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/fscrypt.h> 35 #include <linux/fsnotify.h> 36 #include <linux/lockdep.h> 37 #include <linux/user_namespace.h> 38 #include <linux/fs_context.h> 39 #include <uapi/linux/mount.h> 40 #include "internal.h" 41 42 static int thaw_super_locked(struct super_block *sb); 43 44 static LIST_HEAD(super_blocks); 45 static DEFINE_SPINLOCK(sb_lock); 46 47 static char *sb_writers_name[SB_FREEZE_LEVELS] = { 48 "sb_writers", 49 "sb_pagefaults", 50 "sb_internal", 51 }; 52 53 /* 54 * One thing we have to be careful of with a per-sb shrinker is that we don't 55 * drop the last active reference to the superblock from within the shrinker. 56 * If that happens we could trigger unregistering the shrinker from within the 57 * shrinker path and that leads to deadlock on the shrinker_mutex. Hence we 58 * take a passive reference to the superblock to avoid this from occurring. 59 */ 60 static unsigned long super_cache_scan(struct shrinker *shrink, 61 struct shrink_control *sc) 62 { 63 struct super_block *sb; 64 long fs_objects = 0; 65 long total_objects; 66 long freed = 0; 67 long dentries; 68 long inodes; 69 70 sb = container_of(shrink, struct super_block, s_shrink); 71 72 /* 73 * Deadlock avoidance. We may hold various FS locks, and we don't want 74 * to recurse into the FS that called us in clear_inode() and friends.. 75 */ 76 if (!(sc->gfp_mask & __GFP_FS)) 77 return SHRINK_STOP; 78 79 if (!trylock_super(sb)) 80 return SHRINK_STOP; 81 82 if (sb->s_op->nr_cached_objects) 83 fs_objects = sb->s_op->nr_cached_objects(sb, sc); 84 85 inodes = list_lru_shrink_count(&sb->s_inode_lru, sc); 86 dentries = list_lru_shrink_count(&sb->s_dentry_lru, sc); 87 total_objects = dentries + inodes + fs_objects + 1; 88 if (!total_objects) 89 total_objects = 1; 90 91 /* proportion the scan between the caches */ 92 dentries = mult_frac(sc->nr_to_scan, dentries, total_objects); 93 inodes = mult_frac(sc->nr_to_scan, inodes, total_objects); 94 fs_objects = mult_frac(sc->nr_to_scan, fs_objects, total_objects); 95 96 /* 97 * prune the dcache first as the icache is pinned by it, then 98 * prune the icache, followed by the filesystem specific caches 99 * 100 * Ensure that we always scan at least one object - memcg kmem 101 * accounting uses this to fully empty the caches. 102 */ 103 sc->nr_to_scan = dentries + 1; 104 freed = prune_dcache_sb(sb, sc); 105 sc->nr_to_scan = inodes + 1; 106 freed += prune_icache_sb(sb, sc); 107 108 if (fs_objects) { 109 sc->nr_to_scan = fs_objects + 1; 110 freed += sb->s_op->free_cached_objects(sb, sc); 111 } 112 113 up_read(&sb->s_umount); 114 return freed; 115 } 116 117 static unsigned long super_cache_count(struct shrinker *shrink, 118 struct shrink_control *sc) 119 { 120 struct super_block *sb; 121 long total_objects = 0; 122 123 sb = container_of(shrink, struct super_block, s_shrink); 124 125 /* 126 * We don't call trylock_super() here as it is a scalability bottleneck, 127 * so we're exposed to partial setup state. The shrinker rwsem does not 128 * protect filesystem operations backing list_lru_shrink_count() or 129 * s_op->nr_cached_objects(). Counts can change between 130 * super_cache_count and super_cache_scan, so we really don't need locks 131 * here. 132 * 133 * However, if we are currently mounting the superblock, the underlying 134 * filesystem might be in a state of partial construction and hence it 135 * is dangerous to access it. trylock_super() uses a SB_BORN check to 136 * avoid this situation, so do the same here. The memory barrier is 137 * matched with the one in mount_fs() as we don't hold locks here. 138 */ 139 if (!(sb->s_flags & SB_BORN)) 140 return 0; 141 smp_rmb(); 142 143 if (sb->s_op && sb->s_op->nr_cached_objects) 144 total_objects = sb->s_op->nr_cached_objects(sb, sc); 145 146 total_objects += list_lru_shrink_count(&sb->s_dentry_lru, sc); 147 total_objects += list_lru_shrink_count(&sb->s_inode_lru, sc); 148 149 if (!total_objects) 150 return SHRINK_EMPTY; 151 152 total_objects = vfs_pressure_ratio(total_objects); 153 return total_objects; 154 } 155 156 static void destroy_super_work(struct work_struct *work) 157 { 158 struct super_block *s = container_of(work, struct super_block, 159 destroy_work); 160 int i; 161 162 for (i = 0; i < SB_FREEZE_LEVELS; i++) 163 percpu_free_rwsem(&s->s_writers.rw_sem[i]); 164 kfree(s); 165 } 166 167 static void destroy_super_rcu(struct rcu_head *head) 168 { 169 struct super_block *s = container_of(head, struct super_block, rcu); 170 INIT_WORK(&s->destroy_work, destroy_super_work); 171 schedule_work(&s->destroy_work); 172 } 173 174 /* Free a superblock that has never been seen by anyone */ 175 static void destroy_unused_super(struct super_block *s) 176 { 177 if (!s) 178 return; 179 up_write(&s->s_umount); 180 list_lru_destroy(&s->s_dentry_lru); 181 list_lru_destroy(&s->s_inode_lru); 182 security_sb_free(s); 183 put_user_ns(s->s_user_ns); 184 kfree(s->s_subtype); 185 free_prealloced_shrinker(&s->s_shrink); 186 /* no delays needed */ 187 destroy_super_work(&s->destroy_work); 188 } 189 190 /** 191 * alloc_super - create new superblock 192 * @type: filesystem type superblock should belong to 193 * @flags: the mount flags 194 * @user_ns: User namespace for the super_block 195 * 196 * Allocates and initializes a new &struct super_block. alloc_super() 197 * returns a pointer new superblock or %NULL if allocation had failed. 198 */ 199 static struct super_block *alloc_super(struct file_system_type *type, int flags, 200 struct user_namespace *user_ns) 201 { 202 struct super_block *s = kzalloc(sizeof(struct super_block), GFP_USER); 203 static const struct super_operations default_op; 204 int i; 205 206 if (!s) 207 return NULL; 208 209 INIT_LIST_HEAD(&s->s_mounts); 210 s->s_user_ns = get_user_ns(user_ns); 211 init_rwsem(&s->s_umount); 212 lockdep_set_class(&s->s_umount, &type->s_umount_key); 213 /* 214 * sget() can have s_umount recursion. 215 * 216 * When it cannot find a suitable sb, it allocates a new 217 * one (this one), and tries again to find a suitable old 218 * one. 219 * 220 * In case that succeeds, it will acquire the s_umount 221 * lock of the old one. Since these are clearly distrinct 222 * locks, and this object isn't exposed yet, there's no 223 * risk of deadlocks. 224 * 225 * Annotate this by putting this lock in a different 226 * subclass. 