1 // SPDX-License-Identifier: GPL-2.0 2 /* 3 * linux/fs/super.c 4 * 5 * Copyright (C) 1991, 1992 Linus Torvalds 6 * 7 * super.c contains code to handle: - mount structures 8 * - super-block tables 9 * - filesystem drivers list 10 * - mount system call 11 * - umount system call 12 * - ustat system call 13 * 14 * GK 2/5/95 - Changed to support mounting the root fs via NFS 15 * 16 * Added kerneld support: Jacques Gelinas and Bjorn Ekwall 17 * Added change_root: Werner Almesberger & Hans Lermen, Feb '96 18 * Added options to /proc/mounts: 19 * Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996. 20 * Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998 21 * Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000 22 */ 23 24 #include <linux/export.h> 25 #include <linux/slab.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 <linux/user_namespace.h> 38 #include "internal.h" 39 40 41 static LIST_HEAD(super_blocks); 42 static DEFINE_SPINLOCK(sb_lock); 43 44 static char *sb_writers_name[SB_FREEZE_LEVELS] = { 45 "sb_writers", 46 "sb_pagefaults", 47 "sb_internal", 48 }; 49 50 /* 51 * One thing we have to be careful of with a per-sb shrinker is that we don't 52 * drop the last active reference to the superblock from within the shrinker. 53 * If that happens we could trigger unregistering the shrinker from within the 54 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we 55 * take a passive reference to the superblock to avoid this from occurring. 56 */ 57 static unsigned long super_cache_scan(struct shrinker *shrink, 58 struct shrink_control *sc) 59 { 60 struct super_block *sb; 61 long fs_objects = 0; 62 long total_objects; 63 long freed = 0; 64 long dentries; 65 long inodes; 66 67 sb = container_of(shrink, struct super_block, s_shrink); 68 69 /* 70 * Deadlock avoidance. We may hold various FS locks, and we don't want 71 * to recurse into the FS that called us in clear_inode() and friends.. 72 */ 73 if (!(sc->gfp_mask & __GFP_FS)) 74 return SHRINK_STOP; 75 76 if (!trylock_super(sb)) 77 return SHRINK_STOP; 78 79 if (sb->s_op->nr_cached_objects) 80 fs_objects = sb->s_op->nr_cached_objects(sb, sc); 81 82 inodes = list_lru_shrink_count(&sb->s_inode_lru, sc); 83 dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc); 84 total_objects = dentries + inodes + fs_objects + 1; 85 if (!total_objects) 86 total_objects = 1; 87 88 /* proportion the scan between the caches */ 89 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects); 90 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects); 91 fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects); 92 93 /* 94 * prune the dcache first as the icache is pinned by it, then 95 * prune the icache, followed by the filesystem specific caches 96 * 97 * Ensure that we always scan at least one object - memcg kmem 98 * accounting uses this to fully empty the caches. 99 */ 100 sc->nr_to_scan = dentries + 1; 101 freed = prune_dcache_sb(sb, sc); 102 sc->nr_to_scan = inodes + 1; 103 freed += prune_icache_sb(sb, sc); 104 105 if (fs_objects) { 106 sc->nr_to_scan = fs_objects + 1; 107 freed += sb->s_op->free_cached_objects(sb, sc); 108 } 109 110 up_read(&sb->s_umount); 111 return freed; 112 } 113 114 static unsigned long super_cache_count(struct shrinker *shrink, 115 struct shrink_control *sc) 116 { 117 struct super_block *sb; 118 long total_objects = 0; 119 120 sb = container_of(shrink, struct super_block, s_shrink); 121 122 /* 123 * Don't call trylock_super as it is a potential 124 * scalability bottleneck. The counts could get updated 125 * between super_cache_count and super_cache_scan anyway. 126 * Call to super_cache_count with shrinker_rwsem held 127 * ensures the safety of call to list_lru_shrink_count() and 128 * s_op->nr_cached_objects(). 129 */ 130 if (sb->s_op && sb->s_op->nr_cached_objects) 131 total_objects = sb->s_op->nr_cached_objects(sb, sc); 132 133 total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc); 134 total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc); 135 136 total_objects = vfs_pressure_ratio(total_objects); 137 return total_objects; 138 } 139 140 static void destroy_super_work(struct work_struct *work) 141 { 142 struct super_block *s = container_of(work, struct super_block, 143 destroy_work); 144 int i; 145 146 for (i = 0; i < SB_FREEZE_LEVELS; i++) 147 percpu_free_rwsem(&s->s_writers.rw_sem[i]); 148 kfree(s); 149 } 150 151 static void destroy_super_rcu(struct rcu_head *head) 152 { 153 struct super_block *s = container_of(head, struct super_block, rcu); 154 INIT_WORK(&s->destroy_work, destroy_super_work); 155 schedule_work(&s->destroy_work); 156 } 157 158 /** 159 * destroy_super - frees a superblock 160 * @s: superblock to free 161 * 162 * Frees a superblock. 163 */ 164 static void destroy_super(struct super_block *s) 165 { 166 list_lru_destroy(&s->s_dentry_lru); 167 list_lru_destroy(&s->s_inode_lru); 168 security_sb_free(s); 169 WARN_ON(!list_empty(&s->s_mounts)); 170 put_user_ns(s->s_user_ns); 171 kfree(s->s_subtype); 172 call_rcu(&s->rcu, destroy_super_rcu); 173 } 174 175 /** 176 * alloc_super - create new superblock 177 * @type: filesystem type superblock should belong to 178 * @flags: the mount flags 179 * @user_ns: User namespace for the super_block 180 * 181 * Allocates and initializes a new &struct super_block. alloc_super() 182 * returns a pointer new superblock or %NULL if allocation had failed. 