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
1941/**
1942 * clone_private_mount - create a private clone of a path
1943 * @path: path to clone
1944 *
1945 * This creates a new vfsmount, which will be the clone of @path.  The new mount
1946 * will not be attached anywhere in the namespace and will be private (i.e.
1947 * changes to the originating mount won't be propagated into this).
1948 *
1949 * Release with mntput().
1950 */
1951struct vfsmount *clone_private_mount(const struct path *path)
1952{
1953        struct mount *old_mnt = real_mount(path->mnt);
1954        struct mount *new_mnt;
1955
1956        if (IS_MNT_UNBINDABLE(old_mnt))
1957                return ERR_PTR(-EINVAL);
1958
1959        new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
1960        if (IS_ERR(new_mnt))
1961                return ERR_CAST(new_mnt);
1962
1963        /* Longterm mount to be removed by kern_unmount*() */
1964        new_mnt->mnt_ns = MNT_NS_INTERNAL;
1965
1966        return &new_mnt->mnt;
1967}
1968EXPORT_SYMBOL_GPL(clone_private_mount);
1969
1970int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1971                   struct vfsmount *root)
1972{
1973        struct mount *mnt;
1974        int res = f(root, arg);
1975        if (res)
1976                return res;
1977        list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1978                res = f(&mnt->mnt, arg);
1979                if (res)
1980                        return res;
1981        }
1982        return 0;
1983}
1984
1985static void lock_mnt_tree(struct mount *mnt)
1986{
1987        struct mount *p;
1988
1989        for (p = mnt; p; p = next_mnt(p, mnt)) {
1990                int flags = p->mnt.mnt_flags;
1991                /* Don't allow unprivileged users to change mount flags */
1992                flags |= MNT_LOCK_ATIME;
1993
1994                if (flags & MNT_READONLY)
1995                        flags |= MNT_LOCK_READONLY;
1996
1997                if (flags & MNT_NODEV)
1998                        flags |= MNT_LOCK_NODEV;
1999
2000                if (flags & MNT_NOSUID)
2001                        flags |= MNT_LOCK_NOSUID;
2002
2003                if (flags & MNT_NOEXEC)
2004                        flags |= MNT_LOCK_NOEXEC;
2005                /* Don't allow unprivileged users to reveal what is under a mount */
2006                if (list_empty(&p->mnt_expire))
2007                        flags |= MNT_LOCKED;
2008                p->mnt.mnt_flags = flags;
2009        }
2010}
2011
2012static void cleanup_group_ids(struct mount *mnt, struct mount *end)
2013{
2014        struct mount *p;
2015
2016        for (p = mnt; p != end; p = next_mnt(p, mnt)) {
2017                if (p->mnt_group_id && !IS_MNT_SHARED(p))
2018                        mnt_release_group_id(p);
2019        }
2020}
2021
2022static int invent_group_ids(struct mount *mnt, bool recurse)
2023{
2024        struct mount *p;
2025
2026        for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
2027                if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
2028                        int err = mnt_alloc_group_id(p);
2029                        if (err) {
2030                                cleanup_group_ids(mnt, p);
2031                                return err;
2032                        }
2033                }
2034        }
2035
2036        return 0;
2037}
2038
2039int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
2040{
2041        unsigned int max = READ_ONCE(sysctl_mount_max);
2042        unsigned int mounts = 0, old, pending, sum;
2043        struct mount *p;
2044
2045        for (p = mnt; p; p = next_mnt(p, mnt))
2046                mounts++;
2047
2048        old = ns->mounts;
2049        pending = ns->pending_mounts;
2050        sum = old + pending;
2051        if ((old > sum) ||
2052            (pending > sum) ||
2053            (max < sum) ||
2054            (mounts > (max - sum)))
2055                return -ENOSPC;
2056
2057        ns->pending_mounts = pending + mounts;
2058        return 0;
2059}
2060
2061/*
2062 *  @source_mnt : mount tree to be attached
2063 *  @nd         : place the mount tree @source_mnt is attached
2064 *  @parent_nd  : if non-null, detach the source_mnt from its parent and
2065 *                 store the parent mount and mountpoint dentry.
2066 *                 (done when source_mnt is moved)
2067 *
2068 *  NOTE: in the table below explains the semantics when a source mount
2069 *  of a given type is attached to a destination mount of a given type.
2070 * ---------------------------------------------------------------------------
2071 * |         BIND MOUNT OPERATION                                            |
2072 * |**************************************************************************
2073 * | source-->| shared        |       private  |       slave    | unbindable |
2074 * | dest     |               |                |                |            |
2075 * |   |      |               |                |                |            |
2076 * |   v      |               |                |                |            |
2077 * |**************************************************************************
2078 * |  shared  | shared (++)   |     shared (+) |     shared(+++)|  invalid   |
2079 * |          |               |                |                |            |
2080 * |non-shared| shared (+)    |      private   |      slave (*) |  invalid   |
2081 * ***************************************************************************
2082 * A bind operation clones the source mount and mounts the clone on the
2083 * destination mount.
2084 *
2085 * (++)  the cloned mount is propagated to all the mounts in the propagation
2086 *       tree of the destination mount and the cloned mount is added to
2087 *       the peer group of the source mount.
2088 * (+)   the cloned mount is created under the destination mount and is marked
2089 *       as shared. The cloned mount is added to the peer group of the source
2090 *       mount.
2091 * (+++) the mount is propagated to all the mounts in the propagation tree
2092 *       of the destination mount and the cloned mount is made slave
2093 *       of the same master as that of the source mount. The cloned mount
2094 *       is marked as 'shared and slave'.
2095 * (*)   the cloned mount is made a slave of the same master as that of the
2096 *       source mount.
2097 *
2098 * ---------------------------------------------------------------------------
2099 * |                    MOVE MOUNT OPERATION                                 |
2100 * |**************************************************************************
2101 * | source-->| shared        |       private  |       slave    | unbindable |
2102 * | dest     |               |                |                |            |
2103 * |   |      |               |                |                |            |
2104 * |   v      |               |                |                |            |
2105 * |**************************************************************************
2106 * |  shared  | shared (+)    |     shared (+) |    shared(+++) |  invalid   |
2107 * |          |               |                |                |            |
2108 * |non-shared| shared (+*)   |      private   |    slave (*)   | unbindable |
2109 * ***************************************************************************
2110 *
2111 * (+)  the mount is moved to the destination. And is then propagated to
2112 *      all the mounts in the propagation tree of the destination mount.
2113 * (+*)  the mount is moved to the destination.
2114 * (+++)  the mount is moved to the destination and is then propagated to
2115 *      all the mounts belonging to the destination mount's propagation tree.
2116 *      the mount is marked as 'shared and slave'.
2117 * (*)  the mount continues to be a slave at the new location.
2118 *
2119 * if the source mount is a tree, the operations explained above is
2120 * applied to each mount in the tree.
2121 * Must be called without spinlocks held, since this function can sleep
2122 * in allocations.
2123 */
2124static int attach_recursive_mnt(struct mount *source_mnt,
2125                        struct mount *dest_mnt,
2126                        struct mountpoint *dest_mp,
2127                        bool moving)
2128{
2129        struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2130        HLIST_HEAD(tree_list);
2131        struct mnt_namespace *ns = dest_mnt->mnt_ns;
2132        struct mountpoint *smp;
2133        struct mount *child, *p;
2134        struct hlist_node *n;
2135        int err;
2136
2137        /* Preallocate a mountpoint in case the new mounts need
2138         * to be tucked under other mounts.
2139         */
2140        smp = get_mountpoint(source_mnt->mnt.mnt_root);
2141        if (IS_ERR(smp))
2142                return PTR_ERR(smp);
2143
2144        /* Is there space to add these mounts to the mount namespace? */
2145        if (!moving) {
2146                err = count_mounts(ns, source_mnt);
2147                if (err)
2148                        goto out;
2149        }
2150
2151        if (IS_MNT_SHARED(dest_mnt)) {
2152                err = invent_group_ids(source_mnt, true);
2153                if (err)
2154                        goto out;
2155                err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2156                lock_mount_hash();
2157                if (err)
2158                        goto out_cleanup_ids;
2159                for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2160                        set_mnt_shared(p);
2161        } else {
2162                lock_mount_hash();
2163        }
2164        if (moving) {
2165                unhash_mnt(source_mnt);
2166                attach_mnt(source_mnt, dest_mnt, dest_mp);
2167                touch_mnt_namespace(source_mnt->mnt_ns);
2168        } else {
2169                if (source_mnt->mnt_ns) {
2170                        /* move from anon - the caller will destroy */
2171                        list_del_init(&source_mnt->mnt_ns->list);
2172                }
2173                mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2174                commit_tree(source_mnt);
2175        }
2176
2177        hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2178                struct mount *q;
2179                hlist_del_init(&child->mnt_hash);
2180                q = __lookup_mnt(&child->mnt_parent->mnt,
2181                                 child->mnt_mountpoint);
2182                if (q)
2183                        mnt_change_mountpoint(child, smp, q);
2184                /* Notice when we are propagating across user namespaces */
2185                if (child->mnt_parent->mnt_ns->user_ns != user_ns)
2186                        lock_mnt_tree(child);
2187                child->mnt.mnt_flags &= ~MNT_LOCKED;
2188                commit_tree(child);
2189        }
2190        put_mountpoint(smp);
2191        unlock_mount_hash();
2192
2193        return 0;
2194
2195 out_cleanup_ids:
2196        while (!hlist_empty(&tree_list)) {
2197                child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2198                child->mnt_parent->mnt_ns->pending_mounts = 0;
2199                umount_tree(child, UMOUNT_SYNC);
2200        }
2201        unlock_mount_hash();
2202        cleanup_group_ids(source_mnt, NULL);
2203 out:
2204        ns->pending_mounts = 0;
2205
2206        read_seqlock_excl(&mount_lock);
2207        put_mountpoint(smp);
2208        read_sequnlock_excl(&mount_lock);
2209
2210        return err;
2211}
2212
2213static struct mountpoint *lock_mount(struct path *path)
2214{
2215        struct vfsmount *mnt;
2216        struct dentry *dentry = path->dentry;
2217retry:
2218        inode_lock(dentry->d_inode);
2219        if (unlikely(cant_mount(dentry))) {
2220                inode_unlock(dentry->d_inode);
2221                return ERR_PTR(-ENOENT);
2222        }
2223        namespace_lock();
2224        mnt = lookup_mnt(path);
2225        if (likely(!mnt)) {
2226                struct mountpoint *mp = get_mountpoint(dentry);
2227                if (IS_ERR(mp)) {
2228                        namespace_unlock();
2229                        inode_unlock(dentry->d_inode);
2230                        return mp;
2231                }
2232                return mp;
2233        }
2234        namespace_unlock();
2235        inode_unlock(path->dentry->d_inode);
2236        path_put(path);
2237        path->mnt = mnt;
2238        dentry = path->dentry = dget(mnt->mnt_root);
2239        goto retry;
2240}
2241
2242static void unlock_mount(struct mountpoint *where)
2243{
2244        struct dentry *dentry = where->m_dentry;
2245
2246        read_seqlock_excl(&mount_lock);
2247        put_mountpoint(where);
2248        read_sequnlock_excl(&mount_lock);
2249
2250        namespace_unlock();
2251        inode_unlock(dentry->d_inode);
2252}
2253
2254static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2255{
2256        if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2257                return -EINVAL;
2258
2259        if (d_is_dir(mp->m_dentry) !=
2260              d_is_dir(mnt->mnt.mnt_root))
2261                return -ENOTDIR;
2262
2263        return attach_recursive_mnt(mnt, p, mp, false);
2264}
2265
2266/*
2267 * Sanity check the flags to change_mnt_propagation.
