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#include <linux/mnt_idmapping.h>
  35
  36#include "pnode.h"
  37#include "internal.h"
  38
  39/* Maximum number of mounts in a mount namespace */
  40static unsigned int sysctl_mount_max __read_mostly = 100000;
  41
  42static unsigned int m_hash_mask __read_mostly;
  43static unsigned int m_hash_shift __read_mostly;
  44static unsigned int mp_hash_mask __read_mostly;
  45static unsigned int mp_hash_shift __read_mostly;
  46
  47static __initdata unsigned long mhash_entries;
  48static int __init set_mhash_entries(char *str)
  49{
  50        if (!str)
  51                return 0;
  52        mhash_entries = simple_strtoul(str, &str, 0);
  53        return 1;
  54}
  55__setup("mhash_entries=", set_mhash_entries);
  56
  57static __initdata unsigned long mphash_entries;
  58static int __init set_mphash_entries(char *str)
  59{
  60        if (!str)
  61                return 0;
  62        mphash_entries = simple_strtoul(str, &str, 0);
  63        return 1;
  64}
  65__setup("mphash_entries=", set_mphash_entries);
  66
  67static u64 event;
  68static DEFINE_IDA(mnt_id_ida);
  69static DEFINE_IDA(mnt_group_ida);
  70
  71static struct hlist_head *mount_hashtable __read_mostly;
  72static struct hlist_head *mountpoint_hashtable __read_mostly;
  73static struct kmem_cache *mnt_cache __read_mostly;
  74static DECLARE_RWSEM(namespace_sem);
  75static HLIST_HEAD(unmounted);   /* protected by namespace_sem */
  76static LIST_HEAD(ex_mountpoints); /* protected by namespace_sem */
  77
  78struct mount_kattr {
  79        unsigned int attr_set;
  80        unsigned int attr_clr;
  81        unsigned int propagation;
  82        unsigned int lookup_flags;
  83        bool recurse;
  84        struct user_namespace *mnt_userns;
  85};
  86
  87/* /sys/fs */
  88struct kobject *fs_kobj;
  89EXPORT_SYMBOL_GPL(fs_kobj);
  90
  91/*
  92 * vfsmount lock may be taken for read to prevent changes to the
  93 * vfsmount hash, ie. during mountpoint lookups or walking back
  94 * up the tree.
  95 *
  96 * It should be taken for write in all cases where the vfsmount
  97 * tree or hash is modified or when a vfsmount structure is modified.
  98 */
  99__cacheline_aligned_in_smp DEFINE_SEQLOCK(mount_lock);
 100
 101static inline void lock_mount_hash(void)
 102{
 103        write_seqlock(&mount_lock);
 104}
 105
 106static inline void unlock_mount_hash(void)
 107{
 108        write_sequnlock(&mount_lock);
 109}
 110
 111static inline struct hlist_head *m_hash(struct vfsmount *mnt, struct dentry *dentry)
 112{
 113        unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
 114        tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
 115        tmp = tmp + (tmp >> m_hash_shift);
 116        return &mount_hashtable[tmp & m_hash_mask];
 117}
 118
 119static inline struct hlist_head *mp_hash(struct dentry *dentry)
 120{
 121        unsigned long tmp = ((unsigned long)dentry / L1_CACHE_BYTES);
 122        tmp = tmp + (tmp >> mp_hash_shift);
 123        return &mountpoint_hashtable[tmp & mp_hash_mask];
 124}
 125
 126static int mnt_alloc_id(struct mount *mnt)
 127{
 128        int res = ida_alloc(&mnt_id_ida, GFP_KERNEL);
 129
 130        if (res < 0)
 131                return res;
 132        mnt->mnt_id = res;
 133        return 0;
 134}
 135
 136static void mnt_free_id(struct mount *mnt)
 137{
 138        ida_free(&mnt_id_ida, mnt->mnt_id);
 139}
 140
 141/*
 142 * Allocate a new peer group ID
 143 */
 144static int mnt_alloc_group_id(struct mount *mnt)
 145{
 146        int res = ida_alloc_min(&mnt_group_ida, 1, GFP_KERNEL);
 147
 148        if (res < 0)
 149                return res;
 150        mnt->mnt_group_id = res;
 151        return 0;
 152}
 153
 154/*
 155 * Release a peer group ID
 156 */
 157void mnt_release_group_id(struct mount *mnt)
 158{
 159        ida_free(&mnt_group_ida, mnt->mnt_group_id);
 160        mnt->mnt_group_id = 0;
 161}
 162
 163/*
 164 * vfsmount lock must be held for read
 165 */
 166static inline void mnt_add_count(struct mount *mnt, int n)
 167{
 168#ifdef CONFIG_SMP
 169        this_cpu_add(mnt->mnt_pcp->mnt_count, n);
 170#else
 171        preempt_disable();
 172        mnt->mnt_count += n;
 173        preempt_enable();
 174#endif
 175}
 176
 177/*
 178 * vfsmount lock must be held for write
 179 */
 180int mnt_get_count(struct mount *mnt)
 181{
 182#ifdef CONFIG_SMP
 183        int count = 0;
 184        int cpu;
 185
 186        for_each_possible_cpu(cpu) {
 187                count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
 188        }
 189
 190        return count;
 191#else
 192        return mnt->mnt_count;
 193#endif
 194}
 195
 196static struct mount *alloc_vfsmnt(const char *name)
 197{
 198        struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
 199        if (mnt) {
 200                int err;
 201
 202                err = mnt_alloc_id(mnt);
 203                if (err)
 204                        goto out_free_cache;
 205
 206                if (name) {
 207                        mnt->mnt_devname = kstrdup_const(name,
 208                                                         GFP_KERNEL_ACCOUNT);
 209                        if (!mnt->mnt_devname)
 210                                goto out_free_id;
 211                }
 212
 213#ifdef CONFIG_SMP
 214                mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
 215                if (!mnt->mnt_pcp)
 216                        goto out_free_devname;
 217
 218                this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
 219#else
 220                mnt->mnt_count = 1;
 221                mnt->mnt_writers = 0;
 222#endif
 223
 224                INIT_HLIST_NODE(&mnt->mnt_hash);
 225                INIT_LIST_HEAD(&mnt->mnt_child);
 226                INIT_LIST_HEAD(&mnt->mnt_mounts);
 227                INIT_LIST_HEAD(&mnt->mnt_list);
 228                INIT_LIST_HEAD(&mnt->mnt_expire);
 229                INIT_LIST_HEAD(&mnt->mnt_share);
 230                INIT_LIST_HEAD(&mnt->mnt_slave_list);
 231                INIT_LIST_HEAD(&mnt->mnt_slave);
 232                INIT_HLIST_NODE(&mnt->mnt_mp_list);
 233                INIT_LIST_HEAD(&mnt->mnt_umounting);
 234                INIT_HLIST_HEAD(&mnt->mnt_stuck_children);
 235                mnt->mnt.mnt_userns = &init_user_ns;
 236        }
 237        return mnt;
 238
 239#ifdef CONFIG_SMP
 240out_free_devname:
 241        kfree_const(mnt->mnt_devname);
 242#endif
 243out_free_id:
 244        mnt_free_id(mnt);
 245out_free_cache:
 246        kmem_cache_free(mnt_cache, mnt);
 247        return NULL;
 248}
 249
 250/*
 251 * Most r/o checks on a fs are for operations that take
 252 * discrete amounts of time, like a write() or unlink().
 253 * We must keep track of when those operations start
 254 * (for permission checks) and when they end, so that
 255 * we can determine when writes are able to occur to
 256 * a filesystem.
 257 */
 258/*
 259 * __mnt_is_readonly: check whether a mount is read-only
 260 * @mnt: the mount to check for its write status
 261 *
 262 * This shouldn't be used directly ouside of the VFS.
 263 * It does not guarantee that the filesystem will stay
 264 * r/w, just that it is right *now*.  This can not and
 265 * should not be used in place of IS_RDONLY(inode).
 266 * mnt_want/drop_write() will _keep_ the filesystem
 267 * r/w.
 268 */
 269bool __mnt_is_readonly(struct vfsmount *mnt)
 270{
 271        return (mnt->mnt_flags & MNT_READONLY) || sb_rdonly(mnt->mnt_sb);
 272}
 273EXPORT_SYMBOL_GPL(__mnt_is_readonly);
 274
 275static inline void mnt_inc_writers(struct mount *mnt)
 276{
 277#ifdef CONFIG_SMP
 278        this_cpu_inc(mnt->mnt_pcp->mnt_writers);
 279#else
 280        mnt->mnt_writers++;
 281#endif
 282}
 283
 284static inline void mnt_dec_writers(struct mount *mnt)
 285{
 286#ifdef CONFIG_SMP
 287        this_cpu_dec(mnt->mnt_pcp->mnt_writers);
 288#else
 289        mnt->mnt_writers--;
 290#endif
 291}
 292
 293static unsigned int mnt_get_writers(struct mount *mnt)
 294{
 295#ifdef CONFIG_SMP
 296        unsigned int count = 0;
 297        int cpu;
 298
 299        for_each_possible_cpu(cpu) {
 300                count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
 301        }
 302
 303        return count;
 304#else
 305        return mnt->mnt_writers;
 306#endif
 307}
 308
 309static int mnt_is_readonly(struct vfsmount *mnt)
 310{
 311        if (mnt->mnt_sb->s_readonly_remount)
 312                return 1;
 313        /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
 314        smp_rmb();
 315        return __mnt_is_readonly(mnt);
 316}
 317
 318/*
 319 * Most r/o & frozen checks on a fs are for operations that take discrete
 320 * amounts of time, like a write() or unlink().  We must keep track of when
 321 * those operations start (for permission checks) and when they end, so that we
 322 * can determine when writes are able to occur to a filesystem.
 323 */
 324/**
 325 * __mnt_want_write - get write access to a mount without freeze protection
 326 * @m: the mount on which to take a write
 327 *
 328 * This tells the low-level filesystem that a write is about to be performed to
 329 * it, and makes sure that writes are allowed (mnt it read-write) before
 330 * returning success. This operation does not protect against filesystem being
 331 * frozen. When the write operation is finished, __mnt_drop_write() must be
 332 * called. This is effectively a refcount.
 333 */
 334int __mnt_want_write(struct vfsmount *m)
 335{
 336        struct mount *mnt = real_mount(m);
 337        int ret = 0;
 338
 339        preempt_disable();
 340        mnt_inc_writers(mnt);
 341        /*
 342         * The store to mnt_inc_writers must be visible before we pass
 343         * MNT_WRITE_HOLD loop below, so that the slowpath can see our
 344         * incremented count after it has set MNT_WRITE_HOLD.
 345         */
 346        smp_mb();
 347        while (READ_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
 348                cpu_relax();
 349        /*
 350         * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
 351         * be set to match its requirements. So we must not load that until
 352         * MNT_WRITE_HOLD is cleared.
 353         */
 354        smp_rmb();
 355        if (mnt_is_readonly(m)) {
 356                mnt_dec_writers(mnt);
 357                ret = -EROFS;
 358        }
 359        preempt_enable();
 360
 361        return ret;
 362}
 363
 364/**
 365 * mnt_want_write - get write access to a mount
 366 * @m: the mount on which to take a write
 367 *
 368 * This tells the low-level filesystem that a write is about to be performed to
 369 * it, and makes sure that writes are allowed (mount is read-write, filesystem
 370 * is not frozen) before returning success.  When the write operation is
 371 * finished, mnt_drop_write() must be called.  This is effectively a refcount.
 372 */
 373int mnt_want_write(struct vfsmount *m)
 374{
 375        int ret;
 376
 377        sb_start_write(m->mnt_sb);
 378        ret = __mnt_want_write(m);
 379        if (ret)
 380                sb_end_write(m->mnt_sb);
 381        return ret;
 382}
 383EXPORT_SYMBOL_GPL(mnt_want_write);
 384
 385/**
 386 * __mnt_want_write_file - get write access to a file's mount
 387 * @file: the file who's mount on which to take a write
 388 *
 389 * This is like __mnt_want_write, but if the file is already open for writing it
 390 * skips incrementing mnt_writers (since the open file already has a reference)
 391 * and instead only does the check for emergency r/o remounts.  This must be
 392 * paired with __mnt_drop_write_file.
 393 */
 394int __mnt_want_write_file(struct file *file)
 395{
 396        if (file->f_mode & FMODE_WRITER) {
 397                /*
 398                 * Superblock may have become readonly while there are still
 399                 * writable fd's, e.g. due to a fs error with errors=remount-ro
 400                 */
 401                if (__mnt_is_readonly(file->f_path.mnt))
 402                        return -EROFS;
 403                return 0;
 404        }
 405        return __mnt_want_write(file->f_path.mnt);
 406}
 407
 408/**
 409 * mnt_want_write_file - get write access to a file's mount
 410 * @file: the file who's mount on which to take a write
 411 *
 412 * This is like mnt_want_write, but if the file is already open for writing it
 413 * skips incrementing mnt_writers (since the open file already has a reference)
 414 * and instead only does the freeze protection and the check for emergency r/o
 415 * remounts.  This must be paired with mnt_drop_write_file.
 416 */
 417int mnt_want_write_file(struct file *file)
 418{
 419        int ret;
 420
 421        sb_start_write(file_inode(file)->i_sb);
 422        ret = __mnt_want_write_file(file);
 423        if (ret)
 424                sb_end_write(file_inode(file)->i_sb);
 425        return ret;
 426}
 427EXPORT_SYMBOL_GPL(mnt_want_write_file);
 428
 429/**
 430 * __mnt_drop_write - give up write access to a mount
 431 * @mnt: the mount on which to give up write access
 432 *
 433 * Tells the low-level filesystem that we are done
 434 * performing writes to it.  Must be matched with
 435 * __mnt_want_write() call above.
 436 */
 437void __mnt_drop_write(struct vfsmount *mnt)
 438{
 439        preempt_disable();
 440        mnt_dec_writers(real_mount(mnt));
 441        preempt_enable();
 442}
 443
 444/**
 445 * mnt_drop_write - give up write access to a mount
 446 * @mnt: the mount on which to give up write access
 447 *
 448 * Tells the low-level filesystem that we are done performing writes to it and
 449 * also allows filesystem to be frozen again.  Must be matched with
 450 * mnt_want_write() call above.
 451 */
 452void mnt_drop_write(struct vfsmount *mnt)
 453{
 454        __mnt_drop_write(mnt);
 455        sb_end_write(mnt->mnt_sb);
 456}
 457EXPORT_SYMBOL_GPL(mnt_drop_write);
 458
 459void __mnt_drop_write_file(struct file *file)
 460{
 461        if (!(file->f_mode & FMODE_WRITER))
 462                __mnt_drop_write(file->f_path.mnt);
 463}
 464
 465void mnt_drop_write_file(struct file *file)
 466{
 467        __mnt_drop_write_file(file);
 468        sb_end_write(file_inode(file)->i_sb);
 469}
 470EXPORT_SYMBOL(mnt_drop_write_file);
 471
 472/**
 473 * mnt_hold_writers - prevent write access to the given mount
 474 * @mnt: mnt to prevent write access to
 475 *
 476 * Prevents write access to @mnt if there are no active writers for @mnt.
 477 * This function needs to be called and return successfully before changing
 478 * properties of @mnt that need to remain stable for callers with write access
 479 * to @mnt.
 480 *
 481 * After this functions has been called successfully callers must pair it with
 482 * a call to mnt_unhold_writers() in order to stop preventing write access to
 483 * @mnt.
 484 *
 485 * Context: This function expects lock_mount_hash() to be held serializing
 486 *          setting MNT_WRITE_HOLD.
 487 * Return: On success 0 is returned.
 488 *         On error, -EBUSY is returned.
 489 */
 490static inline int mnt_hold_writers(struct mount *mnt)
 491{
 492        mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
 493        /*
 494         * After storing MNT_WRITE_HOLD, we'll read the counters. This store
 495         * should be visible before we do.
 496         */
 497        smp_mb();
 498
 499        /*
 500         * With writers on hold, if this value is zero, then there are
 501         * definitely no active writers (although held writers may subsequently
 502         * increment the count, they'll have to wait, and decrement it after
 503         * seeing MNT_READONLY).
 504         *
 505         * It is OK to have counter incremented on one CPU and decremented on
 506         * another: the sum will add up correctly. The danger would be when we
 507         * sum up each counter, if we read a counter before it is incremented,
 508         * but then read another CPU's count which it has been subsequently
 509         * decremented from -- we would see more decrements than we should.
 510         * MNT_WRITE_HOLD protects against this scenario, because
 511         * mnt_want_write first increments count, then smp_mb, then spins on
 512         * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
 513         * we're counting up here.
 514         */
 515        if (mnt_get_writers(mnt) > 0)
 516                return -EBUSY;
 517
 518        return 0;
 519}
 520
 521/**
 522 * mnt_unhold_writers - stop preventing write access to the given mount
 523 * @mnt: mnt to stop preventing write access to
 524 *
 525 * Stop preventing write access to @mnt allowing callers to gain write access
 526 * to @mnt again.
 527 *
 528 * This function can only be called after a successful call to
 529 * mnt_hold_writers().
 530 *
 531 * Context: This function expects lock_mount_hash() to be held.
 532 */
 533static inline void mnt_unhold_writers(struct mount *mnt)
 534{
 535        /*
 536         * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
 537         * that become unheld will see MNT_READONLY.
 538         */
 539        smp_wmb();
 540        mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
 541}
 542
 543static int mnt_make_readonly(struct mount *mnt)
 544{
 545        int ret;
 546
 547        ret = mnt_hold_writers(mnt);
 548        if (!ret)
 549                mnt->mnt.mnt_flags |= MNT_READONLY;
 550        mnt_unhold_writers(mnt);
 551        return ret;
 552}
 553
 554int sb_prepare_remount_readonly(struct super_block *sb)
 555{
 556        struct mount *mnt;
 557        int err = 0;
 558
 559        /* Racy optimization.  Recheck the counter under MNT_WRITE_HOLD */
 560        if (atomic_long_read(&sb->s_remove_count))
 561                return -EBUSY;
 562
 563        lock_mount_hash();
 564        list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
 565                if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
 566                        mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
 567                        smp_mb();
 568                        if (mnt_get_writers(mnt) > 0) {
 569                                err = -EBUSY;
 570                                break;
 571                        }
 572                }
 573        }
 574        if (!err && atomic_long_read(&sb->s_remove_count))
 575                err = -EBUSY;
 576
 577        if (!err) {
 578                sb->s_readonly_remount = 1;
 579                smp_wmb();
 580        }
 581        list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
 582                if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
 583                        mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
 584        }
 585        unlock_mount_hash();
 586
 587        return err;
 588}
 589
 590static void free_vfsmnt(struct mount *mnt)
 591{
 592        struct user_namespace *mnt_userns;
 593
 594        mnt_userns = mnt_user_ns(&mnt->mnt);
 595        if (!initial_idmapping(mnt_userns))
 596                put_user_ns(mnt_userns);
 597        kfree_const(mnt->mnt_devname);
 598#ifdef CONFIG_SMP
 599        free_percpu(mnt->mnt_pcp);
 600#endif
 601        kmem_cache_free(mnt_cache, mnt);
 602}
 603
 604static void delayed_free_vfsmnt(struct rcu_head *head)
 605{
 606        free_vfsmnt(container_of(head, struct mount, mnt_rcu));
 607}
 608
 609/* call under rcu_read_lock */
 610int __legitimize_mnt(struct vfsmount *bastard, unsigned seq)
 611{
 612        struct mount *mnt;
 613        if (read_seqretry(&mount_lock, seq))
 614                return 1;
 615        if (bastard == NULL)
 616                return 0;
 617        mnt = real_mount(bastard);
 618        mnt_add_count(mnt, 1);
 619        smp_mb();                       // see mntput_no_expire()
 620        if (likely(!read_seqretry(&mount_lock, seq)))
 621                return 0;
 622        if (bastard->mnt_flags & MNT_SYNC_UMOUNT) {
 623                mnt_add_count(mnt, -1);
 624                return 1;
 625        }
 626        lock_mount_hash();
 627        if (unlikely(bastard->mnt_flags & MNT_DOOMED)) {
 628                mnt_add_count(mnt, -1);
 629                unlock_mount_hash();
 630                return 1;
 631        }
 632        unlock_mount_hash();
 633        /* caller will mntput() */
 634        return -1;
 635}
 636
 637/* call under rcu_read_lock */
 638bool legitimize_mnt(struct vfsmount *bastard, unsigned seq)
 639{
 640        int res = __legitimize_mnt(bastard, seq);
 641        if (likely(!res))
 642                return true;
 643        if (unlikely(res < 0)) {
 644                rcu_read_unlock();
 645                mntput(bastard);
 646                rcu_read_lock();
 647        }
 648        return false;
 649}
 650
 651/*
 652 * find the first mount at @dentry on vfsmount @mnt.
