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/ramfs.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_fs.h>
  25#include "pnode.h"
  26#include "internal.h"
  27
  28#define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
  29#define HASH_SIZE (1UL << HASH_SHIFT)
  30
  31static int event;
  32static DEFINE_IDA(mnt_id_ida);
  33static DEFINE_IDA(mnt_group_ida);
  34static DEFINE_SPINLOCK(mnt_id_lock);
  35static int mnt_id_start = 0;
  36static int mnt_group_start = 1;
  37
  38static struct list_head *mount_hashtable __read_mostly;
  39static struct kmem_cache *mnt_cache __read_mostly;
  40static struct rw_semaphore namespace_sem;
  41
  42/* /sys/fs */
  43struct kobject *fs_kobj;
  44EXPORT_SYMBOL_GPL(fs_kobj);
  45
  46/*
  47 * vfsmount lock may be taken for read to prevent changes to the
  48 * vfsmount hash, ie. during mountpoint lookups or walking back
  49 * up the tree.
  50 *
  51 * It should be taken for write in all cases where the vfsmount
  52 * tree or hash is modified or when a vfsmount structure is modified.
  53 */
  54DEFINE_BRLOCK(vfsmount_lock);
  55
  56static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
  57{
  58        unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
  59        tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
  60        tmp = tmp + (tmp >> HASH_SHIFT);
  61        return tmp & (HASH_SIZE - 1);
  62}
  63
  64#define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
  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->f_dentry->d_inode;
 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        br_write_lock(&vfsmount_lock);
 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        br_write_unlock(&vfsmount_lock);
 494        return ret;
 495}
 496
 497static void __mnt_unmake_readonly(struct mount *mnt)
 498{
 499        br_write_lock(&vfsmount_lock);
 500        mnt->mnt.mnt_flags &= ~MNT_READONLY;
 501        br_write_unlock(&vfsmount_lock);
 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        br_write_lock(&vfsmount_lock);
 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        br_write_unlock(&vfsmount_lock);
 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/*
 551 * find the first or last mount at @dentry on vfsmount @mnt depending on
 552 * @dir. If @dir is set return the first mount else return the last mount.
 553 * vfsmount_lock must be held for read or write.
 554 */
 555struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
 556                              int dir)
 557{
 558        struct list_head *head = mount_hashtable + hash(mnt, dentry);
 559        struct list_head *tmp = head;
 560        struct mount *p, *found = NULL;
 561
 562        for (;;) {
 563                tmp = dir ? tmp->next : tmp->prev;
 564                p = NULL;
 565                if (tmp == head)
 566                        break;
 567                p = list_entry(tmp, struct mount, mnt_hash);
 568                if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) {
 569                        found = p;
 570                        break;
 571                }
 572        }
 573        return found;
 574}
 575
 576/*
 577 * lookup_mnt - Return the first child mount mounted at path
 578 *
 579 * "First" means first mounted chronologically.  If you create the
 580 * following mounts:
 581 *
 582 * mount /dev/sda1 /mnt
 583 * mount /dev/sda2 /mnt
 584 * mount /dev/sda3 /mnt
 585 *
 586 * Then lookup_mnt() on the base /mnt dentry in the root mount will
 587 * return successively the root dentry and vfsmount of /dev/sda1, then
 588 * /dev/sda2, then /dev/sda3, then NULL.
 589 *
 590 * lookup_mnt takes a reference to the found vfsmount.
 591 */
 592struct vfsmount *lookup_mnt(struct path *path)
 593{
 594        struct mount *child_mnt;
 595
 596        br_read_lock(&vfsmount_lock);
 597        child_mnt = __lookup_mnt(path->mnt, path->dentry, 1);
 598        if (child_mnt) {
 599                mnt_add_count(child_mnt, 1);
 600                br_read_unlock(&vfsmount_lock);
 601                return &child_mnt->mnt;
 602        } else {
 603                br_read_unlock(&vfsmount_lock);
 604                return NULL;
 605        }
 606}
 607
 608static inline int check_mnt(struct mount *mnt)
 609{
 610        return mnt->mnt_ns == current->nsproxy->mnt_ns;
 611}
 612
 613/*
 614 * vfsmount lock must be held for write
 615 */
 616static void touch_mnt_namespace(struct mnt_namespace *ns)
 617{
 618        if (ns) {
 619                ns->event = ++event;
 620                wake_up_interruptible(&ns->poll);
 621        }
 622}
 623
 624/*
 625 * vfsmount lock must be held for write
 626 */
 627static void __touch_mnt_namespace(struct mnt_namespace *ns)
 628{
 629        if (ns && ns->event != event) {
 630                ns->event = event;
 631                wake_up_interruptible(&ns->poll);
 632        }
 633}
 634
 635/*
 636 * Clear dentry's mounted state if it has no remaining mounts.
 637 * vfsmount_lock must be held for write.
 638 */
 639static void dentry_reset_mounted(struct dentry *dentry)
 640{
 641        unsigned u;
 642
 643        for (u = 0; u < HASH_SIZE; u++) {
 644                struct mount *p;
 645
 646                list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
 647                        if (p->mnt_mountpoint == dentry)
 648                                return;
 649                }
 650        }
 651        spin_lock(&dentry->d_lock);
 652        dentry->d_flags &= ~DCACHE_MOUNTED;
 653        spin_unlock(&dentry->d_lock);
 654}
 655
 656/*
 657 * vfsmount lock must be held for write
 658 */
 659static void detach_mnt(struct mount *mnt, struct path *old_path)
 660{
 661        old_path->dentry = mnt->mnt_mountpoint;
 662        old_path->mnt = &mnt->mnt_parent->mnt;
 663        mnt->mnt_parent = mnt;
 664        mnt->mnt_mountpoint = mnt->mnt.mnt_root;
 665        list_del_init(&mnt->mnt_child);
 666        list_del_init(&mnt->mnt_hash);
 667        dentry_reset_mounted(old_path->dentry);
 668}
 669
 670/*
 671 * vfsmount lock must be held for write
 672 */
 673void mnt_set_mountpoint(struct mount *mnt, struct dentry *dentry,
 674                        struct mount *child_mnt)
 675{
 676        mnt_add_count(mnt, 1);  /* essentially, that's mntget */
 677        child_mnt->mnt_mountpoint = dget(dentry);
 678        child_mnt->mnt_parent = mnt;
 679        spin_lock(&dentry->d_lock);
 680        dentry->d_flags |= DCACHE_MOUNTED;
 681        spin_unlock(&dentry->d_lock);
 682}
 683
 684/*
 685 * vfsmount lock must be held for write
 686 */
 687static void attach_mnt(struct mount *mnt, struct path *path)
 688{
 689        mnt_set_mountpoint(real_mount(path->mnt), path->dentry, mnt);
 690        list_add_tail(&mnt->mnt_hash, mount_hashtable +
 691                        hash(path->mnt, path->dentry));
 692        list_add_tail(&mnt->mnt_child, &real_mount(path->mnt)->mnt_mounts);
 693}
 694
 695/*
 696 * vfsmount lock must be held for write
 697 */
 698static void commit_tree(struct mount *mnt)
 699{
 700        struct mount *parent = mnt->mnt_parent;
 701        struct mount *m;
 702        LIST_HEAD(head);
 703        struct mnt_namespace *n = parent->mnt_ns;
 704
 705        BUG_ON(parent == mnt);
 706
 707        list_add_tail(&head, &mnt->mnt_list);
 708        list_for_each_entry(m, &head, mnt_list)
 709                m->mnt_ns = n;
 710
 711        list_splice(&head, n->list.prev);
 712
 713        list_add_tail(&mnt->mnt_hash, mount_hashtable +
 714                                hash(&parent->mnt, mnt->mnt_mountpoint));
 715        list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
 716        touch_mnt_namespace(n);
 717}
 718
 719static struct mount *next_mnt(struct mount *p, struct mount *root)
 720{
 721        struct list_head *next = p->mnt_mounts.next;
 722        if (next == &p->mnt_mounts) {
 723                while (1) {
 724                        if (p == root)
 725                                return NULL;
 726                        next = p->mnt_child.next;
 727                        if (next != &p->mnt_parent->mnt_mounts)
 728                                break;
 729                        p = p->mnt_parent;
 730                }
 731        }
 732        return list_entry(next, struct mount, mnt_child);
 733}
 734
 735static struct mount *skip_mnt_tree(struct mount *p)
 736{
 737        struct list_head *prev = p->mnt_mounts.prev;
 738        while (prev != &p->mnt_mounts) {
 739                p = list_entry(prev, struct mount, mnt_child);
 740                prev = p->mnt_mounts.prev;
 741        }
 742        return p;
 743}
 744
 745struct vfsmount *
 746vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
 747{
 748        struct mount *mnt;
 749        struct dentry *root;
 750
 751        if (!type)
 752                return ERR_PTR(-ENODEV);
 753
 754        mnt = alloc_vfsmnt(name);
 755        if (!mnt)
 756                return ERR_PTR(-ENOMEM);
 757
 758        if (flags & MS_KERNMOUNT)
 759                mnt->mnt.mnt_flags = MNT_INTERNAL;
 760
 761        root = mount_fs(type, flags, name, data);
 762        if (IS_ERR(root)) {
