linux/fs/super.c
<<
>>
Prefs
   1/*
   2 *  linux/fs/super.c
   3 *
   4 *  Copyright (C) 1991, 1992  Linus Torvalds
   5 *
   6 *  super.c contains code to handle: - mount structures
   7 *                                   - super-block tables
   8 *                                   - filesystem drivers list
   9 *                                   - mount system call
  10 *                                   - umount system call
  11 *                                   - ustat system call
  12 *
  13 * GK 2/5/95  -  Changed to support mounting the root fs via NFS
  14 *
  15 *  Added kerneld support: Jacques Gelinas and Bjorn Ekwall
  16 *  Added change_root: Werner Almesberger & Hans Lermen, Feb '96
  17 *  Added options to /proc/mounts:
  18 *    Torbjörn Lindh (torbjorn.lindh@gopta.se), April 14, 1996.
  19 *  Added devfs support: Richard Gooch <rgooch@atnf.csiro.au>, 13-JAN-1998
  20 *  Heavily rewritten for 'one fs - one tree' dcache architecture. AV, Mar 2000
  21 */
  22
  23#include <linux/export.h>
  24#include <linux/slab.h>
  25#include <linux/acct.h>
  26#include <linux/blkdev.h>
  27#include <linux/mount.h>
  28#include <linux/security.h>
  29#include <linux/writeback.h>            /* for the emergency remount stuff */
  30#include <linux/idr.h>
  31#include <linux/mutex.h>
  32#include <linux/backing-dev.h>
  33#include <linux/rculist_bl.h>
  34#include <linux/cleancache.h>
  35#include <linux/fsnotify.h>
  36#include "internal.h"
  37
  38
  39LIST_HEAD(super_blocks);
  40DEFINE_SPINLOCK(sb_lock);
  41
  42/*
  43 * One thing we have to be careful of with a per-sb shrinker is that we don't
  44 * drop the last active reference to the superblock from within the shrinker.
  45 * If that happens we could trigger unregistering the shrinker from within the
  46 * shrinker path and that leads to deadlock on the shrinker_rwsem. Hence we
  47 * take a passive reference to the superblock to avoid this from occurring.
  48 */
  49static int prune_super(struct shrinker *shrink, struct shrink_control *sc)
  50{
  51        struct super_block *sb;
  52        int     fs_objects = 0;
  53        int     total_objects;
  54
  55        sb = container_of(shrink, struct super_block, s_shrink);
  56
  57        /*
  58         * Deadlock avoidance.  We may hold various FS locks, and we don't want
  59         * to recurse into the FS that called us in clear_inode() and friends..
  60         */
  61        if (sc->nr_to_scan && !(sc->gfp_mask & __GFP_FS))
  62                return -1;
  63
  64        if (!grab_super_passive(sb))
  65                return !sc->nr_to_scan ? 0 : -1;
  66
  67        if (sb->s_op && sb->s_op->nr_cached_objects)
  68                fs_objects = sb->s_op->nr_cached_objects(sb);
  69
  70        total_objects = sb->s_nr_dentry_unused +
  71                        sb->s_nr_inodes_unused + fs_objects + 1;
  72
  73        if (sc->nr_to_scan) {
  74                int     dentries;
  75                int     inodes;
  76
  77                /* proportion the scan between the caches */
  78                dentries = (sc->nr_to_scan * sb->s_nr_dentry_unused) /
  79                                                        total_objects;
  80                inodes = (sc->nr_to_scan * sb->s_nr_inodes_unused) /
  81                                                        total_objects;
  82                if (fs_objects)
  83                        fs_objects = (sc->nr_to_scan * fs_objects) /
  84                                                        total_objects;
  85                /*
  86                 * prune the dcache first as the icache is pinned by it, then
  87                 * prune the icache, followed by the filesystem specific caches
  88                 */
  89                prune_dcache_sb(sb, dentries);
  90                prune_icache_sb(sb, inodes);
  91
  92                if (fs_objects && sb->s_op->free_cached_objects) {
  93                        sb->s_op->free_cached_objects(sb, fs_objects);
  94                        fs_objects = sb->s_op->nr_cached_objects(sb);
  95                }
  96                total_objects = sb->s_nr_dentry_unused +
  97                                sb->s_nr_inodes_unused + fs_objects;
  98        }
  99
 100        total_objects = (total_objects / 100) * sysctl_vfs_cache_pressure;
 101        drop_super(sb);
 102        return total_objects;
 103}
 104
 105/**
 106 *      alloc_super     -       create new superblock
 107 *      @type:  filesystem type superblock should belong to
 108 *
 109 *      Allocates and initializes a new &struct super_block.  alloc_super()
 110 *      returns a pointer new superblock or %NULL if allocation had failed.
