011 1012 void fuse_conn_put(struct fuse_conn *fc) 1013 { 1014 if (refcount_dec_and_test(&fc->count)) { 1015 struct fuse_iqueue *fiq = &fc->iq; 1016 struct fuse_sync_bucket *bucket; 1017 1018 if (IS_ENABLED(CONFIG_FUSE_DAX)) 1019 fuse_dax_conn_free(fc); 1020 if (fc->timeout.req_timeout) 1021 cancel_delayed_work_sync(&fc->timeout.work); 1022 if (fiq->ops->release) 1023 fiq->ops->release(fiq); 1024 put_pid_ns(fc->pid_ns); 1025 bucket = rcu_dereference_protected(fc->curr_bucket, 1); 1026 if (bucket) { 1027 WARN_ON(atomic_read(&bucket->count) != 1); 1028 kfree(bucket); 1029 } 1030 if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH)) 1031 fuse_backing_files_free(fc); 1032 call_rcu(&fc->rcu, delayed_release); 1033 } 1034 } 1035 EXPORT_SYMBOL_GPL(fuse_conn_put); 1036 1037 struct fuse_conn *fuse_conn_get(struct fuse_conn *fc) 1038 { 1039 refcount_inc(&fc->count); 1040 return fc; 1041 } 1042 EXPORT_SYMBOL_GPL(fuse_conn_get); 1043 1044 static struct inode *fuse_get_root_inode(struct super_block *sb, unsigned int mode) 1045 { 1046 struct fuse_attr attr; 1047 memset(&attr, 0, sizeof(attr)); 1048 1049 attr.mode = mode; 1050 attr.ino = FUSE_ROOT_ID; 1051 attr.nlink = 1; 1052 return fuse_iget(sb, FUSE_ROOT_ID, 0, &attr, 0, 0, 0); 1053 } 1054 1055 struct fuse_inode_handle { 1056 u64 nodeid; 1057 u32 generation; 1058 }; 1059 1060 static struct dentry *fuse_get_dentry(struct super_block *sb, 1061 struct fuse_inode_handle *handle) 1062 { 1063 struct fuse_conn *fc = get_fuse_conn_super(sb); 1064 struct inode *inode; 1065 struct dentry *entry; 1066 int err = -ESTALE; 1067 1068 if (handle->nodeid == 0) 1069 goto out_err; 1070 1071 inode = ilookup5(sb, handle->nodeid, fuse_inode_eq, &handle->nodeid); 1072 if (!inode) { 1073 struct fuse_entry_out outarg; 1074 const struct qstr name = QSTR_INIT(".", 1); 1075 1076 if (!fc->export_support) 1077 goto out_err; 1078 1079 err = fuse_lookup_name(sb, handle->nodeid, &name, &outarg, 1080 &inode); 1081 if (err && err != -ENOENT) 1082 goto out_err; 1083 if (err || !inode) { 1084 err = -ESTALE; 1085 goto out_err; 1086 } 1087 err = -EIO; 1088 if (get_node_id(inode) != handle->nodeid) 1089 goto out_iput; 1090 } 1091 err = -ESTALE; 1092 if (inode->i_generation != handle->generation) 1093 goto out_iput; 1094 1095 entry = d_obtain_alias(inode); 1096 if (!IS_ERR(entry) && get_node_id(inode) != FUSE_ROOT_ID) 1097 fuse_invalidate_entry_cache(entry); 1098 1099 return entry; 1100 1101 out_iput: 1102 iput(inode); 1103 out_err: 1104 return ERR_PTR(err); 1105 } 1106 1107 static int fuse_encode_fh(struct inode *inode, u32 *fh, int *max_len, 1108 struct inode *parent) 1109 { 1110 int len = parent ? 6 : 3; 1111 u64 nodeid; 1112 u32 generation; 1113 1114 if (*max_len < len) { 1115 *max_len = len; 1116 return FILEID_INVALID; 1117 } 1118 1119 nodeid = get_fuse_inode(inode)->nodeid; 1120 generation = inode->i_generation; 1121 1122 fh[0] = (u32)(nodeid >> 32); 1123 fh[1] = (u32)(nodeid & 0xffffffff); 1124 fh[2] = generation; 1125 1126 if (parent) { 1127 nodeid = get_fuse_inode(parent)->nodeid; 1128 generation = parent->i_generation; 1129 1130 fh[3] = (u32)(nodeid >> 32); 1131 fh[4] = (u32)(nodeid & 0xffffffff); 1132 fh[5] = generation; 1133 } 1134 1135 *max_len = len; 1136 return parent ? FILEID_INO64_GEN_PARENT : FILEID_INO64_GEN; 1137 } 1138 1139 static struct dentry *fuse_fh_to_dentry(struct super_block *sb, 1140 struct fid *fid, int fh_len, int fh_type) 1141 { 1142 struct fuse_inode_handle handle; 1143 1144 if ((fh_type != FILEID_INO64_GEN && 1145 fh_type != FILEID_INO64_GEN_PARENT) || fh_len < 3) 1146 return NULL; 1147 1148 handle.nodeid = (u64) fid->raw[0] << 32; 1149 handle.nodeid |= (u64) fid->raw[1]; 1150 handle.generation = fid->raw[2]; 1151 return fuse_get_dentry(sb, &handle); 1152 } 1153 1154 static struct dentry *fuse_fh_to_parent(struct super_block *sb, 1155 struct fid *fid, int fh_len, int fh_type) 1156 { 1157 struct fuse_inode_handle parent; 1158 1159 if (fh_type != FILEID_INO64_GEN_PARENT || fh_len < 6) 1160 return NULL; 1161 1162 parent.nodeid = (u64) fid->raw[3] << 32; 1163 parent.nodeid |= (u64) fid->raw[4]; 1164 parent.generation = fid->raw[5]; 1165 return fuse_get_dentry(sb, &parent); 1166 } 1167 1168 static struct dentry *fuse_get_parent(struct dentry *child) 1169 { 1170 struct inode *child_inode = d_inode(child); 1171 struct fuse_conn *fc = get_fuse_conn(child_inode); 1172 struct inode *inode; 1173 struct dentry *parent; 1174 struct fuse_entry_out outarg; 1175 int err; 1176 1177 if (!fc->export_support) 1178 return ERR_PTR(-ESTALE); 1179 1180 err = fuse_lookup_name(child_inode->i_sb, get_node_id(child_inode), 1181 &dotdot_name, &outarg, &inode); 1182 if (err) { 1183 if (err == -ENOENT) 1184 return ERR_PTR(-ESTALE); 1185 return ERR_PTR(err); 1186 } 1187 1188 parent = d_obtain_alias(inode); 1189 if (!IS_ERR(parent) && get_node_id(inode) != FUSE_ROOT_ID) 1190 fuse_invalidate_entry_cache(parent); 1191 1192 return parent; 1193 } 1194 1195 /* only for fid encoding; no support for file handle */ 1196 static const struct export_operations fuse_export_fid_operations = { 1197 .encode_fh = fuse_encode_fh, 1198 }; 1199 1200 static const struct export_operations fuse_export_operations = { 1201 .fh_to_dentry = fuse_fh_to_dentry, 1202 .fh_to_parent = fuse_fh_to_parent, 1203 .encode_fh = fuse_encode_fh, 1204 .get_parent = fuse_get_parent, 1205 }; 1206 1207 static const struct super_operations fuse_super_operations = { 1208 .alloc_inode = fuse_alloc_inode, 1209 .free_inode = fuse_free_inode, 1210 .evict_inode = fuse_evict_inode, 1211 .write_inode = fuse_write_inode, 1212 .drop_inode = generic_delete_inode, 1213 .umount_begin = fuse_umount_begin, 1214 .statfs = fuse_statfs, 1215 .sync_fs = fuse_sync_fs, 1216 .show_options = fuse_show_options, 1217 }; 1218 1219 static void sanitize_global_limit(unsigned int *limit) 1220 { 1221 /* 1222 * The default maximum number of async requests is calculated to consume 1223 * 1/2^13 of the total memory, assuming 392 bytes per request. 1224 */ 1225 if (*limit == 0) 1226 *limit = ((totalram_pages() << PAGE_SHIFT) >> 13) / 392; 1227 1228 if (*limit >= 1 << 16) 1229 *limit = (1 << 16) - 1; 1230 } 1231 1232 static int set_global_limit(const char *val, const struct kernel_param *kp) 1233 { 1234 int rv; 1235 1236 rv = param_set_uint(val, kp); 1237 if (rv) 1238 return rv; 1239 1240 sanitize_global_limit((unsigned int *)kp->arg); 1241 1242 return 0; 1243 } 1244 1245 static void process_init_limits(struct fuse_conn *fc, struct fuse_init_out *arg) 1246 { 1247 int cap_sys_admin = capable(CAP_SYS_ADMIN); 1248 1249 if (arg->minor < 13) 1250 return; 1251 1252 sanitize_global_limit(&max_user_bgreq); 1253 sanitize_global_limit(&max_user_congthresh); 1254 1255 spin_lock(&fc->bg_lock); 1256 if (arg->max_background) { 1257 fc->max_background = arg->max_background; 1258 1259 if (!cap_sys_admin && fc->max_background > max_user_bgreq) 1260 fc->max_background = max_user_bgreq; 1261 } 1262 if (arg->congestion_threshold) { 1263 fc->congestion_threshold = arg->congestion_threshold; 1264 1265 if (!cap_sys_admin && 1266 fc->congestion_threshold > max_user_congthresh) 1267 fc->congestion_threshold = max_user_congthresh; 1268 } 1269 spin_unlock(&fc->bg_lock); 1270 } 1271 1272 static void set_request_timeout(struct fuse_conn *fc, unsigned int timeout) 1273 { 1274 fc->timeout.req_timeout = secs_to_jiffies(timeout); 1275 INIT_DELAYED_WORK(&fc->timeout.work, fuse_check_timeout); 1276 queue_delayed_work(system_wq, &fc->timeout.work, 1277 fuse_timeout_timer_freq); 1278 } 1279 1280 static void init_server_timeout(struct fuse_conn *fc, unsigned int timeout) 1281 { 1282 if (!timeout && !fuse_max_req_timeout && !fuse_default_req_timeout) 1283 return; 1284 1285 if (!timeout) 1286 timeout = fuse_default_req_timeout; 1287 1288 if (fuse_max_req_timeout) { 1289 if (timeout) 1290 timeout = min(fuse_max_req_timeout, timeout); 1291 else 1292 timeout = fuse_max_req_timeout; 1293 } 1294 1295 timeout = max(FUSE_TIMEOUT_TIMER_FREQ, timeout); 1296 1297 set_request_timeout(fc, timeout); 1298 } 1299 1300 struct fuse_init_args { 1301 struct fuse_args args; 1302 struct fuse_init_in in; 1303 struct fuse_init_out out; 1304 }; 1305 1306 static void process_init_reply(struct fuse_mount *fm, struct fuse_args *args, 1307 int error) 1308 { 1309 struct fuse_conn *fc = fm->fc; 1310 struct fuse_init_args *ia = container_of(args, typeof(*ia), args); 1311 struct fuse_init_out *arg = &ia->out; 1312 bool ok = true; 1313 1314 if (error || arg->major != FUSE_KERNEL_VERSION) 1315 ok = false; 1316 else { 1317 unsigned long ra_pages; 1318 unsigned int timeout = 0; 1319 1320 process_init_limits(fc, arg); 1321 1322 if (arg->minor >= 6) { 1323 u64 flags = arg->flags; 1324 1325 if (flags & FUSE_INIT_EXT) 1326 flags |= (u64) arg->flags2 << 32; 1327 1328 ra_pages = arg->max_readahead / PAGE_SIZE; 1329 if (flags & FUSE_ASYNC_READ) 1330 fc->async_read = 1; 1331 if (!