 763                free_vfsmnt(mnt);
 764                return ERR_CAST(root);
 765        }
 766
 767        mnt->mnt.mnt_root = root;
 768        mnt->mnt.mnt_sb = root->d_sb;
 769        mnt->mnt_mountpoint = mnt->mnt.mnt_root;
 770        mnt->mnt_parent = mnt;
 771        br_write_lock(&vfsmount_lock);
 772        list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
 773        br_write_unlock(&vfsmount_lock);
 774        return &mnt->mnt;
 775}
 776EXPORT_SYMBOL_GPL(vfs_kern_mount);
 777
 778static struct mount *clone_mnt(struct mount *old, struct dentry *root,
 779                                        int flag)
 780{
 781        struct super_block *sb = old->mnt.mnt_sb;
 782        struct mount *mnt;
 783        int err;
 784
 785        mnt = alloc_vfsmnt(old->mnt_devname);
 786        if (!mnt)
 787                return ERR_PTR(-ENOMEM);
 788
 789        if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
 790                mnt->mnt_group_id = 0; /* not a peer of original */
 791        else
 792                mnt->mnt_group_id = old->mnt_group_id;
 793
 794        if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
 795                err = mnt_alloc_group_id(mnt);
 796                if (err)
 797                        goto out_free;
 798        }
 799
 800        mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
 801        atomic_inc(&sb->s_active);
 802        mnt->mnt.mnt_sb = sb;
 803        mnt->mnt.mnt_root = dget(root);
 804        mnt->mnt_mountpoint = mnt->mnt.mnt_root;
 805        mnt->mnt_parent = mnt;
 806        br_write_lock(&vfsmount_lock);
 807        list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
 808        br_write_unlock(&vfsmount_lock);
 809
 810        if ((flag & CL_SLAVE) ||
 811            ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
 812                list_add(&mnt->mnt_slave, &old->mnt_slave_list);
 813                mnt->mnt_master = old;
 814                CLEAR_MNT_SHARED(mnt);
 815        } else if (!(flag & CL_PRIVATE)) {
 816                if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
 817                        list_add(&mnt->mnt_share, &old->mnt_share);
 818                if (IS_MNT_SLAVE(old))
 819                        list_add(&mnt->mnt_slave, &old->mnt_slave);
 820                mnt->mnt_master = old->mnt_master;
 821        }
 822        if (flag & CL_MAKE_SHARED)
 823                set_mnt_shared(mnt);
 824
 825        /* stick the duplicate mount on the same expiry list
 826         * as the original if that was on one */
 827        if (flag & CL_EXPIRE) {
 828                if (!list_empty(&old->mnt_expire))
 829                        list_add(&mnt->mnt_expire, &old->mnt_expire);
 830        }
 831
 832        return mnt;
 833
 834 out_free:
 835        free_vfsmnt(mnt);
 836        return ERR_PTR(err);
 837}
 838
 839static inline void mntfree(struct mount *mnt)
 840{
 841        struct vfsmount *m = &mnt->mnt;
 842        struct super_block *sb = m->mnt_sb;
 843
 844        /*
 845         * This probably indicates that somebody messed
 846         * up a mnt_want/drop_write() pair.  If this
 847         * happens, the filesystem was probably unable
 848         * to make r/w->r/o transitions.
 849         */
 850        /*
 851         * The locking used to deal with mnt_count decrement provides barriers,
 852         * so mnt_get_writers() below is safe.
 853         */
 854        WARN_ON(mnt_get_writers(mnt));
 855        fsnotify_vfsmount_delete(m);
 856        dput(m->mnt_root);
 857        free_vfsmnt(mnt);
 858        deactivate_super(sb);
 859}
 860
 861static void mntput_no_expire(struct mount *mnt)
 862{
 863put_again:
 864#ifdef CONFIG_SMP
 865        br_read_lock(&vfsmount_lock);
 866        if (likely(mnt->mnt_ns)) {
 867                /* shouldn't be the last one */
 868                mnt_add_count(mnt, -1);
 869                br_read_unlock(&vfsmount_lock);
 870                return;
 871        }
 872        br_read_unlock(&vfsmount_lock);
 873
 874        br_write_lock(&vfsmount_lock);
 875        mnt_add_count(mnt, -1);
 876        if (mnt_get_count(mnt)) {
 877                br_write_unlock(&vfsmount_lock);
 878                return;
 879        }
 880#else
 881        mnt_add_count(mnt, -1);
 882        if (likely(mnt_get_count(mnt)))
 883                return;
 884        br_write_lock(&vfsmount_lock);
 885#endif
 886        if (unlikely(mnt->mnt_pinned)) {
 887                mnt_add_count(mnt, mnt->mnt_pinned + 1);
 888                mnt->mnt_pinned = 0;
 889                br_write_unlock(&vfsmount_lock);
 890                acct_auto_close_mnt(&mnt->mnt);
 891                goto put_again;
 892        }
 893
 894        list_del(&mnt->mnt_instance);
 895        br_write_unlock(&vfsmount_lock);
 896        mntfree(mnt);
 897}
 898
 899void mntput(struct vfsmount *mnt)
 900{
 901        if (mnt) {
 902                struct mount *m = real_mount(mnt);
 903                /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
 904                if (unlikely(m->mnt_expiry_mark))
 905                        m->mnt_expiry_mark = 0;
 906                mntput_no_expire(m);
 907        }
 908}
 909EXPORT_SYMBOL(mntput);
 910
 911struct vfsmount *mntget(struct vfsmount *mnt)
 912{
 913        if (mnt)
 914                mnt_add_count(real_mount(mnt), 1);
 915        return mnt;
 916}
 917EXPORT_SYMBOL(mntget);
 918
 919void mnt_pin(struct vfsmount *mnt)
 920{
 921        br_write_lock(&vfsmount_lock);
 922        real_mount(mnt)->mnt_pinned++;
 923        br_write_unlock(&vfsmount_lock);
 924}
 925EXPORT_SYMBOL(mnt_pin);
 926
 927void mnt_unpin(struct vfsmount *m)
 928{
 929        struct mount *mnt = real_mount(m);
 930        br_write_lock(&vfsmount_lock);
 931        if (mnt->mnt_pinned) {
 932                mnt_add_count(mnt, 1);
 933                mnt->mnt_pinned--;
 934        }
 935        br_write_unlock(&vfsmount_lock);
 936}
 937EXPORT_SYMBOL(mnt_unpin);
 938
 939static inline void mangle(struct seq_file *m, const char *s)
 940{
 941        seq_escape(m, s, " \t\n\\");
 942}
 943
 944/*
 945 * Simple .show_options callback for filesystems which don't want to
 946 * implement more complex mount option showing.
 947 *
 948 * See also save_mount_options().
 949 */
 950int generic_show_options(struct seq_file *m, struct dentry *root)
 951{
 952        const char *options;
 953
 954        rcu_read_lock();
 955        options = rcu_dereference(root->d_sb->s_options);
 956
 957        if (options != NULL && options[0]) {
 958                seq_putc(m, ',');
 959                mangle(m, options);
 960        }
 961        rcu_read_unlock();
 962
 963        return 0;
 964}
 965EXPORT_SYMBOL(generic_show_options);
 966
 967/*
 968 * If filesystem uses generic_show_options(), this function should be
 969 * called from the fill_super() callback.
 970 *
 971 * The .remount_fs callback usually needs to be handled in a special
 972 * way, to make sure, that previous options are not overwritten if the
 973 * remount fails.
 974 *
 975 * Also note, that if the filesystem's .remount_fs function doesn't
 976 * reset all options to their default value, but changes only newly
 977 * given options, then the displayed options will not reflect reality
 978 * any more.
 979 */
 980void save_mount_options(struct super_block *sb, char *options)
 981{
 982        BUG_ON(sb->s_options);
 983        rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
 984}
 985EXPORT_SYMBOL(save_mount_options);
 986
 987void replace_mount_options(struct super_block *sb, char *options)
 988{
 989        char *old = sb->s_options;
 990        rcu_assign_pointer(sb->s_options, options);
 991        if (old) {
 992                synchronize_rcu();
 993                kfree(old);
 994        }
 995}
 996EXPORT_SYMBOL(replace_mount_options);
 997
 998#ifdef CONFIG_PROC_FS
 999/* iterator; we want it to have access to namespace_sem, thus here... */
1000static void *m_start(struct seq_file *m, loff_t *pos)
1001{
1002        struct proc_mounts *p = proc_mounts(m);
1003
1004        down_read(&namespace_sem);
1005        return seq_list_start(&p->ns->list, *pos);
1006}
1007
1008static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1009{
1010        struct proc_mounts *p = proc_mounts(m);
1011
1012        return seq_list_next(v, &p->ns->list, pos);
1013}
1014
1015static void m_stop(struct seq_file *m, void *v)
1016{
1017        up_read(&namespace_sem);
1018}
1019
1020static int m_show(struct seq_file *m, void *v)
1021{
1022        struct proc_mounts *p = proc_mounts(m);
1023        struct mount *r = list_entry(v, struct mount, mnt_list);
1024        return p->show(m, &r->mnt);
1025}
1026
1027const struct seq_operations mounts_op = {
1028        .start  = m_start,
1029        .next   = m_next,
1030        .stop   = m_stop,
1031        .show   = m_show,
1032};
1033#endif  /* CONFIG_PROC_FS */
1034
1035/**
1036 * may_umount_tree - check if a mount tree is busy
1037 * @mnt: root of mount tree
1038 *
1039 * This is called to check if a tree of mounts has any
1040 * open files, pwds, chroots or sub mounts that are
1041 * busy.