183 */ 184 static struct super_block *alloc_super(struct file_system_type *type, int flags, 185 struct user_namespace *user_ns) 186 { 187 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER); 188 static const struct super_operations default_op; 189 int i; 190 191 if (!s) 192 return NULL; 193 194 INIT_LIST_HEAD(&s->s_mounts); 195 s->s_user_ns = get_user_ns(user_ns); 196 197 if (security_sb_alloc(s)) 198 goto fail; 199 200 for (i = 0; i < SB_FREEZE_LEVELS; i++) { 201 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i], 202 sb_writers_name[i], 203 &type->s_writers_key[i])) 204 goto fail; 205 } 206 init_waitqueue_head(&s->s_writers.wait_unfrozen); 207 s->s_bdi = &noop_backing_dev_info; 208 s->s_flags = flags; 209 if (s->s_user_ns != &init_user_ns) 210 s->s_iflags |= SB_I_NODEV; 211 INIT_HLIST_NODE(&s->s_instances); 212 INIT_HLIST_BL_HEAD(&s->s_anon); 213 mutex_init(&s->s_sync_lock); 214 INIT_LIST_HEAD(&s->s_inodes); 215 spin_lock_init(&s->s_inode_list_lock); 216 INIT_LIST_HEAD(&s->s_inodes_wb); 217 spin_lock_init(&s->s_inode_wblist_lock); 218 219 if (list_lru_init_memcg(&s->s_dentry_lru)) 220 goto fail; 221 if (list_lru_init_memcg(&s->s_inode_lru)) 222 goto fail; 223 224 init_rwsem(&s->s_umount); 225 lockdep_set_class(&s->s_umount, &type->s_umount_key); 226 /* 227 * sget() can have s_umount recursion. 228 * 229 * When it cannot find a suitable sb, it allocates a new 230 * one (this one), and tries again to find a suitable old 231 * one. 232 * 233 * In case that succeeds, it will acquire the s_umount 234 * lock of the old one. Since these are clearly distrinct 235 * locks, and this object isn't exposed yet, there's no 236 * risk of deadlocks. 237 * 238 * Annotate this by putting this lock in a different 239 * subclass. 240 */ 241 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING); 242 s->s_count = 1; 243 atomic_set(&s->s_active, 1); 244 mutex_init(&s->s_vfs_rename_mutex); 245 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key); 246 init_rwsem(&s->s_dquot.dqio_sem); 247 s->s_maxbytes = MAX_NON_LFS; 248 s->s_op = &default_op; 249 s->s_time_gran = 1000000000; 250 s->cleancache_poolid = CLEANCACHE_NO_POOL; 251 252 s->s_shrink.seeks = DEFAULT_SEEKS; 253 s->s_shrink.scan_objects = super_cache_scan; 254 s->s_shrink.count_objects = super_cache_count; 255 s->s_shrink.batch = 1024; 256 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE; 257 return s; 258 259 fail: 260 destroy_super(s); 261 return NULL; 262 } 263 264 /* Superblock refcounting */ 265 266 /* 267 * Drop a superblock's refcount. The caller must hold sb_lock. 268 */ 269 static void __put_super(struct super_block *sb) 270 { 271 if (!--sb->s_count) { 272 list_del_init(&sb->s_list); 273 destroy_super(sb); 274 } 275 } 276 277 /** 278 * put_super - drop a temporary reference to superblock 279 * @sb: superblock in question 280 * 281 * Drops a temporary reference, frees superblock if there's no 282 * references left. 283 */ 284 static void put_super(struct super_block *sb) 285 { 286 spin_lock(&sb_lock); 287 __put_super(sb); 288 spin_unlock(&sb_lock); 289 } 290 291 292 /** 293 * deactivate_locked_super - drop an active reference to superblock 294 * @s: superblock to deactivate 295 * 296 * Drops an active reference to superblock, converting it into a temporary 297 * one if there is no other active references left. In that case we 298 * tell fs driver to shut it down and drop the temporary reference we 299 * had just acquired. 300 * 301 * Caller holds exclusive lock on superblock; that lock is released. 302 */ 303 void deactivate_locked_super(struct super_block *s) 304 { 305 struct file_system_type *fs = s->s_type; 306 if (atomic_dec_and_test(&s->s_active)) { 307 cleancache_invalidate_fs(s); 308 unregister_shrinker(&s->s_shrink); 309 fs->kill_sb(s); 310 311 /* 312 * Since list_lru_destroy() may sleep, we cannot call it from 313 * put_super(), where we hold the sb_lock. Therefore we destroy 314 * the lru lists right now. 315 */ 316 list_lru_destroy(&s->s_dentry_lru); 317 list_lru_destroy(&s->s_inode_lru); 318 319 put_filesystem(fs); 320 put_super(s); 321 } else { 322 up_write(&s->s_umount); 323 } 324 } 325 326 EXPORT_SYMBOL(deactivate_locked_super); 327 328 /** 329 * deactivate_super - drop an active reference to superblock 330 * @s: superblock to deactivate 331 * 332 * Variant of deactivate_locked_super(), except that superblock is *not* 333 * locked by caller. If we are going to drop the final active reference, 334 * lock will be acquired prior to that. 335 */ 336 void deactivate_super(struct super_block *s) 337 { 338 if (!atomic_add_unless(&s->s_active, -1, 1)) { 339 down_write(&s->s_umount); 340 deactivate_locked_super(s); 341 } 342 } 343 344 EXPORT_SYMBOL(deactivate_super); 345 346 /** 347 * grab_super - acquire an active reference 348 * @s: reference we are trying to make active 349 * 350 * Tries to acquire an active reference. grab_super() is used when we 351 * had just found a superblock in super_blocks or fs_type->fs_supers 352 * and want to turn it into a full-blown active reference. grab_super() 353 * is called with sb_lock held and drops it. Returns 1 in case of 354 * success, 0 if we had failed (superblock contents was already dead or 355 * dying when grab_super() had been called). Note that this is only 356 * called for superblocks not in rundown mode (== ones still on ->fs_supers 357 * of their type), so increment of ->s_count is OK here. 358 */ 359 static int grab_super(struct super_block *s) __releases(sb_lock) 360 { 361 s->s_count++; 362 spin_unlock(&sb_lock); 