 111 */
 112static struct super_block *alloc_super(struct file_system_type *type)
 113{
 114        struct super_block *s = kzalloc(sizeof(struct super_block),  GFP_USER);
 115        static const struct super_operations default_op;
 116
 117        if (s) {
 118                if (security_sb_alloc(s)) {
 119                        kfree(s);
 120                        s = NULL;
 121                        goto out;
 122                }
 123#ifdef CONFIG_SMP
 124                s->s_files = alloc_percpu(struct list_head);
 125                if (!s->s_files) {
 126                        security_sb_free(s);
 127                        kfree(s);
 128                        s = NULL;
 129                        goto out;
 130                } else {
 131                        int i;
 132
 133                        for_each_possible_cpu(i)
 134                                INIT_LIST_HEAD(per_cpu_ptr(s->s_files, i));
 135                }
 136#else
 137                INIT_LIST_HEAD(&s->s_files);
 138#endif
 139                s->s_bdi = &default_backing_dev_info;
 140                INIT_HLIST_NODE(&s->s_instances);
 141                INIT_HLIST_BL_HEAD(&s->s_anon);
 142                INIT_LIST_HEAD(&s->s_inodes);
 143                INIT_LIST_HEAD(&s->s_dentry_lru);
 144                INIT_LIST_HEAD(&s->s_inode_lru);
 145                spin_lock_init(&s->s_inode_lru_lock);
 146                INIT_LIST_HEAD(&s->s_mounts);
 147                init_rwsem(&s->s_umount);
 148                mutex_init(&s->s_lock);
 149                lockdep_set_class(&s->s_umount, &type->s_umount_key);
 150                /*
 151                 * The locking rules for s_lock are up to the
 152                 * filesystem. For example ext3fs has different
 153                 * lock ordering than usbfs:
 154                 */
 155                lockdep_set_class(&s->s_lock, &type->s_lock_key);
 156                /*
 157                 * sget() can have s_umount recursion.
 158                 *
 159                 * When it cannot find a suitable sb, it allocates a new
 160                 * one (this one), and tries again to find a suitable old
 161                 * one.
 162                 *
 163                 * In case that succeeds, it will acquire the s_umount
 164                 * lock of the old one. Since these are clearly distrinct
 165                 * locks, and this object isn't exposed yet, there's no
 166                 * risk of deadlocks.
 167                 *
 168                 * Annotate this by putting this lock in a different
 169                 * subclass.
 170                 */
 171                down_write_nested(&s->s_umount, SINGLE_DEPTH_NESTING);
 172                s->s_count = 1;
 173                atomic_set(&s->s_active, 1);
 174                mutex_init(&s->s_vfs_rename_mutex);
 175                lockdep_set_class(&s->s_vfs_rename_mutex, &type->s_vfs_rename_key);
 176                mutex_init(&s->s_dquot.dqio_mutex);
 177                mutex_init(&s->s_dquot.dqonoff_mutex);
 178                init_rwsem(&s->s_dquot.dqptr_sem);
 179                init_waitqueue_head(&s->s_wait_unfrozen);
 180                s->s_maxbytes = MAX_NON_LFS;
 181                s->s_op = &default_op;
 182                s->s_time_gran = 1000000000;
 183                s->cleancache_poolid = -1;
 184
 185                s->s_shrink.seeks = DEFAULT_SEEKS;
 186                s->s_shrink.shrink = prune_super;
 187                s->s_shrink.batch = 1024;
 188        }
 189out:
 190        return s;
 191}
 192
 193/**
 194 *      destroy_super   -       frees a superblock
 195 *      @s: superblock to free
 196 *
 197 *      Frees a superblock.
 198 */
 199static inline void destroy_super(struct super_block *s)
 200{
 201#ifdef CONFIG_SMP
 202        free_percpu(s->s_files);
 203#endif
 204        security_sb_free(s);
 205        WARN_ON(!list_empty(&s->s_mounts));
 206        kfree(s->s_subtype);
 207        kfree(s->s_options);
 208        kfree(s);
 209}
 210
 211/* Superblock refcounting  */
 212
 213/*
 214 * Drop a superblock's refcount.  The caller must hold sb_lock.
 215 */
 216static void __put_super(struct super_block *sb)
 217{
 218        if (!--sb->s_count) {
 219                list_del_init(&sb->s_list);
 220                destroy_super(sb);
 221        }
 222}
 223
 224/**
 225 *      put_super       -       drop a temporary reference to superblock
 226 *      @sb: superblock in question
 227 *
 228 *      Drops a temporary reference, frees superblock if there's no
 229 *      references left.
 230 */
 231static void put_super(struct super_block *sb)
 232{
 233        spin_lock(&sb_lock);
 234        __put_super(sb);
 235        spin_unlock(&sb_lock);
 236}
 237
 238
 239/**
 240 *      deactivate_locked_super -       drop an active reference to superblock
 241 *      @s: superblock to deactivate
 242 *
 243 *      Drops an active reference to superblock, converting it into a temprory
 244 *      one if there is no other active references left.  In that case we
 245 *      tell fs driver to shut it down and drop the temporary reference we
 246 *      had just acquired.
 247 *
 248 *      Caller holds exclusive lock on superblock; that lock is released.
 249 */
 250void deactivate_locked_super(struct super_block *s)
 251{
 252        struct file_system_type *fs = s->s_type;
 253        if (atomic_dec_and_test(&s->s_active)) {
 254                cleancache_invalidate_fs(s);
 255                fs->kill_sb(s);
 256
 257                /* caches are now gone, we can safely kill the shrinker now */
 258                unregister_shrinker(&s->s_shrink);
 259
 260                /*
 261                 * We need to call rcu_barrier so all the delayed rcu free
 262                 * inodes are flushed before we release the fs module.
 263                 */
 264                rcu_barrier();
 265                put_filesystem(fs);
 266                put_super(s);
 267        } else {
 268                up_write(&s->s_umount);
 269        }
 270}
 271
 272EXPORT_SYMBOL(deactivate_locked_super);
 273
 274/**
 275 *      deactivate_super        -       drop an active reference to superblock
 276 *      @s: superblock to deactivate
 277 *
 278 *      Variant of deactivate_locked_super(), except that superblock is *not*
 279 *      locked by caller.  If we are going to drop the final active reference,
 280 *      lock will be acquired prior to that.