1042 */
1043int may_umount_tree(struct vfsmount *m)
1044{
1045        struct mount *mnt = real_mount(m);
1046        int actual_refs = 0;
1047        int minimum_refs = 0;
1048        struct mount *p;
1049        BUG_ON(!m);
1050
1051        /* write lock needed for mnt_get_count */
1052        br_write_lock(&vfsmount_lock);
1053        for (p = mnt; p; p = next_mnt(p, mnt)) {
1054                actual_refs += mnt_get_count(p);
1055                minimum_refs += 2;
1056        }
1057        br_write_unlock(&vfsmount_lock);
1058
1059        if (actual_refs > minimum_refs)
1060                return 0;
1061
1062        return 1;
1063}
1064
1065EXPORT_SYMBOL(may_umount_tree);
1066
1067/**
1068 * may_umount - check if a mount point is busy
1069 * @mnt: root of mount
1070 *
1071 * This is called to check if a mount point has any
1072 * open files, pwds, chroots or sub mounts. If the
1073 * mount has sub mounts this will return busy
1074 * regardless of whether the sub mounts are busy.
1075 *
1076 * Doesn't take quota and stuff into account. IOW, in some cases it will
1077 * give false negatives. The main reason why it's here is that we need
1078 * a non-destructive way to look for easily umountable filesystems.
1079 */
1080int may_umount(struct vfsmount *mnt)
1081{
1082        int ret = 1;
1083        down_read(&namespace_sem);
1084        br_write_lock(&vfsmount_lock);
1085        if (propagate_mount_busy(real_mount(mnt), 2))
1086                ret = 0;
1087        br_write_unlock(&vfsmount_lock);
1088        up_read(&namespace_sem);
1089        return ret;
1090}
1091
1092EXPORT_SYMBOL(may_umount);
1093
1094void release_mounts(struct list_head *head)
1095{
1096        struct mount *mnt;
1097        while (!list_empty(head)) {
1098                mnt = list_first_entry(head, struct mount, mnt_hash);
1099                list_del_init(&mnt->mnt_hash);
1100                if (mnt_has_parent(mnt)) {
1101                        struct dentry *dentry;
1102                        struct mount *m;
1103
1104                        br_write_lock(&vfsmount_lock);
1105                        dentry = mnt->mnt_mountpoint;
1106                        m = mnt->mnt_parent;
1107                        mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1108                        mnt->mnt_parent = mnt;
1109                        m->mnt_ghosts--;
1110                        br_write_unlock(&vfsmount_lock);
1111                        dput(dentry);
1112                        mntput(&m->mnt);
1113                }
1114                mntput(&mnt->mnt);
1115        }
1116}
1117
1118/*
1119 * vfsmount lock must be held for write
1120 * namespace_sem must be held for write
1121 */
1122void umount_tree(struct mount *mnt, int propagate, struct list_head *kill)
1123{
1124        LIST_HEAD(tmp_list);
1125        struct mount *p;
1126
1127        for (p = mnt; p; p = next_mnt(p, mnt))
1128                list_move(&p->mnt_hash, &tmp_list);
1129
1130        if (propagate)
1131                propagate_umount(&tmp_list);
1132
1133        list_for_each_entry(p, &tmp_list, mnt_hash) {
1134                list_del_init(&p->mnt_expire);
1135                list_del_init(&p->mnt_list);
1136                __touch_mnt_namespace(p->mnt_ns);
1137                p->mnt_ns = NULL;
1138                list_del_init(&p->mnt_child);
1139                if (mnt_has_parent(p)) {
1140                        p->mnt_parent->mnt_ghosts++;
1141                        dentry_reset_mounted(p->mnt_mountpoint);
1142                }
1143                change_mnt_propagation(p, MS_PRIVATE);
1144        }
1145        list_splice(&tmp_list, kill);
1146}
1147
1148static void shrink_submounts(struct mount *mnt, struct list_head *umounts);
1149
1150static int do_umount(struct mount *mnt, int flags)
1151{
1152        struct super_block *sb = mnt->mnt.mnt_sb;
1153        int retval;
1154        LIST_HEAD(umount_list);
1155
1156        retval = security_sb_umount(&mnt->mnt, flags);
1157        if (retval)
1158                return retval;
1159
1160        /*
1161         * Allow userspace to request a mountpoint be expired rather than
1162         * unmounting unconditionally. Unmount only happens if:
1163         *  (1) the mark is already set (the mark is cleared by mntput())
1164         *  (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1165         */
1166        if (flags & MNT_EXPIRE) {
1167                if (&mnt->mnt == current->fs->root.mnt ||
1168                    flags & (MNT_FORCE | MNT_DETACH))
1169                        return -EINVAL;
1170
1171                /*
1172                 * probably don't strictly need the lock here if we examined
1173                 * all race cases, but it's a slowpath.
1174                 */
1175                br_write_lock(&vfsmount_lock);
1176                if (mnt_get_count(mnt) != 2) {
1177                        br_write_unlock(&vfsmount_lock);
1178                        return -EBUSY;
1179                }
1180                br_write_unlock(&vfsmount_lock);
1181
1182                if (!xchg(&mnt->mnt_expiry_mark, 1))
1183                        return -EAGAIN;
1184        }
1185
1186        /*
1187         * If we may have to abort operations to get out of this
1188         * mount, and they will themselves hold resources we must
1189         * allow the fs to do things. In the Unix tradition of
1190         * 'Gee thats tricky lets do it in userspace' the umount_begin
1191         * might fail to complete on the first run through as other tasks
1192         * must return, and the like. Thats for the mount program to worry
1193         * about for the moment.
1194         */
1195
1196        if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1197                sb->s_op->umount_begin(sb);
1198        }
1199
1200        /*
1201         * No sense to grab the lock for this test, but test itself looks
1202         * somewhat bogus. Suggestions for better replacement?
1203         * Ho-hum... In principle, we might treat that as umount + switch
1204         * to rootfs. GC would eventually take care of the old vfsmount.
1205         * Actually it makes sense, especially if rootfs would contain a
1206         * /reboot - static binary that would close all descriptors and
1207         * call reboot(9). Then init(8) could umount root and exec /reboot.
1208         */
1209        if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1210                /*
1211                 * Special case for "unmounting" root ...
1212                 * we just try to remount it readonly.
1213                 */
1214                down_write(&sb->s_umount);
1215                if (!(sb->s_flags & MS_RDONLY))
1216                        retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1217                up_write(&sb->s_umount);
1218                return retval;
1219        }
1220
1221        down_write(&namespace_sem);
1222        br_write_lock(&vfsmount_lock);
1223        event++;
1224
1225        if (!(flags & MNT_DETACH))
1226                shrink_submounts(mnt, &umount_list);
1227
1228        retval = -EBUSY;
1229        if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1230                if (!list_empty(&mnt->mnt_list))
1231                        umount_tree(mnt, 1, &umount_list);
1232                retval = 0;
1233        }
1234        br_write_unlock(&vfsmount_lock);
1235        up_write(&namespace_sem);
1236        release_mounts(&umount_list);
1237        return retval;
1238}
1239
1240/*
1241 * Now umount can handle mount points as well as block devices.
1242 * This is important for filesystems which use unnamed block devices.
1243 *
1244 * We now support a flag for forced unmount like the other 'big iron'
1245 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1246 */
1247
1248SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1249{
1250        struct path path;
1251        struct mount *mnt;
1252        int retval;
1253        int lookup_flags = 0;
1254
1255        if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1256                return -EINVAL;
1257
1258        if (!(flags & UMOUNT_NOFOLLOW))
1259                lookup_flags |= LOOKUP_FOLLOW;
1260
1261        retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1262        if (retval)
1263                goto out;
1264        mnt = real_mount(path.mnt);
1265        retval = -EINVAL;
1266        if (path.dentry != path.mnt->mnt_root)
1267                goto dput_and_out;
1268        if (!check_mnt(mnt))
1269                goto dput_and_out;
1270
1271        retval = -EPERM;
1272        if (!ns_capable(mnt->mnt_ns->user_ns, CAP_SYS_ADMIN))
1273                goto dput_and_out;
1274
1275        retval = do_umount(mnt, flags);
1276dput_and_out:
1277        /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1278        dput(path.dentry);
1279        mntput_no_expire(mnt);
1280out:
1281        return retval;
1282}
1283
1284#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1285
1286/*
1287 *      The 2.0 compatible umount. No flags.
1288 */
1289SYSCALL_DEFINE1(oldumount, char __user *, name)
1290{
1291        return sys_umount(name, 0);
1292}
1293
1294#endif
1295
1296static int mount_is_safe(struct path *path)
1297{
1298        if (ns_capable(real_mount(path->mnt)->mnt_ns->user_ns, CAP_SYS_ADMIN))
1299                return 0;
1300        return -EPERM;
1301#ifdef notyet
1302        if (S_ISLNK(path->dentry->d_inode->i_mode))
1303                return -EPERM;
1304        if (path->dentry->d_inode->i_mode & S_ISVTX) {
1305                if (current_uid() != path->dentry->d_inode->i_uid)
1306                        return -EPERM;
1307        }
1308        if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1309                return -EPERM;
1310        return 0;
1311#endif
1312}
1313
1314static bool mnt_ns_loop(struct path *path)
1315{
1316        /* Could bind mounting the mount namespace inode cause a
1317         * mount namespace loop?