363 down_write(&s->s_umount); 364 if ((s->s_flags & SB_BORN) && atomic_inc_not_zero(&s->s_active)) { 365 put_super(s); 366 return 1; 367 } 368 up_write(&s->s_umount); 369 put_super(s); 370 return 0; 371 } 372 373 /* 374 * trylock_super - try to grab ->s_umount shared 375 * @sb: reference we are trying to grab 376 * 377 * Try to prevent fs shutdown. This is used in places where we 378 * cannot take an active reference but we need to ensure that the 379 * filesystem is not shut down while we are working on it. It returns 380 * false if we cannot acquire s_umount or if we lose the race and 381 * filesystem already got into shutdown, and returns true with the s_umount 382 * lock held in read mode in case of success. On successful return, 383 * the caller must drop the s_umount lock when done. 384 * 385 * Note that unlike get_super() et.al. this one does *not* bump ->s_count. 386 * The reason why it's safe is that we are OK with doing trylock instead 387 * of down_read(). There's a couple of places that are OK with that, but 388 * it's very much not a general-purpose interface. 389 */ 390 bool trylock_super(struct super_block *sb) 391 { 392 if (down_read_trylock(&sb->s_umount)) { 393 if (!hlist_unhashed(&sb->s_instances) && 394 sb->s_root && (sb->s_flags & SB_BORN)) 395 return true; 396 up_read(&sb->s_umount); 397 } 398 399 return false; 400 } 401 402 /** 403 * generic_shutdown_super - common helper for ->kill_sb() 404 * @sb: superblock to kill 405 * 406 * generic_shutdown_super() does all fs-independent work on superblock 407 * shutdown. Typical ->kill_sb() should pick all fs-specific objects 408 * that need destruction out of superblock, call generic_shutdown_super() 409 * and release aforementioned objects. Note: dentries and inodes _are_ 410 * taken care of and do not need specific handling. 411 * 412 * Upon calling this function, the filesystem may no longer alter or 413 * rearrange the set of dentries belonging to this super_block, nor may it 414 * change the attachments of dentries to inodes. 415 */ 416 void generic_shutdown_super(struct super_block *sb) 417 { 418 const struct super_operations *sop = sb->s_op; 419 420 if (sb->s_root) { 421 shrink_dcache_for_umount(sb); 422 sync_filesystem(sb); 423 sb->s_flags &= ~SB_ACTIVE; 424 425 fsnotify_unmount_inodes(sb); 426 cgroup_writeback_umount(); 427 428 evict_inodes(sb); 429 430 if (sb->s_dio_done_wq) { 431 destroy_workqueue(sb->s_dio_done_wq); 432 sb->s_dio_done_wq = NULL; 433 } 434 435 if (sop->put_super) 436 sop->put_super(sb); 437 438 if (!list_empty(&sb->s_inodes)) { 439 printk("VFS: Busy inodes after unmount of %s. " 440 "Self-destruct in 5 seconds. Have a nice day...\n", 441 sb->s_id); 442 } 443 } 444 spin_lock(&sb_lock); 445 /* should be initialized for __put_super_and_need_restart() */ 446 hlist_del_init(&sb->s_instances); 447 spin_unlock(&sb_lock); 448 up_write(&sb->s_umount); 449 if (sb->s_bdi != &noop_backing_dev_info) { 450 bdi_put(sb->s_bdi); 451 sb->s_bdi = &noop_backing_dev_info; 452 } 453 } 454 455 EXPORT_SYMBOL(generic_shutdown_super); 456 457 /** 458 * sget_userns - find or create a superblock 459 * @type: filesystem type superblock should belong to 460 * @test: comparison callback 461 * @set: setup callback 462 * @flags: mount flags 463 * @user_ns: User namespace for the super_block 464 * @data: argument to each of them 465 */ 466 struct super_block *sget_userns(struct file_system_type *type, 467 int (*test)(struct super_block *,void *), 468 int (*set)(struct super_block *,void *), 469 int flags, struct user_namespace *user_ns, 470 void *data) 471 { 472 struct super_block *s = NULL; 473 struct super_block *old; 474 int err; 475 476 if (!(flags & (SB_KERNMOUNT|SB_SUBMOUNT)) && 477 !(type->fs_flags & FS_USERNS_MOUNT) && 478 !capable(CAP_SYS_ADMIN)) 479 return ERR_PTR(-EPERM); 480 retry: 481 spin_lock(&sb_lock); 482 if (test) { 483 hlist_for_each_entry(old, &type->fs_supers, s_instances) { 484 if (!test(old, data)) 485 continue; 486 if (user_ns != old->s_user_ns) { 487 spin_unlock(&sb_lock); 488 if (s) { 489 up_write(&s->s_umount); 490 destroy_super(s); 491 } 492 return ERR_PTR(-EBUSY); 493 } 494 if (!grab_super(old)) 495 goto retry; 496 if (s) { 497 up_write(&s->s_umount); 498 destroy_super(s); 499 s = NULL; 500 } 501 return old; 502 } 503 } 504 if (!s) { 505 spin_unlock(&sb_lock); 506 s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns); 507 if (!s) 508 return ERR_PTR(-ENOMEM); 509 goto retry; 510 } 511 512 err = set(s, data); 513 if (err) { 514 spin_unlock(&sb_lock); 515 up_write(&s->s_umount); 516 destroy_super(s); 517 return ERR_PTR(err); 518 } 519 s->s_type = type; 520 strlcpy(s->s_id, type->name, sizeof(s->s_id)); 521 list_add_tail(&s->s_list, &super_blocks); 522 hlist_add_head(&s->s_instances, &type->fs_supers); 523 spin_unlock(&sb_lock); 524 get_filesystem(type); 525 register_shrinker(&s->s_shrink); 526 return s; 527 } 528 529 EXPORT_SYMBOL(sget_userns); 530 531 /** 532 * sget - find or create a superblock 533 * @type: filesystem type superblock should belong to 534 * @test: comparison callback 535 * @set: setup callback 536 * @flags: mount flags 537 * @data: argument to each of them 538 */ 539 struct super_block *sget(struct file_system_type *type, 540 int (*test)(struct super_block *,void *), 541 int (*set)(struct super_block *,void *), 542 int flags, 543 void *data) 544 { 545 struct user_namespace *user_ns = current_user_ns(); 546 547 /* We don't yet pass the user namespace of the parent 548 * mount through to here so always use &init_user_ns 549 * until that changes. 