2268 */
2269
2270static int flags_to_propagation_type(int ms_flags)
2271{
2272        int type = ms_flags & ~(MS_REC | MS_SILENT);
2273
2274        /* Fail if any non-propagation flags are set */
2275        if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2276                return 0;
2277        /* Only one propagation flag should be set */
2278        if (!is_power_of_2(type))
2279                return 0;
2280        return type;
2281}
2282
2283/*
2284 * recursively change the type of the mountpoint.
2285 */
2286static int do_change_type(struct path *path, int ms_flags)
2287{
2288        struct mount *m;
2289        struct mount *mnt = real_mount(path->mnt);
2290        int recurse = ms_flags & MS_REC;
2291        int type;
2292        int err = 0;
2293
2294        if (path->dentry != path->mnt->mnt_root)
2295                return -EINVAL;
2296
2297        type = flags_to_propagation_type(ms_flags);
2298        if (!type)
2299                return -EINVAL;
2300
2301        namespace_lock();
2302        if (type == MS_SHARED) {
2303                err = invent_group_ids(mnt, recurse);
2304                if (err)
2305                        goto out_unlock;
2306        }
2307
2308        lock_mount_hash();
2309        for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2310                change_mnt_propagation(m, type);
2311        unlock_mount_hash();
2312
2313 out_unlock:
2314        namespace_unlock();
2315        return err;
2316}
2317
2318static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
2319{
2320        struct mount *child;
2321        list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
2322                if (!is_subdir(child->mnt_mountpoint, dentry))
2323                        continue;
2324
2325                if (child->mnt.mnt_flags & MNT_LOCKED)
2326                        return true;
2327        }
2328        return false;
2329}
2330
2331static struct mount *__do_loopback(struct path *old_path, int recurse)
2332{
2333        struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
2334
2335        if (IS_MNT_UNBINDABLE(old))
2336                return mnt;
2337
2338        if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations)
2339                return mnt;
2340
2341        if (!recurse && has_locked_children(old, old_path->dentry))
2342                return mnt;
2343
2344        if (recurse)
2345                mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
2346        else
2347                mnt = clone_mnt(old, old_path->dentry, 0);
2348
2349        if (!IS_ERR(mnt))
2350                mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2351
2352        return mnt;
2353}
2354
2355/*
2356 * do loopback mount.
2357 */
2358static int do_loopback(struct path *path, const char *old_name,
2359                                int recurse)
2360{
2361        struct path old_path;
2362        struct mount *mnt = NULL, *parent;
2363        struct mountpoint *mp;
2364        int err;
2365        if (!old_name || !*old_name)
2366                return -EINVAL;
2367        err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2368        if (err)
2369                return err;
2370
2371        err = -EINVAL;
2372        if (mnt_ns_loop(old_path.dentry))
2373                goto out;
2374
2375        mp = lock_mount(path);
2376        if (IS_ERR(mp)) {
2377                err = PTR_ERR(mp);
2378                goto out;
2379        }
2380
2381        parent = real_mount(path->mnt);
2382        if (!check_mnt(parent))
2383                goto out2;
2384
2385        mnt = __do_loopback(&old_path, recurse);
2386        if (IS_ERR(mnt)) {
2387                err = PTR_ERR(mnt);
2388                goto out2;
2389        }
2390
2391        err = graft_tree(mnt, parent, mp);
2392        if (err) {
2393                lock_mount_hash();
2394                umount_tree(mnt, UMOUNT_SYNC);
2395                unlock_mount_hash();
2396        }
2397out2:
2398        unlock_mount(mp);
2399out:
2400        path_put(&old_path);
2401        return err;
2402}
2403
2404static struct file *open_detached_copy(struct path *path, bool recursive)
2405{
2406        struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2407        struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true);
2408        struct mount *mnt, *p;
2409        struct file *file;
2410
2411        if (IS_ERR(ns))
2412                return ERR_CAST(ns);
2413
2414        namespace_lock();
2415        mnt = __do_loopback(path, recursive);
2416        if (IS_ERR(mnt)) {
2417                namespace_unlock();
2418                free_mnt_ns(ns);
2419                return ERR_CAST(mnt);
2420        }
2421
2422        lock_mount_hash();
2423        for (p = mnt; p; p = next_mnt(p, mnt)) {
2424                p->mnt_ns = ns;
2425                ns->mounts++;
2426        }
2427        ns->root = mnt;
2428        list_add_tail(&ns->list, &mnt->mnt_list);
2429        mntget(&mnt->mnt);
2430        unlock_mount_hash();
2431        namespace_unlock();
2432
2433        mntput(path->mnt);
2434        path->mnt = &mnt->mnt;
2435        file = dentry_open(path, O_PATH, current_cred());
2436        if (IS_ERR(file))
2437                dissolve_on_fput(path->mnt);
2438        else
2439                file->f_mode |= FMODE_NEED_UNMOUNT;
2440        return file;
2441}
2442
2443SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags)
2444{
2445        struct file *file;
2446        struct path path;
2447        int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
2448        bool detached = flags & OPEN_TREE_CLONE;
2449        int error;
2450        int fd;
2451
2452        BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC);
2453
2454        if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE |
2455                      AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE |
2456                      OPEN_TREE_CLOEXEC))
2457                return -EINVAL;
2458
2459        if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE)
2460                return -EINVAL;
2461
2462        if (flags & AT_NO_AUTOMOUNT)
2463                lookup_flags &= ~LOOKUP_AUTOMOUNT;
2464        if (flags & AT_SYMLINK_NOFOLLOW)
2465                lookup_flags &= ~LOOKUP_FOLLOW;
2466        if (flags & AT_EMPTY_PATH)
2467                lookup_flags |= LOOKUP_EMPTY;
2468
2469        if (detached && !may_mount())
2470                return -EPERM;
2471
2472        fd = get_unused_fd_flags(flags & O_CLOEXEC);
2473        if (fd < 0)
2474                return fd;
2475
2476        error = user_path_at(dfd, filename, lookup_flags, &path);
2477        if (unlikely(error)) {
2478                file = ERR_PTR(error);
2479        } else {
2480                if (detached)
2481                        file = open_detached_copy(&path, flags & AT_RECURSIVE);
2482                else
2483                        file = dentry_open(&path, O_PATH, current_cred());
2484                path_put(&path);
2485        }
2486        if (IS_ERR(file)) {
2487                put_unused_fd(fd);
2488                return PTR_ERR(file);
2489        }
2490        fd_install(fd, file);
2491        return fd;
2492}
2493
2494/*
2495 * Don't allow locked mount flags to be cleared.
2496 *
2497 * No locks need to be held here while testing the various MNT_LOCK
2498 * flags because those flags can never be cleared once they are set.
2499 */
2500static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags)
2501{
2502        unsigned int fl = mnt->mnt.mnt_flags;
2503
2504        if ((fl & MNT_LOCK_READONLY) &&
2505            !(mnt_flags & MNT_READONLY))
2506                return false;
2507
2508        if ((fl & MNT_LOCK_NODEV) &&
2509            !(mnt_flags & MNT_NODEV))
2510                return false;
2511
2512        if ((fl & MNT_LOCK_NOSUID) &&
2513            !(mnt_flags & MNT_NOSUID))
2514                return false;
2515
2516        if ((fl & MNT_LOCK_NOEXEC) &&
2517            !(mnt_flags & MNT_NOEXEC))
2518                return false;
2519
2520        if ((fl & MNT_LOCK_ATIME) &&
2521            ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK)))
2522                return false;
2523
2524        return true;
2525}
2526
2527static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags)
2528{
2529        bool readonly_request = (mnt_flags & MNT_READONLY);
2530
2531        if (readonly_request == __mnt_is_readonly(&mnt->mnt))
2532                return 0;
2533
2534        if (readonly_request)
2535                return mnt_make_readonly(mnt);
2536
2537        mnt->mnt.mnt_flags &= ~MNT_READONLY;
2538        return 0;
2539}
2540
2541static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags)
2542{
2543        mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2544        mnt->mnt.mnt_flags = mnt_flags;
2545        touch_mnt_namespace(mnt->mnt_ns);
2546}
2547
2548static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt)
2549{
2550        struct super_block *sb = mnt->mnt_sb;
2551
2552        if (!__mnt_is_readonly(mnt) &&
2553           (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) {
2554                char *buf = (char *)__get_free_page(GFP_KERNEL);
2555                char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM);
2556                struct tm tm;
2557
2558                time64_to_tm(sb->s_time_max, 0, &tm);
2559
2560                pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n",
2561                        sb->s_type->name,
2562                        is_mounted(mnt) ? "remounted" : "mounted",
2563                        mntpath,
2564                        tm.tm_year+1900, (unsigned long long)sb->s_time_max);
2565
2566                free_page((unsigned long)buf);
2567        }
2568}
2569
2570/*
2571 * Handle reconfiguration of the mountpoint only without alteration of the
2572 * superblock it refers to.  This is triggered by specifying MS_REMOUNT|MS_BIND
2573 * to mount(2).