1318         */
1319        struct inode *inode = path->dentry->d_inode;
1320        struct proc_inode *ei;
1321        struct mnt_namespace *mnt_ns;
1322
1323        if (!proc_ns_inode(inode))
1324                return false;
1325
1326        ei = PROC_I(inode);
1327        if (ei->ns_ops != &mntns_operations)
1328                return false;
1329
1330        mnt_ns = ei->ns;
1331        return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1332}
1333
1334struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1335                                        int flag)
1336{
1337        struct mount *res, *p, *q, *r;
1338        struct path path;
1339
1340        if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1341                return ERR_PTR(-EINVAL);
1342
1343        res = q = clone_mnt(mnt, dentry, flag);
1344        if (IS_ERR(q))
1345                return q;
1346
1347        q->mnt_mountpoint = mnt->mnt_mountpoint;
1348
1349        p = mnt;
1350        list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1351                struct mount *s;
1352                if (!is_subdir(r->mnt_mountpoint, dentry))
1353                        continue;
1354
1355                for (s = r; s; s = next_mnt(s, r)) {
1356                        if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1357                                s = skip_mnt_tree(s);
1358                                continue;
1359                        }
1360                        while (p != s->mnt_parent) {
1361                                p = p->mnt_parent;
1362                                q = q->mnt_parent;
1363                        }
1364                        p = s;
1365                        path.mnt = &q->mnt;
1366                        path.dentry = p->mnt_mountpoint;
1367                        q = clone_mnt(p, p->mnt.mnt_root, flag);
1368                        if (IS_ERR(q))
1369                                goto out;
1370                        br_write_lock(&vfsmount_lock);
1371                        list_add_tail(&q->mnt_list, &res->mnt_list);
1372                        attach_mnt(q, &path);
1373                        br_write_unlock(&vfsmount_lock);
1374                }
1375        }
1376        return res;
1377out:
1378        if (res) {
1379                LIST_HEAD(umount_list);
1380                br_write_lock(&vfsmount_lock);
1381                umount_tree(res, 0, &umount_list);
1382                br_write_unlock(&vfsmount_lock);
1383                release_mounts(&umount_list);
1384        }
1385        return q;
1386}
1387
1388/* Caller should check returned pointer for errors */
1389
1390struct vfsmount *collect_mounts(struct path *path)
1391{
1392        struct mount *tree;
1393        down_write(&namespace_sem);
1394        tree = copy_tree(real_mount(path->mnt), path->dentry,
1395                         CL_COPY_ALL | CL_PRIVATE);
1396        up_write(&namespace_sem);
1397        if (IS_ERR(tree))
1398                return NULL;
1399        return &tree->mnt;
1400}
1401
1402void drop_collected_mounts(struct vfsmount *mnt)
1403{
1404        LIST_HEAD(umount_list);
1405        down_write(&namespace_sem);
1406        br_write_lock(&vfsmount_lock);
1407        umount_tree(real_mount(mnt), 0, &umount_list);
1408        br_write_unlock(&vfsmount_lock);
1409        up_write(&namespace_sem);
1410        release_mounts(&umount_list);
1411}
1412
1413int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1414                   struct vfsmount *root)
1415{
1416        struct mount *mnt;
1417        int res = f(root, arg);
1418        if (res)
1419                return res;
1420        list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1421                res = f(&mnt->mnt, arg);
1422                if (res)
1423                        return res;
1424        }
1425        return 0;
1426}
1427
1428static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1429{
1430        struct mount *p;
1431
1432        for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1433                if (p->mnt_group_id && !IS_MNT_SHARED(p))
1434                        mnt_release_group_id(p);
1435        }
1436}
1437
1438static int invent_group_ids(struct mount *mnt, bool recurse)
1439{
1440        struct mount *p;
1441
1442        for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1443                if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1444                        int err = mnt_alloc_group_id(p);
1445                        if (err) {
1446                                cleanup_group_ids(mnt, p);
1447                                return err;
1448                        }
1449                }
1450        }
1451
1452        return 0;
1453}
1454
1455/*
1456 *  @source_mnt : mount tree to be attached
1457 *  @nd         : place the mount tree @source_mnt is attached
1458 *  @parent_nd  : if non-null, detach the source_mnt from its parent and
1459 *                 store the parent mount and mountpoint dentry.
1460 *                 (done when source_mnt is moved)
1461 *
1462 *  NOTE: in the table below explains the semantics when a source mount
1463 *  of a given type is attached to a destination mount of a given type.
1464 * ---------------------------------------------------------------------------
1465 * |         BIND MOUNT OPERATION                                            |
1466 * |**************************************************************************
1467 * | source-->| shared        |       private  |       slave    | unbindable |
1468 * | dest     |               |                |                |            |
1469 * |   |      |               |                |                |            |
1470 * |   v      |               |                |                |            |
1471 * |**************************************************************************
1472 * |  shared  | shared (++)   |     shared (+) |     shared(+++)|  invalid   |
1473 * |          |               |                |                |            |
1474 * |non-shared| shared (+)    |      private   |      slave (*) |  invalid   |
1475 * ***************************************************************************
1476 * A bind operation clones the source mount and mounts the clone on the
1477 * destination mount.
1478 *
1479 * (++)  the cloned mount is propagated to all the mounts in the propagation
1480 *       tree of the destination mount and the cloned mount is added to
1481 *       the peer group of the source mount.
1482 * (+)   the cloned mount is created under the destination mount and is marked
1483 *       as shared. The cloned mount is added to the peer group of the source
1484 *       mount.
1485 * (+++) the mount is propagated to all the mounts in the propagation tree
1486 *       of the destination mount and the cloned mount is made slave
1487 *       of the same master as that of the source mount. The cloned mount
1488 *       is marked as 'shared and slave'.
1489 * (*)   the cloned mount is made a slave of the same master as that of the
1490 *       source mount.
1491 *
1492 * ---------------------------------------------------------------------------
1493 * |                    MOVE MOUNT OPERATION                                 |
1494 * |**************************************************************************
1495 * | source-->| shared        |       private  |       slave    | unbindable |
1496 * | dest     |               |                |                |            |
1497 * |   |      |               |                |                |            |
1498 * |   v      |               |                |                |            |
1499 * |**************************************************************************
1500 * |  shared  | shared (+)    |     shared (+) |    shared(+++) |  invalid   |
1501 * |          |               |                |                |            |
1502 * |non-shared| shared (+*)   |      private   |    slave (*)   | unbindable |
1503 * ***************************************************************************
1504 *
1505 * (+)  the mount is moved to the destination. And is then propagated to
1506 *      all the mounts in the propagation tree of the destination mount.
1507 * (+*)  the mount is moved to the destination.
1508 * (+++)  the mount is moved to the destination and is then propagated to
1509 *      all the mounts belonging to the destination mount's propagation tree.
1510 *      the mount is marked as 'shared and slave'.
1511 * (*)  the mount continues to be a slave at the new location.
1512 *
1513 * if the source mount is a tree, the operations explained above is
1514 * applied to each mount in the tree.
1515 * Must be called without spinlocks held, since this function can sleep
1516 * in allocations.
1517 */
1518static int attach_recursive_mnt(struct mount *source_mnt,
1519                        struct path *path, struct path *parent_path)
1520{
1521        LIST_HEAD(tree_list);
1522        struct mount *dest_mnt = real_mount(path->mnt);
1523        struct dentry *dest_dentry = path->dentry;
1524        struct mount *child, *p;
1525        int err;
1526
1527        if (IS_MNT_SHARED(dest_mnt)) {
1528                err = invent_group_ids(source_mnt, true);
1529                if (err)
1530                        goto out;
1531        }
1532        err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1533        if (err)
1534                goto out_cleanup_ids;
1535
1536        br_write_lock(&vfsmount_lock);
1537
1538        if (IS_MNT_SHARED(dest_mnt)) {
1539                for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1540                        set_mnt_shared(p);
1541        }
1542        if (parent_path) {
1543                detach_mnt(source_mnt, parent_path);
1544                attach_mnt(source_mnt, path);
1545                touch_mnt_namespace(source_mnt->mnt_ns);
1546        } else {
1547                mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1548                commit_tree(source_mnt);
1549        }
1550
1551        list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1552                list_del_init(&child->mnt_hash);
1553                commit_tree(child);
1554        }
1555        br_write_unlock(&vfsmount_lock);
1556
1557        return 0;
1558
1559 out_cleanup_ids:
1560        if (IS_MNT_SHARED(dest_mnt))
1561                cleanup_group_ids(source_mnt, NULL);
1562 out:
1563        return err;
1564}
1565
1566static int lock_mount(struct path *path)
1567{
1568        struct vfsmount *mnt;
1569retry:
1570        mutex_lock(&path->dentry->d_inode->i_mutex);
1571        if (unlikely(cant_mount(path->dentry))) {
1572                mutex_unlock(&path->dentry->d_inode->i_mutex);
1573                return -ENOENT;
1574        }
1575        down_write(&namespace_sem);
1576        mnt = lookup_mnt(path);
1577        if (likely(!mnt))
1578                return 0;
1579        up_write(&namespace_sem);
1580        mutex_unlock(&path->dentry->d_inode->i_mutex);
1581        path_put(path);
1582        path->mnt = mnt;
1583        path->dentry = dget(mnt->mnt_root);
1584        goto retry;
1585}
1586
1587static void unlock_mount(struct path *path)
1588{
1589        up_write(&namespace_sem);
1590        mutex_unlock(&path->dentry->d_inode->i_mutex);
1591}
1592
1593static int graft_tree(struct mount *mnt, struct path *path)
1594{
1595        if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1596                return -EINVAL;
1597
1598        if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1599              S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
1600                return -ENOTDIR;
1601
1602        if (d_unlinked(path->dentry))
1603                return -ENOENT;
1604
1605        return attach_recursive_mnt(mnt, path, NULL);
1606}
1607
1608/*
1609 * Sanity check the flags to change_mnt_propagation.