 653 * call under rcu_read_lock()
 654 */
 655struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
 656{
 657        struct hlist_head *head = m_hash(mnt, dentry);
 658        struct mount *p;
 659
 660        hlist_for_each_entry_rcu(p, head, mnt_hash)
 661                if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry)
 662                        return p;
 663        return NULL;
 664}
 665
 666/*
 667 * lookup_mnt - Return the first child mount mounted at path
 668 *
 669 * "First" means first mounted chronologically.  If you create the
 670 * following mounts:
 671 *
 672 * mount /dev/sda1 /mnt
 673 * mount /dev/sda2 /mnt
 674 * mount /dev/sda3 /mnt
 675 *
 676 * Then lookup_mnt() on the base /mnt dentry in the root mount will
 677 * return successively the root dentry and vfsmount of /dev/sda1, then
 678 * /dev/sda2, then /dev/sda3, then NULL.
 679 *
 680 * lookup_mnt takes a reference to the found vfsmount.
 681 */
 682struct vfsmount *lookup_mnt(const struct path *path)
 683{
 684        struct mount *child_mnt;
 685        struct vfsmount *m;
 686        unsigned seq;
 687
 688        rcu_read_lock();
 689        do {
 690                seq = read_seqbegin(&mount_lock);
 691                child_mnt = __lookup_mnt(path->mnt, path->dentry);
 692                m = child_mnt ? &child_mnt->mnt : NULL;
 693        } while (!legitimize_mnt(m, seq));
 694        rcu_read_unlock();
 695        return m;
 696}
 697
 698static inline void lock_ns_list(struct mnt_namespace *ns)
 699{
 700        spin_lock(&ns->ns_lock);
 701}
 702
 703static inline void unlock_ns_list(struct mnt_namespace *ns)
 704{
 705        spin_unlock(&ns->ns_lock);
 706}
 707
 708static inline bool mnt_is_cursor(struct mount *mnt)
 709{
 710        return mnt->mnt.mnt_flags & MNT_CURSOR;
 711}
 712
 713/*
 714 * __is_local_mountpoint - Test to see if dentry is a mountpoint in the
 715 *                         current mount namespace.
 716 *
 717 * The common case is dentries are not mountpoints at all and that
 718 * test is handled inline.  For the slow case when we are actually
 719 * dealing with a mountpoint of some kind, walk through all of the
 720 * mounts in the current mount namespace and test to see if the dentry
 721 * is a mountpoint.
 722 *
 723 * The mount_hashtable is not usable in the context because we
 724 * need to identify all mounts that may be in the current mount
 725 * namespace not just a mount that happens to have some specified
 726 * parent mount.
 727 */
 728bool __is_local_mountpoint(struct dentry *dentry)
 729{
 730        struct mnt_namespace *ns = current->nsproxy->mnt_ns;
 731        struct mount *mnt;
 732        bool is_covered = false;
 733
 734        down_read(&namespace_sem);
 735        lock_ns_list(ns);
 736        list_for_each_entry(mnt, &ns->list, mnt_list) {
 737                if (mnt_is_cursor(mnt))
 738                        continue;
 739                is_covered = (mnt->mnt_mountpoint == dentry);
 740                if (is_covered)
 741                        break;
 742        }
 743        unlock_ns_list(ns);
 744        up_read(&namespace_sem);
 745
 746        return is_covered;
 747}
 748
 749static struct mountpoint *lookup_mountpoint(struct dentry *dentry)
 750{
 751        struct hlist_head *chain = mp_hash(dentry);
 752        struct mountpoint *mp;
 753
 754        hlist_for_each_entry(mp, chain, m_hash) {
 755                if (mp->m_dentry == dentry) {
 756                        mp->m_count++;
 757                        return mp;
 758                }
 759        }
 760        return NULL;
 761}
 762
 763static struct mountpoint *get_mountpoint(struct dentry *dentry)
 764{
 765        struct mountpoint *mp, *new = NULL;
 766        int ret;
 767
 768        if (d_mountpoint(dentry)) {
 769                /* might be worth a WARN_ON() */
 770                if (d_unlinked(dentry))
 771                        return ERR_PTR(-ENOENT);
 772mountpoint:
 773                read_seqlock_excl(&mount_lock);
 774                mp = lookup_mountpoint(dentry);
 775                read_sequnlock_excl(&mount_lock);
 776                if (mp)
 777                        goto done;
 778        }
 779
 780        if (!new)
 781                new = kmalloc(sizeof(struct mountpoint), GFP_KERNEL);
 782        if (!new)
 783                return ERR_PTR(-ENOMEM);
 784
 785
 786        /* Exactly one processes may set d_mounted */
 787        ret = d_set_mounted(dentry);
 788
 789        /* Someone else set d_mounted? */
 790        if (ret == -EBUSY)
 791                goto mountpoint;
 792
 793        /* The dentry is not available as a mountpoint? */
 794        mp = ERR_PTR(ret);
 795        if (ret)
 796                goto done;
 797
 798        /* Add the new mountpoint to the hash table */
 799        read_seqlock_excl(&mount_lock);
 800        new->m_dentry = dget(dentry);
 801        new->m_count = 1;
 802        hlist_add_head(&new->m_hash, mp_hash(dentry));
 803        INIT_HLIST_HEAD(&new->m_list);
 804        read_sequnlock_excl(&mount_lock);
 805
 806        mp = new;
 807        new = NULL;
 808done:
 809        kfree(new);
 810        return mp;
 811}
 812
 813/*
 814 * vfsmount lock must be held.  Additionally, the caller is responsible
 815 * for serializing calls for given disposal list.
 816 */
 817static void __put_mountpoint(struct mountpoint *mp, struct list_head *list)
 818{
 819        if (!--mp->m_count) {
 820                struct dentry *dentry = mp->m_dentry;
 821                BUG_ON(!hlist_empty(&mp->m_list));
 822                spin_lock(&dentry->d_lock);
 823                dentry->d_flags &= ~DCACHE_MOUNTED;
 824                spin_unlock(&dentry->d_lock);
 825                dput_to_list(dentry, list);
 826                hlist_del(&mp->m_hash);
 827                kfree(mp);
 828        }
 829}
 830
 831/* called with namespace_lock and vfsmount lock */
 832static void put_mountpoint(struct mountpoint *mp)
 833{
 834        __put_mountpoint(mp, &ex_mountpoints);
 835}
 836
 837static inline int check_mnt(struct mount *mnt)
 838{
 839        return mnt->mnt_ns == current->nsproxy->mnt_ns;
 840}
 841
 842/*
 843 * vfsmount lock must be held for write
 844 */
 845static void touch_mnt_namespace(struct mnt_namespace *ns)
 846{
 847        if (ns) {
 848                ns->event = ++event;
 849                wake_up_interruptible(&ns->poll);
 850        }
 851}
 852
 853/*
 854 * vfsmount lock must be held for write
 855 */
 856static void __touch_mnt_namespace(struct mnt_namespace *ns)
 857{
 858        if (ns && ns->event != event) {
 859                ns->event = event;
 860                wake_up_interruptible(&ns->poll);
 861        }
 862}
 863
 864/*
 865 * vfsmount lock must be held for write
 866 */
 867static struct mountpoint *unhash_mnt(struct mount *mnt)
 868{
 869        struct mountpoint *mp;
 870        mnt->mnt_parent = mnt;
 871        mnt->mnt_mountpoint = mnt->mnt.mnt_root;
 872        list_del_init(&mnt->mnt_child);
 873        hlist_del_init_rcu(&mnt->mnt_hash);
 874        hlist_del_init(&mnt->mnt_mp_list);
 875        mp = mnt->mnt_mp;
 876        mnt->mnt_mp = NULL;
 877        return mp;
 878}
 879
 880/*
 881 * vfsmount lock must be held for write
 882 */
 883static void umount_mnt(struct mount *mnt)
 884{
 885        put_mountpoint(unhash_mnt(mnt));
 886}
 887
 888/*
 889 * vfsmount lock must be held for write
 890 */
 891void mnt_set_mountpoint(struct mount *mnt,
 892                        struct mountpoint *mp,
 893                        struct mount *child_mnt)
 894{
 895        mp->m_count++;
 896        mnt_add_count(mnt, 1);  /* essentially, that's mntget */
 897        child_mnt->mnt_mountpoint = mp->m_dentry;
 898        child_mnt->mnt_parent = mnt;
 899        child_mnt->mnt_mp = mp;
 900        hlist_add_head(&child_mnt->mnt_mp_list, &mp->m_list);
 901}
 902
 903static void __attach_mnt(struct mount *mnt, struct mount *parent)
 904{
 905        hlist_add_head_rcu(&mnt->mnt_hash,
 906                           m_hash(&parent->mnt, mnt->mnt_mountpoint));
 907        list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
 908}
 909
 910/*
 911 * vfsmount lock must be held for write
 912 */
 913static void attach_mnt(struct mount *mnt,
 914                        struct mount *parent,
 915                        struct mountpoint *mp)
 916{
 917        mnt_set_mountpoint(parent, mp, mnt);
 918        __attach_mnt(mnt, parent);
 919}
 920
 921void mnt_change_mountpoint(struct mount *parent, struct mountpoint *mp, struct mount *mnt)
 922{
 923        struct mountpoint *old_mp = mnt->mnt_mp;
 924        struct mount *old_parent = mnt->mnt_parent;
 925
 926        list_del_init(&mnt->mnt_child);
 927        hlist_del_init(&mnt->mnt_mp_list);
 928        hlist_del_init_rcu(&mnt->mnt_hash);
 929
 930        attach_mnt(mnt, parent, mp);
 931
 932        put_mountpoint(old_mp);
 933        mnt_add_count(old_parent, -1);
 934}
 935
 936/*
 937 * vfsmount lock must be held for write
 938 */
 939static void commit_tree(struct mount *mnt)
 940{
 941        struct mount *parent = mnt->mnt_parent;
 942        struct mount *m;
 943        LIST_HEAD(head);
 944        struct mnt_namespace *n = parent->mnt_ns;
 945
 946        BUG_ON(parent == mnt);
 947
 948        list_add_tail(&head, &mnt->mnt_list);
 949        list_for_each_entry(m, &head, mnt_list)
 950                m->mnt_ns = n;
 951
 952        list_splice(&head, n->list.prev);
 953
 954        n->mounts += n->pending_mounts;
 955        n->pending_mounts = 0;
 956
 957        __attach_mnt(mnt, parent);
 958        touch_mnt_namespace(n);
 959}
 960
 961static struct mount *next_mnt(struct mount *p, struct mount *root)
 962{
 963        struct list_head *next = p->mnt_mounts.next;
 964        if (next == &p->mnt_mounts) {
 965                while (1) {
 966                        if (p == root)
 967                                return NULL;
 968                        next = p->mnt_child.next;
 969                        if (next != &p->mnt_parent->mnt_mounts)
 970                                break;
 971                        p = p->mnt_parent;
 972                }
 973        }
 974        return list_entry(next, struct mount, mnt_child);
 975}
 976
 977static struct mount *skip_mnt_tree(struct mount *p)
 978{
 979        struct list_head *prev = p->mnt_mounts.prev;
 980        while (prev != &p->mnt_mounts) {
 981                p = list_entry(prev, struct mount, mnt_child);
 982                prev = p->mnt_mounts.prev;
 983        }
 984        return p;
 985}
 986
 987/**
 988 * vfs_create_mount - Create a mount for a configured superblock
 989 * @fc: The configuration context with the superblock attached
 990 *
 991 * Create a mount to an already configured superblock.  If necessary, the
 992 * caller should invoke vfs_get_tree() before calling this.
 993 *
 994 * Note that this does not attach the mount to anything.
 995 */
 996struct vfsmount *vfs_create_mount(struct fs_context *fc)
 997{
 998        struct mount *mnt;
 999        struct user_namespace *fs_userns;
1000
1001        if (!fc->root)
1002                return ERR_PTR(-EINVAL);
1003
1004        mnt = alloc_vfsmnt(fc->source ?: "none");
1005        if (!mnt)
1006                return ERR_PTR(-ENOMEM);
1007
1008        if (fc->sb_flags & SB_KERNMOUNT)
1009                mnt->mnt.mnt_flags = MNT_INTERNAL;
1010
1011        atomic_inc(&fc->root->d_sb->s_active);
1012        mnt->mnt.mnt_sb         = fc->root->d_sb;
1013        mnt->mnt.mnt_root       = dget(fc->root);
1014        mnt->mnt_mountpoint     = mnt->mnt.mnt_root;
1015        mnt->mnt_parent         = mnt;
1016
1017        fs_userns = mnt->mnt.mnt_sb->s_user_ns;
1018        if (!initial_idmapping(fs_userns))
1019                mnt->mnt.mnt_userns = get_user_ns(fs_userns);
1020
1021        lock_mount_hash();
1022        list_add_tail(&mnt->mnt_instance, &mnt->mnt.mnt_sb->s_mounts);
1023        unlock_mount_hash();
1024        return &mnt->mnt;
1025}
1026EXPORT_SYMBOL(vfs_create_mount);
1027
1028struct vfsmount *fc_mount(struct fs_context *fc)
1029{
1030        int err = vfs_get_tree(fc);
1031        if (!err) {
1032                up_write(&fc->root->d_sb->s_umount);
1033                return vfs_create_mount(fc);
1034        }
1035        return ERR_PTR(err);
1036}
1037EXPORT_SYMBOL(fc_mount);
1038
1039struct vfsmount *vfs_kern_mount(struct file_system_type *type,
1040                                int flags, const char *name,
1041                                void *data)
1042{
1043        struct fs_context *fc;
1044        struct vfsmount *mnt;
1045        int ret = 0;
1046
1047        if (!type)
1048                return ERR_PTR(-EINVAL);
1049
1050        fc = fs_context_for_mount(type, flags);
1051        if (IS_ERR(fc))
1052                return ERR_CAST(fc);
1053
1054        if (name)
1055                ret = vfs_parse_fs_string(fc, "source",
1056                                          name, strlen(name));
1057        if (!ret)
1058                ret = parse_monolithic_mount_data(fc, data);
1059        if (!ret)
1060                mnt = fc_mount(fc);
1061        else
1062                mnt = ERR_PTR(ret);
1063
1064        put_fs_context(fc);
1065        return mnt;
1066}
1067EXPORT_SYMBOL_GPL(vfs_kern_mount);
1068
1069struct vfsmount *
1070vfs_submount(const struct dentry *mountpoint, struct file_system_type *type,
1071             const char *name, void *data)
1072{
1073        /* Until it is worked out how to pass the user namespace
1074         * through from the parent mount to the submount don't support
1075         * unprivileged mounts with submounts.
1076         */
1077        if (mountpoint->d_sb->s_user_ns != &init_user_ns)
1078                return ERR_PTR(-EPERM);
1079
1080        return vfs_kern_mount(type, SB_SUBMOUNT, name, data);
1081}
1082EXPORT_SYMBOL_GPL(vfs_submount);
1083
1084static struct mount *clone_mnt(struct mount *old, struct dentry *root,
1085                                        int flag)
1086{
1087        struct super_block *sb = old->mnt.mnt_sb;
1088        struct mount *mnt;
1089        int err;
1090
1091        mnt = alloc_vfsmnt(old->mnt_devname);
1092        if (!mnt)
1093                return ERR_PTR(-ENOMEM);
1094
1095        if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
1096                mnt->mnt_group_id = 0; /* not a peer of original */
1097        else
1098                mnt->mnt_group_id = old->mnt_group_id;
1099
1100        if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
1101                err = mnt_alloc_group_id(mnt);
1102                if (err)
1103                        goto out_free;
1104        }
1105
1106        mnt->mnt.mnt_flags = old->mnt.mnt_flags;
1107        mnt->mnt.mnt_flags &= ~(MNT_WRITE_HOLD|MNT_MARKED|MNT_INTERNAL);
1108
1109        atomic_inc(&sb->s_active);
1110        mnt->mnt.mnt_userns = mnt_user_ns(&old->mnt);
1111        if (!initial_idmapping(mnt->mnt.mnt_userns))
1112                mnt->mnt.mnt_userns = get_user_ns(mnt->mnt.mnt_userns);
1113        mnt->mnt.mnt_sb = sb;
1114        mnt->mnt.mnt_root = dget(root);
1115        mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1116        mnt->mnt_parent = mnt;
1117        lock_mount_hash();
1118        list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
1119        unlock_mount_hash();
1120
1121        if ((flag & CL_SLAVE) ||
1122            ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
1123                list_add(&mnt->mnt_slave, &old->mnt_slave_list);
1124                mnt->mnt_master = old;
1125                CLEAR_MNT_SHARED(mnt);
1126        } else if (!(flag & CL_PRIVATE)) {
1127                if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
1128                        list_add(&mnt->mnt_share, &old->mnt_share);
1129                if (IS_MNT_SLAVE(old))
1130                        list_add(&mnt->mnt_slave, &old->mnt_slave);
1131                mnt->mnt_master = old->mnt_master;
1132        } else {
1133                CLEAR_MNT_SHARED(mnt);
1134        }
1135        if (flag & CL_MAKE_SHARED)
1136                set_mnt_shared(mnt);
1137
1138        /* stick the duplicate mount on the same expiry list
1139         * as the original if that was on one */
1140        if (flag & CL_EXPIRE) {
1141                if (!list_empty(&old->mnt_expire))
1142                        list_add(&mnt->mnt_expire, &old->mnt_expire);
1143        }
1144
1145        return mnt;
1146
1147 out_free:
1148        mnt_free_id(mnt);
1149        free_vfsmnt(mnt);
1150        return ERR_PTR(err);
1151}
1152
1153static void cleanup_mnt(struct mount *mnt)
1154{
1155        struct hlist_node *p;
1156        struct mount *m;
1157        /*
1158         * The warning here probably indicates that somebody messed
1159         * up a mnt_want/drop_write() pair.  If this happens, the
1160         * filesystem was probably unable to make r/w->r/o transitions.
1161         * The locking used to deal with mnt_count decrement provides barriers,
1162         * so mnt_get_writers() below is safe.