2574 */
2575static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags)
2576{
2577        struct super_block *sb = path->mnt->mnt_sb;
2578        struct mount *mnt = real_mount(path->mnt);
2579        int ret;
2580
2581        if (!check_mnt(mnt))
2582                return -EINVAL;
2583
2584        if (path->dentry != mnt->mnt.mnt_root)
2585                return -EINVAL;
2586
2587        if (!can_change_locked_flags(mnt, mnt_flags))
2588                return -EPERM;
2589
2590        /*
2591         * We're only checking whether the superblock is read-only not
2592         * changing it, so only take down_read(&sb->s_umount).
2593         */
2594        down_read(&sb->s_umount);
2595        lock_mount_hash();
2596        ret = change_mount_ro_state(mnt, mnt_flags);
2597        if (ret == 0)
2598                set_mount_attributes(mnt, mnt_flags);
2599        unlock_mount_hash();
2600        up_read(&sb->s_umount);
2601
2602        mnt_warn_timestamp_expiry(path, &mnt->mnt);
2603
2604        return ret;
2605}
2606
2607/*
2608 * change filesystem flags. dir should be a physical root of filesystem.
2609 * If you've mounted a non-root directory somewhere and want to do remount
2610 * on it - tough luck.
2611 */
2612static int do_remount(struct path *path, int ms_flags, int sb_flags,
2613                      int mnt_flags, void *data)
2614{
2615        int err;
2616        struct super_block *sb = path->mnt->mnt_sb;
2617        struct mount *mnt = real_mount(path->mnt);
2618        struct fs_context *fc;
2619
2620        if (!check_mnt(mnt))
2621                return -EINVAL;
2622
2623        if (path->dentry != path->mnt->mnt_root)
2624                return -EINVAL;
2625
2626        if (!can_change_locked_flags(mnt, mnt_flags))
2627                return -EPERM;
2628
2629        fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK);
2630        if (IS_ERR(fc))
2631                return PTR_ERR(fc);
2632
2633        fc->oldapi = true;
2634        err = parse_monolithic_mount_data(fc, data);
2635        if (!err) {
2636                down_write(&sb->s_umount);
2637                err = -EPERM;
2638                if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
2639                        err = reconfigure_super(fc);
2640                        if (!err) {
2641                                lock_mount_hash();
2642                                set_mount_attributes(mnt, mnt_flags);
2643                                unlock_mount_hash();
2644                        }
2645                }
2646                up_write(&sb->s_umount);
2647        }
2648
2649        mnt_warn_timestamp_expiry(path, &mnt->mnt);
2650
2651        put_fs_context(fc);
2652        return err;
2653}
2654
2655static inline int tree_contains_unbindable(struct mount *mnt)
2656{
2657        struct mount *p;
2658        for (p = mnt; p; p = next_mnt(p, mnt)) {
2659                if (IS_MNT_UNBINDABLE(p))
2660                        return 1;
2661        }
2662        return 0;
2663}
2664
2665/*
2666 * Check that there aren't references to earlier/same mount namespaces in the
2667 * specified subtree.  Such references can act as pins for mount namespaces
2668 * that aren't checked by the mount-cycle checking code, thereby allowing
2669 * cycles to be made.
2670 */
2671static bool check_for_nsfs_mounts(struct mount *subtree)
2672{
2673        struct mount *p;
2674        bool ret = false;
2675
2676        lock_mount_hash();
2677        for (p = subtree; p; p = next_mnt(p, subtree))
2678                if (mnt_ns_loop(p->mnt.mnt_root))
2679                        goto out;
2680
2681        ret = true;
2682out:
2683        unlock_mount_hash();
2684        return ret;
2685}
2686
2687static int do_move_mount(struct path *old_path, struct path *new_path)
2688{
2689        struct mnt_namespace *ns;
2690        struct mount *p;
2691        struct mount *old;
2692        struct mount *parent;
2693        struct mountpoint *mp, *old_mp;
2694        int err;
2695        bool attached;
2696
2697        mp = lock_mount(new_path);
2698        if (IS_ERR(mp))
2699                return PTR_ERR(mp);
2700
2701        old = real_mount(old_path->mnt);
2702        p = real_mount(new_path->mnt);
2703        parent = old->mnt_parent;
2704        attached = mnt_has_parent(old);
2705        old_mp = old->mnt_mp;
2706        ns = old->mnt_ns;
2707
2708        err = -EINVAL;
2709        /* The mountpoint must be in our namespace. */
2710        if (!check_mnt(p))
2711                goto out;
2712
2713        /* The thing moved must be mounted... */
2714        if (!is_mounted(&old->mnt))
2715                goto out;
2716
2717        /* ... and either ours or the root of anon namespace */
2718        if (!(attached ? check_mnt(old) : is_anon_ns(ns)))
2719                goto out;
2720
2721        if (old->mnt.mnt_flags & MNT_LOCKED)
2722                goto out;
2723
2724        if (old_path->dentry != old_path->mnt->mnt_root)
2725                goto out;
2726
2727        if (d_is_dir(new_path->dentry) !=
2728            d_is_dir(old_path->dentry))
2729                goto out;
2730        /*
2731         * Don't move a mount residing in a shared parent.
2732         */
2733        if (attached && IS_MNT_SHARED(parent))
2734                goto out;
2735        /*
2736         * Don't move a mount tree containing unbindable mounts to a destination
2737         * mount which is shared.
2738         */
2739        if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2740                goto out;
2741        err = -ELOOP;
2742        if (!check_for_nsfs_mounts(old))
2743                goto out;
2744        for (; mnt_has_parent(p); p = p->mnt_parent)
2745                if (p == old)
2746                        goto out;
2747
2748        err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp,
2749                                   attached);
2750        if (err)
2751                goto out;
2752
2753        /* if the mount is moved, it should no longer be expire
2754         * automatically */
2755        list_del_init(&old->mnt_expire);
2756        if (attached)
2757                put_mountpoint(old_mp);
2758out:
2759        unlock_mount(mp);
2760        if (!err) {
2761                if (attached)
2762                        mntput_no_expire(parent);
2763                else
2764                        free_mnt_ns(ns);
2765        }
2766        return err;
2767}
2768
2769static int do_move_mount_old(struct path *path, const char *old_name)
2770{
2771        struct path old_path;
2772        int err;
2773
2774        if (!old_name || !*old_name)
2775                return -EINVAL;
2776
2777        err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2778        if (err)
2779                return err;
2780
2781        err = do_move_mount(&old_path, path);
2782        path_put(&old_path);
2783        return err;
2784}
2785
2786/*
2787 * add a mount into a namespace's mount tree
2788 */
2789static int do_add_mount(struct mount *newmnt, struct mountpoint *mp,
2790                        struct path *path, int mnt_flags)
2791{
2792        struct mount *parent = real_mount(path->mnt);
2793
2794        mnt_flags &= ~MNT_INTERNAL_FLAGS;
2795
2796        if (unlikely(!check_mnt(parent))) {
2797                /* that's acceptable only for automounts done in private ns */
2798                if (!(mnt_flags & MNT_SHRINKABLE))
2799                        return -EINVAL;
2800                /* ... and for those we'd better have mountpoint still alive */
2801                if (!parent->mnt_ns)
2802                        return -EINVAL;
2803        }
2804
2805        /* Refuse the same filesystem on the same mount point */
2806        if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2807            path->mnt->mnt_root == path->dentry)
2808                return -EBUSY;
2809
2810        if (d_is_symlink(newmnt->mnt.mnt_root))
2811                return -EINVAL;
2812
2813        newmnt->mnt.mnt_flags = mnt_flags;
2814        return graft_tree(newmnt, parent, mp);
2815}
2816
2817static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags);
2818
2819/*
2820 * Create a new mount using a superblock configuration and request it
2821 * be added to the namespace tree.
2822 */
2823static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
2824                           unsigned int mnt_flags)
2825{
2826        struct vfsmount *mnt;
2827        struct mountpoint *mp;
2828        struct super_block *sb = fc->root->d_sb;
2829        int error;
2830
2831        error = security_sb_kern_mount(sb);
2832        if (!error && mount_too_revealing(sb, &mnt_flags))
2833                error = -EPERM;
2834
2835        if (unlikely(error)) {
2836                fc_drop_locked(fc);
2837                return error;
2838        }
2839
2840        up_write(&sb->s_umount);
2841
2842        mnt = vfs_create_mount(fc);
2843        if (IS_ERR(mnt))
2844                return PTR_ERR(mnt);
2845
2846        mnt_warn_timestamp_expiry(mountpoint, mnt);
2847
2848        mp = lock_mount(mountpoint);
2849        if (IS_ERR(mp)) {
2850                mntput(mnt);
2851                return PTR_ERR(mp);
2852        }
2853        error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags);
2854        unlock_mount(mp);
2855        if (error < 0)
2856                mntput(mnt);
2857        return error;
2858}
2859
2860/*
2861 * create a new mount for userspace and request it to be added into the
2862 * namespace's tree
2863 */
2864static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
2865                        int mnt_flags, const char *name, void *data)
2866{
2867        struct file_system_type *type;
2868        struct fs_context *fc;
2869        const char *subtype = NULL;
2870        int err = 0;
2871
2872        if (!fstype)
2873                return -EINVAL;
2874
2875        type = get_fs_type(fstype);
2876        if (!type)
2877                return -ENODEV;
2878
2879        if (type->fs_flags & FS_HAS_SUBTYPE) {
2880                subtype = strchr(fstype, '.');
2881                if (subtype) {
2882                        subtype++;
2883                        if (!*subtype) {
2884                                put_filesystem(type);
2885                                return -EINVAL;
2886                        }
2887                }
2888        }
2889
2890        fc = fs_context_for_mount(type, sb_flags);
2891        put_filesystem(type);
2892        if (IS_ERR(fc))
2893                return PTR_ERR(fc);
2894
2895        if (subtype)
2896                err = vfs_parse_fs_string(fc, "subtype",
2897                                          subtype, strlen(subtype));
2898        if (!err && name)
2899                err = vfs_parse_fs_string(fc, "source", name, strlen(name));
2900        if (!err)
2901                err = parse_monolithic_mount_data(fc, data);
2902        if (!err && !mount_capable(fc))
2903                err = -EPERM;
2904        if (!err)
2905                err = vfs_get_tree(fc);
2906        if (!err)
2907                err = do_new_mount_fc(fc, path, mnt_flags);
2908
2909        put_fs_context(fc);
2910        return err;
2911}
2912
2913int finish_automount(struct vfsmount *m, struct path *path)
2914{
2915        struct dentry *dentry = path->dentry;
2916        struct mountpoint *mp;
2917        struct mount *mnt;
2918        int err;
2919
2920        if (!m)
2921                return 0;
2922        if (IS_ERR(m))
2923                return PTR_ERR(m);
2924
2925        mnt = real_mount(m);
2926        /* The new mount record should have at least 2 refs to prevent it being
2927         * expired before we get a chance to add it
2928         */
2929        BUG_ON(mnt_get_count(mnt) < 2);
2930
2931        if (m->mnt_sb == path->mnt->mnt_sb &&
2932            m->mnt_root == dentry) {
2933                err = -ELOOP;
2934                goto discard;
2935        }
2936
2937        /*
2938         * we don't want to use lock_mount() - in this case finding something
2939         * that overmounts our mountpoint to be means "quitely drop what we've
2940         * got", not "try to mount it on top".
