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_inode(file);
 388
 389        if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
 390                return __mnt_want_write(file->f_path.mnt);
 391        else
 392                return mnt_clone_write(file->f_path.mnt);
 393}
 394
 395/**
 396 * mnt_want_write_file - get write access to a file's mount
 397 * @file: the file who's mount on which to take a write
 398 *
 399 * This is like mnt_want_write, but it takes a file and can
 400 * do some optimisations if the file is open for write already
 401 */
 402int mnt_want_write_file(struct file *file)
 403{
 404        int ret;
 405
 406        sb_start_write(file->f_path.mnt->mnt_sb);
 407        ret = __mnt_want_write_file(file);
 408        if (ret)
 409                sb_end_write(file->f_path.mnt->mnt_sb);
 410        return ret;
 411}
 412EXPORT_SYMBOL_GPL(mnt_want_write_file);
 413
 414/**
 415 * __mnt_drop_write - give up write access to a mount
 416 * @mnt: the mount on which to give up write access
 417 *
 418 * Tells the low-level filesystem that we are done
 419 * performing writes to it.  Must be matched with
 420 * __mnt_want_write() call above.
 421 */
 422void __mnt_drop_write(struct vfsmount *mnt)
 423{
 424        preempt_disable();
 425        mnt_dec_writers(real_mount(mnt));
 426        preempt_enable();
 427}
 428
 429/**
 430 * mnt_drop_write - give up write access to a mount
 431 * @mnt: the mount on which to give up write access
 432 *
 433 * Tells the low-level filesystem that we are done performing writes to it and
 434 * also allows filesystem to be frozen again.  Must be matched with
 435 * mnt_want_write() call above.
 436 */
 437void mnt_drop_write(struct vfsmount *mnt)
 438{
 439        __mnt_drop_write(mnt);
 440        sb_end_write(mnt->mnt_sb);
 441}
 442EXPORT_SYMBOL_GPL(mnt_drop_write);
 443
 444void __mnt_drop_write_file(struct file *file)
 445{
 446        __mnt_drop_write(file->f_path.mnt);
 447}
 448
 449void mnt_drop_write_file(struct file *file)
 450{
 451        mnt_drop_write(file->f_path.mnt);
 452}
 453EXPORT_SYMBOL(mnt_drop_write_file);
 454
 455static int mnt_make_readonly(struct mount *mnt)
 456{
 457        int ret = 0;
 458
 459        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        /* Don't allow unprivileged users to change mount flags */
 802        if ((flag & CL_UNPRIVILEGED) && (mnt->mnt.mnt_flags & MNT_READONLY))
 803                mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
 804
 805        atomic_inc(&sb->s_active);
 806        mnt->mnt.mnt_sb = sb;
 807        mnt->mnt.mnt_root = dget(root);
 808        mnt->mnt_mountpoint = mnt->mnt.mnt_root;
 809        mnt->mnt_parent = mnt;
 810        br_write_lock(&vfsmount_lock);
 811        list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
 812        br_write_unlock(&vfsmount_lock);
 813
 814        if ((flag & CL_SLAVE) ||
 815            ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
 816                list_add(&mnt->mnt_slave, &old->mnt_slave_list);
 817                mnt->mnt_master = old;
 818                CLEAR_MNT_SHARED(mnt);
 819        } else if (!(flag & CL_PRIVATE)) {
 820                if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
 821                        list_add(&mnt->mnt_share, &old->mnt_share);
 822                if (IS_MNT_SLAVE(old))
 823                        list_add(&mnt->mnt_slave, &old->mnt_slave);
 824                mnt->mnt_master = old->mnt_master;
 825        }
 826        if (flag & CL_MAKE_SHARED)
 827                set_mnt_shared(mnt);
 828
 829        /* stick the duplicate mount on the same expiry list
 830         * as the original if that was on one */
 831        if (flag & CL_EXPIRE) {
 832                if (!list_empty(&old->mnt_expire))
 833                        list_add(&mnt->mnt_expire, &old->mnt_expire);
 834        }
 835
 836        return mnt;
 837
 838 out_free:
 839        free_vfsmnt(mnt);
 840        return ERR_PTR(err);
 841}
 842
 843static inline void mntfree(struct mount *mnt)
 844{
 845        struct vfsmount *m = &mnt->mnt;
 846        struct super_block *sb = m->mnt_sb;
 847
 848        /*
 849         * This probably indicates that somebody messed
 850         * up a mnt_want/drop_write() pair.  If this
 851         * happens, the filesystem was probably unable
 852         * to make r/w->r/o transitions.
 853         */
 854        /*
 855         * The locking used to deal with mnt_count decrement provides barriers,
 856         * so mnt_get_writers() below is safe.
 857         */
 858        WARN_ON(mnt_get_writers(mnt));
 859        fsnotify_vfsmount_delete(m);
 860        dput(m->mnt_root);
 861        free_vfsmnt(mnt);
 862        deactivate_super(sb);
 863}
 864
 865static void mntput_no_expire(struct mount *mnt)
 866{
 867put_again:
 868#ifdef CONFIG_SMP
 869        br_read_lock(&vfsmount_lock);
 870        if (likely(mnt->mnt_ns)) {
 871                /* shouldn't be the last one */
 872                mnt_add_count(mnt, -1);
 873                br_read_unlock(&vfsmount_lock);
 874                return;
 875        }
 876        br_read_unlock(&vfsmount_lock);
 877
 878        br_write_lock(&vfsmount_lock);
 879        mnt_add_count(mnt, -1);
 880        if (mnt_get_count(mnt)) {
 881                br_write_unlock(&vfsmount_lock);
 882                return;
 883        }
 884#else
 885        mnt_add_count(mnt, -1);
 886        if (likely(mnt_get_count(mnt)))
 887                return;
 888        br_write_lock(&vfsmount_lock);
 889#endif
 890        if (unlikely(mnt->mnt_pinned)) {
 891                mnt_add_count(mnt, mnt->mnt_pinned + 1);
 892                mnt->mnt_pinned = 0;
 893                br_write_unlock(&vfsmount_lock);
 894                acct_auto_close_mnt(&mnt->mnt);
 895                goto put_again;
 896        }
 897
 898        list_del(&mnt->mnt_instance);
 899        br_write_unlock(&vfsmount_lock);
 900        mntfree(mnt);
 901}
 902
 903void mntput(struct vfsmount *mnt)
 904{
 905        if (mnt) {
 906                struct mount *m = real_mount(mnt);
 907                /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
 908                if (unlikely(m->mnt_expiry_mark))
 909                        m->mnt_expiry_mark = 0;
 910                mntput_no_expire(m);
 911        }
 912}
 913EXPORT_SYMBOL(mntput);
 914
 915struct vfsmount *mntget(struct vfsmount *mnt)
 916{
 917        if (mnt)
 918                mnt_add_count(real_mount(mnt), 1);
 919        return mnt;
 920}
 921EXPORT_SYMBOL(mntget);
 922
 923void mnt_pin(struct vfsmount *mnt)
 924{
 925        br_write_lock(&vfsmount_lock);
 926        real_mount(mnt)->mnt_pinned++;
 927        br_write_unlock(&vfsmount_lock);
 928}
 929EXPORT_SYMBOL(mnt_pin);
 930
 931void mnt_unpin(struct vfsmount *m)
 932{
 933        struct mount *mnt = real_mount(m);
 934        br_write_lock(&vfsmount_lock);
 935        if (mnt->mnt_pinned) {
 936                mnt_add_count(mnt, 1);
 937                mnt->mnt_pinned--;
 938        }
 939        br_write_unlock(&vfsmount_lock);
 940}
 941EXPORT_SYMBOL(mnt_unpin);
 942
 943static inline void mangle(struct seq_file *m, const char *s)
 944{
 945        seq_escape(m, s, " \t\n\\");
 946}
 947
 948/*
 949 * Simple .show_options callback for filesystems which don't want to
 950 * implement more complex mount option showing.
 951 *
 952 * See also save_mount_options().
 953 */
 954int generic_show_options(struct seq_file *m, struct dentry *root)
 955{
 956        const char *options;
 957
 958        rcu_read_lock();
 959        options = rcu_dereference(root->d_sb->s_options);
 960
 961        if (options != NULL && options[0]) {
 962                seq_putc(m, ',');
 963                mangle(m, options);
 964        }
 965        rcu_read_unlock();
 966
 967        return 0;
 968}
 969EXPORT_SYMBOL(generic_show_options);
 970
 971/*
 972 * If filesystem uses generic_show_options(), this function should be
 973 * called from the fill_super() callback.