1163         */
1164        WARN_ON(mnt_get_writers(mnt));
1165        if (unlikely(mnt->mnt_pins.first))
1166                mnt_pin_kill(mnt);
1167        hlist_for_each_entry_safe(m, p, &mnt->mnt_stuck_children, mnt_umount) {
1168                hlist_del(&m->mnt_umount);
1169                mntput(&m->mnt);
1170        }
1171        fsnotify_vfsmount_delete(&mnt->mnt);
1172        dput(mnt->mnt.mnt_root);
1173        deactivate_super(mnt->mnt.mnt_sb);
1174        mnt_free_id(mnt);
1175        call_rcu(&mnt->mnt_rcu, delayed_free_vfsmnt);
1176}
1177
1178static void __cleanup_mnt(struct rcu_head *head)
1179{
1180        cleanup_mnt(container_of(head, struct mount, mnt_rcu));
1181}
1182
1183static LLIST_HEAD(delayed_mntput_list);
1184static void delayed_mntput(struct work_struct *unused)
1185{
1186        struct llist_node *node = llist_del_all(&delayed_mntput_list);
1187        struct mount *m, *t;
1188
1189        llist_for_each_entry_safe(m, t, node, mnt_llist)
1190                cleanup_mnt(m);
1191}
1192static DECLARE_DELAYED_WORK(delayed_mntput_work, delayed_mntput);
1193
1194static void mntput_no_expire(struct mount *mnt)
1195{
1196        LIST_HEAD(list);
1197        int count;
1198
1199        rcu_read_lock();
1200        if (likely(READ_ONCE(mnt->mnt_ns))) {
1201                /*
1202                 * Since we don't do lock_mount_hash() here,
1203                 * ->mnt_ns can change under us.  However, if it's
1204                 * non-NULL, then there's a reference that won't
1205                 * be dropped until after an RCU delay done after
1206                 * turning ->mnt_ns NULL.  So if we observe it
1207                 * non-NULL under rcu_read_lock(), the reference
1208                 * we are dropping is not the final one.
1209                 */
1210                mnt_add_count(mnt, -1);
1211                rcu_read_unlock();
1212                return;
1213        }
1214        lock_mount_hash();
1215        /*
1216         * make sure that if __legitimize_mnt() has not seen us grab
1217         * mount_lock, we'll see their refcount increment here.
1218         */
1219        smp_mb();
1220        mnt_add_count(mnt, -1);
1221        count = mnt_get_count(mnt);
1222        if (count != 0) {
1223                WARN_ON(count < 0);
1224                rcu_read_unlock();
1225                unlock_mount_hash();
1226                return;
1227        }
1228        if (unlikely(mnt->mnt.mnt_flags & MNT_DOOMED)) {
1229                rcu_read_unlock();
1230                unlock_mount_hash();
1231                return;
1232        }
1233        mnt->mnt.mnt_flags |= MNT_DOOMED;
1234        rcu_read_unlock();
1235
1236        list_del(&mnt->mnt_instance);
1237
1238        if (unlikely(!list_empty(&mnt->mnt_mounts))) {
1239                struct mount *p, *tmp;
1240                list_for_each_entry_safe(p, tmp, &mnt->mnt_mounts,  mnt_child) {
1241                        __put_mountpoint(unhash_mnt(p), &list);
1242                        hlist_add_head(&p->mnt_umount, &mnt->mnt_stuck_children);
1243                }
1244        }
1245        unlock_mount_hash();
1246        shrink_dentry_list(&list);
1247
1248        if (likely(!(mnt->mnt.mnt_flags & MNT_INTERNAL))) {
1249                struct task_struct *task = current;
1250                if (likely(!(task->flags & PF_KTHREAD))) {
1251                        init_task_work(&mnt->mnt_rcu, __cleanup_mnt);
1252                        if (!task_work_add(task, &mnt->mnt_rcu, TWA_RESUME))
1253                                return;
1254                }
1255                if (llist_add(&mnt->mnt_llist, &delayed_mntput_list))
1256                        schedule_delayed_work(&delayed_mntput_work, 1);
1257                return;
1258        }
1259        cleanup_mnt(mnt);
1260}
1261
1262void mntput(struct vfsmount *mnt)
1263{
1264        if (mnt) {
1265                struct mount *m = real_mount(mnt);
1266                /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
1267                if (unlikely(m->mnt_expiry_mark))
1268                        m->mnt_expiry_mark = 0;
1269                mntput_no_expire(m);
1270        }
1271}
1272EXPORT_SYMBOL(mntput);
1273
1274struct vfsmount *mntget(struct vfsmount *mnt)
1275{
1276        if (mnt)
1277                mnt_add_count(real_mount(mnt), 1);
1278        return mnt;
1279}
1280EXPORT_SYMBOL(mntget);
1281
1282/**
1283 * path_is_mountpoint() - Check if path is a mount in the current namespace.
1284 * @path: path to check
1285 *
1286 *  d_mountpoint() can only be used reliably to establish if a dentry is
1287 *  not mounted in any namespace and that common case is handled inline.
1288 *  d_mountpoint() isn't aware of the possibility there may be multiple
1289 *  mounts using a given dentry in a different namespace. This function
1290 *  checks if the passed in path is a mountpoint rather than the dentry
1291 *  alone.
1292 */
1293bool path_is_mountpoint(const struct path *path)
1294{
1295        unsigned seq;
1296        bool res;
1297
1298        if (!d_mountpoint(path->dentry))
1299                return false;
1300
1301        rcu_read_lock();
1302        do {
1303                seq = read_seqbegin(&mount_lock);
1304                res = __path_is_mountpoint(path);
1305        } while (read_seqretry(&mount_lock, seq));
1306        rcu_read_unlock();
1307
1308        return res;
1309}
1310EXPORT_SYMBOL(path_is_mountpoint);
1311
1312struct vfsmount *mnt_clone_internal(const struct path *path)
1313{
1314        struct mount *p;
1315        p = clone_mnt(real_mount(path->mnt), path->dentry, CL_PRIVATE);
1316        if (IS_ERR(p))
1317                return ERR_CAST(p);
1318        p->mnt.mnt_flags |= MNT_INTERNAL;
1319        return &p->mnt;
1320}
1321
1322#ifdef CONFIG_PROC_FS
1323static struct mount *mnt_list_next(struct mnt_namespace *ns,
1324                                   struct list_head *p)
1325{
1326        struct mount *mnt, *ret = NULL;
1327
1328        lock_ns_list(ns);
1329        list_for_each_continue(p, &ns->list) {
1330                mnt = list_entry(p, typeof(*mnt), mnt_list);
1331                if (!mnt_is_cursor(mnt)) {
1332                        ret = mnt;
1333                        break;
1334                }
1335        }
1336        unlock_ns_list(ns);
1337
1338        return ret;
1339}
1340
1341/* iterator; we want it to have access to namespace_sem, thus here... */
1342static void *m_start(struct seq_file *m, loff_t *pos)
1343{
1344        struct proc_mounts *p = m->private;
1345        struct list_head *prev;
1346
1347        down_read(&namespace_sem);
1348        if (!*pos) {
1349                prev = &p->ns->list;
1350        } else {
1351                prev = &p->cursor.mnt_list;
1352
1353                /* Read after we'd reached the end? */
1354                if (list_empty(prev))
1355                        return NULL;
1356        }
1357
1358        return mnt_list_next(p->ns, prev);
1359}
1360
1361static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1362{
1363        struct proc_mounts *p = m->private;
1364        struct mount *mnt = v;
1365
1366        ++*pos;
1367        return mnt_list_next(p->ns, &mnt->mnt_list);
1368}
1369
1370static void m_stop(struct seq_file *m, void *v)
1371{
1372        struct proc_mounts *p = m->private;
1373        struct mount *mnt = v;
1374
1375        lock_ns_list(p->ns);
1376        if (mnt)
1377                list_move_tail(&p->cursor.mnt_list, &mnt->mnt_list);
1378        else
1379                list_del_init(&p->cursor.mnt_list);
1380        unlock_ns_list(p->ns);
1381        up_read(&namespace_sem);
1382}
1383
1384static int m_show(struct seq_file *m, void *v)
1385{
1386        struct proc_mounts *p = m->private;
1387        struct mount *r = v;
1388        return p->show(m, &r->mnt);
1389}
1390
1391const struct seq_operations mounts_op = {
1392        .start  = m_start,
1393        .next   = m_next,
1394        .stop   = m_stop,
1395        .show   = m_show,
1396};
1397
1398void mnt_cursor_del(struct mnt_namespace *ns, struct mount *cursor)
1399{
1400        down_read(&namespace_sem);
1401        lock_ns_list(ns);
1402        list_del(&cursor->mnt_list);
1403        unlock_ns_list(ns);
1404        up_read(&namespace_sem);
1405}
1406#endif  /* CONFIG_PROC_FS */
1407
1408/**
1409 * may_umount_tree - check if a mount tree is busy
1410 * @m: root of mount tree
1411 *
1412 * This is called to check if a tree of mounts has any
1413 * open files, pwds, chroots or sub mounts that are
1414 * busy.
1415 */
1416int may_umount_tree(struct vfsmount *m)
1417{
1418        struct mount *mnt = real_mount(m);
1419        int actual_refs = 0;
1420        int minimum_refs = 0;
1421        struct mount *p;
1422        BUG_ON(!m);
1423
1424        /* write lock needed for mnt_get_count */
1425        lock_mount_hash();
1426        for (p = mnt; p; p = next_mnt(p, mnt)) {
1427                actual_refs += mnt_get_count(p);
1428                minimum_refs += 2;
1429        }
1430        unlock_mount_hash();
1431
1432        if (actual_refs > minimum_refs)
1433                return 0;
1434
1435        return 1;
1436}
1437
1438EXPORT_SYMBOL(may_umount_tree);
1439
1440/**
1441 * may_umount - check if a mount point is busy
1442 * @mnt: root of mount
1443 *
1444 * This is called to check if a mount point has any
1445 * open files, pwds, chroots or sub mounts. If the
1446 * mount has sub mounts this will return busy
1447 * regardless of whether the sub mounts are busy.
1448 *
1449 * Doesn't take quota and stuff into account. IOW, in some cases it will
1450 * give false negatives. The main reason why it's here is that we need
1451 * a non-destructive way to look for easily umountable filesystems.
1452 */
1453int may_umount(struct vfsmount *mnt)
1454{
1455        int ret = 1;
1456        down_read(&namespace_sem);
1457        lock_mount_hash();
1458        if (propagate_mount_busy(real_mount(mnt), 2))
1459                ret = 0;
1460        unlock_mount_hash();
1461        up_read(&namespace_sem);
1462        return ret;
1463}
1464
1465EXPORT_SYMBOL(may_umount);
1466
1467static void namespace_unlock(void)
1468{
1469        struct hlist_head head;
1470        struct hlist_node *p;
1471        struct mount *m;
1472        LIST_HEAD(list);
1473
1474        hlist_move_list(&unmounted, &head);
1475        list_splice_init(&ex_mountpoints, &list);
1476
1477        up_write(&namespace_sem);
1478
1479        shrink_dentry_list(&list);
1480
1481        if (likely(hlist_empty(&head)))
1482                return;
1483
1484        synchronize_rcu_expedited();
1485
1486        hlist_for_each_entry_safe(m, p, &head, mnt_umount) {
1487                hlist_del(&m->mnt_umount);
1488                mntput(&m->mnt);
1489        }
1490}
1491
1492static inline void namespace_lock(void)
1493{
1494        down_write(&namespace_sem);
1495}
1496
1497enum umount_tree_flags {
1498        UMOUNT_SYNC = 1,
1499        UMOUNT_PROPAGATE = 2,
1500        UMOUNT_CONNECTED = 4,
1501};
1502
1503static bool disconnect_mount(struct mount *mnt, enum umount_tree_flags how)
1504{
1505        /* Leaving mounts connected is only valid for lazy umounts */
1506        if (how & UMOUNT_SYNC)
1507                return true;
1508
1509        /* A mount without a parent has nothing to be connected to */
1510        if (!mnt_has_parent(mnt))
1511                return true;
1512
1513        /* Because the reference counting rules change when mounts are
1514         * unmounted and connected, umounted mounts may not be
1515         * connected to mounted mounts.
1516         */
1517        if (!(mnt->mnt_parent->mnt.mnt_flags & MNT_UMOUNT))
1518                return true;
1519
1520        /* Has it been requested that the mount remain connected? */
1521        if (how & UMOUNT_CONNECTED)
1522                return false;
1523
1524        /* Is the mount locked such that it needs to remain connected? */
1525        if (IS_MNT_LOCKED(mnt))
1526                return false;
1527
1528        /* By default disconnect the mount */
1529        return true;
1530}
1531
1532/*
1533 * mount_lock must be held
1534 * namespace_sem must be held for write
1535 */
1536static void umount_tree(struct mount *mnt, enum umount_tree_flags how)
1537{
1538        LIST_HEAD(tmp_list);
1539        struct mount *p;
1540
1541        if (how & UMOUNT_PROPAGATE)
1542                propagate_mount_unlock(mnt);
1543
1544        /* Gather the mounts to umount */
1545        for (p = mnt; p; p = next_mnt(p, mnt)) {
1546                p->mnt.mnt_flags |= MNT_UMOUNT;
1547                list_move(&p->mnt_list, &tmp_list);
1548        }
1549
1550        /* Hide the mounts from mnt_mounts */
1551        list_for_each_entry(p, &tmp_list, mnt_list) {
1552                list_del_init(&p->mnt_child);
1553        }
1554
1555        /* Add propogated mounts to the tmp_list */
1556        if (how & UMOUNT_PROPAGATE)
1557                propagate_umount(&tmp_list);
1558
1559        while (!list_empty(&tmp_list)) {
1560                struct mnt_namespace *ns;
1561                bool disconnect;
1562                p = list_first_entry(&tmp_list, struct mount, mnt_list);
1563                list_del_init(&p->mnt_expire);
1564                list_del_init(&p->mnt_list);
1565                ns = p->mnt_ns;
1566                if (ns) {
1567                        ns->mounts--;
1568                        __touch_mnt_namespace(ns);
1569                }
1570                p->mnt_ns = NULL;
1571                if (how & UMOUNT_SYNC)
1572                        p->mnt.mnt_flags |= MNT_SYNC_UMOUNT;
1573
1574                disconnect = disconnect_mount(p, how);
1575                if (mnt_has_parent(p)) {
1576                        mnt_add_count(p->mnt_parent, -1);
1577                        if (!disconnect) {
1578                                /* Don't forget about p */
1579                                list_add_tail(&p->mnt_child, &p->mnt_parent->mnt_mounts);
1580                        } else {
1581                                umount_mnt(p);
1582                        }
1583                }
1584                change_mnt_propagation(p, MS_PRIVATE);
1585                if (disconnect)
1586                        hlist_add_head(&p->mnt_umount, &unmounted);
1587        }
1588}
1589
1590static void shrink_submounts(struct mount *mnt);
1591
1592static int do_umount_root(struct super_block *sb)
1593{
1594        int ret = 0;
1595
1596        down_write(&sb->s_umount);
1597        if (!sb_rdonly(sb)) {
1598                struct fs_context *fc;
1599
1600                fc = fs_context_for_reconfigure(sb->s_root, SB_RDONLY,
1601                                                SB_RDONLY);
1602                if (IS_ERR(fc)) {
1603                        ret = PTR_ERR(fc);
1604                } else {
1605                        ret = parse_monolithic_mount_data(fc, NULL);
1606                        if (!ret)
1607                                ret = reconfigure_super(fc);
1608                        put_fs_context(fc);
1609                }
1610        }
1611        up_write(&sb->s_umount);
1612        return ret;
1613}
1614
1615static int do_umount(struct mount *mnt, int flags)
1616{
1617        struct super_block *sb = mnt->mnt.mnt_sb;
1618        int retval;
1619
1620        retval = security_sb_umount(&mnt->mnt, flags);
1621        if (retval)
1622                return retval;
1623
1624        /*
1625         * Allow userspace to request a mountpoint be expired rather than
1626         * unmounting unconditionally. Unmount only happens if:
1627         *  (1) the mark is already set (the mark is cleared by mntput())
1628         *  (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1629         */
1630        if (flags & MNT_EXPIRE) {
1631                if (&mnt->mnt == current->fs->root.mnt ||
1632                    flags & (MNT_FORCE | MNT_DETACH))
1633                        return -EINVAL;
1634
1635                /*
1636                 * probably don't strictly need the lock here if we examined
1637                 * all race cases, but it's a slowpath.
1638                 */
1639                lock_mount_hash();
1640                if (mnt_get_count(mnt) != 2) {
1641                        unlock_mount_hash();
1642                        return -EBUSY;
1643                }
1644                unlock_mount_hash();
1645
1646                if (!xchg(&mnt->mnt_expiry_mark, 1))
1647                        return -EAGAIN;
1648        }
1649
1650        /*
1651         * If we may have to abort operations to get out of this
1652         * mount, and they will themselves hold resources we must
1653         * allow the fs to do things. In the Unix tradition of
1654         * 'Gee thats tricky lets do it in userspace' the umount_begin
1655         * might fail to complete on the first run through as other tasks
1656         * must return, and the like. Thats for the mount program to worry
1657         * about for the moment.
1658         */
1659
1660        if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1661                sb->s_op->umount_begin(sb);
1662        }
1663
1664        /*
1665         * No sense to grab the lock for this test, but test itself looks
1666         * somewhat bogus. Suggestions for better replacement?
1667         * Ho-hum... In principle, we might treat that as umount + switch
1668         * to rootfs. GC would eventually take care of the old vfsmount.
1669         * Actually it makes sense, especially if rootfs would contain a
1670         * /reboot - static binary that would close all descriptors and
1671         * call reboot(9). Then init(8) could umount root and exec /reboot.
1672         */
1673        if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1674                /*
1675                 * Special case for "unmounting" root ...
1676                 * we just try to remount it readonly.
1677                 */
1678                if (!ns_capable(sb->s_user_ns, CAP_SYS_ADMIN))
1679                        return -EPERM;
1680                return do_umount_root(sb);
1681        }
1682
1683        namespace_lock();
1684        lock_mount_hash();
1685
1686        /* Recheck MNT_LOCKED with the locks held */
1687        retval = -EINVAL;
1688        if (mnt->mnt.mnt_flags & MNT_LOCKED)
1689                goto out;
1690
1691        event++;
1692        if (flags & MNT_DETACH) {
1693                if (!list_empty(&mnt->mnt_list))
1694                        umount_tree(mnt, UMOUNT_PROPAGATE);
1695                retval = 0;
1696        } else {
1697                shrink_submounts(mnt);
1698                retval = -EBUSY;
1699                if (!propagate_mount_busy(mnt, 2)) {
1700                        if (!list_empty(&mnt->mnt_list))
1701                                umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
1702                        retval = 0;
1703                }
1704        }
1705out:
1706        unlock_mount_hash();
1707        namespace_unlock();
1708        return retval;
1709}
1710
1711/*
1712 * __detach_mounts - lazily unmount all mounts on the specified dentry
1713 *
1714 * During unlink, rmdir, and d_drop it is possible to loose the path
1715 * to an existing mountpoint, and wind up leaking the mount.
1716 * detach_mounts allows lazily unmounting those mounts instead of
1717 * leaking them.
1718 *
1719 * The caller may hold dentry->d_inode->i_mutex.