2941         */
2942        inode_lock(dentry->d_inode);
2943        namespace_lock();
2944        if (unlikely(cant_mount(dentry))) {
2945                err = -ENOENT;
2946                goto discard_locked;
2947        }
2948        rcu_read_lock();
2949        if (unlikely(__lookup_mnt(path->mnt, dentry))) {
2950                rcu_read_unlock();
2951                err = 0;
2952                goto discard_locked;
2953        }
2954        rcu_read_unlock();
2955        mp = get_mountpoint(dentry);
2956        if (IS_ERR(mp)) {
2957                err = PTR_ERR(mp);
2958                goto discard_locked;
2959        }
2960
2961        err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2962        unlock_mount(mp);
2963        if (unlikely(err))
2964                goto discard;
2965        mntput(m);
2966        return 0;
2967
2968discard_locked:
2969        namespace_unlock();
2970        inode_unlock(dentry->d_inode);
2971discard:
2972        /* remove m from any expiration list it may be on */
2973        if (!list_empty(&mnt->mnt_expire)) {
2974                namespace_lock();
2975                list_del_init(&mnt->mnt_expire);
2976                namespace_unlock();
2977        }
2978        mntput(m);
2979        mntput(m);
2980        return err;
2981}
2982
2983/**
2984 * mnt_set_expiry - Put a mount on an expiration list
2985 * @mnt: The mount to list.
2986 * @expiry_list: The list to add the mount to.
2987 */
2988void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2989{
2990        namespace_lock();
2991
2992        list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2993
2994        namespace_unlock();
2995}
2996EXPORT_SYMBOL(mnt_set_expiry);
2997
2998/*
2999 * process a list of expirable mountpoints with the intent of discarding any
3000 * mountpoints that aren't in use and haven't been touched since last we came
3001 * here
3002 */
3003void mark_mounts_for_expiry(struct list_head *mounts)
3004{
3005        struct mount *mnt, *next;
3006        LIST_HEAD(graveyard);
3007
3008        if (list_empty(mounts))
3009                return;
3010
3011        namespace_lock();
3012        lock_mount_hash();
3013
3014        /* extract from the expiration list every vfsmount that matches the
3015         * following criteria:
3016         * - only referenced by its parent vfsmount
3017         * - still marked for expiry (marked on the last call here; marks are
3018         *   cleared by mntput())
3019         */
3020        list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
3021                if (!xchg(&mnt->mnt_expiry_mark, 1) ||
3022                        propagate_mount_busy(mnt, 1))
3023                        continue;
3024                list_move(&mnt->mnt_expire, &graveyard);
3025        }
3026        while (!list_empty(&graveyard)) {
3027                mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
3028                touch_mnt_namespace(mnt->mnt_ns);
3029                umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3030        }
3031        unlock_mount_hash();
3032        namespace_unlock();
3033}
3034
3035EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
3036
3037/*
3038 * Ripoff of 'select_parent()'
3039 *
3040 * search the list of submounts for a given mountpoint, and move any
3041 * shrinkable submounts to the 'graveyard' list.
3042 */
3043static int select_submounts(struct mount *parent, struct list_head *graveyard)
3044{
3045        struct mount *this_parent = parent;
3046        struct list_head *next;
3047        int found = 0;
3048
3049repeat:
3050        next = this_parent->mnt_mounts.next;
3051resume:
3052        while (next != &this_parent->mnt_mounts) {
3053                struct list_head *tmp = next;
3054                struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
3055
3056                next = tmp->next;
3057                if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
3058                        continue;
3059                /*
3060                 * Descend a level if the d_mounts list is non-empty.
3061                 */
3062                if (!list_empty(&mnt->mnt_mounts)) {
3063                        this_parent = mnt;
3064                        goto repeat;
3065                }
3066
3067                if (!propagate_mount_busy(mnt, 1)) {
3068                        list_move_tail(&mnt->mnt_expire, graveyard);
3069                        found++;
3070                }
3071        }
3072        /*
3073         * All done at this level ... ascend and resume the search
3074         */
3075        if (this_parent != parent) {
3076                next = this_parent->mnt_child.next;
3077                this_parent = this_parent->mnt_parent;
3078                goto resume;
3079        }
3080        return found;
3081}
3082
3083/*
3084 * process a list of expirable mountpoints with the intent of discarding any
3085 * submounts of a specific parent mountpoint
3086 *
3087 * mount_lock must be held for write
3088 */
3089static void shrink_submounts(struct mount *mnt)
3090{
3091        LIST_HEAD(graveyard);
3092        struct mount *m;
3093
3094        /* extract submounts of 'mountpoint' from the expiration list */
3095        while (select_submounts(mnt, &graveyard)) {
3096                while (!list_empty(&graveyard)) {
3097                        m = list_first_entry(&graveyard, struct mount,
3098                                                mnt_expire);
3099                        touch_mnt_namespace(m->mnt_ns);
3100                        umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3101                }
3102        }
3103}
3104
3105static void *copy_mount_options(const void __user * data)
3106{
3107        char *copy;
3108        unsigned left, offset;
3109
3110        if (!data)
3111                return NULL;
3112
3113        copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
3114        if (!copy)
3115                return ERR_PTR(-ENOMEM);
3116
3117        left = copy_from_user(copy, data, PAGE_SIZE);
3118
3119        /*
3120         * Not all architectures have an exact copy_from_user(). Resort to
3121         * byte at a time.
3122         */
3123        offset = PAGE_SIZE - left;
3124        while (left) {
3125                char c;
3126                if (get_user(c, (const char __user *)data + offset))
3127                        break;
3128                copy[offset] = c;
3129                left--;
3130                offset++;
3131        }
3132
3133        if (left == PAGE_SIZE) {
3134                kfree(copy);
3135                return ERR_PTR(-EFAULT);
3136        }
3137
3138        return copy;
3139}
3140
3141static char *copy_mount_string(const void __user *data)
3142{
3143        return data ? strndup_user(data, PATH_MAX) : NULL;
3144}
3145
3146/*
3147 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3148 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3149 *
3150 * data is a (void *) that can point to any structure up to
3151 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3152 * information (or be NULL).
3153 *
3154 * Pre-0.97 versions of mount() didn't have a flags word.
3155 * When the flags word was introduced its top half was required
3156 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3157 * Therefore, if this magic number is present, it carries no information
3158 * and must be discarded.
3159 */
3160int path_mount(const char *dev_name, struct path *path,
3161                const char *type_page, unsigned long flags, void *data_page)
3162{
3163        unsigned int mnt_flags = 0, sb_flags;
3164        int ret;
3165
3166        /* Discard magic */
3167        if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
3168                flags &= ~MS_MGC_MSK;
3169
3170        /* Basic sanity checks */
3171        if (data_page)
3172                ((char *)data_page)[PAGE_SIZE - 1] = 0;
3173
3174        if (flags & MS_NOUSER)
3175                return -EINVAL;
3176
3177        ret = security_sb_mount(dev_name, path, type_page, flags, data_page);
3178        if (ret)
3179                return ret;
3180        if (!may_mount())
3181                return -EPERM;
3182        if ((flags & SB_MANDLOCK) && !may_mandlock())
3183                return -EPERM;
3184
3185        /* Default to relatime unless overriden */
3186        if (!(flags & MS_NOATIME))
3187                mnt_flags |= MNT_RELATIME;
3188
3189        /* Separate the per-mountpoint flags */
3190        if (flags & MS_NOSUID)
3191                mnt_flags |= MNT_NOSUID;
3192        if (flags & MS_NODEV)
3193                mnt_flags |= MNT_NODEV;
3194        if (flags & MS_NOEXEC)
3195                mnt_flags |= MNT_NOEXEC;
3196        if (flags & MS_NOATIME)
3197                mnt_flags |= MNT_NOATIME;
3198        if (flags & MS_NODIRATIME)
3199                mnt_flags |= MNT_NODIRATIME;
3200        if (flags & MS_STRICTATIME)
3201                mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
3202        if (flags & MS_RDONLY)
3203                mnt_flags |= MNT_READONLY;
3204        if (flags & MS_NOSYMFOLLOW)
3205                mnt_flags |= MNT_NOSYMFOLLOW;
3206
3207        /* The default atime for remount is preservation */
3208        if ((flags & MS_REMOUNT) &&
3209            ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
3210                       MS_STRICTATIME)) == 0)) {
3211                mnt_flags &= ~MNT_ATIME_MASK;
3212                mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK;
3213        }
3214
3215        sb_flags = flags & (SB_RDONLY |
3216                            SB_SYNCHRONOUS |
3217                            SB_MANDLOCK |
3218                            SB_DIRSYNC |
3219                            SB_SILENT |
3220                            SB_POSIXACL |
3221                            SB_LAZYTIME |
3222                            SB_I_VERSION);
3223
3224        if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND))
3225                return do_reconfigure_mnt(path, mnt_flags);
3226        if (flags & MS_REMOUNT)
3227                return do_remount(path, flags, sb_flags, mnt_flags, data_page);
3228        if (flags & MS_BIND)
3229                return do_loopback(path, dev_name, flags & MS_REC);
3230        if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
3231                return do_change_type(path, flags);
3232        if (flags & MS_MOVE)
3233                return do_move_mount_old(path, dev_name);
3234
3235        return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name,
3236                            data_page);
3237}
3238
3239long do_mount(const char *dev_name, const char __user *dir_name,
3240                const char *type_page, unsigned long flags, void *data_page)
3241{
3242        struct path path;
3243        int ret;
3244
3245        ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path);
3246        if (ret)
3247                return ret;
3248        ret = path_mount(dev_name, &path, type_page, flags, data_page);
3249        path_put(&path);
3250        return ret;
3251}
3252
3253static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
3254{
3255        return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
3256}
3257
3258static void dec_mnt_namespaces(struct ucounts *ucounts)
3259{
3260        dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
3261}
3262
3263static void free_mnt_ns(struct mnt_namespace *ns)
3264{
3265        if (!is_anon_ns(ns))
3266                ns_free_inum(&ns->ns);
3267        dec_mnt_namespaces(ns->ucounts);
3268        put_user_ns(ns->user_ns);
3269        kfree(ns);
3270}
3271
3272/*
3273 * Assign a sequence number so we can detect when we attempt to bind
3274 * mount a reference to an older mount namespace into the current
3275 * mount namespace, preventing reference counting loops.  A 64bit
3276 * number incrementing at 10Ghz will take 12,427 years to wrap which
3277 * is effectively never, so we can ignore the possibility.
