linux/fs/namespace.c
<<
>>
Prefs
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *  linux/fs/namespace.c
   4 *
   5 * (C) Copyright Al Viro 2000, 2001
   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/cred.h>
  19#include <linux/idr.h>
  20#include <linux/init.h>         /* init_rootfs */
  21#include <linux/fs_struct.h>    /* get_fs_root et.al. */
  22#include <linux/fsnotify.h>     /* fsnotify_vfsmount_delete */
  23#include <linux/file.h>
  24#include <linux/uaccess.h>
  25#include <linux/proc_ns.h>
  26#include <linux/magic.h>
  27#include <linux/memblock.h>
  28#include <linux/proc_fs.h>
  29#include <linux/task_work.h>
  30#include <linux/sched/task.h>
  31#include <uapi/linux/mount.h>
  32#include <linux/fs_context.h>
  33#include <linux/shmem_fs.h>
  34
  35#include "pnode.h"
  36#include "internal.h"
  37
  38/* Maximum number of mounts in a mount namespace */
  39unsigned int sysctl_mount_max __read_mostly = 100000;
  40
  41static unsigned int m_hash_mask __read_mostly;
  42static unsigned int m_hash_shift __read_mostly;
  43static unsigned int mp_hash_mask __read_mostly;
  44static unsigned int mp_hash_shift __read_mostly;
  45
  46static __initdata unsigned long mhash_entries;
  47static int __init set_mhash_entries(char *str)
  48{
  49        if (!str)
  50                return 0;
  51        mhash_entries = simple_strtoul(str, &str, 0);
  52        return 1;
  53}
  54__setup("mhash_entries=", set_mhash_entries);
  55
  56static __initdata unsigned long mphash_entries;
  57static int __init set_mphash_entries(char *str)
  58{
  59        if (!str)
  60                return 0;
  61        mphash_entries = simple_strtoul(str, &str, 0);
  62        return 1;
  63}
  64__setup("mphash_entries=", set_mphash_entries);
  65
  66static u64 event;
  67static DEFINE_IDA(mnt_id_ida);
  68static DEFINE_IDA(mnt_group_ida);
  69
  70static struct hlist_head *mount_hashtable __read_mostly;
  71static struct hlist_head *mountpoint_hashtable __read_mostly;
  72static struct kmem_cache *mnt_cache __read_mostly;
  73static DECLARE_RWSEM(namespace_sem);
  74static HLIST_HEAD(unmounted);   /* protected by namespace_sem */
  75static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */
  76
  77struct mount_kattr {
  78        unsigned int attr_set;
  79        unsigned int attr_clr;
  80        unsigned int propagation;
  81        unsigned int lookup_flags;
  82        bool recurse;
  83        struct user_namespace *mnt_userns;
  84};
  85
  86/* /sys/fs */
  87struct kobject *fs_kobj;
  88EXPORT_SYMBOL_GPL(fs_kobj);
  89
  90/*
  91 * vfsmount lock may be taken for read to prevent changes to the
  92 * vfsmount hash, ie. during mountpoint lookups or walking back
  93 * up the tree.
  94 *
  95 * It should be taken for write in all cases where the vfsmount
  96 * tree or hash is modified or when a vfsmount structure is modified.
  97 */
  98__cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
  99
 100static inline void lock_mount_hash(void)
 101{
 102        write_seqlock(&mount_lock);
 103}
 104
 105static inline void unlock_mount_hash(void)
 106{
 107        write_sequnlock(&mount_lock);
 108}
 109
 110static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
 111{
 112        unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
 113        tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
 114        tmp = tmp + (tmp >> m_hash_shift);
 115        return &mount_hashtable[tmp & m_hash_mask];
 116}
 117
 118static inline struct hlist_head *mp_hash(struct dentry *dentry)
 119{
 120        unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
 121        tmp = tmp + (tmp >> mp_hash_shift);
 122        return &mountpoint_hashtable[tmp & mp_hash_mask];
 123}
 124
 125static int mnt_alloc_id(struct mount *mnt)
 126{
 127        int res = ida_alloc(&mnt_id_ida, GFP_KERNEL);
 128
 129        if (res < 0)
 130                return res;
 131        mnt->mnt_id = res;
 132        return 0;
 133}
 134
 135static void mnt_free_id(struct mount *mnt)
 136{
 137        ida_free(&mnt_id_ida, mnt->mnt_id);
 138}
 139
 140/*
 141 * Allocate a new peer group ID
 142 */
 143static int mnt_alloc_group_id(struct mount *mnt)
 144{
 145        int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL);
 146
 147        if (res < 0)
 148                return res;
 149        mnt->mnt_group_id = res;
 150        return 0;
 151}
 152
 153/*
 154 * Release a peer group ID
 155 */
 156void mnt_release_group_id(struct mount *mnt)
 157{
 158        ida_free(&mnt_group_ida, mnt->mnt_group_id);
 159        mnt->mnt_group_id = 0;
 160}
 161
 162/*
 163 * vfsmount lock must be held for read
 164 */
 165static inline void mnt_add_count(struct mount *mnt, int n)
 166{
 167#ifdef CONFIG_SMP
 168        this_cpu_add(mnt->mnt_pcp->mnt_count, n);
 169#else
 170        preempt_disable();
 171        mnt->mnt_count += n;
 172        preempt_enable();
 173#endif
 174}
 175
 176/*
 177 * vfsmount lock must be held for write
 178 */
 179int mnt_get_count(struct mount *mnt)
 180{
 181#ifdef CONFIG_SMP
 182        int count = 0;
 183        int cpu;
 184
 185        for_each_possible_cpu(cpu) {
 186                count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
 187        }
 188
 189        return count;
 190#else
 191        return mnt->mnt_count;
 192#endif
 193}
 194
 195static struct mount *alloc_vfsmnt(const char *name)
 196{
 197        struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
 198        if (mnt) {
 199                int err;
 200
 201                err = mnt_alloc_id(mnt);
 202                if (err)
 203                        goto out_free_cache;
 204
 205                if (name) {
 206                        mnt->mnt_devname = kstrdup_const(name, GFP_KERNEL);
 207                        if (!mnt->mnt_devname)
 208                                goto out_free_id;
 209                }
 210
 211#ifdef CONFIG_SMP
 212                mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
 213                if (!mnt->mnt_pcp)
 214                        goto out_free_devname;
 215
 216                this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
 217#else
 218                mnt->mnt_count = 1;
 219                mnt->mnt_writers = 0;
 220#endif
 221
 222                INIT_HLIST_NODE(&mnt->mnt_hash);
 223                INIT_LIST_HEAD(&mnt->mnt_child);
 224                INIT_LIST_HEAD(&mnt->mnt_mounts);
 225                INIT_LIST_HEAD(&mnt->mnt_list);
 226                INIT_LIST_HEAD(&mnt->mnt_expire);
 227                INIT_LIST_HEAD(&mnt->mnt_share);
 228                INIT_LIST_HEAD(&mnt->mnt_slave_list);
 229                INIT_LIST_HEAD(&mnt->mnt_slave);
 230                INIT_HLIST_NODE(&mnt->mnt_mp_list);
 231                INIT_LIST_HEAD(&mnt->mnt_umounting);
 232                INIT_HLIST_HEAD(&mnt->mnt_stuck_children);
 233                mnt->mnt.mnt_userns = &init_user_ns;
 234        }
 235        return mnt;
 236
 237#ifdef CONFIG_SMP
 238out_free_devname:
 239        kfree_const(mnt->mnt_devname);
 240#endif
 241out_free_id:
 242        mnt_free_id(mnt);
 243out_free_cache:
 244        kmem_cache_free(mnt_cache, mnt);
 245        return NULL;
 246}
 247
 248/*
 249 * Most r/o checks on a fs are for operations that take
 250 * discrete amounts of time, like a write() or unlink().
 251 * We must keep track of when those operations start
 252 * (for permission checks) and when they end, so that
 253 * we can determine when writes are able to occur to
 254 * a filesystem.
 255 */
 256/*
 257 * __mnt_is_readonly: check whether a mount is read-only
 258 * @mnt: the mount to check for its write status
 259 *
 260 * This shouldn't be used directly ouside of the VFS.
 261 * It does not guarantee that the filesystem will stay
 262 * r/w, just that it is right *now*.  This can not and
 263 * should not be used in place of IS_RDONLY(inode).
 264 * mnt_want/drop_write() will _keep_ the filesystem
 265 * r/w.
 266 */
 267bool __mnt_is_readonly(struct vfsmount *mnt)
 268{
 269        return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb);
 270}
 271EXPORT_SYMBOL_GPL(__mnt_is_readonly);
 272
 273static inline void mnt_inc_writers(struct mount *mnt)
 274{
 275#ifdef CONFIG_SMP
 276        this_cpu_inc(mnt->mnt_pcp->mnt_writers);
 277#else
 278        mnt->mnt_writers++;
 279#endif
 280}
 281
 282static inline void mnt_dec_writers(struct mount *mnt)
 283{
 284#ifdef CONFIG_SMP
 285        this_cpu_dec(mnt->mnt_pcp->mnt_writers);
 286#else
 287        mnt->mnt_writers--;
 288#endif
 289}
 290
 291static unsigned int mnt_get_writers(struct mount *mnt)
 292{
 293#ifdef CONFIG_SMP
 294        unsigned int count = 0;
 295        int cpu;
 296
 297        for_each_possible_cpu(cpu) {
 298                count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
 299        }
 300
 301        return count;
 302#else
 303        return mnt->mnt_writers;
 304#endif
 305}
 306
 307static int mnt_is_readonly(struct vfsmount *mnt)
 308{
 309        if (mnt->mnt_sb->s_readonly_remount)
 310                return 1;
 311        /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
 312        smp_rmb();
 313        return __mnt_is_readonly(mnt);
 314}
 315
 316/*
 317 * Most r/o & frozen checks on a fs are for operations that take discrete
 318 * amounts of time, like a write() or unlink().  We must keep track of when
 319 * those operations start (for permission checks) and when they end, so that we
 320 * can determine when writes are able to occur to a filesystem.
 321 */
 322/**
 323 * __mnt_want_write - get write access to a mount without freeze protection
 324 * @m: the mount on which to take a write
 325 *
 326 * This tells the low-level filesystem that a write is about to be performed to
 327 * it, and makes sure that writes are allowed (mnt it read-write) before
 328 * returning success. This operation does not protect against filesystem being
 329 * frozen. When the write operation is finished, __mnt_drop_write() must be
 330 * called. This is effectively a refcount.
 331 */
 332int __mnt_want_write(struct vfsmount *m)
 333{
 334        struct mount *mnt = real_mount(m);
 335        int ret = 0;
 336
 337        preempt_disable();
 338        mnt_inc_writers(mnt);
 339        /*
 340         * The store to mnt_inc_writers must be visible before we pass
 341         * MNT_WRITE_HOLD loop below, so that the slowpath can see our
 342         * incremented count after it has set MNT_WRITE_HOLD.
 343         */
 344        smp_mb();
 345        while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
 346                cpu_relax();
 347        /*
 348         * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
 349         * be set to match its requirements. So we must not load that until
 350         * MNT_WRITE_HOLD is cleared.
 351         */
 352        smp_rmb();
 353        if (mnt_is_readonly(m)) {
 354                mnt_dec_writers(mnt);
 355                ret = -EROFS;
 356        }
 357        preempt_enable();
 358
 359        return ret;
 360}
 361
 362/**
 363 * mnt_want_write - get write access to a mount
 364 * @m: the mount on which to take a write
 365 *
 366 * This tells the low-level filesystem that a write is about to be performed to
 367 * it, and makes sure that writes are allowed (mount is read-write, filesystem
 368 * is not frozen) before returning success.  When the write operation is
 369 * finished, mnt_drop_write() must be called.  This is effectively a refcount.
 370 */
 371int mnt_want_write(struct vfsmount *m)
 372{
 373        int ret;
 374
 375        sb_start_write(m->mnt_sb);
 376        ret = __mnt_want_write(m);
 377        if (ret)
 378                sb_end_write(m->mnt_sb);
 379        return ret;
 380}
 381EXPORT_SYMBOL_GPL(mnt_want_write);
 382
 383/**
 384 * __mnt_want_write_file - get write access to a file's mount
 385 * @file: the file who's mount on which to take a write
 386 *
 387 * This is like __mnt_want_write, but if the file is already open for writing it
 388 * skips incrementing mnt_writers (since the open file already has a reference)
 389 * and instead only does the check for emergency r/o remounts.  This must be
 390 * paired with __mnt_drop_write_file.
 391 */
 392int __mnt_want_write_file(struct file *file)
 393{
 394        if (file->f_mode & FMODE_WRITER) {
 395                /*
 396                 * Superblock may have become readonly while there are still
 397                 * writable fd's, e.g. due to a fs error with errors=remount-ro
 398                 */
 399                if (__mnt_is_readonly(file->f_path.mnt))
 400                        return -EROFS;
 401                return 0;
 402        }
 403        return __mnt_want_write(file->f_path.mnt);
 404}
 405
 406/**
 407 * mnt_want_write_file - get write access to a file's mount
 408 * @file: the file who's mount on which to take a write
 409 *
 410 * This is like mnt_want_write, but if the file is already open for writing it
 411 * skips incrementing mnt_writers (since the open file already has a reference)
 412 * and instead only does the freeze protection and the check for emergency r/o
 413 * remounts.  This must be paired with mnt_drop_write_file.
 414 */
 415int mnt_want_write_file(struct file *file)
 416{
 417        int ret;
 418
 419        sb_start_write(file_inode(file)->i_sb);
 420        ret = __mnt_want_write_file(file);
 421        if (ret)
 422                sb_end_write(file_inode(file)->i_sb);
 423        return ret;
 424}
 425EXPORT_SYMBOL_GPL(mnt_want_write_file);
 426
 427/**
 428 * __mnt_drop_write - give up write access to a mount
 429 * @mnt: the mount on which to give up write access
 430 *
 431 * Tells the low-level filesystem that we are done
 432 * performing writes to it.  Must be matched with
 433 * __mnt_want_write() call above.
 434 */
 435void __mnt_drop_write(struct vfsmount *mnt)
 436{
 437        preempt_disable();
 438        mnt_dec_writers(real_mount(mnt));
 439        preempt_enable();
 440}
 441
 442/**
 443 * mnt_drop_write - give up write access to a mount
 444 * @mnt: the mount on which to give up write access
 445 *
 446 * Tells the low-level filesystem that we are done performing writes to it and
 447 * also allows filesystem to be frozen again.  Must be matched with
 448 * mnt_want_write() call above.
 449 */
 450void mnt_drop_write(struct vfsmount *mnt)
 451{
 452        __mnt_drop_write(mnt);
 453        sb_end_write(mnt->mnt_sb);
 454}
 455EXPORT_SYMBOL_GPL(mnt_drop_write);
 456
 457void __mnt_drop_write_file(struct file *file)
 458{
 459        if (!(file->f_mode & FMODE_WRITER))
 460                __mnt_drop_write(file->f_path.mnt);
 461}
 462
 463void mnt_drop_write_file(struct file *file)
 464{
 465        __mnt_drop_write_file(file);
 466        sb_end_write(file_inode(file)->i_sb);
 467}
 468EXPORT_SYMBOL(mnt_drop_write_file);
 469
 470static inline int mnt_hold_writers(struct mount *mnt)
 471{
 472        mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
 473        /*
 474         * After storing MNT_WRITE_HOLD, we'll read the counters. This store
 475         * should be visible before we do.
 476         */
 477        smp_mb();
 478
 479        /*
 480         * With writers on hold, if this value is zero, then there are
 481         * definitely no active writers (although held writers may subsequently
 482         * increment the count, they'll have to wait, and decrement it after
 483         * seeing MNT_READONLY).
 484         *
 485         * It is OK to have counter incremented on one CPU and decremented on
 486         * another: the sum will add up correctly. The danger would be when we
 487         * sum up each counter, if we read a counter before it is incremented,
 488         * but then read another CPU's count which it has been subsequently
 489         * decremented from -- we would see more decrements than we should.
 490         * MNT_WRITE_HOLD protects against this scenario, because
 491         * mnt_want_write first increments count, then smp_mb, then spins on
 492         * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
 493         * we're counting up here.
 494         */
 495        if (mnt_get_writers(mnt) > 0)
 496                return -EBUSY;
 497
 498        return 0;
 499}
 500
 501static inline void mnt_unhold_writers(struct mount *mnt)
 502{
 503        /*
 504         * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
 505         * that become unheld will see MNT_READONLY.
