linux/fs/namespace.c
<<
>>
Prefs
   1/*
   2 *  linux/fs/namespace.c
   3 *
   4 * (C) Copyright Al Viro 2000, 2001
   5 *      Released under GPL v2.
   6 *
   7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
   8 * Heavily rewritten.
   9 */
  10
  11#include <linux/syscalls.h>
  12#include <linux/export.h>
  13#include <linux/capability.h>
  14#include <linux/mnt_namespace.h>
  15#include <linux/user_namespace.h>
  16#include <linux/namei.h>
  17#include <linux/security.h>
  18#include <linux/idr.h>
  19#include <linux/acct.h>         /* acct_auto_close_mnt */
  20#include <linux/init.h>         /* init_rootfs */
  21#include <linux/fs_struct.h>    /* get_fs_root et.al. */
  22#include <linux/fsnotify.h>     /* fsnotify_vfsmount_delete */
  23#include <linux/uaccess.h>
  24#include <linux/proc_ns.h>
  25#include <linux/magic.h>
  26#include "pnode.h"
  27#include "internal.h"
  28
  29#define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
  30#define HASH_SIZE (1UL << HASH_SHIFT)
  31
  32static int event;
  33static DEFINE_IDA(mnt_id_ida);
  34static DEFINE_IDA(mnt_group_ida);
  35static DEFINE_SPINLOCK(mnt_id_lock);
  36static int mnt_id_start = 0;
  37static int mnt_group_start = 1;
  38
  39static struct list_head *mount_hashtable __read_mostly;
  40static struct list_head *mountpoint_hashtable __read_mostly;
  41static struct kmem_cache *mnt_cache __read_mostly;
  42static DECLARE_RWSEM(namespace_sem);
  43
  44/* /sys/fs */
  45struct kobject *fs_kobj;
  46EXPORT_SYMBOL_GPL(fs_kobj);
  47
  48/*
  49 * vfsmount lock may be taken for read to prevent changes to the
  50 * vfsmount hash, ie. during mountpoint lookups or walking back
  51 * up the tree.
  52 *
  53 * It should be taken for write in all cases where the vfsmount
  54 * tree or hash is modified or when a vfsmount structure is modified.
  55 */
  56__cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
  57
  58static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
  59{
  60        unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
  61        tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
  62        tmp = tmp + (tmp >> HASH_SHIFT);
  63        return tmp & (HASH_SIZE - 1);
  64}
  65
  66/*
  67 * allocation is serialized by namespace_sem, but we need the spinlock to
  68 * serialize with freeing.
  69 */
  70static int mnt_alloc_id(struct mount *mnt)
  71{
  72        int res;
  73
  74retry:
  75        ida_pre_get(&mnt_id_ida, GFP_KERNEL);
  76        spin_lock(&mnt_id_lock);
  77        res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
  78        if (!res)
  79                mnt_id_start = mnt->mnt_id + 1;
  80        spin_unlock(&mnt_id_lock);
  81        if (res == -EAGAIN)
  82                goto retry;
  83
  84        return res;
  85}
  86
  87static void mnt_free_id(struct mount *mnt)
  88{
  89        int id = mnt->mnt_id;
  90        spin_lock(&mnt_id_lock);
  91        ida_remove(&mnt_id_ida, id);
  92        if (mnt_id_start > id)
  93                mnt_id_start = id;
  94        spin_unlock(&mnt_id_lock);
  95}
  96
  97/*
  98 * Allocate a new peer group ID
  99 *
 100 * mnt_group_ida is protected by namespace_sem
 101 */
 102static int mnt_alloc_group_id(struct mount *mnt)
 103{
 104        int res;
 105
 106        if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
 107                return -ENOMEM;
 108
 109        res = ida_get_new_above(&mnt_group_ida,
 110                                mnt_group_start,
 111                                &mnt->mnt_group_id);
 112        if (!res)
 113                mnt_group_start = mnt->mnt_group_id + 1;
 114
 115        return res;
 116}
 117
 118/*
 119 * Release a peer group ID
 120 */
 121void mnt_release_group_id(struct mount *mnt)
 122{
 123        int id = mnt->mnt_group_id;
 124        ida_remove(&mnt_group_ida, id);
 125        if (mnt_group_start > id)
 126                mnt_group_start = id;
 127        mnt->mnt_group_id = 0;
 128}
 129
 130/*
 131 * vfsmount lock must be held for read
 132 */
 133static inline void mnt_add_count(struct mount *mnt, int n)
 134{
 135#ifdef CONFIG_SMP
 136        this_cpu_add(mnt->mnt_pcp->mnt_count, n);
 137#else
 138        preempt_disable();
 139        mnt->mnt_count += n;
 140        preempt_enable();
 141#endif
 142}
 143
 144/*
 145 * vfsmount lock must be held for write
 146 */
 147unsigned int mnt_get_count(struct mount *mnt)
 148{
 149#ifdef CONFIG_SMP
 150        unsigned int count = 0;
 151        int cpu;
 152
 153        for_each_possible_cpu(cpu) {
 154                count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
 155        }
 156
 157        return count;
 158#else
 159        return mnt->mnt_count;
 160#endif
 161}
 162
 163static struct mount *alloc_vfsmnt(const char *name)
 164{
 165        struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
 166        if (mnt) {
 167                int err;
 168
 169                err = mnt_alloc_id(mnt);
 170                if (err)
 171                        goto out_free_cache;
 172
 173                if (name) {
 174                        mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
 175                        if (!mnt->mnt_devname)
 176                                goto out_free_id;
 177                }
 178
 179#ifdef CONFIG_SMP
 180                mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
 181                if (!mnt->mnt_pcp)
 182                        goto out_free_devname;
 183
 184                this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
 185#else
 186                mnt->mnt_count = 1;
 187                mnt->mnt_writers = 0;
 188#endif
 189
 190                INIT_LIST_HEAD(&mnt->mnt_hash);
 191                INIT_LIST_HEAD(&mnt->mnt_child);
 192                INIT_LIST_HEAD(&mnt->mnt_mounts);
 193                INIT_LIST_HEAD(&mnt->mnt_list);
 194                INIT_LIST_HEAD(&mnt->mnt_expire);
 195                INIT_LIST_HEAD(&mnt->mnt_share);
 196                INIT_LIST_HEAD(&mnt->mnt_slave_list);
 197                INIT_LIST_HEAD(&mnt->mnt_slave);
 198#ifdef CONFIG_FSNOTIFY
 199                INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
 200#endif
 201        }
 202        return mnt;
 203
 204#ifdef CONFIG_SMP
 205out_free_devname:
 206        kfree(mnt->mnt_devname);
 207#endif
 208out_free_id:
 209        mnt_free_id(mnt);
 210out_free_cache:
 211        kmem_cache_free(mnt_cache, mnt);
 212        return NULL;
 213}
 214
 215/*
 216 * Most r/o checks on a fs are for operations that take
 217 * discrete amounts of time, like a write() or unlink().
 218 * We must keep track of when those operations start
 219 * (for permission checks) and when they end, so that
 220 * we can determine when writes are able to occur to
 221 * a filesystem.
 222 */
 223/*
 224 * __mnt_is_readonly: check whether a mount is read-only
 225 * @mnt: the mount to check for its write status
 226 *
 227 * This shouldn't be used directly ouside of the VFS.
 228 * It does not guarantee that the filesystem will stay
 229 * r/w, just that it is right *now*.  This can not and
 230 * should not be used in place of IS_RDONLY(inode).
 231 * mnt_want/drop_write() will _keep_ the filesystem
 232 * r/w.
 233 */
 234int __mnt_is_readonly(struct vfsmount *mnt)
 235{
 236        if (mnt->mnt_flags & MNT_READONLY)
 237                return 1;
 238        if (mnt->mnt_sb->s_flags & MS_RDONLY)
 239                return 1;
 240        return 0;
 241}
 242EXPORT_SYMBOL_GPL(__mnt_is_readonly);
 243
 244static inline void mnt_inc_writers(struct mount *mnt)
 245{
 246#ifdef CONFIG_SMP
 247        this_cpu_inc(mnt->mnt_pcp->mnt_writers);
 248#else
 249        mnt->mnt_writers++;
 250#endif
 251}
 252
 253static inline void mnt_dec_writers(struct mount *mnt)
 254{
 255#ifdef CONFIG_SMP
 256        this_cpu_dec(mnt->mnt_pcp->mnt_writers);
 257#else
 258        mnt->mnt_writers--;
 259#endif
 260}
 261
 262static unsigned int mnt_get_writers(struct mount *mnt)
 263{
 264#ifdef CONFIG_SMP
 265        unsigned int count = 0;
 266        int cpu;
 267
 268        for_each_possible_cpu(cpu) {
 269                count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
 270        }
 271
 272        return count;
 273#else
 274        return mnt->mnt_writers;
 275#endif
 276}
 277
 278static int mnt_is_readonly(struct vfsmount *mnt)
 279{
 280        if (mnt->mnt_sb->s_readonly_remount)
 281                return 1;
 282        /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
 283        smp_rmb();
 284        return __mnt_is_readonly(mnt);
 285}
 286
 287/*
 288 * Most r/o & frozen checks on a fs are for operations that take discrete
 289 * amounts of time, like a write() or unlink().  We must keep track of when
 290 * those operations start (for permission checks) and when they end, so that we
 291 * can determine when writes are able to occur to a filesystem.
 292 */
 293/**
 294 * __mnt_want_write - get write access to a mount without freeze protection
 295 * @m: the mount on which to take a write
 296 *
 297 * This tells the low-level filesystem that a write is about to be performed to
 298 * it, and makes sure that writes are allowed (mnt it read-write) before
 299 * returning success. This operation does not protect against filesystem being
 300 * frozen. When the write operation is finished, __mnt_drop_write() must be
 301 * called. This is effectively a refcount.
 302 */
 303int __mnt_want_write(struct vfsmount *m)
 304{
 305        struct mount *mnt = real_mount(m);
 306        int ret = 0;
 307
 308        preempt_disable();
 309        mnt_inc_writers(mnt);
 310        /*
 311         * The store to mnt_inc_writers must be visible before we pass
 312         * MNT_WRITE_HOLD loop below, so that the slowpath can see our
 313         * incremented count after it has set MNT_WRITE_HOLD.
 314         */
 315        smp_mb();
 316        while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
 317                cpu_relax();
 318        /*
 319         * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
 320         * be set to match its requirements. So we must not load that until
 321         * MNT_WRITE_HOLD is cleared.
 322         */
 323        smp_rmb();
 324        if (mnt_is_readonly(m)) {
 325                mnt_dec_writers(mnt);
 326                ret = -EROFS;
 327        }
 328        preempt_enable();
 329
 330        return ret;
 331}
 332
 333/**
 334 * mnt_want_write - get write access to a mount
 335 * @m: the mount on which to take a write
 336 *
 337 * This tells the low-level filesystem that a write is about to be performed to
 338 * it, and makes sure that writes are allowed (mount is read-write, filesystem
 339 * is not frozen) before returning success.  When the write operation is
 340 * finished, mnt_drop_write() must be called.  This is effectively a refcount.
 341 */
 342int mnt_want_write(struct vfsmount *m)
 343{
 344        int ret;
 345
 346        sb_start_write(m->mnt_sb);
 347        ret = __mnt_want_write(m);
 348        if (ret)
 349                sb_end_write(m->mnt_sb);
 350        return ret;
 351}
 352EXPORT_SYMBOL_GPL(mnt_want_write);
 353
 354/**
 355 * mnt_clone_write - get write access to a mount
 356 * @mnt: the mount on which to take a write
 357 *
 358 * This is effectively like mnt_want_write, except
 359 * it must only be used to take an extra write reference
 360 * on a mountpoint that we already know has a write reference
 361 * on it. This allows some optimisation.
 362 *
 363 * After finished, mnt_drop_write must be called as usual to
 364 * drop the reference.