227 */ 228 down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING); 229 230 if (security_sb_alloc(s)) 231 goto fail; 232 233 for (i = 0; i < SB_FREEZE_LEVELS; i++) { 234 if (__percpu_init_rwsem(&s->s_writers.rw_sem[i], 235 sb_writers_name[i], 236 &type->s_writers_key[i])) 237 goto fail; 238 } 239 init_waitqueue_head(&s->s_writers.wait_unfrozen); 240 s->s_bdi = &noop_backing_dev_info; 241 s->s_flags = flags; 242 if (s->s_user_ns != &init_user_ns) 243 s->s_iflags |= SB_I_NODEV; 244 INIT_HLIST_NODE(&s->s_instances); 245 INIT_HLIST_BL_HEAD(&s->s_roots); 246 mutex_init(&s->s_sync_lock); 247 INIT_LIST_HEAD(&s->s_inodes); 248 spin_lock_init(&s->s_inode_list_lock); 249 INIT_LIST_HEAD(&s->s_inodes_wb); 250 spin_lock_init(&s->s_inode_wblist_lock); 251 252 s->s_count = 1; 253 atomic_set(&s->s_active, 1); 254 mutex_init(&s->s_vfs_rename_mutex); 255 lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key); 256 init_rwsem(&s->s_dquot.dqio_sem); 257 s->s_maxbytes = MAX_NON_LFS; 258 s->s_op = &default_op; 259 s->s_time_gran = 1000000000; 260 s->s_time_min = TIME64_MIN; 261 s->s_time_max = TIME64_MAX; 262 263 s->s_shrink.seeks = DEFAULT_SEEKS; 264 s->s_shrink.scan_objects = super_cache_scan; 265 s->s_shrink.count_objects = super_cache_count; 266 s->s_shrink.batch = 1024; 267 s->s_shrink.flags = SHRINKER_NUMA_AWARE | SHRINKER_MEMCG_AWARE; 268 if (prealloc_shrinker(&s->s_shrink, "sb-%s", type->name)) 269 goto fail; 270 if (list_lru_init_memcg(&s->s_dentry_lru, &s->s_shrink)) 271 goto fail; 272 if (list_lru_init_memcg(&s->s_inode_lru, &s->s_shrink)) 273 goto fail; 274 return s; 275 276 fail: 277 destroy_unused_super(s); 278 return NULL; 279 } 280 281 /* Superblock refcounting */ 282 283 /* 284 * Drop a superblock's refcount. The caller must hold sb_lock. 285 */ 286 static void __put_super(struct super_block *s) 287 { 288 if (!--s->s_count) { 289 list_del_init(&s->s_list); 290 WARN_ON(s->s_dentry_lru.node); 291 WARN_ON(s->s_inode_lru.node); 292 WARN_ON(!list_empty(&s->s_mounts)); 293 security_sb_free(s); 294 put_user_ns(s->s_user_ns); 295 kfree(s->s_subtype); 296 call_rcu(&s->rcu, destroy_super_rcu); 297 } 298 } 299 300 /** 301 * put_super - drop a temporary reference to superblock 302 * @sb: superblock in question 303 * 304 * Drops a temporary reference, frees superblock if there's no 305 * references left. 306 */ 307 void put_super(struct super_block *sb) 308 { 309 spin_lock(&sb_lock); 310 __put_super(sb); 311 spin_unlock(&sb_lock); 312 } 313 314 315 /** 316 * deactivate_locked_super - drop an active reference to superblock 317 * @s: superblock to deactivate 318 * 319 * Drops an active reference to superblock, converting it into a temporary 320 * one if there is no other active references left. In that case we 321 * tell fs driver to shut it down and drop the temporary reference we 322 * had just acquired. 323 * 324 * Caller holds exclusive lock on superblock; that lock is released. 325 */ 326 void deactivate_locked_super(struct super_block *s) 327 { 328 struct file_system_type *fs = s->s_type; 329 if (atomic_dec_and_test(&s->s_active)) { 330 unregister_shrinker(&s->s_shrink); 331 fs->kill_sb(s); 332 333 /* 334 * Since list_lru_destroy() may sleep, we cannot call it from 335 * put_super(), where we hold the sb_lock. Therefore we destroy 336 * the lru lists right now. 337 */ 338 list_lru_destroy(&s->s_dentry_lru); 339 list_lru_destroy(&s->s_inode_lru); 340 341 put_filesystem(fs); 342 put_super(s); 343 } else { 344 up_write(&s->s_umount); 345 } 346 } 347 348 EXPORT_SYMBOL(deactivate_locked_super); 349 350 /** 351 * deactivate_super - drop an active reference to superblock 352 * @s: superblock to deactivate 353 * 354 * Variant of deactivate_locked_super(), except that superblock is *not* 355 * locked by caller. If we are going to drop the final active reference, 356 * lock will be acquired prior to that. 357 */ 358 void deactivate_super(struct super_block *s) 359 { 360 if (!atomic_add_unless(&s->s_active, -1, 1)) { 361 down_write(&s->s_umount); 362 deactivate_locked_super(s); 363 } 364 } 365 366 EXPORT_SYMBOL(deactivate_super); 367 368 /** 369 * grab_super - acquire an active reference 370 * @s: reference we are trying to make active 371 * 372 * Tries to acquire an active reference. grab_super() is used when we 373 * had just found a superblock in super_blocks or fs_type->fs_supers 374 * and want to turn it into a full-blown active reference. grab_super() 375 * is called with sb_lock held and drops it. Returns 1 in case of 376 * success, 0 if we had failed (superblock contents was already dead or 377 * dying when grab_super() had been called). Note that this is only 378 * called for superblocks not in rundown mode (== ones still on ->fs_supers 379 * of their type), so increment of ->s_count is OK here. 380 */ 381 static int grab_super(struct super_block *s) __releases(sb_lock) 382 { 383 s->s_count++; 384 spin_unlock(&sb_lock); 385 down_write(&s->s_umount); 386 if ((s->s_flags & SB_BORN) && atomic_inc_not_zero(&s->s_active)) { 387 put_super(s); 388 return 1; 389 } 390 up_write(&s->s_umount); 391 put_super(s); 392 return 0; 393 } 394 395 /* 396 * trylock_super - try to grab ->s_umount shared 397 * @sb: reference we are trying to grab 398 * 399 * Try to prevent fs shutdown. This is used in places where we 400 * cannot take an active reference but we need to ensure that the 401 * filesystem is not shut down while we are working on it. It returns 402 * false if we cannot acquire s_umount or if we lose the race and 403 * filesystem already got into shutdown, and returns true with the s_umount 404 * lock held in read mode in case of success. On successful return, 405 * the caller must drop the s_umount lock when done. 406 * 407 * Note that unlike get_super() et.al. this one does *not* bump ->s_count. 408 * The reason why it's safe is that we are OK with doing trylock instead 409 * of down_read(). There's a couple of places that are OK with that, but 410 * it's very much not a general-purpose interface. 411 */ 412 bool trylock_super(struct super_block *sb) 413 { 414 if (down_read_trylock(&sb->s_umount)) { 415 if (!hlist_unhashed(&sb->s_instances) && 416 sb->s_root && (sb->s_flags & SB_BORN)) 417 return true; 418 up_read(&sb->s_umount); 419 } 420 421 return false; 422 } 423 424 /** 425 * retire_super - prevents superblock from being reused 426 * @sb: superblock to retire 427 * 428 * The function marks superblock to be ignored in superblock test, which 429 * prevents it from being reused for any new mounts. If the superblock has 430 * a private bdi, it also unregisters it, but doesn't reduce the refcount 431 * of the superblock to prevent potential races. The refcount is reduced 432 * by generic_shutdown_super(). The function can not be called 433 * concurrently with generic_shutdown_super(). It is safe to call the 434 * function multiple times, subsequent calls have no effect. 