 506         */
 507        smp_wmb();
 508        mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
 509}
 510
 511static int mnt_make_readonly(struct mount *mnt)
 512{
 513        int ret;
 514
 515        ret = mnt_hold_writers(mnt);
 516        if (!ret)
 517                mnt->mnt.mnt_flags |= MNT_READONLY;
 518        mnt_unhold_writers(mnt);
 519        return ret;
 520}
 521
 522int sb_prepare_remount_readonly(struct super_block *sb)
 523{
 524        struct mount *mnt;
 525        int err = 0;
 526
 527        /* Racy optimization.  Recheck the counter under MNT_WRITE_HOLD */
 528        if (atomic_long_read(&sb->s_remove_count))
 529                return -EBUSY;
 530
 531        lock_mount_hash();
 532        list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
 533                if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
 534                        mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
 535                        smp_mb();
 536                        if (mnt_get_writers(mnt) > 0) {
 537                                err = -EBUSY;
 538                                break;
 539                        }
 540                }
 541        }
 542        if (!err && atomic_long_read(&sb->s_remove_count))
 543                err = -EBUSY;
 544
 545        if (!err) {
 546                sb->s_readonly_remount = 1;
 547                smp_wmb();
 548        }
 549        list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
 550                if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
 551                        mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
 552        }
 553        unlock_mount_hash();
 554
 555        return err;
 556}
 557
 558static void free_vfsmnt(struct mount *mnt)
 559{
 560        struct user_namespace *mnt_userns;
 561
 562        mnt_userns = mnt_user_ns(&mnt->mnt);
 563        if (mnt_userns != &init_user_ns)
 564                put_user_ns(mnt_userns);
 565        kfree_const(mnt->mnt_devname);
 566#ifdef CONFIG_SMP
 567        free_percpu(mnt->mnt_pcp);
 568#endif
 569        kmem_cache_free(mnt_cache, mnt);
 570}
 571
 572static void delayed_free_vfsmnt(struct rcu_head *head)
 573{
 574        free_vfsmnt(container_of(head, struct mount, mnt_rcu));
 575}
 576
 577/* call under rcu_read_lock */
 578int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
 579{
 580        struct mount *mnt;
 581        if (read_seqretry(&mount_lock, seq))
 582                return 1;
 583        if (bastard == NULL)
 584                return 0;
 585        mnt = real_mount(bastard);
 586        mnt_add_count(mnt, 1);
 587        smp_mb();                       // see mntput_no_expire()
 588        if (likely(!read_seqretry(&mount_lock, seq)))
 589                return 0;
 590        if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
 591                mnt_add_count(mnt, -1);
 592                return 1;
 593        }
 594        lock_mount_hash();
 595        if (unlikely(bastard->mnt_flags & MNT_DOOMED)) {
 596                mnt_add_count(mnt, -1);
 597                unlock_mount_hash();
 598                return 1;
 599        }
 600        unlock_mount_hash();
 601        /* caller will mntput() */
 602        return -1;
 603}
 604
 605/* call under rcu_read_lock */
 606bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
 607{
 608        int res = __legitimize_mnt(bastard, seq);
 609        if (likely(!res))
 610                return true;
 611        if (unlikely(res < 0)) {
 612                rcu_read_unlock();
 613                mntput(bastard);
 614                rcu_read_lock();
 615        }
 616        return false;
 617}
 618
 619/*
 620 * find the first mount at @dentry on vfsmount @mnt.
 621 * call under rcu_read_lock()
 622 */
 623struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
 624{
 625        struct hlist_head *head = m_hash(mnt, dentry);
 626        struct mount *p;
 627
 628        hlist_for_each_entry_rcu(p, head, mnt_hash)
 629                if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
 630                        return p;
 631        return NULL;
 632}
 633
 634/*
 635 * lookup_mnt - Return the first child mount mounted at path
 636 *
 637 * "First" means first mounted chronologically.  If you create the
 638 * following mounts:
 639 *
 640 * mount /dev/sda1 /mnt
 641 * mount /dev/sda2 /mnt
 642 * mount /dev/sda3 /mnt
 643 *
 644 * Then lookup_mnt() on the base /mnt dentry in the root mount will
 645 * return successively the root dentry and vfsmount of /dev/sda1, then
 646 * /dev/sda2, then /dev/sda3, then NULL.
 647 *
 648 * lookup_mnt takes a reference to the found vfsmount.
 649 */
 650struct vfsmount *lookup_mnt(const struct path *path)
 651{
 652        struct mount *child_mnt;
 653        struct vfsmount *m;
 654        unsigned seq;
 655
 656        rcu_read_lock();
 657        do {
 658                seq = read_seqbegin(&mount_lock);
 659                child_mnt = __lookup_mnt(path->mnt, path->dentry);
 660                m = child_mnt ? &child_mnt->mnt : NULL;
 661        } while (!legitimize_mnt(m, seq));
 662        rcu_read_unlock();
 663        return m;
 664}
 665
 666static inline void lock_ns_list(struct mnt_namespace *ns)
 667{
 668        spin_lock(&ns->ns_lock);
 669}
 670
 671static inline void unlock_ns_list(struct mnt_namespace *ns)
 672{
 673        spin_unlock(&ns->ns_lock);
 674}
 675
 676static inline bool mnt_is_cursor(struct mount *mnt)
 677{
 678        return mnt->mnt.mnt_flags & MNT_CURSOR;
 679}
 680
 681/*
 682 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
 683 *                         current mount namespace.
 684 *
 685 * The common case is dentries are not mountpoints at all and that
 686 * test is handled inline.  For the slow case when we are actually
 687 * dealing with a mountpoint of some kind, walk through all of the
 688 * mounts in the current mount namespace and test to see if the dentry
 689 * is a mountpoint.
 690 *
 691 * The mount_hashtable is not usable in the context because we
 692 * need to identify all mounts that may be in the current mount
 693 * namespace not just a mount that happens to have some specified
 694 * parent mount.
 695 */
 696bool __is_local_mountpoint(struct dentry *dentry)
 697{
 698        struct mnt_namespace *ns = current->nsproxy->mnt_ns;
 699        struct mount *mnt;
 700        bool is_covered = false;
 701
 702        down_read(&namespace_sem);
 703        lock_ns_list(ns);
 704        list_for_each_entry(mnt, &ns->list, mnt_list) {
 705                if (mnt_is_cursor(mnt))
 706                        continue;
 707                is_covered = (mnt->mnt_mountpoint == dentry);
 708                if (is_covered)
 709                        break;
 710        }
 711        unlock_ns_list(ns);
 712        up_read(&namespace_sem);
 713
 714        return is_covered;
 715}
 716
 717static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
 718{
 719        struct hlist_head *chain = mp_hash(dentry);
 720        struct mountpoint *mp;
 721
 722        hlist_for_each_entry(mp, chain, m_hash) {
 723                if (mp->m_dentry == dentry) {
 724                        mp->m_count++;
 725                        return mp;
 726                }
 727        }
 728        return NULL;
 729}
 730
 731static struct mountpoint *get_mountpoint(struct dentry *dentry)
 732{
 733        struct mountpoint *mp, *new = NULL;
 734        int ret;
 735
 736        if (d_mountpoint(dentry)) {
 737                /* might be worth a WARN_ON() */
 738                if (d_unlinked(dentry))
 739                        return ERR_PTR(-ENOENT);
 740mountpoint:
 741                read_seqlock_excl(&mount_lock);
 742                mp = lookup_mountpoint(dentry);
 743                read_sequnlock_excl(&mount_lock);
 744                if (mp)
 745                        goto done;
 746        }
 747
 748        if (!new)
 749                new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
 750        if (!new)
 751                return ERR_PTR(-ENOMEM);
 752
 753
 754        /* Exactly one processes may set d_mounted */
 755        ret = d_set_mounted(dentry);
 756
 757        /* Someone else set d_mounted? */
 758        if (ret == -EBUSY)
 759                goto mountpoint;
 760
 761        /* The dentry is not available as a mountpoint? */
 762        mp = ERR_PTR(ret);
 763        if (ret)
 764                goto done;
 765
 766        /* Add the new mountpoint to the hash table */
 767        read_seqlock_excl(&mount_lock);
 768        new->m_dentry = dget(dentry);
 769        new->m_count = 1;
 770        hlist_add_head(&new->m_hash, mp_hash(dentry));
 771        INIT_HLIST_HEAD(&new->m_list);
 772        read_sequnlock_excl(&mount_lock);
 773
 774        mp = new;
 775        new = NULL;
 776done:
 777        kfree(new);
 778        return mp;
 779}
 780
 781/*
 782 * vfsmount lock must be held.  Additionally, the caller is responsible
 783 * for serializing calls for given disposal list.
 784 */
 785static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
 786{
 787        if (!--mp->m_count) {
 788                struct dentry *dentry = mp->m_dentry;
 789                BUG_ON(!hlist_empty(&mp->m_list));
 790                spin_lock(&dentry->d_lock);
 791                dentry->d_flags &= ~DCACHE_MOUNTED;
 792                spin_unlock(&dentry->d_lock);
 793                dput_to_list(dentry, list);
 794                hlist_del(&mp->m_hash);
 795                kfree(mp);
 796        }
 797}
 798
 799/* called with namespace_lock and vfsmount lock */
 800static void put_mountpoint(struct mountpoint *mp)
 801{
 802        __put_mountpoint(mp, &ex_mountpoints);
 803}
 804
 805static inline int check_mnt(struct mount *mnt)
 806{
 807        return mnt->mnt_ns == current->nsproxy->mnt_ns;
 808}
 809
 810/*
 811 * vfsmount lock must be held for write
 812 */
 813static void touch_mnt_namespace(struct mnt_namespace *ns)
 814{
 815        if (ns) {
 816                ns->event = ++event;
 817                wake_up_interruptible(&ns->poll);
 818        }
 819}
 820
 821/*
 822 * vfsmount lock must be held for write
 823 */
 824static void __touch_mnt_namespace(struct mnt_namespace *ns)
 825{
 826        if (ns && ns->event != event) {
 827                ns->event = event;
 828                wake_up_interruptible(&ns->poll);
 829        }
 830}
 831
 832/*
 833 * vfsmount lock must be held for write
 834 */
 835static struct mountpoint *unhash_mnt(struct mount *mnt)
 836{
 837        struct mountpoint *mp;
 838        mnt->mnt_parent = mnt;
 839        mnt->mnt_mountpoint = mnt->mnt.mnt_root;
 840        list_del_init(&mnt->mnt_child);
 841        hlist_del_init_rcu(&mnt->mnt_hash);
 842        hlist_del_init(&mnt->mnt_mp_list);
 843        mp = mnt->mnt_mp;
 844        mnt->mnt_mp = NULL;
 845        return mp;
 846}
 847
 848/*
 849 * vfsmount lock must be held for write
 850 */
 851static void umount_mnt(struct mount *mnt)
 852{
 853        put_mountpoint(unhash_mnt(mnt));
 854}
 855
 856/*
 857 * vfsmount lock must be held for write
 858 */
 859void mnt_set_mountpoint(struct mount *mnt,
 860                        struct mountpoint *mp,
 861                        struct mount *child_mnt)
 862{
 863        mp->m_count++;
 864        mnt_add_count(mnt, 1);  /* essentially, that's mntget */
 865        child_mnt->mnt_mountpoint = mp->m_dentry;
 866        child_mnt->mnt_parent = mnt;
 867        child_mnt->mnt_mp = mp;
 868        hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
 869}
 870
 871static void __attach_mnt(struct mount *mnt, struct mount *parent)
 872{
 873        hlist_add_head_rcu(&mnt->mnt_hash,
 874                           m_hash(&parent->mnt, mnt->mnt_mountpoint));
 875        list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
 876}
 877
 878/*
 879 * vfsmount lock must be held for write
 880 */
 881static void attach_mnt(struct mount *mnt,
 882                        struct mount *parent,
 883                        struct mountpoint *mp)
 884{
 885        mnt_set_mountpoint(parent, mp, mnt);
 886        __attach_mnt(mnt, parent);
 887}
 888
 889void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
 890{
 891        struct mountpoint *old_mp = mnt->mnt_mp;
 892        struct mount *old_parent = mnt->mnt_parent;
 893
 894        list_del_init(&mnt->mnt_child);
 895        hlist_del_init(&mnt->mnt_mp_list);
 896        hlist_del_init_rcu(&mnt->mnt_hash);
 897
 898        attach_mnt(mnt, parent, mp);
 899
 900        put_mountpoint(old_mp);
 901        mnt_add_count(old_parent, -1);
 902}
 903
 904/*
 905 * vfsmount lock must be held for write
 906 */
 907static void commit_tree(struct mount *mnt)
 908{
 909        struct mount *parent = mnt->mnt_parent;
 910        struct mount *m;
 911        LIST_HEAD(head);
 912        struct mnt_namespace *n = parent->mnt_ns;
 913
 914        BUG_ON(parent == mnt);
 915
 916        list_add_tail(&head, &mnt->mnt_list);
 917        list_for_each_entry(m, &head, mnt_list)
 918                m->mnt_ns = n;
 919
 920        list_splice(&head, n->list.prev);
 921
 922        n->mounts += n->pending_mounts;
 923        n->pending_mounts = 0;
 924
 925        __attach_mnt(mnt, parent);
 926        touch_mnt_namespace(n);
 927}
 928
 929static struct mount *next_mnt(struct mount *p, struct mount *root)
 930{
 931        struct list_head *next = p->mnt_mounts.next;
 932        if (next == &p->mnt_mounts) {
 933                while (1) {
 934                        if (p == root)
 935                                return NULL;
 936                        next = p->mnt_child.next;
 937                        if (next != &p->mnt_parent->mnt_mounts)
 938                                break;
 939                        p = p->mnt_parent;
 940                }
 941        }
 942        return list_entry(next, struct mount, mnt_child);
 943}
 944
 945static struct mount *skip_mnt_tree(struct mount *p)
 946{
 947        struct list_head *prev = p->mnt_mounts.prev;
 948        while (prev != &p->mnt_mounts) {
 949                p = list_entry(prev, struct mount, mnt_child);
 950                prev = p->mnt_mounts.prev;
 951        }
 952        return p;
 953}
 954
 955/**
 956 * vfs_create_mount - Create a mount for a configured superblock
 957 * @fc: The configuration context with the superblock attached
 958 *
 959 * Create a mount to an already configured superblock.  If necessary, the
 960 * caller should invoke vfs_get_tree() before calling this.
 961 *
 962 * Note that this does not attach the mount to anything.
 963 */
 964struct vfsmount *vfs_create_mount(struct fs_context *fc)
 965{
 966        struct mount *mnt;
 967
 968        if (!fc->root)
 969                return ERR_PTR(-EINVAL);
 970
 971        mnt = alloc_vfsmnt(fc->source ?: "none");
 972        if (!mnt)
 973                return ERR_PTR(-ENOMEM);
 974
 975        if (fc->sb_flags & SB_KERNMOUNT)
 976                mnt->mnt.mnt_flags = MNT_INTERNAL;
 977
 978        atomic_inc(&fc->root->d_sb->s_active);
 979        mnt->mnt.mnt_sb         = fc->root->d_sb;
 980        mnt->mnt.mnt_root       = dget(fc->root);
 981        mnt->mnt_mountpoint     = mnt->mnt.mnt_root;
 982        mnt->mnt_parent         = mnt;
 983
 984        lock_mount_hash();
 985        list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts);
 986        unlock_mount_hash();
 987        return &mnt->mnt;
 988}
 989EXPORT_SYMBOL(vfs_create_mount);
 990
 991struct vfsmount *fc_mount(struct fs_context *fc)
 992{
 993        int err = vfs_get_tree(fc);
 994        if (!err) {
 995                up_write(&fc->root->d_sb->s_umount);
 996                return vfs_create_mount(fc);
 997        }
 998        return ERR_PTR(err);
 999}
1000EXPORT_SYMBOL(fc_mount);
1001
1002struct vfsmount *vfs_kern_mount(struct file_system_type *type,
1003                                int flags, const char *name,
1004                                void *data)
1005{
1006        struct fs_context *fc;
1007        struct vfsmount *mnt;
1008        int ret = 0;
1009
1010        if (!type)
1011                return ERR_PTR(-EINVAL);
1012
1013        fc = fs_context_for_mount(type, flags);
1014        if (IS_ERR(fc))
1015                return ERR_CAST(fc);
1016
1017        if (name)
1018                ret = vfs_parse_fs_string(fc, "source",
1019                                          name, strlen(name));
1020        if (!ret)
1021                ret = parse_monolithic_mount_data(fc, data);
1022        if (!ret)
1023                mnt = fc_mount(fc);
1024        else
1025                mnt = ERR_PTR(ret);
1026
1027        put_fs_context(fc);
1028        return mnt;
1029}
1030EXPORT_SYMBOL_GPL(vfs_kern_mount);
1031
1032struct vfsmount *
1033vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1034             const char *name, void *data)
1035{
1036        /* Until it is worked out how to pass the user namespace
1037         * through from the parent mount to the submount don't support
1038         * unprivileged mounts with submounts.
1039         */
1040        if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1041                return ERR_PTR(-EPERM);
1042
1043        return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
1044}
1045EXPORT_SYMBOL_GPL(vfs_submount);
1046
1047static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1048                                        int flag)
1049{
1050        struct super_block *sb = old->mnt.mnt_sb;
1051        struct mount *mnt;
1052        int err;
1053
1054        mnt = alloc_vfsmnt(old->mnt_devname);
1055        if (!mnt)
1056                return ERR_PTR(-ENOMEM);
1057
1058        if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1059                mnt->mnt_group_id = 0; /* not a peer of original */
1060        else
1061                mnt->mnt_group_id = old->mnt_group_id;
1062
1063        if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1064                err = mnt_alloc_group_id(mnt);
1065                if (err)
1066                        goto out_free;
1067        }
1068
1069        mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1070        mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
1071
1072        atomic_inc(&sb->s_active);
1073        mnt->mnt.mnt_userns = mnt_user_ns(&old->mnt);
1074        if (mnt->mnt.mnt_userns != &init_user_ns)
1075                mnt->mnt.mnt_userns = get_user_ns(mnt->mnt.mnt_userns);
1076        mnt->mnt.mnt_sb = sb;
1077        mnt->mnt.mnt_root = dget(root);
1078        mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1079        mnt->mnt_parent = mnt;
1080        lock_mount_hash();
1081        list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1082        unlock_mount_hash();
1083
1084        if ((flag & CL_SLAVE) ||
1085            ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1086                list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1087                mnt->mnt_master = old;
1088                CLEAR_MNT_SHARED(mnt);
1089        } else if (!(flag & CL_PRIVATE)) {
1090                if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1091                        list_add(&mnt->mnt_share, &old->mnt_share);
1092                if (IS_MNT_SLAVE(old))
1093                        list_add(&mnt->mnt_slave, &old->mnt_slave);
1094                mnt->mnt_master = old->mnt_master;
1095        } else {
1096                CLEAR_MNT_SHARED(mnt);
1097        }
1098        if (flag & CL_MAKE_SHARED)
1099                set_mnt_shared(mnt);
1100
1101        /* stick the duplicate mount on the same expiry list
1102         * as the original if that was on one */
1103        if (flag & CL_EXPIRE) {
1104                if (!list_empty(&old->mnt_expire))
1105                        list_add(&mnt->mnt_expire, &old->mnt_expire);
1106        }
1107
1108        return mnt;
1109
1110 out_free:
1111        mnt_free_id(mnt);
1112        free_vfsmnt(mnt);
1113        return ERR_PTR(err);
1114}
1115
1116static void cleanup_mnt(struct mount *mnt)
1117{
1118        struct hlist_node *p;
1119        struct mount *m;
1120        /*
1121         * The warning here probably indicates that somebody messed
1122         * up a mnt_want/drop_write() pair.  If this happens, the
1123         * filesystem was probably unable to make r/w->r/o transitions.
