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