 365 */
 366int mnt_clone_write(struct vfsmount *mnt)
 367{
 368        /* superblock may be r/o */
 369        if (__mnt_is_readonly(mnt))
 370                return -EROFS;
 371        preempt_disable();
 372        mnt_inc_writers(real_mount(mnt));
 373        preempt_enable();
 374        return 0;
 375}
 376EXPORT_SYMBOL_GPL(mnt_clone_write);
 377
 378/**
 379 * __mnt_want_write_file - get write access to a file's mount
 380 * @file: the file who's mount on which to take a write
 381 *
 382 * This is like __mnt_want_write, but it takes a file and can
 383 * do some optimisations if the file is open for write already
 384 */
 385int __mnt_want_write_file(struct file *file)
 386{
 387        struct inode *inode = file_inode(file);
 388
 389        if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
 390                return __mnt_want_write(file->f_path.mnt);
 391        else
 392                return mnt_clone_write(file->f_path.mnt);
 393}
 394
 395/**
 396 * mnt_want_write_file - get write access to a file's mount
 397 * @file: the file who's mount on which to take a write
 398 *
 399 * This is like mnt_want_write, but it takes a file and can
 400 * do some optimisations if the file is open for write already
 401 */
 402int mnt_want_write_file(struct file *file)
 403{
 404        int ret;
 405
 406        sb_start_write(file->f_path.mnt->mnt_sb);
 407        ret = __mnt_want_write_file(file);
 408        if (ret)
 409                sb_end_write(file->f_path.mnt->mnt_sb);
 410        return ret;
 411}
 412EXPORT_SYMBOL_GPL(mnt_want_write_file);
 413
 414/**
 415 * __mnt_drop_write - give up write access to a mount
 416 * @mnt: the mount on which to give up write access
 417 *
 418 * Tells the low-level filesystem that we are done
 419 * performing writes to it.  Must be matched with
 420 * __mnt_want_write() call above.
 421 */
 422void __mnt_drop_write(struct vfsmount *mnt)
 423{
 424        preempt_disable();
 425        mnt_dec_writers(real_mount(mnt));
 426        preempt_enable();
 427}
 428
 429/**
 430 * mnt_drop_write - give up write access to a mount
 431 * @mnt: the mount on which to give up write access
 432 *
 433 * Tells the low-level filesystem that we are done performing writes to it and
 434 * also allows filesystem to be frozen again.  Must be matched with
 435 * mnt_want_write() call above.
 436 */
 437void mnt_drop_write(struct vfsmount *mnt)
 438{
 439        __mnt_drop_write(mnt);
 440        sb_end_write(mnt->mnt_sb);
 441}
 442EXPORT_SYMBOL_GPL(mnt_drop_write);
 443
 444void __mnt_drop_write_file(struct file *file)
 445{
 446        __mnt_drop_write(file->f_path.mnt);
 447}
 448
 449void mnt_drop_write_file(struct file *file)
 450{
 451        mnt_drop_write(file->f_path.mnt);
 452}
 453EXPORT_SYMBOL(mnt_drop_write_file);
 454
 455static int mnt_make_readonly(struct mount *mnt)
 456{
 457        int ret = 0;
 458
 459        lock_mount_hash();
 460        mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
 461        /*
 462         * After storing MNT_WRITE_HOLD, we'll read the counters. This store
 463         * should be visible before we do.
 464         */
 465        smp_mb();
 466
 467        /*
 468         * With writers on hold, if this value is zero, then there are
 469         * definitely no active writers (although held writers may subsequently
 470         * increment the count, they'll have to wait, and decrement it after
 471         * seeing MNT_READONLY).
 472         *
 473         * It is OK to have counter incremented on one CPU and decremented on
 474         * another: the sum will add up correctly. The danger would be when we
 475         * sum up each counter, if we read a counter before it is incremented,
 476         * but then read another CPU's count which it has been subsequently
 477         * decremented from -- we would see more decrements than we should.
 478         * MNT_WRITE_HOLD protects against this scenario, because
 479         * mnt_want_write first increments count, then smp_mb, then spins on
 480         * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
 481         * we're counting up here.
 482         */
 483        if (mnt_get_writers(mnt) > 0)
 484                ret = -EBUSY;
 485        else
 486                mnt->mnt.mnt_flags |= MNT_READONLY;
 487        /*
 488         * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
 489         * that become unheld will see MNT_READONLY.
 490         */
 491        smp_wmb();
 492        mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
 493        unlock_mount_hash();
 494        return ret;
 495}
 496
 497static void __mnt_unmake_readonly(struct mount *mnt)
 498{
 499        lock_mount_hash();
 500        mnt->mnt.mnt_flags &= ~MNT_READONLY;
 501        unlock_mount_hash();
 502}
 503
 504int sb_prepare_remount_readonly(struct super_block *sb)
 505{
 506        struct mount *mnt;
 507        int err = 0;
 508
 509        /* Racy optimization.  Recheck the counter under MNT_WRITE_HOLD */
 510        if (atomic_long_read(&sb->s_remove_count))
 511                return -EBUSY;
 512
 513        lock_mount_hash();
 514        list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
 515                if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
 516                        mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
 517                        smp_mb();
 518                        if (mnt_get_writers(mnt) > 0) {
 519                                err = -EBUSY;
 520                                break;
 521                        }
 522                }
 523        }
 524        if (!err && atomic_long_read(&sb->s_remove_count))
 525                err = -EBUSY;
 526
 527        if (!err) {
 528                sb->s_readonly_remount = 1;
 529                smp_wmb();
 530        }
 531        list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
 532                if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
 533                        mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
 534        }
 535        unlock_mount_hash();
 536
 537        return err;
 538}
 539
 540static void free_vfsmnt(struct mount *mnt)
 541{
 542        kfree(mnt->mnt_devname);
 543        mnt_free_id(mnt);
 544#ifdef CONFIG_SMP
 545        free_percpu(mnt->mnt_pcp);
 546#endif
 547        kmem_cache_free(mnt_cache, mnt);
 548}
 549
 550/* call under rcu_read_lock */
 551bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
 552{
 553        struct mount *mnt;
 554        if (read_seqretry(&mount_lock, seq))
 555                return false;
 556        if (bastard == NULL)
 557                return true;
 558        mnt = real_mount(bastard);
 559        mnt_add_count(mnt, 1);
 560        if (likely(!read_seqretry(&mount_lock, seq)))
 561                return true;
 562        if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
 563                mnt_add_count(mnt, -1);
 564                return false;
 565        }
 566        rcu_read_unlock();
 567        mntput(bastard);
 568        rcu_read_lock();
 569        return false;
 570}
 571
 572/*
 573 * find the first mount at @dentry on vfsmount @mnt.
 574 * call under rcu_read_lock()
 575 */
 576struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
 577{
 578        struct list_head *head = mount_hashtable + hash(mnt, dentry);
 579        struct mount *p;
 580
 581        list_for_each_entry_rcu(p, head, mnt_hash)
 582                if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
 583                        return p;
 584        return NULL;
 585}
 586
 587/*
 588 * find the last mount at @dentry on vfsmount @mnt.
 589 * mount_lock must be held.
 590 */
 591struct mount *__lookup_mnt_last(struct vfsmount *mnt, struct dentry *dentry)
 592{
 593        struct list_head *head = mount_hashtable + hash(mnt, dentry);
 594        struct mount *p;
 595
 596        list_for_each_entry_reverse(p, head, mnt_hash)
 597                if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
 598                        return p;
 599        return NULL;
 600}
 601
 602/*
 603 * lookup_mnt - Return the first child mount mounted at path
 604 *
 605 * "First" means first mounted chronologically.  If you create the
 606 * following mounts:
 607 *
 608 * mount /dev/sda1 /mnt
 609 * mount /dev/sda2 /mnt
 610 * mount /dev/sda3 /mnt
 611 *
 612 * Then lookup_mnt() on the base /mnt dentry in the root mount will
 613 * return successively the root dentry and vfsmount of /dev/sda1, then
 614 * /dev/sda2, then /dev/sda3, then NULL.
 615 *
 616 * lookup_mnt takes a reference to the found vfsmount.
 617 */
 618struct vfsmount *lookup_mnt(struct path *path)
 619{
 620        struct mount *child_mnt;
 621        struct vfsmount *m;
 622        unsigned seq;
 623
 624        rcu_read_lock();
 625        do {
 626                seq = read_seqbegin(&mount_lock);
 627                child_mnt = __lookup_mnt(path->mnt, path->dentry);
 628                m = child_mnt ? &child_mnt->mnt : NULL;
 629        } while (!legitimize_mnt(m, seq));
 630        rcu_read_unlock();
 631        return m;
 632}
 633
 634static struct mountpoint *new_mountpoint(struct dentry *dentry)
 635{
 636        struct list_head *chain = mountpoint_hashtable + hash(NULL, dentry);
 637        struct mountpoint *mp;
 638        int ret;
 639
 640        list_for_each_entry(mp, chain, m_hash) {
 641                if (mp->m_dentry == dentry) {
 642                        /* might be worth a WARN_ON() */
 643                        if (d_unlinked(dentry))
 644                                return ERR_PTR(-ENOENT);
 645                        mp->m_count++;
 646                        return mp;
 647                }
 648        }
 649
 650        mp = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
 651        if (!mp)
 652                return ERR_PTR(-ENOMEM);
 653
 654        ret = d_set_mounted(dentry);
 655        if (ret) {
 656                kfree(mp);
 657                return ERR_PTR(ret);
 658        }
 659
 660        mp->m_dentry = dentry;
 661        mp->m_count = 1;
 662        list_add(&mp->m_hash, chain);
 663        return mp;
 664}
 665
 666static void put_mountpoint(struct mountpoint *mp)
 667{
 668        if (!--mp->m_count) {
 669                struct dentry *dentry = mp->m_dentry;
 670                spin_lock(&dentry->d_lock);
 671                dentry->d_flags &= ~DCACHE_MOUNTED;
 672                spin_unlock(&dentry->d_lock);
 673                list_del(&mp->m_hash);
 674                kfree(mp);
 675        }
 676}
 677
 678static inline int check_mnt(struct mount *mnt)
 679{
 680        return mnt->mnt_ns == current->nsproxy->mnt_ns;
 681}
 682
 683/*
 684 * vfsmount lock must be held for write
 685 */
 686static void touch_mnt_namespace(struct mnt_namespace *ns)
 687{
 688        if (ns) {
 689                ns->event = ++event;
 690                wake_up_interruptible(&ns->poll);
 691        }
 692}
 693
 694/*
 695 * vfsmount lock must be held for write
 696 */
 697static void __touch_mnt_namespace(struct mnt_namespace *ns)
 698{
 699        if (ns && ns->event != event) {
 700                ns->event = event;
 701                wake_up_interruptible(&ns->poll);
 702        }
 703}
 704
 705/*
 706 * vfsmount lock must be held for write
 707 */
 708static void detach_mnt(struct mount *mnt, struct path *old_path)
 709{
 710        old_path->dentry = mnt->mnt_mountpoint;
 711        old_path->mnt = &mnt->mnt_parent->mnt;
 712        mnt->mnt_parent = mnt;
 713        mnt->mnt_mountpoint = mnt->mnt.mnt_root;
 714        list_del_init(&mnt->mnt_child);
 715        list_del_init(&mnt->mnt_hash);
 716        put_mountpoint(mnt->mnt_mp);
 717        mnt->mnt_mp = NULL;
 718}
 719
 720/*
 721 * vfsmount lock must be held for write
 722 */
 723void mnt_set_mountpoint(struct mount *mnt,
 724                        struct mountpoint *mp,
 725                        struct mount *child_mnt)
 726{
 727        mp->m_count++;
 728        mnt_add_count(mnt, 1);  /* essentially, that's mntget */
 729        child_mnt->mnt_mountpoint = dget(mp->m_dentry);
 730        child_mnt->mnt_parent = mnt;
 731        child_mnt->mnt_mp = mp;
 732}
 733
 734/*
 735 * vfsmount lock must be held for write
 736 */