435 * 436 * The marker will affect the re-use only for block-device-based 437 * superblocks. Other superblocks will still get marked if this function 438 * is used, but that will not affect their reusability. 439 */ 440 void retire_super(struct super_block *sb) 441 { 442 WARN_ON(!sb->s_bdev); 443 down_write(&sb->s_umount); 444 if (sb->s_iflags & SB_I_PERSB_BDI) { 445 bdi_unregister(sb->s_bdi); 446 sb->s_iflags &= ~SB_I_PERSB_BDI; 447 } 448 sb->s_iflags |= SB_I_RETIRED; 449 up_write(&sb->s_umount); 450 } 451 EXPORT_SYMBOL(retire_super); 452 453 /** 454 * generic_shutdown_super - common helper for ->kill_sb() 455 * @sb: superblock to kill 456 * 457 * generic_shutdown_super() does all fs-independent work on superblock 458 * shutdown. Typical ->kill_sb() should pick all fs-specific objects 459 * that need destruction out of superblock, call generic_shutdown_super() 460 * and release aforementioned objects. Note: dentries and inodes _are_ 461 * taken care of and do not need specific handling. 462 * 463 * Upon calling this function, the filesystem may no longer alter or 464 * rearrange the set of dentries belonging to this super_block, nor may it 465 * change the attachments of dentries to inodes. 466 */ 467 void generic_shutdown_super(struct super_block *sb) 468 { 469 const struct super_operations *sop = sb->s_op; 470 471 if (sb->s_root) { 472 shrink_dcache_for_umount(sb); 473 sync_filesystem(sb); 474 sb->s_flags &= ~SB_ACTIVE; 475 476 cgroup_writeback_umount(); 477 478 /* Evict all inodes with zero refcount. */ 479 evict_inodes(sb); 480 481 /* 482 * Clean up and evict any inodes that still have references due 483 * to fsnotify or the security policy. 484 */ 485 fsnotify_sb_delete(sb); 486 security_sb_delete(sb); 487 488 /* 489 * Now that all potentially-encrypted inodes have been evicted, 490 * the fscrypt keyring can be destroyed. 491 */ 492 fscrypt_destroy_keyring(sb); 493 494 if (sb->s_dio_done_wq) { 495 destroy_workqueue(sb->s_dio_done_wq); 496 sb->s_dio_done_wq = NULL; 497 } 498 499 if (sop->put_super) 500 sop->put_super(sb); 501 502 if (CHECK_DATA_CORRUPTION(!list_empty(&sb->s_inodes), 503 "VFS: Busy inodes after unmount of %s (%s)", 504 sb->s_id, sb->s_type->name)) { 505 /* 506 * Adding a proper bailout path here would be hard, but 507 * we can at least make it more likely that a later 508 * iput_final() or such crashes cleanly. 509 */ 510 struct inode *inode; 511 512 spin_lock(&sb->s_inode_list_lock); 513 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { 514 inode->i_op = VFS_PTR_POISON; 515 inode->i_sb = VFS_PTR_POISON; 516 inode->i_mapping = VFS_PTR_POISON; 517 } 518 spin_unlock(&sb->s_inode_list_lock); 519 } 520 } 521 spin_lock(&sb_lock); 522 /* should be initialized for __put_super_and_need_restart() */ 523 hlist_del_init(&sb->s_instances); 524 spin_unlock(&sb_lock); 525 up_write(&sb->s_umount); 526 if (sb->s_bdi != &noop_backing_dev_info) { 527 if (sb->s_iflags & SB_I_PERSB_BDI) 528 bdi_unregister(sb->s_bdi); 529 bdi_put(sb->s_bdi); 530 sb->s_bdi = &noop_backing_dev_info; 531 } 532 } 533 534 EXPORT_SYMBOL(generic_shutdown_super); 535 536 bool mount_capable(struct fs_context *fc) 537 { 538 if (!(fc->fs_type->fs_flags & FS_USERNS_MOUNT)) 539 return capable(CAP_SYS_ADMIN); 540 else 541 return ns_capable(fc->user_ns, CAP_SYS_ADMIN); 542 } 543 544 /** 545 * sget_fc - Find or create a superblock 546 * @fc: Filesystem context. 547 * @test: Comparison callback 548 * @set: Setup callback 549 * 550 * Find or create a superblock using the parameters stored in the filesystem 551 * context and the two callback functions. 552 * 553 * If an extant superblock is matched, then that will be returned with an 554 * elevated reference count that the caller must transfer or discard. 555 * 556 * If no match is made, a new superblock will be allocated and basic 557 * initialisation will be performed (s_type, s_fs_info and s_id will be set and 558 * the set() callback will be invoked), the superblock will be published and it 559 * will be returned in a partially constructed state with SB_BORN and SB_ACTIVE 560 * as yet unset. 561 */ 562 struct super_block *sget_fc(struct fs_context *fc, 563 int (*test)(struct super_block *, struct fs_context *), 564 int (*set)(struct super_block *, struct fs_context *)) 565 { 566 struct super_block *s = NULL; 567 struct super_block *old; 568 struct user_namespace *user_ns = fc->global ? &init_user_ns : fc->user_ns; 569 int err; 570 571 retry: 572 spin_lock(&sb_lock); 573 if (test) { 574 hlist_for_each_entry(old, &fc->fs_type->fs_supers, s_instances) { 575 if (test(old, fc)) 576 goto share_extant_sb; 577 } 578 } 579 if (!s) { 580 spin_unlock(&sb_lock); 581 s = alloc_super(fc->fs_type, fc->sb_flags, user_ns); 582 if (!s) 583 return ERR_PTR(-ENOMEM); 584 goto retry; 585 } 586 587 s->s_fs_info = fc->s_fs_info; 588 err = set(s, fc); 589 if (err) { 590 s->s_fs_info = NULL; 591 spin_unlock(&sb_lock); 592 destroy_unused_super(s); 593 return ERR_PTR(err); 594 } 595 fc->s_fs_info = NULL; 596 s->s_type = fc->fs_type; 597 s->s_iflags |= fc->s_iflags; 598 strlcpy(s->s_id, s->s_type->name, sizeof(s->s_id)); 599 list_add_tail(&s->s_list, &super_blocks); 600 hlist_add_head(&s->s_instances, &s->s_type->fs_supers); 601 spin_unlock(&sb_lock); 602 get_filesystem(s->s_type); 603 register_shrinker_prepared(&s->s_shrink); 604 return s; 605 606 share_extant_sb: 607 if (user_ns != old->s_user_ns) { 608 spin_unlock(&sb_lock); 609 destroy_unused_super(s); 610 return ERR_PTR(-EBUSY); 611 } 612 if (!grab_super(old)) 613 goto retry; 614 destroy_unused_super(s); 615 return old; 616 } 617 EXPORT_SYMBOL(sget_fc); 618 619 /** 620 * sget - find or create a superblock 621 * @type: filesystem type superblock should belong to 622 * @test: comparison callback 623 * @set: setup callback 624 * @flags: mount flags 625 * @data: argument to each of them 626 */ 627 struct super_block *sget(struct file_system_type *type, 628 int (*test)(struct super_block *,void *), 629 int (*set)(struct super_block *,void *), 630 int flags, 631 void *data) 632 { 633 struct user_namespace *user_ns = current_user_ns(); 634 struct super_block *s = NULL; 635 struct super_block *old; 636 int err; 637 638 /* We don't yet pass the user namespace of the parent 639 * mount through to here so always use &init_user_ns 640 * until that changes. 