1124         * The locking used to deal with mnt_count decrement provides barriers,
1125         * so mnt_get_writers() below is safe.
1126         */
1127        WARN_ON(mnt_get_writers(mnt));
1128        if (unlikely(mnt->mnt_pins.first))
1129                mnt_pin_kill(mnt);
1130        hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) {
1131                hlist_del(&m->mnt_umount);
1132                mntput(&m->mnt);
1133        }
1134        fsnotify_vfsmount_delete(&mnt->mnt);
1135        dput(mnt->mnt.mnt_root);
1136        deactivate_super(mnt->mnt.mnt_sb);
1137        mnt_free_id(mnt);
1138        call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1139}
1140
1141static void __cleanup_mnt(struct rcu_head *head)
1142{
1143        cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1144}
1145
1146static LLIST_HEAD(delayed_mntput_list);
1147static void delayed_mntput(struct work_struct *unused)
1148{
1149        struct llist_node *node = llist_del_all(&delayed_mntput_list);
1150        struct mount *m, *t;
1151
1152        llist_for_each_entry_safe(m, t, node, mnt_llist)
1153                cleanup_mnt(m);
1154}
1155static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1156
1157static void mntput_no_expire(struct mount *mnt)
1158{
1159        LIST_HEAD(list);
1160        int count;
1161
1162        rcu_read_lock();
1163        if (likely(READ_ONCE(mnt->mnt_ns))) {
1164                /*
1165                 * Since we don't do lock_mount_hash() here,
1166                 * ->mnt_ns can change under us.  However, if it's
1167                 * non-NULL, then there's a reference that won't
1168                 * be dropped until after an RCU delay done after
1169                 * turning ->mnt_ns NULL.  So if we observe it
1170                 * non-NULL under rcu_read_lock(), the reference
1171                 * we are dropping is not the final one.
1172                 */
1173                mnt_add_count(mnt, -1);
1174                rcu_read_unlock();
1175                return;
1176        }
1177        lock_mount_hash();
1178        /*
1179         * make sure that if __legitimize_mnt() has not seen us grab
1180         * mount_lock, we'll see their refcount increment here.
1181         */
1182        smp_mb();
1183        mnt_add_count(mnt, -1);
1184        count = mnt_get_count(mnt);
1185        if (count != 0) {
1186                WARN_ON(count < 0);
1187                rcu_read_unlock();
1188                unlock_mount_hash();
1189                return;
1190        }
1191        if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1192                rcu_read_unlock();
1193                unlock_mount_hash();
1194                return;
1195        }
1196        mnt->mnt.mnt_flags |= MNT_DOOMED;
1197        rcu_read_unlock();
1198
1199        list_del(&mnt->mnt_instance);
1200
1201        if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1202                struct mount *p, *tmp;
1203                list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts,  mnt_child) {
1204                        __put_mountpoint(unhash_mnt(p), &list);
1205                        hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children);
1206                }
1207        }
1208        unlock_mount_hash();
1209        shrink_dentry_list(&list);
1210
1211        if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1212                struct task_struct *task = current;
1213                if (likely(!(task->flags & PF_KTHREAD))) {
1214                        init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1215                        if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME))
1216                                return;
1217                }
1218                if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1219                        schedule_delayed_work(&delayed_mntput_work, 1);
1220                return;
1221        }
1222        cleanup_mnt(mnt);
1223}
1224
1225void mntput(struct vfsmount *mnt)
1226{
1227        if (mnt) {
1228                struct mount *m = real_mount(mnt);
1229                /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1230                if (unlikely(m->mnt_expiry_mark))
1231                        m->mnt_expiry_mark = 0;
1232                mntput_no_expire(m);
1233        }
1234}
1235EXPORT_SYMBOL(mntput);
1236
1237struct vfsmount *mntget(struct vfsmount *mnt)
1238{
1239        if (mnt)
1240                mnt_add_count(real_mount(mnt), 1);
1241        return mnt;
1242}
1243EXPORT_SYMBOL(mntget);
1244
1245/**
1246 * path_is_mountpoint() - Check if path is a mount in the current namespace.
1247 * @path: path to check
1248 *
1249 *  d_mountpoint() can only be used reliably to establish if a dentry is
1250 *  not mounted in any namespace and that common case is handled inline.
1251 *  d_mountpoint() isn't aware of the possibility there may be multiple
1252 *  mounts using a given dentry in a different namespace. This function
1253 *  checks if the passed in path is a mountpoint rather than the dentry
1254 *  alone.
1255 */
1256bool path_is_mountpoint(const struct path *path)
1257{
1258        unsigned seq;
1259        bool res;
1260
1261        if (!d_mountpoint(path->dentry))
1262                return false;
1263
1264        rcu_read_lock();
1265        do {
1266                seq = read_seqbegin(&mount_lock);
1267                res = __path_is_mountpoint(path);
1268        } while (read_seqretry(&mount_lock, seq));
1269        rcu_read_unlock();
1270
1271        return res;
1272}
1273EXPORT_SYMBOL(path_is_mountpoint);
1274
1275struct vfsmount *mnt_clone_internal(const struct path *path)
1276{
1277        struct mount *p;
1278        p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1279        if (IS_ERR(p))
1280                return ERR_CAST(p);
1281        p->mnt.mnt_flags |= MNT_INTERNAL;
1282        return &p->mnt;
1283}
1284
1285#ifdef CONFIG_PROC_FS
1286static struct mount *mnt_list_next(struct mnt_namespace *ns,
1287                                   struct list_head *p)
1288{
1289        struct mount *mnt, *ret = NULL;
1290
1291        lock_ns_list(ns);
1292        list_for_each_continue(p, &ns->list) {
1293                mnt = list_entry(p, typeof(*mnt), mnt_list);
1294                if (!mnt_is_cursor(mnt)) {
1295                        ret = mnt;
1296                        break;
1297                }
1298        }
1299        unlock_ns_list(ns);
1300
1301        return ret;
1302}
1303
1304/* iterator; we want it to have access to namespace_sem, thus here... */
1305static void *m_start(struct seq_file *m, loff_t *pos)
1306{
1307        struct proc_mounts *p = m->private;
1308        struct list_head *prev;
1309
1310        down_read(&namespace_sem);
1311        if (!*pos) {
1312                prev = &p->ns->list;
1313        } else {
1314                prev = &p->cursor.mnt_list;
1315
1316                /* Read after we'd reached the end? */
1317                if (list_empty(prev))
1318                        return NULL;
1319        }
1320
1321        return mnt_list_next(p->ns, prev);
1322}
1323
1324static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1325{
1326        struct proc_mounts *p = m->private;
1327        struct mount *mnt = v;
1328
1329        ++*pos;
1330        return mnt_list_next(p->ns, &mnt->mnt_list);
1331}
1332
1333static void m_stop(struct seq_file *m, void *v)
1334{
1335        struct proc_mounts *p = m->private;
1336        struct mount *mnt = v;
1337
1338        lock_ns_list(p->ns);
1339        if (mnt)
1340                list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list);
1341        else
1342                list_del_init(&p->cursor.mnt_list);
1343        unlock_ns_list(p->ns);
1344        up_read(&namespace_sem);
1345}
1346
1347static int m_show(struct seq_file *m, void *v)
1348{
1349        struct proc_mounts *p = m->private;
1350        struct mount *r = v;
1351        return p->show(m, &r->mnt);
1352}
1353
1354const struct seq_operations mounts_op = {
1355        .start  = m_start,
1356        .next   = m_next,
1357        .stop   = m_stop,
1358        .show   = m_show,
1359};
1360
1361void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor)
1362{
1363        down_read(&namespace_sem);
1364        lock_ns_list(ns);
1365        list_del(&cursor->mnt_list);
1366        unlock_ns_list(ns);
1367        up_read(&namespace_sem);
1368}
1369#endif  /* CONFIG_PROC_FS */
1370
1371/**
1372 * may_umount_tree - check if a mount tree is busy
1373 * @m: root of mount tree
1374 *
1375 * This is called to check if a tree of mounts has any
1376 * open files, pwds, chroots or sub mounts that are
1377 * busy.
1378 */
1379int may_umount_tree(struct vfsmount *m)
1380{
1381        struct mount *mnt = real_mount(m);
1382        int actual_refs = 0;
1383        int minimum_refs = 0;
1384        struct mount *p;
1385        BUG_ON(!m);
1386
1387        /* write lock needed for mnt_get_count */
1388        lock_mount_hash();
1389        for (p = mnt; p; p = next_mnt(p, mnt)) {
1390                actual_refs += mnt_get_count(p);
1391                minimum_refs += 2;
1392        }
1393        unlock_mount_hash();
1394
1395        if (actual_refs > minimum_refs)
1396                return 0;
1397
1398        return 1;
1399}
1400
1401EXPORT_SYMBOL(may_umount_tree);
1402
1403/**
1404 * may_umount - check if a mount point is busy
1405 * @mnt: root of mount
1406 *
1407 * This is called to check if a mount point has any
1408 * open files, pwds, chroots or sub mounts. If the
1409 * mount has sub mounts this will return busy
1410 * regardless of whether the sub mounts are busy.
1411 *
1412 * Doesn't take quota and stuff into account. IOW, in some cases it will
1413 * give false negatives. The main reason why it's here is that we need
1414 * a non-destructive way to look for easily umountable filesystems.
1415 */
1416int may_umount(struct vfsmount *mnt)
1417{
1418        int ret = 1;
1419        down_read(&namespace_sem);
1420        lock_mount_hash();
1421        if (propagate_mount_busy(real_mount(mnt), 2))
1422                ret = 0;
1423        unlock_mount_hash();
1424        up_read(&namespace_sem);
1425        return ret;
1426}
1427
1428EXPORT_SYMBOL(may_umount);
1429
1430static void namespace_unlock(void)
1431{
1432        struct hlist_head head;
1433        struct hlist_node *p;
1434        struct mount *m;
1435        LIST_HEAD(list);
1436
1437        hlist_move_list(&unmounted, &head);
1438        list_splice_init(&ex_mountpoints, &list);
1439
1440        up_write(&namespace_sem);
1441
1442        shrink_dentry_list(&list);
1443
1444        if (likely(hlist_empty(&head)))
1445                return;
1446
1447        synchronize_rcu_expedited();
1448
1449        hlist_for_each_entry_safe(m, p, &head, mnt_umount) {
1450                hlist_del(&m->mnt_umount);
1451                mntput(&m->mnt);
1452        }
1453}
1454
1455static inline void namespace_lock(void)
1456{
1457        down_write(&namespace_sem);
1458}
1459
1460enum umount_tree_flags {
1461        UMOUNT_SYNC = 1,
1462        UMOUNT_PROPAGATE = 2,
1463        UMOUNT_CONNECTED = 4,
1464};
1465
1466static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1467{
1468        /* Leaving mounts connected is only valid for lazy umounts */
1469        if (how & UMOUNT_SYNC)
1470                return true;
1471
1472        /* A mount without a parent has nothing to be connected to */
1473        if (!mnt_has_parent(mnt))
1474                return true;
1475
1476        /* Because the reference counting rules change when mounts are
1477         * unmounted and connected, umounted mounts may not be
1478         * connected to mounted mounts.
1479         */
1480        if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1481                return true;
1482
1483        /* Has it been requested that the mount remain connected? */
1484        if (how & UMOUNT_CONNECTED)
1485                return false;
1486
1487        /* Is the mount locked such that it needs to remain connected? */
1488        if (IS_MNT_LOCKED(mnt))
1489                return false;
1490
1491        /* By default disconnect the mount */
1492        return true;
1493}
1494
1495/*
1496 * mount_lock must be held
1497 * namespace_sem must be held for write
1498 */
1499static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1500{
1501        LIST_HEAD(tmp_list);
1502        struct mount *p;
1503
1504        if (how & UMOUNT_PROPAGATE)
1505                propagate_mount_unlock(mnt);
1506
1507        /* Gather the mounts to umount */
1508        for (p = mnt; p; p = next_mnt(p, mnt)) {
1509                p->mnt.mnt_flags |= MNT_UMOUNT;
1510                list_move(&p->mnt_list, &tmp_list);
1511        }
1512
1513        /* Hide the mounts from mnt_mounts */
1514        list_for_each_entry(p, &tmp_list, mnt_list) {
1515                list_del_init(&p->mnt_child);
1516        }
1517
1518        /* Add propogated mounts to the tmp_list */
1519        if (how & UMOUNT_PROPAGATE)
1520                propagate_umount(&tmp_list);
1521
1522        while (!list_empty(&tmp_list)) {
1523                struct mnt_namespace *ns;
1524                bool disconnect;
1525                p = list_first_entry(&tmp_list, struct mount, mnt_list);
1526                list_del_init(&p->mnt_expire);
1527                list_del_init(&p->mnt_list);
1528                ns = p->mnt_ns;
1529                if (ns) {
1530                        ns->mounts--;
1531                        __touch_mnt_namespace(ns);
1532                }
1533                p->mnt_ns = NULL;
1534                if (how & UMOUNT_SYNC)
1535                        p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1536
1537                disconnect = disconnect_mount(p, how);
1538                if (mnt_has_parent(p)) {
1539                        mnt_add_count(p->mnt_parent, -1);
1540                        if (!disconnect) {
1541                                /* Don't forget about p */
1542                                list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1543                        } else {
1544                                umount_mnt(p);
1545                        }
1546                }
1547                change_mnt_propagation(p, MS_PRIVATE);
1548                if (disconnect)
1549                        hlist_add_head(&p->mnt_umount, &unmounted);
1550        }
1551}
1552
1553static void shrink_submounts(struct mount *mnt);
1554
1555static int do_umount_root(struct super_block *sb)
1556{
1557        int ret = 0;
1558
1559        down_write(&sb->s_umount);
1560        if (!sb_rdonly(sb)) {
1561                struct fs_context *fc;
1562
1563                fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY,
1564                                                SB_RDONLY);
1565                if (IS_ERR(fc)) {
1566                        ret = PTR_ERR(fc);
1567                } else {
1568                        ret = parse_monolithic_mount_data(fc, NULL);
1569                        if (!ret)
1570                                ret = reconfigure_super(fc);
1571                        put_fs_context(fc);
1572                }
1573        }
1574        up_write(&sb->s_umount);
1575        return ret;
1576}
1577
1578static int do_umount(struct mount *mnt, int flags)
1579{
1580        struct super_block *sb = mnt->mnt.mnt_sb;
1581        int retval;
1582
1583        retval = security_sb_umount(&mnt->mnt, flags);
1584        if (retval)
1585                return retval;
1586
1587        /*
1588         * Allow userspace to request a mountpoint be expired rather than
1589         * unmounting unconditionally. Unmount only happens if:
1590         *  (1) the mark is already set (the mark is cleared by mntput())
1591         *  (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1592         */
1593        if (flags & MNT_EXPIRE) {
1594                if (&mnt->mnt == current->fs->root.mnt ||
1595                    flags & (MNT_FORCE | MNT_DETACH))
1596                        return -EINVAL;
1597
1598                /*
1599                 * probably don't strictly need the lock here if we examined
1600                 * all race cases, but it's a slowpath.
1601                 */
1602                lock_mount_hash();
1603                if (mnt_get_count(mnt) != 2) {
1604                        unlock_mount_hash();
1605                        return -EBUSY;
1606                }
1607                unlock_mount_hash();
1608
1609                if (!xchg(&mnt->mnt_expiry_mark, 1))
1610                        return -EAGAIN;
1611        }
1612
1613        /*
1614         * If we may have to abort operations to get out of this
1615         * mount, and they will themselves hold resources we must
1616         * allow the fs to do things. In the Unix tradition of
1617         * 'Gee thats tricky lets do it in userspace' the umount_begin
1618         * might fail to complete on the first run through as other tasks
1619         * must return, and the like. Thats for the mount program to worry
1620         * about for the moment.
1621         */
1622
1623        if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1624                sb->s_op->umount_begin(sb);
1625        }
1626
1627        /*
1628         * No sense to grab the lock for this test, but test itself looks
1629         * somewhat bogus. Suggestions for better replacement?
1630         * Ho-hum... In principle, we might treat that as umount + switch
1631         * to rootfs. GC would eventually take care of the old vfsmount.
1632         * Actually it makes sense, especially if rootfs would contain a
1633         * /reboot - static binary that would close all descriptors and
1634         * call reboot(9). Then init(8) could umount root and exec /reboot.
1635         */
1636        if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1637                /*
1638                 * Special case for "unmounting" root ...
1639                 * we just try to remount it readonly.
1640                 */
1641                if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN))
1642                        return -EPERM;
1643                return do_umount_root(sb);
1644        }
1645
1646        namespace_lock();
1647        lock_mount_hash();
1648
1649        /* Recheck MNT_LOCKED with the locks held */
1650        retval = -EINVAL;
1651        if (mnt->mnt.mnt_flags & MNT_LOCKED)
1652                goto out;
1653
1654        event++;
1655        if (flags & MNT_DETACH) {
1656                if (!list_empty(&mnt->mnt_list))
1657                        umount_tree(mnt, UMOUNT_PROPAGATE);
1658                retval = 0;
1659        } else {
1660                shrink_submounts(mnt);
1661                retval = -EBUSY;
1662                if (!propagate_mount_busy(mnt, 2)) {
1663                        if (!list_empty(&mnt->mnt_list))
1664                                umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1665                        retval = 0;
1666                }
1667        }
1668out:
1669        unlock_mount_hash();
1670        namespace_unlock();
1671        return retval;
1672}
1673
1674/*
1675 * __detach_mounts - lazily unmount all mounts on the specified dentry
1676 *
1677 * During unlink, rmdir, and d_drop it is possible to loose the path
1678 * to an existing mountpoint, and wind up leaking the mount.
1679 * detach_mounts allows lazily unmounting those mounts instead of
1680 * leaking them.
1681 *
1682 * The caller may hold dentry->d_inode->i_mutex.