 974 *
 975 * The .remount_fs callback usually needs to be handled in a special
 976 * way, to make sure, that previous options are not overwritten if the
 977 * remount fails.
 978 *
 979 * Also note, that if the filesystem's .remount_fs function doesn't
 980 * reset all options to their default value, but changes only newly
 981 * given options, then the displayed options will not reflect reality
 982 * any more.
 983 */
 984void save_mount_options(struct super_block *sb, char *options)
 985{
 986        BUG_ON(sb->s_options);
 987        rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
 988}
 989EXPORT_SYMBOL(save_mount_options);
 990
 991void replace_mount_options(struct super_block *sb, char *options)
 992{
 993        char *old = sb->s_options;
 994        rcu_assign_pointer(sb->s_options, options);
 995        if (old) {
 996                synchronize_rcu();
 997                kfree(old);
 998        }
 999}
1000EXPORT_SYMBOL(replace_mount_options);
1001
1002#ifdef CONFIG_PROC_FS
1003/* iterator; we want it to have access to namespace_sem, thus here... */
1004static void *m_start(struct seq_file *m, loff_t *pos)
1005{
1006        struct proc_mounts *p = proc_mounts(m);
1007
1008        down_read(&namespace_sem);
1009        return seq_list_start(&p->ns->list, *pos);
1010}
1011
1012static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1013{
1014        struct proc_mounts *p = proc_mounts(m);
1015
1016        return seq_list_next(v, &p->ns->list, pos);
1017}
1018
1019static void m_stop(struct seq_file *m, void *v)
1020{
1021        up_read(&namespace_sem);
1022}
1023
1024static int m_show(struct seq_file *m, void *v)
1025{
1026        struct proc_mounts *p = proc_mounts(m);
1027        struct mount *r = list_entry(v, struct mount, mnt_list);
1028        return p->show(m, &r->mnt);
1029}
1030
1031const struct seq_operations mounts_op = {
1032        .start  = m_start,
1033        .next   = m_next,
1034        .stop   = m_stop,
1035        .show   = m_show,
1036};
1037#endif  /* CONFIG_PROC_FS */
1038
1039/**
1040 * may_umount_tree - check if a mount tree is busy
1041 * @mnt: root of mount tree
1042 *
1043 * This is called to check if a tree of mounts has any
1044 * open files, pwds, chroots or sub mounts that are
1045 * busy.
1046 */
1047int may_umount_tree(struct vfsmount *m)
1048{
1049        struct mount *mnt = real_mount(m);
1050        int actual_refs = 0;
1051        int minimum_refs = 0;
1052        struct mount *p;
1053        BUG_ON(!m);
1054
1055        /* write lock needed for mnt_get_count */
1056        br_write_lock(&vfsmount_lock);
1057        for (p = mnt; p; p = next_mnt(p, mnt)) {
1058                actual_refs += mnt_get_count(p);
1059                minimum_refs += 2;
1060        }
1061        br_write_unlock(&vfsmount_lock);
1062
1063        if (actual_refs > minimum_refs)
1064                return 0;
1065
1066        return 1;
1067}
1068
1069EXPORT_SYMBOL(may_umount_tree);
1070
1071/**
1072 * may_umount - check if a mount point is busy
1073 * @mnt: root of mount
1074 *
1075 * This is called to check if a mount point has any
1076 * open files, pwds, chroots or sub mounts. If the
1077 * mount has sub mounts this will return busy
1078 * regardless of whether the sub mounts are busy.
1079 *
1080 * Doesn't take quota and stuff into account. IOW, in some cases it will
1081 * give false negatives. The main reason why it's here is that we need
1082 * a non-destructive way to look for easily umountable filesystems.
1083 */
1084int may_umount(struct vfsmount *mnt)
1085{
1086        int ret = 1;
1087        down_read(&namespace_sem);
1088        br_write_lock(&vfsmount_lock);
1089        if (propagate_mount_busy(real_mount(mnt), 2))
1090                ret = 0;
1091        br_write_unlock(&vfsmount_lock);
1092        up_read(&namespace_sem);
1093        return ret;
1094}
1095
1096EXPORT_SYMBOL(may_umount);
1097
1098void release_mounts(struct list_head *head)
1099{
1100        struct mount *mnt;
1101        while (!list_empty(head)) {
1102                mnt = list_first_entry(head, struct mount, mnt_hash);
1103                list_del_init(&mnt->mnt_hash);
1104                if (mnt_has_parent(mnt)) {
1105                        struct dentry *dentry;
1106                        struct mount *m;
1107
1108                        br_write_lock(&vfsmount_lock);
1109                        dentry = mnt->mnt_mountpoint;
1110                        m = mnt->mnt_parent;
1111                        mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1112                        mnt->mnt_parent = mnt;
1113                        m->mnt_ghosts--;
1114                        br_write_unlock(&vfsmount_lock);
1115                        dput(dentry);
1116                        mntput(&m->mnt);
1117                }
1118                mntput(&mnt->mnt);
1119        }
1120}
1121
1122/*
1123 * vfsmount lock must be held for write
1124 * namespace_sem must be held for write
1125 */
1126void umount_tree(struct mount *mnt, int propagate, struct list_head *kill)
1127{
1128        LIST_HEAD(tmp_list);
1129        struct mount *p;
1130
1131        for (p = mnt; p; p = next_mnt(p, mnt))
1132                list_move(&p->mnt_hash, &tmp_list);
1133
1134        if (propagate)
1135                propagate_umount(&tmp_list);
1136
1137        list_for_each_entry(p, &tmp_list, mnt_hash) {
1138                list_del_init(&p->mnt_expire);
1139                list_del_init(&p->mnt_list);
1140                __touch_mnt_namespace(p->mnt_ns);
1141                p->mnt_ns = NULL;
1142                list_del_init(&p->mnt_child);
1143                if (mnt_has_parent(p)) {
1144                        p->mnt_parent->mnt_ghosts++;
1145                        dentry_reset_mounted(p->mnt_mountpoint);
1146                }
1147                change_mnt_propagation(p, MS_PRIVATE);
1148        }
1149        list_splice(&tmp_list, kill);
1150}
1151
1152static void shrink_submounts(struct mount *mnt, struct list_head *umounts);
1153
1154static int do_umount(struct mount *mnt, int flags)
1155{
1156        struct super_block *sb = mnt->mnt.mnt_sb;
1157        int retval;
1158        LIST_HEAD(umount_list);
1159
1160        retval = security_sb_umount(&mnt->mnt, flags);
1161        if (retval)
1162                return retval;
1163
1164        /*
1165         * Allow userspace to request a mountpoint be expired rather than
1166         * unmounting unconditionally. Unmount only happens if:
1167         *  (1) the mark is already set (the mark is cleared by mntput())
1168         *  (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1169         */
1170        if (flags & MNT_EXPIRE) {
1171                if (&mnt->mnt == current->fs->root.mnt ||
1172                    flags & (MNT_FORCE | MNT_DETACH))
1173                        return -EINVAL;
1174
1175                /*
1176                 * probably don't strictly need the lock here if we examined
1177                 * all race cases, but it's a slowpath.
1178                 */
1179                br_write_lock(&vfsmount_lock);
1180                if (mnt_get_count(mnt) != 2) {
1181                        br_write_unlock(&vfsmount_lock);
1182                        return -EBUSY;
1183                }
1184                br_write_unlock(&vfsmount_lock);
1185
1186                if (!xchg(&mnt->mnt_expiry_mark, 1))
1187                        return -EAGAIN;
1188        }
1189
1190        /*
1191         * If we may have to abort operations to get out of this
1192         * mount, and they will themselves hold resources we must
1193         * allow the fs to do things. In the Unix tradition of
1194         * 'Gee thats tricky lets do it in userspace' the umount_begin
1195         * might fail to complete on the first run through as other tasks
1196         * must return, and the like. Thats for the mount program to worry
1197         * about for the moment.
1198         */
1199
1200        if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1201                sb->s_op->umount_begin(sb);
1202        }
1203
1204        /*
1205         * No sense to grab the lock for this test, but test itself looks
1206         * somewhat bogus. Suggestions for better replacement?
1207         * Ho-hum... In principle, we might treat that as umount + switch
1208         * to rootfs. GC would eventually take care of the old vfsmount.
1209         * Actually it makes sense, especially if rootfs would contain a
1210         * /reboot - static binary that would close all descriptors and
1211         * call reboot(9). Then init(8) could umount root and exec /reboot.