3278 */
3279static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
3280
3281static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon)
3282{
3283        struct mnt_namespace *new_ns;
3284        struct ucounts *ucounts;
3285        int ret;
3286
3287        ucounts = inc_mnt_namespaces(user_ns);
3288        if (!ucounts)
3289                return ERR_PTR(-ENOSPC);
3290
3291        new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
3292        if (!new_ns) {
3293                dec_mnt_namespaces(ucounts);
3294                return ERR_PTR(-ENOMEM);
3295        }
3296        if (!anon) {
3297                ret = ns_alloc_inum(&new_ns->ns);
3298                if (ret) {
3299                        kfree(new_ns);
3300                        dec_mnt_namespaces(ucounts);
3301                        return ERR_PTR(ret);
3302                }
3303        }
3304        new_ns->ns.ops = &mntns_operations;
3305        if (!anon)
3306                new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
3307        refcount_set(&new_ns->ns.count, 1);
3308        INIT_LIST_HEAD(&new_ns->list);
3309        init_waitqueue_head(&new_ns->poll);
3310        spin_lock_init(&new_ns->ns_lock);
3311        new_ns->user_ns = get_user_ns(user_ns);
3312        new_ns->ucounts = ucounts;
3313        return new_ns;
3314}
3315
3316__latent_entropy
3317struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
3318                struct user_namespace *user_ns, struct fs_struct *new_fs)
3319{
3320        struct mnt_namespace *new_ns;
3321        struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
3322        struct mount *p, *q;
3323        struct mount *old;
3324        struct mount *new;
3325        int copy_flags;
3326
3327        BUG_ON(!ns);
3328
3329        if (likely(!(flags & CLONE_NEWNS))) {
3330                get_mnt_ns(ns);
3331                return ns;
3332        }
3333
3334        old = ns->root;
3335
3336        new_ns = alloc_mnt_ns(user_ns, false);
3337        if (IS_ERR(new_ns))
3338                return new_ns;
3339
3340        namespace_lock();
3341        /* First pass: copy the tree topology */
3342        copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
3343        if (user_ns != ns->user_ns)
3344                copy_flags |= CL_SHARED_TO_SLAVE;
3345        new = copy_tree(old, old->mnt.mnt_root, copy_flags);
3346        if (IS_ERR(new)) {
3347                namespace_unlock();
3348                free_mnt_ns(new_ns);
3349                return ERR_CAST(new);
3350        }
3351        if (user_ns != ns->user_ns) {
3352                lock_mount_hash();
3353                lock_mnt_tree(new);
3354                unlock_mount_hash();
3355        }
3356        new_ns->root = new;
3357        list_add_tail(&new_ns->list, &new->mnt_list);
3358
3359        /*
3360         * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3361         * as belonging to new namespace.  We have already acquired a private
3362         * fs_struct, so tsk->fs->lock is not needed.
3363         */
3364        p = old;
3365        q = new;
3366        while (p) {
3367                q->mnt_ns = new_ns;
3368                new_ns->mounts++;
3369                if (new_fs) {
3370                        if (&p->mnt == new_fs->root.mnt) {
3371                                new_fs->root.mnt = mntget(&q->mnt);
3372                                rootmnt = &p->mnt;
3373                        }
3374                        if (&p->mnt == new_fs->pwd.mnt) {
3375                                new_fs->pwd.mnt = mntget(&q->mnt);
3376                                pwdmnt = &p->mnt;
3377                        }
3378                }
3379                p = next_mnt(p, old);
3380                q = next_mnt(q, new);
3381                if (!q)
3382                        break;
3383                while (p->mnt.mnt_root != q->mnt.mnt_root)
3384                        p = next_mnt(p, old);
3385        }
3386        namespace_unlock();
3387
3388        if (rootmnt)
3389                mntput(rootmnt);
3390        if (pwdmnt)
3391                mntput(pwdmnt);
3392
3393        return new_ns;
3394}
3395
3396struct dentry *mount_subtree(struct vfsmount *m, const char *name)
3397{
3398        struct mount *mnt = real_mount(m);
3399        struct mnt_namespace *ns;
3400        struct super_block *s;
3401        struct path path;
3402        int err;
3403
3404        ns = alloc_mnt_ns(&init_user_ns, true);
3405        if (IS_ERR(ns)) {
3406                mntput(m);
3407                return ERR_CAST(ns);
3408        }
3409        mnt->mnt_ns = ns;
3410        ns->root = mnt;
3411        ns->mounts++;
3412        list_add(&mnt->mnt_list, &ns->list);
3413
3414        err = vfs_path_lookup(m->mnt_root, m,
3415                        name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3416
3417        put_mnt_ns(ns);
3418
3419        if (err)
3420                return ERR_PTR(err);
3421
3422        /* trade a vfsmount reference for active sb one */
3423        s = path.mnt->mnt_sb;
3424        atomic_inc(&s->s_active);
3425        mntput(path.mnt);
3426        /* lock the sucker */
3427        down_write(&s->s_umount);
3428        /* ... and return the root of (sub)tree on it */
3429        return path.dentry;
3430}
3431EXPORT_SYMBOL(mount_subtree);
3432
3433SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3434                char __user *, type, unsigned long, flags, void __user *, data)
3435{
3436        int ret;
3437        char *kernel_type;
3438        char *kernel_dev;
3439        void *options;
3440
3441        kernel_type = copy_mount_string(type);
3442        ret = PTR_ERR(kernel_type);
3443        if (IS_ERR(kernel_type))
3444                goto out_type;
3445
3446        kernel_dev = copy_mount_string(dev_name);
3447        ret = PTR_ERR(kernel_dev);
3448        if (IS_ERR(kernel_dev))
3449                goto out_dev;
3450
3451        options = copy_mount_options(data);
3452        ret = PTR_ERR(options);
3453        if (IS_ERR(options))
3454                goto out_data;
3455
3456        ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3457
3458        kfree(options);
3459out_data:
3460        kfree(kernel_dev);
3461out_dev:
3462        kfree(kernel_type);
3463out_type:
3464        return ret;
3465}
3466
3467#define FSMOUNT_VALID_FLAGS \
3468        (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV | \
3469         MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME)
3470
3471#define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP)
3472
3473#define MOUNT_SETATTR_PROPAGATION_FLAGS \
3474        (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED)
3475
3476static unsigned int attr_flags_to_mnt_flags(u64 attr_flags)
3477{
3478        unsigned int mnt_flags = 0;
3479
3480        if (attr_flags & MOUNT_ATTR_RDONLY)
3481                mnt_flags |= MNT_READONLY;
3482        if (attr_flags & MOUNT_ATTR_NOSUID)
3483                mnt_flags |= MNT_NOSUID;
3484        if (attr_flags & MOUNT_ATTR_NODEV)
3485                mnt_flags |= MNT_NODEV;
3486        if (attr_flags & MOUNT_ATTR_NOEXEC)
3487                mnt_flags |= MNT_NOEXEC;
3488        if (attr_flags & MOUNT_ATTR_NODIRATIME)
3489                mnt_flags |= MNT_NODIRATIME;
3490
3491        return mnt_flags;
3492}
3493
3494/*
3495 * Create a kernel mount representation for a new, prepared superblock
3496 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3497 */
3498SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
3499                unsigned int, attr_flags)
3500{
3501        struct mnt_namespace *ns;
3502        struct fs_context *fc;
3503        struct file *file;
3504        struct path newmount;
3505        struct mount *mnt;
3506        struct fd f;
3507        unsigned int mnt_flags = 0;
3508        long ret;
3509
3510        if (!may_mount())
3511                return -EPERM;
3512
3513        if ((flags & ~(FSMOUNT_CLOEXEC)) != 0)
3514                return -EINVAL;
3515
3516        if (attr_flags & ~FSMOUNT_VALID_FLAGS)
3517                return -EINVAL;
3518
3519        mnt_flags = attr_flags_to_mnt_flags(attr_flags);
3520
3521        switch (attr_flags & MOUNT_ATTR__ATIME) {
3522        case MOUNT_ATTR_STRICTATIME:
3523                break;
3524        case MOUNT_ATTR_NOATIME:
3525                mnt_flags |= MNT_NOATIME;
3526                break;
3527        case MOUNT_ATTR_RELATIME:
3528                mnt_flags |= MNT_RELATIME;
3529                break;
3530        default:
3531                return -EINVAL;
3532        }
3533
3534        f = fdget(fs_fd);
3535        if (!f.file)
3536                return -EBADF;
3537
3538        ret = -EINVAL;
3539        if (f.file->f_op != &fscontext_fops)
3540                goto err_fsfd;
3541
3542        fc = f.file->private_data;
3543
3544        ret = mutex_lock_interruptible(&fc->uapi_mutex);
3545        if (ret < 0)
3546                goto err_fsfd;
3547
3548        /* There must be a valid superblock or we can't mount it */
3549        ret = -EINVAL;
3550        if (!fc->root)
3551                goto err_unlock;
3552
3553        ret = -EPERM;
3554        if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) {
3555                pr_warn("VFS: Mount too revealing\n");
3556                goto err_unlock;
3557        }
3558
3559        ret = -EBUSY;
3560        if (fc->phase != FS_CONTEXT_AWAITING_MOUNT)
3561                goto err_unlock;
3562
3563        ret = -EPERM;
3564        if ((fc->sb_flags & SB_MANDLOCK) && !may_mandlock())
3565                goto err_unlock;
3566
3567        newmount.mnt = vfs_create_mount(fc);
3568        if (IS_ERR(newmount.mnt)) {
3569                ret = PTR_ERR(newmount.mnt);
3570                goto err_unlock;
3571        }
3572        newmount.dentry = dget(fc->root);
3573        newmount.mnt->mnt_flags = mnt_flags;
3574
3575        /* We've done the mount bit - now move the file context into more or
3576         * less the same state as if we'd done an fspick().  We don't want to
3577         * do any memory allocation or anything like that at this point as we
3578         * don't want to have to handle any errors incurred.