550 */ 551 if (flags & SB_SUBMOUNT) 552 user_ns = &init_user_ns; 553 554 /* Ensure the requestor has permissions over the target filesystem */ 555 if (!(flags & (SB_KERNMOUNT|SB_SUBMOUNT)) && !ns_capable(user_ns, CAP_SYS_ADMIN)) 556 return ERR_PTR(-EPERM); 557 558 return sget_userns(type, test, set, flags, user_ns, data); 559 } 560 561 EXPORT_SYMBOL(sget); 562 563 void drop_super(struct super_block *sb) 564 { 565 up_read(&sb->s_umount); 566 put_super(sb); 567 } 568 569 EXPORT_SYMBOL(drop_super); 570 571 void drop_super_exclusive(struct super_block *sb) 572 { 573 up_write(&sb->s_umount); 574 put_super(sb); 575 } 576 EXPORT_SYMBOL(drop_super_exclusive); 577 578 /** 579 * iterate_supers - call function for all active superblocks 580 * @f: function to call 581 * @arg: argument to pass to it 582 * 583 * Scans the superblock list and calls given function, passing it 584 * locked superblock and given argument. 585 */ 586 void iterate_supers(void (*f)(struct super_block *, void *), void *arg) 587 { 588 struct super_block *sb, *p = NULL; 589 590 spin_lock(&sb_lock); 591 list_for_each_entry(sb, &super_blocks, s_list) { 592 if (hlist_unhashed(&sb->s_instances)) 593 continue; 594 sb->s_count++; 595 spin_unlock(&sb_lock); 596 597 down_read(&sb->s_umount); 598 if (sb->s_root && (sb->s_flags & SB_BORN)) 599 f(sb, arg); 600 up_read(&sb->s_umount); 601 602 spin_lock(&sb_lock); 603 if (p) 604 __put_super(p); 605 p = sb; 606 } 607 if (p) 608 __put_super(p); 609 spin_unlock(&sb_lock); 610 } 611 612 /** 613 * iterate_supers_type - call function for superblocks of given type 614 * @type: fs type 615 * @f: function to call 616 * @arg: argument to pass to it 617 * 618 * Scans the superblock list and calls given function, passing it 619 * locked superblock and given argument. 620 */ 621 void iterate_supers_type(struct file_system_type *type, 622 void (*f)(struct super_block *, void *), void *arg) 623 { 624 struct super_block *sb, *p = NULL; 625 626 spin_lock(&sb_lock); 627 hlist_for_each_entry(sb, &type->fs_supers, s_instances) { 628 sb->s_count++; 629 spin_unlock(&sb_lock); 630 631 down_read(&sb->s_umount); 632 if (sb->s_root && (sb->s_flags & SB_BORN)) 633 f(sb, arg); 634 up_read(&sb->s_umount); 635 636 spin_lock(&sb_lock); 637 if (p) 638 __put_super(p); 639 p = sb; 640 } 641 if (p) 642 __put_super(p); 643 spin_unlock(&sb_lock); 644 } 645 646 EXPORT_SYMBOL(iterate_supers_type); 647 648 static struct super_block *__get_super(struct block_device *bdev, bool excl) 649 { 650 struct super_block *sb; 651 652 if (!bdev) 653 return NULL; 654 655 spin_lock(&sb_lock); 656 rescan: 657 list_for_each_entry(sb, &super_blocks, s_list) { 658 if (hlist_unhashed(&sb->s_instances)) 659 continue; 660 if (sb->s_bdev == bdev) { 661 sb->s_count++; 662 spin_unlock(&sb_lock); 663 if (!excl) 664 down_read(&sb->s_umount); 665 else 666 down_write(&sb->s_umount); 667 /* still alive? */ 668 if (sb->s_root && (sb->s_flags & SB_BORN)) 669 return sb; 670 if (!excl) 671 up_read(&sb->s_umount); 672 else 673 up_write(&sb->s_umount); 674 /* nope, got unmounted */ 675 spin_lock(&sb_lock); 676 __put_super(sb); 677 goto rescan; 678 } 679 } 680 spin_unlock(&sb_lock); 681 return NULL; 682 } 683 684 /** 685 * get_super - get the superblock of a device 686 * @bdev: device to get the superblock for 687 * 688 * Scans the superblock list and finds the superblock of the file system 689 * mounted on the device given. %NULL is returned if no match is found. 690 */ 691 struct super_block *get_super(struct block_device *bdev) 692 { 693 return __get_super(bdev, false); 694 } 695 EXPORT_SYMBOL(get_super); 696 697 static struct super_block *__get_super_thawed(struct block_device *bdev, 698 bool excl) 699 { 700 while (1) { 701 struct super_block *s = __get_super(bdev, excl); 702 if (!s || s->s_writers.frozen == SB_UNFROZEN) 703 return s; 704 if (!excl) 705 up_read(&s->s_umount); 706 else 707 up_write(&s->s_umount); 708 wait_event(s->s_writers.wait_unfrozen, 709 s->s_writers.frozen == SB_UNFROZEN); 710 put_super(s); 711 } 712 } 713 714 /** 715 * get_super_thawed - get thawed superblock of a device 716 * @bdev: device to get the superblock for 717 * 718 * Scans the superblock list and finds the superblock of the file system 719 * mounted on the device. The superblock is returned once it is thawed 720 * (or immediately if it was not frozen). %NULL is returned if no match 721 * is found. 722 */ 723 struct super_block *get_super_thawed(struct block_device *bdev) 724 { 725 return __get_super_thawed(bdev, false); 726 } 727 EXPORT_SYMBOL(get_super_thawed); 728 729 /** 730 * get_super_exclusive_thawed - get thawed superblock of a device 731 * @bdev: device to get the superblock for 732 * 733 * Scans the superblock list and finds the superblock of the file system 734 * mounted on the device. The superblock is returned once it is thawed 735 * (or immediately if it was not frozen) and s_umount semaphore is held 736 * in exclusive mode. %NULL is returned if no match is found. 737 */ 738 struct super_block *get_super_exclusive_thawed(struct block_device *bdev) 739 { 740 return __get_super_thawed(bdev, true); 741 } 742 EXPORT_SYMBOL(get_super_exclusive_thawed); 743 744 /** 745 * get_active_super - get an active reference to the superblock of a device 746 * @bdev: device to get the superblock for 747 * 748 * Scans the superblock list and finds the superblock of the file system 749 * mounted on the device given. Returns the superblock with an active 750 * reference or %NULL if none was found. 