1 /* 2 * linux/fs/super.c 3 * 4 * Copyright (C) 1991, 1992 Linus Torvalds 5 * 6 * super.c contains code to handle: - mount structures 7 * - super-block tables 8 * - filesystem drivers list 9 * - mount system call 10 * - umount system call 11 * - ustat system call 12 * 13 * GK 2/5/95 - Changed to support mounting the root fs via NFS 14 * 15 * Added kerneld support: Jacques Gelinas and Bjorn Ekwall 16 * Added change_root: Werner Almesberger & Hans Lermen, Feb '96 17 * Added options to /proc/mounts: 18 * Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996. 19 * Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998 20 * Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000 21 */ 22 23 #include <linux/export.h> 24 #include <linux/slab.h> 25 #include <linux/acct.h> 26 #include <linux/blkdev.h> 27 #include <linux/mount.h> 28 #include <linux/security.h> 29 #include <linux/writeback.h> /* for the emergency remount stuff */ 30 #include <linux/idr.h> 31 #include <linux/mutex.h> 32 #include <linux/backing-dev.h> 33 #include <linux/rculist_bl.h> 34 #include <linux/cleancache.h> 35 #include <linux/fsnotify.h> 36 #include <linux/lockdep.h> 37 #include "internal.h" 38 39 40 LIST_HEAD(super_blocks); 41 DEFINE_SPINLOCK(sb_lock); 42 43 static char *sb_writers_name[SB_FREEZE_LEVELS] = { 44 "sb_writers", 45 "sb_pagefaults", 46 "sb_internal", 47 }; 48 49 /* 50 * One thing we have to be careful of with a per-sb shrinker is that we don't 51 * drop the last active reference to the superblock from within the shrinker. 52 * If that happens we could trigger unregistering the shrinker from within the 53 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we 54 * take a passive reference to the superblock to avoid this from occurring. 55 */ 56 static int prune_super(struct shrinker *shrink, struct shrink_control *sc) 57 { 58 struct super_block *sb; 59 int fs_objects = 0; 60 int total_objects; 61 62 sb = container_of(shrink, struct super_block, s_shrink); 63 64 /* 65 * Deadlock avoidance. We may hold various FS locks, and we don't want 66 * to recurse into the FS that called us in clear_inode() and friends.. 67 */ 68 if (sc->nr_to_scan && !(sc->gfp_mask & __GFP_FS)) 69 return -1; 70 71 if (!grab_super_passive(sb)) 72 return -1; 73 74 if (sb->s_op && sb->s_op->nr_cached_objects) 75 fs_objects = sb->s_op->nr_cached_objects(sb); 76 77 total_objects = sb->s_nr_dentry_unused + 78 sb->s_nr_inodes_unused + fs_objects + 1; 79 80 if (sc->nr_to_scan) { 81 int dentries; 82 int inodes; 83 84 /* proportion the scan between the caches */ 85 dentries = (sc->nr_to_scan * sb->s_nr_dentry_unused) / 86 total_objects; 87 inodes = (sc->nr_to_scan * sb->s_nr_inodes_unused) / 88 total_objects; 89 if (fs_objects) 90 fs_objects = (sc->nr_to_scan * fs_objects) / 91 total_objects; 92 /* 93 * prune the dcache first as the icache is pinned by it, then 94 * prune the icache, followed by the filesystem specific caches 95 */ 96 prune_dcache_sb(sb, dentries); 97 prune_icache_sb(sb, inodes); 98 99 if (fs_objects && sb->s_op->free_cached_objects) { 100 sb->s_op->free_cached_objects(sb, fs_objects); 101 fs_objects = sb->s_op->nr_cached_objects(sb); 102 } 103 total_objects = sb->s_nr_dentry_unused + 104 sb->s_nr_inodes_unused + fs_objects; 105 } 106 107 total_objects = (total_objects / 100) * sysctl_vfs_cache_pressure; 108 drop_super(sb); 109 return total_objects; 110 } 111 112 static int init_sb_writers(struct super_block *s, struct file_system_type *type) 113 { 114 int err; 115 int i; 116 117 for (i = 0; i < SB_FREEZE_LEVELS; i++) { 118 err = percpu_counter_init(&s->s_writers.counter[i], 0); 119 if (err < 0) 120 goto err_out; 121 lockdep_init_map(&s->s_writers.lock_map[i], sb_writers_name[i], 122 &type->s_writers_key[i], 0); 123 } 124 init_waitqueue_head(&s->s_writers.wait); 125 init_waitqueue_head(&s->s_writers.wait_unfrozen); 126 return 0; 127 err_out: 128 while (--i >= 0) 129 percpu_counter_destroy(&s->s_writers.counter[i]); 130 return err; 131 } 132 133 static void destroy_sb_writers(struct super_block *s) 134 { 135 int i; 136 137 for (i = 0; i < SB_FREEZE_LEVELS; i++) 138 percpu_counter_destroy(&s->s_writers.counter[i]); 139 } 140 141 /** 142 * alloc_super - create new superblock 143 * @type: filesystem type superblock should belong to 144 * @flags: the mount flags 145 * 146 * Allocates and initializes a new &struct super_block. alloc_super() 147 * returns a pointer new superblock or %NULL if allocation had failed. 