3579         */
3580        vfs_clean_context(fc);
3581
3582        ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true);
3583        if (IS_ERR(ns)) {
3584                ret = PTR_ERR(ns);
3585                goto err_path;
3586        }
3587        mnt = real_mount(newmount.mnt);
3588        mnt->mnt_ns = ns;
3589        ns->root = mnt;
3590        ns->mounts = 1;
3591        list_add(&mnt->mnt_list, &ns->list);
3592        mntget(newmount.mnt);
3593
3594        /* Attach to an apparent O_PATH fd with a note that we need to unmount
3595         * it, not just simply put it.
3596         */
3597        file = dentry_open(&newmount, O_PATH, fc->cred);
3598        if (IS_ERR(file)) {
3599                dissolve_on_fput(newmount.mnt);
3600                ret = PTR_ERR(file);
3601                goto err_path;
3602        }
3603        file->f_mode |= FMODE_NEED_UNMOUNT;
3604
3605        ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
3606        if (ret >= 0)
3607                fd_install(ret, file);
3608        else
3609                fput(file);
3610
3611err_path:
3612        path_put(&newmount);
3613err_unlock:
3614        mutex_unlock(&fc->uapi_mutex);
3615err_fsfd:
3616        fdput(f);
3617        return ret;
3618}
3619
3620/*
3621 * Move a mount from one place to another.  In combination with
3622 * fsopen()/fsmount() this is used to install a new mount and in combination
3623 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
3624 * a mount subtree.
3625 *
3626 * Note the flags value is a combination of MOVE_MOUNT_* flags.
3627 */
3628SYSCALL_DEFINE5(move_mount,
3629                int, from_dfd, const char __user *, from_pathname,
3630                int, to_dfd, const char __user *, to_pathname,
3631                unsigned int, flags)
3632{
3633        struct path from_path, to_path;
3634        unsigned int lflags;
3635        int ret = 0;
3636
3637        if (!may_mount())
3638                return -EPERM;
3639
3640        if (flags & ~MOVE_MOUNT__MASK)
3641                return -EINVAL;
3642
3643        /* If someone gives a pathname, they aren't permitted to move
3644         * from an fd that requires unmount as we can't get at the flag
3645         * to clear it afterwards.
3646         */
3647        lflags = 0;
3648        if (flags & MOVE_MOUNT_F_SYMLINKS)      lflags |= LOOKUP_FOLLOW;
3649        if (flags & MOVE_MOUNT_F_AUTOMOUNTS)    lflags |= LOOKUP_AUTOMOUNT;
3650        if (flags & MOVE_MOUNT_F_EMPTY_PATH)    lflags |= LOOKUP_EMPTY;
3651
3652        ret = user_path_at(from_dfd, from_pathname, lflags, &from_path);
3653        if (ret < 0)
3654                return ret;
3655
3656        lflags = 0;
3657        if (flags & MOVE_MOUNT_T_SYMLINKS)      lflags |= LOOKUP_FOLLOW;
3658        if (flags & MOVE_MOUNT_T_AUTOMOUNTS)    lflags |= LOOKUP_AUTOMOUNT;
3659        if (flags & MOVE_MOUNT_T_EMPTY_PATH)    lflags |= LOOKUP_EMPTY;
3660
3661        ret = user_path_at(to_dfd, to_pathname, lflags, &to_path);
3662        if (ret < 0)
3663                goto out_from;
3664
3665        ret = security_move_mount(&from_path, &to_path);
3666        if (ret < 0)
3667                goto out_to;
3668
3669        ret = do_move_mount(&from_path, &to_path);
3670
3671out_to:
3672        path_put(&to_path);
3673out_from:
3674        path_put(&from_path);
3675        return ret;
3676}
3677
3678/*
3679 * Return true if path is reachable from root
3680 *
3681 * namespace_sem or mount_lock is held
3682 */
3683bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3684                         const struct path *root)
3685{
3686        while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3687                dentry = mnt->mnt_mountpoint;
3688                mnt = mnt->mnt_parent;
3689        }
3690        return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3691}
3692
3693bool path_is_under(const struct path *path1, const struct path *path2)
3694{
3695        bool res;
3696        read_seqlock_excl(&mount_lock);
3697        res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3698        read_sequnlock_excl(&mount_lock);
3699        return res;
3700}
3701EXPORT_SYMBOL(path_is_under);
3702
3703/*
3704 * pivot_root Semantics:
3705 * Moves the root file system of the current process to the directory put_old,
3706 * makes new_root as the new root file system of the current process, and sets
3707 * root/cwd of all processes which had them on the current root to new_root.
3708 *
3709 * Restrictions:
3710 * The new_root and put_old must be directories, and  must not be on the
3711 * same file  system as the current process root. The put_old  must  be
3712 * underneath new_root,  i.e. adding a non-zero number of /.. to the string
3713 * pointed to by put_old must yield the same directory as new_root. No other
3714 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3715 *
3716 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3717 * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives
3718 * in this situation.
3719 *
3720 * Notes:
3721 *  - we don't move root/cwd if they are not at the root (reason: if something
3722 *    cared enough to change them, it's probably wrong to force them elsewhere)
3723 *  - it's okay to pick a root that isn't the root of a file system, e.g.
3724 *    /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3725 *    though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3726 *    first.
3727 */
3728SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3729                const char __user *, put_old)
3730{
3731        struct path new, old, root;
3732        struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
3733        struct mountpoint *old_mp, *root_mp;
3734        int error;
3735
3736        if (!may_mount())
3737                return -EPERM;
3738
3739        error = user_path_at(AT_FDCWD, new_root,
3740                             LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new);
3741        if (error)
3742                goto out0;
3743
3744        error = user_path_at(AT_FDCWD, put_old,
3745                             LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old);
3746        if (error)
3747                goto out1;
3748
3749        error = security_sb_pivotroot(&old, &new);
3750        if (error)
3751                goto out2;
3752
3753        get_fs_root(current->fs, &root);
3754        old_mp = lock_mount(&old);
3755        error = PTR_ERR(old_mp);
3756        if (IS_ERR(old_mp))
3757                goto out3;
3758
3759        error = -EINVAL;
3760        new_mnt = real_mount(new.mnt);
3761        root_mnt = real_mount(root.mnt);
3762        old_mnt = real_mount(old.mnt);
3763        ex_parent = new_mnt->mnt_parent;
3764        root_parent = root_mnt->mnt_parent;
3765        if (IS_MNT_SHARED(old_mnt) ||
3766                IS_MNT_SHARED(ex_parent) ||
3767                IS_MNT_SHARED(root_parent))
3768                goto out4;
3769        if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3770                goto out4;
3771        if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3772                goto out4;
3773        error = -ENOENT;
3774        if (d_unlinked(new.dentry))
3775                goto out4;
3776        error = -EBUSY;
3777        if (new_mnt == root_mnt || old_mnt == root_mnt)
3778                goto out4; /* loop, on the same file system  */
3779        error = -EINVAL;
3780        if (root.mnt->mnt_root != root.dentry)
3781                goto out4; /* not a mountpoint */
3782        if (!mnt_has_parent(root_mnt))
3783                goto out4; /* not attached */
3784        if (new.mnt->mnt_root != new.dentry)
3785                goto out4; /* not a mountpoint */
3786        if (!mnt_has_parent(new_mnt))
3787                goto out4; /* not attached */
3788        /* make sure we can reach put_old from new_root */
3789        if (!is_path_reachable(old_mnt, old.dentry, &new))
3790                goto out4;
3791        /* make certain new is below the root */
3792        if (!is_path_reachable(new_mnt, new.dentry, &root))
3793                goto out4;
3794        lock_mount_hash();
3795        umount_mnt(new_mnt);
3796        root_mp = unhash_mnt(root_mnt);  /* we'll need its mountpoint */
3797        if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3798                new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3799                root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3800        }
3801        /* mount old root on put_old */
3802        attach_mnt(root_mnt, old_mnt, old_mp);
3803        /* mount new_root on / */
3804        attach_mnt(new_mnt, root_parent, root_mp);
3805        mnt_add_count(root_parent, -1);
3806        touch_mnt_namespace(current->nsproxy->mnt_ns);
3807        /* A moved mount should not expire automatically */
3808        list_del_init(&new_mnt->mnt_expire);
3809        put_mountpoint(root_mp);
3810        unlock_mount_hash();
3811        chroot_fs_refs(&root, &new);
3812        error = 0;
3813out4:
3814        unlock_mount(old_mp);
3815        if (!error)
3816                mntput_no_expire(ex_parent);
3817out3:
3818        path_put(&root);
3819out2:
3820        path_put(&old);
3821out1:
3822        path_put(&new);
3823out0:
3824        return error;
3825}
3826
3827static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt)
3828{
3829        unsigned int flags = mnt->mnt.mnt_flags;
3830
3831        /*  flags to clear */
3832        flags &= ~kattr->attr_clr;
3833        /* flags to raise */
3834        flags |= kattr->attr_set;
3835
3836        return flags;
3837}
3838
3839static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
3840{
3841        struct vfsmount *m = &mnt->mnt;
3842
3843        if (!kattr->mnt_userns)
3844                return 0;
3845
3846        /*
3847         * Once a mount has been idmapped we don't allow it to change its
3848         * mapping. It makes things simpler and callers can just create
3849         * another bind-mount they can idmap if they want to.