751 */ 752 struct super_block *get_active_super(struct block_device *bdev) 753 { 754 struct super_block *sb; 755 756 if (!bdev) 757 return NULL; 758 759 restart: 760 spin_lock(&sb_lock); 761 list_for_each_entry(sb, &super_blocks, s_list) { 762 if (hlist_unhashed(&sb->s_instances)) 763 continue; 764 if (sb->s_bdev == bdev) { 765 if (!grab_super(sb)) 766 goto restart; 767 up_write(&sb->s_umount); 768 return sb; 769 } 770 } 771 spin_unlock(&sb_lock); 772 return NULL; 773 } 774 775 struct super_block *user_get_super(dev_t dev) 776 { 777 struct super_block *sb; 778 779 spin_lock(&sb_lock); 780 rescan: 781 list_for_each_entry(sb, &super_blocks, s_list) { 782 if (hlist_unhashed(&sb->s_instances)) 783 continue; 784 if (sb->s_dev == dev) { 785 sb->s_count++; 786 spin_unlock(&sb_lock); 787 down_read(&sb->s_umount); 788 /* still alive? */ 789 if (sb->s_root && (sb->s_flags & SB_BORN)) 790 return sb; 791 up_read(&sb->s_umount); 792 /* nope, got unmounted */ 793 spin_lock(&sb_lock); 794 __put_super(sb); 795 goto rescan; 796 } 797 } 798 spin_unlock(&sb_lock); 799 return NULL; 800 } 801 802 /** 803 * do_remount_sb - asks filesystem to change mount options. 804 * @sb: superblock in question 805 * @sb_flags: revised superblock flags 806 * @data: the rest of options 807 * @force: whether or not to force the change 808 * 809 * Alters the mount options of a mounted file system. 810 */ 811 int do_remount_sb(struct super_block *sb, int sb_flags, void *data, int force) 812 { 813 int retval; 814 int remount_ro; 815 816 if (sb->s_writers.frozen != SB_UNFROZEN) 817 return -EBUSY; 818 819 #ifdef CONFIG_BLOCK 820 if (!(sb_flags & SB_RDONLY) && bdev_read_only(sb->s_bdev)) 821 return -EACCES; 822 #endif 823 824 remount_ro = (sb_flags & SB_RDONLY) && !sb_rdonly(sb); 825 826 if (remount_ro) { 827 if (!hlist_empty(&sb->s_pins)) { 828 up_write(&sb->s_umount); 829 group_pin_kill(&sb->s_pins); 830 down_write(&sb->s_umount); 831 if (!sb->s_root) 832 return 0; 833 if (sb->s_writers.frozen != SB_UNFROZEN) 834 return -EBUSY; 835 remount_ro = (sb_flags & SB_RDONLY) && !sb_rdonly(sb); 836 } 837 } 838 shrink_dcache_sb(sb); 839 840 /* If we are remounting RDONLY and current sb is read/write, 841 make sure there are no rw files opened */ 842 if (remount_ro) { 843 if (force) { 844 sb->s_readonly_remount = 1; 845 smp_wmb(); 846 } else { 847 retval = sb_prepare_remount_readonly(sb); 848 if (retval) 849 return retval; 850 } 851 } 852 853 if (sb->s_op->remount_fs) { 854 retval = sb->s_op->remount_fs(sb, &sb_flags, data); 855 if (retval) { 856 if (!force) 857 goto cancel_readonly; 858 /* If forced remount, go ahead despite any errors */ 859 WARN(1, "forced remount of a %s fs returned %i\n", 860 sb->s_type->name, retval); 861 } 862 } 863 sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (sb_flags & MS_RMT_MASK); 864 /* Needs to be ordered wrt mnt_is_readonly() */ 865 smp_wmb(); 866 sb->s_readonly_remount = 0; 867 868 /* 869 * Some filesystems modify their metadata via some other path than the 870 * bdev buffer cache (eg. use a private mapping, or directories in 871 * pagecache, etc). Also file data modifications go via their own 872 * mappings. So If we try to mount readonly then copy the filesystem 873 * from bdev, we could get stale data, so invalidate it to give a best 874 * effort at coherency. 875 */ 876 if (remount_ro && sb->s_bdev) 877 invalidate_bdev(sb->s_bdev); 878 return 0; 879 880 cancel_readonly: 881 sb->s_readonly_remount = 0; 882 return retval; 883 } 884 885 static void do_emergency_remount(struct work_struct *work) 886 { 887 struct super_block *sb, *p = NULL; 888 889 spin_lock(&sb_lock); 890 list_for_each_entry(sb, &super_blocks, s_list) { 891 if (hlist_unhashed(&sb->s_instances)) 892 continue; 893 sb->s_count++; 894 spin_unlock(&sb_lock); 895 down_write(&sb->s_umount); 896 if (sb->s_root && sb->s_bdev && (sb->s_flags & SB_BORN) && 897 !sb_rdonly(sb)) { 898 /* 899 * What lock protects sb->s_flags?? 900 */ 901 do_remount_sb(sb, SB_RDONLY, NULL, 1); 902 } 903 up_write(&sb->s_umount); 904 spin_lock(&sb_lock); 905 if (p) 906 __put_super(p); 907 p = sb; 908 } 909 if (p) 910 __put_super(p); 911 spin_unlock(&sb_lock); 912 kfree(work); 913 printk("Emergency Remount complete\n"); 914 } 915 916 void emergency_remount(void) 917 { 918 struct work_struct *work; 919 920 work = kmalloc(sizeof(*work), GFP_ATOMIC); 921 if (work) { 922 INIT_WORK(work, do_emergency_remount); 923 schedule_work(work); 924 } 925 } 926 927 /* 928 * Unnamed block devices are dummy devices used by virtual 929 * filesystems which don't use real block-devices. -- jrs 930 */ 931 932 static DEFINE_IDA(unnamed_dev_ida); 933 static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */ 934 /* Many userspace utilities consider an FSID of 0 invalid. 