 737static void attach_mnt(struct mount *mnt,
 738                        struct mount *parent,
 739                        struct mountpoint *mp)
 740{
 741        mnt_set_mountpoint(parent, mp, mnt);
 742        list_add_tail(&mnt->mnt_hash, mount_hashtable +
 743                        hash(&parent->mnt, mp->m_dentry));
 744        list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
 745}
 746
 747/*
 748 * vfsmount lock must be held for write
 749 */
 750static void commit_tree(struct mount *mnt)
 751{
 752        struct mount *parent = mnt->mnt_parent;
 753        struct mount *m;
 754        LIST_HEAD(head);
 755        struct mnt_namespace *n = parent->mnt_ns;
 756
 757        BUG_ON(parent == mnt);
 758
 759        list_add_tail(&head, &mnt->mnt_list);
 760        list_for_each_entry(m, &head, mnt_list)
 761                m->mnt_ns = n;
 762
 763        list_splice(&head, n->list.prev);
 764
 765        list_add_tail(&mnt->mnt_hash, mount_hashtable +
 766                                hash(&parent->mnt, mnt->mnt_mountpoint));
 767        list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
 768        touch_mnt_namespace(n);
 769}
 770
 771static struct mount *next_mnt(struct mount *p, struct mount *root)
 772{
 773        struct list_head *next = p->mnt_mounts.next;
 774        if (next == &p->mnt_mounts) {
 775                while (1) {
 776                        if (p == root)
 777                                return NULL;
 778                        next = p->mnt_child.next;
 779                        if (next != &p->mnt_parent->mnt_mounts)
 780                                break;
 781                        p = p->mnt_parent;
 782                }
 783        }
 784        return list_entry(next, struct mount, mnt_child);
 785}
 786
 787static struct mount *skip_mnt_tree(struct mount *p)
 788{
 789        struct list_head *prev = p->mnt_mounts.prev;
 790        while (prev != &p->mnt_mounts) {
 791                p = list_entry(prev, struct mount, mnt_child);
 792                prev = p->mnt_mounts.prev;
 793        }
 794        return p;
 795}
 796
 797struct vfsmount *
 798vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
 799{
 800        struct mount *mnt;
 801        struct dentry *root;
 802
 803        if (!type)
 804                return ERR_PTR(-ENODEV);
 805
 806        mnt = alloc_vfsmnt(name);
 807        if (!mnt)
 808                return ERR_PTR(-ENOMEM);
 809
 810        if (flags & MS_KERNMOUNT)
 811                mnt->mnt.mnt_flags = MNT_INTERNAL;
 812
 813        root = mount_fs(type, flags, name, data);
 814        if (IS_ERR(root)) {
 815                free_vfsmnt(mnt);
 816                return ERR_CAST(root);
 817        }
 818
 819        mnt->mnt.mnt_root = root;
 820        mnt->mnt.mnt_sb = root->d_sb;
 821        mnt->mnt_mountpoint = mnt->mnt.mnt_root;
 822        mnt->mnt_parent = mnt;
 823        lock_mount_hash();
 824        list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
 825        unlock_mount_hash();
 826        return &mnt->mnt;
 827}
 828EXPORT_SYMBOL_GPL(vfs_kern_mount);
 829
 830static struct mount *clone_mnt(struct mount *old, struct dentry *root,
 831                                        int flag)
 832{
 833        struct super_block *sb = old->mnt.mnt_sb;
 834        struct mount *mnt;
 835        int err;
 836
 837        mnt = alloc_vfsmnt(old->mnt_devname);
 838        if (!mnt)
 839                return ERR_PTR(-ENOMEM);
 840
 841        if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
 842                mnt->mnt_group_id = 0; /* not a peer of original */
 843        else
 844                mnt->mnt_group_id = old->mnt_group_id;
 845
 846        if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
 847                err = mnt_alloc_group_id(mnt);
 848                if (err)
 849                        goto out_free;
 850        }
 851
 852        mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
 853        /* Don't allow unprivileged users to change mount flags */
 854        if ((flag & CL_UNPRIVILEGED) && (mnt->mnt.mnt_flags & MNT_READONLY))
 855                mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
 856
 857        /* Don't allow unprivileged users to reveal what is under a mount */
 858        if ((flag & CL_UNPRIVILEGED) && list_empty(&old->mnt_expire))
 859                mnt->mnt.mnt_flags |= MNT_LOCKED;
 860
 861        atomic_inc(&sb->s_active);
 862        mnt->mnt.mnt_sb = sb;
 863        mnt->mnt.mnt_root = dget(root);
 864        mnt->mnt_mountpoint = mnt->mnt.mnt_root;
 865        mnt->mnt_parent = mnt;
 866        lock_mount_hash();
 867        list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
 868        unlock_mount_hash();
 869
 870        if ((flag & CL_SLAVE) ||
 871            ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
 872                list_add(&mnt->mnt_slave, &old->mnt_slave_list);
 873                mnt->mnt_master = old;
 874                CLEAR_MNT_SHARED(mnt);
 875        } else if (!(flag & CL_PRIVATE)) {
 876                if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
 877                        list_add(&mnt->mnt_share, &old->mnt_share);
 878                if (IS_MNT_SLAVE(old))
 879                        list_add(&mnt->mnt_slave, &old->mnt_slave);
 880                mnt->mnt_master = old->mnt_master;
 881        }
 882        if (flag & CL_MAKE_SHARED)
 883                set_mnt_shared(mnt);
 884
 885        /* stick the duplicate mount on the same expiry list
 886         * as the original if that was on one */
 887        if (flag & CL_EXPIRE) {
 888                if (!list_empty(&old->mnt_expire))
 889                        list_add(&mnt->mnt_expire, &old->mnt_expire);
 890        }
 891
 892        return mnt;
 893
 894 out_free:
 895        free_vfsmnt(mnt);
 896        return ERR_PTR(err);
 897}
 898
 899static void delayed_free(struct rcu_head *head)
 900{
 901        struct mount *mnt = container_of(head, struct mount, mnt_rcu);
 902        kfree(mnt->mnt_devname);
 903#ifdef CONFIG_SMP
 904        free_percpu(mnt->mnt_pcp);
 905#endif
 906        kmem_cache_free(mnt_cache, mnt);
 907}
 908
 909static void mntput_no_expire(struct mount *mnt)
 910{
 911put_again:
 912        rcu_read_lock();
 913        mnt_add_count(mnt, -1);
 914        if (likely(mnt->mnt_ns)) { /* shouldn't be the last one */
 915                rcu_read_unlock();
 916                return;
 917        }
 918        lock_mount_hash();
 919        if (mnt_get_count(mnt)) {
 920                rcu_read_unlock();
 921                unlock_mount_hash();
 922                return;
 923        }
 924        if (unlikely(mnt->mnt_pinned)) {
 925                mnt_add_count(mnt, mnt->mnt_pinned + 1);
 926                mnt->mnt_pinned = 0;
 927                rcu_read_unlock();
 928                unlock_mount_hash();
 929                acct_auto_close_mnt(&mnt->mnt);
 930                goto put_again;
 931        }
 932        if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
 933                rcu_read_unlock();
 934                unlock_mount_hash();
 935                return;
 936        }
 937        mnt->mnt.mnt_flags |= MNT_DOOMED;
 938        rcu_read_unlock();
 939
 940        list_del(&mnt->mnt_instance);
 941        unlock_mount_hash();
 942
 943        /*
 944         * This probably indicates that somebody messed
 945         * up a mnt_want/drop_write() pair.  If this
 946         * happens, the filesystem was probably unable
 947         * to make r/w->r/o transitions.
 948         */
 949        /*
 950         * The locking used to deal with mnt_count decrement provides barriers,
 951         * so mnt_get_writers() below is safe.
 952         */
 953        WARN_ON(mnt_get_writers(mnt));
 954        fsnotify_vfsmount_delete(&mnt->mnt);
 955        dput(mnt->mnt.mnt_root);
 956        deactivate_super(mnt->mnt.mnt_sb);
 957        mnt_free_id(mnt);
 958        call_rcu(&mnt->mnt_rcu, delayed_free);
 959}
 960
 961void mntput(struct vfsmount *mnt)
 962{
 963        if (mnt) {
 964                struct mount *m = real_mount(mnt);
 965                /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
 966                if (unlikely(m->mnt_expiry_mark))
 967                        m->mnt_expiry_mark = 0;
 968                mntput_no_expire(m);
 969        }
 970}
 971EXPORT_SYMBOL(mntput);
 972
 973struct vfsmount *mntget(struct vfsmount *mnt)
 974{
 975        if (mnt)
 976                mnt_add_count(real_mount(mnt), 1);
 977        return mnt;
 978}
 979EXPORT_SYMBOL(mntget);
 980
 981void mnt_pin(struct vfsmount *mnt)
 982{
 983        lock_mount_hash();
 984        real_mount(mnt)->mnt_pinned++;
 985        unlock_mount_hash();
 986}
 987EXPORT_SYMBOL(mnt_pin);
 988
 989void mnt_unpin(struct vfsmount *m)
 990{
 991        struct mount *mnt = real_mount(m);
 992        lock_mount_hash();
 993        if (mnt->mnt_pinned) {
 994                mnt_add_count(mnt, 1);
 995                mnt->mnt_pinned--;
 996        }
 997        unlock_mount_hash();
 998}
 999EXPORT_SYMBOL(mnt_unpin);
1000
1001static inline void mangle(struct seq_file *m, const char *s)
1002{
1003        seq_escape(m, s, " \t\n\\");
1004}
1005
1006/*
1007 * Simple .show_options callback for filesystems which don't want to
1008 * implement more complex mount option showing.
1009 *
1010 * See also save_mount_options().
1011 */
1012int generic_show_options(struct seq_file *m, struct dentry *root)
1013{
1014        const char *options;
1015
1016        rcu_read_lock();
1017        options = rcu_dereference(root->d_sb->s_options);
1018
1019        if (options != NULL && options[0]) {
1020                seq_putc(m, ',');
1021                mangle(m, options);
1022        }
1023        rcu_read_unlock();
1024
1025        return 0;
1026}
1027EXPORT_SYMBOL(generic_show_options);
1028
1029/*
1030 * If filesystem uses generic_show_options(), this function should be
1031 * called from the fill_super() callback.
1032 *
1033 * The .remount_fs callback usually needs to be handled in a special
1034 * way, to make sure, that previous options are not overwritten if the
1035 * remount fails.
1036 *
1037 * Also note, that if the filesystem's .remount_fs function doesn't
1038 * reset all options to their default value, but changes only newly
1039 * given options, then the displayed options will not reflect reality
1040 * any more.
1041 */
1042void save_mount_options(struct super_block *sb, char *options)
1043{
1044        BUG_ON(sb->s_options);
1045        rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
1046}
1047EXPORT_SYMBOL(save_mount_options);
1048
1049void replace_mount_options(struct super_block *sb, char *options)
1050{
1051        char *old = sb->s_options;
1052        rcu_assign_pointer(sb->s_options, options);
1053        if (old) {
1054                synchronize_rcu();
1055                kfree(old);
1056        }
1057}
1058EXPORT_SYMBOL(replace_mount_options);
1059
1060#ifdef CONFIG_PROC_FS
1061/* iterator; we want it to have access to namespace_sem, thus here... */
1062static void *m_start(struct seq_file *m, loff_t *pos)
1063{
1064        struct proc_mounts *p = proc_mounts(m);
1065
1066        down_read(&namespace_sem);
1067        return seq_list_start(&p->ns->list, *pos);
1068}
1069
1070static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1071{
1072        struct proc_mounts *p = proc_mounts(m);
1073
1074        return seq_list_next(v, &p->ns->list, pos);
1075}
1076
1077static void m_stop(struct seq_file *m, void *v)
1078{
1079        up_read(&namespace_sem);
1080}
1081
1082static int m_show(struct seq_file *m, void *v)
1083{
1084        struct proc_mounts *p = proc_mounts(m);
1085        struct mount *r = list_entry(v, struct mount, mnt_list);
1086        return p->show(m, &r->mnt);
1087}
1088
1089const struct seq_operations mounts_op = {
1090        .start  = m_start,
1091        .next   = m_next,
1092        .stop   = m_stop,
1093        .show   = m_show,
1094};
1095#endif  /* CONFIG_PROC_FS */
1096
1097/**
1098 * may_umount_tree - check if a mount tree is busy
1099 * @mnt: root of mount tree
1100 *
1101 * This is called to check if a tree of mounts has any
1102 * open files, pwds, chroots or sub mounts that are
1103 * busy.