 281 */
 282void deactivate_super(struct super_block *s)
 283{
 284        if (!atomic_add_unless(&s->s_active, -1, 1)) {
 285                down_write(&s->s_umount);
 286                deactivate_locked_super(s);
 287        }
 288}
 289
 290EXPORT_SYMBOL(deactivate_super);
 291
 292/**
 293 *      grab_super - acquire an active reference
 294 *      @s: reference we are trying to make active
 295 *
 296 *      Tries to acquire an active reference.  grab_super() is used when we
 297 *      had just found a superblock in super_blocks or fs_type->fs_supers
 298 *      and want to turn it into a full-blown active reference.  grab_super()
 299 *      is called with sb_lock held and drops it.  Returns 1 in case of
 300 *      success, 0 if we had failed (superblock contents was already dead or
 301 *      dying when grab_super() had been called).
 302 */
 303static int grab_super(struct super_block *s) __releases(sb_lock)
 304{
 305        if (atomic_inc_not_zero(&s->s_active)) {
 306                spin_unlock(&sb_lock);
 307                return 1;
 308        }
 309        /* it's going away */
 310        s->s_count++;
 311        spin_unlock(&sb_lock);
 312        /* wait for it to die */
 313        down_write(&s->s_umount);
 314        up_write(&s->s_umount);
 315        put_super(s);
 316        return 0;
 317}
 318
 319/*
 320 *      grab_super_passive - acquire a passive reference
 321 *      @s: reference we are trying to grab
 322 *
 323 *      Tries to acquire a passive reference. This is used in places where we
 324 *      cannot take an active reference but we need to ensure that the
 325 *      superblock does not go away while we are working on it. It returns
 326 *      false if a reference was not gained, and returns true with the s_umount
 327 *      lock held in read mode if a reference is gained. On successful return,
 328 *      the caller must drop the s_umount lock and the passive reference when
 329 *      done.
 330 */
 331bool grab_super_passive(struct super_block *sb)
 332{
 333        spin_lock(&sb_lock);
 334        if (hlist_unhashed(&sb->s_instances)) {
 335                spin_unlock(&sb_lock);
 336                return false;
 337        }
 338
 339        sb->s_count++;
 340        spin_unlock(&sb_lock);
 341
 342        if (down_read_trylock(&sb->s_umount)) {
 343                if (sb->s_root && (sb->s_flags & MS_BORN))
 344                        return true;
 345                up_read(&sb->s_umount);
 346        }
 347
 348        put_super(sb);
 349        return false;
 350}
 351
 352/*
 353 * Superblock locking.  We really ought to get rid of these two.
 354 */
 355void lock_super(struct super_block * sb)
 356{
 357        mutex_lock(&sb->s_lock);
 358}
 359
 360void unlock_super(struct super_block * sb)
 361{
 362        mutex_unlock(&sb->s_lock);
 363}
 364
 365EXPORT_SYMBOL(lock_super);
 366EXPORT_SYMBOL(unlock_super);
 367
 368/**
 369 *      generic_shutdown_super  -       common helper for ->kill_sb()
 370 *      @sb: superblock to kill
 371 *
 372 *      generic_shutdown_super() does all fs-independent work on superblock
 373 *      shutdown.  Typical ->kill_sb() should pick all fs-specific objects
 374 *      that need destruction out of superblock, call generic_shutdown_super()
 375 *      and release aforementioned objects.  Note: dentries and inodes _are_
 376 *      taken care of and do not need specific handling.
 377 *
 378 *      Upon calling this function, the filesystem may no longer alter or
 379 *      rearrange the set of dentries belonging to this super_block, nor may it
 380 *      change the attachments of dentries to inodes.
 381 */
 382void generic_shutdown_super(struct super_block *sb)
 383{
 384        const struct super_operations *sop = sb->s_op;
 385
 386        if (sb->s_root) {
 387                shrink_dcache_for_umount(sb);
 388                sync_filesystem(sb);
 389                sb->s_flags &= ~MS_ACTIVE;
 390
 391                fsnotify_unmount_inodes(&sb->s_inodes);
 392
 393                evict_inodes(sb);
 394
 395                if (sop->put_super)
 396                        sop->put_super(sb);
 397
 398                if (!list_empty(&sb->s_inodes)) {
 399                        printk("VFS: Busy inodes after unmount of %s. "
 400                           "Self-destruct in 5 seconds.  Have a nice day...\n",
 401                           sb->s_id);
 402                }
 403        }
 404        spin_lock(&sb_lock);
 405        /* should be initialized for __put_super_and_need_restart() */
 406        hlist_del_init(&sb->s_instances);
 407        spin_unlock(&sb_lock);
 408        up_write(&sb->s_umount);
 409}
 410
 411EXPORT_SYMBOL(generic_shutdown_super);
 412
 413/**
 414 *      sget    -       find or create a superblock
 415 *      @type:  filesystem type superblock should belong to
 416 *      @test:  comparison callback
 417 *      @set:   setup callback
 418 *      @data:  argument to each of them
 419 */
 420struct super_block *sget(struct file_system_type *type,
 421                        int (*test)(struct super_block *,void *),
 422                        int (*set)(struct super_block *,void *),
 423                        void *data)
 424{
 425        struct super_block *s = NULL;
 426        struct hlist_node *node;
 427        struct super_block *old;
 428        int err;
 429
 430retry:
 431        spin_lock(&sb_lock);
 432        if (test) {
 433                hlist_for_each_entry(old, node, &type->fs_supers, s_instances) {
 434                        if (!test(old, data))
 435                                continue;
 436                        if (!grab_super(old))
 437                                goto retry;
 438                        if (s) {
 439                                up_write(&s->s_umount);
 440                                destroy_super(s);
 441                                s = NULL;
 442                        }
 443                        down_write(&old->s_umount);
 444                        if (unlikely(!(old->s_flags & MS_BORN))) {
 445                                deactivate_locked_super(old);
 446                                goto retry;
 447                        }
 448                        return old;
 449                }
 450        }
 451        if (!s) {
 452                spin_unlock(&sb_lock);
 453                s = alloc_super(type);
 454                if (!s)
 455                        return ERR_PTR(-ENOMEM);
 456                goto retry;
 457        }
 458                
 459        err = set(s, data);
 460        if (err) {
 461                spin_unlock(&sb_lock);
 462                up_write(&s->s_umount);
 463                destroy_super(s);
 464                return ERR_PTR(err);
 465        }
 466        s->s_type = type;
 467        strlcpy(s->s_id, type->name, sizeof(s->s_id));
 468        list_add_tail(&s->s_list, &super_blocks);
 469        hlist_add_head(&s->s_instances, &type->fs_supers);
 470        spin_unlock(&sb_lock);
 471        get_filesystem(type);
 472        register_shrinker(&s->s_shrink);
 473        return s;
 474}
 475
 476EXPORT_SYMBOL(sget);
 477
 478void drop_super(struct super_block *sb)
 479{
 480        up_read(&sb->s_umount);
 481        put_super(sb);
 482}
 483
 484EXPORT_SYMBOL(drop_super);
 485
 486/**
 487 * sync_supers - helper for periodic superblock writeback
 488 *
 489 * Call the write_super method if present on all dirty superblocks in
 490 * the system.  This is for the periodic writeback used by most older
 491 * filesystems.  For data integrity superblock writeback use
 492 * sync_filesystems() instead.