641 */ 642 if (flags & SB_SUBMOUNT) 643 user_ns = &init_user_ns; 644 645 retry: 646 spin_lock(&sb_lock); 647 if (test) { 648 hlist_for_each_entry(old, &type->fs_supers, s_instances) { 649 if (!test(old, data)) 650 continue; 651 if (user_ns != old->s_user_ns) { 652 spin_unlock(&sb_lock); 653 destroy_unused_super(s); 654 return ERR_PTR(-EBUSY); 655 } 656 if (!grab_super(old)) 657 goto retry; 658 destroy_unused_super(s); 659 return old; 660 } 661 } 662 if (!s) { 663 spin_unlock(&sb_lock); 664 s = alloc_super(type, (flags & ~SB_SUBMOUNT), user_ns); 665 if (!s) 666 return ERR_PTR(-ENOMEM); 667 goto retry; 668 } 669 670 err = set(s, data); 671 if (err) { 672 spin_unlock(&sb_lock); 673 destroy_unused_super(s); 674 return ERR_PTR(err); 675 } 676 s->s_type = type; 677 strlcpy(s->s_id, type->name, sizeof(s->s_id)); 678 list_add_tail(&s->s_list, &super_blocks); 679 hlist_add_head(&s->s_instances, &type->fs_supers); 680 spin_unlock(&sb_lock); 681 get_filesystem(type); 682 register_shrinker_prepared(&s->s_shrink); 683 return s; 684 } 685 EXPORT_SYMBOL(sget); 686 687 void drop_super(struct super_block *sb) 688 { 689 up_read(&sb->s_umount); 690 put_super(sb); 691 } 692 693 EXPORT_SYMBOL(drop_super); 694 695 void drop_super_exclusive(struct super_block *sb) 696 { 697 up_write(&sb->s_umount); 698 put_super(sb); 699 } 700 EXPORT_SYMBOL(drop_super_exclusive); 701 702 static void __iterate_supers(void (*f)(struct super_block *)) 703 { 704 struct super_block *sb, *p = NULL; 705 706 spin_lock(&sb_lock); 707 list_for_each_entry(sb, &super_blocks, s_list) { 708 if (hlist_unhashed(&sb->s_instances)) 709 continue; 710 sb->s_count++; 711 spin_unlock(&sb_lock); 712 713 f(sb); 714 715 spin_lock(&sb_lock); 716 if (p) 717 __put_super(p); 718 p = sb; 719 } 720 if (p) 721 __put_super(p); 722 spin_unlock(&sb_lock); 723 } 724 /** 725 * iterate_supers - call function for all active superblocks 726 * @f: function to call 727 * @arg: argument to pass to it 728 * 729 * Scans the superblock list and calls given function, passing it 730 * locked superblock and given argument. 731 */ 732 void iterate_supers(void (*f)(struct super_block *, void *), void *arg) 733 { 734 struct super_block *sb, *p = NULL; 735 736 spin_lock(&sb_lock); 737 list_for_each_entry(sb, &super_blocks, s_list) { 738 if (hlist_unhashed(&sb->s_instances)) 739 continue; 740 sb->s_count++; 741 spin_unlock(&sb_lock); 742 743 down_read(&sb->s_umount); 744 if (sb->s_root && (sb->s_flags & SB_BORN)) 745 f(sb, arg); 746 up_read(&sb->s_umount); 747 748 spin_lock(&sb_lock); 749 if (p) 750 __put_super(p); 751 p = sb; 752 } 753 if (p) 754 __put_super(p); 755 spin_unlock(&sb_lock); 756 } 757 758 /** 759 * iterate_supers_type - call function for superblocks of given type 760 * @type: fs type 761 * @f: function to call 762 * @arg: argument to pass to it 763 * 764 * Scans the superblock list and calls given function, passing it 765 * locked superblock and given argument. 766 */ 767 void iterate_supers_type(struct file_system_type *type, 768 void (*f)(struct super_block *, void *), void *arg) 769 { 770 struct super_block *sb, *p = NULL; 771 772 spin_lock(&sb_lock); 773 hlist_for_each_entry(sb, &type->fs_supers, s_instances) { 774 sb->s_count++; 775 spin_unlock(&sb_lock); 776 777 down_read(&sb->s_umount); 778 if (sb->s_root && (sb->s_flags & SB_BORN)) 779 f(sb, arg); 780 up_read(&sb->s_umount); 781 782 spin_lock(&sb_lock); 783 if (p) 784 __put_super(p); 785 p = sb; 786 } 787 if (p) 788 __put_super(p); 789 spin_unlock(&sb_lock); 790 } 791 792 EXPORT_SYMBOL(iterate_supers_type); 793 794 /** 795 * get_super - get the superblock of a device 796 * @bdev: device to get the superblock for 797 * 798 * Scans the superblock list and finds the superblock of the file system 799 * mounted on the device given. %NULL is returned if no match is found. 800 */ 801 struct super_block *get_super(struct block_device *bdev) 802 { 803 struct super_block *sb; 804 805 if (!bdev) 806 return NULL; 807 808 spin_lock(&sb_lock); 809 rescan: 810 list_for_each_entry(sb, &super_blocks, s_list) { 811 if (hlist_unhashed(&sb->s_instances)) 812 continue; 813 if (sb->s_bdev == bdev) { 814 sb->s_count++; 815 spin_unlock(&sb_lock); 816 down_read(&sb->s_umount); 817 /* still alive? */ 818 if (sb->s_root && (sb->s_flags & SB_BORN)) 819 return sb; 820 up_read(&sb->s_umount); 821 /* nope, got unmounted */ 822 spin_lock(&sb_lock); 823 __put_super(sb); 824 goto rescan; 825 } 826 } 827 spin_unlock(&sb_lock); 828 return NULL; 829 } 830 831 /** 832 * get_active_super - get an active reference to the superblock of a device 833 * @bdev: device to get the superblock for 834 * 835 * Scans the superblock list and finds the superblock of the file system 836 * mounted on the device given. Returns the superblock with an active 837 * reference or %NULL if none was found. 838 */ 839 struct super_block *get_active_super(struct block_device *bdev) 840 { 841 struct super_block *sb; 842 843 if (!bdev) 844 return NULL; 845 846 restart: 847 spin_lock(&sb_lock); 848 list_for_each_entry(sb, &super_blocks, s_list) { 849 if (hlist_unhashed(&sb->s_instances)) 850 continue; 851 if (sb->s_bdev == bdev) { 852 if (!grab_super(sb)) 853 goto restart; 854 up_write(&sb->s_umount); 855 return sb; 856 } 857 } 858 spin_unlock(&sb_lock); 859 return NULL; 860 } 861 862 struct super_block *user_get_super(dev_t dev, bool excl) 863 { 864 struct super_block *sb; 865 866 spin_lock(&sb_lock); 867 rescan: 868 list_for_each_entry(sb, &super_blocks, s_list) { 869 if (hlist_unhashed(&sb->s_instances)) 870 continue; 871 if (sb->s_dev == dev) { 872 sb->s_count++; 873 spin_unlock(&sb_lock); 874 if (excl) 875 down_write(&sb->s_umount); 876 else 877 down_read(&sb->s_umount); 878 /* still alive? */ 879 if (sb->s_root && (sb->s_flags & SB_BORN)) 880 return sb; 881 if (excl) 882 up_write(&sb->s_umount); 883 else 884 up_read(&sb->s_umount); 885 /* nope, got unmounted */ 886 spin_lock(&sb_lock); 887 __put_super(sb); 888 goto rescan; 889 } 890 } 891 spin_unlock(&sb_lock); 892 return NULL; 893 } 894 895 /** 896 * reconfigure_super - asks filesystem to change superblock parameters 897 * @fc: The superblock and configuration 898 * 899 * Alters the configuration parameters of a live superblock. 