1212         */
1213        if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1214                /*
1215                 * Special case for "unmounting" root ...
1216                 * we just try to remount it readonly.
1217                 */
1218                down_write(&sb->s_umount);
1219                if (!(sb->s_flags & MS_RDONLY))
1220                        retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1221                up_write(&sb->s_umount);
1222                return retval;
1223        }
1224
1225        down_write(&namespace_sem);
1226        br_write_lock(&vfsmount_lock);
1227        event++;
1228
1229        if (!(flags & MNT_DETACH))
1230                shrink_submounts(mnt, &umount_list);
1231
1232        retval = -EBUSY;
1233        if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1234                if (!list_empty(&mnt->mnt_list))
1235                        umount_tree(mnt, 1, &umount_list);
1236                retval = 0;
1237        }
1238        br_write_unlock(&vfsmount_lock);
1239        up_write(&namespace_sem);
1240        release_mounts(&umount_list);
1241        return retval;
1242}
1243
1244/* 
1245 * Is the caller allowed to modify his namespace?
1246 */
1247static inline bool may_mount(void)
1248{
1249        return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1250}
1251
1252/*
1253 * Now umount can handle mount points as well as block devices.
1254 * This is important for filesystems which use unnamed block devices.
1255 *
1256 * We now support a flag for forced unmount like the other 'big iron'
1257 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1258 */
1259
1260SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1261{
1262        struct path path;
1263        struct mount *mnt;
1264        int retval;
1265        int lookup_flags = 0;
1266
1267        if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1268                return -EINVAL;
1269
1270        if (!may_mount())
1271                return -EPERM;
1272
1273        if (!(flags & UMOUNT_NOFOLLOW))
1274                lookup_flags |= LOOKUP_FOLLOW;
1275
1276        retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1277        if (retval)
1278                goto out;
1279        mnt = real_mount(path.mnt);
1280        retval = -EINVAL;
1281        if (path.dentry != path.mnt->mnt_root)
1282                goto dput_and_out;
1283        if (!check_mnt(mnt))
1284                goto dput_and_out;
1285
1286        retval = do_umount(mnt, flags);
1287dput_and_out:
1288        /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1289        dput(path.dentry);
1290        mntput_no_expire(mnt);
1291out:
1292        return retval;
1293}
1294
1295#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1296
1297/*
1298 *      The 2.0 compatible umount. No flags.
1299 */
1300SYSCALL_DEFINE1(oldumount, char __user *, name)
1301{
1302        return sys_umount(name, 0);
1303}
1304
1305#endif
1306
1307static bool mnt_ns_loop(struct path *path)
1308{
1309        /* Could bind mounting the mount namespace inode cause a
1310         * mount namespace loop?
1311         */
1312        struct inode *inode = path->dentry->d_inode;
1313        struct proc_inode *ei;
1314        struct mnt_namespace *mnt_ns;
1315
1316        if (!proc_ns_inode(inode))
1317                return false;
1318
1319        ei = PROC_I(inode);
1320        if (ei->ns_ops != &mntns_operations)
1321                return false;
1322
1323        mnt_ns = ei->ns;
1324        return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1325}
1326
1327struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1328                                        int flag)
1329{
1330        struct mount *res, *p, *q, *r;
1331        struct path path;
1332
1333        if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1334                return ERR_PTR(-EINVAL);
1335
1336        res = q = clone_mnt(mnt, dentry, flag);
1337        if (IS_ERR(q))
1338                return q;
1339
1340        q->mnt_mountpoint = mnt->mnt_mountpoint;
1341
1342        p = mnt;
1343        list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1344                struct mount *s;
1345                if (!is_subdir(r->mnt_mountpoint, dentry))
1346                        continue;
1347
1348                for (s = r; s; s = next_mnt(s, r)) {
1349                        if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1350                                s = skip_mnt_tree(s);
1351                                continue;
1352                        }
1353                        while (p != s->mnt_parent) {
1354                                p = p->mnt_parent;
1355                                q = q->mnt_parent;
1356                        }
1357                        p = s;
1358                        path.mnt = &q->mnt;
1359                        path.dentry = p->mnt_mountpoint;
1360                        q = clone_mnt(p, p->mnt.mnt_root, flag);
1361                        if (IS_ERR(q))
1362                                goto out;
1363                        br_write_lock(&vfsmount_lock);
1364                        list_add_tail(&q->mnt_list, &res->mnt_list);
1365                        attach_mnt(q, &path);
1366                        br_write_unlock(&vfsmount_lock);
1367                }
1368        }
1369        return res;
1370out:
1371        if (res) {
1372                LIST_HEAD(umount_list);
1373                br_write_lock(&vfsmount_lock);
1374                umount_tree(res, 0, &umount_list);
1375                br_write_unlock(&vfsmount_lock);
1376                release_mounts(&umount_list);
1377        }
1378        return q;
1379}
1380
1381/* Caller should check returned pointer for errors */
1382
1383struct vfsmount *collect_mounts(struct path *path)
1384{
1385        struct mount *tree;
1386        down_write(&namespace_sem);
1387        tree = copy_tree(real_mount(path->mnt), path->dentry,
1388                         CL_COPY_ALL | CL_PRIVATE);
1389        up_write(&namespace_sem);
1390        if (IS_ERR(tree))
1391                return NULL;
1392        return &tree->mnt;
1393}
1394
1395void drop_collected_mounts(struct vfsmount *mnt)
1396{
1397        LIST_HEAD(umount_list);
1398        down_write(&namespace_sem);
1399        br_write_lock(&vfsmount_lock);
1400        umount_tree(real_mount(mnt), 0, &umount_list);
1401        br_write_unlock(&vfsmount_lock);
1402        up_write(&namespace_sem);
1403        release_mounts(&umount_list);
1404}
1405
1406int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1407                   struct vfsmount *root)
1408{
1409        struct mount *mnt;
1410        int res = f(root, arg);
1411        if (res)
1412                return res;
1413        list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1414                res = f(&mnt->mnt, arg);
1415                if (res)
1416                        return res;
1417        }
1418        return 0;
1419}
1420
1421static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1422{
1423        struct mount *p;
1424
1425        for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1426                if (p->mnt_group_id && !IS_MNT_SHARED(p))
1427                        mnt_release_group_id(p);
1428        }
1429}
1430
1431static int invent_group_ids(struct mount *mnt, bool recurse)
1432{
1433        struct mount *p;
1434
1435        for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1436                if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1437                        int err = mnt_alloc_group_id(p);
1438                        if (err) {
1439                                cleanup_group_ids(mnt, p);
1440                                return err;
1441                        }
1442                }
1443        }
1444
1445        return 0;
1446}
1447
1448/*
1449 *  @source_mnt : mount tree to be attached
1450 *  @nd         : place the mount tree @source_mnt is attached
1451 *  @parent_nd  : if non-null, detach the source_mnt from its parent and
1452 *                 store the parent mount and mountpoint dentry.
1453 *                 (done when source_mnt is moved)
1454 *
1455 *  NOTE: in the table below explains the semantics when a source mount
1456 *  of a given type is attached to a destination mount of a given type.
1457 * ---------------------------------------------------------------------------
1458 * |         BIND MOUNT OPERATION                                            |
1459 * |**************************************************************************
1460 * | source-->| shared        |       private  |       slave    | unbindable |
1461 * | dest     |               |                |                |            |
1462 * |   |      |               |                |                |            |
1463 * |   v      |               |                |                |            |
1464 * |**************************************************************************
1465 * |  shared  | shared (++)   |     shared (+) |     shared(+++)|  invalid   |
1466 * |          |               |                |                |            |
1467 * |non-shared| shared (+)    |      private   |      slave (*) |  invalid   |
1468 * ***************************************************************************
1469 * A bind operation clones the source mount and mounts the clone on the
1470 * destination mount.
1471 *
1472 * (++)  the cloned mount is propagated to all the mounts in the propagation
1473 *       tree of the destination mount and the cloned mount is added to
1474 *       the peer group of the source mount.
1475 * (+)   the cloned mount is created under the destination mount and is marked
1476 *       as shared. The cloned mount is added to the peer group of the source
1477 *       mount.
1478 * (+++) the mount is propagated to all the mounts in the propagation tree
1479 *       of the destination mount and the cloned mount is made slave
1480 *       of the same master as that of the source mount. The cloned mount
1481 *       is marked as 'shared and slave'.
1482 * (*)   the cloned mount is made a slave of the same master as that of the
1483 *       source mount.