1610 */
1611
1612static int flags_to_propagation_type(int flags)
1613{
1614        int type = flags & ~(MS_REC | MS_SILENT);
1615
1616        /* Fail if any non-propagation flags are set */
1617        if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1618                return 0;
1619        /* Only one propagation flag should be set */
1620        if (!is_power_of_2(type))
1621                return 0;
1622        return type;
1623}
1624
1625/*
1626 * recursively change the type of the mountpoint.
1627 */
1628static int do_change_type(struct path *path, int flag)
1629{
1630        struct mount *m;
1631        struct mount *mnt = real_mount(path->mnt);
1632        int recurse = flag & MS_REC;
1633        int type;
1634        int err = 0;
1635
1636        if (!ns_capable(mnt->mnt_ns->user_ns, CAP_SYS_ADMIN))
1637                return -EPERM;
1638
1639        if (path->dentry != path->mnt->mnt_root)
1640                return -EINVAL;
1641
1642        type = flags_to_propagation_type(flag);
1643        if (!type)
1644                return -EINVAL;
1645
1646        down_write(&namespace_sem);
1647        if (type == MS_SHARED) {
1648                err = invent_group_ids(mnt, recurse);
1649                if (err)
1650                        goto out_unlock;
1651        }
1652
1653        br_write_lock(&vfsmount_lock);
1654        for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1655                change_mnt_propagation(m, type);
1656        br_write_unlock(&vfsmount_lock);
1657
1658 out_unlock:
1659        up_write(&namespace_sem);
1660        return err;
1661}
1662
1663/*
1664 * do loopback mount.
1665 */
1666static int do_loopback(struct path *path, const char *old_name,
1667                                int recurse)
1668{
1669        LIST_HEAD(umount_list);
1670        struct path old_path;
1671        struct mount *mnt = NULL, *old;
1672        int err = mount_is_safe(path);
1673        if (err)
1674                return err;
1675        if (!old_name || !*old_name)
1676                return -EINVAL;
1677        err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1678        if (err)
1679                return err;
1680
1681        err = -EINVAL;
1682        if (mnt_ns_loop(&old_path))
1683                goto out; 
1684
1685        err = lock_mount(path);
1686        if (err)
1687                goto out;
1688
1689        old = real_mount(old_path.mnt);
1690
1691        err = -EINVAL;
1692        if (IS_MNT_UNBINDABLE(old))
1693                goto out2;
1694
1695        if (!check_mnt(real_mount(path->mnt)) || !check_mnt(old))
1696                goto out2;
1697
1698        if (recurse)
1699                mnt = copy_tree(old, old_path.dentry, 0);
1700        else
1701                mnt = clone_mnt(old, old_path.dentry, 0);
1702
1703        if (IS_ERR(mnt)) {
1704                err = PTR_ERR(mnt);
1705                goto out;
1706        }
1707
1708        err = graft_tree(mnt, path);
1709        if (err) {
1710                br_write_lock(&vfsmount_lock);
1711                umount_tree(mnt, 0, &umount_list);
1712                br_write_unlock(&vfsmount_lock);
1713        }
1714out2:
1715        unlock_mount(path);
1716        release_mounts(&umount_list);
1717out:
1718        path_put(&old_path);
1719        return err;
1720}
1721
1722static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1723{
1724        int error = 0;
1725        int readonly_request = 0;
1726
1727        if (ms_flags & MS_RDONLY)
1728                readonly_request = 1;
1729        if (readonly_request == __mnt_is_readonly(mnt))
1730                return 0;
1731
1732        if (readonly_request)
1733                error = mnt_make_readonly(real_mount(mnt));
1734        else
1735                __mnt_unmake_readonly(real_mount(mnt));
1736        return error;
1737}
1738
1739/*
1740 * change filesystem flags. dir should be a physical root of filesystem.
1741 * If you've mounted a non-root directory somewhere and want to do remount
1742 * on it - tough luck.
1743 */
1744static int do_remount(struct path *path, int flags, int mnt_flags,
1745                      void *data)
1746{
1747        int err;
1748        struct super_block *sb = path->mnt->mnt_sb;
1749        struct mount *mnt = real_mount(path->mnt);
1750
1751        if (!capable(CAP_SYS_ADMIN))
1752                return -EPERM;
1753
1754        if (!check_mnt(mnt))
1755                return -EINVAL;
1756
1757        if (path->dentry != path->mnt->mnt_root)
1758                return -EINVAL;
1759
1760        err = security_sb_remount(sb, data);
1761        if (err)
1762                return err;
1763
1764        down_write(&sb->s_umount);
1765        if (flags & MS_BIND)
1766                err = change_mount_flags(path->mnt, flags);
1767        else
1768                err = do_remount_sb(sb, flags, data, 0);
1769        if (!err) {
1770                br_write_lock(&vfsmount_lock);
1771                mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK;
1772                mnt->mnt.mnt_flags = mnt_flags;
1773                br_write_unlock(&vfsmount_lock);
1774        }
1775        up_write(&sb->s_umount);
1776        if (!err) {
1777                br_write_lock(&vfsmount_lock);
1778                touch_mnt_namespace(mnt->mnt_ns);
1779                br_write_unlock(&vfsmount_lock);
1780        }
1781        return err;
1782}
1783
1784static inline int tree_contains_unbindable(struct mount *mnt)
1785{
1786        struct mount *p;
1787        for (p = mnt; p; p = next_mnt(p, mnt)) {
1788                if (IS_MNT_UNBINDABLE(p))
1789                        return 1;
1790        }
1791        return 0;
1792}
1793
1794static int do_move_mount(struct path *path, const char *old_name)
1795{
1796        struct path old_path, parent_path;
1797        struct mount *p;
1798        struct mount *old;
1799        int err = 0;
1800        if (!ns_capable(real_mount(path->mnt)->mnt_ns->user_ns, CAP_SYS_ADMIN))
1801                return -EPERM;
1802        if (!old_name || !*old_name)
1803                return -EINVAL;
1804        err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1805        if (err)
1806                return err;
1807
1808        err = lock_mount(path);
1809        if (err < 0)
1810                goto out;
1811
1812        old = real_mount(old_path.mnt);
1813        p = real_mount(path->mnt);
1814
1815        err = -EINVAL;
1816        if (!check_mnt(p) || !check_mnt(old))
1817                goto out1;
1818
1819        if (d_unlinked(path->dentry))
1820                goto out1;
1821
1822        err = -EINVAL;
1823        if (old_path.dentry != old_path.mnt->mnt_root)
1824                goto out1;
1825
1826        if (!mnt_has_parent(old))
1827                goto out1;
1828
1829        if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1830              S_ISDIR(old_path.dentry->d_inode->i_mode))
1831                goto out1;
1832        /*
1833         * Don't move a mount residing in a shared parent.
1834         */
1835        if (IS_MNT_SHARED(old->mnt_parent))
1836                goto out1;
1837        /*
1838         * Don't move a mount tree containing unbindable mounts to a destination
1839         * mount which is shared.
1840         */
1841        if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
1842                goto out1;
1843        err = -ELOOP;
1844        for (; mnt_has_parent(p); p = p->mnt_parent)
1845                if (p == old)
1846                        goto out1;
1847
1848        err = attach_recursive_mnt(old, path, &parent_path);
1849        if (err)
1850                goto out1;
1851
1852        /* if the mount is moved, it should no longer be expire
1853         * automatically */
1854        list_del_init(&old->mnt_expire);
1855out1:
1856        unlock_mount(path);
1857out:
1858        if (!err)
1859                path_put(&parent_path);
1860        path_put(&old_path);
1861        return err;
1862}
1863
1864static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1865{
1866        int err;
1867        const char *subtype = strchr(fstype, '.');
1868        if (subtype) {
1869                subtype++;
1870                err = -EINVAL;
1871                if (!subtype[0])
1872                        goto err;
1873        } else
1874                subtype = "";
1875
1876        mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1877        err = -ENOMEM;
1878        if (!mnt->mnt_sb->s_subtype)
1879                goto err;
1880        return mnt;
1881
1882 err:
1883        mntput(mnt);
1884        return ERR_PTR(err);
1885}
1886
1887/*
1888 * add a mount into a namespace's mount tree
1889 */
1890static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
1891{
1892        int err;
1893
1894        mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1895
1896        err = lock_mount(path);
1897        if (err)
1898                return err;
1899
1900        err = -EINVAL;
1901        if (unlikely(!check_mnt(real_mount(path->mnt)))) {
1902                /* that's acceptable only for automounts done in private ns */
1903                if (!(mnt_flags & MNT_SHRINKABLE))
1904                        goto unlock;
1905                /* ... and for those we'd better have mountpoint still alive */
1906                if (!real_mount(path->mnt)->mnt_ns)
1907                        goto unlock;
1908        }
1909
1910        /* Refuse the same filesystem on the same mount point */
1911        err = -EBUSY;
1912        if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
1913            path->mnt->mnt_root == path->dentry)
1914                goto unlock;
1915
1916        err = -EINVAL;
1917        if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
1918                goto unlock;
1919
1920        newmnt->mnt.mnt_flags = mnt_flags;
1921        err = graft_tree(newmnt, path);
1922
1923unlock:
1924        unlock_mount(path);
1925        return err;
1926}
1927
1928/*
1929 * create a new mount for userspace and request it to be added into the
1930 * namespace's tree
1931 */
1932static int do_new_mount(struct path *path, const char *fstype, int flags,
1933                        int mnt_flags, const char *name, void *data)
1934{
1935        struct file_system_type *type;
1936        struct user_namespace *user_ns;
1937        struct vfsmount *mnt;
1938        int err;
1939
1940        if (!fstype)
1941                return -EINVAL;
1942
1943        /* we need capabilities... */
1944        user_ns = real_mount(path->mnt)->mnt_ns->user_ns;
1945        if (!ns_capable(user_ns, CAP_SYS_ADMIN))
1946                return -EPERM;
1947
1948        type = get_fs_type(fstype);
1949        if (!type)
1950                return -ENODEV;
1951
1952        if (user_ns != &init_user_ns) {
1953                if (!(type->fs_flags & FS_USERNS_MOUNT)) {
1954                        put_filesystem(type);
1955                        return -EPERM;
1956                }
1957                /* Only in special cases allow devices from mounts
1958                 * created outside the initial user namespace.