935 * Always return at least 1 from get_anon_bdev. 936 */ 937 static int unnamed_dev_start = 1; 938 939 int get_anon_bdev(dev_t *p) 940 { 941 int dev; 942 int error; 943 944 retry: 945 if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0) 946 return -ENOMEM; 947 spin_lock(&unnamed_dev_lock); 948 error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev); 949 if (!error) 950 unnamed_dev_start = dev + 1; 951 spin_unlock(&unnamed_dev_lock); 952 if (error == -EAGAIN) 953 /* We raced and lost with another CPU. */ 954 goto retry; 955 else if (error) 956 return -EAGAIN; 957 958 if (dev >= (1 << MINORBITS)) { 959 spin_lock(&unnamed_dev_lock); 960 ida_remove(&unnamed_dev_ida, dev); 961 if (unnamed_dev_start > dev) 962 unnamed_dev_start = dev; 963 spin_unlock(&unnamed_dev_lock); 964 return -EMFILE; 965 } 966 *p = MKDEV(0, dev & MINORMASK); 967 return 0; 968 } 969 EXPORT_SYMBOL(get_anon_bdev); 970 971 void free_anon_bdev(dev_t dev) 972 { 973 int slot = MINOR(dev); 974 spin_lock(&unnamed_dev_lock); 975 ida_remove(&unnamed_dev_ida, slot); 976 if (slot < unnamed_dev_start) 977 unnamed_dev_start = slot; 978 spin_unlock(&unnamed_dev_lock); 979 } 980 EXPORT_SYMBOL(free_anon_bdev); 981 982 int set_anon_super(struct super_block *s, void *data) 983 { 984 return get_anon_bdev(&s->s_dev); 985 } 986 987 EXPORT_SYMBOL(set_anon_super); 988 989 void kill_anon_super(struct super_block *sb) 990 { 991 dev_t dev = sb->s_dev; 992 generic_shutdown_super(sb); 993 free_anon_bdev(dev); 994 } 995 996 EXPORT_SYMBOL(kill_anon_super); 997 998 void kill_litter_super(struct super_block *sb) 999 { 1000 if (sb->s_root) 1001 d_genocide(sb->s_root); 1002 kill_anon_super(sb); 1003 } 1004 1005 EXPORT_SYMBOL(kill_litter_super); 1006 1007 static int ns_test_super(struct super_block *sb, void *data) 1008 { 1009 return sb->s_fs_info == data; 1010 } 1011 1012 static int ns_set_super(struct super_block *sb, void *data) 1013 { 1014 sb->s_fs_info = data; 1015 return set_anon_super(sb, NULL); 1016 } 1017 1018 struct dentry *mount_ns(struct file_system_type *fs_type, 1019 int flags, void *data, void *ns, struct user_namespace *user_ns, 1020 int (*fill_super)(struct super_block *, void *, int)) 1021 { 1022 struct super_block *sb; 1023 1024 /* Don't allow mounting unless the caller has CAP_SYS_ADMIN 1025 * over the namespace. 1026 */ 1027 if (!(flags & SB_KERNMOUNT) && !ns_capable(user_ns, CAP_SYS_ADMIN)) 1028 return ERR_PTR(-EPERM); 1029 1030 sb = sget_userns(fs_type, ns_test_super, ns_set_super, flags, 1031 user_ns, ns); 1032 if (IS_ERR(sb)) 1033 return ERR_CAST(sb); 1034 1035 if (!sb->s_root) { 1036 int err; 1037 err = fill_super(sb, data, flags & SB_SILENT ? 1 : 0); 1038 if (err) { 1039 deactivate_locked_super(sb); 1040 return ERR_PTR(err); 1041 } 1042 1043 sb->s_flags |= SB_ACTIVE; 1044 } 1045 1046 return dget(sb->s_root); 1047 } 1048 1049 EXPORT_SYMBOL(mount_ns); 1050 1051 #ifdef CONFIG_BLOCK 1052 static int set_bdev_super(struct super_block *s, void *data) 1053 { 1054 s->s_bdev = data; 1055 s->s_dev = s->s_bdev->bd_dev; 1056 s->s_bdi = bdi_get(s->s_bdev->bd_bdi); 1057 1058 return 0; 1059 } 1060 1061 static int test_bdev_super(struct super_block *s, void *data) 1062 { 1063 return (void *)s->s_bdev == data; 1064 } 1065 1066 struct dentry *mount_bdev(struct file_system_type *fs_type, 1067 int flags, const char *dev_name, void *data, 1068 int (*fill_super)(struct super_block *, void *, int)) 1069 { 1070 struct block_device *bdev; 1071 struct super_block *s; 1072 fmode_t mode = FMODE_READ | FMODE_EXCL; 1073 int error = 0; 1074 1075 if (!(flags & SB_RDONLY)) 1076 mode |= FMODE_WRITE; 1077 1078 bdev = blkdev_get_by_path(dev_name, mode, fs_type); 1079 if (IS_ERR(bdev)) 1080 return ERR_CAST(bdev); 1081 1082 /* 1083 * once the super is inserted into the list by sget, s_umount 1084 * will protect the lockfs code from trying to start a snapshot 1085 * while we are mounting 1086 */ 1087 mutex_lock(&bdev->bd_fsfreeze_mutex); 1088 if (bdev->bd_fsfreeze_count > 0) { 1089 mutex_unlock(&bdev->bd_fsfreeze_mutex); 1090 error = -EBUSY; 1091 goto error_bdev; 1092 } 1093 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | SB_NOSEC, 1094 bdev); 1095 mutex_unlock(&bdev->bd_fsfreeze_mutex); 1096 if (IS_ERR(s)) 1097 goto error_s; 1098 1099 if (s->s_root) { 1100 if ((flags ^ s->s_flags) & SB_RDONLY) { 1101 deactivate_locked_super(s); 1102 error = -EBUSY; 1103 goto error_bdev; 1104 } 1105 1106 /* 1107 * s_umount nests inside bd_mutex during 1108 * __invalidate_device(). blkdev_put() acquires 1109 * bd_mutex and can't be called under s_umount. Drop 1110 * s_umount temporarily. This is safe as we're 1111 * holding an active reference. 1112 */ 1113 up_write(&s->s_umount); 1114 blkdev_put(bdev, mode); 1115 down_write(&s->s_umount); 1116 } else { 1117 s->s_mode = mode; 1118 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev); 1119 sb_set_blocksize(s, block_size(bdev)); 1120 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0); 1121 if (error) { 1122 deactivate_locked_super(s); 1123 goto error; 1124 } 1125 1126 s->s_flags |= SB_ACTIVE; 1127 bdev->bd_super = s; 1128 } 1129 1130 return dget(s->s_root); 1131 1132 error_s: 1133 error = PTR_ERR(s); 1134 error_bdev: 1135 blkdev_put(bdev, mode); 1136 error: 1137 return ERR_PTR(error); 1138 } 1139 EXPORT_SYMBOL(mount_bdev); 1140 1141 void kill_block_super(struct super_block *sb) 1142 { 1143 struct block_device *bdev = sb->s_bdev; 1144 fmode_t mode = sb->s_mode; 1145 1146 bdev->bd_super = NULL; 1147 generic_shutdown_super(sb); 1148 sync_blockdev(bdev); 1149 WARN_ON_ONCE(!