3850         */
3851        if (mnt_user_ns(m) != &init_user_ns)
3852                return -EPERM;
3853
3854        /* The underlying filesystem doesn't support idmapped mounts yet. */
3855        if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP))
3856                return -EINVAL;
3857
3858        /* Don't yet support filesystem mountable in user namespaces. */
3859        if (m->mnt_sb->s_user_ns != &init_user_ns)
3860                return -EINVAL;
3861
3862        /* We're not controlling the superblock. */
3863        if (!capable(CAP_SYS_ADMIN))
3864                return -EPERM;
3865
3866        /* Mount has already been visible in the filesystem hierarchy. */
3867        if (!is_anon_ns(mnt->mnt_ns))
3868                return -EINVAL;
3869
3870        return 0;
3871}
3872
3873static struct mount *mount_setattr_prepare(struct mount_kattr *kattr,
3874                                           struct mount *mnt, int *err)
3875{
3876        struct mount *m = mnt, *last = NULL;
3877
3878        if (!is_mounted(&m->mnt)) {
3879                *err = -EINVAL;
3880                goto out;
3881        }
3882
3883        if (!(mnt_has_parent(m) ? check_mnt(m) : is_anon_ns(m->mnt_ns))) {
3884                *err = -EINVAL;
3885                goto out;
3886        }
3887
3888        do {
3889                unsigned int flags;
3890
3891                flags = recalc_flags(kattr, m);
3892                if (!can_change_locked_flags(m, flags)) {
3893                        *err = -EPERM;
3894                        goto out;
3895                }
3896
3897                *err = can_idmap_mount(kattr, m);
3898                if (*err)
3899                        goto out;
3900
3901                last = m;
3902
3903                if ((kattr->attr_set & MNT_READONLY) &&
3904                    !(m->mnt.mnt_flags & MNT_READONLY)) {
3905                        *err = mnt_hold_writers(m);
3906                        if (*err)
3907                                goto out;
3908                }
3909        } while (kattr->recurse && (m = next_mnt(m, mnt)));
3910
3911out:
3912        return last;
3913}
3914
3915static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
3916{
3917        struct user_namespace *mnt_userns;
3918
3919        if (!kattr->mnt_userns)
3920                return;
3921
3922        mnt_userns = get_user_ns(kattr->mnt_userns);
3923        /* Pairs with smp_load_acquire() in mnt_user_ns(). */
3924        smp_store_release(&mnt->mnt.mnt_userns, mnt_userns);
3925}
3926
3927static void mount_setattr_commit(struct mount_kattr *kattr,
3928                                 struct mount *mnt, struct mount *last,
3929                                 int err)
3930{
3931        struct mount *m = mnt;
3932
3933        do {
3934                if (!err) {
3935                        unsigned int flags;
3936
3937                        do_idmap_mount(kattr, m);
3938                        flags = recalc_flags(kattr, m);
3939                        WRITE_ONCE(m->mnt.mnt_flags, flags);
3940                }
3941
3942                /*
3943                 * We either set MNT_READONLY above so make it visible
3944                 * before ~MNT_WRITE_HOLD or we failed to recursively
3945                 * apply mount options.
3946                 */
3947                if ((kattr->attr_set & MNT_READONLY) &&
3948                    (m->mnt.mnt_flags & MNT_WRITE_HOLD))
3949                        mnt_unhold_writers(m);
3950
3951                if (!err && kattr->propagation)
3952                        change_mnt_propagation(m, kattr->propagation);
3953
3954                /*
3955                 * On failure, only cleanup until we found the first mount
3956                 * we failed to handle.
3957                 */
3958                if (err && m == last)
3959                        break;
3960        } while (kattr->recurse && (m = next_mnt(m, mnt)));
3961
3962        if (!err)
3963                touch_mnt_namespace(mnt->mnt_ns);
3964}
3965
3966static int do_mount_setattr(struct path *path, struct mount_kattr *kattr)
3967{
3968        struct mount *mnt = real_mount(path->mnt), *last = NULL;
3969        int err = 0;
3970
3971        if (path->dentry != mnt->mnt.mnt_root)
3972                return -EINVAL;
3973
3974        if (kattr->propagation) {
3975                /*
3976                 * Only take namespace_lock() if we're actually changing
3977                 * propagation.
3978                 */
3979                namespace_lock();
3980                if (kattr->propagation == MS_SHARED) {
3981                        err = invent_group_ids(mnt, kattr->recurse);
3982                        if (err) {
3983                                namespace_unlock();
3984                                return err;
3985                        }
3986                }
3987        }
3988
3989        lock_mount_hash();
3990
3991        /*
3992         * Get the mount tree in a shape where we can change mount
3993         * properties without failure.
3994         */
3995        last = mount_setattr_prepare(kattr, mnt, &err);
3996        if (last) /* Commit all changes or revert to the old state. */
3997                mount_setattr_commit(kattr, mnt, last, err);
3998
3999        unlock_mount_hash();
4000
4001        if (kattr->propagation) {
4002                namespace_unlock();
4003                if (err)
4004                        cleanup_group_ids(mnt, NULL);
4005        }
4006
4007        return err;
4008}
4009
4010static int build_mount_idmapped(const struct mount_attr *attr, size_t usize,
4011                                struct mount_kattr *kattr, unsigned int flags)
4012{
4013        int err = 0;
4014        struct ns_common *ns;
4015        struct user_namespace *mnt_userns;
4016        struct file *file;
4017
4018        if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP))
4019                return 0;
4020
4021        /*
4022         * We currently do not support clearing an idmapped mount. If this ever
4023         * is a use-case we can revisit this but for now let's keep it simple
4024         * and not allow it.
4025         */
4026        if (attr->attr_clr & MOUNT_ATTR_IDMAP)
4027                return -EINVAL;
4028
4029        if (attr->userns_fd > INT_MAX)
4030                return -EINVAL;
4031
4032        file = fget(attr->userns_fd);
4033        if (!file)
4034                return -EBADF;
4035
4036        if (!proc_ns_file(file)) {
4037                err = -EINVAL;
4038                goto out_fput;
4039        }
4040
4041        ns = get_proc_ns(file_inode(file));
4042        if (ns->ops->type != CLONE_NEWUSER) {
4043                err = -EINVAL;
4044                goto out_fput;
4045        }
4046
4047        /*
4048         * The init_user_ns is used to indicate that a vfsmount is not idmapped.
4049         * This is simpler than just having to treat NULL as unmapped. Users
4050         * wanting to idmap a mount to init_user_ns can just use a namespace
4051         * with an identity mapping.
4052         */
4053        mnt_userns = container_of(ns, struct user_namespace, ns);
4054        if (mnt_userns == &init_user_ns) {
4055                err = -EPERM;
4056                goto out_fput;
4057        }
4058        kattr->mnt_userns = get_user_ns(mnt_userns);
4059
4060out_fput:
4061        fput(file);
4062        return err;
4063}
4064
4065static int build_mount_kattr(const struct mount_attr *attr, size_t usize,
4066                             struct mount_kattr *kattr, unsigned int flags)
4067{
4068        unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
4069
4070        if (flags & AT_NO_AUTOMOUNT)
4071                lookup_flags &= ~LOOKUP_AUTOMOUNT;
4072        if (flags & AT_SYMLINK_NOFOLLOW)
4073                lookup_flags &= ~LOOKUP_FOLLOW;
4074        if (flags & AT_EMPTY_PATH)
4075                lookup_flags |= LOOKUP_EMPTY;
4076
4077        *kattr = (struct mount_kattr) {
4078                .lookup_flags   = lookup_flags,
4079                .recurse        = !!(flags & AT_RECURSIVE),
4080        };
4081
4082        if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS)
4083                return -EINVAL;
4084        if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1)
4085                return -EINVAL;
4086        kattr->propagation = attr->propagation;
4087
4088        if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS)
4089                return -EINVAL;
4090
4091        kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set);
4092        kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr);
4093
4094        /*
4095         * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap,
4096         * users wanting to transition to a different atime setting cannot
4097         * simply specify the atime setting in @attr_set, but must also
4098         * specify MOUNT_ATTR__ATIME in the @attr_clr field.
4099         * So ensure that MOUNT_ATTR__ATIME can't be partially set in
4100         * @attr_clr and that @attr_set can't have any atime bits set if
4101         * MOUNT_ATTR__ATIME isn't set in @attr_clr.
4102         */
4103        if (attr->attr_clr & MOUNT_ATTR__ATIME) {
4104                if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME)
4105                        return -EINVAL;
4106
4107                /*
4108                 * Clear all previous time settings as they are mutually
4109                 * exclusive.