(flags & FUSE_POSIX_LOCKS)) 1332 fc->no_lock = 1; 1333 if (arg->minor >= 17) { 1334 if (!(flags & FUSE_FLOCK_LOCKS)) 1335 fc->no_flock = 1; 1336 } else { 1337 if (!(flags & FUSE_POSIX_LOCKS)) 1338 fc->no_flock = 1; 1339 } 1340 if (flags & FUSE_ATOMIC_O_TRUNC) 1341 fc->atomic_o_trunc = 1; 1342 if (arg->minor >= 9) { 1343 /* LOOKUP has dependency on proto version */ 1344 if (flags & FUSE_EXPORT_SUPPORT) 1345 fc->export_support = 1; 1346 } 1347 if (flags & FUSE_BIG_WRITES) 1348 fc->big_writes = 1; 1349 if (flags & FUSE_DONT_MASK) 1350 fc->dont_mask = 1; 1351 if (flags & FUSE_AUTO_INVAL_DATA) 1352 fc->auto_inval_data = 1; 1353 else if (flags & FUSE_EXPLICIT_INVAL_DATA) 1354 fc->explicit_inval_data = 1; 1355 if (flags & FUSE_DO_READDIRPLUS) { 1356 fc->do_readdirplus = 1; 1357 if (flags & FUSE_READDIRPLUS_AUTO) 1358 fc->readdirplus_auto = 1; 1359 } 1360 if (flags & FUSE_ASYNC_DIO) 1361 fc->async_dio = 1; 1362 if (flags & FUSE_WRITEBACK_CACHE) 1363 fc->writeback_cache = 1; 1364 if (flags & FUSE_PARALLEL_DIROPS) 1365 fc->parallel_dirops = 1; 1366 if (flags & FUSE_HANDLE_KILLPRIV) 1367 fc->handle_killpriv = 1; 1368 if (arg->time_gran && arg->time_gran <= 1000000000) 1369 fm->sb->s_time_gran = arg->time_gran; 1370 if ((flags & FUSE_POSIX_ACL)) { 1371 fc->default_permissions = 1; 1372 fc->posix_acl = 1; 1373 } 1374 if (flags & FUSE_CACHE_SYMLINKS) 1375 fc->cache_symlinks = 1; 1376 if (flags & FUSE_ABORT_ERROR) 1377 fc->abort_err = 1; 1378 if (flags & FUSE_MAX_PAGES) { 1379 fc->max_pages = 1380 min_t(unsigned int, fc->max_pages_limit, 1381 max_t(unsigned int, arg->max_pages, 1)); 1382 1383 /* 1384 * PATH_MAX file names might need two pages for 1385 * ops like rename 1386 */ 1387 if (fc->max_pages > 1) 1388 fc->name_max = FUSE_NAME_MAX; 1389 } 1390 if (IS_ENABLED(CONFIG_FUSE_DAX)) { 1391 if (flags & FUSE_MAP_ALIGNMENT && 1392 !fuse_dax_check_alignment(fc, arg->map_alignment)) { 1393 ok = false; 1394 } 1395 if (flags & FUSE_HAS_INODE_DAX) 1396 fc->inode_dax = 1; 1397 } 1398 if (flags & FUSE_HANDLE_KILLPRIV_V2) { 1399 fc->handle_killpriv_v2 = 1; 1400 fm->sb->s_flags |= SB_NOSEC; 1401 } 1402 if (flags & FUSE_SETXATTR_EXT) 1403 fc->setxattr_ext = 1; 1404 if (flags & FUSE_SECURITY_CTX) 1405 fc->init_security = 1; 1406 if (flags & FUSE_CREATE_SUPP_GROUP) 1407 fc->create_supp_group = 1; 1408 if (flags & FUSE_DIRECT_IO_ALLOW_MMAP) 1409 fc->direct_io_allow_mmap = 1; 1410 /* 1411 * max_stack_depth is the max stack depth of FUSE fs, 1412 * so it has to be at least 1 to support passthrough 1413 * to backing files. 1414 * 1415 * with max_stack_depth > 1, the backing files can be 1416 * on a stacked fs (e.g. overlayfs) themselves and with 1417 * max_stack_depth == 1, FUSE fs can be stacked as the 1418 * underlying fs of a stacked fs (e.g. overlayfs). 1419 * 1420 * Also don't allow the combination of FUSE_PASSTHROUGH 1421 * and FUSE_WRITEBACK_CACHE, current design doesn't handle 1422 * them together. 1423 */ 1424 if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH) && 1425 (flags & FUSE_PASSTHROUGH) && 1426 arg->max_stack_depth > 0 && 1427 arg->max_stack_depth <= FILESYSTEM_MAX_STACK_DEPTH && 1428 !(flags & FUSE_WRITEBACK_CACHE)) { 1429 fc->passthrough = 1; 1430 fc->max_stack_depth = arg->max_stack_depth; 1431 fm->sb->s_stack_depth = arg->max_stack_depth; 1432 } 1433 if (flags & FUSE_NO_EXPORT_SUPPORT) 1434 fm->sb->s_export_op = &fuse_export_fid_operations; 1435 if (flags & FUSE_ALLOW_IDMAP) { 1436 if (fc->default_permissions) 1437 fm->sb->s_iflags &= ~SB_I_NOIDMAP; 1438 else 1439 ok = false; 1440 } 1441 if (flags & FUSE_OVER_IO_URING && fuse_uring_enabled()) 1442 fc->io_uring = 1; 1443 1444 if (flags & FUSE_REQUEST_TIMEOUT) 1445 timeout = arg->request_timeout; 1446 } else { 1447 ra_pages = fc->max_read / PAGE_SIZE; 1448 fc->no_lock = 1; 1449 fc->no_flock = 1; 1450 } 1451 1452 init_server_timeout(fc, timeout); 1453 1454 fm->sb->s_bdi->ra_pages = 1455 min(fm->sb->s_bdi->ra_pages, ra_pages); 1456 fc->minor = arg->minor; 1457 fc->max_write = arg->minor < 5 ? 