148 */ 149 static struct super_block *alloc_super(struct file_system_type *type, int flags) 150 { 151 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER); 152 static const struct super_operations default_op; 153 154 if (s) { 155 if (security_sb_alloc(s)) { 156 /* 157 * We cannot call security_sb_free() without 158 * security_sb_alloc() succeeding. So bail out manually 159 */ 160 kfree(s); 161 s = NULL; 162 goto out; 163 } 164 #ifdef CONFIG_SMP 165 s->s_files = alloc_percpu(struct list_head); 166 if (!s->s_files) 167 goto err_out; 168 else { 169 int i; 170 171 for_each_possible_cpu(i) 172 INIT_LIST_HEAD(per_cpu_ptr(s->s_files, i)); 173 } 174 #else 175 INIT_LIST_HEAD(&s->s_files); 176 #endif 177 if (init_sb_writers(s, type)) 178 goto err_out; 179 s->s_flags = flags; 180 s->s_bdi = &default_backing_dev_info; 181 INIT_HLIST_NODE(&s->s_instances); 182 INIT_HLIST_BL_HEAD(&s->s_anon); 183 INIT_LIST_HEAD(&s->s_inodes); 184 INIT_LIST_HEAD(&s->s_dentry_lru); 185 INIT_LIST_HEAD(&s->s_inode_lru); 186 spin_lock_init(&s->s_inode_lru_lock); 187 INIT_LIST_HEAD(&s->s_mounts); 188 init_rwsem(&s->s_umount); 189 lockdep_set_class(&s->s_umount, &type->s_umount_key); 190 /* 191 * sget() can have s_umount recursion. 192 * 193 * When it cannot find a suitable sb, it allocates a new 194 * one (this one), and tries again to find a suitable old 195 * one. 196 * 197 * In case that succeeds, it will acquire the s_umount 198 * lock of the old one. Since these are clearly distrinct 199 * locks, and this object isn't exposed yet, there's no 200 * risk of deadlocks. 201 * 202 * Annotate this by putting this lock in a different 203 * subclass. 204 */ 205 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING); 206 s->s_count = 1; 207 atomic_set(&s->s_active, 1); 208 mutex_init(&s->s_vfs_rename_mutex); 209 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key); 210 mutex_init(&s->s_dquot.dqio_mutex); 211 mutex_init(&s->s_dquot.dqonoff_mutex); 212 init_rwsem(&s->s_dquot.dqptr_sem); 213 s->s_maxbytes = MAX_NON_LFS; 214 s->s_op = &default_op; 215 s->s_time_gran = 1000000000; 216 s->cleancache_poolid = -1; 217 218 s->s_shrink.seeks = DEFAULT_SEEKS; 219 s->s_shrink.shrink = prune_super; 220 s->s_shrink.batch = 1024; 221 } 222 out: 223 return s; 224 err_out: 225 security_sb_free(s); 226 #ifdef CONFIG_SMP 227 if (s->s_files) 228 free_percpu(s->s_files); 229 #endif 230 destroy_sb_writers(s); 231 kfree(s); 232 s = NULL; 233 goto out; 234 } 235 236 /** 237 * destroy_super - frees a superblock 238 * @s: superblock to free 239 * 240 * Frees a superblock. 241 */ 242 static inline void destroy_super(struct super_block *s) 243 { 244 #ifdef CONFIG_SMP 245 free_percpu(s->s_files); 246 #endif 247 destroy_sb_writers(s); 248 security_sb_free(s); 249 WARN_ON(!list_empty(&s->s_mounts)); 250 kfree(s->s_subtype); 251 kfree(s->s_options); 252 kfree(s); 253 } 254 255 /* Superblock refcounting */ 256 257 /* 258 * Drop a superblock's refcount. The caller must hold sb_lock. 259 */ 260 static void __put_super(struct super_block *sb) 261 { 262 if (!--sb->s_count) { 263 list_del_init(&sb->s_list); 264 destroy_super(sb); 265 } 266 } 267 268 /** 269 * put_super - drop a temporary reference to superblock 270 * @sb: superblock in question 271 * 272 * Drops a temporary reference, frees superblock if there's no 273 * references left. 274 */ 275 static void put_super(struct super_block *sb) 276 { 277 spin_lock(&sb_lock); 278 __put_super(sb); 279 spin_unlock(&sb_lock); 280 } 281 282 283 /** 284 * deactivate_locked_super - drop an active reference to superblock 285 * @s: superblock to deactivate 286 * 287 * Drops an active reference to superblock, converting it into a temprory 288 * one if there is no other active references left. In that case we 289 * tell fs driver to shut it down and drop the temporary reference we 290 * had just acquired. 291 * 292 * Caller holds exclusive lock on superblock; that lock is released. 293 */ 294 void deactivate_locked_super(struct super_block *s) 295 { 296 struct file_system_type *fs = s->s_type; 297 if (atomic_dec_and_test(&s->s_active)) { 298 cleancache_invalidate_fs(s); 299 fs->kill_sb(s); 300 301 /* caches are now gone, we can safely kill the shrinker now */ 302 unregister_shrinker(&s->s_shrink); 303 put_filesystem(fs); 304 put_super(s); 305 } else { 306 up_write(&s->s_umount); 307 } 308 } 309 310 EXPORT_SYMBOL(deactivate_locked_super); 311 312 /** 313 * deactivate_super - drop an active reference to superblock 314 * @s: superblock to deactivate 315 * 316 * Variant of deactivate_locked_super(), except that superblock is *not* 317 * locked by caller. If we are going to drop the final active reference, 318 * lock will be acquired prior to that. 319 */ 320 void deactivate_super(struct super_block *s) 321 { 322 if (!atomic_add_unless(&s->s_active, -1, 1)) { 323 down_write(&s->s_umount); 324 deactivate_locked_super(s); 325 } 326 } 327 328 EXPORT_SYMBOL(deactivate_super); 329 330 /** 331 * grab_super - acquire an active reference 332 * @s: reference we are trying to make active 333 * 334 * Tries to acquire an active reference. grab_super() is used when we 335 * had just found a superblock in super_blocks or fs_type->fs_supers 336 * and want to turn it into a full-blown active reference. grab_super() 337 * is called with sb_lock held and drops it. Returns 1 in case of 338 * success, 0 if we had failed (superblock contents was already dead or 339 * dying when grab_super() had been called). 