900 */ 901 int reconfigure_super(struct fs_context *fc) 902 { 903 struct super_block *sb = fc->root->d_sb; 904 int retval; 905 bool remount_ro = false; 906 bool force = fc->sb_flags & SB_FORCE; 907 908 if (fc->sb_flags_mask & ~MS_RMT_MASK) 909 return -EINVAL; 910 if (sb->s_writers.frozen != SB_UNFROZEN) 911 return -EBUSY; 912 913 retval = security_sb_remount(sb, fc->security); 914 if (retval) 915 return retval; 916 917 if (fc->sb_flags_mask & SB_RDONLY) { 918 #ifdef CONFIG_BLOCK 919 if (!(fc->sb_flags & SB_RDONLY) && sb->s_bdev && 920 bdev_read_only(sb->s_bdev)) 921 return -EACCES; 922 #endif 923 924 remount_ro = (fc->sb_flags & SB_RDONLY) && !sb_rdonly(sb); 925 } 926 927 if (remount_ro) { 928 if (!hlist_empty(&sb->s_pins)) { 929 up_write(&sb->s_umount); 930 group_pin_kill(&sb->s_pins); 931 down_write(&sb->s_umount); 932 if (!sb->s_root) 933 return 0; 934 if (sb->s_writers.frozen != SB_UNFROZEN) 935 return -EBUSY; 936 remount_ro = !sb_rdonly(sb); 937 } 938 } 939 shrink_dcache_sb(sb); 940 941 /* If we are reconfiguring to RDONLY and current sb is read/write, 942 * make sure there are no files open for writing. 943 */ 944 if (remount_ro) { 945 if (force) { 946 sb->s_readonly_remount = 1; 947 smp_wmb(); 948 } else { 949 retval = sb_prepare_remount_readonly(sb); 950 if (retval) 951 return retval; 952 } 953 } 954 955 if (fc->ops->reconfigure) { 956 retval = fc->ops->reconfigure(fc); 957 if (retval) { 958 if (!force) 959 goto cancel_readonly; 960 /* If forced remount, go ahead despite any errors */ 961 WARN(1, "forced remount of a %s fs returned %i\n", 962 sb->s_type->name, retval); 963 } 964 } 965 966 WRITE_ONCE(sb->s_flags, ((sb->s_flags & ~fc->sb_flags_mask) | 967 (fc->sb_flags & fc->sb_flags_mask))); 968 /* Needs to be ordered wrt mnt_is_readonly() */ 969 smp_wmb(); 970 sb->s_readonly_remount = 0; 971 972 /* 973 * Some filesystems modify their metadata via some other path than the 974 * bdev buffer cache (eg. use a private mapping, or directories in 975 * pagecache, etc). Also file data modifications go via their own 976 * mappings. So If we try to mount readonly then copy the filesystem 977 * from bdev, we could get stale data, so invalidate it to give a best 978 * effort at coherency. 979 */ 980 if (remount_ro && sb->s_bdev) 981 invalidate_bdev(sb->s_bdev); 982 return 0; 983 984 cancel_readonly: 985 sb->s_readonly_remount = 0; 986 return retval; 987 } 988 989 static void do_emergency_remount_callback(struct super_block *sb) 990 { 991 down_write(&sb->s_umount); 992 if (sb->s_root && sb->s_bdev && (sb->s_flags & SB_BORN) && 993 !sb_rdonly(sb)) { 994 struct fs_context *fc; 995 996 fc = fs_context_for_reconfigure(sb->s_root, 997 SB_RDONLY | SB_FORCE, SB_RDONLY); 998 if (!IS_ERR(fc)) { 999 if (parse_monolithic_mount_data(fc, NULL) == 0) 1000 (void)reconfigure_super(fc); 1001 put_fs_context(fc); 1002 } 1003 } 1004 up_write(&sb->s_umount); 1005 } 1006 1007 static void do_emergency_remount(struct work_struct *work) 1008 { 1009 __iterate_supers(do_emergency_remount_callback); 1010 kfree(work); 1011 printk("Emergency Remount complete\n"); 1012 } 1013 1014 void emergency_remount(void) 1015 { 1016 struct work_struct *work; 1017 1018 work = kmalloc(sizeof(*work), GFP_ATOMIC); 1019 if (work) { 1020 INIT_WORK(work, do_emergency_remount); 1021 schedule_work(work); 1022 } 1023 } 1024 1025 static void do_thaw_all_callback(struct super_block *sb) 1026 { 1027 down_write(&sb->s_umount); 1028 if (sb->s_root && sb->s_flags & SB_BORN) { 1029 emergency_thaw_bdev(sb); 1030 thaw_super_locked(sb); 1031 } else { 1032 up_write(&sb->s_umount); 1033 } 1034 } 1035 1036 static void do_thaw_all(struct work_struct *work) 1037 { 1038 __iterate_supers(do_thaw_all_callback); 1039 kfree(work); 1040 printk(KERN_WARNING "Emergency Thaw complete\n"); 1041 } 1042 1043 /** 1044 * emergency_thaw_all -- forcibly thaw every frozen filesystem 1045 * 1046 * Used for emergency unfreeze of all filesystems via SysRq 1047 */ 1048 void emergency_thaw_all(void) 1049 { 1050 struct work_struct *work; 1051 1052 work = kmalloc(sizeof(*work), GFP_ATOMIC); 1053 if (work) { 1054 INIT_WORK(work, do_thaw_all); 1055 schedule_work(work); 1056 } 1057 } 1058 1059 static DEFINE_IDA(unnamed_dev_ida); 1060 1061 /** 1062 * get_anon_bdev - Allocate a block device for filesystems which don't have one. 1063 * @p: Pointer to a dev_t. 1064 * 1065 * Filesystems which don't use real block devices can call this function 1066 * to allocate a virtual block device. 1067 * 1068 * Context: Any context. Frequently called while holding sb_lock. 1069 * Return: 0 on success, -EMFILE if there are no anonymous bdevs left 1070 * or -ENOMEM if memory allocation failed. 1071 */ 1072 int get_anon_bdev(dev_t *p) 1073 { 1074 int dev; 1075 1076 /* 1077 * Many userspace utilities consider an FSID of 0 invalid. 1078 * Always return at least 1 from get_anon_bdev. 1079 */ 1080 dev = ida_alloc_range(&unnamed_dev_ida, 1, (1 << MINORBITS) - 1, 1081 GFP_ATOMIC); 1082 if (dev == -ENOSPC) 1083 dev = -EMFILE; 1084 if (dev < 0) 1085 return dev; 1086 1087 *p = MKDEV(0, dev); 1088 return 0; 1089 } 1090 EXPORT_SYMBOL(get_anon_bdev); 1091 1092 void free_anon_bdev(dev_t dev) 1093 { 1094 ida_free(&unnamed_dev_ida, MINOR(dev)); 1095 } 1096 EXPORT_SYMBOL(free_anon_bdev); 1097 1098 int set_anon_super(struct super_block *s, void *data) 1099 { 1100 return get_anon_bdev(&s->s_dev); 1101 } 1102 EXPORT_SYMBOL(set_anon_super); 1103 1104 void kill_anon_super(struct super_block *sb) 1105 { 1106 dev_t dev = sb->s_dev; 1107 generic_shutdown_super(sb); 1108 free_anon_bdev(dev); 1109 } 1110 EXPORT_SYMBOL(kill_anon_super); 1111 1112 void kill_litter_super(struct super_block *sb) 1113 { 1114 if (sb->s_root) 1115 d_genocide(sb->s_root); 1116 kill_anon_super(sb); 1117 } 1118 EXPORT_SYMBOL(kill_litter_super); 1119 1120 int set_anon_super_fc(struct super_block *sb, struct fs_context *fc) 1121 { 1122 return set_anon_super(sb, NULL); 1123 } 1124 EXPORT_SYMBOL(set_anon_super_fc); 1125 1126 static int test_keyed_super(struct super_block *sb, struct fs_context *fc) 1127 { 1128 return sb->s_fs_info == fc->s_fs_info; 1129 } 1130 1131 static int test_single_super(struct super_block *s, struct fs_context *fc) 1132 { 1133 return 1; 1134 } 1135 1136 static int vfs_get_super(struct fs_context *fc, bool reconf, 1137 int (*test)(struct super_block *, struct fs_context *), 1138 int (*fill_super)(struct super_block *sb, 1139 struct fs_context *fc)) 1140 { 1141 struct super_block *sb; 1142 int err; 1143 1144 sb = sget_fc(fc, test, set_anon_super_fc); 1145 if (IS_ERR(sb)) 1146 return PTR_ERR(sb); 1147 1148 if (!sb->s_root) { 1149 err = fill_super(sb, fc); 1150 if (err) 1151 goto error; 1152 1153 sb->s_flags |= SB_ACTIVE; 1154 fc->root = dget(sb->s_root); 1155 } else { 1156 fc->root = dget(sb->s_root); 1157 if (reconf) { 1158 err = reconfigure_super(fc); 1159 if (err < 0) { 1160 dput(fc->root); 1161 fc->root = NULL; 1162 goto error; 1163 } 1164 } 1165 } 1166 1167 return 0; 1168 1169 error: 1170 deactivate_locked_super(sb); 1171 return err; 1172 } 1173 1174 int get_tree_nodev(struct fs_context *fc, 1175 int (*fill_super)(struct super_block *sb, 1176 struct fs_context *fc)) 1177 { 1178 return vfs_get_super(fc, false, NULL, fill_super); 1179 } 1180 EXPORT_SYMBOL(get_tree_nodev); 1181 1182 int get_tree_single(struct fs_context *fc, 1183 int (*fill_super)(struct super_block *sb, 1184 struct fs_context *fc)) 1185 { 1186 return vfs_get_super(fc, false, test_single_super, fill_super); 1187 } 1188 EXPORT_SYMBOL(get_tree_single); 1189 1190 int get_tree_single_reconf(struct fs_context *fc, 1191 int (*fill_super)(struct super_block *sb, 1192 struct fs_context *fc)) 1193 { 1194 return vfs_get_super(fc, true, test_single_super, fill_super); 1195 } 1196 EXPORT_SYMBOL(get_tree_single_reconf); 1197 1198 int get_tree_keyed(struct fs_context *fc, 1199 int (*fill_super)(struct super_block *sb, 1200 struct fs_context *fc), 1201 void *key) 1202 { 1203 fc->s_fs_info = key; 1204 return vfs_get_super(fc, false, test_keyed_super, fill_super); 1205 } 1206 EXPORT_SYMBOL(get_tree_keyed); 1207 1208 #ifdef CONFIG_BLOCK 1209 1210 static int set_bdev_super(struct super_block *s, void *data) 1211 { 1212 s->s_bdev = data; 1213 s->s_dev = s->s_bdev->bd_dev; 1214 s->s_bdi = bdi_get(s->s_bdev->bd_disk->bdi); 1215 1216 if (bdev_stable_writes(s->s_bdev)) 1217 s->s_iflags |= SB_I_STABLE_WRITES; 1218 return 0; 1219 } 1220 1221 static int set_bdev_super_fc(struct super_block *s, struct fs_context *fc) 1222 { 1223 return set_bdev_super(s, fc->sget_key); 1224 } 1225 1226 static int test_bdev_super_fc(struct super_block *s, struct fs_context *fc) 1227 { 1228 return !(s->s_iflags & SB_I_RETIRED) && s->s_bdev == fc->sget_key; 1229 } 1230 1231 /** 1232 * get_tree_bdev - Get a superblock based on a single block device 1233 * @fc: The filesystem context holding the parameters 1234 * @fill_super: Helper to initialise a new superblock 1235 */ 1236 int get_tree_bdev(struct fs_context *fc, 1237 int (*fill_super)(struct super_block *, 1238 struct fs_context *)) 1239 { 1240 struct block_device *bdev; 1241 struct super_block *s; 1242 fmode_t mode = FMODE_READ | FMODE_EXCL; 1243 int error = 0; 1244 1245 if (!(fc->sb_flags & SB_RDONLY)) 1246 mode |= FMODE_WRITE; 1247 1248 if (!fc->source) 1249 return invalf(fc, "No source specified"); 1250 1251 bdev = blkdev_get_by_path(fc->source, mode, fc->fs_type); 1252 if (IS_ERR(bdev)) { 1253 errorf(fc, "%s: Can't open blockdev", fc->source); 1254 return PTR_ERR(bdev); 1255 } 1256 1257 /* Once the superblock is inserted into the list by sget_fc(), s_umount 1258 * will protect the lockfs code from trying to start a snapshot while 1259 * we are mounting 1260 */ 1261 mutex_lock(&bdev->bd_fsfreeze_mutex); 1262 if (bdev->bd_fsfreeze_count > 0) { 1263 mutex_unlock(&bdev->bd_fsfreeze_mutex); 1264 warnf(fc, "%pg: Can't mount, blockdev is frozen", bdev); 1265 blkdev_put(bdev, mode); 1266 return -EBUSY; 1267 } 1268 1269 fc->sb_flags |= SB_NOSEC; 1270 fc->sget_key = bdev; 1271 s = sget_fc(fc, test_bdev_super_fc, set_bdev_super_fc); 1272 mutex_unlock(&bdev->bd_fsfreeze_mutex); 1273 if (IS_ERR(s)) { 1274 blkdev_put(bdev, mode); 1275 return PTR_ERR(s); 1276 } 1277 1278 if (s->s_root) { 1279 /* Don't summarily change the RO/RW state. */ 1280 if ((fc->sb_flags ^ s->s_flags) & SB_RDONLY) { 1281 warnf(fc, "%pg: Can't mount, would change RO state", bdev); 1282 deactivate_locked_super(s); 1283 blkdev_put(bdev, mode); 1284 return -EBUSY; 1285 } 1286 1287 /* 1288 * s_umount nests inside open_mutex during 1289 * __invalidate_device(). blkdev_put() acquires 1290 * open_mutex and can't be called under s_umount. Drop 1291 * s_umount temporarily. This is safe as we're 1292 * holding an active reference. 1293 */ 1294 up_write(&s->s_umount); 1295 blkdev_put(bdev, mode); 1296 down_write(&s->s_umount); 1297 } else { 1298 s->s_mode = mode; 1299 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev); 1300 shrinker_debugfs_rename(&s->s_shrink, "sb-%s:%s", 1301 fc->fs_type->name, s->s_id); 1302 sb_set_blocksize(s, block_size(bdev)); 1303 error = fill_super(s, fc); 1304 if (error) { 1305 deactivate_locked_super(s); 1306 return error; 1307 } 1308 1309 s->s_flags |= SB_ACTIVE; 1310 bdev->bd_super = s; 1311 } 1312 1313 BUG_ON(fc->root); 1314 fc->root = dget(s->s_root); 1315 return 0; 1316 } 1317 EXPORT_SYMBOL(get_tree_bdev); 1318 1319 static int test_bdev_super(struct super_block *s, void *data) 1320 { 1321 return !(s->s_iflags & SB_I_RETIRED) && (void *)s->s_bdev == data; 1322 } 1323 1324 struct dentry *mount_bdev(struct file_system_type *fs_type, 1325 int flags, const char *dev_name, void *data, 1326 int (*fill_super)(struct super_block *, void *, int)) 1327 { 1328 struct block_device *bdev; 1329 struct super_block *s; 1330 fmode_t mode = FMODE_READ | FMODE_EXCL; 1331 int error = 0; 1332 1333 if (!(flags & SB_RDONLY)) 1334 mode |= FMODE_WRITE; 1335 1336 bdev = blkdev_get_by_path(dev_name, mode, fs_type); 1337 if (IS_ERR(bdev)) 1338 return ERR_CAST(bdev); 1339 1340 /* 1341 * once the super is inserted into the list by sget, s_umount 1342 * will protect the lockfs code from trying to start a snapshot 1343 * while we are mounting 1344 */ 1345 mutex_lock(&bdev->bd_fsfreeze_mutex); 1346 if (bdev->bd_fsfreeze_count > 0) { 1347 mutex_unlock(&bdev->bd_fsfreeze_mutex); 1348 error = -EBUSY; 1349 goto error_bdev; 1350 } 1351 s = sget(fs_type, test_bdev_super, set_bdev_super, flags | SB_NOSEC, 1352 bdev); 1353 mutex_unlock(&bdev->bd_fsfreeze_mutex); 1354 if (IS_ERR(s)) 1355 goto error_s; 1356 1357 if (s->s_root) { 1358 if ((flags ^ s->s_flags) & SB_RDONLY) { 1359 deactivate_locked_super(s); 1360 error = -EBUSY; 1361 goto error_bdev; 1362 } 1363 1364 /* 1365 * s_umount nests inside open_mutex during 1366 * __invalidate_device(). blkdev_put() acquires 1367 * open_mutex and can't be called under s_umount. Drop 1368 * s_umount temporarily. This is safe as we're 1369 * holding an active reference. 