1683 */
1684void __detach_mounts(struct dentry *dentry)
1685{
1686        struct mountpoint *mp;
1687        struct mount *mnt;
1688
1689        namespace_lock();
1690        lock_mount_hash();
1691        mp = lookup_mountpoint(dentry);
1692        if (!mp)
1693                goto out_unlock;
1694
1695        event++;
1696        while (!hlist_empty(&mp->m_list)) {
1697                mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1698                if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1699                        umount_mnt(mnt);
1700                        hlist_add_head(&mnt->mnt_umount, &unmounted);
1701                }
1702                else umount_tree(mnt, UMOUNT_CONNECTED);
1703        }
1704        put_mountpoint(mp);
1705out_unlock:
1706        unlock_mount_hash();
1707        namespace_unlock();
1708}
1709
1710/*
1711 * Is the caller allowed to modify his namespace?
1712 */
1713static inline bool may_mount(void)
1714{
1715        return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1716}
1717
1718#ifdef  CONFIG_MANDATORY_FILE_LOCKING
1719static inline bool may_mandlock(void)
1720{
1721        return capable(CAP_SYS_ADMIN);
1722}
1723#else
1724static inline bool may_mandlock(void)
1725{
1726        pr_warn("VFS: \"mand\" mount option not supported");
1727        return false;
1728}
1729#endif
1730
1731static int can_umount(const struct path *path, int flags)
1732{
1733        struct mount *mnt = real_mount(path->mnt);
1734
1735        if (!may_mount())
1736                return -EPERM;
1737        if (path->dentry != path->mnt->mnt_root)
1738                return -EINVAL;
1739        if (!check_mnt(mnt))
1740                return -EINVAL;
1741        if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
1742                return -EINVAL;
1743        if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1744                return -EPERM;
1745        return 0;
1746}
1747
1748// caller is responsible for flags being sane
1749int path_umount(struct path *path, int flags)
1750{
1751        struct mount *mnt = real_mount(path->mnt);
1752        int ret;
1753
1754        ret = can_umount(path, flags);
1755        if (!ret)
1756                ret = do_umount(mnt, flags);
1757
1758        /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1759        dput(path->dentry);
1760        mntput_no_expire(mnt);
1761        return ret;
1762}
1763
1764static int ksys_umount(char __user *name, int flags)
1765{
1766        int lookup_flags = LOOKUP_MOUNTPOINT;
1767        struct path path;
1768        int ret;
1769
1770        // basic validity checks done first
1771        if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1772                return -EINVAL;
1773
1774        if (!(flags & UMOUNT_NOFOLLOW))
1775                lookup_flags |= LOOKUP_FOLLOW;
1776        ret = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1777        if (ret)
1778                return ret;
1779        return path_umount(&path, flags);
1780}
1781
1782SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1783{
1784        return ksys_umount(name, flags);
1785}
1786
1787#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1788
1789/*
1790 *      The 2.0 compatible umount. No flags.
1791 */
1792SYSCALL_DEFINE1(oldumount, char __user *, name)
1793{
1794        return ksys_umount(name, 0);
1795}
1796
1797#endif
1798
1799static bool is_mnt_ns_file(struct dentry *dentry)
1800{
1801        /* Is this a proxy for a mount namespace? */
1802        return dentry->d_op == &ns_dentry_operations &&
1803               dentry->d_fsdata == &mntns_operations;
1804}
1805
1806static struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1807{
1808        return container_of(ns, struct mnt_namespace, ns);
1809}
1810
1811struct ns_common *from_mnt_ns(struct mnt_namespace *mnt)
1812{
1813        return &mnt->ns;
1814}
1815
1816static bool mnt_ns_loop(struct dentry *dentry)
1817{
1818        /* Could bind mounting the mount namespace inode cause a
1819         * mount namespace loop?
1820         */
1821        struct mnt_namespace *mnt_ns;
1822        if (!is_mnt_ns_file(dentry))
1823                return false;
1824
1825        mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1826        return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1827}
1828
1829struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1830                                        int flag)
1831{
1832        struct mount *res, *p, *q, *r, *parent;
1833
1834        if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1835                return ERR_PTR(-EINVAL);
1836
1837        if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1838                return ERR_PTR(-EINVAL);
1839
1840        res = q = clone_mnt(mnt, dentry, flag);
1841        if (IS_ERR(q))
1842                return q;
1843
1844        q->mnt_mountpoint = mnt->mnt_mountpoint;
1845
1846        p = mnt;
1847        list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1848                struct mount *s;
1849                if (!is_subdir(r->mnt_mountpoint, dentry))
1850                        continue;
1851
1852                for (s = r; s; s = next_mnt(s, r)) {
1853                        if (!(flag & CL_COPY_UNBINDABLE) &&
1854                            IS_MNT_UNBINDABLE(s)) {
1855                                if (s->mnt.mnt_flags & MNT_LOCKED) {
1856                                        /* Both unbindable and locked. */
1857                                        q = ERR_PTR(-EPERM);
1858                                        goto out;
1859                                } else {
1860                                        s = skip_mnt_tree(s);
1861                                        continue;
1862                                }
1863                        }
1864                        if (!(flag & CL_COPY_MNT_NS_FILE) &&
1865                            is_mnt_ns_file(s->mnt.mnt_root)) {
1866                                s = skip_mnt_tree(s);
1867                                continue;
1868                        }
1869                        while (p != s->mnt_parent) {
1870                                p = p->mnt_parent;
1871                                q = q->mnt_parent;
1872                        }
1873                        p = s;
1874                        parent = q;
1875                        q = clone_mnt(p, p->mnt.mnt_root, flag);
1876                        if (IS_ERR(q))
1877                                goto out;
1878                        lock_mount_hash();
1879                        list_add_tail(&q->mnt_list, &res->mnt_list);
1880                        attach_mnt(q, parent, p->mnt_mp);
1881                        unlock_mount_hash();
1882                }
1883        }
1884        return res;
1885out:
1886        if (res) {
1887                lock_mount_hash();
1888                umount_tree(res, UMOUNT_SYNC);
1889                unlock_mount_hash();
1890        }
1891        return q;
1892}
1893
1894/* Caller should check returned pointer for errors */
1895
1896struct vfsmount *collect_mounts(const struct path *path)
1897{
1898        struct mount *tree;
1899        namespace_lock();
1900        if (!check_mnt(real_mount(path->mnt)))
1901                tree = ERR_PTR(-EINVAL);
1902        else
1903                tree = copy_tree(real_mount(path->mnt), path->dentry,
1904                                 CL_COPY_ALL | CL_PRIVATE);
1905        namespace_unlock();
1906        if (IS_ERR(tree))
1907                return ERR_CAST(tree);
1908        return &tree->mnt;
1909}
1910
1911static void free_mnt_ns(struct mnt_namespace *);
1912static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
1913
1914void dissolve_on_fput(struct vfsmount *mnt)
1915{
1916        struct mnt_namespace *ns;
1917        namespace_lock();
1918        lock_mount_hash();
1919        ns = real_mount(mnt)->mnt_ns;
1920        if (ns) {
1921                if (is_anon_ns(ns))
1922                        umount_tree(real_mount(mnt), UMOUNT_CONNECTED);
1923                else
1924                        ns = NULL;
1925        }
1926        unlock_mount_hash();
1927        namespace_unlock();
1928        if (ns)
1929                free_mnt_ns(ns);
1930}
1931
1932void drop_collected_mounts(struct vfsmount *mnt)
1933{
1934        namespace_lock();
1935        lock_mount_hash();
1936        umount_tree(real_mount(mnt), 0);
1937        unlock_mount_hash();
1938        namespace_unlock();
1939}
1940
1941static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
1942{
1943        struct mount *child;
1944
1945        list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
1946                if (!is_subdir(child->mnt_mountpoint, dentry))
1947                        continue;
1948
1949                if (child->mnt.mnt_flags & MNT_LOCKED)
1950                        return true;
1951        }
1952        return false;
1953}
1954
1955/**
1956 * clone_private_mount - create a private clone of a path
1957 * @path: path to clone
1958 *
1959 * This creates a new vfsmount, which will be the clone of @path.  The new mount
1960 * will not be attached anywhere in the namespace and will be private (i.e.
1961 * changes to the originating mount won't be propagated into this).
1962 *
1963 * Release with mntput().
1964 */
1965struct vfsmount *clone_private_mount(const struct path *path)
1966{
1967        struct mount *old_mnt = real_mount(path->mnt);
1968        struct mount *new_mnt;
1969
1970        down_read(&namespace_sem);
1971        if (IS_MNT_UNBINDABLE(old_mnt))
1972                goto invalid;
1973
1974        if (!check_mnt(old_mnt))
1975                goto invalid;
1976
1977        if (has_locked_children(old_mnt, path->dentry))
1978                goto invalid;
1979
1980        new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1981        up_read(&namespace_sem);
1982
1983        if (IS_ERR(new_mnt))
1984                return ERR_CAST(new_mnt);
1985
1986        /* Longterm mount to be removed by kern_unmount*() */
1987        new_mnt->mnt_ns = MNT_NS_INTERNAL;
1988
1989        return &new_mnt->mnt;
1990
1991invalid:
1992        up_read(&namespace_sem);
1993        return ERR_PTR(-EINVAL);
1994}
1995EXPORT_SYMBOL_GPL(clone_private_mount);
1996
1997int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1998                   struct vfsmount *root)
1999{
2000        struct mount *mnt;
2001        int res = f(root, arg);
2002        if (res)
2003                return res;
2004        list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
2005                res = f(&mnt->mnt, arg);
2006                if (res)
2007                        return res;
2008        }
2009        return 0;
2010}
2011
2012static void lock_mnt_tree(struct mount *mnt)
2013{
2014        struct mount *p;
2015
2016        for (p = mnt; p; p = next_mnt(p, mnt)) {
2017                int flags = p->mnt.mnt_flags;
2018                /* Don't allow unprivileged users to change mount flags */
2019                flags |= MNT_LOCK_ATIME;
2020
2021                if (flags & MNT_READONLY)
2022                        flags |= MNT_LOCK_READONLY;
2023
2024                if (flags & MNT_NODEV)
2025                        flags |= MNT_LOCK_NODEV;
2026
2027                if (flags & MNT_NOSUID)
2028                        flags |= MNT_LOCK_NOSUID;
2029
2030                if (flags & MNT_NOEXEC)
2031                        flags |= MNT_LOCK_NOEXEC;
2032                /* Don't allow unprivileged users to reveal what is under a mount */
2033                if (list_empty(&p->mnt_expire))
2034                        flags |= MNT_LOCKED;
2035                p->mnt.mnt_flags = flags;
2036        }
2037}
2038
2039static void cleanup_group_ids(struct mount *mnt, struct mount *end)
2040{
2041        struct mount *p;
2042
2043        for (p = mnt; p != end; p = next_mnt(p, mnt)) {
2044                if (p->mnt_group_id && !IS_MNT_SHARED(p))
2045                        mnt_release_group_id(p);
2046        }
2047}
2048
2049static int invent_group_ids(struct mount *mnt, bool recurse)
2050{
2051        struct mount *p;
2052
2053        for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
2054                if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
2055                        int err = mnt_alloc_group_id(p);
2056                        if (err) {
2057                                cleanup_group_ids(mnt, p);
2058                                return err;
2059                        }
2060                }
2061        }
2062
2063        return 0;
2064}
2065
2066int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
2067{
2068        unsigned int max = READ_ONCE(sysctl_mount_max);
2069        unsigned int mounts = 0, old, pending, sum;
2070        struct mount *p;
2071
2072        for (p = mnt; p; p = next_mnt(p, mnt))
2073                mounts++;
2074
2075        old = ns->mounts;
2076        pending = ns->pending_mounts;
2077        sum = old + pending;
2078        if ((old > sum) ||
2079            (pending > sum) ||
2080            (max < sum) ||
2081            (mounts > (max - sum)))
2082                return -ENOSPC;
2083
2084        ns->pending_mounts = pending + mounts;
2085        return 0;
2086}
2087
2088/*
2089 *  @source_mnt : mount tree to be attached
2090 *  @nd         : place the mount tree @source_mnt is attached
2091 *  @parent_nd  : if non-null, detach the source_mnt from its parent and
2092 *                 store the parent mount and mountpoint dentry.
2093 *                 (done when source_mnt is moved)
2094 *
2095 *  NOTE: in the table below explains the semantics when a source mount
2096 *  of a given type is attached to a destination mount of a given type.
2097 * ---------------------------------------------------------------------------
2098 * |         BIND MOUNT OPERATION                                            |
2099 * |**************************************************************************
2100 * | source-->| shared        |       private  |       slave    | unbindable |
2101 * | dest     |               |                |                |            |
2102 * |   |      |               |                |                |            |
2103 * |   v      |               |                |                |            |
2104 * |**************************************************************************
2105 * |  shared  | shared (++)   |     shared (+) |     shared(+++)|  invalid   |
2106 * |          |               |                |                |            |
2107 * |non-shared| shared (+)    |      private   |      slave (*) |  invalid   |
2108 * ***************************************************************************
2109 * A bind operation clones the source mount and mounts the clone on the
2110 * destination mount.
2111 *
2112 * (++)  the cloned mount is propagated to all the mounts in the propagation
2113 *       tree of the destination mount and the cloned mount is added to
2114 *       the peer group of the source mount.
2115 * (+)   the cloned mount is created under the destination mount and is marked
2116 *       as shared. The cloned mount is added to the peer group of the source
2117 *       mount.
2118 * (+++) the mount is propagated to all the mounts in the propagation tree
2119 *       of the destination mount and the cloned mount is made slave
2120 *       of the same master as that of the source mount. The cloned mount
2121 *       is marked as 'shared and slave'.
2122 * (*)   the cloned mount is made a slave of the same master as that of the
2123 *       source mount.
2124 *
2125 * ---------------------------------------------------------------------------
2126 * |                    MOVE MOUNT OPERATION                                 |
2127 * |**************************************************************************
2128 * | source-->| shared        |       private  |       slave    | unbindable |
2129 * | dest     |               |                |                |            |
2130 * |   |      |               |                |                |            |
2131 * |   v      |               |                |                |            |
2132 * |**************************************************************************
2133 * |  shared  | shared (+)    |     shared (+) |    shared(+++) |  invalid   |
2134 * |          |               |                |                |            |
2135 * |non-shared| shared (+*)   |      private   |    slave (*)   | unbindable |
2136 * ***************************************************************************
2137 *
2138 * (+)  the mount is moved to the destination. And is then propagated to
2139 *      all the mounts in the propagation tree of the destination mount.
2140 * (+*)  the mount is moved to the destination.
2141 * (+++)  the mount is moved to the destination and is then propagated to
2142 *      all the mounts belonging to the destination mount's propagation tree.
2143 *      the mount is marked as 'shared and slave'.
2144 * (*)  the mount continues to be a slave at the new location.
2145 *
2146 * if the source mount is a tree, the operations explained above is
2147 * applied to each mount in the tree.
2148 * Must be called without spinlocks held, since this function can sleep
2149 * in allocations.
2150 */
2151static int attach_recursive_mnt(struct mount *source_mnt,
2152                        struct mount *dest_mnt,
2153                        struct mountpoint *dest_mp,
2154                        bool moving)
2155{
2156        struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2157        HLIST_HEAD(tree_list);
2158        struct mnt_namespace *ns = dest_mnt->mnt_ns;
2159        struct mountpoint *smp;
2160        struct mount *child, *p;
2161        struct hlist_node *n;
2162        int err;
2163
2164        /* Preallocate a mountpoint in case the new mounts need
2165         * to be tucked under other mounts.
2166         */
2167        smp = get_mountpoint(source_mnt->mnt.mnt_root);
2168        if (IS_ERR(smp))
2169                return PTR_ERR(smp);
2170
2171        /* Is there space to add these mounts to the mount namespace? */
2172        if (!moving) {
2173                err = count_mounts(ns, source_mnt);
2174                if (err)
2175                        goto out;
2176        }
2177
2178        if (IS_MNT_SHARED(dest_mnt)) {
2179                err = invent_group_ids(source_mnt, true);
2180                if (err)
2181                        goto out;
2182                err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2183                lock_mount_hash();
2184                if (err)
2185                        goto out_cleanup_ids;
2186                for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2187                        set_mnt_shared(p);
2188        } else {
2189                lock_mount_hash();
2190        }
2191        if (moving) {
2192                unhash_mnt(source_mnt);
2193                attach_mnt(source_mnt, dest_mnt, dest_mp);
2194                touch_mnt_namespace(source_mnt->mnt_ns);
2195        } else {
2196                if (source_mnt->mnt_ns) {
2197                        /* move from anon - the caller will destroy */
2198                        list_del_init(&source_mnt->mnt_ns->list);
2199                }
2200                mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2201                commit_tree(source_mnt);
2202        }
2203
2204        hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2205                struct mount *q;
2206                hlist_del_init(&child->mnt_hash);
2207                q = __lookup_mnt(&child->mnt_parent->mnt,
2208                                 child->mnt_mountpoint);
2209                if (q)
2210                        mnt_change_mountpoint(child, smp, q);
2211                /* Notice when we are propagating across user namespaces */
2212                if (child->mnt_parent->mnt_ns->user_ns != user_ns)
2213                        lock_mnt_tree(child);
2214                child->mnt.mnt_flags &= ~MNT_LOCKED;
2215                commit_tree(child);
2216        }
2217        put_mountpoint(smp);
2218        unlock_mount_hash();
2219
2220        return 0;
2221
2222 out_cleanup_ids:
2223        while (!hlist_empty(&tree_list)) {
2224                child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2225                child->mnt_parent->mnt_ns->pending_mounts = 0;
2226                umount_tree(child, UMOUNT_SYNC);
2227        }
2228        unlock_mount_hash();
2229        cleanup_group_ids(source_mnt, NULL);
2230 out:
2231        ns->pending_mounts = 0;
2232
2233        read_seqlock_excl(&mount_lock);
2234        put_mountpoint(smp);
2235        read_sequnlock_excl(&mount_lock);
2236
2237        return err;
2238}
2239
2240static struct mountpoint *lock_mount(struct path *path)
2241{
2242        struct vfsmount *mnt;
2243        struct dentry *dentry = path->dentry;
2244retry:
2245        inode_lock(dentry->d_inode);
2246        if (unlikely(cant_mount(dentry))) {
2247                inode_unlock(dentry->d_inode);
2248                return ERR_PTR(-ENOENT);
2249        }
2250        namespace_lock();
2251        mnt = lookup_mnt(path);
2252        if (likely(!mnt)) {
2253                struct mountpoint *mp = get_mountpoint(dentry);
2254                if (IS_ERR(mp)) {
2255                        namespace_unlock();
2256                        inode_unlock(dentry->d_inode);
2257                        return mp;
2258                }
2259                return mp;
2260        }
2261        namespace_unlock();
2262        inode_unlock(path->dentry->d_inode);
2263        path_put(path);
2264        path->mnt = mnt;
2265        dentry = path->dentry = dget(mnt->mnt_root);
2266        goto retry;
2267}
2268
2269static void unlock_mount(struct mountpoint *where)
2270{
2271        struct dentry *dentry = where->m_dentry;
2272
2273        read_seqlock_excl(&mount_lock);
2274        put_mountpoint(where);
2275        read_sequnlock_excl(&mount_lock);
2276
2277        namespace_unlock();
2278        inode_unlock(dentry->d_inode);
2279}
2280
2281static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2282{
2283        if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2284                return -EINVAL;
2285
2286        if (d_is_dir(mp->m_dentry) !=
2287              d_is_dir(mnt->mnt.mnt_root))
2288                return -ENOTDIR;
2289
2290        return attach_recursive_mnt(mnt, p, mp, false);
2291}
2292
2293/*
2294 * Sanity check the flags to change_mnt_propagation.