1104 */
1105int may_umount_tree(struct vfsmount *m)
1106{
1107        struct mount *mnt = real_mount(m);
1108        int actual_refs = 0;
1109        int minimum_refs = 0;
1110        struct mount *p;
1111        BUG_ON(!m);
1112
1113        /* write lock needed for mnt_get_count */
1114        lock_mount_hash();
1115        for (p = mnt; p; p = next_mnt(p, mnt)) {
1116                actual_refs += mnt_get_count(p);
1117                minimum_refs += 2;
1118        }
1119        unlock_mount_hash();
1120
1121        if (actual_refs > minimum_refs)
1122                return 0;
1123
1124        return 1;
1125}
1126
1127EXPORT_SYMBOL(may_umount_tree);
1128
1129/**
1130 * may_umount - check if a mount point is busy
1131 * @mnt: root of mount
1132 *
1133 * This is called to check if a mount point has any
1134 * open files, pwds, chroots or sub mounts. If the
1135 * mount has sub mounts this will return busy
1136 * regardless of whether the sub mounts are busy.
1137 *
1138 * Doesn't take quota and stuff into account. IOW, in some cases it will
1139 * give false negatives. The main reason why it's here is that we need
1140 * a non-destructive way to look for easily umountable filesystems.
1141 */
1142int may_umount(struct vfsmount *mnt)
1143{
1144        int ret = 1;
1145        down_read(&namespace_sem);
1146        lock_mount_hash();
1147        if (propagate_mount_busy(real_mount(mnt), 2))
1148                ret = 0;
1149        unlock_mount_hash();
1150        up_read(&namespace_sem);
1151        return ret;
1152}
1153
1154EXPORT_SYMBOL(may_umount);
1155
1156static LIST_HEAD(unmounted);    /* protected by namespace_sem */
1157
1158static void namespace_unlock(void)
1159{
1160        struct mount *mnt;
1161        LIST_HEAD(head);
1162
1163        if (likely(list_empty(&unmounted))) {
1164                up_write(&namespace_sem);
1165                return;
1166        }
1167
1168        list_splice_init(&unmounted, &head);
1169        up_write(&namespace_sem);
1170
1171        synchronize_rcu();
1172
1173        while (!list_empty(&head)) {
1174                mnt = list_first_entry(&head, struct mount, mnt_hash);
1175                list_del_init(&mnt->mnt_hash);
1176                if (mnt->mnt_ex_mountpoint.mnt)
1177                        path_put(&mnt->mnt_ex_mountpoint);
1178                mntput(&mnt->mnt);
1179        }
1180}
1181
1182static inline void namespace_lock(void)
1183{
1184        down_write(&namespace_sem);
1185}
1186
1187/*
1188 * mount_lock must be held
1189 * namespace_sem must be held for write
1190 * how = 0 => just this tree, don't propagate
1191 * how = 1 => propagate; we know that nobody else has reference to any victims
1192 * how = 2 => lazy umount
1193 */
1194void umount_tree(struct mount *mnt, int how)
1195{
1196        LIST_HEAD(tmp_list);
1197        struct mount *p;
1198
1199        for (p = mnt; p; p = next_mnt(p, mnt))
1200                list_move(&p->mnt_hash, &tmp_list);
1201
1202        if (how)
1203                propagate_umount(&tmp_list);
1204
1205        list_for_each_entry(p, &tmp_list, mnt_hash) {
1206                list_del_init(&p->mnt_expire);
1207                list_del_init(&p->mnt_list);
1208                __touch_mnt_namespace(p->mnt_ns);
1209                p->mnt_ns = NULL;
1210                if (how < 2)
1211                        p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1212                list_del_init(&p->mnt_child);
1213                if (mnt_has_parent(p)) {
1214                        put_mountpoint(p->mnt_mp);
1215                        /* move the reference to mountpoint into ->mnt_ex_mountpoint */
1216                        p->mnt_ex_mountpoint.dentry = p->mnt_mountpoint;
1217                        p->mnt_ex_mountpoint.mnt = &p->mnt_parent->mnt;
1218                        p->mnt_mountpoint = p->mnt.mnt_root;
1219                        p->mnt_parent = p;
1220                        p->mnt_mp = NULL;
1221                }
1222                change_mnt_propagation(p, MS_PRIVATE);
1223        }
1224        list_splice(&tmp_list, &unmounted);
1225}
1226
1227static void shrink_submounts(struct mount *mnt);
1228
1229static int do_umount(struct mount *mnt, int flags)
1230{
1231        struct super_block *sb = mnt->mnt.mnt_sb;
1232        int retval;
1233
1234        retval = security_sb_umount(&mnt->mnt, flags);
1235        if (retval)
1236                return retval;
1237
1238        /*
1239         * Allow userspace to request a mountpoint be expired rather than
1240         * unmounting unconditionally. Unmount only happens if:
1241         *  (1) the mark is already set (the mark is cleared by mntput())
1242         *  (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1243         */
1244        if (flags & MNT_EXPIRE) {
1245                if (&mnt->mnt == current->fs->root.mnt ||
1246                    flags & (MNT_FORCE | MNT_DETACH))
1247                        return -EINVAL;
1248
1249                /*
1250                 * probably don't strictly need the lock here if we examined
1251                 * all race cases, but it's a slowpath.
1252                 */
1253                lock_mount_hash();
1254                if (mnt_get_count(mnt) != 2) {
1255                        unlock_mount_hash();
1256                        return -EBUSY;
1257                }
1258                unlock_mount_hash();
1259
1260                if (!xchg(&mnt->mnt_expiry_mark, 1))
1261                        return -EAGAIN;
1262        }
1263
1264        /*
1265         * If we may have to abort operations to get out of this
1266         * mount, and they will themselves hold resources we must
1267         * allow the fs to do things. In the Unix tradition of
1268         * 'Gee thats tricky lets do it in userspace' the umount_begin
1269         * might fail to complete on the first run through as other tasks
1270         * must return, and the like. Thats for the mount program to worry
1271         * about for the moment.
1272         */
1273
1274        if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1275                sb->s_op->umount_begin(sb);
1276        }
1277
1278        /*
1279         * No sense to grab the lock for this test, but test itself looks
1280         * somewhat bogus. Suggestions for better replacement?
1281         * Ho-hum... In principle, we might treat that as umount + switch
1282         * to rootfs. GC would eventually take care of the old vfsmount.
1283         * Actually it makes sense, especially if rootfs would contain a
1284         * /reboot - static binary that would close all descriptors and
1285         * call reboot(9). Then init(8) could umount root and exec /reboot.
1286         */
1287        if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1288                /*
1289                 * Special case for "unmounting" root ...
1290                 * we just try to remount it readonly.
1291                 */
1292                down_write(&sb->s_umount);
1293                if (!(sb->s_flags & MS_RDONLY))
1294                        retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1295                up_write(&sb->s_umount);
1296                return retval;
1297        }
1298
1299        namespace_lock();
1300        lock_mount_hash();
1301        event++;
1302
1303        if (flags & MNT_DETACH) {
1304                if (!list_empty(&mnt->mnt_list))
1305                        umount_tree(mnt, 2);
1306                retval = 0;
1307        } else {
1308                shrink_submounts(mnt);
1309                retval = -EBUSY;
1310                if (!propagate_mount_busy(mnt, 2)) {
1311                        if (!list_empty(&mnt->mnt_list))
1312                                umount_tree(mnt, 1);
1313                        retval = 0;
1314                }
1315        }
1316        unlock_mount_hash();
1317        namespace_unlock();
1318        return retval;
1319}
1320
1321/* 
1322 * Is the caller allowed to modify his namespace?
1323 */
1324static inline bool may_mount(void)
1325{
1326        return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1327}
1328
1329/*
1330 * Now umount can handle mount points as well as block devices.
1331 * This is important for filesystems which use unnamed block devices.
1332 *
1333 * We now support a flag for forced unmount like the other 'big iron'
1334 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1335 */
1336
1337SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1338{
1339        struct path path;
1340        struct mount *mnt;
1341        int retval;
1342        int lookup_flags = 0;
1343
1344        if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1345                return -EINVAL;
1346
1347        if (!may_mount())
1348                return -EPERM;
1349
1350        if (!(flags & UMOUNT_NOFOLLOW))
1351                lookup_flags |= LOOKUP_FOLLOW;
1352
1353        retval = user_path_mountpoint_at(AT_FDCWD, name, lookup_flags, &path);
1354        if (retval)
1355                goto out;
1356        mnt = real_mount(path.mnt);
1357        retval = -EINVAL;
1358        if (path.dentry != path.mnt->mnt_root)
1359                goto dput_and_out;
1360        if (!check_mnt(mnt))
1361                goto dput_and_out;
1362        if (mnt->mnt.mnt_flags & MNT_LOCKED)
1363                goto dput_and_out;
1364
1365        retval = do_umount(mnt, flags);
1366dput_and_out:
1367        /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1368        dput(path.dentry);
1369        mntput_no_expire(mnt);
1370out:
1371        return retval;
1372}
1373
1374#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1375
1376/*
1377 *      The 2.0 compatible umount. No flags.
1378 */
1379SYSCALL_DEFINE1(oldumount, char __user *, name)
1380{
1381        return sys_umount(name, 0);
1382}
1383
1384#endif
1385
1386static bool is_mnt_ns_file(struct dentry *dentry)
1387{
1388        /* Is this a proxy for a mount namespace? */
1389        struct inode *inode = dentry->d_inode;
1390        struct proc_ns *ei;
1391
1392        if (!proc_ns_inode(inode))
1393                return false;
1394
1395        ei = get_proc_ns(inode);
1396        if (ei->ns_ops != &mntns_operations)
1397                return false;
1398
1399        return true;
1400}
1401
1402static bool mnt_ns_loop(struct dentry *dentry)
1403{
1404        /* Could bind mounting the mount namespace inode cause a
1405         * mount namespace loop?