1484 *
1485 * ---------------------------------------------------------------------------
1486 * |                    MOVE MOUNT OPERATION                                 |
1487 * |**************************************************************************
1488 * | source-->| shared        |       private  |       slave    | unbindable |
1489 * | dest     |               |                |                |            |
1490 * |   |      |               |                |                |            |
1491 * |   v      |               |                |                |            |
1492 * |**************************************************************************
1493 * |  shared  | shared (+)    |     shared (+) |    shared(+++) |  invalid   |
1494 * |          |               |                |                |            |
1495 * |non-shared| shared (+*)   |      private   |    slave (*)   | unbindable |
1496 * ***************************************************************************
1497 *
1498 * (+)  the mount is moved to the destination. And is then propagated to
1499 *      all the mounts in the propagation tree of the destination mount.
1500 * (+*)  the mount is moved to the destination.
1501 * (+++)  the mount is moved to the destination and is then propagated to
1502 *      all the mounts belonging to the destination mount's propagation tree.
1503 *      the mount is marked as 'shared and slave'.
1504 * (*)  the mount continues to be a slave at the new location.
1505 *
1506 * if the source mount is a tree, the operations explained above is
1507 * applied to each mount in the tree.
1508 * Must be called without spinlocks held, since this function can sleep
1509 * in allocations.
1510 */
1511static int attach_recursive_mnt(struct mount *source_mnt,
1512                        struct path *path, struct path *parent_path)
1513{
1514        LIST_HEAD(tree_list);
1515        struct mount *dest_mnt = real_mount(path->mnt);
1516        struct dentry *dest_dentry = path->dentry;
1517        struct mount *child, *p;
1518        int err;
1519
1520        if (IS_MNT_SHARED(dest_mnt)) {
1521                err = invent_group_ids(source_mnt, true);
1522                if (err)
1523                        goto out;
1524        }
1525        err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1526        if (err)
1527                goto out_cleanup_ids;
1528
1529        br_write_lock(&vfsmount_lock);
1530
1531        if (IS_MNT_SHARED(dest_mnt)) {
1532                for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1533                        set_mnt_shared(p);
1534        }
1535        if (parent_path) {
1536                detach_mnt(source_mnt, parent_path);
1537                attach_mnt(source_mnt, path);
1538                touch_mnt_namespace(source_mnt->mnt_ns);
1539        } else {
1540                mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1541                commit_tree(source_mnt);
1542        }
1543
1544        list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1545                list_del_init(&child->mnt_hash);
1546                commit_tree(child);
1547        }
1548        br_write_unlock(&vfsmount_lock);
1549
1550        return 0;
1551
1552 out_cleanup_ids:
1553        if (IS_MNT_SHARED(dest_mnt))
1554                cleanup_group_ids(source_mnt, NULL);
1555 out:
1556        return err;
1557}
1558
1559static int lock_mount(struct path *path)
1560{
1561        struct vfsmount *mnt;
1562retry:
1563        mutex_lock(&path->dentry->d_inode->i_mutex);
1564        if (unlikely(cant_mount(path->dentry))) {
1565                mutex_unlock(&path->dentry->d_inode->i_mutex);
1566                return -ENOENT;
1567        }
1568        down_write(&namespace_sem);
1569        mnt = lookup_mnt(path);
1570        if (likely(!mnt))
1571                return 0;
1572        up_write(&namespace_sem);
1573        mutex_unlock(&path->dentry->d_inode->i_mutex);
1574        path_put(path);
1575        path->mnt = mnt;
1576        path->dentry = dget(mnt->mnt_root);
1577        goto retry;
1578}
1579
1580static void unlock_mount(struct path *path)
1581{
1582        up_write(&namespace_sem);
1583        mutex_unlock(&path->dentry->d_inode->i_mutex);
1584}
1585
1586static int graft_tree(struct mount *mnt, struct path *path)
1587{
1588        if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1589                return -EINVAL;
1590
1591        if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1592              S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
1593                return -ENOTDIR;
1594
1595        if (d_unlinked(path->dentry))
1596                return -ENOENT;
1597
1598        return attach_recursive_mnt(mnt, path, NULL);
1599}
1600
1601/*
1602 * Sanity check the flags to change_mnt_propagation.
1603 */
1604
1605static int flags_to_propagation_type(int flags)
1606{
1607        int type = flags & ~(MS_REC | MS_SILENT);
1608
1609        /* Fail if any non-propagation flags are set */
1610        if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1611                return 0;
1612        /* Only one propagation flag should be set */
1613        if (!is_power_of_2(type))
1614                return 0;
1615        return type;
1616}
1617
1618/*
1619 * recursively change the type of the mountpoint.
1620 */
1621static int do_change_type(struct path *path, int flag)
1622{
1623        struct mount *m;
1624        struct mount *mnt = real_mount(path->mnt);
1625        int recurse = flag & MS_REC;
1626        int type;
1627        int err = 0;
1628
1629        if (path->dentry != path->mnt->mnt_root)
1630                return -EINVAL;
1631
1632        type = flags_to_propagation_type(flag);
1633        if (!type)
1634                return -EINVAL;
1635
1636        down_write(&namespace_sem);
1637        if (type == MS_SHARED) {
1638                err = invent_group_ids(mnt, recurse);
1639                if (err)
1640                        goto out_unlock;
1641        }
1642
1643        br_write_lock(&vfsmount_lock);
1644        for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1645                change_mnt_propagation(m, type);
1646        br_write_unlock(&vfsmount_lock);
1647
1648 out_unlock:
1649        up_write(&namespace_sem);
1650        return err;
1651}
1652
1653/*
1654 * do loopback mount.
1655 */
1656static int do_loopback(struct path *path, const char *old_name,
1657                                int recurse)
1658{
1659        LIST_HEAD(umount_list);
1660        struct path old_path;
1661        struct mount *mnt = NULL, *old;
1662        int err;
1663        if (!old_name || !*old_name)
1664                return -EINVAL;
1665        err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1666        if (err)
1667                return err;
1668
1669        err = -EINVAL;
1670        if (mnt_ns_loop(&old_path))
1671                goto out; 
1672
1673        err = lock_mount(path);
1674        if (err)
1675                goto out;
1676
1677        old = real_mount(old_path.mnt);
1678
1679        err = -EINVAL;
1680        if (IS_MNT_UNBINDABLE(old))
1681                goto out2;
1682
1683        if (!check_mnt(real_mount(path->mnt)) || !check_mnt(old))
1684                goto out2;
1685
1686        if (recurse)
1687                mnt = copy_tree(old, old_path.dentry, 0);
1688        else
1689                mnt = clone_mnt(old, old_path.dentry, 0);
1690
1691        if (IS_ERR(mnt)) {
1692                err = PTR_ERR(mnt);
1693                goto out2;
1694        }
1695
1696        err = graft_tree(mnt, path);
1697        if (err) {
1698                br_write_lock(&vfsmount_lock);
1699                umount_tree(mnt, 0, &umount_list);
1700                br_write_unlock(&vfsmount_lock);
1701        }
1702out2:
1703        unlock_mount(path);
1704        release_mounts(&umount_list);
1705out:
1706        path_put(&old_path);
1707        return err;
1708}
1709
1710static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1711{
1712        int error = 0;
1713        int readonly_request = 0;
1714
1715        if (ms_flags & MS_RDONLY)
1716                readonly_request = 1;
1717        if (readonly_request == __mnt_is_readonly(mnt))
1718                return 0;
1719
1720        if (mnt->mnt_flags & MNT_LOCK_READONLY)
1721                return -EPERM;
1722
1723        if (readonly_request)
1724                error = mnt_make_readonly(real_mount(mnt));
1725        else
1726                __mnt_unmake_readonly(real_mount(mnt));
1727        return error;
1728}
1729
1730/*
1731 * change filesystem flags. dir should be a physical root of filesystem.
1732 * If you've mounted a non-root directory somewhere and want to do remount
1733 * on it - tough luck.