 493 *
 494 * Note: check the dirty flag before waiting, so we don't
 495 * hold up the sync while mounting a device. (The newly
 496 * mounted device won't need syncing.)
 497 */
 498void sync_supers(void)
 499{
 500        struct super_block *sb, *p = NULL;
 501
 502        spin_lock(&sb_lock);
 503        list_for_each_entry(sb, &super_blocks, s_list) {
 504                if (hlist_unhashed(&sb->s_instances))
 505                        continue;
 506                if (sb->s_op->write_super && sb->s_dirt) {
 507                        sb->s_count++;
 508                        spin_unlock(&sb_lock);
 509
 510                        down_read(&sb->s_umount);
 511                        if (sb->s_root && sb->s_dirt && (sb->s_flags & MS_BORN))
 512                                sb->s_op->write_super(sb);
 513                        up_read(&sb->s_umount);
 514
 515                        spin_lock(&sb_lock);
 516                        if (p)
 517                                __put_super(p);
 518                        p = sb;
 519                }
 520        }
 521        if (p)
 522                __put_super(p);
 523        spin_unlock(&sb_lock);
 524}
 525
 526/**
 527 *      iterate_supers - call function for all active superblocks
 528 *      @f: function to call
 529 *      @arg: argument to pass to it
 530 *
 531 *      Scans the superblock list and calls given function, passing it
 532 *      locked superblock and given argument.
 533 */
 534void iterate_supers(void (*f)(struct super_block *, void *), void *arg)
 535{
 536        struct super_block *sb, *p = NULL;
 537
 538        spin_lock(&sb_lock);
 539        list_for_each_entry(sb, &super_blocks, s_list) {
 540                if (hlist_unhashed(&sb->s_instances))
 541                        continue;
 542                sb->s_count++;
 543                spin_unlock(&sb_lock);
 544
 545                down_read(&sb->s_umount);
 546                if (sb->s_root && (sb->s_flags & MS_BORN))
 547                        f(sb, arg);
 548                up_read(&sb->s_umount);
 549
 550                spin_lock(&sb_lock);
 551                if (p)
 552                        __put_super(p);
 553                p = sb;
 554        }
 555        if (p)
 556                __put_super(p);
 557        spin_unlock(&sb_lock);
 558}
 559
 560/**
 561 *      iterate_supers_type - call function for superblocks of given type
 562 *      @type: fs type
 563 *      @f: function to call
 564 *      @arg: argument to pass to it
 565 *
 566 *      Scans the superblock list and calls given function, passing it
 567 *      locked superblock and given argument.
 568 */
 569void iterate_supers_type(struct file_system_type *type,
 570        void (*f)(struct super_block *, void *), void *arg)
 571{
 572        struct super_block *sb, *p = NULL;
 573        struct hlist_node *node;
 574
 575        spin_lock(&sb_lock);
 576        hlist_for_each_entry(sb, node, &type->fs_supers, s_instances) {
 577                sb->s_count++;
 578                spin_unlock(&sb_lock);
 579
 580                down_read(&sb->s_umount);
 581                if (sb->s_root && (sb->s_flags & MS_BORN))
 582                        f(sb, arg);
 583                up_read(&sb->s_umount);
 584
 585                spin_lock(&sb_lock);
 586                if (p)
 587                        __put_super(p);
 588                p = sb;
 589        }
 590        if (p)
 591                __put_super(p);
 592        spin_unlock(&sb_lock);
 593}
 594
 595EXPORT_SYMBOL(iterate_supers_type);
 596
 597/**
 598 *      get_super - get the superblock of a device
 599 *      @bdev: device to get the superblock for
 600 *      
 601 *      Scans the superblock list and finds the superblock of the file system
 602 *      mounted on the device given. %NULL is returned if no match is found.