1370 */ 1371 up_write(&s->s_umount); 1372 blkdev_put(bdev, mode); 1373 down_write(&s->s_umount); 1374 } else { 1375 s->s_mode = mode; 1376 snprintf(s->s_id, sizeof(s->s_id), "%pg", bdev); 1377 shrinker_debugfs_rename(&s->s_shrink, "sb-%s:%s", 1378 fs_type->name, s->s_id); 1379 sb_set_blocksize(s, block_size(bdev)); 1380 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0); 1381 if (error) { 1382 deactivate_locked_super(s); 1383 goto error; 1384 } 1385 1386 s->s_flags |= SB_ACTIVE; 1387 bdev->bd_super = s; 1388 } 1389 1390 return dget(s->s_root); 1391 1392 error_s: 1393 error = PTR_ERR(s); 1394 error_bdev: 1395 blkdev_put(bdev, mode); 1396 error: 1397 return ERR_PTR(error); 1398 } 1399 EXPORT_SYMBOL(mount_bdev); 1400 1401 void kill_block_super(struct super_block *sb) 1402 { 1403 struct block_device *bdev = sb->s_bdev; 1404 fmode_t mode = sb->s_mode; 1405 1406 bdev->bd_super = NULL; 1407 generic_shutdown_super(sb); 1408 sync_blockdev(bdev); 1409 WARN_ON_ONCE(!(mode & FMODE_EXCL)); 1410 blkdev_put(bdev, mode | FMODE_EXCL); 1411 } 1412 1413 EXPORT_SYMBOL(kill_block_super); 1414 #endif 1415 1416 struct dentry *mount_nodev(struct file_system_type *fs_type, 1417 int flags, void *data, 1418 int (*fill_super)(struct super_block *, void *, int)) 1419 { 1420 int error; 1421 struct super_block *s = sget(fs_type, NULL, set_anon_super, flags, NULL); 1422 1423 if (IS_ERR(s)) 1424 return ERR_CAST(s); 1425 1426 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0); 1427 if (error) { 1428 deactivate_locked_super(s); 1429 return ERR_PTR(error); 1430 } 1431 s->s_flags |= SB_ACTIVE; 1432 return dget(s->s_root); 1433 } 1434 EXPORT_SYMBOL(mount_nodev); 1435 1436 int reconfigure_single(struct super_block *s, 1437 int flags, void *data) 1438 { 1439 struct fs_context *fc; 1440 int ret; 1441 1442 /* The caller really need to be passing fc down into mount_single(), 1443 * then a chunk of this can be removed. [Bollocks -- AV] 1444 * Better yet, reconfiguration shouldn't happen, but rather the second 1445 * mount should be rejected if the parameters are not compatible. 1446 */ 1447 fc = fs_context_for_reconfigure(s->s_root, flags, MS_RMT_MASK); 1448 if (IS_ERR(fc)) 1449 return PTR_ERR(fc); 1450 1451 ret = parse_monolithic_mount_data(fc, data); 1452 if (ret < 0) 1453 goto out; 1454 1455 ret = reconfigure_super(fc); 1456 out: 1457 put_fs_context(fc); 1458 return ret; 1459 } 1460 1461 static int compare_single(struct super_block *s, void *p) 1462 { 1463 return 1; 1464 } 1465 1466 struct dentry *mount_single(struct file_system_type *fs_type, 1467 int flags, void *data, 1468 int (*fill_super)(struct super_block *, void *, int)) 1469 { 1470 struct super_block *s; 1471 int error; 1472 1473 s = sget(fs_type, compare_single, set_anon_super, flags, NULL); 1474 if (IS_ERR(s)) 1475 return ERR_CAST(s); 1476 if (!s->s_root) { 1477 error = fill_super(s, data, flags & SB_SILENT ? 1 : 0); 1478 if (!error) 1479 s->s_flags |= SB_ACTIVE; 1480 } else { 1481 error = reconfigure_single(s, flags, data); 1482 } 1483 if (unlikely(error)) { 1484 deactivate_locked_super(s); 1485 return ERR_PTR(error); 1486 } 1487 return dget(s->s_root); 1488 } 1489 EXPORT_SYMBOL(mount_single); 1490 1491 /** 1492 * vfs_get_tree - Get the mountable root 1493 * @fc: The superblock configuration context. 1494 * 1495 * The filesystem is invoked to get or create a superblock which can then later 1496 * be used for mounting. The filesystem places a pointer to the root to be 1497 * used for mounting in @fc->root. 1498 */ 1499 int vfs_get_tree(struct fs_context *fc) 1500 { 1501 struct super_block *sb; 1502 int error; 1503 1504 if (fc->root) 1505 return -EBUSY; 1506 1507 /* Get the mountable root in fc->root, with a ref on the root and a ref 1508 * on the superblock. 1509 */ 1510 error = fc->ops->get_tree(fc); 1511 if (error < 0) 1512 return error; 1513 1514 if (!fc->root) { 1515 pr_err("Filesystem %s get_tree() didn't set fc->root\n", 1516 fc->fs_type->name); 1517 /* We don't know what the locking state of the superblock is - 1518 * if there is a superblock. 1519 */ 1520 BUG(); 1521 } 1522 1523 sb = fc->root->d_sb; 1524 WARN_ON(!sb->s_bdi); 1525 1526 /* 1527 * Write barrier is for super_cache_count(). We place it before setting 1528 * SB_BORN as the data dependency between the two functions is the 1529 * superblock structure contents that we just set up, not the SB_BORN 1530 * flag. 1531 */ 1532 smp_wmb(); 1533 sb->s_flags |= SB_BORN; 1534 1535 error = security_sb_set_mnt_opts(sb, fc->security, 0, NULL); 1536 if (unlikely(error)) { 1537 fc_drop_locked(fc); 1538 return error; 1539 } 1540 1541 /* 1542 * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE 1543 * but s_maxbytes was an unsigned long long for many releases. Throw 1544 * this warning for a little while to try and catch filesystems that 1545 * violate this rule. 1546 */ 1547 WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to " 1548 "negative value (%lld)\n", fc->fs_type->name, sb->s_maxbytes); 1549 1550 return 0; 1551 } 1552 EXPORT_SYMBOL(vfs_get_tree); 1553 1554 /* 1555 * Setup private BDI for given superblock. It gets automatically cleaned up 1556 * in generic_shutdown_super(). 1557 */ 1558 int super_setup_bdi_name(struct super_block *sb, char *fmt, ...) 1559 { 1560 struct backing_dev_info *bdi; 1561 int err; 1562 va_list args; 1563 1564 bdi = bdi_alloc(NUMA_NO_NODE); 1565 if (!bdi) 1566 return -ENOMEM; 1567 1568 va_start(args, fmt); 1569 err = bdi_register_va(bdi, fmt, args); 1570 va_end(args); 1571 if (err) { 1572 bdi_put(bdi); 1573 return err; 1574 } 1575 WARN_ON(sb->s_bdi != &noop_backing_dev_info); 1576 sb->s_bdi = bdi; 1577 sb->s_iflags |= SB_I_PERSB_BDI; 1578 1579 return 0; 1580 } 1581 EXPORT_SYMBOL(super_setup_bdi_name); 1582 1583 /* 1584 * Setup private BDI for given superblock. I gets automatically cleaned up 1585 * in generic_shutdown_super(). 1586 */ 1587 int super_setup_bdi(struct super_block *sb) 1588 { 1589 static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0); 1590 1591 return super_setup_bdi_name(sb, "%.28s-%ld", sb->s_type->name, 1592 atomic_long_inc_return(&bdi_seq)); 1593 } 1594 EXPORT_SYMBOL(super_setup_bdi); 1595 1596 /** 1597 * sb_wait_write - wait until all writers to given file system finish 1598 * @sb: the super for which we wait 1599 * @level: type of writers we wait for (normal vs page fault) 1600 * 1601 * This function waits until there are no writers of given type to given file 1602 * system. 1603 */ 1604 static void sb_wait_write(struct super_block *sb, int level) 1605 { 1606 percpu_down_write(sb->s_writers.rw_sem + level-1); 1607 } 1608 1609 /* 1610 * We are going to return to userspace and forget about these locks, the 1611 * ownership goes to the caller of thaw_super() which does unlock(). 