340 */ 341 static int grab_super(struct super_block *s) __releases(sb_lock) 342 { 343 if (atomic_inc_not_zero(&s->s_active)) { 344 spin_unlock(&sb_lock); 345 return 1; 346 } 347 /* it's going away */ 348 s->s_count++; 349 spin_unlock(&sb_lock); 350 /* wait for it to die */ 351 down_write(&s->s_umount); 352 up_write(&s->s_umount); 353 put_super(s); 354 return 0; 355 } 356 357 /* 358 * grab_super_passive - acquire a passive reference 359 * @sb: reference we are trying to grab 360 * 361 * Tries to acquire a passive reference. This is used in places where we 362 * cannot take an active reference but we need to ensure that the 363 * superblock does not go away while we are working on it. It returns 364 * false if a reference was not gained, and returns true with the s_umount 365 * lock held in read mode if a reference is gained. On successful return, 366 * the caller must drop the s_umount lock and the passive reference when 367 * done. 368 */ 369 bool grab_super_passive(struct super_block *sb) 370 { 371 spin_lock(&sb_lock); 372 if (hlist_unhashed(&sb->s_instances)) { 373 spin_unlock(&sb_lock); 374 return false; 375 } 376 377 sb->s_count++; 378 spin_unlock(&sb_lock); 379 380 if (down_read_trylock(&sb->s_umount)) { 381 if (sb->s_root && (sb->s_flags & MS_BORN)) 382 return true; 383 up_read(&sb->s_umount); 384 } 385 386 put_super(sb); 387 return false; 388 } 389 390 /** 391 * generic_shutdown_super - common helper for ->kill_sb() 392 * @sb: superblock to kill 393 * 394 * generic_shutdown_super() does all fs-independent work on superblock 395 * shutdown. Typical ->kill_sb() should pick all fs-specific objects 396 * that need destruction out of superblock, call generic_shutdown_super() 397 * and release aforementioned objects. Note: dentries and inodes _are_ 398 * taken care of and do not need specific handling. 399 * 400 * Upon calling this function, the filesystem may no longer alter or 401 * rearrange the set of dentries belonging to this super_block, nor may it 402 * change the attachments of dentries to inodes. 403 */ 404 void generic_shutdown_super(struct super_block *sb) 405 { 406 const struct super_operations *sop = sb->s_op; 407 408 if (sb->s_root) { 409 shrink_dcache_for_umount(sb); 410 sync_filesystem(sb); 411 sb->s_flags &= ~MS_ACTIVE; 412 413 fsnotify_unmount_inodes(&sb->s_inodes); 414 415 evict_inodes(sb); 416 417 if (sop->put_super) 418 sop->put_super(sb); 419 420 if (!list_empty(&sb->s_inodes)) { 421 printk("VFS: Busy inodes after unmount of %s. " 422 "Self-destruct in 5 seconds. Have a nice day...\n", 423 sb->s_id); 424 } 425 } 426 spin_lock(&sb_lock); 427 /* should be initialized for __put_super_and_need_restart() */ 428 hlist_del_init(&sb->s_instances); 429 spin_unlock(&sb_lock); 430 up_write(&sb->s_umount); 431 } 432 433 EXPORT_SYMBOL(generic_shutdown_super); 434 435 /** 436 * sget - find or create a superblock 437 * @type: filesystem type superblock should belong to 438 * @test: comparison callback 439 * @set: setup callback 440 * @flags: mount flags 441 * @data: argument to each of them 442 */ 443 struct super_block *sget(struct file_system_type *type, 444 int (*test)(struct super_block *,void *), 445 int (*set)(struct super_block *,void *), 446 int flags, 447 void *data) 448 { 449 struct super_block *s = NULL; 450 struct hlist_node *node; 451 struct super_block *old; 452 int err; 453 454 retry: 455 spin_lock(&sb_lock); 456 if (test) { 457 hlist_for_each_entry(old, node, &type->fs_supers, s_instances) { 458 if (!test(old, data)) 459 continue; 460 if (!grab_super(old)) 461 goto retry; 462 if (s) { 463 up_write(&s->s_umount); 464 destroy_super(s); 465 s = NULL; 466 } 467 down_write(&old->s_umount); 468 if (unlikely(!(old->s_flags & MS_BORN))) { 469 deactivate_locked_super(old); 470 goto retry; 471 } 472 return old; 473 } 474 } 475 if (!s) { 476 spin_unlock(&sb_lock); 477 s = alloc_super(type, flags); 478 if (!s) 479 return ERR_PTR(-ENOMEM); 480 goto retry; 481 } 482 483 err = set(s, data); 484 if (err) { 485 spin_unlock(&sb_lock); 486 up_write(&s->s_umount); 487 destroy_super(s); 488 return ERR_PTR(err); 489 } 490 s->s_type = type; 491 strlcpy(s->s_id, type->name, sizeof(s->s_id)); 492 list_add_tail(&s->s_list, &super_blocks); 493 hlist_add_head(&s->s_instances, &type->fs_supers); 494 spin_unlock(&sb_lock); 495 get_filesystem(type); 496 register_shrinker(&s->s_shrink); 497 return s; 498 } 499 500 EXPORT_SYMBOL(sget); 501 502 void drop_super(struct super_block *sb) 503 { 504 up_read(&sb->s_umount); 505 put_super(sb); 506 } 507 508 EXPORT_SYMBOL(drop_super); 509 510 /** 511 * iterate_supers - call function for all active superblocks 512 * @f: function to call 513 * @arg: argument to pass to it 514 * 515 * Scans the superblock list and calls given function, passing it 516 * locked superblock and given argument. 