1406         */
1407        struct mnt_namespace *mnt_ns;
1408        if (!is_mnt_ns_file(dentry))
1409                return false;
1410
1411        mnt_ns = get_proc_ns(dentry->d_inode)->ns;
1412        return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1413}
1414
1415struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1416                                        int flag)
1417{
1418        struct mount *res, *p, *q, *r, *parent;
1419
1420        if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1421                return ERR_PTR(-EINVAL);
1422
1423        if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1424                return ERR_PTR(-EINVAL);
1425
1426        res = q = clone_mnt(mnt, dentry, flag);
1427        if (IS_ERR(q))
1428                return q;
1429
1430        q->mnt.mnt_flags &= ~MNT_LOCKED;
1431        q->mnt_mountpoint = mnt->mnt_mountpoint;
1432
1433        p = mnt;
1434        list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1435                struct mount *s;
1436                if (!is_subdir(r->mnt_mountpoint, dentry))
1437                        continue;
1438
1439                for (s = r; s; s = next_mnt(s, r)) {
1440                        if (!(flag & CL_COPY_UNBINDABLE) &&
1441                            IS_MNT_UNBINDABLE(s)) {
1442                                s = skip_mnt_tree(s);
1443                                continue;
1444                        }
1445                        if (!(flag & CL_COPY_MNT_NS_FILE) &&
1446                            is_mnt_ns_file(s->mnt.mnt_root)) {
1447                                s = skip_mnt_tree(s);
1448                                continue;
1449                        }
1450                        while (p != s->mnt_parent) {
1451                                p = p->mnt_parent;
1452                                q = q->mnt_parent;
1453                        }
1454                        p = s;
1455                        parent = q;
1456                        q = clone_mnt(p, p->mnt.mnt_root, flag);
1457                        if (IS_ERR(q))
1458                                goto out;
1459                        lock_mount_hash();
1460                        list_add_tail(&q->mnt_list, &res->mnt_list);
1461                        attach_mnt(q, parent, p->mnt_mp);
1462                        unlock_mount_hash();
1463                }
1464        }
1465        return res;
1466out:
1467        if (res) {
1468                lock_mount_hash();
1469                umount_tree(res, 0);
1470                unlock_mount_hash();
1471        }
1472        return q;
1473}
1474
1475/* Caller should check returned pointer for errors */
1476
1477struct vfsmount *collect_mounts(struct path *path)
1478{
1479        struct mount *tree;
1480        namespace_lock();
1481        tree = copy_tree(real_mount(path->mnt), path->dentry,
1482                         CL_COPY_ALL | CL_PRIVATE);
1483        namespace_unlock();
1484        if (IS_ERR(tree))
1485                return ERR_CAST(tree);
1486        return &tree->mnt;
1487}
1488
1489void drop_collected_mounts(struct vfsmount *mnt)
1490{
1491        namespace_lock();
1492        lock_mount_hash();
1493        umount_tree(real_mount(mnt), 0);
1494        unlock_mount_hash();
1495        namespace_unlock();
1496}
1497
1498int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1499                   struct vfsmount *root)
1500{
1501        struct mount *mnt;
1502        int res = f(root, arg);
1503        if (res)
1504                return res;
1505        list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1506                res = f(&mnt->mnt, arg);
1507                if (res)
1508                        return res;
1509        }
1510        return 0;
1511}
1512
1513static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1514{
1515        struct mount *p;
1516
1517        for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1518                if (p->mnt_group_id && !IS_MNT_SHARED(p))
1519                        mnt_release_group_id(p);
1520        }
1521}
1522
1523static int invent_group_ids(struct mount *mnt, bool recurse)
1524{
1525        struct mount *p;
1526
1527        for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1528                if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1529                        int err = mnt_alloc_group_id(p);
1530                        if (err) {
1531                                cleanup_group_ids(mnt, p);
1532                                return err;
1533                        }
1534                }
1535        }
1536
1537        return 0;
1538}
1539
1540/*
1541 *  @source_mnt : mount tree to be attached
1542 *  @nd         : place the mount tree @source_mnt is attached
1543 *  @parent_nd  : if non-null, detach the source_mnt from its parent and
1544 *                 store the parent mount and mountpoint dentry.
1545 *                 (done when source_mnt is moved)
1546 *
1547 *  NOTE: in the table below explains the semantics when a source mount
1548 *  of a given type is attached to a destination mount of a given type.
1549 * ---------------------------------------------------------------------------
1550 * |         BIND MOUNT OPERATION                                            |
1551 * |**************************************************************************
1552 * | source-->| shared        |       private  |       slave    | unbindable |
1553 * | dest     |               |                |                |            |
1554 * |   |      |               |                |                |            |
1555 * |   v      |               |                |                |            |
1556 * |**************************************************************************
1557 * |  shared  | shared (++)   |     shared (+) |     shared(+++)|  invalid   |
1558 * |          |               |                |                |            |
1559 * |non-shared| shared (+)    |      private   |      slave (*) |  invalid   |
1560 * ***************************************************************************
1561 * A bind operation clones the source mount and mounts the clone on the
1562 * destination mount.
1563 *
1564 * (++)  the cloned mount is propagated to all the mounts in the propagation
1565 *       tree of the destination mount and the cloned mount is added to
1566 *       the peer group of the source mount.
1567 * (+)   the cloned mount is created under the destination mount and is marked
1568 *       as shared. The cloned mount is added to the peer group of the source
1569 *       mount.
1570 * (+++) the mount is propagated to all the mounts in the propagation tree
1571 *       of the destination mount and the cloned mount is made slave
1572 *       of the same master as that of the source mount. The cloned mount
1573 *       is marked as 'shared and slave'.
1574 * (*)   the cloned mount is made a slave of the same master as that of the
1575 *       source mount.
1576 *
1577 * ---------------------------------------------------------------------------
1578 * |                    MOVE MOUNT OPERATION                                 |
1579 * |**************************************************************************
1580 * | source-->| shared        |       private  |       slave    | unbindable |
1581 * | dest     |               |                |                |            |
1582 * |   |      |               |                |                |            |
1583 * |   v      |               |                |                |            |
1584 * |**************************************************************************
1585 * |  shared  | shared (+)    |     shared (+) |    shared(+++) |  invalid   |
1586 * |          |               |                |                |            |
1587 * |non-shared| shared (+*)   |      private   |    slave (*)   | unbindable |
1588 * ***************************************************************************
1589 *
1590 * (+)  the mount is moved to the destination. And is then propagated to
1591 *      all the mounts in the propagation tree of the destination mount.
1592 * (+*)  the mount is moved to the destination.
1593 * (+++)  the mount is moved to the destination and is then propagated to
1594 *      all the mounts belonging to the destination mount's propagation tree.
1595 *      the mount is marked as 'shared and slave'.
1596 * (*)  the mount continues to be a slave at the new location.
1597 *
1598 * if the source mount is a tree, the operations explained above is
1599 * applied to each mount in the tree.
1600 * Must be called without spinlocks held, since this function can sleep
1601 * in allocations.
1602 */
1603static int attach_recursive_mnt(struct mount *source_mnt,
1604                        struct mount *dest_mnt,
1605                        struct mountpoint *dest_mp,
1606                        struct path *parent_path)
1607{
1608        LIST_HEAD(tree_list);
1609        struct mount *child, *p;
1610        int err;
1611
1612        if (IS_MNT_SHARED(dest_mnt)) {
1613                err = invent_group_ids(source_mnt, true);
1614                if (err)
1615                        goto out;
1616        }
1617        err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
1618        if (err)
1619                goto out_cleanup_ids;
1620
1621        lock_mount_hash();
1622
1623        if (IS_MNT_SHARED(dest_mnt)) {
1624                for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1625                        set_mnt_shared(p);
1626        }
1627        if (parent_path) {
1628                detach_mnt(source_mnt, parent_path);
1629                attach_mnt(source_mnt, dest_mnt, dest_mp);
1630                touch_mnt_namespace(source_mnt->mnt_ns);
1631        } else {
1632                mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
1633                commit_tree(source_mnt);
1634        }
1635
1636        list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1637                list_del_init(&child->mnt_hash);
1638                commit_tree(child);
1639        }
1640        unlock_mount_hash();
1641
1642        return 0;
1643
1644 out_cleanup_ids:
1645        if (IS_MNT_SHARED(dest_mnt))
1646                cleanup_group_ids(source_mnt, NULL);
1647 out:
1648        return err;
1649}
1650
1651static struct mountpoint *lock_mount(struct path *path)
1652{
1653        struct vfsmount *mnt;
1654        struct dentry *dentry = path->dentry;
1655retry:
1656        mutex_lock(&dentry->d_inode->i_mutex);
1657        if (unlikely(cant_mount(dentry))) {
1658                mutex_unlock(&dentry->d_inode->i_mutex);
1659                return ERR_PTR(-ENOENT);
1660        }
1661        namespace_lock();
1662        mnt = lookup_mnt(path);
1663        if (likely(!mnt)) {
1664                struct mountpoint *mp = new_mountpoint(dentry);
1665                if (IS_ERR(mp)) {
1666                        namespace_unlock();
1667                        mutex_unlock(&dentry->d_inode->i_mutex);
1668                        return mp;
1669                }
1670                return mp;
1671        }
1672        namespace_unlock();
1673        mutex_unlock(&path->dentry->d_inode->i_mutex);
1674        path_put(path);
1675        path->mnt = mnt;
1676        dentry = path->dentry = dget(mnt->mnt_root);
1677        goto retry;
1678}
1679
1680static void unlock_mount(struct mountpoint *where)
1681{
1682        struct dentry *dentry = where->m_dentry;
1683        put_mountpoint(where);
1684        namespace_unlock();
1685        mutex_unlock(&dentry->d_inode->i_mutex);
1686}
1687
1688static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
1689{
1690        if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1691                return -EINVAL;
1692
1693        if (S_ISDIR(mp->m_dentry->d_inode->i_mode) !=
1694              S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
1695                return -ENOTDIR;
1696
1697        return attach_recursive_mnt(mnt, p, mp, NULL);
1698}
1699
1700/*
1701 * Sanity check the flags to change_mnt_propagation.
1702 */
1703
1704static int flags_to_propagation_type(int flags)
1705{
1706        int type = flags & ~(MS_REC | MS_SILENT);
1707
1708        /* Fail if any non-propagation flags are set */
1709        if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1710                return 0;
1711        /* Only one propagation flag should be set */
1712        if (!is_power_of_2(type))
1713                return 0;
1714        return type;
1715}
1716
1717/*
1718 * recursively change the type of the mountpoint.
1719 */
1720static int do_change_type(struct path *path, int flag)
1721{
1722        struct mount *m;
1723        struct mount *mnt = real_mount(path->mnt);
1724        int recurse = flag & MS_REC;
1725        int type;
1726        int err = 0;
1727
1728        if (path->dentry != path->mnt->mnt_root)
1729                return -EINVAL;
1730
1731        type = flags_to_propagation_type(flag);
1732        if (!type)
1733                return -EINVAL;
1734
1735        namespace_lock();
1736        if (type == MS_SHARED) {
1737                err = invent_group_ids(mnt, recurse);
1738                if (err)
1739                        goto out_unlock;
1740        }
1741
1742        lock_mount_hash();
1743        for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1744                change_mnt_propagation(m, type);
1745        unlock_mount_hash();
1746
1747 out_unlock:
1748        namespace_unlock();
1749        return err;
1750}
1751
1752static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
1753{
1754        struct mount *child;
1755        list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
1756                if (!is_subdir(child->mnt_mountpoint, dentry))
1757                        continue;
1758
1759                if (child->mnt.mnt_flags & MNT_LOCKED)
1760                        return true;
1761        }
1762        return false;
1763}
1764
1765/*
1766 * do loopback mount.