1734 */
1735static int do_remount(struct path *path, int flags, int mnt_flags,
1736                      void *data)
1737{
1738        int err;
1739        struct super_block *sb = path->mnt->mnt_sb;
1740        struct mount *mnt = real_mount(path->mnt);
1741
1742        if (!check_mnt(mnt))
1743                return -EINVAL;
1744
1745        if (path->dentry != path->mnt->mnt_root)
1746                return -EINVAL;
1747
1748        err = security_sb_remount(sb, data);
1749        if (err)
1750                return err;
1751
1752        down_write(&sb->s_umount);
1753        if (flags & MS_BIND)
1754                err = change_mount_flags(path->mnt, flags);
1755        else if (!capable(CAP_SYS_ADMIN))
1756                err = -EPERM;
1757        else
1758                err = do_remount_sb(sb, flags, data, 0);
1759        if (!err) {
1760                br_write_lock(&vfsmount_lock);
1761                mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK;
1762                mnt->mnt.mnt_flags = mnt_flags;
1763                br_write_unlock(&vfsmount_lock);
1764        }
1765        up_write(&sb->s_umount);
1766        if (!err) {
1767                br_write_lock(&vfsmount_lock);
1768                touch_mnt_namespace(mnt->mnt_ns);
1769                br_write_unlock(&vfsmount_lock);
1770        }
1771        return err;
1772}
1773
1774static inline int tree_contains_unbindable(struct mount *mnt)
1775{
1776        struct mount *p;
1777        for (p = mnt; p; p = next_mnt(p, mnt)) {
1778                if (IS_MNT_UNBINDABLE(p))
1779                        return 1;
1780        }
1781        return 0;
1782}
1783
1784static int do_move_mount(struct path *path, const char *old_name)
1785{
1786        struct path old_path, parent_path;
1787        struct mount *p;
1788        struct mount *old;
1789        int err;
1790        if (!old_name || !*old_name)
1791                return -EINVAL;
1792        err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1793        if (err)
1794                return err;
1795
1796        err = lock_mount(path);
1797        if (err < 0)
1798                goto out;
1799
1800        old = real_mount(old_path.mnt);
1801        p = real_mount(path->mnt);
1802
1803        err = -EINVAL;
1804        if (!check_mnt(p) || !check_mnt(old))
1805                goto out1;
1806
1807        if (d_unlinked(path->dentry))
1808                goto out1;
1809
1810        err = -EINVAL;
1811        if (old_path.dentry != old_path.mnt->mnt_root)
1812                goto out1;
1813
1814        if (!mnt_has_parent(old))
1815                goto out1;
1816
1817        if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1818              S_ISDIR(old_path.dentry->d_inode->i_mode))
1819                goto out1;
1820        /*
1821         * Don't move a mount residing in a shared parent.
1822         */
1823        if (IS_MNT_SHARED(old->mnt_parent))
1824                goto out1;
1825        /*
1826         * Don't move a mount tree containing unbindable mounts to a destination
1827         * mount which is shared.
1828         */
1829        if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
1830                goto out1;
1831        err = -ELOOP;
1832        for (; mnt_has_parent(p); p = p->mnt_parent)
1833                if (p == old)
1834                        goto out1;
1835
1836        err = attach_recursive_mnt(old, path, &parent_path);
1837        if (err)
1838                goto out1;
1839
1840        /* if the mount is moved, it should no longer be expire
1841         * automatically */
1842        list_del_init(&old->mnt_expire);
1843out1:
1844        unlock_mount(path);
1845out:
1846        if (!err)
1847                path_put(&parent_path);
1848        path_put(&old_path);
1849        return err;
1850}
1851
1852static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1853{
1854        int err;
1855        const char *subtype = strchr(fstype, '.');
1856        if (subtype) {
1857                subtype++;
1858                err = -EINVAL;
1859                if (!subtype[0])
1860                        goto err;
1861        } else
1862                subtype = "";
1863
1864        mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1865        err = -ENOMEM;
1866        if (!mnt->mnt_sb->s_subtype)
1867                goto err;
1868        return mnt;
1869
1870 err:
1871        mntput(mnt);
1872        return ERR_PTR(err);
1873}
1874
1875/*
1876 * add a mount into a namespace's mount tree
1877 */
1878static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
1879{
1880        int err;
1881
1882        mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1883
1884        err = lock_mount(path);
1885        if (err)
1886                return err;
1887
1888        err = -EINVAL;
1889        if (unlikely(!check_mnt(real_mount(path->mnt)))) {
1890                /* that's acceptable only for automounts done in private ns */
1891                if (!(mnt_flags & MNT_SHRINKABLE))
1892                        goto unlock;
1893                /* ... and for those we'd better have mountpoint still alive */
1894                if (!real_mount(path->mnt)->mnt_ns)
1895                        goto unlock;
1896        }
1897
1898        /* Refuse the same filesystem on the same mount point */
1899        err = -EBUSY;
1900        if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
1901            path->mnt->mnt_root == path->dentry)
1902                goto unlock;
1903
1904        err = -EINVAL;
1905        if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
1906                goto unlock;
1907
1908        newmnt->mnt.mnt_flags = mnt_flags;
1909        err = graft_tree(newmnt, path);
1910
1911unlock:
1912        unlock_mount(path);
1913        return err;
1914}
1915
1916/*
1917 * create a new mount for userspace and request it to be added into the
1918 * namespace's tree
1919 */
1920static int do_new_mount(struct path *path, const char *fstype, int flags,
1921                        int mnt_flags, const char *name, void *data)
1922{
1923        struct file_system_type *type;
1924        struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
1925        struct vfsmount *mnt;
1926        int err;
1927
1928        if (!fstype)
1929                return -EINVAL;
1930
1931        type = get_fs_type(fstype);
1932        if (!type)
1933                return -ENODEV;
1934
1935        if (user_ns != &init_user_ns) {
1936                if (!(type->fs_flags & FS_USERNS_MOUNT)) {
1937                        put_filesystem(type);
1938                        return -EPERM;
1939                }
1940                /* Only in special cases allow devices from mounts
1941                 * created outside the initial user namespace.
1942                 */
1943                if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
1944                        flags |= MS_NODEV;
1945                        mnt_flags |= MNT_NODEV;
1946                }
1947        }
1948
1949        mnt = vfs_kern_mount(type, flags, name, data);
1950        if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1951            !mnt->mnt_sb->s_subtype)
1952                mnt = fs_set_subtype(mnt, fstype);
1953
1954        put_filesystem(type);
1955        if (IS_ERR(mnt))
1956                return PTR_ERR(mnt);
1957
1958        err = do_add_mount(real_mount(mnt), path, mnt_flags);
1959        if (err)
1960                mntput(mnt);
1961        return err;
1962}
1963
1964int finish_automount(struct vfsmount *m, struct path *path)
1965{
1966        struct mount *mnt = real_mount(m);
1967        int err;
1968        /* The new mount record should have at least 2 refs to prevent it being
1969         * expired before we get a chance to add it
1970         */
1971        BUG_ON(mnt_get_count(mnt) < 2);
1972
1973        if (m->mnt_sb == path->mnt->mnt_sb &&
1974            m->mnt_root == path->dentry) {
1975                err = -ELOOP;
1976                goto fail;
1977        }
1978
1979        err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
1980        if (!err)
1981                return 0;
1982fail:
1983        /* remove m from any expiration list it may be on */
1984        if (!list_empty(&mnt->mnt_expire)) {
1985                down_write(&namespace_sem);
1986                br_write_lock(&vfsmount_lock);
1987                list_del_init(&mnt->mnt_expire);
1988                br_write_unlock(&vfsmount_lock);
1989                up_write(&namespace_sem);
1990        }
1991        mntput(m);
1992        mntput(m);
1993        return err;
1994}
1995
1996/**
1997 * mnt_set_expiry - Put a mount on an expiration list
1998 * @mnt: The mount to list.
1999 * @expiry_list: The list to add the mount to.
2000 */
2001void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2002{
2003        down_write(&namespace_sem);
2004        br_write_lock(&vfsmount_lock);
2005
2006        list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2007
2008        br_write_unlock(&vfsmount_lock);
2009        up_write(&namespace_sem);
2010}
2011EXPORT_SYMBOL(mnt_set_expiry);
2012
2013/*
2014 * process a list of expirable mountpoints with the intent of discarding any
2015 * mountpoints that aren't in use and haven't been touched since last we came
2016 * here
2017 */
2018void mark_mounts_for_expiry(struct list_head *mounts)
2019{
2020        struct mount *mnt, *next;
2021        LIST_HEAD(graveyard);
2022        LIST_HEAD(umounts);
2023
2024        if (list_empty(mounts))
2025                return;
2026
2027        down_write(&namespace_sem);
2028        br_write_lock(&vfsmount_lock);
2029
2030        /* extract from the expiration list every vfsmount that matches the
2031         * following criteria:
2032         * - only referenced by its parent vfsmount
2033         * - still marked for expiry (marked on the last call here; marks are
2034         *   cleared by mntput())
2035         */
2036        list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2037                if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2038                        propagate_mount_busy(mnt, 1))
2039                        continue;
2040                list_move(&mnt->mnt_expire, &graveyard);
2041        }
2042        while (!list_empty(&graveyard)) {
2043                mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2044                touch_mnt_namespace(mnt->mnt_ns);
2045                umount_tree(mnt, 1, &umounts);
2046        }
2047        br_write_unlock(&vfsmount_lock);
2048        up_write(&namespace_sem);
2049
2050        release_mounts(&umounts);
2051}
2052
2053EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2054
2055/*
2056 * Ripoff of 'select_parent()'
2057 *
2058 * search the list of submounts for a given mountpoint, and move any
2059 * shrinkable submounts to the 'graveyard' list.