517 */ 518 void iterate_supers(void (*f)(struct super_block *, void *), void *arg) 519 { 520 struct super_block *sb, *p = NULL; 521 522 spin_lock(&sb_lock); 523 list_for_each_entry(sb, &super_blocks, s_list) { 524 if (hlist_unhashed(&sb->s_instances)) 525 continue; 526 sb->s_count++; 527 spin_unlock(&sb_lock); 528 529 down_read(&sb->s_umount); 530 if (sb->s_root && (sb->s_flags & MS_BORN)) 531 f(sb, arg); 532 up_read(&sb->s_umount); 533 534 spin_lock(&sb_lock); 535 if (p) 536 __put_super(p); 537 p = sb; 538 } 539 if (p) 540 __put_super(p); 541 spin_unlock(&sb_lock); 542 } 543 544 /** 545 * iterate_supers_type - call function for superblocks of given type 546 * @type: fs type 547 * @f: function to call 548 * @arg: argument to pass to it 549 * 550 * Scans the superblock list and calls given function, passing it 551 * locked superblock and given argument. 552 */ 553 void iterate_supers_type(struct file_system_type *type, 554 void (*f)(struct super_block *, void *), void *arg) 555 { 556 struct super_block *sb, *p = NULL; 557 struct hlist_node *node; 558 559 spin_lock(&sb_lock); 560 hlist_for_each_entry(sb, node, &type->fs_supers, s_instances) { 561 sb->s_count++; 562 spin_unlock(&sb_lock); 563 564 down_read(&sb->s_umount); 565 if (sb->s_root && (sb->s_flags & MS_BORN)) 566 f(sb, arg); 567 up_read(&sb->s_umount); 568 569 spin_lock(&sb_lock); 570 if (p) 571 __put_super(p); 572 p = sb; 573 } 574 if (p) 575 __put_super(p); 576 spin_unlock(&sb_lock); 577 } 578 579 EXPORT_SYMBOL(iterate_supers_type); 580 581 /** 582 * get_super - get the superblock of a device 583 * @bdev: device to get the superblock for 584 * 585 * Scans the superblock list and finds the superblock of the file system 586 * mounted on the device given. %NULL is returned if no match is found. 587 */ 588 589 struct super_block *get_super(struct block_device *bdev) 590 { 591 struct super_block *sb; 592 593 if (!bdev) 594 return NULL; 595 596 spin_lock(&sb_lock); 597 rescan: 598 list_for_each_entry(sb, &super_blocks, s_list) { 599 if (hlist_unhashed(&sb->s_instances)) 600 continue; 601 if (sb->s_bdev == bdev) { 602 sb->s_count++; 603 spin_unlock(&sb_lock); 604 down_read(&sb->s_umount); 605 /* still alive? */ 606 if (sb->s_root && (sb->s_flags & MS_BORN)) 607 return sb; 608 up_read(&sb->s_umount); 609 /* nope, got unmounted */ 610 spin_lock(&sb_lock); 611 __put_super(sb); 612 goto rescan; 613 } 614 } 615 spin_unlock(&sb_lock); 616 return NULL; 617 } 618 619 EXPORT_SYMBOL(get_super); 620 621 /** 622 * get_super_thawed - get thawed superblock of a device 623 * @bdev: device to get the superblock for 624 * 625 * Scans the superblock list and finds the superblock of the file system 626 * mounted on the device. The superblock is returned once it is thawed 627 * (or immediately if it was not frozen). %NULL is returned if no match 628 * is found. 629 */ 630 struct super_block *get_super_thawed(struct block_device *bdev) 631 { 632 while (1) { 633 struct super_block *s = get_super(bdev); 634 if (!s || s->s_writers.frozen == SB_UNFROZEN) 635 return s; 636 up_read(&s->s_umount); 637 wait_event(s->s_writers.wait_unfrozen, 638 s->s_writers.frozen == SB_UNFROZEN); 639 put_super(s); 640 } 641 } 642 EXPORT_SYMBOL(get_super_thawed); 643 644 /** 645 * get_active_super - get an active reference to the superblock of a device 646 * @bdev: device to get the superblock for 647 * 648 * Scans the superblock list and finds the superblock of the file system 649 * mounted on the device given. Returns the superblock with an active 650 * reference or %NULL if none was found. 651 */ 652 struct super_block *get_active_super(struct block_device *bdev) 653 { 654 struct super_block *sb; 655 656 if (!bdev) 657 return NULL; 658 659 restart: 660 spin_lock(&sb_lock); 661 list_for_each_entry(sb, &super_blocks, s_list) { 662 if (hlist_unhashed(&sb->s_instances)) 663 continue; 664 if (sb->s_bdev == bdev) { 665 if (grab_super(sb)) /* drops sb_lock */ 666 return sb; 667 else 668 goto restart; 669 } 670 } 671 spin_unlock(&sb_lock); 672 return NULL; 673 } 674 675 struct super_block *user_get_super(dev_t dev) 676 { 677 struct super_block *sb; 678 679 spin_lock(&sb_lock); 680 rescan: 681 list_for_each_entry(sb, &super_blocks, s_list) { 682 if (hlist_unhashed(&sb->s_instances)) 683 continue; 684 if (sb->s_dev == dev) { 685 sb->s_count++; 686 spin_unlock(&sb_lock); 687 down_read(&sb->s_umount); 688 /* still alive? */ 689 if (sb->s_root && (sb->s_flags & MS_BORN)) 690 return sb; 691 up_read(&sb->s_umount); 692 /* nope, got unmounted */ 693 spin_lock(&sb_lock); 694 __put_super(sb); 695 goto rescan; 696 } 697 } 698 spin_unlock(&sb_lock); 699 return NULL; 700 } 701 702 /** 703 * do_remount_sb - asks filesystem to change mount options. 704 * @sb: superblock in question 705 * @flags: numeric part of options 706 * @data: the rest of options 707 * @force: whether or not to force the change 708 * 709 * Alters the mount options of a mounted file system. 