1720 */
1721void __detach_mounts(struct dentry *dentry)
1722{
1723        struct mountpoint *mp;
1724        struct mount *mnt;
1725
1726        namespace_lock();
1727        lock_mount_hash();
1728        mp = lookup_mountpoint(dentry);
1729        if (!mp)
1730                goto out_unlock;
1731
1732        event++;
1733        while (!hlist_empty(&mp->m_list)) {
1734                mnt = hlist_entry(mp->m_list.first, struct mount, mnt_mp_list);
1735                if (mnt->mnt.mnt_flags & MNT_UMOUNT) {
1736                        umount_mnt(mnt);
1737                        hlist_add_head(&mnt->mnt_umount, &unmounted);
1738                }
1739                else umount_tree(mnt, UMOUNT_CONNECTED);
1740        }
1741        put_mountpoint(mp);
1742out_unlock:
1743        unlock_mount_hash();
1744        namespace_unlock();
1745}
1746
1747/*
1748 * Is the caller allowed to modify his namespace?
1749 */
1750static inline bool may_mount(void)
1751{
1752        return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1753}
1754
1755static void warn_mandlock(void)
1756{
1757        pr_warn_once("=======================================================\n"
1758                     "WARNING: The mand mount option has been deprecated and\n"
1759                     "         and is ignored by this kernel. Remove the mand\n"
1760                     "         option from the mount to silence this warning.\n"
1761                     "=======================================================\n");
1762}
1763
1764static int can_umount(const struct path *path, int flags)
1765{
1766        struct mount *mnt = real_mount(path->mnt);
1767
1768        if (!may_mount())
1769                return -EPERM;
1770        if (path->dentry != path->mnt->mnt_root)
1771                return -EINVAL;
1772        if (!check_mnt(mnt))
1773                return -EINVAL;
1774        if (mnt->mnt.mnt_flags & MNT_LOCKED) /* Check optimistically */
1775                return -EINVAL;
1776        if (flags & MNT_FORCE && !capable(CAP_SYS_ADMIN))
1777                return -EPERM;
1778        return 0;
1779}
1780
1781// caller is responsible for flags being sane
1782int path_umount(struct path *path, int flags)
1783{
1784        struct mount *mnt = real_mount(path->mnt);
1785        int ret;
1786
1787        ret = can_umount(path, flags);
1788        if (!ret)
1789                ret = do_umount(mnt, flags);
1790
1791        /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1792        dput(path->dentry);
1793        mntput_no_expire(mnt);
1794        return ret;
1795}
1796
1797static int ksys_umount(char __user *name, int flags)
1798{
1799        int lookup_flags = LOOKUP_MOUNTPOINT;
1800        struct path path;
1801        int ret;
1802
1803        // basic validity checks done first
1804        if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1805                return -EINVAL;
1806
1807        if (!(flags & UMOUNT_NOFOLLOW))
1808                lookup_flags |= LOOKUP_FOLLOW;
1809        ret = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1810        if (ret)
1811                return ret;
1812        return path_umount(&path, flags);
1813}
1814
1815SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1816{
1817        return ksys_umount(name, flags);
1818}
1819
1820#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1821
1822/*
1823 *      The 2.0 compatible umount. No flags.
1824 */
1825SYSCALL_DEFINE1(oldumount, char __user *, name)
1826{
1827        return ksys_umount(name, 0);
1828}
1829
1830#endif
1831
1832static bool is_mnt_ns_file(struct dentry *dentry)
1833{
1834        /* Is this a proxy for a mount namespace? */
1835        return dentry->d_op == &ns_dentry_operations &&
1836               dentry->d_fsdata == &mntns_operations;
1837}
1838
1839static struct mnt_namespace *to_mnt_ns(struct ns_common *ns)
1840{
1841        return container_of(ns, struct mnt_namespace, ns);
1842}
1843
1844struct ns_common *from_mnt_ns(struct mnt_namespace *mnt)
1845{
1846        return &mnt->ns;
1847}
1848
1849static bool mnt_ns_loop(struct dentry *dentry)
1850{
1851        /* Could bind mounting the mount namespace inode cause a
1852         * mount namespace loop?
1853         */
1854        struct mnt_namespace *mnt_ns;
1855        if (!is_mnt_ns_file(dentry))
1856                return false;
1857
1858        mnt_ns = to_mnt_ns(get_proc_ns(dentry->d_inode));
1859        return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1860}
1861
1862struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1863                                        int flag)
1864{
1865        struct mount *res, *p, *q, *r, *parent;
1866
1867        if (!(flag & CL_COPY_UNBINDABLE) && IS_MNT_UNBINDABLE(mnt))
1868                return ERR_PTR(-EINVAL);
1869
1870        if (!(flag & CL_COPY_MNT_NS_FILE) && is_mnt_ns_file(dentry))
1871                return ERR_PTR(-EINVAL);
1872
1873        res = q = clone_mnt(mnt, dentry, flag);
1874        if (IS_ERR(q))
1875                return q;
1876
1877        q->mnt_mountpoint = mnt->mnt_mountpoint;
1878
1879        p = mnt;
1880        list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1881                struct mount *s;
1882                if (!is_subdir(r->mnt_mountpoint, dentry))
1883                        continue;
1884
1885                for (s = r; s; s = next_mnt(s, r)) {
1886                        if (!(flag & CL_COPY_UNBINDABLE) &&
1887                            IS_MNT_UNBINDABLE(s)) {
1888                                if (s->mnt.mnt_flags & MNT_LOCKED) {
1889                                        /* Both unbindable and locked. */
1890                                        q = ERR_PTR(-EPERM);
1891                                        goto out;
1892                                } else {
1893                                        s = skip_mnt_tree(s);
1894                                        continue;
1895                                }
1896                        }
1897                        if (!(flag & CL_COPY_MNT_NS_FILE) &&
1898                            is_mnt_ns_file(s->mnt.mnt_root)) {
1899                                s = skip_mnt_tree(s);
1900                                continue;
1901                        }
1902                        while (p != s->mnt_parent) {
1903                                p = p->mnt_parent;
1904                                q = q->mnt_parent;
1905                        }
1906                        p = s;
1907                        parent = q;
1908                        q = clone_mnt(p, p->mnt.mnt_root, flag);
1909                        if (IS_ERR(q))
1910                                goto out;
1911                        lock_mount_hash();
1912                        list_add_tail(&q->mnt_list, &res->mnt_list);
1913                        attach_mnt(q, parent, p->mnt_mp);
1914                        unlock_mount_hash();
1915                }
1916        }
1917        return res;
1918out:
1919        if (res) {
1920                lock_mount_hash();
1921                umount_tree(res, UMOUNT_SYNC);
1922                unlock_mount_hash();
1923        }
1924        return q;
1925}
1926
1927/* Caller should check returned pointer for errors */
1928
1929struct vfsmount *collect_mounts(const struct path *path)
1930{
1931        struct mount *tree;
1932        namespace_lock();
1933        if (!check_mnt(real_mount(path->mnt)))
1934                tree = ERR_PTR(-EINVAL);
1935        else
1936                tree = copy_tree(real_mount(path->mnt), path->dentry,
1937                                 CL_COPY_ALL | CL_PRIVATE);
1938        namespace_unlock();
1939        if (IS_ERR(tree))
1940                return ERR_CAST(tree);
1941        return &tree->mnt;
1942}
1943
1944static void free_mnt_ns(struct mnt_namespace *);
1945static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *, bool);
1946
1947void dissolve_on_fput(struct vfsmount *mnt)
1948{
1949        struct mnt_namespace *ns;
1950        namespace_lock();
1951        lock_mount_hash();
1952        ns = real_mount(mnt)->mnt_ns;
1953        if (ns) {
1954                if (is_anon_ns(ns))
1955                        umount_tree(real_mount(mnt), UMOUNT_CONNECTED);
1956                else
1957                        ns = NULL;
1958        }
1959        unlock_mount_hash();
1960        namespace_unlock();
1961        if (ns)
1962                free_mnt_ns(ns);
1963}
1964
1965void drop_collected_mounts(struct vfsmount *mnt)
1966{
1967        namespace_lock();
1968        lock_mount_hash();
1969        umount_tree(real_mount(mnt), 0);
1970        unlock_mount_hash();
1971        namespace_unlock();
1972}
1973
1974static bool has_locked_children(struct mount *mnt, struct dentry *dentry)
1975{
1976        struct mount *child;
1977
1978        list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
1979                if (!is_subdir(child->mnt_mountpoint, dentry))
1980                        continue;
1981
1982                if (child->mnt.mnt_flags & MNT_LOCKED)
1983                        return true;
1984        }
1985        return false;
1986}
1987
1988/**
1989 * clone_private_mount - create a private clone of a path
1990 * @path: path to clone
1991 *
1992 * This creates a new vfsmount, which will be the clone of @path.  The new mount
1993 * will not be attached anywhere in the namespace and will be private (i.e.
1994 * changes to the originating mount won't be propagated into this).
1995 *
1996 * Release with mntput().
1997 */
1998struct vfsmount *clone_private_mount(const struct path *path)
1999{
2000        struct mount *old_mnt = real_mount(path->mnt);
2001        struct mount *new_mnt;
2002
2003        down_read(&namespace_sem);
2004        if (IS_MNT_UNBINDABLE(old_mnt))
2005                goto invalid;
2006
2007        if (!check_mnt(old_mnt))
2008                goto invalid;
2009
2010        if (has_locked_children(old_mnt, path->dentry))
2011                goto invalid;
2012
2013        new_mnt = clone_mnt(old_mnt, path->dentry, CL_PRIVATE);
2014        up_read(&namespace_sem);
2015
2016        if (IS_ERR(new_mnt))
2017                return ERR_CAST(new_mnt);
2018
2019        /* Longterm mount to be removed by kern_unmount*() */
2020        new_mnt->mnt_ns = MNT_NS_INTERNAL;
2021
2022        return &new_mnt->mnt;
2023
2024invalid:
2025        up_read(&namespace_sem);
2026        return ERR_PTR(-EINVAL);
2027}
2028EXPORT_SYMBOL_GPL(clone_private_mount);
2029
2030int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
2031                   struct vfsmount *root)
2032{
2033        struct mount *mnt;
2034        int res = f(root, arg);
2035        if (res)
2036                return res;
2037        list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
2038                res = f(&mnt->mnt, arg);
2039                if (res)
2040                        return res;
2041        }
2042        return 0;
2043}
2044
2045static void lock_mnt_tree(struct mount *mnt)
2046{
2047        struct mount *p;
2048
2049        for (p = mnt; p; p = next_mnt(p, mnt)) {
2050                int flags = p->mnt.mnt_flags;
2051                /* Don't allow unprivileged users to change mount flags */
2052                flags |= MNT_LOCK_ATIME;
2053
2054                if (flags & MNT_READONLY)
2055                        flags |= MNT_LOCK_READONLY;
2056
2057                if (flags & MNT_NODEV)
2058                        flags |= MNT_LOCK_NODEV;
2059
2060                if (flags & MNT_NOSUID)
2061                        flags |= MNT_LOCK_NOSUID;
2062
2063                if (flags & MNT_NOEXEC)
2064                        flags |= MNT_LOCK_NOEXEC;
2065                /* Don't allow unprivileged users to reveal what is under a mount */
2066                if (list_empty(&p->mnt_expire))
2067                        flags |= MNT_LOCKED;
2068                p->mnt.mnt_flags = flags;
2069        }
2070}
2071
2072static void cleanup_group_ids(struct mount *mnt, struct mount *end)
2073{
2074        struct mount *p;
2075
2076        for (p = mnt; p != end; p = next_mnt(p, mnt)) {
2077                if (p->mnt_group_id && !IS_MNT_SHARED(p))
2078                        mnt_release_group_id(p);
2079        }
2080}
2081
2082static int invent_group_ids(struct mount *mnt, bool recurse)
2083{
2084        struct mount *p;
2085
2086        for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
2087                if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
2088                        int err = mnt_alloc_group_id(p);
2089                        if (err) {
2090                                cleanup_group_ids(mnt, p);
2091                                return err;
2092                        }
2093                }
2094        }
2095
2096        return 0;
2097}
2098
2099int count_mounts(struct mnt_namespace *ns, struct mount *mnt)
2100{
2101        unsigned int max = READ_ONCE(sysctl_mount_max);
2102        unsigned int mounts = 0, old, pending, sum;
2103        struct mount *p;
2104
2105        for (p = mnt; p; p = next_mnt(p, mnt))
2106                mounts++;
2107
2108        old = ns->mounts;
2109        pending = ns->pending_mounts;
2110        sum = old + pending;
2111        if ((old > sum) ||
2112            (pending > sum) ||
2113            (max < sum) ||
2114            (mounts > (max - sum)))
2115                return -ENOSPC;
2116
2117        ns->pending_mounts = pending + mounts;
2118        return 0;
2119}
2120
2121/*
2122 *  @source_mnt : mount tree to be attached
2123 *  @nd         : place the mount tree @source_mnt is attached
2124 *  @parent_nd  : if non-null, detach the source_mnt from its parent and
2125 *                 store the parent mount and mountpoint dentry.
2126 *                 (done when source_mnt is moved)
2127 *
2128 *  NOTE: in the table below explains the semantics when a source mount
2129 *  of a given type is attached to a destination mount of a given type.
2130 * ---------------------------------------------------------------------------
2131 * |         BIND MOUNT OPERATION                                            |
2132 * |**************************************************************************
2133 * | source-->| shared        |       private  |       slave    | unbindable |
2134 * | dest     |               |                |                |            |
2135 * |   |      |               |                |                |            |
2136 * |   v      |               |                |                |            |
2137 * |**************************************************************************
2138 * |  shared  | shared (++)   |     shared (+) |     shared(+++)|  invalid   |
2139 * |          |               |                |                |            |
2140 * |non-shared| shared (+)    |      private   |      slave (*) |  invalid   |
2141 * ***************************************************************************
2142 * A bind operation clones the source mount and mounts the clone on the
2143 * destination mount.
2144 *
2145 * (++)  the cloned mount is propagated to all the mounts in the propagation
2146 *       tree of the destination mount and the cloned mount is added to
2147 *       the peer group of the source mount.
2148 * (+)   the cloned mount is created under the destination mount and is marked
2149 *       as shared. The cloned mount is added to the peer group of the source
2150 *       mount.
2151 * (+++) the mount is propagated to all the mounts in the propagation tree
2152 *       of the destination mount and the cloned mount is made slave
2153 *       of the same master as that of the source mount. The cloned mount
2154 *       is marked as 'shared and slave'.
2155 * (*)   the cloned mount is made a slave of the same master as that of the
2156 *       source mount.
2157 *
2158 * ---------------------------------------------------------------------------
2159 * |                    MOVE MOUNT OPERATION                                 |
2160 * |**************************************************************************
2161 * | source-->| shared        |       private  |       slave    | unbindable |
2162 * | dest     |               |                |                |            |
2163 * |   |      |               |                |                |            |
2164 * |   v      |               |                |                |            |
2165 * |**************************************************************************
2166 * |  shared  | shared (+)    |     shared (+) |    shared(+++) |  invalid   |
2167 * |          |               |                |                |            |
2168 * |non-shared| shared (+*)   |      private   |    slave (*)   | unbindable |
2169 * ***************************************************************************
2170 *
2171 * (+)  the mount is moved to the destination. And is then propagated to
2172 *      all the mounts in the propagation tree of the destination mount.
2173 * (+*)  the mount is moved to the destination.
2174 * (+++)  the mount is moved to the destination and is then propagated to
2175 *      all the mounts belonging to the destination mount's propagation tree.
2176 *      the mount is marked as 'shared and slave'.
2177 * (*)  the mount continues to be a slave at the new location.
2178 *
2179 * if the source mount is a tree, the operations explained above is
2180 * applied to each mount in the tree.
2181 * Must be called without spinlocks held, since this function can sleep
2182 * in allocations.
2183 */
2184static int attach_recursive_mnt(struct mount *source_mnt,
2185                        struct mount *dest_mnt,
2186                        struct mountpoint *dest_mp,
2187                        bool moving)
2188{
2189        struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2190        HLIST_HEAD(tree_list);
2191        struct mnt_namespace *ns = dest_mnt->mnt_ns;
2192        struct mountpoint *smp;
2193        struct mount *child, *p;
2194        struct hlist_node *n;
2195        int err;
2196
2197        /* Preallocate a mountpoint in case the new mounts need
2198         * to be tucked under other mounts.
2199         */
2200        smp = get_mountpoint(source_mnt->mnt.mnt_root);
2201        if (IS_ERR(smp))
2202                return PTR_ERR(smp);
2203
2204        /* Is there space to add these mounts to the mount namespace? */
2205        if (!moving) {
2206                err = count_mounts(ns, source_mnt);
2207                if (err)
2208                        goto out;
2209        }
2210
2211        if (IS_MNT_SHARED(dest_mnt)) {
2212                err = invent_group_ids(source_mnt, true);
2213                if (err)
2214                        goto out;
2215                err = propagate_mnt(dest_mnt, dest_mp, source_mnt, &tree_list);
2216                lock_mount_hash();
2217                if (err)
2218                        goto out_cleanup_ids;
2219                for (p = source_mnt; p; p = next_mnt(p, source_mnt))
2220                        set_mnt_shared(p);
2221        } else {
2222                lock_mount_hash();
2223        }
2224        if (moving) {
2225                unhash_mnt(source_mnt);
2226                attach_mnt(source_mnt, dest_mnt, dest_mp);
2227                touch_mnt_namespace(source_mnt->mnt_ns);
2228        } else {
2229                if (source_mnt->mnt_ns) {
2230                        /* move from anon - the caller will destroy */
2231                        list_del_init(&source_mnt->mnt_ns->list);
2232                }
2233                mnt_set_mountpoint(dest_mnt, dest_mp, source_mnt);
2234                commit_tree(source_mnt);
2235        }
2236
2237        hlist_for_each_entry_safe(child, n, &tree_list, mnt_hash) {
2238                struct mount *q;
2239                hlist_del_init(&child->mnt_hash);
2240                q = __lookup_mnt(&child->mnt_parent->mnt,
2241                                 child->mnt_mountpoint);
2242                if (q)
2243                        mnt_change_mountpoint(child, smp, q);
2244                /* Notice when we are propagating across user namespaces */
2245                if (child->mnt_parent->mnt_ns->user_ns != user_ns)
2246                        lock_mnt_tree(child);
2247                child->mnt.mnt_flags &= ~MNT_LOCKED;
2248                commit_tree(child);
2249        }
2250        put_mountpoint(smp);
2251        unlock_mount_hash();
2252
2253        return 0;
2254
2255 out_cleanup_ids:
2256        while (!hlist_empty(&tree_list)) {
2257                child = hlist_entry(tree_list.first, struct mount, mnt_hash);
2258                child->mnt_parent->mnt_ns->pending_mounts = 0;
2259                umount_tree(child, UMOUNT_SYNC);
2260        }
2261        unlock_mount_hash();
2262        cleanup_group_ids(source_mnt, NULL);
2263 out:
2264        ns->pending_mounts = 0;
2265
2266        read_seqlock_excl(&mount_lock);
2267        put_mountpoint(smp);
2268        read_sequnlock_excl(&mount_lock);
2269
2270        return err;
2271}
2272
2273static struct mountpoint *lock_mount(struct path *path)
2274{
2275        struct vfsmount *mnt;
2276        struct dentry *dentry = path->dentry;
2277retry:
2278        inode_lock(dentry->d_inode);
2279        if (unlikely(cant_mount(dentry))) {
2280                inode_unlock(dentry->d_inode);
2281                return ERR_PTR(-ENOENT);
2282        }
2283        namespace_lock();
2284        mnt = lookup_mnt(path);
2285        if (likely(!mnt)) {
2286                struct mountpoint *mp = get_mountpoint(dentry);
2287                if (IS_ERR(mp)) {
2288                        namespace_unlock();
2289                        inode_unlock(dentry->d_inode);
2290                        return mp;
2291                }
2292                return mp;
2293        }
2294        namespace_unlock();
2295        inode_unlock(path->dentry->d_inode);
2296        path_put(path);
2297        path->mnt = mnt;
2298        dentry = path->dentry = dget(mnt->mnt_root);
2299        goto retry;
2300}
2301
2302static void unlock_mount(struct mountpoint *where)
2303{
2304        struct dentry *dentry = where->m_dentry;
2305
2306        read_seqlock_excl(&mount_lock);
2307        put_mountpoint(where);
2308        read_sequnlock_excl(&mount_lock);
2309
2310        namespace_unlock();
2311        inode_unlock(dentry->d_inode);
2312}
2313
2314static int graft_tree(struct mount *mnt, struct mount *p, struct mountpoint *mp)
2315{
2316        if (mnt->mnt.mnt_sb->s_flags & SB_NOUSER)
2317                return -EINVAL;
2318
2319        if (d_is_dir(mp->m_dentry) !=
2320              d_is_dir(mnt->mnt.mnt_root))
2321                return -ENOTDIR;
2322
2323        return attach_recursive_mnt(mnt, p, mp, false);
2324}
2325
2326/*
2327 * Sanity check the flags to change_mnt_propagation.