1959                 */
1960                if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
1961                        flags |= MS_NODEV;
1962                        mnt_flags |= MNT_NODEV;
1963                }
1964        }
1965
1966        mnt = vfs_kern_mount(type, flags, name, data);
1967        if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1968            !mnt->mnt_sb->s_subtype)
1969                mnt = fs_set_subtype(mnt, fstype);
1970
1971        put_filesystem(type);
1972        if (IS_ERR(mnt))
1973                return PTR_ERR(mnt);
1974
1975        err = do_add_mount(real_mount(mnt), path, mnt_flags);
1976        if (err)
1977                mntput(mnt);
1978        return err;
1979}
1980
1981int finish_automount(struct vfsmount *m, struct path *path)
1982{
1983        struct mount *mnt = real_mount(m);
1984        int err;
1985        /* The new mount record should have at least 2 refs to prevent it being
1986         * expired before we get a chance to add it
1987         */
1988        BUG_ON(mnt_get_count(mnt) < 2);
1989
1990        if (m->mnt_sb == path->mnt->mnt_sb &&
1991            m->mnt_root == path->dentry) {
1992                err = -ELOOP;
1993                goto fail;
1994        }
1995
1996        err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
1997        if (!err)
1998                return 0;
1999fail:
2000        /* remove m from any expiration list it may be on */
2001        if (!list_empty(&mnt->mnt_expire)) {
2002                down_write(&namespace_sem);
2003                br_write_lock(&vfsmount_lock);
2004                list_del_init(&mnt->mnt_expire);
2005                br_write_unlock(&vfsmount_lock);
2006                up_write(&namespace_sem);
2007        }
2008        mntput(m);
2009        mntput(m);
2010        return err;
2011}
2012
2013/**
2014 * mnt_set_expiry - Put a mount on an expiration list
2015 * @mnt: The mount to list.
2016 * @expiry_list: The list to add the mount to.
2017 */
2018void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2019{
2020        down_write(&namespace_sem);
2021        br_write_lock(&vfsmount_lock);
2022
2023        list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2024
2025        br_write_unlock(&vfsmount_lock);
2026        up_write(&namespace_sem);
2027}
2028EXPORT_SYMBOL(mnt_set_expiry);
2029
2030/*
2031 * process a list of expirable mountpoints with the intent of discarding any
2032 * mountpoints that aren't in use and haven't been touched since last we came
2033 * here
2034 */
2035void mark_mounts_for_expiry(struct list_head *mounts)
2036{
2037        struct mount *mnt, *next;
2038        LIST_HEAD(graveyard);
2039        LIST_HEAD(umounts);
2040
2041        if (list_empty(mounts))
2042                return;
2043
2044        down_write(&namespace_sem);
2045        br_write_lock(&vfsmount_lock);
2046
2047        /* extract from the expiration list every vfsmount that matches the
2048         * following criteria:
2049         * - only referenced by its parent vfsmount
2050         * - still marked for expiry (marked on the last call here; marks are
2051         *   cleared by mntput())
2052         */
2053        list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2054                if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2055                        propagate_mount_busy(mnt, 1))
2056                        continue;
2057                list_move(&mnt->mnt_expire, &graveyard);
2058        }
2059        while (!list_empty(&graveyard)) {
2060                mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2061                touch_mnt_namespace(mnt->mnt_ns);
2062                umount_tree(mnt, 1, &umounts);
2063        }
2064        br_write_unlock(&vfsmount_lock);
2065        up_write(&namespace_sem);
2066
2067        release_mounts(&umounts);
2068}
2069
2070EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2071
2072/*
2073 * Ripoff of 'select_parent()'
2074 *
2075 * search the list of submounts for a given mountpoint, and move any
2076 * shrinkable submounts to the 'graveyard' list.
2077 */
2078static int select_submounts(struct mount *parent, struct list_head *graveyard)
2079{
2080        struct mount *this_parent = parent;
2081        struct list_head *next;
2082        int found = 0;
2083
2084repeat:
2085        next = this_parent->mnt_mounts.next;
2086resume:
2087        while (next != &this_parent->mnt_mounts) {
2088                struct list_head *tmp = next;
2089                struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2090
2091                next = tmp->next;
2092                if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2093                        continue;
2094                /*
2095                 * Descend a level if the d_mounts list is non-empty.
2096                 */
2097                if (!list_empty(&mnt->mnt_mounts)) {
2098                        this_parent = mnt;
2099                        goto repeat;
2100                }
2101
2102                if (!propagate_mount_busy(mnt, 1)) {
2103                        list_move_tail(&mnt->mnt_expire, graveyard);
2104                        found++;
2105                }
2106        }
2107        /*
2108         * All done at this level ... ascend and resume the search
2109         */
2110        if (this_parent != parent) {
2111                next = this_parent->mnt_child.next;
2112                this_parent = this_parent->mnt_parent;
2113                goto resume;
2114        }
2115        return found;
2116}
2117
2118/*
2119 * process a list of expirable mountpoints with the intent of discarding any
2120 * submounts of a specific parent mountpoint
2121 *
2122 * vfsmount_lock must be held for write
2123 */
2124static void shrink_submounts(struct mount *mnt, struct list_head *umounts)
2125{
2126        LIST_HEAD(graveyard);
2127        struct mount *m;
2128
2129        /* extract submounts of 'mountpoint' from the expiration list */
2130        while (select_submounts(mnt, &graveyard)) {
2131                while (!list_empty(&graveyard)) {
2132                        m = list_first_entry(&graveyard, struct mount,
2133                                                mnt_expire);
2134                        touch_mnt_namespace(m->mnt_ns);
2135                        umount_tree(m, 1, umounts);
2136                }
2137        }
2138}
2139
2140/*
2141 * Some copy_from_user() implementations do not return the exact number of
2142 * bytes remaining to copy on a fault.  But copy_mount_options() requires that.
2143 * Note that this function differs from copy_from_user() in that it will oops
2144 * on bad values of `to', rather than returning a short copy.
2145 */
2146static long exact_copy_from_user(void *to, const void __user * from,
2147                                 unsigned long n)
2148{
2149        char *t = to;
2150        const char __user *f = from;
2151        char c;
2152
2153        if (!access_ok(VERIFY_READ, from, n))
2154                return n;
2155
2156        while (n) {
2157                if (__get_user(c, f)) {
2158                        memset(t, 0, n);
2159                        break;
2160                }
2161                *t++ = c;
2162                f++;
2163                n--;
2164        }
2165        return n;
2166}
2167
2168int copy_mount_options(const void __user * data, unsigned long *where)
2169{
2170        int i;
2171        unsigned long page;
2172        unsigned long size;
2173
2174        *where = 0;
2175        if (!data)
2176                return 0;
2177
2178        if (!(page = __get_free_page(GFP_KERNEL)))
2179                return -ENOMEM;
2180
2181        /* We only care that *some* data at the address the user
2182         * gave us is valid.  Just in case, we'll zero
2183         * the remainder of the page.
2184         */
2185        /* copy_from_user cannot cross TASK_SIZE ! */
2186        size = TASK_SIZE - (unsigned long)data;
2187        if (size > PAGE_SIZE)
2188                size = PAGE_SIZE;
2189
2190        i = size - exact_copy_from_user((void *)page, data, size);
2191        if (!i) {
2192                free_page(page);
2193                return -EFAULT;
2194        }
2195        if (i != PAGE_SIZE)
2196                memset((char *)page + i, 0, PAGE_SIZE - i);
2197        *where = page;
2198        return 0;
2199}
2200
2201int copy_mount_string(const void __user *data, char **where)
2202{
2203        char *tmp;
2204
2205        if (!data) {
2206                *where = NULL;
2207                return 0;
2208        }
2209
2210        tmp = strndup_user(data, PAGE_SIZE);
2211        if (IS_ERR(tmp))
2212                return PTR_ERR(tmp);
2213
2214        *where = tmp;
2215        return 0;
2216}
2217
2218/*
2219 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2220 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2221 *
2222 * data is a (void *) that can point to any structure up to
2223 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2224 * information (or be NULL).
2225 *
2226 * Pre-0.97 versions of mount() didn't have a flags word.
2227 * When the flags word was introduced its top half was required
2228 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2229 * Therefore, if this magic number is present, it carries no information
2230 * and must be discarded.