710 */ 711 int do_remount_sb(struct super_block *sb, int flags, void *data, int force) 712 { 713 int retval; 714 int remount_ro; 715 716 if (sb->s_writers.frozen != SB_UNFROZEN) 717 return -EBUSY; 718 719 #ifdef CONFIG_BLOCK 720 if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev)) 721 return -EACCES; 722 #endif 723 724 if (flags & MS_RDONLY) 725 acct_auto_close(sb); 726 shrink_dcache_sb(sb); 727 sync_filesystem(sb); 728 729 remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY); 730 731 /* If we are remounting RDONLY and current sb is read/write, 732 make sure there are no rw files opened */ 733 if (remount_ro) { 734 if (force) { 735 mark_files_ro(sb); 736 } else { 737 retval = sb_prepare_remount_readonly(sb); 738 if (retval) 739 return retval; 740 } 741 } 742 743 if (sb->s_op->remount_fs) { 744 retval = sb->s_op->remount_fs(sb, &flags, data); 745 if (retval) { 746 if (!force) 747 goto cancel_readonly; 748 /* If forced remount, go ahead despite any errors */ 749 WARN(1, "forced remount of a %s fs returned %i\n", 750 sb->s_type->name, retval); 751 } 752 } 753 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK); 754 /* Needs to be ordered wrt mnt_is_readonly() */ 755 smp_wmb(); 756 sb->s_readonly_remount = 0; 757 758 /* 759 * Some filesystems modify their metadata via some other path than the 760 * bdev buffer cache (eg. use a private mapping, or directories in 761 * pagecache, etc). Also file data modifications go via their own 762 * mappings. So If we try to mount readonly then copy the filesystem 763 * from bdev, we could get stale data, so invalidate it to give a best 764 * effort at coherency. 765 */ 766 if (remount_ro && sb->s_bdev) 767 invalidate_bdev(sb->s_bdev); 768 return 0; 769 770 cancel_readonly: 771 sb->s_readonly_remount = 0; 772 return retval; 773 } 774 775 static void do_emergency_remount(struct work_struct *work) 776 { 777 struct super_block *sb, *p = NULL; 778 779 spin_lock(&sb_lock); 780 list_for_each_entry(sb, &super_blocks, s_list) { 781 if (hlist_unhashed(&sb->s_instances)) 782 continue; 783 sb->s_count++; 784 spin_unlock(&sb_lock); 785 down_write(&sb->s_umount); 786 if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) && 787 !(sb->s_flags & MS_RDONLY)) { 788 /* 789 * What lock protects sb->s_flags?? 790 */ 791 do_remount_sb(sb, MS_RDONLY, NULL, 1); 792 } 793 up_write(&sb->s_umount); 794 spin_lock(&sb_lock); 795 if (p) 796 __put_super(p); 797 p = sb; 798 } 799 if (p) 800 __put_super(p); 801 spin_unlock(&sb_lock); 802 kfree(work); 803 printk("Emergency Remount complete\n"); 804 } 805 806 void emergency_remount(void) 807 { 808 struct work_struct *work; 809 810 work = kmalloc(sizeof(*work), GFP_ATOMIC); 811 if (work) { 812 INIT_WORK(work, do_emergency_remount); 813 schedule_work(work); 814 } 815 } 816 817 /* 818 * Unnamed block devices are dummy devices used by virtual 819 * filesystems which don't use real block-devices. -- jrs 820 */ 821 822 static DEFINE_IDA(unnamed_dev_ida); 823 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */ 824 static int unnamed_dev_start = 0; /* don't bother trying below it */ 825 826 int get_anon_bdev(dev_t *p) 827 { 828 int dev; 829 int error; 830 831 retry: 832 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0) 833 return -ENOMEM; 834 spin_lock(&unnamed_dev_lock); 835 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev); 836 if (!error) 837 unnamed_dev_start = dev + 1; 838 spin_unlock(&unnamed_dev_lock); 839 if (error == -EAGAIN) 840 /* We raced and lost with another CPU. */ 841 goto retry; 842 else if (error) 843 return -EAGAIN; 844 845 if ((dev & MAX_IDR_MASK) == (1 << MINORBITS)) { 846 spin_lock(&unnamed_dev_lock); 847 ida_remove(&unnamed_dev_ida, dev); 848 if (unnamed_dev_start > dev) 849 unnamed_dev_start = dev; 850 spin_unlock(&unnamed_dev_lock); 851 return -EMFILE; 852 } 853 *p = MKDEV(0, dev & MINORMASK); 854 return 0; 855 } 856 EXPORT_SYMBOL(get_anon_bdev); 857 858 void free_anon_bdev(dev_t dev) 859 { 860 int slot = MINOR(dev); 861 spin_lock(&unnamed_dev_lock); 862 ida_remove(&unnamed_dev_ida, slot); 863 if (slot < unnamed_dev_start) 864 unnamed_dev_start = slot; 865 spin_unlock(&unnamed_dev_lock); 866 } 867 EXPORT_SYMBOL(free_anon_bdev); 868 869 int set_anon_super(struct super_block *s, void *data) 870 { 871 int error = get_anon_bdev(&s->s_dev); 872 if (!error) 873 s->s_bdi = &noop_backing_dev_info; 874 return error; 875 } 876 877 EXPORT_SYMBOL(set_anon_super); 878 879 void kill_anon_super(struct super_block *sb) 880 { 881 dev_t dev = sb->s_dev; 882 generic_shutdown_super(sb); 883 free_anon_bdev(dev); 884 } 885 886 EXPORT_SYMBOL(kill_anon_super); 887 888 void kill_litter_super(struct super_block *sb) 889 { 890 if (sb->s_root) 891 d_genocide(sb->s_root); 892 kill_anon_super(sb); 893 } 894 895 EXPORT_SYMBOL(kill_litter_super); 896 897 static int