2328 */
2329
2330static int flags_to_propagation_type(int ms_flags)
2331{
2332        int type = ms_flags & ~(MS_REC | MS_SILENT);
2333
2334        /* Fail if any non-propagation flags are set */
2335        if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2336                return 0;
2337        /* Only one propagation flag should be set */
2338        if (!is_power_of_2(type))
2339                return 0;
2340        return type;
2341}
2342
2343/*
2344 * recursively change the type of the mountpoint.
2345 */
2346static int do_change_type(struct path *path, int ms_flags)
2347{
2348        struct mount *m;
2349        struct mount *mnt = real_mount(path->mnt);
2350        int recurse = ms_flags & MS_REC;
2351        int type;
2352        int err = 0;
2353
2354        if (path->dentry != path->mnt->mnt_root)
2355                return -EINVAL;
2356
2357        type = flags_to_propagation_type(ms_flags);
2358        if (!type)
2359                return -EINVAL;
2360
2361        namespace_lock();
2362        if (type == MS_SHARED) {
2363                err = invent_group_ids(mnt, recurse);
2364                if (err)
2365                        goto out_unlock;
2366        }
2367
2368        lock_mount_hash();
2369        for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
2370                change_mnt_propagation(m, type);
2371        unlock_mount_hash();
2372
2373 out_unlock:
2374        namespace_unlock();
2375        return err;
2376}
2377
2378static struct mount *__do_loopback(struct path *old_path, int recurse)
2379{
2380        struct mount *mnt = ERR_PTR(-EINVAL), *old = real_mount(old_path->mnt);
2381
2382        if (IS_MNT_UNBINDABLE(old))
2383                return mnt;
2384
2385        if (!check_mnt(old) && old_path->dentry->d_op != &ns_dentry_operations)
2386                return mnt;
2387
2388        if (!recurse && has_locked_children(old, old_path->dentry))
2389                return mnt;
2390
2391        if (recurse)
2392                mnt = copy_tree(old, old_path->dentry, CL_COPY_MNT_NS_FILE);
2393        else
2394                mnt = clone_mnt(old, old_path->dentry, 0);
2395
2396        if (!IS_ERR(mnt))
2397                mnt->mnt.mnt_flags &= ~MNT_LOCKED;
2398
2399        return mnt;
2400}
2401
2402/*
2403 * do loopback mount.
2404 */
2405static int do_loopback(struct path *path, const char *old_name,
2406                                int recurse)
2407{
2408        struct path old_path;
2409        struct mount *mnt = NULL, *parent;
2410        struct mountpoint *mp;
2411        int err;
2412        if (!old_name || !*old_name)
2413                return -EINVAL;
2414        err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
2415        if (err)
2416                return err;
2417
2418        err = -EINVAL;
2419        if (mnt_ns_loop(old_path.dentry))
2420                goto out;
2421
2422        mp = lock_mount(path);
2423        if (IS_ERR(mp)) {
2424                err = PTR_ERR(mp);
2425                goto out;
2426        }
2427
2428        parent = real_mount(path->mnt);
2429        if (!check_mnt(parent))
2430                goto out2;
2431
2432        mnt = __do_loopback(&old_path, recurse);
2433        if (IS_ERR(mnt)) {
2434                err = PTR_ERR(mnt);
2435                goto out2;
2436        }
2437
2438        err = graft_tree(mnt, parent, mp);
2439        if (err) {
2440                lock_mount_hash();
2441                umount_tree(mnt, UMOUNT_SYNC);
2442                unlock_mount_hash();
2443        }
2444out2:
2445        unlock_mount(mp);
2446out:
2447        path_put(&old_path);
2448        return err;
2449}
2450
2451static struct file *open_detached_copy(struct path *path, bool recursive)
2452{
2453        struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
2454        struct mnt_namespace *ns = alloc_mnt_ns(user_ns, true);
2455        struct mount *mnt, *p;
2456        struct file *file;
2457
2458        if (IS_ERR(ns))
2459                return ERR_CAST(ns);
2460
2461        namespace_lock();
2462        mnt = __do_loopback(path, recursive);
2463        if (IS_ERR(mnt)) {
2464                namespace_unlock();
2465                free_mnt_ns(ns);
2466                return ERR_CAST(mnt);
2467        }
2468
2469        lock_mount_hash();
2470        for (p = mnt; p; p = next_mnt(p, mnt)) {
2471                p->mnt_ns = ns;
2472                ns->mounts++;
2473        }
2474        ns->root = mnt;
2475        list_add_tail(&ns->list, &mnt->mnt_list);
2476        mntget(&mnt->mnt);
2477        unlock_mount_hash();
2478        namespace_unlock();
2479
2480        mntput(path->mnt);
2481        path->mnt = &mnt->mnt;
2482        file = dentry_open(path, O_PATH, current_cred());
2483        if (IS_ERR(file))
2484                dissolve_on_fput(path->mnt);
2485        else
2486                file->f_mode |= FMODE_NEED_UNMOUNT;
2487        return file;
2488}
2489
2490SYSCALL_DEFINE3(open_tree, int, dfd, const char __user *, filename, unsigned, flags)
2491{
2492        struct file *file;
2493        struct path path;
2494        int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
2495        bool detached = flags & OPEN_TREE_CLONE;
2496        int error;
2497        int fd;
2498
2499        BUILD_BUG_ON(OPEN_TREE_CLOEXEC != O_CLOEXEC);
2500
2501        if (flags & ~(AT_EMPTY_PATH | AT_NO_AUTOMOUNT | AT_RECURSIVE |
2502                      AT_SYMLINK_NOFOLLOW | OPEN_TREE_CLONE |
2503                      OPEN_TREE_CLOEXEC))
2504                return -EINVAL;
2505
2506        if ((flags & (AT_RECURSIVE | OPEN_TREE_CLONE)) == AT_RECURSIVE)
2507                return -EINVAL;
2508
2509        if (flags & AT_NO_AUTOMOUNT)
2510                lookup_flags &= ~LOOKUP_AUTOMOUNT;
2511        if (flags & AT_SYMLINK_NOFOLLOW)
2512                lookup_flags &= ~LOOKUP_FOLLOW;
2513        if (flags & AT_EMPTY_PATH)
2514                lookup_flags |= LOOKUP_EMPTY;
2515
2516        if (detached && !may_mount())
2517                return -EPERM;
2518
2519        fd = get_unused_fd_flags(flags & O_CLOEXEC);
2520        if (fd < 0)
2521                return fd;
2522
2523        error = user_path_at(dfd, filename, lookup_flags, &path);
2524        if (unlikely(error)) {
2525                file = ERR_PTR(error);
2526        } else {
2527                if (detached)
2528                        file = open_detached_copy(&path, flags & AT_RECURSIVE);
2529                else
2530                        file = dentry_open(&path, O_PATH, current_cred());
2531                path_put(&path);
2532        }
2533        if (IS_ERR(file)) {
2534                put_unused_fd(fd);
2535                return PTR_ERR(file);
2536        }
2537        fd_install(fd, file);
2538        return fd;
2539}
2540
2541/*
2542 * Don't allow locked mount flags to be cleared.
2543 *
2544 * No locks need to be held here while testing the various MNT_LOCK
2545 * flags because those flags can never be cleared once they are set.
2546 */
2547static bool can_change_locked_flags(struct mount *mnt, unsigned int mnt_flags)
2548{
2549        unsigned int fl = mnt->mnt.mnt_flags;
2550
2551        if ((fl & MNT_LOCK_READONLY) &&
2552            !(mnt_flags & MNT_READONLY))
2553                return false;
2554
2555        if ((fl & MNT_LOCK_NODEV) &&
2556            !(mnt_flags & MNT_NODEV))
2557                return false;
2558
2559        if ((fl & MNT_LOCK_NOSUID) &&
2560            !(mnt_flags & MNT_NOSUID))
2561                return false;
2562
2563        if ((fl & MNT_LOCK_NOEXEC) &&
2564            !(mnt_flags & MNT_NOEXEC))
2565                return false;
2566
2567        if ((fl & MNT_LOCK_ATIME) &&
2568            ((fl & MNT_ATIME_MASK) != (mnt_flags & MNT_ATIME_MASK)))
2569                return false;
2570
2571        return true;
2572}
2573
2574static int change_mount_ro_state(struct mount *mnt, unsigned int mnt_flags)
2575{
2576        bool readonly_request = (mnt_flags & MNT_READONLY);
2577
2578        if (readonly_request == __mnt_is_readonly(&mnt->mnt))
2579                return 0;
2580
2581        if (readonly_request)
2582                return mnt_make_readonly(mnt);
2583
2584        mnt->mnt.mnt_flags &= ~MNT_READONLY;
2585        return 0;
2586}
2587
2588static void set_mount_attributes(struct mount *mnt, unsigned int mnt_flags)
2589{
2590        mnt_flags |= mnt->mnt.mnt_flags & ~MNT_USER_SETTABLE_MASK;
2591        mnt->mnt.mnt_flags = mnt_flags;
2592        touch_mnt_namespace(mnt->mnt_ns);
2593}
2594
2595static void mnt_warn_timestamp_expiry(struct path *mountpoint, struct vfsmount *mnt)
2596{
2597        struct super_block *sb = mnt->mnt_sb;
2598
2599        if (!__mnt_is_readonly(mnt) &&
2600           (ktime_get_real_seconds() + TIME_UPTIME_SEC_MAX > sb->s_time_max)) {
2601                char *buf = (char *)__get_free_page(GFP_KERNEL);
2602                char *mntpath = buf ? d_path(mountpoint, buf, PAGE_SIZE) : ERR_PTR(-ENOMEM);
2603                struct tm tm;
2604
2605                time64_to_tm(sb->s_time_max, 0, &tm);
2606
2607                pr_warn("%s filesystem being %s at %s supports timestamps until %04ld (0x%llx)\n",
2608                        sb->s_type->name,
2609                        is_mounted(mnt) ? "remounted" : "mounted",
2610                        mntpath,
2611                        tm.tm_year+1900, (unsigned long long)sb->s_time_max);
2612
2613                free_page((unsigned long)buf);
2614        }
2615}
2616
2617/*
2618 * Handle reconfiguration of the mountpoint only without alteration of the
2619 * superblock it refers to.  This is triggered by specifying MS_REMOUNT|MS_BIND
2620 * to mount(2).
2621 */
2622static int do_reconfigure_mnt(struct path *path, unsigned int mnt_flags)
2623{
2624        struct super_block *sb = path->mnt->mnt_sb;
2625        struct mount *mnt = real_mount(path->mnt);
2626        int ret;
2627
2628        if (!check_mnt(mnt))
2629                return -EINVAL;
2630
2631        if (path->dentry != mnt->mnt.mnt_root)
2632                return -EINVAL;
2633
2634        if (!can_change_locked_flags(mnt, mnt_flags))
2635                return -EPERM;
2636
2637        /*
2638         * We're only checking whether the superblock is read-only not
2639         * changing it, so only take down_read(&sb->s_umount).
2640         */
2641        down_read(&sb->s_umount);
2642        lock_mount_hash();
2643        ret = change_mount_ro_state(mnt, mnt_flags);
2644        if (ret == 0)
2645                set_mount_attributes(mnt, mnt_flags);
2646        unlock_mount_hash();
2647        up_read(&sb->s_umount);
2648
2649        mnt_warn_timestamp_expiry(path, &mnt->mnt);
2650
2651        return ret;
2652}
2653
2654/*
2655 * change filesystem flags. dir should be a physical root of filesystem.
2656 * If you've mounted a non-root directory somewhere and want to do remount
2657 * on it - tough luck.
2658 */
2659static int do_remount(struct path *path, int ms_flags, int sb_flags,
2660                      int mnt_flags, void *data)
2661{
2662        int err;
2663        struct super_block *sb = path->mnt->mnt_sb;
2664        struct mount *mnt = real_mount(path->mnt);
2665        struct fs_context *fc;
2666
2667        if (!check_mnt(mnt))
2668                return -EINVAL;
2669
2670        if (path->dentry != path->mnt->mnt_root)
2671                return -EINVAL;
2672
2673        if (!can_change_locked_flags(mnt, mnt_flags))
2674                return -EPERM;
2675
2676        fc = fs_context_for_reconfigure(path->dentry, sb_flags, MS_RMT_MASK);
2677        if (IS_ERR(fc))
2678                return PTR_ERR(fc);
2679
2680        fc->oldapi = true;
2681        err = parse_monolithic_mount_data(fc, data);
2682        if (!err) {
2683                down_write(&sb->s_umount);
2684                err = -EPERM;
2685                if (ns_capable(sb->s_user_ns, CAP_SYS_ADMIN)) {
2686                        err = reconfigure_super(fc);
2687                        if (!err) {
2688                                lock_mount_hash();
2689                                set_mount_attributes(mnt, mnt_flags);
2690                                unlock_mount_hash();
2691                        }
2692                }
2693                up_write(&sb->s_umount);
2694        }
2695
2696        mnt_warn_timestamp_expiry(path, &mnt->mnt);
2697
2698        put_fs_context(fc);
2699        return err;
2700}
2701
2702static inline int tree_contains_unbindable(struct mount *mnt)
2703{
2704        struct mount *p;
2705        for (p = mnt; p; p = next_mnt(p, mnt)) {
2706                if (IS_MNT_UNBINDABLE(p))
2707                        return 1;
2708        }
2709        return 0;
2710}
2711
2712/*
2713 * Check that there aren't references to earlier/same mount namespaces in the
2714 * specified subtree.  Such references can act as pins for mount namespaces
2715 * that aren't checked by the mount-cycle checking code, thereby allowing
2716 * cycles to be made.
2717 */
2718static bool check_for_nsfs_mounts(struct mount *subtree)
2719{
2720        struct mount *p;
2721        bool ret = false;
2722
2723        lock_mount_hash();
2724        for (p = subtree; p; p = next_mnt(p, subtree))
2725                if (mnt_ns_loop(p->mnt.mnt_root))
2726                        goto out;
2727
2728        ret = true;
2729out:
2730        unlock_mount_hash();
2731        return ret;
2732}
2733
2734static int do_set_group(struct path *from_path, struct path *to_path)
2735{
2736        struct mount *from, *to;
2737        int err;
2738
2739        from = real_mount(from_path->mnt);
2740        to = real_mount(to_path->mnt);
2741
2742        namespace_lock();
2743
2744        err = -EINVAL;
2745        /* To and From must be mounted */
2746        if (!is_mounted(&from->mnt))
2747                goto out;
2748        if (!is_mounted(&to->mnt))
2749                goto out;
2750
2751        err = -EPERM;
2752        /* We should be allowed to modify mount namespaces of both mounts */
2753        if (!ns_capable(from->mnt_ns->user_ns, CAP_SYS_ADMIN))
2754                goto out;
2755        if (!ns_capable(to->mnt_ns->user_ns, CAP_SYS_ADMIN))
2756                goto out;
2757
2758        err = -EINVAL;
2759        /* To and From paths should be mount roots */
2760        if (from_path->dentry != from_path->mnt->mnt_root)
2761                goto out;
2762        if (to_path->dentry != to_path->mnt->mnt_root)
2763                goto out;
2764
2765        /* Setting sharing groups is only allowed across same superblock */
2766        if (from->mnt.mnt_sb != to->mnt.mnt_sb)
2767                goto out;
2768
2769        /* From mount root should be wider than To mount root */
2770        if (!is_subdir(to->mnt.mnt_root, from->mnt.mnt_root))
2771                goto out;
2772
2773        /* From mount should not have locked children in place of To's root */
2774        if (has_locked_children(from, to->mnt.mnt_root))
2775                goto out;
2776
2777        /* Setting sharing groups is only allowed on private mounts */
2778        if (IS_MNT_SHARED(to) || IS_MNT_SLAVE(to))
2779                goto out;
2780
2781        /* From should not be private */
2782        if (!IS_MNT_SHARED(from) && !IS_MNT_SLAVE(from))
2783                goto out;
2784
2785        if (IS_MNT_SLAVE(from)) {
2786                struct mount *m = from->mnt_master;
2787
2788                list_add(&to->mnt_slave, &m->mnt_slave_list);
2789                to->mnt_master = m;
2790        }
2791
2792        if (IS_MNT_SHARED(from)) {
2793                to->mnt_group_id = from->mnt_group_id;
2794                list_add(&to->mnt_share, &from->mnt_share);
2795                lock_mount_hash();
2796                set_mnt_shared(to);
2797                unlock_mount_hash();
2798        }
2799
2800        err = 0;
2801out:
2802        namespace_unlock();
2803        return err;
2804}
2805
2806static int do_move_mount(struct path *old_path, struct path *new_path)
2807{
2808        struct mnt_namespace *ns;
2809        struct mount *p;
2810        struct mount *old;
2811        struct mount *parent;
2812        struct mountpoint *mp, *old_mp;
2813        int err;
2814        bool attached;
2815
2816        mp = lock_mount(new_path);
2817        if (IS_ERR(mp))
2818                return PTR_ERR(mp);
2819
2820        old = real_mount(old_path->mnt);
2821        p = real_mount(new_path->mnt);
2822        parent = old->mnt_parent;
2823        attached = mnt_has_parent(old);
2824        old_mp = old->mnt_mp;
2825        ns = old->mnt_ns;
2826
2827        err = -EINVAL;
2828        /* The mountpoint must be in our namespace. */
2829        if (!check_mnt(p))
2830                goto out;
2831
2832        /* The thing moved must be mounted... */
2833        if (!is_mounted(&old->mnt))
2834                goto out;
2835
2836        /* ... and either ours or the root of anon namespace */
2837        if (!(attached ? check_mnt(old) : is_anon_ns(ns)))
2838                goto out;
2839
2840        if (old->mnt.mnt_flags & MNT_LOCKED)
2841                goto out;
2842
2843        if (old_path->dentry != old_path->mnt->mnt_root)
2844                goto out;
2845
2846        if (d_is_dir(new_path->dentry) !=
2847            d_is_dir(old_path->dentry))
2848                goto out;
2849        /*
2850         * Don't move a mount residing in a shared parent.
2851         */
2852        if (attached && IS_MNT_SHARED(parent))
2853                goto out;
2854        /*
2855         * Don't move a mount tree containing unbindable mounts to a destination
2856         * mount which is shared.