2231 */
2232long do_mount(const char *dev_name, const char *dir_name,
2233                const char *type_page, unsigned long flags, void *data_page)
2234{
2235        struct path path;
2236        int retval = 0;
2237        int mnt_flags = 0;
2238
2239        /* Discard magic */
2240        if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2241                flags &= ~MS_MGC_MSK;
2242
2243        /* Basic sanity checks */
2244
2245        if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2246                return -EINVAL;
2247
2248        if (data_page)
2249                ((char *)data_page)[PAGE_SIZE - 1] = 0;
2250
2251        /* ... and get the mountpoint */
2252        retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2253        if (retval)
2254                return retval;
2255
2256        retval = security_sb_mount(dev_name, &path,
2257                                   type_page, flags, data_page);
2258        if (retval)
2259                goto dput_out;
2260
2261        /* Default to relatime unless overriden */
2262        if (!(flags & MS_NOATIME))
2263                mnt_flags |= MNT_RELATIME;
2264
2265        /* Separate the per-mountpoint flags */
2266        if (flags & MS_NOSUID)
2267                mnt_flags |= MNT_NOSUID;
2268        if (flags & MS_NODEV)
2269                mnt_flags |= MNT_NODEV;
2270        if (flags & MS_NOEXEC)
2271                mnt_flags |= MNT_NOEXEC;
2272        if (flags & MS_NOATIME)
2273                mnt_flags |= MNT_NOATIME;
2274        if (flags & MS_NODIRATIME)
2275                mnt_flags |= MNT_NODIRATIME;
2276        if (flags & MS_STRICTATIME)
2277                mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2278        if (flags & MS_RDONLY)
2279                mnt_flags |= MNT_READONLY;
2280
2281        flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2282                   MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2283                   MS_STRICTATIME);
2284
2285        if (flags & MS_REMOUNT)
2286                retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2287                                    data_page);
2288        else if (flags & MS_BIND)
2289                retval = do_loopback(&path, dev_name, flags & MS_REC);
2290        else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2291                retval = do_change_type(&path, flags);
2292        else if (flags & MS_MOVE)
2293                retval = do_move_mount(&path, dev_name);
2294        else
2295                retval = do_new_mount(&path, type_page, flags, mnt_flags,
2296                                      dev_name, data_page);
2297dput_out:
2298        path_put(&path);
2299        return retval;
2300}
2301
2302static void free_mnt_ns(struct mnt_namespace *ns)
2303{
2304        proc_free_inum(ns->proc_inum);
2305        put_user_ns(ns->user_ns);
2306        kfree(ns);
2307}
2308
2309/*
2310 * Assign a sequence number so we can detect when we attempt to bind
2311 * mount a reference to an older mount namespace into the current
2312 * mount namespace, preventing reference counting loops.  A 64bit
2313 * number incrementing at 10Ghz will take 12,427 years to wrap which
2314 * is effectively never, so we can ignore the possibility.
2315 */
2316static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2317
2318static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2319{
2320        struct mnt_namespace *new_ns;
2321        int ret;
2322
2323        new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2324        if (!new_ns)
2325                return ERR_PTR(-ENOMEM);
2326        ret = proc_alloc_inum(&new_ns->proc_inum);
2327        if (ret) {
2328                kfree(new_ns);
2329                return ERR_PTR(ret);
2330        }
2331        new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2332        atomic_set(&new_ns->count, 1);
2333        new_ns->root = NULL;
2334        INIT_LIST_HEAD(&new_ns->list);
2335        init_waitqueue_head(&new_ns->poll);
2336        new_ns->event = 0;
2337        new_ns->user_ns = get_user_ns(user_ns);
2338        return new_ns;
2339}
2340
2341/*
2342 * Allocate a new namespace structure and populate it with contents
2343 * copied from the namespace of the passed in task structure.
2344 */
2345static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2346                struct user_namespace *user_ns, struct fs_struct *fs)
2347{
2348        struct mnt_namespace *new_ns;
2349        struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2350        struct mount *p, *q;
2351        struct mount *old = mnt_ns->root;
2352        struct mount *new;
2353        int copy_flags;
2354
2355        new_ns = alloc_mnt_ns(user_ns);
2356        if (IS_ERR(new_ns))
2357                return new_ns;
2358
2359        down_write(&namespace_sem);
2360        /* First pass: copy the tree topology */
2361        copy_flags = CL_COPY_ALL | CL_EXPIRE;
2362        if (user_ns != mnt_ns->user_ns)
2363                copy_flags |= CL_SHARED_TO_SLAVE;
2364        new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2365        if (IS_ERR(new)) {
2366                up_write(&namespace_sem);
2367                free_mnt_ns(new_ns);
2368                return ERR_CAST(new);
2369        }
2370        new_ns->root = new;
2371        br_write_lock(&vfsmount_lock);
2372        list_add_tail(&new_ns->list, &new->mnt_list);
2373        br_write_unlock(&vfsmount_lock);
2374
2375        /*
2376         * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2377         * as belonging to new namespace.  We have already acquired a private
2378         * fs_struct, so tsk->fs->lock is not needed.
2379         */
2380        p = old;
2381        q = new;
2382        while (p) {
2383                q->mnt_ns = new_ns;
2384                if (fs) {
2385                        if (&p->mnt == fs->root.mnt) {
2386                                fs->root.mnt = mntget(&q->mnt);
2387                                rootmnt = &p->mnt;
2388                        }
2389                        if (&p->mnt == fs->pwd.mnt) {
2390                                fs->pwd.mnt = mntget(&q->mnt);
2391                                pwdmnt = &p->mnt;
2392                        }
2393                }
2394                p = next_mnt(p, old);
2395                q = next_mnt(q, new);
2396        }
2397        up_write(&namespace_sem);
2398
2399        if (rootmnt)
2400                mntput(rootmnt);
2401        if (pwdmnt)
2402                mntput(pwdmnt);
2403
2404        return new_ns;
2405}
2406
2407struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2408                struct user_namespace *user_ns, struct fs_struct *new_fs)
2409{
2410        struct mnt_namespace *new_ns;
2411
2412        BUG_ON(!ns);
2413        get_mnt_ns(ns);
2414
2415        if (!(flags & CLONE_NEWNS))
2416                return ns;
2417
2418        new_ns = dup_mnt_ns(ns, user_ns, new_fs);
2419
2420        put_mnt_ns(ns);
2421        return new_ns;
2422}
2423
2424/**
2425 * create_mnt_ns - creates a private namespace and adds a root filesystem
2426 * @mnt: pointer to the new root filesystem mountpoint
2427 */
2428static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2429{
2430        struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2431        if (!IS_ERR(new_ns)) {
2432                struct mount *mnt = real_mount(m);
2433                mnt->mnt_ns = new_ns;
2434                new_ns->root = mnt;
2435                list_add(&new_ns->list, &mnt->mnt_list);
2436        } else {
2437                mntput(m);
2438        }
2439        return new_ns;
2440}
2441
2442struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2443{
2444        struct mnt_namespace *ns;
2445        struct super_block *s;
2446        struct path path;
2447        int err;
2448
2449        ns = create_mnt_ns(mnt);
2450        if (IS_ERR(ns))
2451                return ERR_CAST(ns);
2452
2453        err = vfs_path_lookup(mnt->mnt_root, mnt,
2454                        name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2455
2456        put_mnt_ns(ns);
2457
2458        if (err)
2459                return ERR_PTR(err);
2460
2461        /* trade a vfsmount reference for active sb one */
2462        s = path.mnt->mnt_sb;
2463        atomic_inc(&s->s_active);
2464        mntput(path.mnt);
2465        /* lock the sucker */
2466        down_write(&s->s_umount);
2467        /* ... and return the root of (sub)tree on it */
2468        return path.dentry;
2469}
2470EXPORT_SYMBOL(mount_subtree);
2471
2472SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2473                char __user *, type, unsigned long, flags, void __user *, data)
2474{
2475        int ret;
2476        char *kernel_type;
2477        struct filename *kernel_dir;
2478        char *kernel_dev;
2479        unsigned long data_page;
2480
2481        ret = copy_mount_string(type, &kernel_type);
2482        if (ret < 0)
2483                goto out_type;
2484
2485        kernel_dir = getname(dir_name);
2486        if (IS_ERR(kernel_dir)) {
2487                ret = PTR_ERR(kernel_dir);
2488                goto out_dir;
2489        }
2490
2491        ret = copy_mount_string(dev_name, &kernel_dev);
2492        if (ret < 0)
2493                goto out_dev;
2494
2495        ret = copy_mount_options(data, &data_page);
2496        if (ret < 0)
2497                goto out_data;
2498
2499        ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags,
2500                (void *) data_page);
2501
2502        free_page(data_page);
2503out_data:
2504        kfree(kernel_dev);
2505out_dev:
2506        putname(kernel_dir);
2507out_dir:
2508        kfree(kernel_type);
2509out_type:
2510        return ret;
2511}
2512
2513/*
2514 * Return true if path is reachable from root
2515 *
2516 * namespace_sem or vfsmount_lock is held
2517 */
2518bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2519                         const struct path *root)
2520{
2521        while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2522                dentry = mnt->mnt_mountpoint;
2523                mnt = mnt->mnt_parent;
2524        }
2525        return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2526}
2527
2528int path_is_under(struct path *path1, struct path *path2)
2529{
2530        int res;
2531        br_read_lock(&vfsmount_lock);
2532        res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2533        br_read_unlock(&vfsmount_lock);
2534        return res;
2535}
2536EXPORT_SYMBOL(path_is_under);
2537
2538/*
2539 * pivot_root Semantics:
2540 * Moves the root file system of the current process to the directory put_old,
2541 * makes new_root as the new root file system of the current process, and sets
2542 * root/cwd of all processes which had them on the current root to new_root.