 603 */
 604
 605struct super_block *get_super(struct block_device *bdev)
 606{
 607        struct super_block *sb;
 608
 609        if (!bdev)
 610                return NULL;
 611
 612        spin_lock(&sb_lock);
 613rescan:
 614        list_for_each_entry(sb, &super_blocks, s_list) {
 615                if (hlist_unhashed(&sb->s_instances))
 616                        continue;
 617                if (sb->s_bdev == bdev) {
 618                        sb->s_count++;
 619                        spin_unlock(&sb_lock);
 620                        down_read(&sb->s_umount);
 621                        /* still alive? */
 622                        if (sb->s_root && (sb->s_flags & MS_BORN))
 623                                return sb;
 624                        up_read(&sb->s_umount);
 625                        /* nope, got unmounted */
 626                        spin_lock(&sb_lock);
 627                        __put_super(sb);
 628                        goto rescan;
 629                }
 630        }
 631        spin_unlock(&sb_lock);
 632        return NULL;
 633}
 634
 635EXPORT_SYMBOL(get_super);
 636
 637/**
 638 *      get_super_thawed - get thawed superblock of a device
 639 *      @bdev: device to get the superblock for
 640 *
 641 *      Scans the superblock list and finds the superblock of the file system
 642 *      mounted on the device. The superblock is returned once it is thawed
 643 *      (or immediately if it was not frozen). %NULL is returned if no match
 644 *      is found.
 645 */
 646struct super_block *get_super_thawed(struct block_device *bdev)
 647{
 648        while (1) {
 649                struct super_block *s = get_super(bdev);
 650                if (!s || s->s_frozen == SB_UNFROZEN)
 651                        return s;
 652                up_read(&s->s_umount);
 653                vfs_check_frozen(s, SB_FREEZE_WRITE);
 654                put_super(s);
 655        }
 656}
 657EXPORT_SYMBOL(get_super_thawed);
 658
 659/**
 660 * get_active_super - get an active reference to the superblock of a device
 661 * @bdev: device to get the superblock for
 662 *
 663 * Scans the superblock list and finds the superblock of the file system
 664 * mounted on the device given.  Returns the superblock with an active
 665 * reference or %NULL if none was found.
 666 */
 667struct super_block *get_active_super(struct block_device *bdev)
 668{
 669        struct super_block *sb;
 670
 671        if (!bdev)
 672                return NULL;
 673
 674restart:
 675        spin_lock(&sb_lock);
 676        list_for_each_entry(sb, &super_blocks, s_list) {
 677                if (hlist_unhashed(&sb->s_instances))
 678                        continue;
 679                if (sb->s_bdev == bdev) {
 680                        if (grab_super(sb)) /* drops sb_lock */
 681                                return sb;
 682                        else
 683                                goto restart;
 684                }
 685        }
 686        spin_unlock(&sb_lock);
 687        return NULL;
 688}
 689 
 690struct super_block *user_get_super(dev_t dev)
 691{
 692        struct super_block *sb;
 693
 694        spin_lock(&sb_lock);
 695rescan:
 696        list_for_each_entry(sb, &super_blocks, s_list) {
 697                if (hlist_unhashed(&sb->s_instances))
 698                        continue;
 699                if (sb->s_dev ==  dev) {
 700                        sb->s_count++;
 701                        spin_unlock(&sb_lock);
 702                        down_read(&sb->s_umount);
 703                        /* still alive? */
 704                        if (sb->s_root && (sb->s_flags & MS_BORN))
 705                                return sb;
 706                        up_read(&sb->s_umount);
 707                        /* nope, got unmounted */
 708                        spin_lock(&sb_lock);
 709                        __put_super(sb);
 710                        goto rescan;
 711                }
 712        }
 713        spin_unlock(&sb_lock);
 714        return NULL;
 715}
 716
 717/**
 718 *      do_remount_sb - asks filesystem to change mount options.
 719 *      @sb:    superblock in question
 720 *      @flags: numeric part of options
 721 *      @data:  the rest of options
 722 *      @force: whether or not to force the change
 723 *
 724 *      Alters the mount options of a mounted file system.
 725 */
 726int do_remount_sb(struct super_block *sb, int flags, void *data, int force)
 727{
 728        int retval;
 729        int remount_ro;
 730
 731        if (sb->s_frozen != SB_UNFROZEN)
 732                return -EBUSY;
 733
 734#ifdef CONFIG_BLOCK
 735        if (!(flags & MS_RDONLY) && bdev_read_only(sb->s_bdev))
 736                return -EACCES;
 737#endif
 738
 739        if (flags & MS_RDONLY)
 740                acct_auto_close(sb);
 741        shrink_dcache_sb(sb);
 742        sync_filesystem(sb);
 743
 744        remount_ro = (flags & MS_RDONLY) && !(sb->s_flags & MS_RDONLY);
 745
 746        /* If we are remounting RDONLY and current sb is read/write,
 747           make sure there are no rw files opened */
 748        if (remount_ro) {
 749                if (force) {
 750                        mark_files_ro(sb);
 751                } else {
 752                        retval = sb_prepare_remount_readonly(sb);
 753                        if (retval)
 754                                return retval;
 755                }
 756        }
 757
 758        if (sb->s_op->remount_fs) {
 759                retval = sb->s_op->remount_fs(sb, &flags, data);
 760                if (retval) {
 761                        if (!force)
 762                                goto cancel_readonly;
 763                        /* If forced remount, go ahead despite any errors */
 764                        WARN(1, "forced remount of a %s fs returned %i\n",
 765                             sb->s_type->name, retval);
 766                }
 767        }
 768        sb->s_flags = (sb->s_flags & ~MS_RMT_MASK) | (flags & MS_RMT_MASK);
 769        /* Needs to be ordered wrt mnt_is_readonly() */
 770        smp_wmb();
 771        sb->s_readonly_remount = 0;
 772
 773        /*
 774         * Some filesystems modify their metadata via some other path than the
 775         * bdev buffer cache (eg. use a private mapping, or directories in
 776         * pagecache, etc). Also file data modifications go via their own
 777         * mappings. So If we try to mount readonly then copy the filesystem
 778         * from bdev, we could get stale data, so invalidate it to give a best
 779         * effort at coherency.