2295 */
2296
2297static int flags_to_propagation_type(int ms_flags)
2298{
2299        int type = ms_flags & ~(MS_REC | MS_SILENT);
2300
2301        /* Fail if any non-propagation flags are set */
2302        if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2303                return 0;
2304        /* Only one propagation flag should be set */
2305        if (!is_power_of_2(type))
2306                return 0;
2307        return type;
2308}
2309
2310/*
2311 * recursively change the type of the mountpoint.
2312 */
2313static int do_change_type(struct path *path, int ms_flags)
2314{
2315        struct mount *m;
2316        struct mount *mnt = real_mount(path->mnt);
2317        int recurse = ms_flags & MS_REC;
2318        int type;
2319        int err = 0;
2320
2321        if (path->dentry != path->mnt->mnt_root)
2322                return -EINVAL;
2323
2324        type = flags_to_propagation_type(ms_flags);
2325        if (!type)
2326                return -EINVAL;
2327
2328        namespace_lock();
2329        if (type == MS_SHARED) {
2330                err = invent_group_ids(mnt, recurse);
2331                if (err)
2332                        goto out_unlock;
2333        }
2334
2335        lock_mount_hash();
2336        for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2337                change_mnt_propagation(m, type);
2338        unlock_mount_hash();
2339
2340 out_unlock:
2341        namespace_unlock();
2342        return err;
2343}
2344
2345static struct mount *__do_loopback(struct path *old_path, int recurse)
2346{
2347        struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
2348
2349        if (IS_MNT_UNBINDABLE(old))
2350                return mnt;
2351
2352        if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations)
2353                return mnt;
2354
2355        if (!recurse && has_locked_children(old, old_path->dentry))
2356                return mnt;
2357
2358        if (recurse)
2359                mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
2360        else
2361                mnt = clone_mnt(old, old_path->dentry, 0);
2362
2363        if (!IS_ERR(mnt))
2364                mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2365
2366        return mnt;
2367}
2368
2369/*
2370 * do loopback mount.
2371 */
2372static int do_loopback(struct path *path, const char *old_name,
2373                                int recurse)
2374{
2375        struct path old_path;
2376        struct mount *mnt = NULL, *parent;
2377        struct mountpoint *mp;
2378        int err;
2379        if (!old_name || !*old_name)
2380                return -EINVAL;
2381        err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2382        if (err)
2383                return err;
2384
2385        err = -EINVAL;
2386        if (mnt_ns_loop(old_path.dentry))
2387                goto out;
2388
2389        mp = lock_mount(path);
2390        if (IS_ERR(mp)) {
2391                err = PTR_ERR(mp);
2392                goto out;
2393        }
2394
2395        parent = real_mount(path->mnt);
2396        if (!check_mnt(parent))
2397                goto out2;
2398
2399        mnt = __do_loopback(&old_path, recurse);
2400        if (IS_ERR(mnt)) {
2401                err = PTR_ERR(mnt);
2402                goto out2;
2403        }
2404
2405        err = graft_tree(mnt, parent, mp);
2406        if (err) {
2407                lock_mount_hash();
2408                umount_tree(mnt, UMOUNT_SYNC);
2409                unlock_mount_hash();
2410        }
2411out2:
2412        unlock_mount(mp);
2413out:
2414        path_put(&old_path);
2415        return err;
2416}
2417
2418static struct file *open_detached_copy(struct path *path, bool recursive)
2419{
2420        struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2421        struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true);
2422        struct mount *mnt, *p;
2423        struct file *file;
2424
2425        if (IS_ERR(ns))
2426                return ERR_CAST(ns);
2427
2428        namespace_lock();
2429        mnt = __do_loopback(path, recursive);
2430        if (IS_ERR(mnt)) {
2431                namespace_unlock();
2432                free_mnt_ns(ns);
2433                return ERR_CAST(mnt);
2434        }
2435
2436        lock_mount_hash();
2437        for (p = mnt; p; p = next_mnt(p, mnt)) {
2438                p->mnt_ns = ns;
2439                ns->mounts++;
2440        }
2441        ns->root = mnt;
2442        list_add_tail(&ns->list, &mnt->mnt_list);
2443        mntget(&mnt->mnt);
2444        unlock_mount_hash();
2445        namespace_unlock();
2446
2447        mntput(path->mnt);
2448        path->mnt = &mnt->mnt;
2449        file = dentry_open(path, O_PATH, current_cred());
2450        if (IS_ERR(file))
2451                dissolve_on_fput(path->mnt);
2452        else
2453                file->f_mode |= FMODE_NEED_UNMOUNT;
2454        return file;
2455}
2456
2457SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags)
2458{
2459        struct file *file;
2460        struct path path;
2461        int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
2462        bool detached = flags & OPEN_TREE_CLONE;
2463        int error;
2464        int fd;
2465
2466        BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC);
2467
2468        if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE |
2469                      AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE |
2470                      OPEN_TREE_CLOEXEC))
2471                return -EINVAL;
2472
2473        if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE)
2474                return -EINVAL;
2475
2476        if (flags & AT_NO_AUTOMOUNT)
2477                lookup_flags &= ~LOOKUP_AUTOMOUNT;
2478        if (flags & AT_SYMLINK_NOFOLLOW)
2479                lookup_flags &= ~LOOKUP_FOLLOW;
2480        if (flags & AT_EMPTY_PATH)
2481                lookup_flags |= LOOKUP_EMPTY;
2482
2483        if (detached && !may_mount())
2484                return -EPERM;
2485
2486        fd = get_unused_fd_flags(flags & O_CLOEXEC);
2487        if (fd < 0)
2488                return fd;
2489
2490        error = user_path_at(dfd, filename, lookup_flags, &path);
2491        if (unlikely(error)) {
2492                file = ERR_PTR(error);
2493        } else {
2494                if (detached)
2495                        file = open_detached_copy(&path, flags & AT_RECURSIVE);
2496                else
2497                        file = dentry_open(&path, O_PATH, current_cred());
2498                path_put(&path);
2499        }
2500        if (IS_ERR(file)) {
2501                put_unused_fd(fd);
2502                return PTR_ERR(file);
2503        }
2504        fd_install(fd, file);
2505        return fd;
2506}
2507
2508/*
2509 * Don't allow locked mount flags to be cleared.
2510 *
2511 * No locks need to be held here while testing the various MNT_LOCK
2512 * flags because those flags can never be cleared once they are set.
2513 */
2514static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags)
2515{
2516        unsigned int fl = mnt->mnt.mnt_flags;
2517
2518        if ((fl & MNT_LOCK_READONLY) &&
2519            !(mnt_flags & MNT_READONLY))
2520                return false;
2521
2522        if ((fl & MNT_LOCK_NODEV) &&
2523            !(mnt_flags & MNT_NODEV))
2524                return false;
2525
2526        if ((fl & MNT_LOCK_NOSUID) &&
2527            !(mnt_flags & MNT_NOSUID))
2528                return false;
2529
2530        if ((fl & MNT_LOCK_NOEXEC) &&
2531            !(mnt_flags & MNT_NOEXEC))
2532                return false;
2533
2534        if ((fl & MNT_LOCK_ATIME) &&
2535            ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK)))
2536                return false;
2537
2538        return true;
2539}
2540
2541static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags)
2542{
2543        bool readonly_request = (mnt_flags & MNT_READONLY);
2544
2545        if (readonly_request == __mnt_is_readonly(&mnt->mnt))
2546                return 0;
2547
2548        if (readonly_request)
2549                return mnt_make_readonly(mnt);
2550
2551        mnt->mnt.mnt_flags &= ~MNT_READONLY;
2552        return 0;
2553}
2554
2555static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags)
2556{
2557        mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2558        mnt->mnt.mnt_flags = mnt_flags;
2559        touch_mnt_namespace(mnt->mnt_ns);
2560}
2561
2562static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt)
2563{
2564        struct super_block *sb = mnt->mnt_sb;
2565
2566        if (!__mnt_is_readonly(mnt) &&
2567           (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) {
2568                char *buf = (char *)__get_free_page(GFP_KERNEL);
2569                char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM);
2570                struct tm tm;
2571
2572                time64_to_tm(sb->s_time_max, 0, &tm);
2573
2574                pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n",
2575                        sb->s_type->name,
2576                        is_mounted(mnt) ? "remounted" : "mounted",
2577                        mntpath,
2578                        tm.tm_year+1900, (unsigned long long)sb->s_time_max);
2579
2580                free_page((unsigned long)buf);
2581        }
2582}
2583
2584/*
2585 * Handle reconfiguration of the mountpoint only without alteration of the
2586 * superblock it refers to.  This is triggered by specifying MS_REMOUNT|MS_BIND
2587 * to mount(2).
2588 */
2589static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags)
2590{
2591        struct super_block *sb = path->mnt->mnt_sb;
2592        struct mount *mnt = real_mount(path->mnt);
2593        int ret;
2594
2595        if (!check_mnt(mnt))
2596                return -EINVAL;
2597
2598        if (path->dentry != mnt->mnt.mnt_root)
2599                return -EINVAL;
2600
2601        if (!can_change_locked_flags(mnt, mnt_flags))
2602                return -EPERM;
2603
2604        /*
2605         * We're only checking whether the superblock is read-only not
2606         * changing it, so only take down_read(&sb->s_umount).
2607         */
2608        down_read(&sb->s_umount);
2609        lock_mount_hash();
2610        ret = change_mount_ro_state(mnt, mnt_flags);
2611        if (ret == 0)
2612                set_mount_attributes(mnt, mnt_flags);
2613        unlock_mount_hash();
2614        up_read(&sb->s_umount);
2615
2616        mnt_warn_timestamp_expiry(path, &mnt->mnt);
2617
2618        return ret;
2619}
2620
2621/*
2622 * change filesystem flags. dir should be a physical root of filesystem.
2623 * If you've mounted a non-root directory somewhere and want to do remount
2624 * on it - tough luck.
2625 */
2626static int do_remount(struct path *path, int ms_flags, int sb_flags,
2627                      int mnt_flags, void *data)
2628{
2629        int err;
2630        struct super_block *sb = path->mnt->mnt_sb;
2631        struct mount *mnt = real_mount(path->mnt);
2632        struct fs_context *fc;
2633
2634        if (!check_mnt(mnt))
2635                return -EINVAL;
2636
2637        if (path->dentry != path->mnt->mnt_root)
2638                return -EINVAL;
2639
2640        if (!can_change_locked_flags(mnt, mnt_flags))
2641                return -EPERM;
2642
2643        fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK);
2644        if (IS_ERR(fc))
2645                return PTR_ERR(fc);
2646
2647        fc->oldapi = true;
2648        err = parse_monolithic_mount_data(fc, data);
2649        if (!err) {
2650                down_write(&sb->s_umount);
2651                err = -EPERM;
2652                if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
2653                        err = reconfigure_super(fc);
2654                        if (!err) {
2655                                lock_mount_hash();
2656                                set_mount_attributes(mnt, mnt_flags);
2657                                unlock_mount_hash();
2658                        }
2659                }
2660                up_write(&sb->s_umount);
2661        }
2662
2663        mnt_warn_timestamp_expiry(path, &mnt->mnt);
2664
2665        put_fs_context(fc);
2666        return err;
2667}
2668
2669static inline int tree_contains_unbindable(struct mount *mnt)
2670{
2671        struct mount *p;
2672        for (p = mnt; p; p = next_mnt(p, mnt)) {
2673                if (IS_MNT_UNBINDABLE(p))
2674                        return 1;
2675        }
2676        return 0;
2677}
2678
2679/*
2680 * Check that there aren't references to earlier/same mount namespaces in the
2681 * specified subtree.  Such references can act as pins for mount namespaces
2682 * that aren't checked by the mount-cycle checking code, thereby allowing
2683 * cycles to be made.
2684 */
2685static bool check_for_nsfs_mounts(struct mount *subtree)
2686{
2687        struct mount *p;
2688        bool ret = false;
2689
2690        lock_mount_hash();
2691        for (p = subtree; p; p = next_mnt(p, subtree))
2692                if (mnt_ns_loop(p->mnt.mnt_root))
2693                        goto out;
2694
2695        ret = true;
2696out:
2697        unlock_mount_hash();
2698        return ret;
2699}
2700
2701static int do_move_mount(struct path *old_path, struct path *new_path)
2702{
2703        struct mnt_namespace *ns;
2704        struct mount *p;
2705        struct mount *old;
2706        struct mount *parent;
2707        struct mountpoint *mp, *old_mp;
2708        int err;
2709        bool attached;
2710
2711        mp = lock_mount(new_path);
2712        if (IS_ERR(mp))
2713                return PTR_ERR(mp);
2714
2715        old = real_mount(old_path->mnt);
2716        p = real_mount(new_path->mnt);
2717        parent = old->mnt_parent;
2718        attached = mnt_has_parent(old);
2719        old_mp = old->mnt_mp;
2720        ns = old->mnt_ns;
2721
2722        err = -EINVAL;
2723        /* The mountpoint must be in our namespace. */
2724        if (!check_mnt(p))
2725                goto out;
2726
2727        /* The thing moved must be mounted... */
2728        if (!is_mounted(&old->mnt))
2729                goto out;
2730
2731        /* ... and either ours or the root of anon namespace */
2732        if (!(attached ? check_mnt(old) : is_anon_ns(ns)))
2733                goto out;
2734
2735        if (old->mnt.mnt_flags & MNT_LOCKED)
2736                goto out;
2737
2738        if (old_path->dentry != old_path->mnt->mnt_root)
2739                goto out;
2740
2741        if (d_is_dir(new_path->dentry) !=
2742            d_is_dir(old_path->dentry))
2743                goto out;
2744        /*
2745         * Don't move a mount residing in a shared parent.
2746         */
2747        if (attached && IS_MNT_SHARED(parent))
2748                goto out;
2749        /*
2750         * Don't move a mount tree containing unbindable mounts to a destination
2751         * mount which is shared.
2752         */
2753        if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2754                goto out;
2755        err = -ELOOP;
2756        if (!check_for_nsfs_mounts(old))
2757                goto out;
2758        for (; mnt_has_parent(p); p = p->mnt_parent)
2759                if (p == old)
2760                        goto out;
2761
2762        err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp,
2763                                   attached);
2764        if (err)
2765                goto out;
2766
2767        /* if the mount is moved, it should no longer be expire
2768         * automatically */
2769        list_del_init(&old->mnt_expire);
2770        if (attached)
2771                put_mountpoint(old_mp);
2772out:
2773        unlock_mount(mp);
2774        if (!err) {
2775                if (attached)
2776                        mntput_no_expire(parent);
2777                else
2778                        free_mnt_ns(ns);
2779        }
2780        return err;
2781}
2782
2783static int do_move_mount_old(struct path *path, const char *old_name)
2784{
2785        struct path old_path;
2786        int err;
2787
2788        if (!old_name || !*old_name)
2789                return -EINVAL;
2790
2791        err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2792        if (err)
2793                return err;
2794
2795        err = do_move_mount(&old_path, path);
2796        path_put(&old_path);
2797        return err;
2798}
2799
2800/*
2801 * add a mount into a namespace's mount tree
2802 */
2803static int do_add_mount(struct mount *newmnt, struct mountpoint *mp,
2804                        struct path *path, int mnt_flags)
2805{
2806        struct mount *parent = real_mount(path->mnt);
2807
2808        mnt_flags &= ~MNT_INTERNAL_FLAGS;
2809
2810        if (unlikely(!check_mnt(parent))) {
2811                /* that's acceptable only for automounts done in private ns */
2812                if (!(mnt_flags & MNT_SHRINKABLE))
2813                        return -EINVAL;
2814                /* ... and for those we'd better have mountpoint still alive */
2815                if (!parent->mnt_ns)
2816                        return -EINVAL;
2817        }
2818
2819        /* Refuse the same filesystem on the same mount point */
2820        if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2821            path->mnt->mnt_root == path->dentry)
2822                return -EBUSY;
2823
2824        if (d_is_symlink(newmnt->mnt.mnt_root))
2825                return -EINVAL;
2826
2827        newmnt->mnt.mnt_flags = mnt_flags;
2828        return graft_tree(newmnt, parent, mp);
2829}
2830
2831static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags);
2832
2833/*
2834 * Create a new mount using a superblock configuration and request it
2835 * be added to the namespace tree.