2543 *
2544 * Restrictions:
2545 * The new_root and put_old must be directories, and  must not be on the
2546 * same file  system as the current process root. The put_old  must  be
2547 * underneath new_root,  i.e. adding a non-zero number of /.. to the string
2548 * pointed to by put_old must yield the same directory as new_root. No other
2549 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2550 *
2551 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2552 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2553 * in this situation.
2554 *
2555 * Notes:
2556 *  - we don't move root/cwd if they are not at the root (reason: if something
2557 *    cared enough to change them, it's probably wrong to force them elsewhere)
2558 *  - it's okay to pick a root that isn't the root of a file system, e.g.
2559 *    /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2560 *    though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2561 *    first.
2562 */
2563SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2564                const char __user *, put_old)
2565{
2566        struct path new, old, parent_path, root_parent, root;
2567        struct mount *new_mnt, *root_mnt;
2568        int error;
2569
2570        if (!ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN))
2571                return -EPERM;
2572
2573        error = user_path_dir(new_root, &new);
2574        if (error)
2575                goto out0;
2576
2577        error = user_path_dir(put_old, &old);
2578        if (error)
2579                goto out1;
2580
2581        error = security_sb_pivotroot(&old, &new);
2582        if (error)
2583                goto out2;
2584
2585        get_fs_root(current->fs, &root);
2586        error = lock_mount(&old);
2587        if (error)
2588                goto out3;
2589
2590        error = -EINVAL;
2591        new_mnt = real_mount(new.mnt);
2592        root_mnt = real_mount(root.mnt);
2593        if (IS_MNT_SHARED(real_mount(old.mnt)) ||
2594                IS_MNT_SHARED(new_mnt->mnt_parent) ||
2595                IS_MNT_SHARED(root_mnt->mnt_parent))
2596                goto out4;
2597        if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
2598                goto out4;
2599        error = -ENOENT;
2600        if (d_unlinked(new.dentry))
2601                goto out4;
2602        if (d_unlinked(old.dentry))
2603                goto out4;
2604        error = -EBUSY;
2605        if (new.mnt == root.mnt ||
2606            old.mnt == root.mnt)
2607                goto out4; /* loop, on the same file system  */
2608        error = -EINVAL;
2609        if (root.mnt->mnt_root != root.dentry)
2610                goto out4; /* not a mountpoint */
2611        if (!mnt_has_parent(root_mnt))
2612                goto out4; /* not attached */
2613        if (new.mnt->mnt_root != new.dentry)
2614                goto out4; /* not a mountpoint */
2615        if (!mnt_has_parent(new_mnt))
2616                goto out4; /* not attached */
2617        /* make sure we can reach put_old from new_root */
2618        if (!is_path_reachable(real_mount(old.mnt), old.dentry, &new))
2619                goto out4;
2620        br_write_lock(&vfsmount_lock);
2621        detach_mnt(new_mnt, &parent_path);
2622        detach_mnt(root_mnt, &root_parent);
2623        /* mount old root on put_old */
2624        attach_mnt(root_mnt, &old);
2625        /* mount new_root on / */
2626        attach_mnt(new_mnt, &root_parent);
2627        touch_mnt_namespace(current->nsproxy->mnt_ns);
2628        br_write_unlock(&vfsmount_lock);
2629        chroot_fs_refs(&root, &new);
2630        error = 0;
2631out4:
2632        unlock_mount(&old);
2633        if (!error) {
2634                path_put(&root_parent);
2635                path_put(&parent_path);
2636        }
2637out3:
2638        path_put(&root);
2639out2:
2640        path_put(&old);
2641out1:
2642        path_put(&new);
2643out0:
2644        return error;
2645}
2646
2647static void __init init_mount_tree(void)
2648{
2649        struct vfsmount *mnt;
2650        struct mnt_namespace *ns;
2651        struct path root;
2652        struct file_system_type *type;
2653
2654        type = get_fs_type("rootfs");
2655        if (!type)
2656                panic("Can't find rootfs type");
2657        mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
2658        put_filesystem(type);
2659        if (IS_ERR(mnt))
2660                panic("Can't create rootfs");
2661
2662        ns = create_mnt_ns(mnt);
2663        if (IS_ERR(ns))
2664                panic("Can't allocate initial namespace");
2665
2666        init_task.nsproxy->mnt_ns = ns;
2667        get_mnt_ns(ns);
2668
2669        root.mnt = mnt;
2670        root.dentry = mnt->mnt_root;
2671
2672        set_fs_pwd(current->fs, &root);
2673        set_fs_root(current->fs, &root);
2674}
2675
2676void __init mnt_init(void)
2677{
2678        unsigned u;
2679        int err;
2680
2681        init_rwsem(&namespace_sem);
2682
2683        mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
2684                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2685
2686        mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2687
2688        if (!mount_hashtable)
2689                panic("Failed to allocate mount hash table\n");
2690
2691        printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
2692
2693        for (u = 0; u < HASH_SIZE; u++)
2694                INIT_LIST_HEAD(&mount_hashtable[u]);
2695
2696        br_lock_init(&vfsmount_lock);
2697
2698        err = sysfs_init();
2699        if (err)
2700                printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2701                        __func__, err);
2702        fs_kobj = kobject_create_and_add("fs", NULL);
2703        if (!fs_kobj)
2704                printk(KERN_WARNING "%s: kobj create error\n", __func__);
2705        init_rootfs();
2706        init_mount_tree();
2707}
2708
2709void put_mnt_ns(struct mnt_namespace *ns)
2710{
2711        LIST_HEAD(umount_list);
2712
2713        if (!atomic_dec_and_test(&ns->count))
2714                return;
2715        down_write(&namespace_sem);
2716        br_write_lock(&vfsmount_lock);
2717        umount_tree(ns->root, 0, &umount_list);
2718        br_write_unlock(&vfsmount_lock);
2719        up_write(&namespace_sem);
2720        release_mounts(&umount_list);
2721        free_mnt_ns(ns);
2722}
2723
2724struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2725{
2726        struct vfsmount *mnt;
2727        mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2728        if (!IS_ERR(mnt)) {
2729                /*
2730                 * it is a longterm mount, don't release mnt until
2731                 * we unmount before file sys is unregistered
2732                */
2733                real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
2734        }
2735        return mnt;
2736}
2737EXPORT_SYMBOL_GPL(kern_mount_data);
2738
2739void kern_unmount(struct vfsmount *mnt)
2740{
2741        /* release long term mount so mount point can be released */
2742        if (!IS_ERR_OR_NULL(mnt)) {
2743                br_write_lock(&vfsmount_lock);
2744                real_mount(mnt)->mnt_ns = NULL;
2745                br_write_unlock(&vfsmount_lock);
2746                mntput(mnt);
2747        }
2748}
2749EXPORT_SYMBOL(kern_unmount);
2750
2751bool our_mnt(struct vfsmount *mnt)
2752{
2753        return check_mnt(real_mount(mnt));
2754}
2755
2756static void *mntns_get(struct task_struct *task)
2757{
2758        struct mnt_namespace *ns = NULL;
2759        struct nsproxy *nsproxy;
2760
2761        rcu_read_lock();
2762        nsproxy = task_nsproxy(task);
2763        if (nsproxy) {
2764                ns = nsproxy->mnt_ns;
2765                get_mnt_ns(ns);
2766        }
2767        rcu_read_unlock();
2768
2769        return ns;
2770}
2771
2772static void mntns_put(void *ns)
2773{
2774        put_mnt_ns(ns);
2775}
2776
2777static int mntns_install(struct nsproxy *nsproxy, void *ns)
2778{
2779        struct fs_struct *fs = current->fs;
2780        struct mnt_namespace *mnt_ns = ns;
2781        struct path root;
2782
2783        if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
2784            !nsown_capable(CAP_SYS_CHROOT) ||
2785            !nsown_capable(CAP_SYS_ADMIN))
2786                return -EPERM;
2787
2788        if (fs->users != 1)
2789                return -EINVAL;
2790
2791        get_mnt_ns(mnt_ns);
2792        put_mnt_ns(nsproxy->mnt_ns);
2793        nsproxy->mnt_ns = mnt_ns;
2794
2795        /* Find the root */
2796        root.mnt    = &mnt_ns->root->mnt;
2797        root.dentry = mnt_ns->root->mnt.mnt_root;
2798        path_get(&root);
2799        while(d_mountpoint(root.dentry) && follow_down_one(&root))
2800                ;
2801
2802        /* Update the pwd and root */
2803        set_fs_pwd(fs, &root);
2804        set_fs_root(fs, &root);
2805
2806        path_put(&root);
2807        return 0;
2808}
2809
2810static unsigned int mntns_inum(void *ns)
2811{
2812        struct mnt_namespace *mnt_ns = ns;
2813        return mnt_ns->proc_inum;
2814}
2815
2816const struct proc_ns_operations mntns_operations = {
2817        .name           = "mnt",
2818        .type           = CLONE_NEWNS,
2819        .get            = mntns_get,
2820        .put            = mntns_put,
2821        .install        = mntns_install,
2822        .inum           = mntns_inum,
2823};
2824
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.