2060 */
2061static int select_submounts(struct mount *parent, struct list_head *graveyard)
2062{
2063        struct mount *this_parent = parent;
2064        struct list_head *next;
2065        int found = 0;
2066
2067repeat:
2068        next = this_parent->mnt_mounts.next;
2069resume:
2070        while (next != &this_parent->mnt_mounts) {
2071                struct list_head *tmp = next;
2072                struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2073
2074                next = tmp->next;
2075                if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2076                        continue;
2077                /*
2078                 * Descend a level if the d_mounts list is non-empty.
2079                 */
2080                if (!list_empty(&mnt->mnt_mounts)) {
2081                        this_parent = mnt;
2082                        goto repeat;
2083                }
2084
2085                if (!propagate_mount_busy(mnt, 1)) {
2086                        list_move_tail(&mnt->mnt_expire, graveyard);
2087                        found++;
2088                }
2089        }
2090        /*
2091         * All done at this level ... ascend and resume the search
2092         */
2093        if (this_parent != parent) {
2094                next = this_parent->mnt_child.next;
2095                this_parent = this_parent->mnt_parent;
2096                goto resume;
2097        }
2098        return found;
2099}
2100
2101/*
2102 * process a list of expirable mountpoints with the intent of discarding any
2103 * submounts of a specific parent mountpoint
2104 *
2105 * vfsmount_lock must be held for write
2106 */
2107static void shrink_submounts(struct mount *mnt, struct list_head *umounts)
2108{
2109        LIST_HEAD(graveyard);
2110        struct mount *m;
2111
2112        /* extract submounts of 'mountpoint' from the expiration list */
2113        while (select_submounts(mnt, &graveyard)) {
2114                while (!list_empty(&graveyard)) {
2115                        m = list_first_entry(&graveyard, struct mount,
2116                                                mnt_expire);
2117                        touch_mnt_namespace(m->mnt_ns);
2118                        umount_tree(m, 1, umounts);
2119                }
2120        }
2121}
2122
2123/*
2124 * Some copy_from_user() implementations do not return the exact number of
2125 * bytes remaining to copy on a fault.  But copy_mount_options() requires that.
2126 * Note that this function differs from copy_from_user() in that it will oops
2127 * on bad values of `to', rather than returning a short copy.
2128 */
2129static long exact_copy_from_user(void *to, const void __user * from,
2130                                 unsigned long n)
2131{
2132        char *t = to;
2133        const char __user *f = from;
2134        char c;
2135
2136        if (!access_ok(VERIFY_READ, from, n))
2137                return n;
2138
2139        while (n) {
2140                if (__get_user(c, f)) {
2141                        memset(t, 0, n);
2142                        break;
2143                }
2144                *t++ = c;
2145                f++;
2146                n--;
2147        }
2148        return n;
2149}
2150
2151int copy_mount_options(const void __user * data, unsigned long *where)
2152{
2153        int i;
2154        unsigned long page;
2155        unsigned long size;
2156
2157        *where = 0;
2158        if (!data)
2159                return 0;
2160
2161        if (!(page = __get_free_page(GFP_KERNEL)))
2162                return -ENOMEM;
2163
2164        /* We only care that *some* data at the address the user
2165         * gave us is valid.  Just in case, we'll zero
2166         * the remainder of the page.
2167         */
2168        /* copy_from_user cannot cross TASK_SIZE ! */
2169        size = TASK_SIZE - (unsigned long)data;
2170        if (size > PAGE_SIZE)
2171                size = PAGE_SIZE;
2172
2173        i = size - exact_copy_from_user((void *)page, data, size);
2174        if (!i) {
2175                free_page(page);
2176                return -EFAULT;
2177        }
2178        if (i != PAGE_SIZE)
2179                memset((char *)page + i, 0, PAGE_SIZE - i);
2180        *where = page;
2181        return 0;
2182}
2183
2184int copy_mount_string(const void __user *data, char **where)
2185{
2186        char *tmp;
2187
2188        if (!data) {
2189                *where = NULL;
2190                return 0;
2191        }
2192
2193        tmp = strndup_user(data, PAGE_SIZE);
2194        if (IS_ERR(tmp))
2195                return PTR_ERR(tmp);
2196
2197        *where = tmp;
2198        return 0;
2199}
2200
2201/*
2202 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2203 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2204 *
2205 * data is a (void *) that can point to any structure up to
2206 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2207 * information (or be NULL).
2208 *
2209 * Pre-0.97 versions of mount() didn't have a flags word.
2210 * When the flags word was introduced its top half was required
2211 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2212 * Therefore, if this magic number is present, it carries no information
2213 * and must be discarded.
2214 */
2215long do_mount(const char *dev_name, const char *dir_name,
2216                const char *type_page, unsigned long flags, void *data_page)
2217{
2218        struct path path;
2219        int retval = 0;
2220        int mnt_flags = 0;
2221
2222        /* Discard magic */
2223        if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2224                flags &= ~MS_MGC_MSK;
2225
2226        /* Basic sanity checks */
2227
2228        if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2229                return -EINVAL;
2230
2231        if (data_page)
2232                ((char *)data_page)[PAGE_SIZE - 1] = 0;
2233
2234        /* ... and get the mountpoint */
2235        retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2236        if (retval)
2237                return retval;
2238
2239        retval = security_sb_mount(dev_name, &path,
2240                                   type_page, flags, data_page);
2241        if (retval)
2242                goto dput_out;
2243
2244        if (!may_mount())
2245                return -EPERM;
2246
2247        /* Default to relatime unless overriden */
2248        if (!(flags & MS_NOATIME))
2249                mnt_flags |= MNT_RELATIME;
2250
2251        /* Separate the per-mountpoint flags */
2252        if (flags & MS_NOSUID)
2253                mnt_flags |= MNT_NOSUID;
2254        if (flags & MS_NODEV)
2255                mnt_flags |= MNT_NODEV;
2256        if (flags & MS_NOEXEC)
2257                mnt_flags |= MNT_NOEXEC;
2258        if (flags & MS_NOATIME)
2259                mnt_flags |= MNT_NOATIME;
2260        if (flags & MS_NODIRATIME)
2261                mnt_flags |= MNT_NODIRATIME;
2262        if (flags & MS_STRICTATIME)
2263                mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2264        if (flags & MS_RDONLY)
2265                mnt_flags |= MNT_READONLY;
2266
2267        flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2268                   MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2269                   MS_STRICTATIME);
2270
2271        if (flags & MS_REMOUNT)
2272                retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2273                                    data_page);
2274        else if (flags & MS_BIND)
2275                retval = do_loopback(&path, dev_name, flags & MS_REC);
2276        else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2277                retval = do_change_type(&path, flags);
2278        else if (flags & MS_MOVE)
2279                retval = do_move_mount(&path, dev_name);
2280        else
2281                retval = do_new_mount(&path, type_page, flags, mnt_flags,
2282                                      dev_name, data_page);
2283dput_out:
2284        path_put(&path);
2285        return retval;
2286}
2287
2288static void free_mnt_ns(struct mnt_namespace *ns)
2289{
2290        proc_free_inum(ns->proc_inum);
2291        put_user_ns(ns->user_ns);
2292        kfree(ns);
2293}
2294
2295/*
2296 * Assign a sequence number so we can detect when we attempt to bind
2297 * mount a reference to an older mount namespace into the current
2298 * mount namespace, preventing reference counting loops.  A 64bit
2299 * number incrementing at 10Ghz will take 12,427 years to wrap which
2300 * is effectively never, so we can ignore the possibility.