4096 : arg->max_write; 1458 fc->max_write = max_t(unsigned, 4096, fc->max_write); 1459 fc->conn_init = 1; 1460 } 1461 kfree(ia); 1462 1463 if (!ok) { 1464 fc->conn_init = 0; 1465 fc->conn_error = 1; 1466 } 1467 1468 fuse_set_initialized(fc); 1469 wake_up_all(&fc->blocked_waitq); 1470 } 1471 1472 void fuse_send_init(struct fuse_mount *fm) 1473 { 1474 struct fuse_init_args *ia; 1475 u64 flags; 1476 1477 ia = kzalloc(sizeof(*ia), GFP_KERNEL | __GFP_NOFAIL); 1478 1479 ia->in.major = FUSE_KERNEL_VERSION; 1480 ia->in.minor = FUSE_KERNEL_MINOR_VERSION; 1481 ia->in.max_readahead = fm->sb->s_bdi->ra_pages * PAGE_SIZE; 1482 flags = 1483 FUSE_ASYNC_READ | FUSE_POSIX_LOCKS | FUSE_ATOMIC_O_TRUNC | 1484 FUSE_EXPORT_SUPPORT | FUSE_BIG_WRITES | FUSE_DONT_MASK | 1485 FUSE_SPLICE_WRITE | FUSE_SPLICE_MOVE | FUSE_SPLICE_READ | 1486 FUSE_FLOCK_LOCKS | FUSE_HAS_IOCTL_DIR | FUSE_AUTO_INVAL_DATA | 1487 FUSE_DO_READDIRPLUS | FUSE_READDIRPLUS_AUTO | FUSE_ASYNC_DIO | 1488 FUSE_WRITEBACK_CACHE | FUSE_NO_OPEN_SUPPORT | 1489 FUSE_PARALLEL_DIROPS | FUSE_HANDLE_KILLPRIV | FUSE_POSIX_ACL | 1490 FUSE_ABORT_ERROR | FUSE_MAX_PAGES | FUSE_CACHE_SYMLINKS | 1491 FUSE_NO_OPENDIR_SUPPORT | FUSE_EXPLICIT_INVAL_DATA | 1492 FUSE_HANDLE_KILLPRIV_V2 | FUSE_SETXATTR_EXT | FUSE_INIT_EXT | 1493 FUSE_SECURITY_CTX | FUSE_CREATE_SUPP_GROUP | 1494 FUSE_HAS_EXPIRE_ONLY | FUSE_DIRECT_IO_ALLOW_MMAP | 1495 FUSE_NO_EXPORT_SUPPORT | FUSE_HAS_RESEND | FUSE_ALLOW_IDMAP | 1496 FUSE_REQUEST_TIMEOUT; 1497 #ifdef CONFIG_FUSE_DAX 1498 if (fm->fc->dax) 1499 flags |= FUSE_MAP_ALIGNMENT; 1500 if (fuse_is_inode_dax_mode(fm->fc->dax_mode)) 1501 flags |= FUSE_HAS_INODE_DAX; 1502 #endif 1503 if (fm->fc->auto_submounts) 1504 flags |= FUSE_SUBMOUNTS; 1505 if (IS_ENABLED(CONFIG_FUSE_PASSTHROUGH)) 1506 flags |= FUSE_PASSTHROUGH; 1507 1508 /* 1509 * This is just an information flag for fuse server. No need to check 1510 * the reply - server is either sending IORING_OP_URING_CMD or not. 1511 */ 1512 if (fuse_uring_enabled()) 1513 flags |= FUSE_OVER_IO_URING; 1514 1515 ia->in.flags = flags; 1516 ia->in.flags2 = flags >> 32; 1517 1518 ia->args.opcode = FUSE_INIT; 1519 ia->args.in_numargs = 1; 1520 ia->args.in_args[0].size = sizeof(ia->in); 1521 ia->args.in_args[0].value = &ia->in; 1522 ia->args.out_numargs = 1; 1523 /* Variable length argument used for backward compatibility 1524 with interface version < 7.5. Rest of init_out is zeroed 1525 by do_get_request(), so a short reply is not a problem */ 1526 ia->args.out_argvar = true; 1527 ia->args.out_args[0].size = sizeof(ia->out); 1528 ia->args.out_args[0].value = &ia->out; 1529 ia->args.force = true; 1530 ia->args.nocreds = true; 1531 ia->args.end = process_init_reply; 1532 1533 if (fuse_simple_background(fm, &ia->args, GFP_KERNEL) != 0) 1534 process_init_reply(fm, &ia->args, -ENOTCONN); 1535 } 1536 EXPORT_SYMBOL_GPL(fuse_send_init); 1537 1538 void fuse_free_conn(struct fuse_conn *fc) 1539 { 1540 WARN_ON(!list_empty(&fc->devices)); 1541 kfree(fc); 1542 } 1543 EXPORT_SYMBOL_GPL(fuse_free_conn); 1544 1545 static int fuse_bdi_init(struct fuse_conn *fc, struct super_block *sb) 1546 { 1547 int err; 1548 char *suffix = ""; 1549 1550 if (sb->s_bdev) { 1551 suffix = "-fuseblk"; 1552 /* 1553 * sb->s_bdi points to blkdev's bdi however we want to redirect 1554 * it to our private bdi... 1555 */ 1556 bdi_put(sb->s_bdi); 1557 sb->s_bdi = &noop_backing_dev_info; 1558 } 1559 err = super_setup_bdi_name(sb, "%u:%u%s", MAJOR(fc->dev), 1560 MINOR(fc->dev), suffix); 1561 if (err) 1562 return err; 1563 1564 /* fuse does it's own writeback accounting */ 1565 sb->s_bdi->capabilities &= ~BDI_CAP_WRITEBACK_ACCT; 1566 sb->s_bdi->capabilities |= BDI_CAP_STRICTLIMIT; 1567 1568 /* 1569 * For a single fuse filesystem use max 1% of dirty + 1570 * writeback threshold. 1571 * 1572 * This gives about 1M of write buffer for memory maps on a 1573 * machine with 1G and 10% dirty_ratio, which should be more 1574 * than enough. 