(mode & FMODE_EXCL)); 1150 blkdev_put(bdev, mode | FMODE_EXCL); 1151 } 1152 1153 EXPORT_SYMBOL(kill_block_super); 1154 #endif 1155 1156 struct dentry *mount_nodev(struct file_system_type *fs_type, 1157 int flags, void *data, 1158 int (*fill_super)(struct super_block *, void *, int)) 1159 { 1160 int error; 1161 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL); 1162 1163 if (IS_ERR(s)) 1164 return ERR_CAST(s); 1165 1166 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0); 1167 if (error) { 1168 deactivate_locked_super(s); 1169 return ERR_PTR(error); 1170 } 1171 s->s_flags |= SB_ACTIVE; 1172 return dget(s->s_root); 1173 } 1174 EXPORT_SYMBOL(mount_nodev); 1175 1176 static int compare_single(struct super_block *s, void *p) 1177 { 1178 return 1; 1179 } 1180 1181 struct dentry *mount_single(struct file_system_type *fs_type, 1182 int flags, void *data, 1183 int (*fill_super)(struct super_block *, void *, int)) 1184 { 1185 struct super_block *s; 1186 int error; 1187 1188 s = sget(fs_type, compare_single, set_anon_super, flags, NULL); 1189 if (IS_ERR(s)) 1190 return ERR_CAST(s); 1191 if (!s->s_root) { 1192 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0); 1193 if (error) { 1194 deactivate_locked_super(s); 1195 return ERR_PTR(error); 1196 } 1197 s->s_flags |= SB_ACTIVE; 1198 } else { 1199 do_remount_sb(s, flags, data, 0); 1200 } 1201 return dget(s->s_root); 1202 } 1203 EXPORT_SYMBOL(mount_single); 1204 1205 struct dentry * 1206 mount_fs(struct file_system_type *type, int flags, const char *name, void *data) 1207 { 1208 struct dentry *root; 1209 struct super_block *sb; 1210 char *secdata = NULL; 1211 int error = -ENOMEM; 1212 1213 if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) { 1214 secdata = alloc_secdata(); 1215 if (!secdata) 1216 goto out; 1217 1218 error = security_sb_copy_data(data, secdata); 1219 if (error) 1220 goto out_free_secdata; 1221 } 1222 1223 root = type->mount(type, flags, name, data); 1224 if (IS_ERR(root)) { 1225 error = PTR_ERR(root); 1226 goto out_free_secdata; 1227 } 1228 sb = root->d_sb; 1229 BUG_ON(!sb); 1230 WARN_ON(!sb->s_bdi); 1231 sb->s_flags |= SB_BORN; 1232 1233 error = security_sb_kern_mount(sb, flags, secdata); 1234 if (error) 1235 goto out_sb; 1236 1237 /* 1238 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE 1239 * but s_maxbytes was an unsigned long long for many releases. Throw 1240 * this warning for a little while to try and catch filesystems that 1241 * violate this rule. 1242 */ 1243 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to " 1244 "negative value (%lld)\n", type->name, sb->s_maxbytes); 1245 1246 up_write(&sb->s_umount); 1247 free_secdata(secdata); 1248 return root; 1249 out_sb: 1250 dput(root); 1251 deactivate_locked_super(sb); 1252 out_free_secdata: 1253 free_secdata(secdata); 1254 out: 1255 return ERR_PTR(error); 1256 } 1257 1258 /* 1259 * Setup private BDI for given superblock. It gets automatically cleaned up 1260 * in generic_shutdown_super(). 1261 */ 1262 int super_setup_bdi_name(struct super_block *sb, char *fmt, ...) 1263 { 1264 struct backing_dev_info *bdi; 1265 int err; 1266 va_list args; 1267 1268 bdi = bdi_alloc(GFP_KERNEL); 1269 if (!bdi) 1270 return -ENOMEM; 1271 1272 bdi->name = sb->s_type->name; 1273 1274 va_start(args, fmt); 1275 err = bdi_register_va(bdi, fmt, args); 1276 va_end(args); 1277 if (err) { 1278 bdi_put(bdi); 1279 return err; 1280 } 1281 WARN_ON(sb->s_bdi != &noop_backing_dev_info); 1282 sb->s_bdi = bdi; 1283 1284 return 0; 1285 } 1286 EXPORT_SYMBOL(super_setup_bdi_name); 1287 1288 /* 1289 * Setup private BDI for given superblock. I gets automatically cleaned up 1290 * in generic_shutdown_super(). 1291 */ 1292 int super_setup_bdi(struct super_block *sb) 1293 { 1294 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0); 1295 1296 return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name, 1297 atomic_long_inc_return(&bdi_seq)); 1298 } 1299 EXPORT_SYMBOL(super_setup_bdi); 1300 1301 /* 1302 * This is an internal function, please use sb_end_{write,pagefault,intwrite} 1303 * instead. 1304 */ 1305 void __sb_end_write(struct super_block *sb, int level) 1306 { 1307 percpu_up_read(sb->s_writers.rw_sem + level-1); 1308 } 1309 EXPORT_SYMBOL(__sb_end_write); 1310 1311 /* 1312 * This is an internal function, please use sb_start_{write,pagefault,intwrite} 1313 * instead. 1314 */ 1315 int __sb_start_write(struct super_block *sb, int level, bool wait) 1316 { 1317 bool force_trylock = false; 1318 int ret = 1; 1319 1320 #ifdef CONFIG_LOCKDEP 1321 /* 1322 * We want lockdep to tell us about possible deadlocks with freezing 1323 * but it's it bit tricky to properly instrument it. Getting a freeze 1324 * protection works as getting a read lock but there are subtle 1325 * problems. XFS for example gets freeze protection on internal level 1326 * twice in some cases, which is OK only because we already hold a 1327 * freeze protection also on higher level. Due to these cases we have 1328 * to use wait == F (trylock mode) which must not fail. 1329 */ 1330 if (wait) { 1331 int i; 1332 1333 for (i = 0; i < level - 1; i++) 1334 if (percpu_rwsem_is_held(sb->s_writers.rw_sem + i)) { 1335 force_trylock = true; 1336 break; 1337 } 1338 } 1339 #endif 1340 if (wait && !force_trylock) 1341 percpu_down_read(sb->s_writers.rw_sem + level-1); 1342 else 1343 ret = percpu_down_read_trylock(sb->s_writers.rw_sem + level-1); 1344 1345 WARN_ON(force_trylock && !ret); 1346 return ret; 1347 } 1348 EXPORT_SYMBOL(__sb_start_write); 1349 1350 /** 1351 * sb_wait_write - wait until all writers to given file system finish 1352 * @sb: the super for which we wait 1353 * @level: type of writers we wait for (normal vs page fault) 1354 * 1355 * This function waits until there are no writers of given type to given file 1356 * system. 