2301 */
2302static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2303
2304static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2305{
2306        struct mnt_namespace *new_ns;
2307        int ret;
2308
2309        new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2310        if (!new_ns)
2311                return ERR_PTR(-ENOMEM);
2312        ret = proc_alloc_inum(&new_ns->proc_inum);
2313        if (ret) {
2314                kfree(new_ns);
2315                return ERR_PTR(ret);
2316        }
2317        new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2318        atomic_set(&new_ns->count, 1);
2319        new_ns->root = NULL;
2320        INIT_LIST_HEAD(&new_ns->list);
2321        init_waitqueue_head(&new_ns->poll);
2322        new_ns->event = 0;
2323        new_ns->user_ns = get_user_ns(user_ns);
2324        return new_ns;
2325}
2326
2327/*
2328 * Allocate a new namespace structure and populate it with contents
2329 * copied from the namespace of the passed in task structure.
2330 */
2331static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2332                struct user_namespace *user_ns, struct fs_struct *fs)
2333{
2334        struct mnt_namespace *new_ns;
2335        struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2336        struct mount *p, *q;
2337        struct mount *old = mnt_ns->root;
2338        struct mount *new;
2339        int copy_flags;
2340
2341        new_ns = alloc_mnt_ns(user_ns);
2342        if (IS_ERR(new_ns))
2343                return new_ns;
2344
2345        down_write(&namespace_sem);
2346        /* First pass: copy the tree topology */
2347        copy_flags = CL_COPY_ALL | CL_EXPIRE;
2348        if (user_ns != mnt_ns->user_ns)
2349                copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2350        new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2351        if (IS_ERR(new)) {
2352                up_write(&namespace_sem);
2353                free_mnt_ns(new_ns);
2354                return ERR_CAST(new);
2355        }
2356        new_ns->root = new;
2357        br_write_lock(&vfsmount_lock);
2358        list_add_tail(&new_ns->list, &new->mnt_list);
2359        br_write_unlock(&vfsmount_lock);
2360
2361        /*
2362         * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2363         * as belonging to new namespace.  We have already acquired a private
2364         * fs_struct, so tsk->fs->lock is not needed.
2365         */
2366        p = old;
2367        q = new;
2368        while (p) {
2369                q->mnt_ns = new_ns;
2370                if (fs) {
2371                        if (&p->mnt == fs->root.mnt) {
2372                                fs->root.mnt = mntget(&q->mnt);
2373                                rootmnt = &p->mnt;
2374                        }
2375                        if (&p->mnt == fs->pwd.mnt) {
2376                                fs->pwd.mnt = mntget(&q->mnt);
2377                                pwdmnt = &p->mnt;
2378                        }
2379                }
2380                p = next_mnt(p, old);
2381                q = next_mnt(q, new);
2382        }
2383        up_write(&namespace_sem);
2384
2385        if (rootmnt)
2386                mntput(rootmnt);
2387        if (pwdmnt)
2388                mntput(pwdmnt);
2389
2390        return new_ns;
2391}
2392
2393struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2394                struct user_namespace *user_ns, struct fs_struct *new_fs)
2395{
2396        struct mnt_namespace *new_ns;
2397
2398        BUG_ON(!ns);
2399        get_mnt_ns(ns);
2400
2401        if (!(flags & CLONE_NEWNS))
2402                return ns;
2403
2404        new_ns = dup_mnt_ns(ns, user_ns, new_fs);
2405
2406        put_mnt_ns(ns);
2407        return new_ns;
2408}
2409
2410/**
2411 * create_mnt_ns - creates a private namespace and adds a root filesystem
2412 * @mnt: pointer to the new root filesystem mountpoint
2413 */
2414static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2415{
2416        struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2417        if (!IS_ERR(new_ns)) {
2418                struct mount *mnt = real_mount(m);
2419                mnt->mnt_ns = new_ns;
2420                new_ns->root = mnt;
2421                list_add(&new_ns->list, &mnt->mnt_list);
2422        } else {
2423                mntput(m);
2424        }
2425        return new_ns;
2426}
2427
2428struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2429{
2430        struct mnt_namespace *ns;
2431        struct super_block *s;
2432        struct path path;
2433        int err;
2434
2435        ns = create_mnt_ns(mnt);
2436        if (IS_ERR(ns))
2437                return ERR_CAST(ns);
2438
2439        err = vfs_path_lookup(mnt->mnt_root, mnt,
2440                        name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2441
2442        put_mnt_ns(ns);
2443
2444        if (err)
2445                return ERR_PTR(err);
2446
2447        /* trade a vfsmount reference for active sb one */
2448        s = path.mnt->mnt_sb;
2449        atomic_inc(&s->s_active);
2450        mntput(path.mnt);
2451        /* lock the sucker */
2452        down_write(&s->s_umount);
2453        /* ... and return the root of (sub)tree on it */
2454        return path.dentry;
2455}
2456EXPORT_SYMBOL(mount_subtree);
2457
2458SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2459                char __user *, type, unsigned long, flags, void __user *, data)
2460{
2461        int ret;
2462        char *kernel_type;
2463        struct filename *kernel_dir;
2464        char *kernel_dev;
2465        unsigned long data_page;
2466
2467        ret = copy_mount_string(type, &kernel_type);
2468        if (ret < 0)
2469                goto out_type;
2470
2471        kernel_dir = getname(dir_name);
2472        if (IS_ERR(kernel_dir)) {
2473                ret = PTR_ERR(kernel_dir);
2474                goto out_dir;
2475        }
2476
2477        ret = copy_mount_string(dev_name, &kernel_dev);
2478        if (ret < 0)
2479                goto out_dev;
2480
2481        ret = copy_mount_options(data, &data_page);
2482        if (ret < 0)
2483                goto out_data;
2484
2485        ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags,
2486                (void *) data_page);
2487
2488        free_page(data_page);
2489out_data:
2490        kfree(kernel_dev);
2491out_dev:
2492        putname(kernel_dir);
2493out_dir:
2494        kfree(kernel_type);
2495out_type:
2496        return ret;
2497}
2498
2499/*
2500 * Return true if path is reachable from root
2501 *
2502 * namespace_sem or vfsmount_lock is held
2503 */
2504bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2505                         const struct path *root)
2506{
2507        while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2508                dentry = mnt->mnt_mountpoint;
2509                mnt = mnt->mnt_parent;
2510        }
2511        return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2512}
2513
2514int path_is_under(struct path *path1, struct path *path2)
2515{
2516        int res;
2517        br_read_lock(&vfsmount_lock);
2518        res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2519        br_read_unlock(&vfsmount_lock);
2520        return res;
2521}
2522EXPORT_SYMBOL(path_is_under);
2523
2524/*
2525 * pivot_root Semantics:
2526 * Moves the root file system of the current process to the directory put_old,
2527 * makes new_root as the new root file system of the current process, and sets
2528 * root/cwd of all processes which had them on the current root to new_root.
2529 *
2530 * Restrictions:
2531 * The new_root and put_old must be directories, and  must not be on the
2532 * same file  system as the current process root. The put_old  must  be
2533 * underneath new_root,  i.e. adding a non-zero number of /.. to the string
2534 * pointed to by put_old must yield the same directory as new_root. No other
2535 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2536 *
2537 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2538 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2539 * in this situation.
2540 *
2541 * Notes:
2542 *  - we don't move root/cwd if they are not at the root (reason: if something
2543 *    cared enough to change them, it's probably wrong to force them elsewhere)
2544 *  - it's okay to pick a root that isn't the root of a file system, e.g.
2545 *    /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2546 *    though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2547 *    first.