1575 * 1576 * Privileged users can raise it by writing to 1577 * 1578 * /sys/class/bdi/<bdi>/max_ratio 1579 */ 1580 bdi_set_max_ratio(sb->s_bdi, 1); 1581 1582 return 0; 1583 } 1584 1585 struct fuse_dev *fuse_dev_alloc(void) 1586 { 1587 struct fuse_dev *fud; 1588 struct list_head *pq; 1589 1590 fud = kzalloc(sizeof(struct fuse_dev), GFP_KERNEL); 1591 if (!fud) 1592 return NULL; 1593 1594 pq = kcalloc(FUSE_PQ_HASH_SIZE, sizeof(struct list_head), GFP_KERNEL); 1595 if (!pq) { 1596 kfree(fud); 1597 return NULL; 1598 } 1599 1600 fud->pq.processing = pq; 1601 fuse_pqueue_init(&fud->pq); 1602 1603 return fud; 1604 } 1605 EXPORT_SYMBOL_GPL(fuse_dev_alloc); 1606 1607 void fuse_dev_install(struct fuse_dev *fud, struct fuse_conn *fc) 1608 { 1609 fud->fc = fuse_conn_get(fc); 1610 spin_lock(&fc->lock); 1611 list_add_tail(&fud->entry, &fc->devices); 1612 spin_unlock(&fc->lock); 1613 } 1614 EXPORT_SYMBOL_GPL(fuse_dev_install); 1615 1616 struct fuse_dev *fuse_dev_alloc_install(struct fuse_conn *fc) 1617 { 1618 struct fuse_dev *fud; 1619 1620 fud = fuse_dev_alloc(); 1621 if (!fud) 1622 return NULL; 1623 1624 fuse_dev_install(fud, fc); 1625 return fud; 1626 } 1627 EXPORT_SYMBOL_GPL(fuse_dev_alloc_install); 1628 1629 void fuse_dev_free(struct fuse_dev *fud) 1630 { 1631 struct fuse_conn *fc = fud->fc; 1632 1633 if (fc) { 1634 spin_lock(&fc->lock); 1635 list_del(&fud->entry); 1636 spin_unlock(&fc->lock); 1637 1638 fuse_conn_put(fc); 1639 } 1640 kfree(fud->pq.processing); 1641 kfree(fud); 1642 } 1643 EXPORT_SYMBOL_GPL(fuse_dev_free); 1644 1645 static void fuse_fill_attr_from_inode(struct fuse_attr *attr, 1646 const struct fuse_inode *fi) 1647 { 1648 struct timespec64 atime = inode_get_atime(&fi->inode); 1649 struct timespec64 mtime = inode_get_mtime(&fi->inode); 1650 struct timespec64 ctime = inode_get_ctime(&fi->inode); 1651 1652 *attr = (struct fuse_attr){ 1653 .ino = fi->inode.i_ino, 1654 .size = fi->inode.i_size, 1655 .blocks = fi->inode.i_blocks, 1656 .atime = atime.tv_sec, 1657 .mtime = mtime.tv_sec, 1658 .ctime = ctime.tv_sec, 1659 .atimensec = atime.tv_nsec, 1660 .mtimensec = mtime.tv_nsec, 1661 .ctimensec = ctime.tv_nsec, 1662 .mode = fi->inode.i_mode, 1663 .nlink = fi->inode.i_nlink, 1664 .uid = __kuid_val(fi->inode.i_uid), 1665 .gid = __kgid_val(fi->inode.i_gid), 1666 .rdev = fi->inode.i_rdev, 1667 .blksize = 1u << fi->inode.i_blkbits, 1668 }; 1669 } 1670 1671 static void fuse_sb_defaults(struct super_block *sb) 1672 { 1673 sb->s_magic = FUSE_SUPER_MAGIC; 1674 sb->s_op = &fuse_super_operations; 1675 sb->s_xattr = fuse_xattr_handlers; 1676 sb->s_maxbytes = MAX_LFS_FILESIZE; 1677 sb->s_time_gran = 1; 1678 sb->s_export_op = &fuse_export_operations; 1679 sb->s_iflags |= SB_I_IMA_UNVERIFIABLE_SIGNATURE; 1680 sb->s_iflags |= SB_I_NOIDMAP; 1681 if (sb->s_user_ns != &init_user_ns) 1682 sb->s_iflags |= SB_I_UNTRUSTED_MOUNTER; 1683 sb->s_flags &= ~(SB_NOSEC | SB_I_VERSION); 1684 } 1685 1686 static int fuse_fill_super_submount(struct super_block *sb, 1687 struct fuse_inode *parent_fi) 1688 { 1689 struct fuse_mount *fm = get_fuse_mount_super(sb); 1690 struct super_block *parent_sb = parent_fi->inode.i_sb; 1691 struct fuse_attr root_attr; 1692 struct inode *root; 1693 struct fuse_submount_lookup *sl; 1694 struct fuse_inode *fi; 1695 1696 fuse_sb_defaults(sb); 1697 fm->sb = sb; 1698 1699 WARN_ON(sb->s_bdi != &noop_backing_dev_info); 1700 sb->s_bdi = bdi_get(parent_sb->s_bdi); 1701 1702 sb->s_xattr = parent_sb->s_xattr; 1703 sb->s_export_op = parent_sb->s_export_op; 1704 sb->s_time_gran = parent_sb->s_time_gran; 1705 sb->s_blocksize = parent_sb->s_blocksize; 1706 sb->s_blocksize_bits = parent_sb->s_blocksize_bits; 1707 sb->s_subtype = kstrdup(parent_sb->s_subtype, GFP_KERNEL); 1708 if (parent_sb->s_subtype && !sb->s_subtype) 1709 return -ENOMEM; 1710 1711 fuse_fill_attr_from_inode(&root_attr, parent_fi); 1712 root = fuse_iget(sb, parent_fi->nodeid, 0, &root_attr, 0, 0, 1713 fuse_get_evict_ctr(fm->fc)); 1714 /* 1715 * This inode is just a duplicate, so it is not looked up and 1716 * its nlookup should not be incremented. fuse_iget() does 1717 * that, though, so undo it here. 1718 */ 1719 fi = get_fuse_inode(root); 1720 fi->nlookup--; 1721 1722 sb->s_d_op = &fuse_dentry_operations; 1723 sb->s_root = d_make_root(root); 1724 if (!sb->s_root) 1725 return -ENOMEM; 1726 1727 /* 1728 * Grab the parent's submount_lookup pointer and take a 1729 * reference on the shared nlookup from the parent. This is to 1730 * prevent the last forget for this nodeid from getting 1731 * triggered until all users have finished with it. 1732 */ 1733 sl = parent_fi->submount_lookup; 1734 WARN_ON(!sl); 1735 if (sl) { 1736 refcount_inc(&sl->count); 1737 fi->submount_lookup = sl; 1738 } 1739 1740 return 0; 1741 } 1742 1743 /* Filesystem context private data holds the FUSE inode of the mount point */ 1744 static int fuse_get_tree_submount(struct fs_context *fsc) 1745 { 1746 struct fuse_mount *fm; 1747 struct