1767 */
1768static int do_loopback(struct path *path, const char *old_name,
1769                                int recurse)
1770{
1771        struct path old_path;
1772        struct mount *mnt = NULL, *old, *parent;
1773        struct mountpoint *mp;
1774        int err;
1775        if (!old_name || !*old_name)
1776                return -EINVAL;
1777        err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1778        if (err)
1779                return err;
1780
1781        err = -EINVAL;
1782        if (mnt_ns_loop(old_path.dentry))
1783                goto out; 
1784
1785        mp = lock_mount(path);
1786        err = PTR_ERR(mp);
1787        if (IS_ERR(mp))
1788                goto out;
1789
1790        old = real_mount(old_path.mnt);
1791        parent = real_mount(path->mnt);
1792
1793        err = -EINVAL;
1794        if (IS_MNT_UNBINDABLE(old))
1795                goto out2;
1796
1797        if (!check_mnt(parent) || !check_mnt(old))
1798                goto out2;
1799
1800        if (!recurse && has_locked_children(old, old_path.dentry))
1801                goto out2;
1802
1803        if (recurse)
1804                mnt = copy_tree(old, old_path.dentry, CL_COPY_MNT_NS_FILE);
1805        else
1806                mnt = clone_mnt(old, old_path.dentry, 0);
1807
1808        if (IS_ERR(mnt)) {
1809                err = PTR_ERR(mnt);
1810                goto out2;
1811        }
1812
1813        mnt->mnt.mnt_flags &= ~MNT_LOCKED;
1814
1815        err = graft_tree(mnt, parent, mp);
1816        if (err) {
1817                lock_mount_hash();
1818                umount_tree(mnt, 0);
1819                unlock_mount_hash();
1820        }
1821out2:
1822        unlock_mount(mp);
1823out:
1824        path_put(&old_path);
1825        return err;
1826}
1827
1828static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1829{
1830        int error = 0;
1831        int readonly_request = 0;
1832
1833        if (ms_flags & MS_RDONLY)
1834                readonly_request = 1;
1835        if (readonly_request == __mnt_is_readonly(mnt))
1836                return 0;
1837
1838        if (mnt->mnt_flags & MNT_LOCK_READONLY)
1839                return -EPERM;
1840
1841        if (readonly_request)
1842                error = mnt_make_readonly(real_mount(mnt));
1843        else
1844                __mnt_unmake_readonly(real_mount(mnt));
1845        return error;
1846}
1847
1848/*
1849 * change filesystem flags. dir should be a physical root of filesystem.
1850 * If you've mounted a non-root directory somewhere and want to do remount
1851 * on it - tough luck.
1852 */
1853static int do_remount(struct path *path, int flags, int mnt_flags,
1854                      void *data)
1855{
1856        int err;
1857        struct super_block *sb = path->mnt->mnt_sb;
1858        struct mount *mnt = real_mount(path->mnt);
1859
1860        if (!check_mnt(mnt))
1861                return -EINVAL;
1862
1863        if (path->dentry != path->mnt->mnt_root)
1864                return -EINVAL;
1865
1866        err = security_sb_remount(sb, data);
1867        if (err)
1868                return err;
1869
1870        down_write(&sb->s_umount);
1871        if (flags & MS_BIND)
1872                err = change_mount_flags(path->mnt, flags);
1873        else if (!capable(CAP_SYS_ADMIN))
1874                err = -EPERM;
1875        else
1876                err = do_remount_sb(sb, flags, data, 0);
1877        if (!err) {
1878                lock_mount_hash();
1879                mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK;
1880                mnt->mnt.mnt_flags = mnt_flags;
1881                touch_mnt_namespace(mnt->mnt_ns);
1882                unlock_mount_hash();
1883        }
1884        up_write(&sb->s_umount);
1885        return err;
1886}
1887
1888static inline int tree_contains_unbindable(struct mount *mnt)
1889{
1890        struct mount *p;
1891        for (p = mnt; p; p = next_mnt(p, mnt)) {
1892                if (IS_MNT_UNBINDABLE(p))
1893                        return 1;
1894        }
1895        return 0;
1896}
1897
1898static int do_move_mount(struct path *path, const char *old_name)
1899{
1900        struct path old_path, parent_path;
1901        struct mount *p;
1902        struct mount *old;
1903        struct mountpoint *mp;
1904        int err;
1905        if (!old_name || !*old_name)
1906                return -EINVAL;
1907        err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1908        if (err)
1909                return err;
1910
1911        mp = lock_mount(path);
1912        err = PTR_ERR(mp);
1913        if (IS_ERR(mp))
1914                goto out;
1915
1916        old = real_mount(old_path.mnt);
1917        p = real_mount(path->mnt);
1918
1919        err = -EINVAL;
1920        if (!check_mnt(p) || !check_mnt(old))
1921                goto out1;
1922
1923        if (old->mnt.mnt_flags & MNT_LOCKED)
1924                goto out1;
1925
1926        err = -EINVAL;
1927        if (old_path.dentry != old_path.mnt->mnt_root)
1928                goto out1;
1929
1930        if (!mnt_has_parent(old))
1931                goto out1;
1932
1933        if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1934              S_ISDIR(old_path.dentry->d_inode->i_mode))
1935                goto out1;
1936        /*
1937         * Don't move a mount residing in a shared parent.
1938         */
1939        if (IS_MNT_SHARED(old->mnt_parent))
1940                goto out1;
1941        /*
1942         * Don't move a mount tree containing unbindable mounts to a destination
1943         * mount which is shared.
1944         */
1945        if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
1946                goto out1;
1947        err = -ELOOP;
1948        for (; mnt_has_parent(p); p = p->mnt_parent)
1949                if (p == old)
1950                        goto out1;
1951
1952        err = attach_recursive_mnt(old, real_mount(path->mnt), mp, &parent_path);
1953        if (err)
1954                goto out1;
1955
1956        /* if the mount is moved, it should no longer be expire
1957         * automatically */
1958        list_del_init(&old->mnt_expire);
1959out1:
1960        unlock_mount(mp);
1961out:
1962        if (!err)
1963                path_put(&parent_path);
1964        path_put(&old_path);
1965        return err;
1966}
1967
1968static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1969{
1970        int err;
1971        const char *subtype = strchr(fstype, '.');
1972        if (subtype) {
1973                subtype++;
1974                err = -EINVAL;
1975                if (!subtype[0])
1976                        goto err;
1977        } else
1978                subtype = "";
1979
1980        mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1981        err = -ENOMEM;
1982        if (!mnt->mnt_sb->s_subtype)
1983                goto err;
1984        return mnt;
1985
1986 err:
1987        mntput(mnt);
1988        return ERR_PTR(err);
1989}
1990
1991/*
1992 * add a mount into a namespace's mount tree
1993 */
1994static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
1995{
1996        struct mountpoint *mp;
1997        struct mount *parent;
1998        int err;
1999
2000        mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL | MNT_DOOMED | MNT_SYNC_UMOUNT);
2001
2002        mp = lock_mount(path);
2003        if (IS_ERR(mp))
2004                return PTR_ERR(mp);
2005
2006        parent = real_mount(path->mnt);
2007        err = -EINVAL;
2008        if (unlikely(!check_mnt(parent))) {
2009                /* that's acceptable only for automounts done in private ns */
2010                if (!(mnt_flags & MNT_SHRINKABLE))
2011                        goto unlock;
2012                /* ... and for those we'd better have mountpoint still alive */
2013                if (!parent->mnt_ns)
2014                        goto unlock;
2015        }
2016
2017        /* Refuse the same filesystem on the same mount point */
2018        err = -EBUSY;
2019        if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2020            path->mnt->mnt_root == path->dentry)
2021                goto unlock;
2022
2023        err = -EINVAL;
2024        if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
2025                goto unlock;
2026
2027        newmnt->mnt.mnt_flags = mnt_flags;
2028        err = graft_tree(newmnt, parent, mp);
2029
2030unlock:
2031        unlock_mount(mp);
2032        return err;
2033}
2034
2035/*
2036 * create a new mount for userspace and request it to be added into the
2037 * namespace's tree
2038 */
2039static int do_new_mount(struct path *path, const char *fstype, int flags,
2040                        int mnt_flags, const char *name, void *data)
2041{
2042        struct file_system_type *type;
2043        struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2044        struct vfsmount *mnt;
2045        int err;
2046
2047        if (!fstype)
2048                return -EINVAL;
2049
2050        type = get_fs_type(fstype);
2051        if (!type)
2052                return -ENODEV;
2053
2054        if (user_ns != &init_user_ns) {
2055                if (!(type->fs_flags & FS_USERNS_MOUNT)) {
2056                        put_filesystem(type);
2057                        return -EPERM;
2058                }
2059                /* Only in special cases allow devices from mounts
2060                 * created outside the initial user namespace.
2061                 */
2062                if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
2063                        flags |= MS_NODEV;
2064                        mnt_flags |= MNT_NODEV;
2065                }
2066        }
2067
2068        mnt = vfs_kern_mount(type, flags, name, data);
2069        if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
2070            !mnt->mnt_sb->s_subtype)
2071                mnt = fs_set_subtype(mnt, fstype);
2072
2073        put_filesystem(type);
2074        if (IS_ERR(mnt))
2075                return PTR_ERR(mnt);
2076
2077        err = do_add_mount(real_mount(mnt), path, mnt_flags);
2078        if (err)
2079                mntput(mnt);
2080        return err;
2081}
2082
2083int finish_automount(struct vfsmount *m, struct path *path)
2084{
2085        struct mount *mnt = real_mount(m);
2086        int err;
2087        /* The new mount record should have at least 2 refs to prevent it being
2088         * expired before we get a chance to add it
2089         */
2090        BUG_ON(mnt_get_count(mnt) < 2);
2091
2092        if (m->mnt_sb == path->mnt->mnt_sb &&
2093            m->mnt_root == path->dentry) {
2094                err = -ELOOP;
2095                goto fail;
2096        }
2097
2098        err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2099        if (!err)
2100                return 0;
2101fail:
2102        /* remove m from any expiration list it may be on */
2103        if (!list_empty(&mnt->mnt_expire)) {
2104                namespace_lock();
2105                list_del_init(&mnt->mnt_expire);
2106                namespace_unlock();
2107        }
2108        mntput(m);
2109        mntput(m);
2110        return err;
2111}
2112
2113/**
2114 * mnt_set_expiry - Put a mount on an expiration list
2115 * @mnt: The mount to list.
2116 * @expiry_list: The list to add the mount to.
2117 */
2118void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2119{
2120        namespace_lock();
2121
2122        list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2123
2124        namespace_unlock();
2125}
2126EXPORT_SYMBOL(mnt_set_expiry);
2127
2128/*
2129 * process a list of expirable mountpoints with the intent of discarding any
2130 * mountpoints that aren't in use and haven't been touched since last we came
2131 * here
2132 */
2133void mark_mounts_for_expiry(struct list_head *mounts)
2134{
2135        struct mount *mnt, *next;
2136        LIST_HEAD(graveyard);
2137
2138        if (list_empty(mounts))
2139                return;
2140
2141        namespace_lock();
2142        lock_mount_hash();
2143
2144        /* extract from the expiration list every vfsmount that matches the
2145         * following criteria:
2146         * - only referenced by its parent vfsmount
2147         * - still marked for expiry (marked on the last call here; marks are
2148         *   cleared by mntput())
2149         */
2150        list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2151                if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2152                        propagate_mount_busy(mnt, 1))
2153                        continue;
2154                list_move(&mnt->mnt_expire, &graveyard);
2155        }
2156        while (!list_empty(&graveyard)) {
2157                mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2158                touch_mnt_namespace(mnt->mnt_ns);
2159                umount_tree(mnt, 1);
2160        }
2161        unlock_mount_hash();
2162        namespace_unlock();
2163}
2164
2165EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2166
2167/*
2168 * Ripoff of 'select_parent()'
2169 *
2170 * search the list of submounts for a given mountpoint, and move any
2171 * shrinkable submounts to the 'graveyard' list.
2172 */
2173static int select_submounts(struct mount *parent, struct list_head *graveyard)
2174{
2175        struct mount *this_parent = parent;
2176        struct list_head *next;
2177        int found = 0;
2178
2179repeat:
2180        next = this_parent->mnt_mounts.next;
2181resume:
2182        while (next != &this_parent->mnt_mounts) {
2183                struct list_head *tmp = next;
2184                struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2185
2186                next = tmp->next;
2187                if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2188                        continue;
2189                /*
2190                 * Descend a level if the d_mounts list is non-empty.