ns_test_super(struct super_block *sb, void *data) 898 { 899 return sb->s_fs_info == data; 900 } 901 902 static int ns_set_super(struct super_block *sb, void *data) 903 { 904 sb->s_fs_info = data; 905 return set_anon_super(sb, NULL); 906 } 907 908 struct dentry *mount_ns(struct file_system_type *fs_type, int flags, 909 void *data, int (*fill_super)(struct super_block *, void *, int)) 910 { 911 struct super_block *sb; 912 913 sb = sget(fs_type, ns_test_super, ns_set_super, flags, data); 914 if (IS_ERR(sb)) 915 return ERR_CAST(sb); 916 917 if (!sb->s_root) { 918 int err; 919 err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0); 920 if (err) { 921 deactivate_locked_super(sb); 922 return ERR_PTR(err); 923 } 924 925 sb->s_flags |= MS_ACTIVE; 926 } 927 928 return dget(sb->s_root); 929 } 930 931 EXPORT_SYMBOL(mount_ns); 932 933 #ifdef CONFIG_BLOCK 934 static int set_bdev_super(struct super_block *s, void *data) 935 { 936 s->s_bdev = data; 937 s->s_dev = s->s_bdev->bd_dev; 938 939 /* 940 * We set the bdi here to the queue backing, file systems can 941 * overwrite this in ->fill_super() 942 */ 943 s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info; 944 return 0; 945 } 946 947 static int test_bdev_super(struct super_block *s, void *data) 948 { 949 return (void *)s->s_bdev == data; 950 } 951 952 struct dentry *mount_bdev(struct file_system_type *fs_type, 953 int flags, const char *dev_name, void *data, 954 int (*fill_super)(struct super_block *, void *, int)) 955 { 956 struct block_device *bdev; 957 struct super_block *s; 958 fmode_t mode = FMODE_READ | FMODE_EXCL; 959 int error = 0; 960 961 if (!(flags & MS_RDONLY)) 962 mode |= FMODE_WRITE; 963 964 bdev = blkdev_get_by_path(dev_name, mode, fs_type); 965 if (IS_ERR(bdev)) 966 return ERR_CAST(bdev); 967 968 /* 969 * once the super is inserted into the list by sget, s_umount 970 * will protect the lockfs code from trying to start a snapshot 971 * while we are mounting 972 */ 973 mutex_lock(&bdev->bd_fsfreeze_mutex); 974 if (bdev->bd_fsfreeze_count > 0) { 975 mutex_unlock(&bdev->bd_fsfreeze_mutex); 976 error = -EBUSY; 977 goto error_bdev; 978 } 979 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | MS_NOSEC, 980 bdev); 981 mutex_unlock(&bdev->bd_fsfreeze_mutex); 982 if (IS_ERR(s)) 983 goto error_s; 984 985 if (s->s_root) { 986 if ((flags ^ s->s_flags) & MS_RDONLY) { 987 deactivate_locked_super(s); 988 error = -EBUSY; 989 goto error_bdev; 990 } 991 992 /* 993 * s_umount nests inside bd_mutex during 994 * __invalidate_device(). blkdev_put() acquires 995 * bd_mutex and can't be called under s_umount. Drop 996 * s_umount temporarily. This is safe as we're 997 * holding an active reference. 998 */ 999 up_write(&s->s_umount); 1000 blkdev_put(bdev, mode); 1001 down_write(&s->s_umount); 1002 } else { 1003 char b[BDEVNAME_SIZE]; 1004 1005 s->s_mode = mode; 1006 strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id)); 1007 sb_set_blocksize(s, block_size(bdev)); 1008 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0); 1009 if (error) { 1010 deactivate_locked_super(s); 1011 goto error; 1012 } 1013 1014 s->s_flags |= MS_ACTIVE; 1015 bdev->bd_super = s; 1016 } 1017 1018 return dget(s->s_root); 1019 1020 error_s: 1021 error = PTR_ERR(s); 1022 error_bdev: 1023 blkdev_put(bdev, mode); 1024 error: 1025 return ERR_PTR(error); 1026 } 1027 EXPORT_SYMBOL(mount_bdev); 1028 1029 void kill_block_super(struct super_block *sb) 1030 { 1031 struct block_device *bdev = sb->s_bdev; 1032 fmode_t mode = sb->s_mode; 1033 1034 bdev->bd_super = NULL; 1035 generic_shutdown_super(sb); 1036 sync_blockdev(bdev); 1037 WARN_ON_ONCE(!(mode & FMODE_EXCL)); 1038 blkdev_put(bdev, mode | FMODE_EXCL); 1039 } 1040 1041 EXPORT_SYMBOL(kill_block_super); 1042 #endif 1043 1044 struct dentry *mount_nodev(struct file_system_type *fs_type, 1045 int flags, void *data, 1046 int (*fill_super)(struct super_block *, void *, int)) 1047 { 1048 int error; 1049 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL); 1050 1051 if (IS_ERR(s)) 1052 return ERR_CAST(s); 1053 1054 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0); 1055 if (error) { 1056 deactivate_locked_super(s); 1057 return ERR_PTR(error); 1058 } 1059 s->s_flags |= MS_ACTIVE; 1060 return dget(s->s_root); 1061 } 1062 EXPORT_SYMBOL(mount_nodev); 1063 1064 static int compare_single(struct super_block *s, void *p) 1065 { 1066 return 1; 1067 } 1068 1069 struct dentry *mount_single(struct file_system_type *fs_type, 1070 int flags, void *data, 1071 int (*fill_super)(struct super_block *, void *, int)) 1072 { 1073 struct super_block *s; 1074 int error; 1075 1076 s = sget(fs_type, compare_single, set_anon_super, flags, NULL); 1077 if (IS_ERR(s)) 1078 return ERR_CAST(s); 1079 if (!s->s_root) { 1080 error = fill_super(s, data, flags & MS_SILENT ? 