fuse_inode *mp_fi = fsc->fs_private; 1748 struct fuse_conn *fc = get_fuse_conn(&mp_fi->inode); 1749 struct super_block *sb; 1750 int err; 1751 1752 fm = kzalloc(sizeof(struct fuse_mount), GFP_KERNEL); 1753 if (!fm) 1754 return -ENOMEM; 1755 1756 fm->fc = fuse_conn_get(fc); 1757 fsc->s_fs_info = fm; 1758 sb = sget_fc(fsc, NULL, set_anon_super_fc); 1759 if (fsc->s_fs_info) 1760 fuse_mount_destroy(fm); 1761 if (IS_ERR(sb)) 1762 return PTR_ERR(sb); 1763 1764 /* Initialize superblock, making @mp_fi its root */ 1765 err = fuse_fill_super_submount(sb, mp_fi); 1766 if (err) { 1767 deactivate_locked_super(sb); 1768 return err; 1769 } 1770 1771 down_write(&fc->killsb); 1772 list_add_tail(&fm->fc_entry, &fc->mounts); 1773 up_write(&fc->killsb); 1774 1775 sb->s_flags |= SB_ACTIVE; 1776 fsc->root = dget(sb->s_root); 1777 1778 return 0; 1779 } 1780 1781 static const struct fs_context_operations fuse_context_submount_ops = { 1782 .get_tree = fuse_get_tree_submount, 1783 }; 1784 1785 int fuse_init_fs_context_submount(struct fs_context *fsc) 1786 { 1787 fsc->ops = &fuse_context_submount_ops; 1788 return 0; 1789 } 1790 EXPORT_SYMBOL_GPL(fuse_init_fs_context_submount); 1791 1792 int fuse_fill_super_common(struct super_block *sb, struct fuse_fs_context *ctx) 1793 { 1794 struct fuse_dev *fud = NULL; 1795 struct fuse_mount *fm = get_fuse_mount_super(sb); 1796 struct fuse_conn *fc = fm->fc; 1797 struct inode *root; 1798 struct dentry *root_dentry; 1799 int err; 1800 1801 err = -EINVAL; 1802 if (sb->s_flags & SB_MANDLOCK) 1803 goto err; 1804 1805 rcu_assign_pointer(fc->curr_bucket, fuse_sync_bucket_alloc()); 1806 fuse_sb_defaults(sb); 1807 1808 if (ctx->is_bdev) { 1809 #ifdef CONFIG_BLOCK 1810 err = -EINVAL; 1811 if (!sb_set_blocksize(sb, ctx->blksize)) 1812 goto err; 1813 #endif 1814 } else { 1815 sb->s_blocksize = PAGE_SIZE; 1816 sb->s_blocksize_bits = PAGE_SHIFT; 1817 } 1818 1819 sb->s_subtype = ctx->subtype; 1820 ctx->subtype = NULL; 1821 if (IS_ENABLED(CONFIG_FUSE_DAX)) { 1822 err = fuse_dax_conn_alloc(fc, ctx->dax_mode, ctx->dax_dev); 1823 if (err) 1824 goto err; 1825 } 1826 1827 if (ctx->fudptr) { 1828 err = -ENOMEM; 1829 fud = fuse_dev_alloc_install(fc); 1830 if (!fud) 1831 goto err_free_dax; 1832 } 1833 1834 fc->dev = sb->s_dev; 1835 fm->sb = sb; 1836 err = fuse_bdi_init(fc, sb); 1837 if (err) 1838 goto err_dev_free; 1839 1840 /* Handle umasking inside the fuse code */ 1841 if (sb->s_flags & SB_POSIXACL) 1842 fc->dont_mask = 1; 1843 sb->s_flags |= SB_POSIXACL; 1844 1845 fc->default_permissions = ctx->default_permissions; 1846 fc->allow_other = ctx->allow_other; 1847 fc->user_id = ctx->user_id; 1848 fc->group_id = ctx->group_id; 1849 fc->legacy_opts_show = ctx->legacy_opts_show; 1850 fc->max_read = max_t(unsigned int, 4096, ctx->max_read); 1851 fc->destroy = ctx->destroy; 1852 fc->no_control = ctx->no_control; 1853 fc->no_force_umount = ctx->no_force_umount; 1854 1855 err = -ENOMEM; 1856 root = fuse_get_root_inode(sb, ctx->rootmode); 1857 sb->s_d_op = &fuse_root_dentry_operations; 1858 root_dentry = d_make_root(root); 1859 if (!root_dentry) 1860 goto err_dev_free; 1861 /* Root dentry doesn't have .d_revalidate */ 1862 sb->s_d_op = &fuse_dentry_operations; 1863 1864 mutex_lock(&fuse_mutex); 1865 err = -EINVAL; 1866 if (ctx->fudptr && *ctx->fudptr) 1867 goto err_unlock; 1868 1869 err = fuse_ctl_add_conn(fc); 1870 if (err) 1871 goto err_unlock; 1872 1873 list_add_tail(&fc->entry, &fuse_conn_list); 1874 sb->s_root = root_dentry; 1875 if (ctx->fudptr) 1876 *ctx->fudptr = fud; 1877 mutex_unlock(&fuse_mutex); 1878 return 0; 1879 1880 err_unlock: 1881 mutex_unlock(&fuse_mutex); 1882 dput(root_dentry); 1883 err_dev_free: 1884 if (fud) 1885 fuse_dev_free(fud); 1886 err_free_dax: 1887 if (IS_ENABLED(CONFIG_FUSE_DAX)) 1888 fuse_dax_conn_free(fc); 1889 err: 1890 return err; 1891 } 1892 EXPORT_SYMBOL_GPL(fuse_fill_super_common); 1893 1894 static int fuse_fill_super(struct super_block *sb, struct fs_context *fsc) 1895 { 1896 struct fuse_fs_context *ctx = fsc->fs_private; 1897 int err; 1898 1899 if (!ctx->file || !ctx->rootmode_present || 1900 !ctx->user_id_present || !ctx->group_id_present) 1901 return -EINVAL; 1902 1903 /* 1904 * Require mount to happen from the same user namespace which 1905 * opened /dev/fuse to prevent potential attacks. 