2191                 */
2192                if (!list_empty(&mnt->mnt_mounts)) {
2193                        this_parent = mnt;
2194                        goto repeat;
2195                }
2196
2197                if (!propagate_mount_busy(mnt, 1)) {
2198                        list_move_tail(&mnt->mnt_expire, graveyard);
2199                        found++;
2200                }
2201        }
2202        /*
2203         * All done at this level ... ascend and resume the search
2204         */
2205        if (this_parent != parent) {
2206                next = this_parent->mnt_child.next;
2207                this_parent = this_parent->mnt_parent;
2208                goto resume;
2209        }
2210        return found;
2211}
2212
2213/*
2214 * process a list of expirable mountpoints with the intent of discarding any
2215 * submounts of a specific parent mountpoint
2216 *
2217 * mount_lock must be held for write
2218 */
2219static void shrink_submounts(struct mount *mnt)
2220{
2221        LIST_HEAD(graveyard);
2222        struct mount *m;
2223
2224        /* extract submounts of 'mountpoint' from the expiration list */
2225        while (select_submounts(mnt, &graveyard)) {
2226                while (!list_empty(&graveyard)) {
2227                        m = list_first_entry(&graveyard, struct mount,
2228                                                mnt_expire);
2229                        touch_mnt_namespace(m->mnt_ns);
2230                        umount_tree(m, 1);
2231                }
2232        }
2233}
2234
2235/*
2236 * Some copy_from_user() implementations do not return the exact number of
2237 * bytes remaining to copy on a fault.  But copy_mount_options() requires that.
2238 * Note that this function differs from copy_from_user() in that it will oops
2239 * on bad values of `to', rather than returning a short copy.
2240 */
2241static long exact_copy_from_user(void *to, const void __user * from,
2242                                 unsigned long n)
2243{
2244        char *t = to;
2245        const char __user *f = from;
2246        char c;
2247
2248        if (!access_ok(VERIFY_READ, from, n))
2249                return n;
2250
2251        while (n) {
2252                if (__get_user(c, f)) {
2253                        memset(t, 0, n);
2254                        break;
2255                }
2256                *t++ = c;
2257                f++;
2258                n--;
2259        }
2260        return n;
2261}
2262
2263int copy_mount_options(const void __user * data, unsigned long *where)
2264{
2265        int i;
2266        unsigned long page;
2267        unsigned long size;
2268
2269        *where = 0;
2270        if (!data)
2271                return 0;
2272
2273        if (!(page = __get_free_page(GFP_KERNEL)))
2274                return -ENOMEM;
2275
2276        /* We only care that *some* data at the address the user
2277         * gave us is valid.  Just in case, we'll zero
2278         * the remainder of the page.
2279         */
2280        /* copy_from_user cannot cross TASK_SIZE ! */
2281        size = TASK_SIZE - (unsigned long)data;
2282        if (size > PAGE_SIZE)
2283                size = PAGE_SIZE;
2284
2285        i = size - exact_copy_from_user((void *)page, data, size);
2286        if (!i) {
2287                free_page(page);
2288                return -EFAULT;
2289        }
2290        if (i != PAGE_SIZE)
2291                memset((char *)page + i, 0, PAGE_SIZE - i);
2292        *where = page;
2293        return 0;
2294}
2295
2296int copy_mount_string(const void __user *data, char **where)
2297{
2298        char *tmp;
2299
2300        if (!data) {
2301                *where = NULL;
2302                return 0;
2303        }
2304
2305        tmp = strndup_user(data, PAGE_SIZE);
2306        if (IS_ERR(tmp))
2307                return PTR_ERR(tmp);
2308
2309        *where = tmp;
2310        return 0;
2311}
2312
2313/*
2314 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2315 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2316 *
2317 * data is a (void *) that can point to any structure up to
2318 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2319 * information (or be NULL).
2320 *
2321 * Pre-0.97 versions of mount() didn't have a flags word.
2322 * When the flags word was introduced its top half was required
2323 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2324 * Therefore, if this magic number is present, it carries no information
2325 * and must be discarded.
2326 */
2327long do_mount(const char *dev_name, const char *dir_name,
2328                const char *type_page, unsigned long flags, void *data_page)
2329{
2330        struct path path;
2331        int retval = 0;
2332        int mnt_flags = 0;
2333
2334        /* Discard magic */
2335        if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2336                flags &= ~MS_MGC_MSK;
2337
2338        /* Basic sanity checks */
2339
2340        if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2341                return -EINVAL;
2342
2343        if (data_page)
2344                ((char *)data_page)[PAGE_SIZE - 1] = 0;
2345
2346        /* ... and get the mountpoint */
2347        retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2348        if (retval)
2349                return retval;
2350
2351        retval = security_sb_mount(dev_name, &path,
2352                                   type_page, flags, data_page);
2353        if (!retval && !may_mount())
2354                retval = -EPERM;
2355        if (retval)
2356                goto dput_out;
2357
2358        /* Default to relatime unless overriden */
2359        if (!(flags & MS_NOATIME))
2360                mnt_flags |= MNT_RELATIME;
2361
2362        /* Separate the per-mountpoint flags */
2363        if (flags & MS_NOSUID)
2364                mnt_flags |= MNT_NOSUID;
2365        if (flags & MS_NODEV)
2366                mnt_flags |= MNT_NODEV;
2367        if (flags & MS_NOEXEC)
2368                mnt_flags |= MNT_NOEXEC;
2369        if (flags & MS_NOATIME)
2370                mnt_flags |= MNT_NOATIME;
2371        if (flags & MS_NODIRATIME)
2372                mnt_flags |= MNT_NODIRATIME;
2373        if (flags & MS_STRICTATIME)
2374                mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2375        if (flags & MS_RDONLY)
2376                mnt_flags |= MNT_READONLY;
2377
2378        flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2379                   MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2380                   MS_STRICTATIME);
2381
2382        if (flags & MS_REMOUNT)
2383                retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2384                                    data_page);
2385        else if (flags & MS_BIND)
2386                retval = do_loopback(&path, dev_name, flags & MS_REC);
2387        else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2388                retval = do_change_type(&path, flags);
2389        else if (flags & MS_MOVE)
2390                retval = do_move_mount(&path, dev_name);
2391        else
2392                retval = do_new_mount(&path, type_page, flags, mnt_flags,
2393                                      dev_name, data_page);
2394dput_out:
2395        path_put(&path);
2396        return retval;
2397}
2398
2399static void free_mnt_ns(struct mnt_namespace *ns)
2400{
2401        proc_free_inum(ns->proc_inum);
2402        put_user_ns(ns->user_ns);
2403        kfree(ns);
2404}
2405
2406/*
2407 * Assign a sequence number so we can detect when we attempt to bind
2408 * mount a reference to an older mount namespace into the current
2409 * mount namespace, preventing reference counting loops.  A 64bit
2410 * number incrementing at 10Ghz will take 12,427 years to wrap which
2411 * is effectively never, so we can ignore the possibility.
2412 */
2413static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2414
2415static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2416{
2417        struct mnt_namespace *new_ns;
2418        int ret;
2419
2420        new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2421        if (!new_ns)
2422                return ERR_PTR(-ENOMEM);
2423        ret = proc_alloc_inum(&new_ns->proc_inum);
2424        if (ret) {
2425                kfree(new_ns);
2426                return ERR_PTR(ret);
2427        }
2428        new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2429        atomic_set(&new_ns->count, 1);
2430        new_ns->root = NULL;
2431        INIT_LIST_HEAD(&new_ns->list);
2432        init_waitqueue_head(&new_ns->poll);
2433        new_ns->event = 0;
2434        new_ns->user_ns = get_user_ns(user_ns);
2435        return new_ns;
2436}
2437
2438struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2439                struct user_namespace *user_ns, struct fs_struct *new_fs)
2440{
2441        struct mnt_namespace *new_ns;
2442        struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2443        struct mount *p, *q;
2444        struct mount *old;
2445        struct mount *new;
2446        int copy_flags;
2447
2448        BUG_ON(!ns);
2449
2450        if (likely(!(flags & CLONE_NEWNS))) {
2451                get_mnt_ns(ns);
2452                return ns;
2453        }
2454
2455        old = ns->root;
2456
2457        new_ns = alloc_mnt_ns(user_ns);
2458        if (IS_ERR(new_ns))
2459                return new_ns;
2460
2461        namespace_lock();
2462        /* First pass: copy the tree topology */
2463        copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
2464        if (user_ns != ns->user_ns)
2465                copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2466        new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2467        if (IS_ERR(new)) {
2468                namespace_unlock();
2469                free_mnt_ns(new_ns);
2470                return ERR_CAST(new);
2471        }
2472        new_ns->root = new;
2473        list_add_tail(&new_ns->list, &new->mnt_list);
2474
2475        /*
2476         * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2477         * as belonging to new namespace.  We have already acquired a private
2478         * fs_struct, so tsk->fs->lock is not needed.
2479         */
2480        p = old;
2481        q = new;
2482        while (p) {
2483                q->mnt_ns = new_ns;
2484                if (new_fs) {
2485                        if (&p->mnt == new_fs->root.mnt) {
2486                                new_fs->root.mnt = mntget(&q->mnt);
2487                                rootmnt = &p->mnt;
2488                        }
2489                        if (&p->mnt == new_fs->pwd.mnt) {
2490                                new_fs->pwd.mnt = mntget(&q->mnt);
2491                                pwdmnt = &p->mnt;
2492                        }
2493                }
2494                p = next_mnt(p, old);
2495                q = next_mnt(q, new);
2496                if (!q)
2497                        break;
2498                while (p->mnt.mnt_root != q->mnt.mnt_root)
2499                        p = next_mnt(p, old);
2500        }
2501        namespace_unlock();
2502
2503        if (rootmnt)
2504                mntput(rootmnt);
2505        if (pwdmnt)
2506                mntput(pwdmnt);
2507
2508        return new_ns;
2509}
2510
2511/**
2512 * create_mnt_ns - creates a private namespace and adds a root filesystem
2513 * @mnt: pointer to the new root filesystem mountpoint
2514 */
2515static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2516{
2517        struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2518        if (!IS_ERR(new_ns)) {
2519                struct mount *mnt = real_mount(m);
2520                mnt->mnt_ns = new_ns;
2521                new_ns->root = mnt;
2522                list_add(&mnt->mnt_list, &new_ns->list);
2523        } else {
2524                mntput(m);
2525        }
2526        return new_ns;
2527}
2528
2529struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2530{
2531        struct mnt_namespace *ns;
2532        struct super_block *s;
2533        struct path path;
2534        int err;
2535
2536        ns = create_mnt_ns(mnt);
2537        if (IS_ERR(ns))
2538                return ERR_CAST(ns);
2539
2540        err = vfs_path_lookup(mnt->mnt_root, mnt,
2541                        name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2542
2543        put_mnt_ns(ns);
2544
2545        if (err)
2546                return ERR_PTR(err);
2547
2548        /* trade a vfsmount reference for active sb one */
2549        s = path.mnt->mnt_sb;
2550        atomic_inc(&s->s_active);
2551        mntput(path.mnt);
2552        /* lock the sucker */
2553        down_write(&s->s_umount);
2554        /* ... and return the root of (sub)tree on it */
2555        return path.dentry;
2556}
2557EXPORT_SYMBOL(mount_subtree);
2558
2559SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2560                char __user *, type, unsigned long, flags, void __user *, data)
2561{
2562        int ret;
2563        char *kernel_type;
2564        struct filename *kernel_dir;
2565        char *kernel_dev;
2566        unsigned long data_page;
2567
2568        ret = copy_mount_string(type, &kernel_type);
2569        if (ret < 0)
2570                goto out_type;
2571
2572        kernel_dir = getname(dir_name);
2573        if (IS_ERR(kernel_dir)) {
2574                ret = PTR_ERR(kernel_dir);
2575                goto out_dir;
2576        }
2577
2578        ret = copy_mount_string(dev_name, &kernel_dev);
2579        if (ret < 0)
2580                goto out_dev;
2581
2582        ret = copy_mount_options(data, &data_page);
2583        if (ret < 0)
2584                goto out_data;
2585
2586        ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags,
2587                (void *) data_page);
2588
2589        free_page(data_page);
2590out_data:
2591        kfree(kernel_dev);
2592out_dev:
2593        putname(kernel_dir);
2594out_dir:
2595        kfree(kernel_type);
2596out_type:
2597        return ret;
2598}
2599
2600/*
2601 * Return true if path is reachable from root
2602 *
2603 * namespace_sem or mount_lock is held
2604 */
2605bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2606                         const struct path *root)
2607{
2608        while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2609                dentry = mnt->mnt_mountpoint;
2610                mnt = mnt->mnt_parent;
2611        }
2612        return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2613}
2614
2615int path_is_under(struct path *path1, struct path *path2)
2616{
2617        int res;
2618        read_seqlock_excl(&mount_lock);
2619        res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2620        read_sequnlock_excl(&mount_lock);
2621        return res;
2622}
2623EXPORT_SYMBOL(path_is_under);
2624
2625/*
2626 * pivot_root Semantics:
2627 * Moves the root file system of the current process to the directory put_old,
2628 * makes new_root as the new root file system of the current process, and sets
2629 * root/cwd of all processes which had them on the current root to new_root.