 780         */
 781        if (remount_ro && sb->s_bdev)
 782                invalidate_bdev(sb->s_bdev);
 783        return 0;
 784
 785cancel_readonly:
 786        sb->s_readonly_remount = 0;
 787        return retval;
 788}
 789
 790static void do_emergency_remount(struct work_struct *work)
 791{
 792        struct super_block *sb, *p = NULL;
 793
 794        spin_lock(&sb_lock);
 795        list_for_each_entry(sb, &super_blocks, s_list) {
 796                if (hlist_unhashed(&sb->s_instances))
 797                        continue;
 798                sb->s_count++;
 799                spin_unlock(&sb_lock);
 800                down_write(&sb->s_umount);
 801                if (sb->s_root && sb->s_bdev && (sb->s_flags & MS_BORN) &&
 802                    !(sb->s_flags & MS_RDONLY)) {
 803                        /*
 804                         * What lock protects sb->s_flags??
 805                         */
 806                        do_remount_sb(sb, MS_RDONLY, NULL, 1);
 807                }
 808                up_write(&sb->s_umount);
 809                spin_lock(&sb_lock);
 810                if (p)
 811                        __put_super(p);
 812                p = sb;
 813        }
 814        if (p)
 815                __put_super(p);
 816        spin_unlock(&sb_lock);
 817        kfree(work);
 818        printk("Emergency Remount complete\n");
 819}
 820
 821void emergency_remount(void)
 822{
 823        struct work_struct *work;
 824
 825        work = kmalloc(sizeof(*work), GFP_ATOMIC);
 826        if (work) {
 827                INIT_WORK(work, do_emergency_remount);
 828                schedule_work(work);
 829        }
 830}
 831
 832/*
 833 * Unnamed block devices are dummy devices used by virtual
 834 * filesystems which don't use real block-devices.  -- jrs
 835 */
 836
 837static DEFINE_IDA(unnamed_dev_ida);
 838static DEFINE_SPINLOCK(unnamed_dev_lock);/* protects the above */
 839static int unnamed_dev_start = 0; /* don't bother trying below it */
 840
 841int get_anon_bdev(dev_t *p)
 842{
 843        int dev;
 844        int error;
 845
 846 retry:
 847        if (ida_pre_get(&unnamed_dev_ida, GFP_ATOMIC) == 0)
 848                return -ENOMEM;
 849        spin_lock(&unnamed_dev_lock);
 850        error = ida_get_new_above(&unnamed_dev_ida, unnamed_dev_start, &dev);
 851        if (!error)
 852                unnamed_dev_start = dev + 1;
 853        spin_unlock(&unnamed_dev_lock);
 854        if (error == -EAGAIN)
 855                /* We raced and lost with another CPU. */
 856                goto retry;
 857        else if (error)
 858                return -EAGAIN;
 859
 860        if ((dev & MAX_ID_MASK) == (1 << MINORBITS)) {
 861                spin_lock(&unnamed_dev_lock);
 862                ida_remove(&unnamed_dev_ida, dev);
 863                if (unnamed_dev_start > dev)
 864                        unnamed_dev_start = dev;
 865                spin_unlock(&unnamed_dev_lock);
 866                return -EMFILE;
 867        }
 868        *p = MKDEV(0, dev & MINORMASK);
 869        return 0;
 870}
 871EXPORT_SYMBOL(get_anon_bdev);
 872
 873void free_anon_bdev(dev_t dev)
 874{
 875        int slot = MINOR(dev);
 876        spin_lock(&unnamed_dev_lock);
 877        ida_remove(&unnamed_dev_ida, slot);
 878        if (slot < unnamed_dev_start)
 879                unnamed_dev_start = slot;
 880        spin_unlock(&unnamed_dev_lock);
 881}
 882EXPORT_SYMBOL(free_anon_bdev);
 883
 884int set_anon_super(struct super_block *s, void *data)
 885{
 886        int error = get_anon_bdev(&s->s_dev);
 887        if (!error)
 888                s->s_bdi = &noop_backing_dev_info;
 889        return error;
 890}
 891
 892EXPORT_SYMBOL(set_anon_super);
 893
 894void kill_anon_super(struct super_block *sb)
 895{
 896        dev_t dev = sb->s_dev;
 897        generic_shutdown_super(sb);
 898        free_anon_bdev(dev);
 899}
 900
 901EXPORT_SYMBOL(kill_anon_super);
 902
 903void kill_litter_super(struct super_block *sb)
 904{
 905        if (sb->s_root)
 906                d_genocide(sb->s_root);
 907        kill_anon_super(sb);
 908}
 909
 910EXPORT_SYMBOL(kill_litter_super);
 911
 912static int ns_test_super(struct super_block *sb, void *data)
 913{
 914        return sb->s_fs_info == data;
 915}
 916
 917static int ns_set_super(struct super_block *sb, void *data)
 918{
 919        sb->s_fs_info = data;
 920        return set_anon_super(sb, NULL);
 921}
 922
 923struct dentry *mount_ns(struct file_system_type *fs_type, int flags,
 924        void *data, int (*fill_super)(struct super_block *, void *, int))
 925{
 926        struct super_block *sb;
 927
 928        sb = sget(fs_type, ns_test_super, ns_set_super, data);
 929        if (IS_ERR(sb))
 930                return ERR_CAST(sb);
 931
 932        if (!sb->s_root) {
 933                int err;
 934                sb->s_flags = flags;
 935                err = fill_super(sb, data, flags & MS_SILENT ? 1 : 0);
 936                if (err) {
 937                        deactivate_locked_super(sb);
 938                        return ERR_PTR(err);
 939                }
 940
 941                sb->s_flags |= MS_ACTIVE;
 942        }
 943
 944        return dget(sb->s_root);
 945}