1906 */ 1907 if ((ctx->file->f_op != &fuse_dev_operations) || 1908 (ctx->file->f_cred->user_ns != sb->s_user_ns)) 1909 return -EINVAL; 1910 ctx->fudptr = &ctx->file->private_data; 1911 1912 err = fuse_fill_super_common(sb, ctx); 1913 if (err) 1914 return err; 1915 /* file->private_data shall be visible on all CPUs after this */ 1916 smp_mb(); 1917 fuse_send_init(get_fuse_mount_super(sb)); 1918 return 0; 1919 } 1920 1921 /* 1922 * This is the path where user supplied an already initialized fuse dev. In 1923 * this case never create a new super if the old one is gone. 1924 */ 1925 static int fuse_set_no_super(struct super_block *sb, struct fs_context *fsc) 1926 { 1927 return -ENOTCONN; 1928 } 1929 1930 static int fuse_test_super(struct super_block *sb, struct fs_context *fsc) 1931 { 1932 1933 return fsc->sget_key == get_fuse_conn_super(sb); 1934 } 1935 1936 static int fuse_get_tree(struct fs_context *fsc) 1937 { 1938 struct fuse_fs_context *ctx = fsc->fs_private; 1939 struct fuse_dev *fud; 1940 struct fuse_conn *fc; 1941 struct fuse_mount *fm; 1942 struct super_block *sb; 1943 int err; 1944 1945 fc = kmalloc(sizeof(*fc), GFP_KERNEL); 1946 if (!fc) 1947 return -ENOMEM; 1948 1949 fm = kzalloc(sizeof(*fm), GFP_KERNEL); 1950 if (!fm) { 1951 kfree(fc); 1952 return -ENOMEM; 1953 } 1954 1955 fuse_conn_init(fc, fm, fsc->user_ns, &fuse_dev_fiq_ops, NULL); 1956 fc->release = fuse_free_conn; 1957 1958 fsc->s_fs_info = fm; 1959 1960 if (ctx->fd_present) 1961 ctx->file = fget(ctx->fd); 1962 1963 if (IS_ENABLED(CONFIG_BLOCK) && ctx->is_bdev) { 1964 err = get_tree_bdev(fsc, fuse_fill_super); 1965 goto out; 1966 } 1967 /* 1968 * While block dev mount can be initialized with a dummy device fd 1969 * (found by device name), normal fuse mounts can't 1970 */ 1971 err = -EINVAL; 1972 if (!ctx->file) 1973 goto out; 1974 1975 /* 1976 * Allow creating a fuse mount with an already initialized fuse 1977 * connection 1978 */ 1979 fud = READ_ONCE(ctx->file->private_data); 1980 if (ctx->file->f_op == &fuse_dev_operations && fud) { 1981 fsc->sget_key = fud->fc; 1982 sb = sget_fc(fsc, fuse_test_super, fuse_set_no_super); 1983 err = PTR_ERR_OR_ZERO(sb); 1984 if (!IS_ERR(sb)) 1985 fsc->root = dget(sb->s_root); 1986 } else { 1987 err = get_tree_nodev(fsc, fuse_fill_super); 1988 } 1989 out: 1990 if (fsc->s_fs_info) 1991 fuse_mount_destroy(fm); 1992 if (ctx->file) 1993 fput(ctx->file); 1994 return err; 1995 } 1996 1997 static const struct fs_context_operations fuse_context_ops = { 1998 .free = fuse_free_fsc, 1999 .parse_param = fuse_parse_param, 2000 .reconfigure = fuse_reconfigure, 2001 .get_tree = fuse_get_tree, 2002 }; 2003 2004 /* 2005 * Set up the filesystem mount context. 2006 */ 2007 static int fuse_init_fs_context(struct fs_context *fsc) 2008 { 2009 struct fuse_fs_context *ctx; 2010 2011 ctx = kzalloc(sizeof(struct fuse_fs_context), GFP_KERNEL); 2012 if (!ctx) 2013 return -ENOMEM; 2014 2015 ctx->max_read = ~0; 2016 ctx->blksize = FUSE_DEFAULT_BLKSIZE; 2017 ctx->legacy_opts_show = true; 2018 2019 #ifdef CONFIG_BLOCK 2020 if (fsc->fs_type == &fuseblk_fs_type) { 2021 ctx->is_bdev = true; 2022 ctx->destroy = true; 2023 } 2024 #endif 2025 2026 fsc->fs_private = ctx; 2027 fsc->ops = &fuse_context_ops; 2028 return 0; 2029 } 2030 2031 bool fuse_mount_remove(struct fuse_mount *fm) 2032 { 2033 struct fuse_conn *fc = fm->fc; 2034 bool last = false; 2035 2036 down_write(&fc->killsb); 2037 list_del_init(&fm->fc_entry); 2038 if (list_empty(&fc->mounts)) 2039 last = true; 2040 up_write(&fc->killsb); 2041 2042 return last; 2043 } 2044 EXPORT_SYMBOL_GPL(fuse_mount_remove); 2045 2046 void fuse_conn_destroy(struct fuse_mount *fm) 2047 { 2048 struct fuse_conn *fc = fm->fc; 2049 2050 if (fc->destroy) 2051 fuse_send_destroy(fm); 2052 2053 fuse_abort_conn(fc); 2054 fuse_wait_aborted(fc); 2055 2056 if (!list_empty(&fc->entry)) { 2057 mutex_lock(&fuse_mutex); 2058 list_del(&fc->entry); 2059 fuse_ctl_remove_conn(fc); 2060 mutex_unlock(&fuse_mutex); 2061 } 2062 } 2063 EXPORT_SYMBOL_GPL(fuse_conn_destroy); 2064 2065 static void fuse_sb_destroy(struct super_block *sb) 2066 { 2067 struct fuse_mount *fm = get_fuse_mount_super(sb); 2068 bool last; 2069 2070 if (sb->s_root) { 2071 last = fuse_mount_remove(fm); 2072 if (last) 2073 fuse_conn_destroy(fm); 2074 } 2075 } 2076 2077 void fuse_mount_destroy(struct fuse_mount *fm) 2078 { 2079 fuse_conn_put(fm->fc); 2080 kfree_rcu(fm, rcu); 2081 } 2082 