1 : 0); 1081 if (error) { 1082 deactivate_locked_super(s); 1083 return ERR_PTR(error); 1084 } 1085 s->s_flags |= MS_ACTIVE; 1086 } else { 1087 do_remount_sb(s, flags, data, 0); 1088 } 1089 return dget(s->s_root); 1090 } 1091 EXPORT_SYMBOL(mount_single); 1092 1093 struct dentry * 1094 mount_fs(struct file_system_type *type, int flags, const char *name, void *data) 1095 { 1096 struct dentry *root; 1097 struct super_block *sb; 1098 char *secdata = NULL; 1099 int error = -ENOMEM; 1100 1101 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) { 1102 secdata = alloc_secdata(); 1103 if (!secdata) 1104 goto out; 1105 1106 error = security_sb_copy_data(data, secdata); 1107 if (error) 1108 goto out_free_secdata; 1109 } 1110 1111 root = type->mount(type, flags, name, data); 1112 if (IS_ERR(root)) { 1113 error = PTR_ERR(root); 1114 goto out_free_secdata; 1115 } 1116 sb = root->d_sb; 1117 BUG_ON(!sb); 1118 WARN_ON(!sb->s_bdi); 1119 WARN_ON(sb->s_bdi == &default_backing_dev_info); 1120 sb->s_flags |= MS_BORN; 1121 1122 error = security_sb_kern_mount(sb, flags, secdata); 1123 if (error) 1124 goto out_sb; 1125 1126 /* 1127 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE 1128 * but s_maxbytes was an unsigned long long for many releases. Throw 1129 * this warning for a little while to try and catch filesystems that 1130 * violate this rule. 1131 */ 1132 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to " 1133 "negative value (%lld)\n", type->name, sb->s_maxbytes); 1134 1135 up_write(&sb->s_umount); 1136 free_secdata(secdata); 1137 return root; 1138 out_sb: 1139 dput(root); 1140 deactivate_locked_super(sb); 1141 out_free_secdata: 1142 free_secdata(secdata); 1143 out: 1144 return ERR_PTR(error); 1145 } 1146 1147 /* 1148 * This is an internal function, please use sb_end_{write,pagefault,intwrite} 1149 * instead. 1150 */ 1151 void __sb_end_write(struct super_block *sb, int level) 1152 { 1153 percpu_counter_dec(&sb->s_writers.counter[level-1]); 1154 /* 1155 * Make sure s_writers are updated before we wake up waiters in 1156 * freeze_super(). 1157 */ 1158 smp_mb(); 1159 if (waitqueue_active(&sb->s_writers.wait)) 1160 wake_up(&sb->s_writers.wait); 1161 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _RET_IP_); 1162 } 1163 EXPORT_SYMBOL(__sb_end_write); 1164 1165 #ifdef CONFIG_LOCKDEP 1166 /* 1167 * We want lockdep to tell us about possible deadlocks with freezing but 1168 * it's it bit tricky to properly instrument it. Getting a freeze protection 1169 * works as getting a read lock but there are subtle problems. XFS for example 1170 * gets freeze protection on internal level twice in some cases, which is OK 1171 * only because we already hold a freeze protection also on higher level. Due 1172 * to these cases we have to tell lockdep we are doing trylock when we 1173 * already hold a freeze protection for a higher freeze level. 1174 */ 1175 static void acquire_freeze_lock(struct super_block *sb, int level, bool trylock, 1176 unsigned long ip) 1177 { 1178 int i; 1179 1180 if (!trylock) { 1181 for (i = 0; i < level - 1; i++) 1182 if (lock_is_held(&sb->s_writers.lock_map[i])) { 1183 trylock = true; 1184 break; 1185 } 1186 } 1187 rwsem_acquire_read(&sb->s_writers.lock_map[level-1], 0, trylock, ip); 1188 } 1189 #endif 1190 1191 /* 1192 * This is an internal function, please use sb_start_{write,pagefault,intwrite} 1193 * instead. 1194 */ 1195 int __sb_start_write(struct super_block *sb, int level, bool wait) 1196 { 1197 retry: 1198 if (unlikely(sb->s_writers.frozen >= level)) { 1199 if (!wait) 1200 return 0; 1201 wait_event(sb->s_writers.wait_unfrozen, 1202 sb->s_writers.frozen < level); 1203 } 1204 1205 #ifdef CONFIG_LOCKDEP 1206 acquire_freeze_lock(sb, level, !wait, _RET_IP_); 1207 #endif 1208 percpu_counter_inc(&sb->s_writers.counter[level-1]); 1209 /* 1210 * Make sure counter is updated before we check for frozen. 1211 * freeze_super() first sets frozen and then checks the counter. 1212 */ 1213 smp_mb(); 1214 if (unlikely(sb->s_writers.frozen >= level)) { 1215 __sb_end_write(sb, level); 1216 goto retry; 1217 } 1218 return 1; 1219 } 1220 EXPORT_SYMBOL(__sb_start_write); 1221 1222 /** 1223 * sb_wait_write - wait until all writers to given file system finish 1224 * @sb: the super for which we wait 1225 * @level: type of writers we wait for (normal vs page fault) 1226 * 1227 * This function waits until there are no writers of given type to given file 1228 * system. Caller of this function should make sure there can be no new writers 1229 * of type @level before calling this function. Otherwise this function can 1230 * livelock. 