 946
 947EXPORT_SYMBOL(mount_ns);
 948
 949#ifdef CONFIG_BLOCK
 950static int set_bdev_super(struct super_block *s, void *data)
 951{
 952        s->s_bdev = data;
 953        s->s_dev = s->s_bdev->bd_dev;
 954
 955        /*
 956         * We set the bdi here to the queue backing, file systems can
 957         * overwrite this in ->fill_super()
 958         */
 959        s->s_bdi = &bdev_get_queue(s->s_bdev)->backing_dev_info;
 960        return 0;
 961}
 962
 963static int test_bdev_super(struct super_block *s, void *data)
 964{
 965        return (void *)s->s_bdev == data;
 966}
 967
 968struct dentry *mount_bdev(struct file_system_type *fs_type,
 969        int flags, const char *dev_name, void *data,
 970        int (*fill_super)(struct super_block *, void *, int))
 971{
 972        struct block_device *bdev;
 973        struct super_block *s;
 974        fmode_t mode = FMODE_READ | FMODE_EXCL;
 975        int error = 0;
 976
 977        if (!(flags & MS_RDONLY))
 978                mode |= FMODE_WRITE;
 979
 980        bdev = blkdev_get_by_path(dev_name, mode, fs_type);
 981        if (IS_ERR(bdev))
 982                return ERR_CAST(bdev);
 983
 984        /*
 985         * once the super is inserted into the list by sget, s_umount
 986         * will protect the lockfs code from trying to start a snapshot
 987         * while we are mounting
 988         */
 989        mutex_lock(&bdev->bd_fsfreeze_mutex);
 990        if (bdev->bd_fsfreeze_count > 0) {
 991                mutex_unlock(&bdev->bd_fsfreeze_mutex);
 992                error = -EBUSY;
 993                goto error_bdev;
 994        }
 995        s = sget(fs_type, test_bdev_super, set_bdev_super, bdev);
 996        mutex_unlock(&bdev->bd_fsfreeze_mutex);
 997        if (IS_ERR(s))
 998                goto error_s;
 999
1000        if (s->s_root) {
1001                if ((flags ^ s->s_flags) & MS_RDONLY) {
1002                        deactivate_locked_super(s);
1003                        error = -EBUSY;
1004                        goto error_bdev;
1005                }
1006
1007                /*
1008                 * s_umount nests inside bd_mutex during
1009                 * __invalidate_device().  blkdev_put() acquires
1010                 * bd_mutex and can't be called under s_umount.  Drop
1011                 * s_umount temporarily.  This is safe as we're
1012                 * holding an active reference.
1013                 */
1014                up_write(&s->s_umount);
1015                blkdev_put(bdev, mode);
1016                down_write(&s->s_umount);
1017        } else {
1018                char b[BDEVNAME_SIZE];
1019
1020                s->s_flags = flags | MS_NOSEC;
1021                s->s_mode = mode;
1022                strlcpy(s->s_id, bdevname(bdev, b), sizeof(s->s_id));
1023                sb_set_blocksize(s, block_size(bdev));
1024                error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1025                if (error) {
1026                        deactivate_locked_super(s);
1027                        goto error;
1028                }
1029
1030                s->s_flags |= MS_ACTIVE;
1031                bdev->bd_super = s;
1032        }
1033
1034        return dget(s->s_root);
1035
1036error_s:
1037        error = PTR_ERR(s);
1038error_bdev:
1039        blkdev_put(bdev, mode);
1040error:
1041        return ERR_PTR(error);
1042}
1043EXPORT_SYMBOL(mount_bdev);
1044
1045void kill_block_super(struct super_block *sb)
1046{
1047        struct block_device *bdev = sb->s_bdev;
1048        fmode_t mode = sb->s_mode;
1049
1050        bdev->bd_super = NULL;
1051        generic_shutdown_super(sb);
1052        sync_blockdev(bdev);
1053        WARN_ON_ONCE(!(mode & FMODE_EXCL));
1054        blkdev_put(bdev, mode | FMODE_EXCL);
1055}
1056
1057EXPORT_SYMBOL(kill_block_super);
1058#endif
1059
1060struct dentry *mount_nodev(struct file_system_type *fs_type,
1061        int flags, void *data,
1062        int (*fill_super)(struct super_block *, void *, int))
1063{
1064        int error;
1065        struct super_block *s = sget(fs_type, NULL, set_anon_super, NULL);
1066
1067        if (IS_ERR(s))
1068                return ERR_CAST(s);
1069
1070        s->s_flags = flags;
1071
1072        error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1073        if (error) {
1074                deactivate_locked_super(s);
1075                return ERR_PTR(error);
1076        }
1077        s->s_flags |= MS_ACTIVE;
1078        return dget(s->s_root);
1079}
1080EXPORT_SYMBOL(mount_nodev);
1081
1082static int compare_single(struct super_block *s, void *p)
1083{
1084        return 1;
1085}
1086
1087struct dentry *mount_single(struct file_system_type *fs_type,
1088        int flags, void *data,
1089        int (*fill_super)(struct super_block *, void *, int))
1090{
1091        struct super_block *s;
1092        int error;
1093
1094        s = sget(fs_type, compare_single, set_anon_super, NULL);
1095        if (IS_ERR(s))
1096                return ERR_CAST(s);
1097        if (!s->s_root) {
1098                s->s_flags = flags;
1099                error = fill_super(s, data, flags & MS_SILENT ? 1 : 0);
1100                if (error) {
1101                        deactivate_locked_super(s);