1357 */ 1358 static void sb_wait_write(struct super_block *sb, int level) 1359 { 1360 percpu_down_write(sb->s_writers.rw_sem + level-1); 1361 } 1362 1363 /* 1364 * We are going to return to userspace and forget about these locks, the 1365 * ownership goes to the caller of thaw_super() which does unlock(). 1366 */ 1367 static void lockdep_sb_freeze_release(struct super_block *sb) 1368 { 1369 int level; 1370 1371 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--) 1372 percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_); 1373 } 1374 1375 /* 1376 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb). 1377 */ 1378 static void lockdep_sb_freeze_acquire(struct super_block *sb) 1379 { 1380 int level; 1381 1382 for (level = 0; level < SB_FREEZE_LEVELS; ++level) 1383 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_); 1384 } 1385 1386 static void sb_freeze_unlock(struct super_block *sb) 1387 { 1388 int level; 1389 1390 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--) 1391 percpu_up_write(sb->s_writers.rw_sem + level); 1392 } 1393 1394 /** 1395 * freeze_super - lock the filesystem and force it into a consistent state 1396 * @sb: the super to lock 1397 * 1398 * Syncs the super to make sure the filesystem is consistent and calls the fs's 1399 * freeze_fs. Subsequent calls to this without first thawing the fs will return 1400 * -EBUSY. 1401 * 1402 * During this function, sb->s_writers.frozen goes through these values: 1403 * 1404 * SB_UNFROZEN: File system is normal, all writes progress as usual. 1405 * 1406 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New 1407 * writes should be blocked, though page faults are still allowed. We wait for 1408 * all writes to complete and then proceed to the next stage. 1409 * 1410 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked 1411 * but internal fs threads can still modify the filesystem (although they 1412 * should not dirty new pages or inodes), writeback can run etc. After waiting 1413 * for all running page faults we sync the filesystem which will clean all 1414 * dirty pages and inodes (no new dirty pages or inodes can be created when 1415 * sync is running). 1416 * 1417 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs 1418 * modification are blocked (e.g. XFS preallocation truncation on inode 1419 * reclaim). This is usually implemented by blocking new transactions for 1420 * filesystems that have them and need this additional guard. After all 1421 * internal writers are finished we call ->freeze_fs() to finish filesystem 1422 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is 1423 * mostly auxiliary for filesystems to verify they do not modify frozen fs. 1424 * 1425 * sb->s_writers.frozen is protected by sb->s_umount. 1426 */ 1427 int freeze_super(struct super_block *sb) 1428 { 1429 int ret; 1430 1431 atomic_inc(&sb->s_active); 1432 down_write(&sb->s_umount); 1433 if (sb->s_writers.frozen != SB_UNFROZEN) { 1434 deactivate_locked_super(sb); 1435 return -EBUSY; 1436 } 1437 1438 if (!(sb->s_flags & SB_BORN)) { 1439 up_write(&sb->s_umount); 1440 return 0; /* sic - it's "nothing to do" */ 1441 } 1442 1443 if (sb_rdonly(sb)) { 1444 /* Nothing to do really... */ 1445 sb->s_writers.frozen = SB_FREEZE_COMPLETE; 1446 up_write(&sb->s_umount); 1447 return 0; 1448 } 1449 1450 sb->s_writers.frozen = SB_FREEZE_WRITE; 1451 /* Release s_umount to preserve sb_start_write -> s_umount ordering */ 1452 up_write(&sb->s_umount); 1453 sb_wait_write(sb, SB_FREEZE_WRITE); 1454 down_write(&sb->s_umount); 1455 1456 /* Now we go and block page faults... */ 1457 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT; 1458 sb_wait_write(sb, SB_FREEZE_PAGEFAULT); 1459 1460 /* All writers are done so after syncing there won't be dirty data */ 1461 sync_filesystem(sb); 1462 1463 /* Now wait for internal filesystem counter */ 1464 sb->s_writers.frozen = SB_FREEZE_FS; 1465 sb_wait_write(sb, SB_FREEZE_FS); 1466 1467 if (sb->s_op->freeze_fs) { 1468 ret = sb->s_op->freeze_fs(sb); 1469 if (ret) { 1470 printk(KERN_ERR 1471 "VFS:Filesystem freeze failed\n"); 1472 sb->s_writers.frozen = SB_UNFROZEN; 1473 sb_freeze_unlock(sb); 1474 wake_up(&sb->s_writers.wait_unfrozen); 1475 deactivate_locked_super(sb); 1476 return ret; 1477 } 1478 } 1479 /* 1480 * For debugging purposes so that fs can warn if it sees write activity 1481 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super(). 1482 */ 1483 sb->s_writers.frozen = SB_FREEZE_COMPLETE; 1484 lockdep_sb_freeze_release(sb); 1485 up_write(&sb->s_umount); 1486 return 0; 1487 } 1488 EXPORT_SYMBOL(freeze_super); 1489 1490 /** 1491 * thaw_super -- unlock filesystem 1492 * @sb: the super to thaw 1493 * 1494 * Unlocks the filesystem and marks it writeable again after freeze_super(). 1495 */ 1496 int thaw_super(struct super_block *sb) 1497 { 1498 int error; 1499 1500 down_write(&sb->s_umount); 1501 if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) { 1502 up_write(&sb->s_umount); 1503 return -EINVAL; 1504 } 1505 1506 if (sb_rdonly(sb)) { 1507 sb->s_writers.frozen = SB_UNFROZEN; 1508 goto out; 1509 } 1510 1511 lockdep_sb_freeze_acquire(sb); 1512 1513 if (sb->s_op->unfreeze_fs) { 1514 error = sb->s_op->unfreeze_fs(sb); 1515 if (error) { 1516 printk(KERN_ERR 1517 "VFS:Filesystem thaw failed\n"); 1518 lockdep_sb_freeze_release(sb); 1519 up_write(&sb->s_umount); 1520 return error; 1521 } 1522 } 1523 1524 sb->s_writers.frozen = SB_UNFROZEN; 1525 sb_freeze_unlock(sb); 1526 out: 1527 wake_up(&sb->s_writers.wait_unfrozen); 1528 deactivate_locked_super(sb); 1529 return 0; 1530 } 1531 EXPORT_SYMBOL(thaw_super); 1532