2630 *
2631 * Restrictions:
2632 * The new_root and put_old must be directories, and  must not be on the
2633 * same file  system as the current process root. The put_old  must  be
2634 * underneath new_root,  i.e. adding a non-zero number of /.. to the string
2635 * pointed to by put_old must yield the same directory as new_root. No other
2636 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2637 *
2638 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2639 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2640 * in this situation.
2641 *
2642 * Notes:
2643 *  - we don't move root/cwd if they are not at the root (reason: if something
2644 *    cared enough to change them, it's probably wrong to force them elsewhere)
2645 *  - it's okay to pick a root that isn't the root of a file system, e.g.
2646 *    /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2647 *    though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2648 *    first.
2649 */
2650SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2651                const char __user *, put_old)
2652{
2653        struct path new, old, parent_path, root_parent, root;
2654        struct mount *new_mnt, *root_mnt, *old_mnt;
2655        struct mountpoint *old_mp, *root_mp;
2656        int error;
2657
2658        if (!may_mount())
2659                return -EPERM;
2660
2661        error = user_path_dir(new_root, &new);
2662        if (error)
2663                goto out0;
2664
2665        error = user_path_dir(put_old, &old);
2666        if (error)
2667                goto out1;
2668
2669        error = security_sb_pivotroot(&old, &new);
2670        if (error)
2671                goto out2;
2672
2673        get_fs_root(current->fs, &root);
2674        old_mp = lock_mount(&old);
2675        error = PTR_ERR(old_mp);
2676        if (IS_ERR(old_mp))
2677                goto out3;
2678
2679        error = -EINVAL;
2680        new_mnt = real_mount(new.mnt);
2681        root_mnt = real_mount(root.mnt);
2682        old_mnt = real_mount(old.mnt);
2683        if (IS_MNT_SHARED(old_mnt) ||
2684                IS_MNT_SHARED(new_mnt->mnt_parent) ||
2685                IS_MNT_SHARED(root_mnt->mnt_parent))
2686                goto out4;
2687        if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
2688                goto out4;
2689        if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
2690                goto out4;
2691        error = -ENOENT;
2692        if (d_unlinked(new.dentry))
2693                goto out4;
2694        error = -EBUSY;
2695        if (new_mnt == root_mnt || old_mnt == root_mnt)
2696                goto out4; /* loop, on the same file system  */
2697        error = -EINVAL;
2698        if (root.mnt->mnt_root != root.dentry)
2699                goto out4; /* not a mountpoint */
2700        if (!mnt_has_parent(root_mnt))
2701                goto out4; /* not attached */
2702        root_mp = root_mnt->mnt_mp;
2703        if (new.mnt->mnt_root != new.dentry)
2704                goto out4; /* not a mountpoint */
2705        if (!mnt_has_parent(new_mnt))
2706                goto out4; /* not attached */
2707        /* make sure we can reach put_old from new_root */
2708        if (!is_path_reachable(old_mnt, old.dentry, &new))
2709                goto out4;
2710        root_mp->m_count++; /* pin it so it won't go away */
2711        lock_mount_hash();
2712        detach_mnt(new_mnt, &parent_path);
2713        detach_mnt(root_mnt, &root_parent);
2714        if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
2715                new_mnt->mnt.mnt_flags |= MNT_LOCKED;
2716                root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2717        }
2718        /* mount old root on put_old */
2719        attach_mnt(root_mnt, old_mnt, old_mp);
2720        /* mount new_root on / */
2721        attach_mnt(new_mnt, real_mount(root_parent.mnt), root_mp);
2722        touch_mnt_namespace(current->nsproxy->mnt_ns);
2723        unlock_mount_hash();
2724        chroot_fs_refs(&root, &new);
2725        put_mountpoint(root_mp);
2726        error = 0;
2727out4:
2728        unlock_mount(old_mp);
2729        if (!error) {
2730                path_put(&root_parent);
2731                path_put(&parent_path);
2732        }
2733out3:
2734        path_put(&root);
2735out2:
2736        path_put(&old);
2737out1:
2738        path_put(&new);
2739out0:
2740        return error;
2741}
2742
2743static void __init init_mount_tree(void)
2744{
2745        struct vfsmount *mnt;
2746        struct mnt_namespace *ns;
2747        struct path root;
2748        struct file_system_type *type;
2749
2750        type = get_fs_type("rootfs");
2751        if (!type)
2752                panic("Can't find rootfs type");
2753        mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
2754        put_filesystem(type);
2755        if (IS_ERR(mnt))
2756                panic("Can't create rootfs");
2757
2758        ns = create_mnt_ns(mnt);
2759        if (IS_ERR(ns))
2760                panic("Can't allocate initial namespace");
2761
2762        init_task.nsproxy->mnt_ns = ns;
2763        get_mnt_ns(ns);
2764
2765        root.mnt = mnt;
2766        root.dentry = mnt->mnt_root;
2767
2768        set_fs_pwd(current->fs, &root);
2769        set_fs_root(current->fs, &root);
2770}
2771
2772void __init mnt_init(void)
2773{
2774        unsigned u;
2775        int err;
2776
2777        mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
2778                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2779
2780        mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2781        mountpoint_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2782
2783        if (!mount_hashtable || !mountpoint_hashtable)
2784                panic("Failed to allocate mount hash table\n");
2785
2786        printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
2787
2788        for (u = 0; u < HASH_SIZE; u++)
2789                INIT_LIST_HEAD(&mount_hashtable[u]);
2790        for (u = 0; u < HASH_SIZE; u++)
2791                INIT_LIST_HEAD(&mountpoint_hashtable[u]);
2792
2793        err = sysfs_init();
2794        if (err)
2795                printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2796                        __func__, err);
2797        fs_kobj = kobject_create_and_add("fs", NULL);
2798        if (!fs_kobj)
2799                printk(KERN_WARNING "%s: kobj create error\n", __func__);
2800        init_rootfs();
2801        init_mount_tree();
2802}
2803
2804void put_mnt_ns(struct mnt_namespace *ns)
2805{
2806        if (!atomic_dec_and_test(&ns->count))
2807                return;
2808        drop_collected_mounts(&ns->root->mnt);
2809        free_mnt_ns(ns);
2810}
2811
2812struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2813{
2814        struct vfsmount *mnt;
2815        mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2816        if (!IS_ERR(mnt)) {
2817                /*
2818                 * it is a longterm mount, don't release mnt until
2819                 * we unmount before file sys is unregistered
2820                */
2821                real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
2822        }
2823        return mnt;
2824}
2825EXPORT_SYMBOL_GPL(kern_mount_data);
2826
2827void kern_unmount(struct vfsmount *mnt)
2828{
2829        /* release long term mount so mount point can be released */
2830        if (!IS_ERR_OR_NULL(mnt)) {
2831                real_mount(mnt)->mnt_ns = NULL;
2832                synchronize_rcu();      /* yecchhh... */
2833                mntput(mnt);
2834        }
2835}
2836EXPORT_SYMBOL(kern_unmount);
2837
2838bool our_mnt(struct vfsmount *mnt)
2839{
2840        return check_mnt(real_mount(mnt));
2841}
2842
2843bool current_chrooted(void)
2844{
2845        /* Does the current process have a non-standard root */
2846        struct path ns_root;
2847        struct path fs_root;
2848        bool chrooted;
2849
2850        /* Find the namespace root */
2851        ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
2852        ns_root.dentry = ns_root.mnt->mnt_root;
2853        path_get(&ns_root);
2854        while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
2855                ;
2856
2857        get_fs_root(current->fs, &fs_root);
2858
2859        chrooted = !path_equal(&fs_root, &ns_root);
2860
2861        path_put(&fs_root);
2862        path_put(&ns_root);
2863
2864        return chrooted;
2865}
2866
2867bool fs_fully_visible(struct file_system_type *type)
2868{
2869        struct mnt_namespace *ns = current->nsproxy->mnt_ns;
2870        struct mount *mnt;
2871        bool visible = false;
2872
2873        if (unlikely(!ns))
2874                return false;
2875
2876        down_read(&namespace_sem);
2877        list_for_each_entry(mnt, &ns->list, mnt_list) {
2878                struct mount *child;
2879                if (mnt->mnt.mnt_sb->s_type != type)
2880                        continue;
2881
2882                /* This mount is not fully visible if there are any child mounts
2883                 * that cover anything except for empty directories.
2884                 */
2885                list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2886                        struct inode *inode = child->mnt_mountpoint->d_inode;
2887                        if (!S_ISDIR(inode->i_mode))
2888                                goto next;
2889                        if (inode->i_nlink > 2)
2890                                goto next;
2891                }
2892                visible = true;
2893                goto found;
2894        next:   ;
2895        }
2896found:
2897        up_read(&namespace_sem);
2898        return visible;
2899}
2900
2901static void *mntns_get(struct task_struct *task)
2902{
2903        struct mnt_namespace *ns = NULL;
2904        struct nsproxy *nsproxy;
2905
2906        rcu_read_lock();
2907        nsproxy = task_nsproxy(task);
2908        if (nsproxy) {
2909                ns = nsproxy->mnt_ns;
2910                get_mnt_ns(ns);
2911        }
2912        rcu_read_unlock();
2913
2914        return ns;
2915}
2916
2917static void mntns_put(void *ns)
2918{
2919        put_mnt_ns(ns);
2920}
2921
2922static int mntns_install(struct nsproxy *nsproxy, void *ns)
2923{
2924        struct fs_struct *fs = current->fs;
2925        struct mnt_namespace *mnt_ns = ns;
2926        struct path root;
2927
2928        if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
2929            !ns_capable(current_user_ns(), CAP_SYS_CHROOT) ||
2930            !ns_capable(current_user_ns(), CAP_SYS_ADMIN))
2931                return -EPERM;
2932
2933        if (fs->users != 1)
2934                return -EINVAL;
2935
2936        get_mnt_ns(mnt_ns);
2937        put_mnt_ns(nsproxy->mnt_ns);
2938        nsproxy->mnt_ns = mnt_ns;
2939
2940        /* Find the root */
2941        root.mnt    = &mnt_ns->root->mnt;
2942        root.dentry = mnt_ns->root->mnt.mnt_root;
2943        path_get(&root);
2944        while(d_mountpoint(root.dentry) && follow_down_one(&root))
2945                ;
2946
2947        /* Update the pwd and root */
2948        set_fs_pwd(fs, &root);
2949        set_fs_root(fs, &root);
2950
2951        path_put(&root);
2952        return 0;
2953}
2954
2955static unsigned int mntns_inum(void *ns)
2956{
2957        struct mnt_namespace *mnt_ns = ns;
2958        return mnt_ns->proc_inum;
2959}
2960
2961const struct proc_ns_operations mntns_operations = {
2962        .name           = "mnt",
2963        .type           = CLONE_NEWNS,
2964        .get            = mntns_get,
2965        .put            = mntns_put,
2966        .install        = mntns_install,
2967        .inum           = mntns_inum,
2968};
2969