4110                 */
4111                kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME;
4112                switch (attr->attr_set & MOUNT_ATTR__ATIME) {
4113                case MOUNT_ATTR_RELATIME:
4114                        kattr->attr_set |= MNT_RELATIME;
4115                        break;
4116                case MOUNT_ATTR_NOATIME:
4117                        kattr->attr_set |= MNT_NOATIME;
4118                        break;
4119                case MOUNT_ATTR_STRICTATIME:
4120                        break;
4121                default:
4122                        return -EINVAL;
4123                }
4124        } else {
4125                if (attr->attr_set & MOUNT_ATTR__ATIME)
4126                        return -EINVAL;
4127        }
4128
4129        return build_mount_idmapped(attr, usize, kattr, flags);
4130}
4131
4132static void finish_mount_kattr(struct mount_kattr *kattr)
4133{
4134        put_user_ns(kattr->mnt_userns);
4135        kattr->mnt_userns = NULL;
4136}
4137
4138SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path,
4139                unsigned int, flags, struct mount_attr __user *, uattr,
4140                size_t, usize)
4141{
4142        int err;
4143        struct path target;
4144        struct mount_attr attr;
4145        struct mount_kattr kattr;
4146
4147        BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0);
4148
4149        if (flags & ~(AT_EMPTY_PATH |
4150                      AT_RECURSIVE |
4151                      AT_SYMLINK_NOFOLLOW |
4152                      AT_NO_AUTOMOUNT))
4153                return -EINVAL;
4154
4155        if (unlikely(usize > PAGE_SIZE))
4156                return -E2BIG;
4157        if (unlikely(usize < MOUNT_ATTR_SIZE_VER0))
4158                return -EINVAL;
4159
4160        if (!may_mount())
4161                return -EPERM;
4162
4163        err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize);
4164        if (err)
4165                return err;
4166
4167        /* Don't bother walking through the mounts if this is a nop. */
4168        if (attr.attr_set == 0 &&
4169            attr.attr_clr == 0 &&
4170            attr.propagation == 0)
4171                return 0;
4172
4173        err = build_mount_kattr(&attr, usize, &kattr, flags);
4174        if (err)
4175                return err;
4176
4177        err = user_path_at(dfd, path, kattr.lookup_flags, &target);
4178        if (err)
4179                return err;
4180
4181        err = do_mount_setattr(&target, &kattr);
4182        finish_mount_kattr(&kattr);
4183        path_put(&target);
4184        return err;
4185}
4186
4187static void __init init_mount_tree(void)
4188{
4189        struct vfsmount *mnt;
4190        struct mount *m;
4191        struct mnt_namespace *ns;
4192        struct path root;
4193
4194        mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL);
4195        if (IS_ERR(mnt))
4196                panic("Can't create rootfs");
4197
4198        ns = alloc_mnt_ns(&init_user_ns, false);
4199        if (IS_ERR(ns))
4200                panic("Can't allocate initial namespace");
4201        m = real_mount(mnt);
4202        m->mnt_ns = ns;
4203        ns->root = m;
4204        ns->mounts = 1;
4205        list_add(&m->mnt_list, &ns->list);
4206        init_task.nsproxy->mnt_ns = ns;
4207        get_mnt_ns(ns);
4208
4209        root.mnt = mnt;
4210        root.dentry = mnt->mnt_root;
4211        mnt->mnt_flags |= MNT_LOCKED;
4212
4213        set_fs_pwd(current->fs, &root);
4214        set_fs_root(current->fs, &root);
4215}
4216
4217void __init mnt_init(void)
4218{
4219        int err;
4220
4221        mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
4222                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
4223
4224        mount_hashtable = alloc_large_system_hash("Mount-cache",
4225                                sizeof(struct hlist_head),
4226                                mhash_entries, 19,
4227                                HASH_ZERO,
4228                                &m_hash_shift, &m_hash_mask, 0, 0);
4229        mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
4230                                sizeof(struct hlist_head),
4231                                mphash_entries, 19,
4232                                HASH_ZERO,
4233                                &mp_hash_shift, &mp_hash_mask, 0, 0);
4234
4235        if (!mount_hashtable || !mountpoint_hashtable)
4236                panic("Failed to allocate mount hash table\n");
4237
4238        kernfs_init();
4239
4240        err = sysfs_init();
4241        if (err)
4242                printk(KERN_WARNING "%s: sysfs_init error: %d\n",
4243                        __func__, err);
4244        fs_kobj = kobject_create_and_add("fs", NULL);
4245        if (!fs_kobj)
4246                printk(KERN_WARNING "%s: kobj create error\n", __func__);
4247        shmem_init();
4248        init_rootfs();
4249        init_mount_tree();
4250}
4251
4252void put_mnt_ns(struct mnt_namespace *ns)
4253{
4254        if (!refcount_dec_and_test(&ns->ns.count))
4255                return;
4256        drop_collected_mounts(&ns->root->mnt);
4257        free_mnt_ns(ns);
4258}
4259
4260struct vfsmount *kern_mount(struct file_system_type *type)
4261{
4262        struct vfsmount *mnt;
4263        mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
4264        if (!IS_ERR(mnt)) {
4265                /*
4266                 * it is a longterm mount, don't release mnt until
4267                 * we unmount before file sys is unregistered
4268                */
4269                real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
4270        }
4271        return mnt;
4272}
4273EXPORT_SYMBOL_GPL(kern_mount);
4274
4275void kern_unmount(struct vfsmount *mnt)
4276{
4277        /* release long term mount so mount point can be released */
4278        if (!IS_ERR_OR_NULL(mnt)) {
4279                real_mount(mnt)->mnt_ns = NULL;
4280                synchronize_rcu();      /* yecchhh... */
4281                mntput(mnt);
4282        }
4283}
4284EXPORT_SYMBOL(kern_unmount);
4285
4286void kern_unmount_array(struct vfsmount *mnt[], unsigned int num)
4287{
4288        unsigned int i;
4289
4290        for (i = 0; i < num; i++)
4291                if (mnt[i])
4292                        real_mount(mnt[i])->mnt_ns = NULL;
4293        synchronize_rcu_expedited();
4294        for (i = 0; i < num; i++)
4295                mntput(mnt[i]);
4296}
4297EXPORT_SYMBOL(kern_unmount_array);
4298
4299bool our_mnt(struct vfsmount *mnt)
4300{
4301        return check_mnt(real_mount(mnt));
4302}
4303
4304bool current_chrooted(void)
4305{
4306        /* Does the current process have a non-standard root */
4307        struct path ns_root;
4308        struct path fs_root;
4309        bool chrooted;
4310
4311        /* Find the namespace root */
4312        ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
4313        ns_root.dentry = ns_root.mnt->mnt_root;
4314        path_get(&ns_root);
4315        while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
4316                ;
4317
4318        get_fs_root(current->fs, &fs_root);
4319
4320        chrooted = !path_equal(&fs_root, &ns_root);
4321
4322        path_put(&fs_root);
4323        path_put(&ns_root);
4324
4325        return chrooted;
4326}
4327
4328static bool mnt_already_visible(struct mnt_namespace *ns,
4329                                const struct super_block *sb,
4330                                int *new_mnt_flags)
4331{
4332        int new_flags = *new_mnt_flags;
4333        struct mount *mnt;
4334        bool visible = false;
4335
4336        down_read(&namespace_sem);
4337        lock_ns_list(ns);
4338        list_for_each_entry(mnt, &ns->list, mnt_list) {
4339                struct mount *child;
4340                int mnt_flags;
4341
4342                if (mnt_is_cursor(mnt))
4343                        continue;
4344
4345                if (mnt->mnt.mnt_sb->s_type != sb->s_type)
4346                        continue;
4347
4348                /* This mount is not fully visible if it's root directory
4349                 * is not the root directory of the filesystem.
4350                 */
4351                if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
4352                        continue;
4353
4354                /* A local view of the mount flags */
4355                mnt_flags = mnt->mnt.mnt_flags;
4356
4357                /* Don't miss readonly hidden in the superblock flags */
4358                if (sb_rdonly(mnt->mnt.mnt_sb))
4359                        mnt_flags |= MNT_LOCK_READONLY;
4360
4361                /* Verify the mount flags are equal to or more permissive
4362                 * than the proposed new mount.
4363                 */
4364                if ((mnt_flags & MNT_LOCK_READONLY) &&
4365                    !(new_flags & MNT_READONLY))
4366                        continue;
4367                if ((mnt_flags & MNT_LOCK_ATIME) &&
4368                    ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
4369                        continue;
4370
4371                /* This mount is not fully visible if there are any
4372                 * locked child mounts that cover anything except for
4373                 * empty directories.
4374                 */
4375                list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
4376                        struct inode *inode = child->mnt_mountpoint->d_inode;
4377                        /* Only worry about locked mounts */
4378                        if (!(child->mnt.mnt_flags & MNT_LOCKED))
4379                                continue;
4380                        /* Is the directory permanetly empty? */
4381                        if (!is_empty_dir_inode(inode))
4382                                goto next;
4383                }
4384                /* Preserve the locked attributes */
4385                *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
4386                                               MNT_LOCK_ATIME);
4387                visible = true;
4388                goto found;
4389        next:   ;
4390        }
4391found:
4392        unlock_ns_list(ns);
4393        up_read(&namespace_sem);
4394        return visible;
4395}
4396
4397static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags)
4398{
4399        const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
4400        struct mnt_namespace *ns = current->nsproxy->mnt_ns;
4401        unsigned long s_iflags;
4402
4403        if (ns->user_ns == &init_user_ns)
4404                return false;
4405
4406        /* Can this filesystem be too revealing? */
4407        s_iflags = sb->s_iflags;
4408        if (!(s_iflags & SB_I_USERNS_VISIBLE))
4409                return false;
4410
4411        if ((s_iflags & required_iflags) != required_iflags) {
4412                WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
4413                          required_iflags);
4414                return true;
4415        }
4416
4417        return !mnt_already_visible(ns, sb, new_mnt_flags);
4418}
4419
4420bool mnt_may_suid(struct vfsmount *mnt)
4421{
4422        /*
4423         * Foreign mounts (accessed via fchdir or through /proc
4424         * symlinks) are always treated as if they are nosuid.  This
4425         * prevents namespaces from trusting potentially unsafe
4426         * suid/sgid bits, file caps, or security labels that originate
4427         * in other namespaces.
4428         */
4429        return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
4430               current_in_userns(mnt->mnt_sb->s_user_ns);
4431}
4432
4433static struct ns_common *mntns_get(struct task_struct *task)
4434{
4435        struct ns_common *ns = NULL;
4436        struct nsproxy *nsproxy;
4437
4438        task_lock(task);
4439        nsproxy = task->nsproxy;
4440        if (nsproxy) {
4441                ns = &nsproxy->mnt_ns->ns;
4442                get_mnt_ns(to_mnt_ns(ns));
4443        }
4444        task_unlock(task);
4445
4446        return ns;
4447}
4448
4449static void mntns_put(struct ns_common *ns)
4450{
4451        put_mnt_ns(to_mnt_ns(ns));
4452}
4453
4454static int mntns_install(struct nsset *nsset, struct ns_common *ns)
4455{
4456        struct nsproxy *nsproxy = nsset->nsproxy;
4457        struct fs_struct *fs = nsset->fs;
4458        struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
4459        struct user_namespace *user_ns = nsset->cred->user_ns;
4460        struct path root;
4461        int err;
4462
4463        if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
4464            !ns_capable(user_ns, CAP_SYS_CHROOT) ||
4465            !ns_capable(user_ns, CAP_SYS_ADMIN))
4466                return -EPERM;
4467
4468        if (is_anon_ns(mnt_ns))
4469                return -EINVAL;
4470
4471        if (fs->users != 1)
4472                return -EINVAL;
4473
4474        get_mnt_ns(mnt_ns);
4475        old_mnt_ns = nsproxy->mnt_ns;
4476        nsproxy->mnt_ns = mnt_ns;
4477
4478        /* Find the root */
4479        err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
4480                                "/", LOOKUP_DOWN, &root);
4481        if (err) {
4482                /* revert to old namespace */
4483                nsproxy->mnt_ns = old_mnt_ns;
4484                put_mnt_ns(mnt_ns);
4485                return err;
4486        }
4487
4488        put_mnt_ns(old_mnt_ns);
4489
4490        /* Update the pwd and root */
4491        set_fs_pwd(fs, &root);
4492        set_fs_root(fs, &root);
4493
4494        path_put(&root);
4495        return 0;
4496}
4497
4498static struct user_namespace *mntns_owner(struct ns_common *ns)
4499{
4500        return to_mnt_ns(ns)->user_ns;
4501}
4502
4503const struct proc_ns_operations mntns_operations = {
4504        .name           = "mnt",
4505        .type           = CLONE_NEWNS,
4506        .get            = mntns_get,
4507        .put            = mntns_put,
4508        .install        = mntns_install,
4509        .owner          = mntns_owner,
4510};
4511