2857         */
2858        if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
2859                goto out;
2860        err = -ELOOP;
2861        if (!check_for_nsfs_mounts(old))
2862                goto out;
2863        for (; mnt_has_parent(p); p = p->mnt_parent)
2864                if (p == old)
2865                        goto out;
2866
2867        err = attach_recursive_mnt(old, real_mount(new_path->mnt), mp,
2868                                   attached);
2869        if (err)
2870                goto out;
2871
2872        /* if the mount is moved, it should no longer be expire
2873         * automatically */
2874        list_del_init(&old->mnt_expire);
2875        if (attached)
2876                put_mountpoint(old_mp);
2877out:
2878        unlock_mount(mp);
2879        if (!err) {
2880                if (attached)
2881                        mntput_no_expire(parent);
2882                else
2883                        free_mnt_ns(ns);
2884        }
2885        return err;
2886}
2887
2888static int do_move_mount_old(struct path *path, const char *old_name)
2889{
2890        struct path old_path;
2891        int err;
2892
2893        if (!old_name || !*old_name)
2894                return -EINVAL;
2895
2896        err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
2897        if (err)
2898                return err;
2899
2900        err = do_move_mount(&old_path, path);
2901        path_put(&old_path);
2902        return err;
2903}
2904
2905/*
2906 * add a mount into a namespace's mount tree
2907 */
2908static int do_add_mount(struct mount *newmnt, struct mountpoint *mp,
2909                        struct path *path, int mnt_flags)
2910{
2911        struct mount *parent = real_mount(path->mnt);
2912
2913        mnt_flags &= ~MNT_INTERNAL_FLAGS;
2914
2915        if (unlikely(!check_mnt(parent))) {
2916                /* that's acceptable only for automounts done in private ns */
2917                if (!(mnt_flags & MNT_SHRINKABLE))
2918                        return -EINVAL;
2919                /* ... and for those we'd better have mountpoint still alive */
2920                if (!parent->mnt_ns)
2921                        return -EINVAL;
2922        }
2923
2924        /* Refuse the same filesystem on the same mount point */
2925        if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
2926            path->mnt->mnt_root == path->dentry)
2927                return -EBUSY;
2928
2929        if (d_is_symlink(newmnt->mnt.mnt_root))
2930                return -EINVAL;
2931
2932        newmnt->mnt.mnt_flags = mnt_flags;
2933        return graft_tree(newmnt, parent, mp);
2934}
2935
2936static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags);
2937
2938/*
2939 * Create a new mount using a superblock configuration and request it
2940 * be added to the namespace tree.
2941 */
2942static int do_new_mount_fc(struct fs_context *fc, struct path *mountpoint,
2943                           unsigned int mnt_flags)
2944{
2945        struct vfsmount *mnt;
2946        struct mountpoint *mp;
2947        struct super_block *sb = fc->root->d_sb;
2948        int error;
2949
2950        error = security_sb_kern_mount(sb);
2951        if (!error && mount_too_revealing(sb, &mnt_flags))
2952                error = -EPERM;
2953
2954        if (unlikely(error)) {
2955                fc_drop_locked(fc);
2956                return error;
2957        }
2958
2959        up_write(&sb->s_umount);
2960
2961        mnt = vfs_create_mount(fc);
2962        if (IS_ERR(mnt))
2963                return PTR_ERR(mnt);
2964
2965        mnt_warn_timestamp_expiry(mountpoint, mnt);
2966
2967        mp = lock_mount(mountpoint);
2968        if (IS_ERR(mp)) {
2969                mntput(mnt);
2970                return PTR_ERR(mp);
2971        }
2972        error = do_add_mount(real_mount(mnt), mp, mountpoint, mnt_flags);
2973        unlock_mount(mp);
2974        if (error < 0)
2975                mntput(mnt);
2976        return error;
2977}
2978
2979/*
2980 * create a new mount for userspace and request it to be added into the
2981 * namespace's tree
2982 */
2983static int do_new_mount(struct path *path, const char *fstype, int sb_flags,
2984                        int mnt_flags, const char *name, void *data)
2985{
2986        struct file_system_type *type;
2987        struct fs_context *fc;
2988        const char *subtype = NULL;
2989        int err = 0;
2990
2991        if (!fstype)
2992                return -EINVAL;
2993
2994        type = get_fs_type(fstype);
2995        if (!type)
2996                return -ENODEV;
2997
2998        if (type->fs_flags & FS_HAS_SUBTYPE) {
2999                subtype = strchr(fstype, '.');
3000                if (subtype) {
3001                        subtype++;
3002                        if (!*subtype) {
3003                                put_filesystem(type);
3004                                return -EINVAL;
3005                        }
3006                }
3007        }
3008
3009        fc = fs_context_for_mount(type, sb_flags);
3010        put_filesystem(type);
3011        if (IS_ERR(fc))
3012                return PTR_ERR(fc);
3013
3014        if (subtype)
3015                err = vfs_parse_fs_string(fc, "subtype",
3016                                          subtype, strlen(subtype));
3017        if (!err && name)
3018                err = vfs_parse_fs_string(fc, "source", name, strlen(name));
3019        if (!err)
3020                err = parse_monolithic_mount_data(fc, data);
3021        if (!err && !mount_capable(fc))
3022                err = -EPERM;
3023        if (!err)
3024                err = vfs_get_tree(fc);
3025        if (!err)
3026                err = do_new_mount_fc(fc, path, mnt_flags);
3027
3028        put_fs_context(fc);
3029        return err;
3030}
3031
3032int finish_automount(struct vfsmount *m, struct path *path)
3033{
3034        struct dentry *dentry = path->dentry;
3035        struct mountpoint *mp;
3036        struct mount *mnt;
3037        int err;
3038
3039        if (!m)
3040                return 0;
3041        if (IS_ERR(m))
3042                return PTR_ERR(m);
3043
3044        mnt = real_mount(m);
3045        /* The new mount record should have at least 2 refs to prevent it being
3046         * expired before we get a chance to add it
3047         */
3048        BUG_ON(mnt_get_count(mnt) < 2);
3049
3050        if (m->mnt_sb == path->mnt->mnt_sb &&
3051            m->mnt_root == dentry) {
3052                err = -ELOOP;
3053                goto discard;
3054        }
3055
3056        /*
3057         * we don't want to use lock_mount() - in this case finding something
3058         * that overmounts our mountpoint to be means "quitely drop what we've
3059         * got", not "try to mount it on top".
3060         */
3061        inode_lock(dentry->d_inode);
3062        namespace_lock();
3063        if (unlikely(cant_mount(dentry))) {
3064                err = -ENOENT;
3065                goto discard_locked;
3066        }
3067        rcu_read_lock();
3068        if (unlikely(__lookup_mnt(path->mnt, dentry))) {
3069                rcu_read_unlock();
3070                err = 0;
3071                goto discard_locked;
3072        }
3073        rcu_read_unlock();
3074        mp = get_mountpoint(dentry);
3075        if (IS_ERR(mp)) {
3076                err = PTR_ERR(mp);
3077                goto discard_locked;
3078        }
3079
3080        err = do_add_mount(mnt, mp, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
3081        unlock_mount(mp);
3082        if (unlikely(err))
3083                goto discard;
3084        mntput(m);
3085        return 0;
3086
3087discard_locked:
3088        namespace_unlock();
3089        inode_unlock(dentry->d_inode);
3090discard:
3091        /* remove m from any expiration list it may be on */
3092        if (!list_empty(&mnt->mnt_expire)) {
3093                namespace_lock();
3094                list_del_init(&mnt->mnt_expire);
3095                namespace_unlock();
3096        }
3097        mntput(m);
3098        mntput(m);
3099        return err;
3100}
3101
3102/**
3103 * mnt_set_expiry - Put a mount on an expiration list
3104 * @mnt: The mount to list.
3105 * @expiry_list: The list to add the mount to.
3106 */
3107void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
3108{
3109        namespace_lock();
3110
3111        list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
3112
3113        namespace_unlock();
3114}
3115EXPORT_SYMBOL(mnt_set_expiry);
3116
3117/*
3118 * process a list of expirable mountpoints with the intent of discarding any
3119 * mountpoints that aren't in use and haven't been touched since last we came
3120 * here
3121 */
3122void mark_mounts_for_expiry(struct list_head *mounts)
3123{
3124        struct mount *mnt, *next;
3125        LIST_HEAD(graveyard);
3126
3127        if (list_empty(mounts))
3128                return;
3129
3130        namespace_lock();
3131        lock_mount_hash();
3132
3133        /* extract from the expiration list every vfsmount that matches the
3134         * following criteria:
3135         * - only referenced by its parent vfsmount
3136         * - still marked for expiry (marked on the last call here; marks are
3137         *   cleared by mntput())
3138         */
3139        list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
3140                if (!xchg(&mnt->mnt_expiry_mark, 1) ||
3141                        propagate_mount_busy(mnt, 1))
3142                        continue;
3143                list_move(&mnt->mnt_expire, &graveyard);
3144        }
3145        while (!list_empty(&graveyard)) {
3146                mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
3147                touch_mnt_namespace(mnt->mnt_ns);
3148                umount_tree(mnt, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3149        }
3150        unlock_mount_hash();
3151        namespace_unlock();
3152}
3153
3154EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
3155
3156/*
3157 * Ripoff of 'select_parent()'
3158 *
3159 * search the list of submounts for a given mountpoint, and move any
3160 * shrinkable submounts to the 'graveyard' list.
3161 */
3162static int select_submounts(struct mount *parent, struct list_head *graveyard)
3163{
3164        struct mount *this_parent = parent;
3165        struct list_head *next;
3166        int found = 0;
3167
3168repeat:
3169        next = this_parent->mnt_mounts.next;
3170resume:
3171        while (next != &this_parent->mnt_mounts) {
3172                struct list_head *tmp = next;
3173                struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
3174
3175                next = tmp->next;
3176                if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
3177                        continue;
3178                /*
3179                 * Descend a level if the d_mounts list is non-empty.
3180                 */
3181                if (!list_empty(&mnt->mnt_mounts)) {
3182                        this_parent = mnt;
3183                        goto repeat;
3184                }
3185
3186                if (!propagate_mount_busy(mnt, 1)) {
3187                        list_move_tail(&mnt->mnt_expire, graveyard);
3188                        found++;
3189                }
3190        }
3191        /*
3192         * All done at this level ... ascend and resume the search
3193         */
3194        if (this_parent != parent) {
3195                next = this_parent->mnt_child.next;
3196                this_parent = this_parent->mnt_parent;
3197                goto resume;
3198        }
3199        return found;
3200}
3201
3202/*
3203 * process a list of expirable mountpoints with the intent of discarding any
3204 * submounts of a specific parent mountpoint
3205 *
3206 * mount_lock must be held for write
3207 */
3208static void shrink_submounts(struct mount *mnt)
3209{
3210        LIST_HEAD(graveyard);
3211        struct mount *m;
3212
3213        /* extract submounts of 'mountpoint' from the expiration list */
3214        while (select_submounts(mnt, &graveyard)) {
3215                while (!list_empty(&graveyard)) {
3216                        m = list_first_entry(&graveyard, struct mount,
3217                                                mnt_expire);
3218                        touch_mnt_namespace(m->mnt_ns);
3219                        umount_tree(m, UMOUNT_PROPAGATE|UMOUNT_SYNC);
3220                }
3221        }
3222}
3223
3224static void *copy_mount_options(const void __user * data)
3225{
3226        char *copy;
3227        unsigned left, offset;
3228
3229        if (!data)
3230                return NULL;
3231
3232        copy = kmalloc(PAGE_SIZE, GFP_KERNEL);
3233        if (!copy)
3234                return ERR_PTR(-ENOMEM);
3235
3236        left = copy_from_user(copy, data, PAGE_SIZE);
3237
3238        /*
3239         * Not all architectures have an exact copy_from_user(). Resort to
3240         * byte at a time.
3241         */
3242        offset = PAGE_SIZE - left;
3243        while (left) {
3244                char c;
3245                if (get_user(c, (const char __user *)data + offset))
3246                        break;
3247                copy[offset] = c;
3248                left--;
3249                offset++;
3250        }
3251
3252        if (left == PAGE_SIZE) {
3253                kfree(copy);
3254                return ERR_PTR(-EFAULT);
3255        }
3256
3257        return copy;
3258}
3259
3260static char *copy_mount_string(const void __user *data)
3261{
3262        return data ? strndup_user(data, PATH_MAX) : NULL;
3263}
3264
3265/*
3266 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
3267 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
3268 *
3269 * data is a (void *) that can point to any structure up to
3270 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
3271 * information (or be NULL).
3272 *
3273 * Pre-0.97 versions of mount() didn't have a flags word.
3274 * When the flags word was introduced its top half was required
3275 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
3276 * Therefore, if this magic number is present, it carries no information
3277 * and must be discarded.
3278 */
3279int path_mount(const char *dev_name, struct path *path,
3280                const char *type_page, unsigned long flags, void *data_page)
3281{
3282        unsigned int mnt_flags = 0, sb_flags;
3283        int ret;
3284
3285        /* Discard magic */
3286        if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
3287                flags &= ~MS_MGC_MSK;
3288
3289        /* Basic sanity checks */
3290        if (data_page)
3291                ((char *)data_page)[PAGE_SIZE - 1] = 0;
3292
3293        if (flags & MS_NOUSER)
3294                return -EINVAL;
3295
3296        ret = security_sb_mount(dev_name, path, type_page, flags, data_page);
3297        if (ret)
3298                return ret;
3299        if (!may_mount())
3300                return -EPERM;
3301        if (flags & SB_MANDLOCK)
3302                warn_mandlock();
3303
3304        /* Default to relatime unless overriden */
3305        if (!(flags & MS_NOATIME))
3306                mnt_flags |= MNT_RELATIME;
3307
3308        /* Separate the per-mountpoint flags */
3309        if (flags & MS_NOSUID)
3310                mnt_flags |= MNT_NOSUID;
3311        if (flags & MS_NODEV)
3312                mnt_flags |= MNT_NODEV;
3313        if (flags & MS_NOEXEC)
3314                mnt_flags |= MNT_NOEXEC;
3315        if (flags & MS_NOATIME)
3316                mnt_flags |= MNT_NOATIME;
3317        if (flags & MS_NODIRATIME)
3318                mnt_flags |= MNT_NODIRATIME;
3319        if (flags & MS_STRICTATIME)
3320                mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
3321        if (flags & MS_RDONLY)
3322                mnt_flags |= MNT_READONLY;
3323        if (flags & MS_NOSYMFOLLOW)
3324                mnt_flags |= MNT_NOSYMFOLLOW;
3325
3326        /* The default atime for remount is preservation */
3327        if ((flags & MS_REMOUNT) &&
3328            ((flags & (MS_NOATIME | MS_NODIRATIME | MS_RELATIME |
3329                       MS_STRICTATIME)) == 0)) {
3330                mnt_flags &= ~MNT_ATIME_MASK;
3331                mnt_flags |= path->mnt->mnt_flags & MNT_ATIME_MASK;
3332        }
3333
3334        sb_flags = flags & (SB_RDONLY |
3335                            SB_SYNCHRONOUS |
3336                            SB_MANDLOCK |
3337                            SB_DIRSYNC |
3338                            SB_SILENT |
3339                            SB_POSIXACL |
3340                            SB_LAZYTIME |
3341                            SB_I_VERSION);
3342
3343        if ((flags & (MS_REMOUNT | MS_BIND)) == (MS_REMOUNT | MS_BIND))
3344                return do_reconfigure_mnt(path, mnt_flags);
3345        if (flags & MS_REMOUNT)
3346                return do_remount(path, flags, sb_flags, mnt_flags, data_page);
3347        if (flags & MS_BIND)
3348                return do_loopback(path, dev_name, flags & MS_REC);
3349        if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
3350                return do_change_type(path, flags);
3351        if (flags & MS_MOVE)
3352                return do_move_mount_old(path, dev_name);
3353
3354        return do_new_mount(path, type_page, sb_flags, mnt_flags, dev_name,
3355                            data_page);
3356}
3357
3358long do_mount(const char *dev_name, const char __user *dir_name,
3359                const char *type_page, unsigned long flags, void *data_page)
3360{
3361        struct path path;
3362        int ret;
3363
3364        ret = user_path_at(AT_FDCWD, dir_name, LOOKUP_FOLLOW, &path);
3365        if (ret)
3366                return ret;
3367        ret = path_mount(dev_name, &path, type_page, flags, data_page);
3368        path_put(&path);
3369        return ret;
3370}
3371
3372static struct ucounts *inc_mnt_namespaces(struct user_namespace *ns)
3373{
3374        return inc_ucount(ns, current_euid(), UCOUNT_MNT_NAMESPACES);
3375}
3376
3377static void dec_mnt_namespaces(struct ucounts *ucounts)
3378{
3379        dec_ucount(ucounts, UCOUNT_MNT_NAMESPACES);
3380}
3381
3382static void free_mnt_ns(struct mnt_namespace *ns)
3383{
3384        if (!is_anon_ns(ns))
3385                ns_free_inum(&ns->ns);
3386        dec_mnt_namespaces(ns->ucounts);
3387        put_user_ns(ns->user_ns);
3388        kfree(ns);
3389}
3390
3391/*
3392 * Assign a sequence number so we can detect when we attempt to bind
3393 * mount a reference to an older mount namespace into the current
3394 * mount namespace, preventing reference counting loops.  A 64bit
3395 * number incrementing at 10Ghz will take 12,427 years to wrap which
3396 * is effectively never, so we can ignore the possibility.
3397 */
3398static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
3399
3400static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns, bool anon)
3401{
3402        struct mnt_namespace *new_ns;
3403        struct ucounts *ucounts;
3404        int ret;
3405
3406        ucounts = inc_mnt_namespaces(user_ns);
3407        if (!ucounts)
3408                return ERR_PTR(-ENOSPC);
3409
3410        new_ns = kzalloc(sizeof(struct mnt_namespace), GFP_KERNEL_ACCOUNT);
3411        if (!new_ns) {
3412                dec_mnt_namespaces(ucounts);
3413                return ERR_PTR(-ENOMEM);
3414        }
3415        if (!anon) {
3416                ret = ns_alloc_inum(&new_ns->ns);
3417                if (ret) {
3418                        kfree(new_ns);
3419                        dec_mnt_namespaces(ucounts);
3420                        return ERR_PTR(ret);
3421                }
3422        }
3423        new_ns->ns.ops = &mntns_operations;
3424        if (!anon)
3425                new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
3426        refcount_set(&new_ns->ns.count, 1);
3427        INIT_LIST_HEAD(&new_ns->list);
3428        init_waitqueue_head(&new_ns->poll);
3429        spin_lock_init(&new_ns->ns_lock);
3430        new_ns->user_ns = get_user_ns(user_ns);
3431        new_ns->ucounts = ucounts;
3432        return new_ns;
3433}
3434
3435__latent_entropy
3436struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
3437                struct user_namespace *user_ns, struct fs_struct *new_fs)
3438{
3439        struct mnt_namespace *new_ns;
3440        struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
3441        struct mount *p, *q;
3442        struct mount *old;
3443        struct mount *new;
3444        int copy_flags;
3445
3446        BUG_ON(!ns);
3447
3448        if (likely(!(flags & CLONE_NEWNS))) {
3449                get_mnt_ns(ns);
3450                return ns;
3451        }
3452
3453        old = ns->root;
3454
3455        new_ns = alloc_mnt_ns(user_ns, false);
3456        if (IS_ERR(new_ns))
3457                return new_ns;
3458
3459        namespace_lock();
3460        /* First pass: copy the tree topology */
3461        copy_flags = CL_COPY_UNBINDABLE | CL_EXPIRE;
3462        if (user_ns != ns->user_ns)
3463                copy_flags |= CL_SHARED_TO_SLAVE;
3464        new = copy_tree(old, old->mnt.mnt_root, copy_flags);
3465        if (IS_ERR(new)) {
3466                namespace_unlock();
3467                free_mnt_ns(new_ns);
3468                return ERR_CAST(new);
3469        }
3470        if (user_ns != ns->user_ns) {
3471                lock_mount_hash();
3472                lock_mnt_tree(new);
3473                unlock_mount_hash();
3474        }
3475        new_ns->root = new;
3476        list_add_tail(&new_ns->list, &new->mnt_list);
3477
3478        /*
3479         * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
3480         * as belonging to new namespace.  We have already acquired a private
3481         * fs_struct, so tsk->fs->lock is not needed.