2836 */
2837static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
2838                           unsigned int mnt_flags)
2839{
2840        struct vfsmount *mnt;
2841        struct mountpoint *mp;
2842        struct super_block *sb = fc->root->d_sb;
2843        int error;
2844
2845        error = security_sb_kern_mount(sb);
2846        if (!error && mount_too_revealing(sb, &mnt_flags))
2847                error = -EPERM;
2848
2849        if (unlikely(error)) {
2850                fc_drop_locked(fc);
2851                return error;
2852        }
2853
2854        up_write(&sb->s_umount);
2855
2856        mnt = vfs_create_mount(fc);
2857        if (IS_ERR(mnt))
2858                return PTR_ERR(mnt);
2859
2860        mnt_warn_timestamp_expiry(mountpoint, mnt);
2861
2862        mp = lock_mount(mountpoint);
2863        if (IS_ERR(mp)) {
2864                mntput(mnt);
2865                return PTR_ERR(mp);
2866        }
2867        error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags);
2868        unlock_mount(mp);
2869        if (error < 0)
2870                mntput(mnt);
2871        return error;
2872}
2873
2874/*
2875 * create a new mount for userspace and request it to be added into the
2876 * namespace's tree
2877 */
2878static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
2879                        int mnt_flags, const char *name, void *data)
2880{
2881        struct file_system_type *type;
2882        struct fs_context *fc;
2883        const char *subtype = NULL;
2884        int err = 0;
2885
2886        if (!fstype)
2887                return -EINVAL;
2888
2889        type = get_fs_type(fstype);
2890        if (!type)
2891                return -ENODEV;
2892
2893        if (type->fs_flags & FS_HAS_SUBTYPE) {
2894                subtype = strchr(fstype, '.');
2895                if (subtype) {
2896                        subtype++;
2897                        if (!*subtype) {
2898                                put_filesystem(type);
2899                                return -EINVAL;
2900                        }
2901                }
2902        }
2903
2904        fc = fs_context_for_mount(type, sb_flags);
2905        put_filesystem(type);
2906        if (IS_ERR(fc))
2907                return PTR_ERR(fc);
2908
2909        if (subtype)
2910                err = vfs_parse_fs_string(fc, "subtype",
2911                                          subtype, strlen(subtype));
2912        if (!err && name)
2913                err = vfs_parse_fs_string(fc, "source", name, strlen(name));
2914        if (!err)
2915                err = parse_monolithic_mount_data(fc, data);
2916        if (!err && !mount_capable(fc))
2917                err = -EPERM;
2918        if (!err)
2919                err = vfs_get_tree(fc);
2920        if (!err)
2921                err = do_new_mount_fc(fc, path, mnt_flags);
2922
2923        put_fs_context(fc);
2924        return err;
2925}
2926
2927int finish_automount(struct vfsmount *m, struct path *path)
2928{
2929        struct dentry *dentry = path->dentry;
2930        struct mountpoint *mp;
2931        struct mount *mnt;
2932        int err;
2933
2934        if (!m)
2935                return 0;
2936        if (IS_ERR(m))
2937                return PTR_ERR(m);
2938
2939        mnt = real_mount(m);
2940        /* The new mount record should have at least 2 refs to prevent it being
2941         * expired before we get a chance to add it
2942         */
2943        BUG_ON(mnt_get_count(mnt) < 2);
2944
2945        if (m->mnt_sb == path->mnt->mnt_sb &&
2946            m->mnt_root == dentry) {
2947                err = -ELOOP;
2948                goto discard;
2949        }
2950
2951        /*
2952         * we don't want to use lock_mount() - in this case finding something
2953         * that overmounts our mountpoint to be means "quitely drop what we've
2954         * got", not "try to mount it on top".
2955         */
2956        inode_lock(dentry->d_inode);
2957        namespace_lock();
2958        if (unlikely(cant_mount(dentry))) {
2959                err = -ENOENT;
2960                goto discard_locked;
2961        }
2962        rcu_read_lock();
2963        if (unlikely(__lookup_mnt(path->mnt, dentry))) {
2964                rcu_read_unlock();
2965                err = 0;
2966                goto discard_locked;
2967        }
2968        rcu_read_unlock();
2969        mp = get_mountpoint(dentry);
2970        if (IS_ERR(mp)) {
2971                err = PTR_ERR(mp);
2972                goto discard_locked;
2973        }
2974
2975        err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2976        unlock_mount(mp);
2977        if (unlikely(err))
2978                goto discard;
2979        mntput(m);
2980        return 0;
2981
2982discard_locked:
2983        namespace_unlock();
2984        inode_unlock(dentry->d_inode);
2985discard:
2986        /* remove m from any expiration list it may be on */
2987        if (!list_empty(&mnt->mnt_expire)) {
2988                namespace_lock();
2989                list_del_init(&mnt->mnt_expire);
2990                namespace_unlock();
2991        }
2992        mntput(m);
2993        mntput(m);
2994        return err;
2995}
2996
2997/**
2998 * mnt_set_expiry - Put a mount on an expiration list
2999 * @mnt: The mount to list.
3000 * @expiry_list: The list to add the mount to.
3001 */
3002void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
3003{
3004        namespace_lock();
3005
3006        list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
3007
3008        namespace_unlock();
3009}
3010EXPORT_SYMBOL(mnt_set_expiry);
3011
3012/*
3013 * process a list of expirable mountpoints with the intent of discarding any
3014 * mountpoints that aren't in use and haven't been touched since last we came
3015 * here
3016 */
3017void mark_mounts_for_expiry(struct list_head *mounts)
3018{
3019        struct mount *mnt, *next;
3020        LIST_HEAD(graveyard);
3021
3022        if (list_empty(mounts))
3023                return;
3024
3025        namespace_lock();
3026        lock_mount_hash();
3027
3028        /* extract from the expiration list every vfsmount that matches the
3029         * following criteria:
3030         * - only referenced by its parent vfsmount
3031         * - still marked for expiry (marked on the last call here; marks are
3032         *   cleared by mntput())
3033         */
3034        list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
3035                if (!xchg(&mnt->mnt_expiry_mark, 1) ||
3036                        propagate_mount_busy(mnt, 1))
3037                        continue;
3038                list_move(&mnt->mnt_expire, &graveyard);
3039        }
3040        while (!list_empty(&graveyard)) {
3041                mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
3042                touch_mnt_namespace(mnt->mnt_ns);
3043                umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3044        }
3045        unlock_mount_hash();
3046        namespace_unlock();
3047}
3048
3049EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
3050
3051/*
3052 * Ripoff of 'select_parent()'
3053 *
3054 * search the list of submounts for a given mountpoint, and move any
3055 * shrinkable submounts to the 'graveyard' list.
3056 */
3057static int select_submounts(struct mount *parent, struct list_head *graveyard)
3058{
3059        struct mount *this_parent = parent;
3060        struct list_head *next;
3061        int found = 0;
3062
3063repeat:
3064        next = this_parent->mnt_mounts.next;
3065resume:
3066        while (next != &this_parent->mnt_mounts) {
3067                struct list_head *tmp = next;
3068                struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
3069
3070                next = tmp->next;
3071                if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
3072                        continue;
3073                /*
3074                 * Descend a level if the d_mounts list is non-empty.
3075                 */
3076                if (!list_empty(&mnt->mnt_mounts)) {
3077                        this_parent = mnt;
3078                        goto repeat;
3079                }
3080
3081                if (!propagate_mount_busy(mnt, 1)) {
3082                        list_move_tail(&mnt->mnt_expire, graveyard);
3083                        found++;
3084                }
3085        }
3086        /*
3087         * All done at this level ... ascend and resume the search
3088         */
3089        if (this_parent != parent) {
3090                next = this_parent->mnt_child.next;
3091                this_parent = this_parent->mnt_parent;
3092                goto resume;
3093        }
3094        return found;
3095}
3096
3097/*
3098 * process a list of expirable mountpoints with the intent of discarding any
3099 * submounts of a specific parent mountpoint
3100 *
3101 * mount_lock must be held for write
3102 */
3103static void shrink_submounts(struct mount *mnt)
3104{
3105        LIST_HEAD(graveyard);
3106        struct mount *m;
3107
3108        /* extract submounts of 'mountpoint' from the expiration list */
3109        while (select_submounts(mnt, &graveyard)) {
3110                while (!list_empty(&graveyard)) {
3111                        m = list_first_entry(&graveyard, struct mount,
3112                                                mnt_expire);
3113                        touch_mnt_namespace(m->mnt_ns);
3114                        umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3115                }
3116        }
3117}
3118
3119static void *copy_mount_options(const void __user * data)
3120{
3121        char *copy;
3122        unsigned left, offset;
3123
3124        if (!data)
3125                return NULL;
3126
3127        copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
3128        if (!copy)
3129                return ERR_PTR(-ENOMEM);
3130
3131        left = copy_from_user(copy, data, PAGE_SIZE);
3132
3133        /*
3134         * Not all architectures have an exact copy_from_user(). Resort to
3135         * byte at a time.
3136         */
3137        offset = PAGE_SIZE - left;
3138        while (left) {
3139                char c;
3140                if (get_user(c, (const char __user *)data + offset))
3141                        break;
3142                copy[offset] = c;
3143                left--;
3144                offset++;
3145        }
3146
3147        if (left == PAGE_SIZE) {
3148                kfree(copy);
3149                return ERR_PTR(-EFAULT);
3150        }
3151
3152        return copy;
3153}
3154
3155static char *copy_mount_string(const void __user *data)
3156{
3157        return data ? strndup_user(data, PATH_MAX) : NULL;
3158}
3159
3160/*
3161 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3162 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3163 *
3164 * data is a (void *) that can point to any structure up to
3165 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3166 * information (or be NULL).
3167 *
3168 * Pre-0.97 versions of mount() didn't have a flags word.
3169 * When the flags word was introduced its top half was required
3170 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3171 * Therefore, if this magic number is present, it carries no information
3172 * and must be discarded.
3173 */
3174int path_mount(const char *dev_name, struct path *path,
3175                const char *type_page, unsigned long flags, void *data_page)
3176{
3177        unsigned int mnt_flags = 0, sb_flags;
3178        int ret;
3179
3180        /* Discard magic */
3181        if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
3182                flags &= ~MS_MGC_MSK;
3183
3184        /* Basic sanity checks */
3185        if (data_page)
3186                ((char *)data_page)[PAGE_SIZE - 1] = 0;
3187
3188        if (flags & MS_NOUSER)
3189                return -EINVAL;
3190
3191        ret = security_sb_mount(dev_name, path, type_page, flags, data_page);
3192        if (ret)
3193                return ret;
3194        if (!may_mount())
3195                return -EPERM;
3196        if ((flags & SB_MANDLOCK) && !may_mandlock())
3197                return -EPERM;
3198
3199        /* Default to relatime unless overriden */
3200        if (!(flags & MS_NOATIME))
3201                mnt_flags |= MNT_RELATIME;
3202
3203        /* Separate the per-mountpoint flags */
3204        if (flags & MS_NOSUID)
3205                mnt_flags |= MNT_NOSUID;
3206        if (flags & MS_NODEV)
3207                mnt_flags |= MNT_NODEV;
3208        if (flags & MS_NOEXEC)
3209                mnt_flags |= MNT_NOEXEC;
3210        if (flags & MS_NOATIME)
3211                mnt_flags |= MNT_NOATIME;
3212        if (flags & MS_NODIRATIME)
3213                mnt_flags |= MNT_NODIRATIME;
3214        if (flags & MS_STRICTATIME)
3215                mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
3216        if (flags & MS_RDONLY)
3217                mnt_flags |= MNT_READONLY;
3218        if (flags & MS_NOSYMFOLLOW)
3219                mnt_flags |= MNT_NOSYMFOLLOW;
3220
3221        /* The default atime for remount is preservation */
3222        if ((flags & MS_REMOUNT) &&
3223            ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
3224                       MS_STRICTATIME)) == 0)) {
3225                mnt_flags &= ~MNT_ATIME_MASK;
3226                mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK;
3227        }
3228
3229        sb_flags = flags & (SB_RDONLY |
3230                            SB_SYNCHRONOUS |
3231                            SB_MANDLOCK |
3232                            SB_DIRSYNC |
3233                            SB_SILENT |
3234                            SB_POSIXACL |
3235                            SB_LAZYTIME |
3236                            SB_I_VERSION);
3237
3238        if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND))
3239                return do_reconfigure_mnt(path, mnt_flags);
3240        if (flags & MS_REMOUNT)
3241                return do_remount(path, flags, sb_flags, mnt_flags, data_page);
3242        if (flags & MS_BIND)
3243                return do_loopback(path, dev_name, flags & MS_REC);
3244        if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
3245                return do_change_type(path, flags);
3246        if (flags & MS_MOVE)
3247                return do_move_mount_old(path, dev_name);
3248
3249        return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name,
3250                            data_page);
3251}
3252
3253long do_mount(const char *dev_name, const char __user *dir_name,
3254                const char *type_page, unsigned long flags, void *data_page)
3255{
3256        struct path path;
3257        int ret;
3258
3259        ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path);
3260        if (ret)
3261                return ret;
3262        ret = path_mount(dev_name, &path, type_page, flags, data_page);
3263        path_put(&path);
3264        return ret;
3265}
3266
3267static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
3268{
3269        return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
3270}
3271
3272static void dec_mnt_namespaces(struct ucounts *ucounts)
3273{
3274        dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
3275}
3276
3277static void free_mnt_ns(struct mnt_namespace *ns)
3278{
3279        if (!is_anon_ns(ns))
3280                ns_free_inum(&ns->ns);
3281        dec_mnt_namespaces(ns->ucounts);
3282        put_user_ns(ns->user_ns);
3283        kfree(ns);
3284}
3285
3286/*
3287 * Assign a sequence number so we can detect when we attempt to bind
3288 * mount a reference to an older mount namespace into the current
3289 * mount namespace, preventing reference counting loops.  A 64bit
3290 * number incrementing at 10Ghz will take 12,427 years to wrap which
3291 * is effectively never, so we can ignore the possibility.
3292 */
3293static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
3294
3295static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon)
3296{
3297        struct mnt_namespace *new_ns;
3298        struct ucounts *ucounts;
3299        int ret;
3300
3301        ucounts = inc_mnt_namespaces(user_ns);
3302        if (!ucounts)
3303                return ERR_PTR(-ENOSPC);
3304
3305        new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
3306        if (!new_ns) {
3307                dec_mnt_namespaces(ucounts);
3308                return ERR_PTR(-ENOMEM);
3309        }
3310        if (!anon) {
3311                ret = ns_alloc_inum(&new_ns->ns);
3312                if (ret) {
3313                        kfree(new_ns);
3314                        dec_mnt_namespaces(ucounts);
3315                        return ERR_PTR(ret);
3316                }
3317        }
3318        new_ns->ns.ops = &mntns_operations;
3319        if (!anon)
3320                new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
3321        refcount_set(&new_ns->ns.count, 1);
3322        INIT_LIST_HEAD(&new_ns->list);
3323        init_waitqueue_head(&new_ns->poll);
3324        spin_lock_init(&new_ns->ns_lock);
3325        new_ns->user_ns = get_user_ns(user_ns);
3326        new_ns->ucounts = ucounts;
3327        return new_ns;
3328}
3329
3330__latent_entropy
3331struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
3332                struct user_namespace *user_ns, struct fs_struct *new_fs)
3333{
3334        struct mnt_namespace *new_ns;
3335        struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
3336        struct mount *p, *q;
3337        struct mount *old;
3338        struct mount *new;
3339        int copy_flags;
3340
3341        BUG_ON(!ns);
3342
3343        if (likely(!(flags & CLONE_NEWNS))) {
3344                get_mnt_ns(ns);
3345                return ns;
3346        }
3347
3348        old = ns->root;
3349
3350        new_ns = alloc_mnt_ns(user_ns, false);
3351        if (IS_ERR(new_ns))
3352                return new_ns;
3353
3354        namespace_lock();
3355        /* First pass: copy the tree topology */
3356        copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
3357        if (user_ns != ns->user_ns)
3358                copy_flags |= CL_SHARED_TO_SLAVE;
3359        new = copy_tree(old, old->mnt.mnt_root, copy_flags);
3360        if (IS_ERR(new)) {
3361                namespace_unlock();
3362                free_mnt_ns(new_ns);
3363                return ERR_CAST(new);
3364        }
3365        if (user_ns != ns->user_ns) {
3366                lock_mount_hash();
3367                lock_mnt_tree(new);
3368                unlock_mount_hash();
3369        }
3370        new_ns->root = new;
3371        list_add_tail(&new_ns->list, &new->mnt_list);
3372
3373        /*
3374         * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3375         * as belonging to new namespace.  We have already acquired a private
3376         * fs_struct, so tsk->fs->lock is not needed.