2548 */
2549SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2550                const char __user *, put_old)
2551{
2552        struct path new, old, parent_path, root_parent, root;
2553        struct mount *new_mnt, *root_mnt;
2554        int error;
2555
2556        if (!may_mount())
2557                return -EPERM;
2558
2559        error = user_path_dir(new_root, &new);
2560        if (error)
2561                goto out0;
2562
2563        error = user_path_dir(put_old, &old);
2564        if (error)
2565                goto out1;
2566
2567        error = security_sb_pivotroot(&old, &new);
2568        if (error)
2569                goto out2;
2570
2571        get_fs_root(current->fs, &root);
2572        error = lock_mount(&old);
2573        if (error)
2574                goto out3;
2575
2576        error = -EINVAL;
2577        new_mnt = real_mount(new.mnt);
2578        root_mnt = real_mount(root.mnt);
2579        if (IS_MNT_SHARED(real_mount(old.mnt)) ||
2580                IS_MNT_SHARED(new_mnt->mnt_parent) ||
2581                IS_MNT_SHARED(root_mnt->mnt_parent))
2582                goto out4;
2583        if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
2584                goto out4;
2585        error = -ENOENT;
2586        if (d_unlinked(new.dentry))
2587                goto out4;
2588        if (d_unlinked(old.dentry))
2589                goto out4;
2590        error = -EBUSY;
2591        if (new.mnt == root.mnt ||
2592            old.mnt == root.mnt)
2593                goto out4; /* loop, on the same file system  */
2594        error = -EINVAL;
2595        if (root.mnt->mnt_root != root.dentry)
2596                goto out4; /* not a mountpoint */
2597        if (!mnt_has_parent(root_mnt))
2598                goto out4; /* not attached */
2599        if (new.mnt->mnt_root != new.dentry)
2600                goto out4; /* not a mountpoint */
2601        if (!mnt_has_parent(new_mnt))
2602                goto out4; /* not attached */
2603        /* make sure we can reach put_old from new_root */
2604        if (!is_path_reachable(real_mount(old.mnt), old.dentry, &new))
2605                goto out4;
2606        br_write_lock(&vfsmount_lock);
2607        detach_mnt(new_mnt, &parent_path);
2608        detach_mnt(root_mnt, &root_parent);
2609        /* mount old root on put_old */
2610        attach_mnt(root_mnt, &old);
2611        /* mount new_root on / */
2612        attach_mnt(new_mnt, &root_parent);
2613        touch_mnt_namespace(current->nsproxy->mnt_ns);
2614        br_write_unlock(&vfsmount_lock);
2615        chroot_fs_refs(&root, &new);
2616        error = 0;
2617out4:
2618        unlock_mount(&old);
2619        if (!error) {
2620                path_put(&root_parent);
2621                path_put(&parent_path);
2622        }
2623out3:
2624        path_put(&root);
2625out2:
2626        path_put(&old);
2627out1:
2628        path_put(&new);
2629out0:
2630        return error;
2631}
2632
2633static void __init init_mount_tree(void)
2634{
2635        struct vfsmount *mnt;
2636        struct mnt_namespace *ns;
2637        struct path root;
2638        struct file_system_type *type;
2639
2640        type = get_fs_type("rootfs");
2641        if (!type)
2642                panic("Can't find rootfs type");
2643        mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
2644        put_filesystem(type);
2645        if (IS_ERR(mnt))
2646                panic("Can't create rootfs");
2647
2648        ns = create_mnt_ns(mnt);
2649        if (IS_ERR(ns))
2650                panic("Can't allocate initial namespace");
2651
2652        init_task.nsproxy->mnt_ns = ns;
2653        get_mnt_ns(ns);
2654
2655        root.mnt = mnt;
2656        root.dentry = mnt->mnt_root;
2657
2658        set_fs_pwd(current->fs, &root);
2659        set_fs_root(current->fs, &root);
2660}
2661
2662void __init mnt_init(void)
2663{
2664        unsigned u;
2665        int err;
2666
2667        init_rwsem(&namespace_sem);
2668
2669        mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
2670                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2671
2672        mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2673
2674        if (!mount_hashtable)
2675                panic("Failed to allocate mount hash table\n");
2676
2677        printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
2678
2679        for (u = 0; u < HASH_SIZE; u++)
2680                INIT_LIST_HEAD(&mount_hashtable[u]);
2681
2682        br_lock_init(&vfsmount_lock);
2683
2684        err = sysfs_init();
2685        if (err)
2686                printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2687                        __func__, err);
2688        fs_kobj = kobject_create_and_add("fs", NULL);
2689        if (!fs_kobj)
2690                printk(KERN_WARNING "%s: kobj create error\n", __func__);
2691        init_rootfs();
2692        init_mount_tree();
2693}
2694
2695void put_mnt_ns(struct mnt_namespace *ns)
2696{
2697        LIST_HEAD(umount_list);
2698
2699        if (!atomic_dec_and_test(&ns->count))
2700                return;
2701        down_write(&namespace_sem);
2702        br_write_lock(&vfsmount_lock);
2703        umount_tree(ns->root, 0, &umount_list);
2704        br_write_unlock(&vfsmount_lock);
2705        up_write(&namespace_sem);
2706        release_mounts(&umount_list);
2707        free_mnt_ns(ns);
2708}
2709
2710struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2711{
2712        struct vfsmount *mnt;
2713        mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2714        if (!IS_ERR(mnt)) {
2715                /*
2716                 * it is a longterm mount, don't release mnt until
2717                 * we unmount before file sys is unregistered
2718                */
2719                real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
2720        }
2721        return mnt;
2722}
2723EXPORT_SYMBOL_GPL(kern_mount_data);
2724
2725void kern_unmount(struct vfsmount *mnt)
2726{
2727        /* release long term mount so mount point can be released */
2728        if (!IS_ERR_OR_NULL(mnt)) {
2729                br_write_lock(&vfsmount_lock);
2730                real_mount(mnt)->mnt_ns = NULL;
2731                br_write_unlock(&vfsmount_lock);
2732                mntput(mnt);
2733        }
2734}
2735EXPORT_SYMBOL(kern_unmount);
2736
2737bool our_mnt(struct vfsmount *mnt)
2738{
2739        return check_mnt(real_mount(mnt));
2740}
2741
2742bool current_chrooted(void)
2743{
2744        /* Does the current process have a non-standard root */
2745        struct path ns_root;
2746        struct path fs_root;
2747        bool chrooted;
2748
2749        /* Find the namespace root */
2750        ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
2751        ns_root.dentry = ns_root.mnt->mnt_root;
2752        path_get(&ns_root);
2753        while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
2754                ;
2755
2756        get_fs_root(current->fs, &fs_root);
2757
2758        chrooted = !path_equal(&fs_root, &ns_root);
2759
2760        path_put(&fs_root);
2761        path_put(&ns_root);
2762
2763        return chrooted;
2764}
2765
2766void update_mnt_policy(struct user_namespace *userns)
2767{
2768        struct mnt_namespace *ns = current->nsproxy->mnt_ns;
2769        struct mount *mnt;
2770
2771        down_read(&namespace_sem);
2772        list_for_each_entry(mnt, &ns->list, mnt_list) {
2773                switch (mnt->mnt.mnt_sb->s_magic) {
2774                case SYSFS_MAGIC:
2775                        userns->may_mount_sysfs = true;
2776                        break;
2777                case PROC_SUPER_MAGIC:
2778                        userns->may_mount_proc = true;
2779                        break;
2780                }
2781                if (userns->may_mount_sysfs && userns->may_mount_proc)
2782                        break;
2783        }
2784        up_read(&namespace_sem);
2785}
2786
2787static void *mntns_get(struct task_struct *task)
2788{
2789        struct mnt_namespace *ns = NULL;
2790        struct nsproxy *nsproxy;
2791
2792        rcu_read_lock();
2793        nsproxy = task_nsproxy(task);
2794        if (nsproxy) {
2795                ns = nsproxy->mnt_ns;
2796                get_mnt_ns(ns);
2797        }
2798        rcu_read_unlock();
2799
2800        return ns;
2801}
2802
2803static void mntns_put(void *ns)
2804{
2805        put_mnt_ns(ns);
2806}
2807
2808static int mntns_install(struct nsproxy *nsproxy, void *ns)
2809{
2810        struct fs_struct *fs = current->fs;
2811        struct mnt_namespace *mnt_ns = ns;
2812        struct path root;
2813
2814        if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
2815            !nsown_capable(CAP_SYS_CHROOT) ||
2816            !nsown_capable(CAP_SYS_ADMIN))
2817                return -EPERM;
2818
2819        if (fs->users != 1)
2820                return -EINVAL;
2821
2822        get_mnt_ns(mnt_ns);
2823        put_mnt_ns(nsproxy->mnt_ns);
2824        nsproxy->mnt_ns = mnt_ns;
2825
2826        /* Find the root */
2827        root.mnt    = &mnt_ns->root->mnt;
2828        root.dentry = mnt_ns->root->mnt.mnt_root;
2829        path_get(&root);
2830        while(d_mountpoint(root.dentry) && follow_down_one(&root))
2831                ;
2832
2833        /* Update the pwd and root */
2834        set_fs_pwd(fs, &root);
2835        set_fs_root(fs, &root);
2836
2837        path_put(&root);
2838        return 0;
2839}
2840
2841static unsigned int mntns_inum(void *ns)
2842{
2843        struct mnt_namespace *mnt_ns = ns;
2844        return mnt_ns->proc_inum;
2845}
2846
2847const struct proc_ns_operations mntns_operations = {
2848        .name           = "mnt",
2849        .type           = CLONE_NEWNS,
2850        .get            = mntns_get,
2851        .put            = mntns_put,
2852        .install        = mntns_install,
2853        .inum           = mntns_inum,
2854};
2855
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.