1231 */ 1232 static void sb_wait_write(struct super_block *sb, int level) 1233 { 1234 s64 writers; 1235 1236 /* 1237 * We just cycle-through lockdep here so that it does not complain 1238 * about returning with lock to userspace 1239 */ 1240 rwsem_acquire(&sb->s_writers.lock_map[level-1], 0, 0, _THIS_IP_); 1241 rwsem_release(&sb->s_writers.lock_map[level-1], 1, _THIS_IP_); 1242 1243 do { 1244 DEFINE_WAIT(wait); 1245 1246 /* 1247 * We use a barrier in prepare_to_wait() to separate setting 1248 * of frozen and checking of the counter 1249 */ 1250 prepare_to_wait(&sb->s_writers.wait, &wait, 1251 TASK_UNINTERRUPTIBLE); 1252 1253 writers = percpu_counter_sum(&sb->s_writers.counter[level-1]); 1254 if (writers) 1255 schedule(); 1256 1257 finish_wait(&sb->s_writers.wait, &wait); 1258 } while (writers); 1259 } 1260 1261 /** 1262 * freeze_super - lock the filesystem and force it into a consistent state 1263 * @sb: the super to lock 1264 * 1265 * Syncs the super to make sure the filesystem is consistent and calls the fs's 1266 * freeze_fs. Subsequent calls to this without first thawing the fs will return 1267 * -EBUSY. 1268 * 1269 * During this function, sb->s_writers.frozen goes through these values: 1270 * 1271 * SB_UNFROZEN: File system is normal, all writes progress as usual. 1272 * 1273 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New 1274 * writes should be blocked, though page faults are still allowed. We wait for 1275 * all writes to complete and then proceed to the next stage. 1276 * 1277 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked 1278 * but internal fs threads can still modify the filesystem (although they 1279 * should not dirty new pages or inodes), writeback can run etc. After waiting 1280 * for all running page faults we sync the filesystem which will clean all 1281 * dirty pages and inodes (no new dirty pages or inodes can be created when 1282 * sync is running). 1283 * 1284 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs 1285 * modification are blocked (e.g. XFS preallocation truncation on inode 1286 * reclaim). This is usually implemented by blocking new transactions for 1287 * filesystems that have them and need this additional guard. After all 1288 * internal writers are finished we call ->freeze_fs() to finish filesystem 1289 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is 1290 * mostly auxiliary for filesystems to verify they do not modify frozen fs. 1291 * 1292 * sb->s_writers.frozen is protected by sb->s_umount. 1293 */ 1294 int freeze_super(struct super_block *sb) 1295 { 1296 int ret; 1297 1298 atomic_inc(&sb->s_active); 1299 down_write(&sb->s_umount); 1300 if (sb->s_writers.frozen != SB_UNFROZEN) { 1301 deactivate_locked_super(sb); 1302 return -EBUSY; 1303 } 1304 1305 if (!(sb->s_flags & MS_BORN)) { 1306 up_write(&sb->s_umount); 1307 return 0; /* sic - it's "nothing to do" */ 1308 } 1309 1310 if (sb->s_flags & MS_RDONLY) { 1311 /* Nothing to do really... */ 1312 sb->s_writers.frozen = SB_FREEZE_COMPLETE; 1313 up_write(&sb->s_umount); 1314 return 0; 1315 } 1316 1317 /* From now on, no new normal writers can start */ 1318 sb->s_writers.frozen = SB_FREEZE_WRITE; 1319 smp_wmb(); 1320 1321 /* Release s_umount to preserve sb_start_write -> s_umount ordering */ 1322 up_write(&sb->s_umount); 1323 1324 sb_wait_write(sb, SB_FREEZE_WRITE); 1325 1326 /* Now we go and block page faults... */ 1327 down_write(&sb->s_umount); 1328 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT; 1329 smp_wmb(); 1330 1331 sb_wait_write(sb, SB_FREEZE_PAGEFAULT); 1332 1333 /* All writers are done so after syncing there won't be dirty data */ 1334 sync_filesystem(sb); 1335 1336 /* Now wait for internal filesystem counter */ 1337 sb->s_writers.frozen = SB_FREEZE_FS; 1338 smp_wmb(); 1339 sb_wait_write(sb, SB_FREEZE_FS); 1340 1341 if (sb->s_op->freeze_fs) { 1342 ret = sb->s_op->freeze_fs(sb); 1343 if (ret) { 1344 printk(KERN_ERR 1345 "VFS:Filesystem freeze failed\n"); 1346 sb->s_writers.frozen = SB_UNFROZEN; 1347 smp_wmb(); 1348 wake_up(&sb->s_writers.wait_unfrozen); 1349 deactivate_locked_super(sb); 1350 return ret; 1351 } 1352 } 1353 /* 1354 * This is just for debugging purposes so that fs can warn if it 1355 * sees write activity when frozen is set to SB_FREEZE_COMPLETE. 1356 */ 1357 sb->s_writers.frozen = SB_FREEZE_COMPLETE; 1358 up_write(&sb->s_umount); 1359 return 0; 1360 } 1361 EXPORT_SYMBOL(freeze_super); 1362 1363 /** 1364 * thaw_super -- unlock filesystem 1365 * @sb: the super to thaw 1366 * 1367 * Unlocks the filesystem and marks it writeable again after freeze_super(). 1368 */ 1369 int thaw_super(struct super_block *sb) 1370 { 1371 int error; 1372 1373 down_write(&sb->s_umount); 1374 if (sb->s_writers.frozen == SB_UNFROZEN) { 1375 up_write(&sb->s_umount); 1376 return -EINVAL; 1377 } 1378 1379 if (sb->s_flags & MS_RDONLY) 1380 goto out; 1381 1382 if (sb->s_op->unfreeze_fs) { 1383 error = sb->s_op->unfreeze_fs(sb); 1384 if (error) { 1385 printk(KERN_ERR 1386 "VFS:Filesystem thaw failed\n"); 1387 up_write(&sb->s_umount); 1388 return error; 1389 } 1390 } 1391 1392 out: 1393 sb->s_writers.frozen = SB_UNFROZEN; 1394 smp_wmb(); 1395 wake_up(&sb->s_writers.wait_unfrozen); 1396 deactivate_locked_super(sb); 1397 1398 return 0; 1399 } 1400 EXPORT_SYMBOL(thaw_super); 1401