3377         */
3378        p = old;
3379        q = new;
3380        while (p) {
3381                q->mnt_ns = new_ns;
3382                new_ns->mounts++;
3383                if (new_fs) {
3384                        if (&p->mnt == new_fs->root.mnt) {
3385                                new_fs->root.mnt = mntget(&q->mnt);
3386                                rootmnt = &p->mnt;
3387                        }
3388                        if (&p->mnt == new_fs->pwd.mnt) {
3389                                new_fs->pwd.mnt = mntget(&q->mnt);
3390                                pwdmnt = &p->mnt;
3391                        }
3392                }
3393                p = next_mnt(p, old);
3394                q = next_mnt(q, new);
3395                if (!q)
3396                        break;
3397                while (p->mnt.mnt_root != q->mnt.mnt_root)
3398                        p = next_mnt(p, old);
3399        }
3400        namespace_unlock();
3401
3402        if (rootmnt)
3403                mntput(rootmnt);
3404        if (pwdmnt)
3405                mntput(pwdmnt);
3406
3407        return new_ns;
3408}
3409
3410struct dentry *mount_subtree(struct vfsmount *m, const char *name)
3411{
3412        struct mount *mnt = real_mount(m);
3413        struct mnt_namespace *ns;
3414        struct super_block *s;
3415        struct path path;
3416        int err;
3417
3418        ns = alloc_mnt_ns(&init_user_ns, true);
3419        if (IS_ERR(ns)) {
3420                mntput(m);
3421                return ERR_CAST(ns);
3422        }
3423        mnt->mnt_ns = ns;
3424        ns->root = mnt;
3425        ns->mounts++;
3426        list_add(&mnt->mnt_list, &ns->list);
3427
3428        err = vfs_path_lookup(m->mnt_root, m,
3429                        name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3430
3431        put_mnt_ns(ns);
3432
3433        if (err)
3434                return ERR_PTR(err);
3435
3436        /* trade a vfsmount reference for active sb one */
3437        s = path.mnt->mnt_sb;
3438        atomic_inc(&s->s_active);
3439        mntput(path.mnt);
3440        /* lock the sucker */
3441        down_write(&s->s_umount);
3442        /* ... and return the root of (sub)tree on it */
3443        return path.dentry;
3444}
3445EXPORT_SYMBOL(mount_subtree);
3446
3447SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3448                char __user *, type, unsigned long, flags, void __user *, data)
3449{
3450        int ret;
3451        char *kernel_type;
3452        char *kernel_dev;
3453        void *options;
3454
3455        kernel_type = copy_mount_string(type);
3456        ret = PTR_ERR(kernel_type);
3457        if (IS_ERR(kernel_type))
3458                goto out_type;
3459
3460        kernel_dev = copy_mount_string(dev_name);
3461        ret = PTR_ERR(kernel_dev);
3462        if (IS_ERR(kernel_dev))
3463                goto out_dev;
3464
3465        options = copy_mount_options(data);
3466        ret = PTR_ERR(options);
3467        if (IS_ERR(options))
3468                goto out_data;
3469
3470        ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3471
3472        kfree(options);
3473out_data:
3474        kfree(kernel_dev);
3475out_dev:
3476        kfree(kernel_type);
3477out_type:
3478        return ret;
3479}
3480
3481#define FSMOUNT_VALID_FLAGS \
3482        (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \
3483         MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME)
3484
3485#define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP)
3486
3487#define MOUNT_SETATTR_PROPAGATION_FLAGS \
3488        (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED)
3489
3490static unsigned int attr_flags_to_mnt_flags(u64 attr_flags)
3491{
3492        unsigned int mnt_flags = 0;
3493
3494        if (attr_flags & MOUNT_ATTR_RDONLY)
3495                mnt_flags |= MNT_READONLY;
3496        if (attr_flags & MOUNT_ATTR_NOSUID)
3497                mnt_flags |= MNT_NOSUID;
3498        if (attr_flags & MOUNT_ATTR_NODEV)
3499                mnt_flags |= MNT_NODEV;
3500        if (attr_flags & MOUNT_ATTR_NOEXEC)
3501                mnt_flags |= MNT_NOEXEC;
3502        if (attr_flags & MOUNT_ATTR_NODIRATIME)
3503                mnt_flags |= MNT_NODIRATIME;
3504
3505        return mnt_flags;
3506}
3507
3508/*
3509 * Create a kernel mount representation for a new, prepared superblock
3510 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3511 */
3512SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
3513                unsigned int, attr_flags)
3514{
3515        struct mnt_namespace *ns;
3516        struct fs_context *fc;
3517        struct file *file;
3518        struct path newmount;
3519        struct mount *mnt;
3520        struct fd f;
3521        unsigned int mnt_flags = 0;
3522        long ret;
3523
3524        if (!may_mount())
3525                return -EPERM;
3526
3527        if ((flags & ~(FSMOUNT_CLOEXEC)) != 0)
3528                return -EINVAL;
3529
3530        if (attr_flags & ~FSMOUNT_VALID_FLAGS)
3531                return -EINVAL;
3532
3533        mnt_flags = attr_flags_to_mnt_flags(attr_flags);
3534
3535        switch (attr_flags & MOUNT_ATTR__ATIME) {
3536        case MOUNT_ATTR_STRICTATIME:
3537                break;
3538        case MOUNT_ATTR_NOATIME:
3539                mnt_flags |= MNT_NOATIME;
3540                break;
3541        case MOUNT_ATTR_RELATIME:
3542                mnt_flags |= MNT_RELATIME;
3543                break;
3544        default:
3545                return -EINVAL;
3546        }
3547
3548        f = fdget(fs_fd);
3549        if (!f.file)
3550                return -EBADF;
3551
3552        ret = -EINVAL;
3553        if (f.file->f_op != &fscontext_fops)
3554                goto err_fsfd;
3555
3556        fc = f.file->private_data;
3557
3558        ret = mutex_lock_interruptible(&fc->uapi_mutex);
3559        if (ret < 0)
3560                goto err_fsfd;
3561
3562        /* There must be a valid superblock or we can't mount it */
3563        ret = -EINVAL;
3564        if (!fc->root)
3565                goto err_unlock;
3566
3567        ret = -EPERM;
3568        if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) {
3569                pr_warn("VFS: Mount too revealing\n");
3570                goto err_unlock;
3571        }
3572
3573        ret = -EBUSY;
3574        if (fc->phase != FS_CONTEXT_AWAITING_MOUNT)
3575                goto err_unlock;
3576
3577        ret = -EPERM;
3578        if ((fc->sb_flags & SB_MANDLOCK) && !may_mandlock())
3579                goto err_unlock;
3580
3581        newmount.mnt = vfs_create_mount(fc);
3582        if (IS_ERR(newmount.mnt)) {
3583                ret = PTR_ERR(newmount.mnt);
3584                goto err_unlock;
3585        }
3586        newmount.dentry = dget(fc->root);
3587        newmount.mnt->mnt_flags = mnt_flags;
3588
3589        /* We've done the mount bit - now move the file context into more or
3590         * less the same state as if we'd done an fspick().  We don't want to
3591         * do any memory allocation or anything like that at this point as we
3592         * don't want to have to handle any errors incurred.
3593         */
3594        vfs_clean_context(fc);
3595
3596        ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true);
3597        if (IS_ERR(ns)) {
3598                ret = PTR_ERR(ns);
3599                goto err_path;
3600        }
3601        mnt = real_mount(newmount.mnt);
3602        mnt->mnt_ns = ns;
3603        ns->root = mnt;
3604        ns->mounts = 1;
3605        list_add(&mnt->mnt_list, &ns->list);
3606        mntget(newmount.mnt);
3607
3608        /* Attach to an apparent O_PATH fd with a note that we need to unmount
3609         * it, not just simply put it.
3610         */
3611        file = dentry_open(&newmount, O_PATH, fc->cred);
3612        if (IS_ERR(file)) {
3613                dissolve_on_fput(newmount.mnt);
3614                ret = PTR_ERR(file);
3615                goto err_path;
3616        }
3617        file->f_mode |= FMODE_NEED_UNMOUNT;
3618
3619        ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
3620        if (ret >= 0)
3621                fd_install(ret, file);
3622        else
3623                fput(file);
3624
3625err_path:
3626        path_put(&newmount);
3627err_unlock:
3628        mutex_unlock(&fc->uapi_mutex);
3629err_fsfd:
3630        fdput(f);
3631        return ret;
3632}
3633
3634/*
3635 * Move a mount from one place to another.  In combination with
3636 * fsopen()/fsmount() this is used to install a new mount and in combination
3637 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
3638 * a mount subtree.
3639 *
3640 * Note the flags value is a combination of MOVE_MOUNT_* flags.
3641 */
3642SYSCALL_DEFINE5(move_mount,
3643                int, from_dfd, const char __user *, from_pathname,
3644                int, to_dfd, const char __user *, to_pathname,
3645                unsigned int, flags)
3646{
3647        struct path from_path, to_path;
3648        unsigned int lflags;
3649        int ret = 0;
3650
3651        if (!may_mount())
3652                return -EPERM;
3653
3654        if (flags & ~MOVE_MOUNT__MASK)
3655                return -EINVAL;
3656
3657        /* If someone gives a pathname, they aren't permitted to move
3658         * from an fd that requires unmount as we can't get at the flag
3659         * to clear it afterwards.
3660         */
3661        lflags = 0;
3662        if (flags & MOVE_MOUNT_F_SYMLINKS)      lflags |= LOOKUP_FOLLOW;
3663        if (flags & MOVE_MOUNT_F_AUTOMOUNTS)    lflags |= LOOKUP_AUTOMOUNT;
3664        if (flags & MOVE_MOUNT_F_EMPTY_PATH)    lflags |= LOOKUP_EMPTY;
3665
3666        ret = user_path_at(from_dfd, from_pathname, lflags, &from_path);
3667        if (ret < 0)
3668                return ret;
3669
3670        lflags = 0;
3671        if (flags & MOVE_MOUNT_T_SYMLINKS)      lflags |= LOOKUP_FOLLOW;
3672        if (flags & MOVE_MOUNT_T_AUTOMOUNTS)    lflags |= LOOKUP_AUTOMOUNT;
3673        if (flags & MOVE_MOUNT_T_EMPTY_PATH)    lflags |= LOOKUP_EMPTY;
3674
3675        ret = user_path_at(to_dfd, to_pathname, lflags, &to_path);
3676        if (ret < 0)
3677                goto out_from;
3678
3679        ret = security_move_mount(&from_path, &to_path);
3680        if (ret < 0)
3681                goto out_to;
3682
3683        ret = do_move_mount(&from_path, &to_path);
3684
3685out_to:
3686        path_put(&to_path);
3687out_from:
3688        path_put(&from_path);
3689        return ret;
3690}
3691
3692/*
3693 * Return true if path is reachable from root
3694 *
3695 * namespace_sem or mount_lock is held
3696 */
3697bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3698                         const struct path *root)
3699{
3700        while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3701                dentry = mnt->mnt_mountpoint;
3702                mnt = mnt->mnt_parent;
3703        }
3704        return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3705}
3706
3707bool path_is_under(const struct path *path1, const struct path *path2)
3708{
3709        bool res;
3710        read_seqlock_excl(&mount_lock);
3711        res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3712        read_sequnlock_excl(&mount_lock);
3713        return res;
3714}
3715EXPORT_SYMBOL(path_is_under);
3716
3717/*
3718 * pivot_root Semantics:
3719 * Moves the root file system of the current process to the directory put_old,
3720 * makes new_root as the new root file system of the current process, and sets
3721 * root/cwd of all processes which had them on the current root to new_root.
3722 *
3723 * Restrictions:
3724 * The new_root and put_old must be directories, and  must not be on the
3725 * same file  system as the current process root. The put_old  must  be
3726 * underneath new_root,  i.e. adding a non-zero number of /.. to the string
3727 * pointed to by put_old must yield the same directory as new_root. No other
3728 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3729 *
3730 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3731 * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives
3732 * in this situation.
3733 *
3734 * Notes:
3735 *  - we don't move root/cwd if they are not at the root (reason: if something
3736 *    cared enough to change them, it's probably wrong to force them elsewhere)
3737 *  - it's okay to pick a root that isn't the root of a file system, e.g.
3738 *    /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3739 *    though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3740 *    first.
3741 */
3742SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3743                const char __user *, put_old)
3744{
3745        struct path new, old, root;
3746        struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
3747        struct mountpoint *old_mp, *root_mp;
3748        int error;
3749
3750        if (!may_mount())
3751                return -EPERM;
3752
3753        error = user_path_at(AT_FDCWD, new_root,
3754                             LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new);
3755        if (error)
3756                goto out0;
3757
3758        error = user_path_at(AT_FDCWD, put_old,
3759                             LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old);
3760        if (error)
3761                goto out1;
3762
3763        error = security_sb_pivotroot(&old, &new);
3764        if (error)
3765                goto out2;
3766
3767        get_fs_root(current->fs, &root);
3768        old_mp = lock_mount(&old);
3769        error = PTR_ERR(old_mp);
3770        if (IS_ERR(old_mp))
3771                goto out3;
3772
3773        error = -EINVAL;
3774        new_mnt = real_mount(new.mnt);
3775        root_mnt = real_mount(root.mnt);
3776        old_mnt = real_mount(old.mnt);
3777        ex_parent = new_mnt->mnt_parent;
3778        root_parent = root_mnt->mnt_parent;
3779        if (IS_MNT_SHARED(old_mnt) ||
3780                IS_MNT_SHARED(ex_parent) ||
3781                IS_MNT_SHARED(root_parent))
3782                goto out4;
3783        if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3784                goto out4;
3785        if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3786                goto out4;
3787        error = -ENOENT;
3788        if (d_unlinked(new.dentry))
3789                goto out4;
3790        error = -EBUSY;
3791        if (new_mnt == root_mnt || old_mnt == root_mnt)
3792                goto out4; /* loop, on the same file system  */
3793        error = -EINVAL;
3794        if (root.mnt->mnt_root != root.dentry)
3795                goto out4; /* not a mountpoint */
3796        if (!mnt_has_parent(root_mnt))
3797                goto out4; /* not attached */
3798        if (new.mnt->mnt_root != new.dentry)
3799                goto out4; /* not a mountpoint */
3800        if (!mnt_has_parent(new_mnt))
3801                goto out4; /* not attached */
3802        /* make sure we can reach put_old from new_root */
3803        if (!is_path_reachable(old_mnt, old.dentry, &new))
3804                goto out4;
3805        /* make certain new is below the root */
3806        if (!is_path_reachable(new_mnt, new.dentry, &root))
3807                goto out4;
3808        lock_mount_hash();
3809        umount_mnt(new_mnt);
3810        root_mp = unhash_mnt(root_mnt);  /* we'll need its mountpoint */
3811        if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3812                new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3813                root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3814        }
3815        /* mount old root on put_old */
3816        attach_mnt(root_mnt, old_mnt, old_mp);
3817        /* mount new_root on / */
3818        attach_mnt(new_mnt, root_parent, root_mp);
3819        mnt_add_count(root_parent, -1);
3820        touch_mnt_namespace(current->nsproxy->mnt_ns);
3821        /* A moved mount should not expire automatically */
3822        list_del_init(&new_mnt->mnt_expire);
3823        put_mountpoint(root_mp);
3824        unlock_mount_hash();
3825        chroot_fs_refs(&root, &new);
3826        error = 0;
3827out4:
3828        unlock_mount(old_mp);
3829        if (!error)
3830                mntput_no_expire(ex_parent);
3831out3:
3832        path_put(&root);
3833out2:
3834        path_put(&old);
3835out1:
3836        path_put(&new);
3837out0:
3838        return error;
3839}
3840
3841static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt)
3842{
3843        unsigned int flags = mnt->mnt.mnt_flags;
3844
3845        /*  flags to clear */
3846        flags &= ~kattr->attr_clr;
3847        /* flags to raise */
3848        flags |= kattr->attr_set;
3849
3850        return flags;
3851}
3852
3853static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
3854{
3855        struct vfsmount *m = &mnt->mnt;
3856
3857        if (!kattr->mnt_userns)
3858                return 0;
3859
3860        /*
3861         * Once a mount has been idmapped we don't allow it to change its
3862         * mapping. It makes things simpler and callers can just create
3863         * another bind-mount they can idmap if they want to.
3864         */
3865        if (mnt_user_ns(m) != &init_user_ns)
3866                return -EPERM;
3867
3868        /* The underlying filesystem doesn't support idmapped mounts yet. */
3869        if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP))
3870                return -EINVAL;
3871
3872        /* Don't yet support filesystem mountable in user namespaces. */
3873        if (m->mnt_sb->s_user_ns != &init_user_ns)
3874                return -EINVAL;
3875
3876        /* We're not controlling the superblock. */
3877        if (!capable(CAP_SYS_ADMIN))
3878                return -EPERM;
3879
3880        /* Mount has already been visible in the filesystem hierarchy. */
3881        if (!is_anon_ns(mnt->mnt_ns))
3882                return -EINVAL;
3883
3884        return 0;
3885}
3886
3887static struct mount *mount_setattr_prepare(struct mount_kattr *kattr,
3888                                           struct mount *mnt, int *err)
3889{
3890        struct mount *m = mnt, *last = NULL;
3891
3892        if (!is_mounted(&m->mnt)) {
3893                *err = -EINVAL;
3894                goto out;
3895        }
3896
3897        if (!(mnt_has_parent(m) ? check_mnt(m) : is_anon_ns(m->mnt_ns))) {
3898                *err = -EINVAL;
3899                goto out;
3900        }
3901
3902        do {
3903                unsigned int flags;
3904
3905                flags = recalc_flags(kattr, m);
3906                if (!can_change_locked_flags(m, flags)) {
3907                        *err = -EPERM;
3908                        goto out;
3909                }
3910
3911                *err = can_idmap_mount(kattr, m);
3912                if (*err)
3913                        goto out;
3914
3915                last = m;
3916
3917                if ((kattr->attr_set & MNT_READONLY) &&
3918                    !(m->mnt.mnt_flags & MNT_READONLY)) {
3919                        *err = mnt_hold_writers(m);
3920                        if (*err)
3921                                goto out;
3922                }
3923        } while (kattr->recurse && (m = next_mnt(m, mnt)));
3924
3925out:
3926        return last;
3927}
3928
3929static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
3930{
3931        struct user_namespace *mnt_userns;
3932
3933        if (!kattr->mnt_userns)
3934                return;
3935
3936        mnt_userns = get_user_ns(kattr->mnt_userns);
3937        /* Pairs with smp_load_acquire() in mnt_user_ns(). */
3938        smp_store_release(&mnt->mnt.mnt_userns, mnt_userns);
3939}
3940
3941static void mount_setattr_commit(struct mount_kattr *kattr,
3942                                 struct mount *mnt, struct mount *last,
3943                                 int err)
3944{
3945        struct mount *m = mnt;
3946
3947        do {
3948                if (!err) {
3949                        unsigned int flags;
3950
3951                        do_idmap_mount(kattr, m);
3952                        flags = recalc_flags(kattr, m);
3953                        WRITE_ONCE(m->mnt.mnt_flags, flags);
3954                }
3955
3956                /*
3957                 * We either set MNT_READONLY above so make it visible
3958                 * before ~MNT_WRITE_HOLD or we failed to recursively
3959                 * apply mount options.