EXPORT_SYMBOL(fuse_mount_destroy); 2083 2084 static void fuse_kill_sb_anon(struct super_block *sb) 2085 { 2086 fuse_sb_destroy(sb); 2087 kill_anon_super(sb); 2088 fuse_mount_destroy(get_fuse_mount_super(sb)); 2089 } 2090 2091 static struct file_system_type fuse_fs_type = { 2092 .owner = THIS_MODULE, 2093 .name = "fuse", 2094 .fs_flags = FS_HAS_SUBTYPE | FS_USERNS_MOUNT | FS_ALLOW_IDMAP, 2095 .init_fs_context = fuse_init_fs_context, 2096 .parameters = fuse_fs_parameters, 2097 .kill_sb = fuse_kill_sb_anon, 2098 }; 2099 MODULE_ALIAS_FS("fuse"); 2100 2101 #ifdef CONFIG_BLOCK 2102 static void fuse_kill_sb_blk(struct super_block *sb) 2103 { 2104 fuse_sb_destroy(sb); 2105 kill_block_super(sb); 2106 fuse_mount_destroy(get_fuse_mount_super(sb)); 2107 } 2108 2109 static struct file_system_type fuseblk_fs_type = { 2110 .owner = THIS_MODULE, 2111 .name = "fuseblk", 2112 .init_fs_context = fuse_init_fs_context, 2113 .parameters = fuse_fs_parameters, 2114 .kill_sb = fuse_kill_sb_blk, 2115 .fs_flags = FS_REQUIRES_DEV | FS_HAS_SUBTYPE | FS_ALLOW_IDMAP, 2116 }; 2117 MODULE_ALIAS_FS("fuseblk"); 2118 2119 static inline int register_fuseblk(void) 2120 { 2121 return register_filesystem(&fuseblk_fs_type); 2122 } 2123 2124 static inline void unregister_fuseblk(void) 2125 { 2126 unregister_filesystem(&fuseblk_fs_type); 2127 } 2128 #else 2129 static inline int register_fuseblk(void) 2130 { 2131 return 0; 2132 } 2133 2134 static inline void unregister_fuseblk(void) 2135 { 2136 } 2137 #endif 2138 2139 static void fuse_inode_init_once(void *foo) 2140 { 2141 struct inode *inode = foo; 2142 2143 inode_init_once(inode); 2144 } 2145 2146 static int __init fuse_fs_init(void) 2147 { 2148 int err; 2149 2150 fuse_inode_cachep = kmem_cache_create("fuse_inode", 2151 sizeof(struct fuse_inode), 0, 2152 SLAB_HWCACHE_ALIGN|SLAB_ACCOUNT|SLAB_RECLAIM_ACCOUNT, 2153 fuse_inode_init_once); 2154 err = -ENOMEM; 2155 if (!fuse_inode_cachep) 2156 goto out; 2157 2158 err = register_fuseblk(); 2159 if (err) 2160 goto out2; 2161 2162 err = register_filesystem(&fuse_fs_type); 2163 if (err) 2164 goto out3; 2165 2166 err = fuse_sysctl_register(); 2167 if (err) 2168 goto out4; 2169 2170 return 0; 2171 2172 out4: 2173 unregister_filesystem(&fuse_fs_type); 2174 out3: 2175 unregister_fuseblk(); 2176 out2: 2177 kmem_cache_destroy(fuse_inode_cachep); 2178 out: 2179 return err; 2180 } 2181 2182 static void fuse_fs_cleanup(void) 2183 { 2184 fuse_sysctl_unregister(); 2185 unregister_filesystem(&fuse_fs_type); 2186 unregister_fuseblk(); 2187 2188 /* 2189 * Make sure all delayed rcu free inodes are flushed before we 2190 * destroy cache. 2191 */ 2192 rcu_barrier(); 2193 kmem_cache_destroy(fuse_inode_cachep); 2194 } 2195 2196 static struct kobject *fuse_kobj; 2197 2198 static int fuse_sysfs_init(void) 2199 { 2200 int err; 2201 2202 fuse_kobj = kobject_create_and_add("fuse", fs_kobj); 2203 if (!fuse_kobj) { 2204 err = -ENOMEM; 2205 goto out_err; 2206 } 2207 2208 err = sysfs_create_mount_point(fuse_kobj, "connections"); 2209 if (err) 2210 goto out_fuse_unregister; 2211 2212 return 0; 2213 2214 out_fuse_unregister: 2215 kobject_put(fuse_kobj); 2216 out_err: 2217 return err; 2218 } 2219 2220 static void fuse_sysfs_cleanup(void) 2221 { 2222 sysfs_remove_mount_point(fuse_kobj, "connections"); 2223 kobject_put(fuse_kobj); 2224 } 2225 2226 static int __init fuse_init(void) 2227 { 2228 int res; 2229 2230 pr_info("init (API version %i.%i)\n", 2231 FUSE_KERNEL_VERSION, FUSE_KERNEL_MINOR_VERSION); 2232 2233 INIT_LIST_HEAD(&fuse_conn_list); 2234 res = fuse_fs_init(); 2235 if (res) 2236 goto err; 2237 2238 res = fuse_dev_init(); 2239 if (res) 2240 goto err_fs_cleanup; 2241 2242 res = fuse_sysfs_init(); 2243 if (res) 2244 goto err_dev_cleanup; 2245 2246 res = fuse_ctl_init(); 2247 if (res) 2248 goto err_sysfs_cleanup; 2249 2250 sanitize_global_limit(&max_user_bgreq); 2251 sanitize_global_limit(&max_user_congthresh); 2252 2253 return 0; 2254 2255 err_sysfs_cleanup: 2256 fuse_sysfs_cleanup(); 2257 err_dev_cleanup: 2258 fuse_dev_cleanup(); 2259 err_fs_cleanup: 2260 fuse_fs_cleanup(); 2261 err: 2262 return res; 2263 } 2264 2265 static void __exit fuse_exit(void) 2266 { 2267 pr_debug("exit\n"); 2268 2269 fuse_ctl_cleanup(); 2270 fuse_sysfs_cleanup(); 2271 fuse_fs_cleanup(); 2272 fuse_dev_cleanup(); 2273 } 2274 2275 module_init(fuse_init); 2276 module_exit(fuse_exit); 2277