1612 */ 1613 static void lockdep_sb_freeze_release(struct super_block *sb) 1614 { 1615 int level; 1616 1617 for (level = SB_FREEZE_LEVELS - 1; level >= 0; level--) 1618 percpu_rwsem_release(sb->s_writers.rw_sem + level, 0, _THIS_IP_); 1619 } 1620 1621 /* 1622 * Tell lockdep we are holding these locks before we call ->unfreeze_fs(sb). 1623 */ 1624 static void lockdep_sb_freeze_acquire(struct super_block *sb) 1625 { 1626 int level; 1627 1628 for (level = 0; level < SB_FREEZE_LEVELS; ++level) 1629 percpu_rwsem_acquire(sb->s_writers.rw_sem + level, 0, _THIS_IP_); 1630 } 1631 1632 static void sb_freeze_unlock(struct super_block *sb, int level) 1633 { 1634 for (level--; level >= 0; level--) 1635 percpu_up_write(sb->s_writers.rw_sem + level); 1636 } 1637 1638 /** 1639 * freeze_super - lock the filesystem and force it into a consistent state 1640 * @sb: the super to lock 1641 * 1642 * Syncs the super to make sure the filesystem is consistent and calls the fs's 1643 * freeze_fs. Subsequent calls to this without first thawing the fs will return 1644 * -EBUSY. 1645 * 1646 * During this function, sb->s_writers.frozen goes through these values: 1647 * 1648 * SB_UNFROZEN: File system is normal, all writes progress as usual. 1649 * 1650 * SB_FREEZE_WRITE: The file system is in the process of being frozen. New 1651 * writes should be blocked, though page faults are still allowed. We wait for 1652 * all writes to complete and then proceed to the next stage. 1653 * 1654 * SB_FREEZE_PAGEFAULT: Freezing continues. Now also page faults are blocked 1655 * but internal fs threads can still modify the filesystem (although they 1656 * should not dirty new pages or inodes), writeback can run etc. After waiting 1657 * for all running page faults we sync the filesystem which will clean all 1658 * dirty pages and inodes (no new dirty pages or inodes can be created when 1659 * sync is running). 1660 * 1661 * SB_FREEZE_FS: The file system is frozen. Now all internal sources of fs 1662 * modification are blocked (e.g. XFS preallocation truncation on inode 1663 * reclaim). This is usually implemented by blocking new transactions for 1664 * filesystems that have them and need this additional guard. After all 1665 * internal writers are finished we call ->freeze_fs() to finish filesystem 1666 * freezing. Then we transition to SB_FREEZE_COMPLETE state. This state is 1667 * mostly auxiliary for filesystems to verify they do not modify frozen fs. 1668 * 1669 * sb->s_writers.frozen is protected by sb->s_umount. 1670 */ 1671 int freeze_super(struct super_block *sb) 1672 { 1673 int ret; 1674 1675 atomic_inc(&sb->s_active); 1676 down_write(&sb->s_umount); 1677 if (sb->s_writers.frozen != SB_UNFROZEN) { 1678 deactivate_locked_super(sb); 1679 return -EBUSY; 1680 } 1681 1682 if (!(sb->s_flags & SB_BORN)) { 1683 up_write(&sb->s_umount); 1684 return 0; /* sic - it's "nothing to do" */ 1685 } 1686 1687 if (sb_rdonly(sb)) { 1688 /* Nothing to do really... */ 1689 sb->s_writers.frozen = SB_FREEZE_COMPLETE; 1690 up_write(&sb->s_umount); 1691 return 0; 1692 } 1693 1694 sb->s_writers.frozen = SB_FREEZE_WRITE; 1695 /* Release s_umount to preserve sb_start_write -> s_umount ordering */ 1696 up_write(&sb->s_umount); 1697 sb_wait_write(sb, SB_FREEZE_WRITE); 1698 down_write(&sb->s_umount); 1699 1700 /* Now we go and block page faults... */ 1701 sb->s_writers.frozen = SB_FREEZE_PAGEFAULT; 1702 sb_wait_write(sb, SB_FREEZE_PAGEFAULT); 1703 1704 /* All writers are done so after syncing there won't be dirty data */ 1705 ret = sync_filesystem(sb); 1706 if (ret) { 1707 sb->s_writers.frozen = SB_UNFROZEN; 1708 sb_freeze_unlock(sb, SB_FREEZE_PAGEFAULT); 1709 wake_up(&sb->s_writers.wait_unfrozen); 1710 deactivate_locked_super(sb); 1711 return ret; 1712 } 1713 1714 /* Now wait for internal filesystem counter */ 1715 sb->s_writers.frozen = SB_FREEZE_FS; 1716 sb_wait_write(sb, SB_FREEZE_FS); 1717 1718 if (sb->s_op->freeze_fs) { 1719 ret = sb->s_op->freeze_fs(sb); 1720 if (ret) { 1721 printk(KERN_ERR 1722 "VFS:Filesystem freeze failed\n"); 1723 sb->s_writers.frozen = SB_UNFROZEN; 1724 sb_freeze_unlock(sb, SB_FREEZE_FS); 1725 wake_up(&sb->s_writers.wait_unfrozen); 1726 deactivate_locked_super(sb); 1727 return ret; 1728 } 1729 } 1730 /* 1731 * For debugging purposes so that fs can warn if it sees write activity 1732 * when frozen is set to SB_FREEZE_COMPLETE, and for thaw_super(). 1733 */ 1734 sb->s_writers.frozen = SB_FREEZE_COMPLETE; 1735 lockdep_sb_freeze_release(sb); 1736 up_write(&sb->s_umount); 1737 return 0; 1738 } 1739 EXPORT_SYMBOL(freeze_super); 1740 1741 static int thaw_super_locked(struct super_block *sb) 1742 { 1743 int error; 1744 1745 if (sb->s_writers.frozen != SB_FREEZE_COMPLETE) { 1746 up_write(&sb->s_umount); 1747 return -EINVAL; 1748 } 1749 1750 if (sb_rdonly(sb)) { 1751 sb->s_writers.frozen = SB_UNFROZEN; 1752 goto out; 1753 } 1754 1755 lockdep_sb_freeze_acquire(sb); 1756 1757 if (sb->s_op->unfreeze_fs) { 1758 error = sb->s_op->unfreeze_fs(sb); 1759 if (error) { 1760 printk(KERN_ERR 1761 "VFS:Filesystem thaw failed\n"); 1762 lockdep_sb_freeze_release(sb); 1763 up_write(&sb->s_umount); 1764 return error; 1765 } 1766 } 1767 1768 sb->s_writers.frozen = SB_UNFROZEN; 1769 sb_freeze_unlock(sb, SB_FREEZE_FS); 1770 out: 1771 wake_up(&sb->s_writers.wait_unfrozen); 1772 deactivate_locked_super(sb); 1773 return 0; 1774 } 1775 1776 /** 1777 * thaw_super -- unlock filesystem 1778 * @sb: the super to thaw 1779 * 1780 * Unlocks the filesystem and marks it writeable again after freeze_super(). 1781 */ 1782 int thaw_super(struct super_block *sb) 1783 { 1784 down_write(&sb->s_umount); 1785 return thaw_super_locked(sb); 1786 } 1787 EXPORT_SYMBOL(thaw_super); 1788 1789 /* 1790 * Create workqueue for deferred direct IO completions. We allocate the 1791 * workqueue when it's first needed. This avoids creating workqueue for 1792 * filesystems that don't need it and also allows us to create the workqueue 1793 * late enough so the we can include s_id in the name of the workqueue. 1794 */ 1795 int sb_init_dio_done_wq(struct super_block *sb) 1796 { 1797 struct workqueue_struct *old; 1798 struct workqueue_struct *wq = alloc_workqueue("dio/%s", 1799 WQ_MEM_RECLAIM, 0, 1800 sb->s_id); 1801 if (!wq) 1802 return -ENOMEM; 1803 /* 1804 * This has to be atomic as more DIOs can race to create the workqueue 1805 */ 1806 old = cmpxchg(&sb->s_dio_done_wq, NULL, wq); 1807 /* Someone created workqueue before us? Free ours... */ 1808 if (old) 1809 destroy_workqueue(wq); 1810 return 0; 1811 } 1812