3960                 */
3961                if ((kattr->attr_set & MNT_READONLY) &&
3962                    (m->mnt.mnt_flags & MNT_WRITE_HOLD))
3963                        mnt_unhold_writers(m);
3964
3965                if (!err && kattr->propagation)
3966                        change_mnt_propagation(m, kattr->propagation);
3967
3968                /*
3969                 * On failure, only cleanup until we found the first mount
3970                 * we failed to handle.
3971                 */
3972                if (err && m == last)
3973                        break;
3974        } while (kattr->recurse && (m = next_mnt(m, mnt)));
3975
3976        if (!err)
3977                touch_mnt_namespace(mnt->mnt_ns);
3978}
3979
3980static int do_mount_setattr(struct path *path, struct mount_kattr *kattr)
3981{
3982        struct mount *mnt = real_mount(path->mnt), *last = NULL;
3983        int err = 0;
3984
3985        if (path->dentry != mnt->mnt.mnt_root)
3986                return -EINVAL;
3987
3988        if (kattr->propagation) {
3989                /*
3990                 * Only take namespace_lock() if we're actually changing
3991                 * propagation.
3992                 */
3993                namespace_lock();
3994                if (kattr->propagation == MS_SHARED) {
3995                        err = invent_group_ids(mnt, kattr->recurse);
3996                        if (err) {
3997                                namespace_unlock();
3998                                return err;
3999                        }
4000                }
4001        }
4002
4003        lock_mount_hash();
4004
4005        /*
4006         * Get the mount tree in a shape where we can change mount
4007         * properties without failure.
4008         */
4009        last = mount_setattr_prepare(kattr, mnt, &err);
4010        if (last) /* Commit all changes or revert to the old state. */
4011                mount_setattr_commit(kattr, mnt, last, err);
4012
4013        unlock_mount_hash();
4014
4015        if (kattr->propagation) {
4016                namespace_unlock();
4017                if (err)
4018                        cleanup_group_ids(mnt, NULL);
4019        }
4020
4021        return err;
4022}
4023
4024static int build_mount_idmapped(const struct mount_attr *attr, size_t usize,
4025                                struct mount_kattr *kattr, unsigned int flags)
4026{
4027        int err = 0;
4028        struct ns_common *ns;
4029        struct user_namespace *mnt_userns;
4030        struct file *file;
4031
4032        if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP))
4033                return 0;
4034
4035        /*
4036         * We currently do not support clearing an idmapped mount. If this ever
4037         * is a use-case we can revisit this but for now let's keep it simple
4038         * and not allow it.
4039         */
4040        if (attr->attr_clr & MOUNT_ATTR_IDMAP)
4041                return -EINVAL;
4042
4043        if (attr->userns_fd > INT_MAX)
4044                return -EINVAL;
4045
4046        file = fget(attr->userns_fd);
4047        if (!file)
4048                return -EBADF;
4049
4050        if (!proc_ns_file(file)) {
4051                err = -EINVAL;
4052                goto out_fput;
4053        }
4054
4055        ns = get_proc_ns(file_inode(file));
4056        if (ns->ops->type != CLONE_NEWUSER) {
4057                err = -EINVAL;
4058                goto out_fput;
4059        }
4060
4061        /*
4062         * The init_user_ns is used to indicate that a vfsmount is not idmapped.
4063         * This is simpler than just having to treat NULL as unmapped. Users
4064         * wanting to idmap a mount to init_user_ns can just use a namespace
4065         * with an identity mapping.
4066         */
4067        mnt_userns = container_of(ns, struct user_namespace, ns);
4068        if (mnt_userns == &init_user_ns) {
4069                err = -EPERM;
4070                goto out_fput;
4071        }
4072        kattr->mnt_userns = get_user_ns(mnt_userns);
4073
4074out_fput:
4075        fput(file);
4076        return err;
4077}
4078
4079static int build_mount_kattr(const struct mount_attr *attr, size_t usize,
4080                             struct mount_kattr *kattr, unsigned int flags)
4081{
4082        unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
4083
4084        if (flags & AT_NO_AUTOMOUNT)
4085                lookup_flags &= ~LOOKUP_AUTOMOUNT;
4086        if (flags & AT_SYMLINK_NOFOLLOW)
4087                lookup_flags &= ~LOOKUP_FOLLOW;
4088        if (flags & AT_EMPTY_PATH)
4089                lookup_flags |= LOOKUP_EMPTY;
4090
4091        *kattr = (struct mount_kattr) {
4092                .lookup_flags   = lookup_flags,
4093                .recurse        = !!(flags & AT_RECURSIVE),
4094        };
4095
4096        if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS)
4097                return -EINVAL;
4098        if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1)
4099                return -EINVAL;
4100        kattr->propagation = attr->propagation;
4101
4102        if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS)
4103                return -EINVAL;
4104
4105        kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set);
4106        kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr);
4107
4108        /*
4109         * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap,
4110         * users wanting to transition to a different atime setting cannot
4111         * simply specify the atime setting in @attr_set, but must also
4112         * specify MOUNT_ATTR__ATIME in the @attr_clr field.
4113         * So ensure that MOUNT_ATTR__ATIME can't be partially set in
4114         * @attr_clr and that @attr_set can't have any atime bits set if
4115         * MOUNT_ATTR__ATIME isn't set in @attr_clr.
4116         */
4117        if (attr->attr_clr & MOUNT_ATTR__ATIME) {
4118                if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME)
4119                        return -EINVAL;
4120
4121                /*
4122                 * Clear all previous time settings as they are mutually
4123                 * exclusive.
4124                 */
4125                kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME;
4126                switch (attr->attr_set & MOUNT_ATTR__ATIME) {
4127                case MOUNT_ATTR_RELATIME:
4128                        kattr->attr_set |= MNT_RELATIME;
4129                        break;
4130                case MOUNT_ATTR_NOATIME:
4131                        kattr->attr_set |= MNT_NOATIME;
4132                        break;
4133                case MOUNT_ATTR_STRICTATIME:
4134                        break;
4135                default:
4136                        return -EINVAL;
4137                }
4138        } else {
4139                if (attr->attr_set & MOUNT_ATTR__ATIME)
4140                        return -EINVAL;
4141        }
4142
4143        return build_mount_idmapped(attr, usize, kattr, flags);
4144}
4145
4146static void finish_mount_kattr(struct mount_kattr *kattr)
4147{
4148        put_user_ns(kattr->mnt_userns);
4149        kattr->mnt_userns = NULL;
4150}
4151
4152SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path,
4153                unsigned int, flags, struct mount_attr __user *, uattr,
4154                size_t, usize)
4155{
4156        int err;
4157        struct path target;
4158        struct mount_attr attr;
4159        struct mount_kattr kattr;
4160
4161        BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0);
4162
4163        if (flags & ~(AT_EMPTY_PATH |
4164                      AT_RECURSIVE |
4165                      AT_SYMLINK_NOFOLLOW |
4166                      AT_NO_AUTOMOUNT))
4167                return -EINVAL;
4168
4169        if (unlikely(usize > PAGE_SIZE))
4170                return -E2BIG;
4171        if (unlikely(usize < MOUNT_ATTR_SIZE_VER0))
4172                return -EINVAL;
4173
4174        if (!may_mount())
4175                return -EPERM;
4176
4177        err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize);
4178        if (err)
4179                return err;
4180
4181        /* Don't bother walking through the mounts if this is a nop. */
4182        if (attr.attr_set == 0 &&
4183            attr.attr_clr == 0 &&
4184            attr.propagation == 0)
4185                return 0;
4186
4187        err = build_mount_kattr(&attr, usize, &kattr, flags);
4188        if (err)
4189                return err;
4190
4191        err = user_path_at(dfd, path, kattr.lookup_flags, &target);
4192        if (err)
4193                return err;
4194
4195        err = do_mount_setattr(&target, &kattr);
4196        finish_mount_kattr(&kattr);
4197        path_put(&target);
4198        return err;
4199}
4200
4201static void __init init_mount_tree(void)
4202{
4203        struct vfsmount *mnt;
4204        struct mount *m;
4205        struct mnt_namespace *ns;
4206        struct path root;
4207
4208        mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL);
4209        if (IS_ERR(mnt))
4210                panic("Can't create rootfs");
4211
4212        ns = alloc_mnt_ns(&init_user_ns, false);
4213        if (IS_ERR(ns))
4214                panic("Can't allocate initial namespace");
4215        m = real_mount(mnt);
4216        m->mnt_ns = ns;
4217        ns->root = m;
4218        ns->mounts = 1;
4219        list_add(&m->mnt_list, &ns->list);
4220        init_task.nsproxy->mnt_ns = ns;
4221        get_mnt_ns(ns);
4222
4223        root.mnt = mnt;
4224        root.dentry = mnt->mnt_root;
4225        mnt->mnt_flags |= MNT_LOCKED;
4226
4227        set_fs_pwd(current->fs, &root);
4228        set_fs_root(current->fs, &root);
4229}
4230
4231void __init mnt_init(void)
4232{
4233        int err;
4234
4235        mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
4236                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
4237
4238        mount_hashtable = alloc_large_system_hash("Mount-cache",
4239                                sizeof(struct hlist_head),
4240                                mhash_entries, 19,
4241                                HASH_ZERO,
4242                                &m_hash_shift, &m_hash_mask, 0, 0);
4243        mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
4244                                sizeof(struct hlist_head),
4245                                mphash_entries, 19,
4246                                HASH_ZERO,
4247                                &mp_hash_shift, &mp_hash_mask, 0, 0);
4248
4249        if (!mount_hashtable || !mountpoint_hashtable)
4250                panic("Failed to allocate mount hash table\n");
4251
4252        kernfs_init();
4253
4254        err = sysfs_init();
4255        if (err)
4256                printk(KERN_WARNING "%s: sysfs_init error: %d\n",
4257                        __func__, err);
4258        fs_kobj = kobject_create_and_add("fs", NULL);
4259        if (!fs_kobj)
4260                printk(KERN_WARNING "%s: kobj create error\n", __func__);
4261        shmem_init();
4262        init_rootfs();
4263        init_mount_tree();
4264}
4265
4266void put_mnt_ns(struct mnt_namespace *ns)
4267{
4268        if (!refcount_dec_and_test(&ns->ns.count))
4269                return;
4270        drop_collected_mounts(&ns->root->mnt);
4271        free_mnt_ns(ns);
4272}
4273
4274struct vfsmount *kern_mount(struct file_system_type *type)
4275{
4276        struct vfsmount *mnt;
4277        mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
4278        if (!IS_ERR(mnt)) {
4279                /*
4280                 * it is a longterm mount, don't release mnt until
4281                 * we unmount before file sys is unregistered
4282                */
4283                real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
4284        }
4285        return mnt;
4286}
4287EXPORT_SYMBOL_GPL(kern_mount);
4288
4289void kern_unmount(struct vfsmount *mnt)
4290{
4291        /* release long term mount so mount point can be released */
4292        if (!IS_ERR_OR_NULL(mnt)) {
4293                real_mount(mnt)->mnt_ns = NULL;
4294                synchronize_rcu();      /* yecchhh... */
4295                mntput(mnt);
4296        }
4297}
4298EXPORT_SYMBOL(kern_unmount);
4299
4300void kern_unmount_array(struct vfsmount *mnt[], unsigned int num)
4301{
4302        unsigned int i;
4303
4304        for (i = 0; i < num; i++)
4305                if (mnt[i])
4306                        real_mount(mnt[i])->mnt_ns = NULL;
4307        synchronize_rcu_expedited();
4308        for (i = 0; i < num; i++)
4309                mntput(mnt[i]);
4310}
4311EXPORT_SYMBOL(kern_unmount_array);
4312
4313bool our_mnt(struct vfsmount *mnt)
4314{
4315        return check_mnt(real_mount(mnt));
4316}
4317
4318bool current_chrooted(void)
4319{
4320        /* Does the current process have a non-standard root */
4321        struct path ns_root;
4322        struct path fs_root;
4323        bool chrooted;
4324
4325        /* Find the namespace root */
4326        ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
4327        ns_root.dentry = ns_root.mnt->mnt_root;
4328        path_get(&ns_root);
4329        while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
4330                ;
4331
4332        get_fs_root(current->fs, &fs_root);
4333
4334        chrooted = !path_equal(&fs_root, &ns_root);
4335
4336        path_put(&fs_root);
4337        path_put(&ns_root);
4338
4339        return chrooted;
4340}
4341
4342static bool mnt_already_visible(struct mnt_namespace *ns,
4343                                const struct super_block *sb,
4344                                int *new_mnt_flags)
4345{
4346        int new_flags = *new_mnt_flags;
4347        struct mount *mnt;
4348        bool visible = false;
4349
4350        down_read(&namespace_sem);
4351        lock_ns_list(ns);
4352        list_for_each_entry(mnt, &ns->list, mnt_list) {
4353                struct mount *child;
4354                int mnt_flags;
4355
4356                if (mnt_is_cursor(mnt))
4357                        continue;
4358
4359                if (mnt->mnt.mnt_sb->s_type != sb->s_type)
4360                        continue;
4361
4362                /* This mount is not fully visible if it's root directory
4363                 * is not the root directory of the filesystem.
4364                 */
4365                if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
4366                        continue;
4367
4368                /* A local view of the mount flags */
4369                mnt_flags = mnt->mnt.mnt_flags;
4370
4371                /* Don't miss readonly hidden in the superblock flags */
4372                if (sb_rdonly(mnt->mnt.mnt_sb))
4373                        mnt_flags |= MNT_LOCK_READONLY;
4374
4375                /* Verify the mount flags are equal to or more permissive
4376                 * than the proposed new mount.
4377                 */
4378                if ((mnt_flags & MNT_LOCK_READONLY) &&
4379                    !(new_flags & MNT_READONLY))
4380                        continue;
4381                if ((mnt_flags & MNT_LOCK_ATIME) &&
4382                    ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
4383                        continue;
4384
4385                /* This mount is not fully visible if there are any
4386                 * locked child mounts that cover anything except for
4387                 * empty directories.
4388                 */
4389                list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
4390                        struct inode *inode = child->mnt_mountpoint->d_inode;
4391                        /* Only worry about locked mounts */
4392                        if (!(child->mnt.mnt_flags & MNT_LOCKED))
4393                                continue;
4394                        /* Is the directory permanetly empty? */
4395                        if (!is_empty_dir_inode(inode))
4396                                goto next;
4397                }
4398                /* Preserve the locked attributes */
4399                *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
4400                                               MNT_LOCK_ATIME);
4401                visible = true;
4402                goto found;
4403        next:   ;
4404        }
4405found:
4406        unlock_ns_list(ns);
4407        up_read(&namespace_sem);
4408        return visible;
4409}
4410
4411static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags)
4412{
4413        const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
4414        struct mnt_namespace *ns = current->nsproxy->mnt_ns;
4415        unsigned long s_iflags;
4416
4417        if (ns->user_ns == &init_user_ns)
4418                return false;
4419
4420        /* Can this filesystem be too revealing? */
4421        s_iflags = sb->s_iflags;
4422        if (!(s_iflags & SB_I_USERNS_VISIBLE))
4423                return false;
4424
4425        if ((s_iflags & required_iflags) != required_iflags) {
4426                WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
4427                          required_iflags);
4428                return true;
4429        }
4430
4431        return !mnt_already_visible(ns, sb, new_mnt_flags);
4432}
4433
4434bool mnt_may_suid(struct vfsmount *mnt)
4435{
4436        /*
4437         * Foreign mounts (accessed via fchdir or through /proc
4438         * symlinks) are always treated as if they are nosuid.  This
4439         * prevents namespaces from trusting potentially unsafe
4440         * suid/sgid bits, file caps, or security labels that originate
4441         * in other namespaces.
4442         */
4443        return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
4444               current_in_userns(mnt->mnt_sb->s_user_ns);
4445}
4446
4447static struct ns_common *mntns_get(struct task_struct *task)
4448{
4449        struct ns_common *ns = NULL;
4450        struct nsproxy *nsproxy;
4451
4452        task_lock(task);
4453        nsproxy = task->nsproxy;
4454        if (nsproxy) {
4455                ns = &nsproxy->mnt_ns->ns;
4456                get_mnt_ns(to_mnt_ns(ns));
4457        }
4458        task_unlock(task);
4459
4460        return ns;
4461}
4462
4463static void mntns_put(struct ns_common *ns)
4464{
4465        put_mnt_ns(to_mnt_ns(ns));
4466}
4467
4468static int mntns_install(struct nsset *nsset, struct ns_common *ns)
4469{
4470        struct nsproxy *nsproxy = nsset->nsproxy;
4471        struct fs_struct *fs = nsset->fs;
4472        struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
4473        struct user_namespace *user_ns = nsset->cred->user_ns;
4474        struct path root;
4475        int err;
4476
4477        if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
4478            !ns_capable(user_ns, CAP_SYS_CHROOT) ||
4479            !ns_capable(user_ns, CAP_SYS_ADMIN))
4480                return -EPERM;
4481
4482        if (is_anon_ns(mnt_ns))
4483                return -EINVAL;
4484
4485        if (fs->users != 1)
4486                return -EINVAL;
4487
4488        get_mnt_ns(mnt_ns);
4489        old_mnt_ns = nsproxy->mnt_ns;
4490        nsproxy->mnt_ns = mnt_ns;
4491
4492        /* Find the root */
4493        err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
4494                                "/", LOOKUP_DOWN, &root);
4495        if (err) {
4496                /* revert to old namespace */
4497                nsproxy->mnt_ns = old_mnt_ns;
4498                put_mnt_ns(mnt_ns);
4499                return err;
4500        }
4501
4502        put_mnt_ns(old_mnt_ns);
4503
4504        /* Update the pwd and root */
4505        set_fs_pwd(fs, &root);
4506        set_fs_root(fs, &root);
4507
4508        path_put(&root);
4509        return 0;
4510}
4511
4512static struct user_namespace *mntns_owner(struct ns_common *ns)
4513{
4514        return to_mnt_ns(ns)->user_ns;
4515}
4516
4517const struct proc_ns_operations mntns_operations = {
4518        .name           = "mnt",
4519        .type           = CLONE_NEWNS,
4520        .get            = mntns_get,
4521        .put            = mntns_put,
4522        .install        = mntns_install,
4523        .owner          = mntns_owner,
4524};
4525