3482         */
3483        p = old;
3484        q = new;
3485        while (p) {
3486                q->mnt_ns = new_ns;
3487                new_ns->mounts++;
3488                if (new_fs) {
3489                        if (&p->mnt == new_fs->root.mnt) {
3490                                new_fs->root.mnt = mntget(&q->mnt);
3491                                rootmnt = &p->mnt;
3492                        }
3493                        if (&p->mnt == new_fs->pwd.mnt) {
3494                                new_fs->pwd.mnt = mntget(&q->mnt);
3495                                pwdmnt = &p->mnt;
3496                        }
3497                }
3498                p = next_mnt(p, old);
3499                q = next_mnt(q, new);
3500                if (!q)
3501                        break;
3502                while (p->mnt.mnt_root != q->mnt.mnt_root)
3503                        p = next_mnt(p, old);
3504        }
3505        namespace_unlock();
3506
3507        if (rootmnt)
3508                mntput(rootmnt);
3509        if (pwdmnt)
3510                mntput(pwdmnt);
3511
3512        return new_ns;
3513}
3514
3515struct dentry *mount_subtree(struct vfsmount *m, const char *name)
3516{
3517        struct mount *mnt = real_mount(m);
3518        struct mnt_namespace *ns;
3519        struct super_block *s;
3520        struct path path;
3521        int err;
3522
3523        ns = alloc_mnt_ns(&init_user_ns, true);
3524        if (IS_ERR(ns)) {
3525                mntput(m);
3526                return ERR_CAST(ns);
3527        }
3528        mnt->mnt_ns = ns;
3529        ns->root = mnt;
3530        ns->mounts++;
3531        list_add(&mnt->mnt_list, &ns->list);
3532
3533        err = vfs_path_lookup(m->mnt_root, m,
3534                        name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
3535
3536        put_mnt_ns(ns);
3537
3538        if (err)
3539                return ERR_PTR(err);
3540
3541        /* trade a vfsmount reference for active sb one */
3542        s = path.mnt->mnt_sb;
3543        atomic_inc(&s->s_active);
3544        mntput(path.mnt);
3545        /* lock the sucker */
3546        down_write(&s->s_umount);
3547        /* ... and return the root of (sub)tree on it */
3548        return path.dentry;
3549}
3550EXPORT_SYMBOL(mount_subtree);
3551
3552SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
3553                char __user *, type, unsigned long, flags, void __user *, data)
3554{
3555        int ret;
3556        char *kernel_type;
3557        char *kernel_dev;
3558        void *options;
3559
3560        kernel_type = copy_mount_string(type);
3561        ret = PTR_ERR(kernel_type);
3562        if (IS_ERR(kernel_type))
3563                goto out_type;
3564
3565        kernel_dev = copy_mount_string(dev_name);
3566        ret = PTR_ERR(kernel_dev);
3567        if (IS_ERR(kernel_dev))
3568                goto out_dev;
3569
3570        options = copy_mount_options(data);
3571        ret = PTR_ERR(options);
3572        if (IS_ERR(options))
3573                goto out_data;
3574
3575        ret = do_mount(kernel_dev, dir_name, kernel_type, flags, options);
3576
3577        kfree(options);
3578out_data:
3579        kfree(kernel_dev);
3580out_dev:
3581        kfree(kernel_type);
3582out_type:
3583        return ret;
3584}
3585
3586#define FSMOUNT_VALID_FLAGS                                                    \
3587        (MOUNT_ATTR_RDONLY | MOUNT_ATTR_NOSUID | MOUNT_ATTR_NODEV |            \
3588         MOUNT_ATTR_NOEXEC | MOUNT_ATTR__ATIME | MOUNT_ATTR_NODIRATIME |       \
3589         MOUNT_ATTR_NOSYMFOLLOW)
3590
3591#define MOUNT_SETATTR_VALID_FLAGS (FSMOUNT_VALID_FLAGS | MOUNT_ATTR_IDMAP)
3592
3593#define MOUNT_SETATTR_PROPAGATION_FLAGS \
3594        (MS_UNBINDABLE | MS_PRIVATE | MS_SLAVE | MS_SHARED)
3595
3596static unsigned int attr_flags_to_mnt_flags(u64 attr_flags)
3597{
3598        unsigned int mnt_flags = 0;
3599
3600        if (attr_flags & MOUNT_ATTR_RDONLY)
3601                mnt_flags |= MNT_READONLY;
3602        if (attr_flags & MOUNT_ATTR_NOSUID)
3603                mnt_flags |= MNT_NOSUID;
3604        if (attr_flags & MOUNT_ATTR_NODEV)
3605                mnt_flags |= MNT_NODEV;
3606        if (attr_flags & MOUNT_ATTR_NOEXEC)
3607                mnt_flags |= MNT_NOEXEC;
3608        if (attr_flags & MOUNT_ATTR_NODIRATIME)
3609                mnt_flags |= MNT_NODIRATIME;
3610        if (attr_flags & MOUNT_ATTR_NOSYMFOLLOW)
3611                mnt_flags |= MNT_NOSYMFOLLOW;
3612
3613        return mnt_flags;
3614}
3615
3616/*
3617 * Create a kernel mount representation for a new, prepared superblock
3618 * (specified by fs_fd) and attach to an open_tree-like file descriptor.
3619 */
3620SYSCALL_DEFINE3(fsmount, int, fs_fd, unsigned int, flags,
3621                unsigned int, attr_flags)
3622{
3623        struct mnt_namespace *ns;
3624        struct fs_context *fc;
3625        struct file *file;
3626        struct path newmount;
3627        struct mount *mnt;
3628        struct fd f;
3629        unsigned int mnt_flags = 0;
3630        long ret;
3631
3632        if (!may_mount())
3633                return -EPERM;
3634
3635        if ((flags & ~(FSMOUNT_CLOEXEC)) != 0)
3636                return -EINVAL;
3637
3638        if (attr_flags & ~FSMOUNT_VALID_FLAGS)
3639                return -EINVAL;
3640
3641        mnt_flags = attr_flags_to_mnt_flags(attr_flags);
3642
3643        switch (attr_flags & MOUNT_ATTR__ATIME) {
3644        case MOUNT_ATTR_STRICTATIME:
3645                break;
3646        case MOUNT_ATTR_NOATIME:
3647                mnt_flags |= MNT_NOATIME;
3648                break;
3649        case MOUNT_ATTR_RELATIME:
3650                mnt_flags |= MNT_RELATIME;
3651                break;
3652        default:
3653                return -EINVAL;
3654        }
3655
3656        f = fdget(fs_fd);
3657        if (!f.file)
3658                return -EBADF;
3659
3660        ret = -EINVAL;
3661        if (f.file->f_op != &fscontext_fops)
3662                goto err_fsfd;
3663
3664        fc = f.file->private_data;
3665
3666        ret = mutex_lock_interruptible(&fc->uapi_mutex);
3667        if (ret < 0)
3668                goto err_fsfd;
3669
3670        /* There must be a valid superblock or we can't mount it */
3671        ret = -EINVAL;
3672        if (!fc->root)
3673                goto err_unlock;
3674
3675        ret = -EPERM;
3676        if (mount_too_revealing(fc->root->d_sb, &mnt_flags)) {
3677                pr_warn("VFS: Mount too revealing\n");
3678                goto err_unlock;
3679        }
3680
3681        ret = -EBUSY;
3682        if (fc->phase != FS_CONTEXT_AWAITING_MOUNT)
3683                goto err_unlock;
3684
3685        if (fc->sb_flags & SB_MANDLOCK)
3686                warn_mandlock();
3687
3688        newmount.mnt = vfs_create_mount(fc);
3689        if (IS_ERR(newmount.mnt)) {
3690                ret = PTR_ERR(newmount.mnt);
3691                goto err_unlock;
3692        }
3693        newmount.dentry = dget(fc->root);
3694        newmount.mnt->mnt_flags = mnt_flags;
3695
3696        /* We've done the mount bit - now move the file context into more or
3697         * less the same state as if we'd done an fspick().  We don't want to
3698         * do any memory allocation or anything like that at this point as we
3699         * don't want to have to handle any errors incurred.
3700         */
3701        vfs_clean_context(fc);
3702
3703        ns = alloc_mnt_ns(current->nsproxy->mnt_ns->user_ns, true);
3704        if (IS_ERR(ns)) {
3705                ret = PTR_ERR(ns);
3706                goto err_path;
3707        }
3708        mnt = real_mount(newmount.mnt);
3709        mnt->mnt_ns = ns;
3710        ns->root = mnt;
3711        ns->mounts = 1;
3712        list_add(&mnt->mnt_list, &ns->list);
3713        mntget(newmount.mnt);
3714
3715        /* Attach to an apparent O_PATH fd with a note that we need to unmount
3716         * it, not just simply put it.
3717         */
3718        file = dentry_open(&newmount, O_PATH, fc->cred);
3719        if (IS_ERR(file)) {
3720                dissolve_on_fput(newmount.mnt);
3721                ret = PTR_ERR(file);
3722                goto err_path;
3723        }
3724        file->f_mode |= FMODE_NEED_UNMOUNT;
3725
3726        ret = get_unused_fd_flags((flags & FSMOUNT_CLOEXEC) ? O_CLOEXEC : 0);
3727        if (ret >= 0)
3728                fd_install(ret, file);
3729        else
3730                fput(file);
3731
3732err_path:
3733        path_put(&newmount);
3734err_unlock:
3735        mutex_unlock(&fc->uapi_mutex);
3736err_fsfd:
3737        fdput(f);
3738        return ret;
3739}
3740
3741/*
3742 * Move a mount from one place to another.  In combination with
3743 * fsopen()/fsmount() this is used to install a new mount and in combination
3744 * with open_tree(OPEN_TREE_CLONE [| AT_RECURSIVE]) it can be used to copy
3745 * a mount subtree.
3746 *
3747 * Note the flags value is a combination of MOVE_MOUNT_* flags.
3748 */
3749SYSCALL_DEFINE5(move_mount,
3750                int, from_dfd, const char __user *, from_pathname,
3751                int, to_dfd, const char __user *, to_pathname,
3752                unsigned int, flags)
3753{
3754        struct path from_path, to_path;
3755        unsigned int lflags;
3756        int ret = 0;
3757
3758        if (!may_mount())
3759                return -EPERM;
3760
3761        if (flags & ~MOVE_MOUNT__MASK)
3762                return -EINVAL;
3763
3764        /* If someone gives a pathname, they aren't permitted to move
3765         * from an fd that requires unmount as we can't get at the flag
3766         * to clear it afterwards.
3767         */
3768        lflags = 0;
3769        if (flags & MOVE_MOUNT_F_SYMLINKS)      lflags |= LOOKUP_FOLLOW;
3770        if (flags & MOVE_MOUNT_F_AUTOMOUNTS)    lflags |= LOOKUP_AUTOMOUNT;
3771        if (flags & MOVE_MOUNT_F_EMPTY_PATH)    lflags |= LOOKUP_EMPTY;
3772
3773        ret = user_path_at(from_dfd, from_pathname, lflags, &from_path);
3774        if (ret < 0)
3775                return ret;
3776
3777        lflags = 0;
3778        if (flags & MOVE_MOUNT_T_SYMLINKS)      lflags |= LOOKUP_FOLLOW;
3779        if (flags & MOVE_MOUNT_T_AUTOMOUNTS)    lflags |= LOOKUP_AUTOMOUNT;
3780        if (flags & MOVE_MOUNT_T_EMPTY_PATH)    lflags |= LOOKUP_EMPTY;
3781
3782        ret = user_path_at(to_dfd, to_pathname, lflags, &to_path);
3783        if (ret < 0)
3784                goto out_from;
3785
3786        ret = security_move_mount(&from_path, &to_path);
3787        if (ret < 0)
3788                goto out_to;
3789
3790        if (flags & MOVE_MOUNT_SET_GROUP)
3791                ret = do_set_group(&from_path, &to_path);
3792        else
3793                ret = do_move_mount(&from_path, &to_path);
3794
3795out_to:
3796        path_put(&to_path);
3797out_from:
3798        path_put(&from_path);
3799        return ret;
3800}
3801
3802/*
3803 * Return true if path is reachable from root
3804 *
3805 * namespace_sem or mount_lock is held
3806 */
3807bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
3808                         const struct path *root)
3809{
3810        while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
3811                dentry = mnt->mnt_mountpoint;
3812                mnt = mnt->mnt_parent;
3813        }
3814        return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
3815}
3816
3817bool path_is_under(const struct path *path1, const struct path *path2)
3818{
3819        bool res;
3820        read_seqlock_excl(&mount_lock);
3821        res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
3822        read_sequnlock_excl(&mount_lock);
3823        return res;
3824}
3825EXPORT_SYMBOL(path_is_under);
3826
3827/*
3828 * pivot_root Semantics:
3829 * Moves the root file system of the current process to the directory put_old,
3830 * makes new_root as the new root file system of the current process, and sets
3831 * root/cwd of all processes which had them on the current root to new_root.
3832 *
3833 * Restrictions:
3834 * The new_root and put_old must be directories, and  must not be on the
3835 * same file  system as the current process root. The put_old  must  be
3836 * underneath new_root,  i.e. adding a non-zero number of /.. to the string
3837 * pointed to by put_old must yield the same directory as new_root. No other
3838 * file system may be mounted on put_old. After all, new_root is a mountpoint.
3839 *
3840 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
3841 * See Documentation/filesystems/ramfs-rootfs-initramfs.rst for alternatives
3842 * in this situation.
3843 *
3844 * Notes:
3845 *  - we don't move root/cwd if they are not at the root (reason: if something
3846 *    cared enough to change them, it's probably wrong to force them elsewhere)
3847 *  - it's okay to pick a root that isn't the root of a file system, e.g.
3848 *    /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
3849 *    though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
3850 *    first.
3851 */
3852SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
3853                const char __user *, put_old)
3854{
3855        struct path new, old, root;
3856        struct mount *new_mnt, *root_mnt, *old_mnt, *root_parent, *ex_parent;
3857        struct mountpoint *old_mp, *root_mp;
3858        int error;
3859
3860        if (!may_mount())
3861                return -EPERM;
3862
3863        error = user_path_at(AT_FDCWD, new_root,
3864                             LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &new);
3865        if (error)
3866                goto out0;
3867
3868        error = user_path_at(AT_FDCWD, put_old,
3869                             LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old);
3870        if (error)
3871                goto out1;
3872
3873        error = security_sb_pivotroot(&old, &new);
3874        if (error)
3875                goto out2;
3876
3877        get_fs_root(current->fs, &root);
3878        old_mp = lock_mount(&old);
3879        error = PTR_ERR(old_mp);
3880        if (IS_ERR(old_mp))
3881                goto out3;
3882
3883        error = -EINVAL;
3884        new_mnt = real_mount(new.mnt);
3885        root_mnt = real_mount(root.mnt);
3886        old_mnt = real_mount(old.mnt);
3887        ex_parent = new_mnt->mnt_parent;
3888        root_parent = root_mnt->mnt_parent;
3889        if (IS_MNT_SHARED(old_mnt) ||
3890                IS_MNT_SHARED(ex_parent) ||
3891                IS_MNT_SHARED(root_parent))
3892                goto out4;
3893        if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
3894                goto out4;
3895        if (new_mnt->mnt.mnt_flags & MNT_LOCKED)
3896                goto out4;
3897        error = -ENOENT;
3898        if (d_unlinked(new.dentry))
3899                goto out4;
3900        error = -EBUSY;
3901        if (new_mnt == root_mnt || old_mnt == root_mnt)
3902                goto out4; /* loop, on the same file system  */
3903        error = -EINVAL;
3904        if (root.mnt->mnt_root != root.dentry)
3905                goto out4; /* not a mountpoint */
3906        if (!mnt_has_parent(root_mnt))
3907                goto out4; /* not attached */
3908        if (new.mnt->mnt_root != new.dentry)
3909                goto out4; /* not a mountpoint */
3910        if (!mnt_has_parent(new_mnt))
3911                goto out4; /* not attached */
3912        /* make sure we can reach put_old from new_root */
3913        if (!is_path_reachable(old_mnt, old.dentry, &new))
3914                goto out4;
3915        /* make certain new is below the root */
3916        if (!is_path_reachable(new_mnt, new.dentry, &root))
3917                goto out4;
3918        lock_mount_hash();
3919        umount_mnt(new_mnt);
3920        root_mp = unhash_mnt(root_mnt);  /* we'll need its mountpoint */
3921        if (root_mnt->mnt.mnt_flags & MNT_LOCKED) {
3922                new_mnt->mnt.mnt_flags |= MNT_LOCKED;
3923                root_mnt->mnt.mnt_flags &= ~MNT_LOCKED;
3924        }
3925        /* mount old root on put_old */
3926        attach_mnt(root_mnt, old_mnt, old_mp);
3927        /* mount new_root on / */
3928        attach_mnt(new_mnt, root_parent, root_mp);
3929        mnt_add_count(root_parent, -1);
3930        touch_mnt_namespace(current->nsproxy->mnt_ns);
3931        /* A moved mount should not expire automatically */
3932        list_del_init(&new_mnt->mnt_expire);
3933        put_mountpoint(root_mp);
3934        unlock_mount_hash();
3935        chroot_fs_refs(&root, &new);
3936        error = 0;
3937out4:
3938        unlock_mount(old_mp);
3939        if (!error)
3940                mntput_no_expire(ex_parent);
3941out3:
3942        path_put(&root);
3943out2:
3944        path_put(&old);
3945out1:
3946        path_put(&new);
3947out0:
3948        return error;
3949}
3950
3951static unsigned int recalc_flags(struct mount_kattr *kattr, struct mount *mnt)
3952{
3953        unsigned int flags = mnt->mnt.mnt_flags;
3954
3955        /*  flags to clear */
3956        flags &= ~kattr->attr_clr;
3957        /* flags to raise */
3958        flags |= kattr->attr_set;
3959
3960        return flags;
3961}
3962
3963static int can_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
3964{
3965        struct vfsmount *m = &mnt->mnt;
3966        struct user_namespace *fs_userns = m->mnt_sb->s_user_ns;
3967
3968        if (!kattr->mnt_userns)
3969                return 0;
3970
3971        /*
3972         * Creating an idmapped mount with the filesystem wide idmapping
3973         * doesn't make sense so block that. We don't allow mushy semantics.
3974         */
3975        if (kattr->mnt_userns == fs_userns)
3976                return -EINVAL;
3977
3978        /*
3979         * Once a mount has been idmapped we don't allow it to change its
3980         * mapping. It makes things simpler and callers can just create
3981         * another bind-mount they can idmap if they want to.