1102                        return ERR_PTR(error);
1103                }
1104                s->s_flags |= MS_ACTIVE;
1105        } else {
1106                do_remount_sb(s, flags, data, 0);
1107        }
1108        return dget(s->s_root);
1109}
1110EXPORT_SYMBOL(mount_single);
1111
1112struct dentry *
1113mount_fs(struct file_system_type *type, int flags, const char *name, void *data)
1114{
1115        struct dentry *root;
1116        struct super_block *sb;
1117        char *secdata = NULL;
1118        int error = -ENOMEM;
1119
1120        if (data && !(type->fs_flags & FS_BINARY_MOUNTDATA)) {
1121                secdata = alloc_secdata();
1122                if (!secdata)
1123                        goto out;
1124
1125                error = security_sb_copy_data(data, secdata);
1126                if (error)
1127                        goto out_free_secdata;
1128        }
1129
1130        root = type->mount(type, flags, name, data);
1131        if (IS_ERR(root)) {
1132                error = PTR_ERR(root);
1133                goto out_free_secdata;
1134        }
1135        sb = root->d_sb;
1136        BUG_ON(!sb);
1137        WARN_ON(!sb->s_bdi);
1138        WARN_ON(sb->s_bdi == &default_backing_dev_info);
1139        sb->s_flags |= MS_BORN;
1140
1141        error = security_sb_kern_mount(sb, flags, secdata);
1142        if (error)
1143                goto out_sb;
1144
1145        /*
1146         * filesystems should never set s_maxbytes larger than MAX_LFS_FILESIZE
1147         * but s_maxbytes was an unsigned long long for many releases. Throw
1148         * this warning for a little while to try and catch filesystems that
1149         * violate this rule.
1150         */
1151        WARN((sb->s_maxbytes < 0), "%s set sb->s_maxbytes to "
1152                "negative value (%lld)\n", type->name, sb->s_maxbytes);
1153
1154        up_write(&sb->s_umount);
1155        free_secdata(secdata);
1156        return root;
1157out_sb:
1158        dput(root);
1159        deactivate_locked_super(sb);
1160out_free_secdata:
1161        free_secdata(secdata);
1162out:
1163        return ERR_PTR(error);
1164}
1165
1166/**
1167 * freeze_super - lock the filesystem and force it into a consistent state
1168 * @sb: the super to lock
1169 *
1170 * Syncs the super to make sure the filesystem is consistent and calls the fs's
1171 * freeze_fs.  Subsequent calls to this without first thawing the fs will return
1172 * -EBUSY.
1173 */
1174int freeze_super(struct super_block *sb)
1175{
1176        int ret;
1177
1178        atomic_inc(&sb->s_active);
1179        down_write(&sb->s_umount);
1180        if (sb->s_frozen) {
1181                deactivate_locked_super(sb);
1182                return -EBUSY;
1183        }
1184
1185        if (!(sb->s_flags & MS_BORN)) {
1186                up_write(&sb->s_umount);
1187                return 0;       /* sic - it's "nothing to do" */
1188        }
1189
1190        if (sb->s_flags & MS_RDONLY) {
1191                sb->s_frozen = SB_FREEZE_TRANS;
1192                smp_wmb();
1193                up_write(&sb->s_umount);
1194                return 0;
1195        }
1196
1197        sb->s_frozen = SB_FREEZE_WRITE;
1198        smp_wmb();
1199
1200        sync_filesystem(sb);
1201
1202        sb->s_frozen = SB_FREEZE_TRANS;
1203        smp_wmb();
1204
1205        sync_blockdev(sb->s_bdev);
1206        if (sb->s_op->freeze_fs) {
1207                ret = sb->s_op->freeze_fs(sb);
1208                if (ret) {
1209                        printk(KERN_ERR
1210                                "VFS:Filesystem freeze failed\n");
1211                        sb->s_frozen = SB_UNFROZEN;
1212                        smp_wmb();
1213                        wake_up(&sb->s_wait_unfrozen);
1214                        deactivate_locked_super(sb);
1215                        return ret;
1216                }
1217        }
1218        up_write(&sb->s_umount);
1219        return 0;
1220}
1221EXPORT_SYMBOL(freeze_super);
1222
1223/**
1224 * thaw_super -- unlock filesystem
1225 * @sb: the super to thaw
1226 *
1227 * Unlocks the filesystem and marks it writeable again after freeze_super().
1228 */
1229int thaw_super(struct super_block *sb)
1230{
1231        int error;
1232
1233        down_write(&sb->s_umount);
1234        if (sb->s_frozen == SB_UNFROZEN) {
1235                up_write(&sb->s_umount);
1236                return -EINVAL;
1237        }
1238
1239        if (sb->s_flags & MS_RDONLY)
1240                goto out;
1241
1242        if (sb->s_op->unfreeze_fs) {
1243                error = sb->s_op->unfreeze_fs(sb);
1244                if (error) {
1245                        printk(KERN_ERR
1246                                "VFS:Filesystem thaw failed\n");
1247                        sb->s_frozen = SB_FREEZE_TRANS;
1248                        up_write(&sb->s_umount);
1249                        return error;
1250                }
1251        }
1252
1253out:
1254        sb->s_frozen = SB_UNFROZEN;
1255        smp_wmb();
1256        wake_up(&sb->s_wait_unfrozen);
1257        deactivate_locked_super(sb);
1258
1259        return 0;
1260}
1261EXPORT_SYMBOL(thaw_super);
1262
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.