3982         */
3983        if (is_idmapped_mnt(m))
3984                return -EPERM;
3985
3986        /* The underlying filesystem doesn't support idmapped mounts yet. */
3987        if (!(m->mnt_sb->s_type->fs_flags & FS_ALLOW_IDMAP))
3988                return -EINVAL;
3989
3990        /* We're not controlling the superblock. */
3991        if (!ns_capable(fs_userns, CAP_SYS_ADMIN))
3992                return -EPERM;
3993
3994        /* Mount has already been visible in the filesystem hierarchy. */
3995        if (!is_anon_ns(mnt->mnt_ns))
3996                return -EINVAL;
3997
3998        return 0;
3999}
4000
4001static struct mount *mount_setattr_prepare(struct mount_kattr *kattr,
4002                                           struct mount *mnt, int *err)
4003{
4004        struct mount *m = mnt, *last = NULL;
4005
4006        if (!is_mounted(&m->mnt)) {
4007                *err = -EINVAL;
4008                goto out;
4009        }
4010
4011        if (!(mnt_has_parent(m) ? check_mnt(m) : is_anon_ns(m->mnt_ns))) {
4012                *err = -EINVAL;
4013                goto out;
4014        }
4015
4016        do {
4017                unsigned int flags;
4018
4019                flags = recalc_flags(kattr, m);
4020                if (!can_change_locked_flags(m, flags)) {
4021                        *err = -EPERM;
4022                        goto out;
4023                }
4024
4025                *err = can_idmap_mount(kattr, m);
4026                if (*err)
4027                        goto out;
4028
4029                last = m;
4030
4031                if ((kattr->attr_set & MNT_READONLY) &&
4032                    !(m->mnt.mnt_flags & MNT_READONLY)) {
4033                        *err = mnt_hold_writers(m);
4034                        if (*err)
4035                                goto out;
4036                }
4037        } while (kattr->recurse && (m = next_mnt(m, mnt)));
4038
4039out:
4040        return last;
4041}
4042
4043static void do_idmap_mount(const struct mount_kattr *kattr, struct mount *mnt)
4044{
4045        struct user_namespace *mnt_userns, *old_mnt_userns;
4046
4047        if (!kattr->mnt_userns)
4048                return;
4049
4050        /*
4051         * We're the only ones able to change the mount's idmapping. So
4052         * mnt->mnt.mnt_userns is stable and we can retrieve it directly.
4053         */
4054        old_mnt_userns = mnt->mnt.mnt_userns;
4055
4056        mnt_userns = get_user_ns(kattr->mnt_userns);
4057        /* Pairs with smp_load_acquire() in mnt_user_ns(). */
4058        smp_store_release(&mnt->mnt.mnt_userns, mnt_userns);
4059
4060        /*
4061         * If this is an idmapped filesystem drop the reference we've taken
4062         * in vfs_create_mount() before.
4063         */
4064        if (!initial_idmapping(old_mnt_userns))
4065                put_user_ns(old_mnt_userns);
4066}
4067
4068static void mount_setattr_commit(struct mount_kattr *kattr,
4069                                 struct mount *mnt, struct mount *last,
4070                                 int err)
4071{
4072        struct mount *m = mnt;
4073
4074        do {
4075                if (!err) {
4076                        unsigned int flags;
4077
4078                        do_idmap_mount(kattr, m);
4079                        flags = recalc_flags(kattr, m);
4080                        WRITE_ONCE(m->mnt.mnt_flags, flags);
4081                }
4082
4083                /*
4084                 * We either set MNT_READONLY above so make it visible
4085                 * before ~MNT_WRITE_HOLD or we failed to recursively
4086                 * apply mount options.
4087                 */
4088                if ((kattr->attr_set & MNT_READONLY) &&
4089                    (m->mnt.mnt_flags & MNT_WRITE_HOLD))
4090                        mnt_unhold_writers(m);
4091
4092                if (!err && kattr->propagation)
4093                        change_mnt_propagation(m, kattr->propagation);
4094
4095                /*
4096                 * On failure, only cleanup until we found the first mount
4097                 * we failed to handle.
4098                 */
4099                if (err && m == last)
4100                        break;
4101        } while (kattr->recurse && (m = next_mnt(m, mnt)));
4102
4103        if (!err)
4104                touch_mnt_namespace(mnt->mnt_ns);
4105}
4106
4107static int do_mount_setattr(struct path *path, struct mount_kattr *kattr)
4108{
4109        struct mount *mnt = real_mount(path->mnt), *last = NULL;
4110        int err = 0;
4111
4112        if (path->dentry != mnt->mnt.mnt_root)
4113                return -EINVAL;
4114
4115        if (kattr->propagation) {
4116                /*
4117                 * Only take namespace_lock() if we're actually changing
4118                 * propagation.
4119                 */
4120                namespace_lock();
4121                if (kattr->propagation == MS_SHARED) {
4122                        err = invent_group_ids(mnt, kattr->recurse);
4123                        if (err) {
4124                                namespace_unlock();
4125                                return err;
4126                        }
4127                }
4128        }
4129
4130        lock_mount_hash();
4131
4132        /*
4133         * Get the mount tree in a shape where we can change mount
4134         * properties without failure.
4135         */
4136        last = mount_setattr_prepare(kattr, mnt, &err);
4137        if (last) /* Commit all changes or revert to the old state. */
4138                mount_setattr_commit(kattr, mnt, last, err);
4139
4140        unlock_mount_hash();
4141
4142        if (kattr->propagation) {
4143                namespace_unlock();
4144                if (err)
4145                        cleanup_group_ids(mnt, NULL);
4146        }
4147
4148        return err;
4149}
4150
4151static int build_mount_idmapped(const struct mount_attr *attr, size_t usize,
4152                                struct mount_kattr *kattr, unsigned int flags)
4153{
4154        int err = 0;
4155        struct ns_common *ns;
4156        struct user_namespace *mnt_userns;
4157        struct file *file;
4158
4159        if (!((attr->attr_set | attr->attr_clr) & MOUNT_ATTR_IDMAP))
4160                return 0;
4161
4162        /*
4163         * We currently do not support clearing an idmapped mount. If this ever
4164         * is a use-case we can revisit this but for now let's keep it simple
4165         * and not allow it.
4166         */
4167        if (attr->attr_clr & MOUNT_ATTR_IDMAP)
4168                return -EINVAL;
4169
4170        if (attr->userns_fd > INT_MAX)
4171                return -EINVAL;
4172
4173        file = fget(attr->userns_fd);
4174        if (!file)
4175                return -EBADF;
4176
4177        if (!proc_ns_file(file)) {
4178                err = -EINVAL;
4179                goto out_fput;
4180        }
4181
4182        ns = get_proc_ns(file_inode(file));
4183        if (ns->ops->type != CLONE_NEWUSER) {
4184                err = -EINVAL;
4185                goto out_fput;
4186        }
4187
4188        /*
4189         * The initial idmapping cannot be used to create an idmapped
4190         * mount. We use the initial idmapping as an indicator of a mount
4191         * that is not idmapped. It can simply be passed into helpers that
4192         * are aware of idmapped mounts as a convenient shortcut. A user
4193         * can just create a dedicated identity mapping to achieve the same
4194         * result.
4195         */
4196        mnt_userns = container_of(ns, struct user_namespace, ns);
4197        if (initial_idmapping(mnt_userns)) {
4198                err = -EPERM;
4199                goto out_fput;
4200        }
4201        kattr->mnt_userns = get_user_ns(mnt_userns);
4202
4203out_fput:
4204        fput(file);
4205        return err;
4206}
4207
4208static int build_mount_kattr(const struct mount_attr *attr, size_t usize,
4209                             struct mount_kattr *kattr, unsigned int flags)
4210{
4211        unsigned int lookup_flags = LOOKUP_AUTOMOUNT | LOOKUP_FOLLOW;
4212
4213        if (flags & AT_NO_AUTOMOUNT)
4214                lookup_flags &= ~LOOKUP_AUTOMOUNT;
4215        if (flags & AT_SYMLINK_NOFOLLOW)
4216                lookup_flags &= ~LOOKUP_FOLLOW;
4217        if (flags & AT_EMPTY_PATH)
4218                lookup_flags |= LOOKUP_EMPTY;
4219
4220        *kattr = (struct mount_kattr) {
4221                .lookup_flags   = lookup_flags,
4222                .recurse        = !!(flags & AT_RECURSIVE),
4223        };
4224
4225        if (attr->propagation & ~MOUNT_SETATTR_PROPAGATION_FLAGS)
4226                return -EINVAL;
4227        if (hweight32(attr->propagation & MOUNT_SETATTR_PROPAGATION_FLAGS) > 1)
4228                return -EINVAL;
4229        kattr->propagation = attr->propagation;
4230
4231        if ((attr->attr_set | attr->attr_clr) & ~MOUNT_SETATTR_VALID_FLAGS)
4232                return -EINVAL;
4233
4234        kattr->attr_set = attr_flags_to_mnt_flags(attr->attr_set);
4235        kattr->attr_clr = attr_flags_to_mnt_flags(attr->attr_clr);
4236
4237        /*
4238         * Since the MOUNT_ATTR_<atime> values are an enum, not a bitmap,
4239         * users wanting to transition to a different atime setting cannot
4240         * simply specify the atime setting in @attr_set, but must also
4241         * specify MOUNT_ATTR__ATIME in the @attr_clr field.
4242         * So ensure that MOUNT_ATTR__ATIME can't be partially set in
4243         * @attr_clr and that @attr_set can't have any atime bits set if
4244         * MOUNT_ATTR__ATIME isn't set in @attr_clr.
4245         */
4246        if (attr->attr_clr & MOUNT_ATTR__ATIME) {
4247                if ((attr->attr_clr & MOUNT_ATTR__ATIME) != MOUNT_ATTR__ATIME)
4248                        return -EINVAL;
4249
4250                /*
4251                 * Clear all previous time settings as they are mutually
4252                 * exclusive.
4253                 */
4254                kattr->attr_clr |= MNT_RELATIME | MNT_NOATIME;
4255                switch (attr->attr_set & MOUNT_ATTR__ATIME) {
4256                case MOUNT_ATTR_RELATIME:
4257                        kattr->attr_set |= MNT_RELATIME;
4258                        break;
4259                case MOUNT_ATTR_NOATIME:
4260                        kattr->attr_set |= MNT_NOATIME;
4261                        break;
4262                case MOUNT_ATTR_STRICTATIME:
4263                        break;
4264                default:
4265                        return -EINVAL;
4266                }
4267        } else {
4268                if (attr->attr_set & MOUNT_ATTR__ATIME)
4269                        return -EINVAL;
4270        }
4271
4272        return build_mount_idmapped(attr, usize, kattr, flags);
4273}
4274
4275static void finish_mount_kattr(struct mount_kattr *kattr)
4276{
4277        put_user_ns(kattr->mnt_userns);
4278        kattr->mnt_userns = NULL;
4279}
4280
4281SYSCALL_DEFINE5(mount_setattr, int, dfd, const char __user *, path,
4282                unsigned int, flags, struct mount_attr __user *, uattr,
4283                size_t, usize)
4284{
4285        int err;
4286        struct path target;
4287        struct mount_attr attr;
4288        struct mount_kattr kattr;
4289
4290        BUILD_BUG_ON(sizeof(struct mount_attr) != MOUNT_ATTR_SIZE_VER0);
4291
4292        if (flags & ~(AT_EMPTY_PATH |
4293                      AT_RECURSIVE |
4294                      AT_SYMLINK_NOFOLLOW |
4295                      AT_NO_AUTOMOUNT))
4296                return -EINVAL;
4297
4298        if (unlikely(usize > PAGE_SIZE))
4299                return -E2BIG;
4300        if (unlikely(usize < MOUNT_ATTR_SIZE_VER0))
4301                return -EINVAL;
4302
4303        if (!may_mount())
4304                return -EPERM;
4305
4306        err = copy_struct_from_user(&attr, sizeof(attr), uattr, usize);
4307        if (err)
4308                return err;
4309
4310        /* Don't bother walking through the mounts if this is a nop. */
4311        if (attr.attr_set == 0 &&
4312            attr.attr_clr == 0 &&
4313            attr.propagation == 0)
4314                return 0;
4315
4316        err = build_mount_kattr(&attr, usize, &kattr, flags);
4317        if (err)
4318                return err;
4319
4320        err = user_path_at(dfd, path, kattr.lookup_flags, &target);
4321        if (!err) {
4322                err = do_mount_setattr(&target, &kattr);
4323                path_put(&target);
4324        }
4325        finish_mount_kattr(&kattr);
4326        return err;
4327}
4328
4329static void __init init_mount_tree(void)
4330{
4331        struct vfsmount *mnt;
4332        struct mount *m;
4333        struct mnt_namespace *ns;
4334        struct path root;
4335
4336        mnt = vfs_kern_mount(&rootfs_fs_type, 0, "rootfs", NULL);
4337        if (IS_ERR(mnt))
4338                panic("Can't create rootfs");
4339
4340        ns = alloc_mnt_ns(&init_user_ns, false);
4341        if (IS_ERR(ns))
4342                panic("Can't allocate initial namespace");
4343        m = real_mount(mnt);
4344        m->mnt_ns = ns;
4345        ns->root = m;
4346        ns->mounts = 1;
4347        list_add(&m->mnt_list, &ns->list);
4348        init_task.nsproxy->mnt_ns = ns;
4349        get_mnt_ns(ns);
4350
4351        root.mnt = mnt;
4352        root.dentry = mnt->mnt_root;
4353        mnt->mnt_flags |= MNT_LOCKED;
4354
4355        set_fs_pwd(current->fs, &root);
4356        set_fs_root(current->fs, &root);
4357}
4358
4359void __init mnt_init(void)
4360{
4361        int err;
4362
4363        mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
4364                        0, SLAB_HWCACHE_ALIGN|SLAB_PANIC|SLAB_ACCOUNT, NULL);
4365
4366        mount_hashtable = alloc_large_system_hash("Mount-cache",
4367                                sizeof(struct hlist_head),
4368                                mhash_entries, 19,
4369                                HASH_ZERO,
4370                                &m_hash_shift, &m_hash_mask, 0, 0);
4371        mountpoint_hashtable = alloc_large_system_hash("Mountpoint-cache",
4372                                sizeof(struct hlist_head),
4373                                mphash_entries, 19,
4374                                HASH_ZERO,
4375                                &mp_hash_shift, &mp_hash_mask, 0, 0);
4376
4377        if (!mount_hashtable || !mountpoint_hashtable)
4378                panic("Failed to allocate mount hash table\n");
4379
4380        kernfs_init();
4381
4382        err = sysfs_init();
4383        if (err)
4384                printk(KERN_WARNING "%s: sysfs_init error: %d\n",
4385                        __func__, err);
4386        fs_kobj = kobject_create_and_add("fs", NULL);
4387        if (!fs_kobj)
4388                printk(KERN_WARNING "%s: kobj create error\n", __func__);
4389        shmem_init();
4390        init_rootfs();
4391        init_mount_tree();
4392}
4393
4394void put_mnt_ns(struct mnt_namespace *ns)
4395{
4396        if (!refcount_dec_and_test(&ns->ns.count))
4397                return;
4398        drop_collected_mounts(&ns->root->mnt);
4399        free_mnt_ns(ns);
4400}
4401
4402struct vfsmount *kern_mount(struct file_system_type *type)
4403{
4404        struct vfsmount *mnt;
4405        mnt = vfs_kern_mount(type, SB_KERNMOUNT, type->name, NULL);
4406        if (!IS_ERR(mnt)) {
4407                /*
4408                 * it is a longterm mount, don't release mnt until
4409                 * we unmount before file sys is unregistered
4410                */
4411                real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
4412        }
4413        return mnt;
4414}
4415EXPORT_SYMBOL_GPL(kern_mount);
4416
4417void kern_unmount(struct vfsmount *mnt)
4418{
4419        /* release long term mount so mount point can be released */
4420        if (!IS_ERR_OR_NULL(mnt)) {
4421                real_mount(mnt)->mnt_ns = NULL;
4422                synchronize_rcu();      /* yecchhh... */
4423                mntput(mnt);
4424        }
4425}
4426EXPORT_SYMBOL(kern_unmount);
4427
4428void kern_unmount_array(struct vfsmount *mnt[], unsigned int num)
4429{
4430        unsigned int i;
4431
4432        for (i = 0; i < num; i++)
4433                if (mnt[i])
4434                        real_mount(mnt[i])->mnt_ns = NULL;
4435        synchronize_rcu_expedited();
4436        for (i = 0; i < num; i++)
4437                mntput(mnt[i]);
4438}
4439EXPORT_SYMBOL(kern_unmount_array);
4440
4441bool our_mnt(struct vfsmount *mnt)
4442{
4443        return check_mnt(real_mount(mnt));
4444}
4445
4446bool current_chrooted(void)
4447{
4448        /* Does the current process have a non-standard root */
4449        struct path ns_root;
4450        struct path fs_root;
4451        bool chrooted;
4452
4453        /* Find the namespace root */
4454        ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
4455        ns_root.dentry = ns_root.mnt->mnt_root;
4456        path_get(&ns_root);
4457        while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
4458                ;
4459
4460        get_fs_root(current->fs, &fs_root);
4461
4462        chrooted = !path_equal(&fs_root, &ns_root);
4463
4464        path_put(&fs_root);
4465        path_put(&ns_root);
4466
4467        return chrooted;
4468}
4469
4470static bool mnt_already_visible(struct mnt_namespace *ns,
4471                                const struct super_block *sb,
4472                                int *new_mnt_flags)
4473{
4474        int new_flags = *new_mnt_flags;
4475        struct mount *mnt;
4476        bool visible = false;
4477
4478        down_read(&namespace_sem);
4479        lock_ns_list(ns);
4480        list_for_each_entry(mnt, &ns->list, mnt_list) {
4481                struct mount *child;
4482                int mnt_flags;
4483
4484                if (mnt_is_cursor(mnt))
4485                        continue;
4486
4487                if (mnt->mnt.mnt_sb->s_type != sb->s_type)
4488                        continue;
4489
4490                /* This mount is not fully visible if it's root directory
4491                 * is not the root directory of the filesystem.
4492                 */
4493                if (mnt->mnt.mnt_root != mnt->mnt.mnt_sb->s_root)
4494                        continue;
4495
4496                /* A local view of the mount flags */
4497                mnt_flags = mnt->mnt.mnt_flags;
4498
4499                /* Don't miss readonly hidden in the superblock flags */
4500                if (sb_rdonly(mnt->mnt.mnt_sb))
4501                        mnt_flags |= MNT_LOCK_READONLY;
4502
4503                /* Verify the mount flags are equal to or more permissive
4504                 * than the proposed new mount.
4505                 */
4506                if ((mnt_flags & MNT_LOCK_READONLY) &&
4507                    !(new_flags & MNT_READONLY))
4508                        continue;
4509                if ((mnt_flags & MNT_LOCK_ATIME) &&
4510                    ((mnt_flags & MNT_ATIME_MASK) != (new_flags & MNT_ATIME_MASK)))
4511                        continue;
4512
4513                /* This mount is not fully visible if there are any
4514                 * locked child mounts that cover anything except for
4515                 * empty directories.
4516                 */
4517                list_for_each_entry(child, &mnt->mnt_mounts, mnt_child) {
4518                        struct inode *inode = child->mnt_mountpoint->d_inode;
4519                        /* Only worry about locked mounts */
4520                        if (!(child->mnt.mnt_flags & MNT_LOCKED))
4521                                continue;
4522                        /* Is the directory permanetly empty? */
4523                        if (!is_empty_dir_inode(inode))
4524                                goto next;
4525                }
4526                /* Preserve the locked attributes */
4527                *new_mnt_flags |= mnt_flags & (MNT_LOCK_READONLY | \
4528                                               MNT_LOCK_ATIME);
4529                visible = true;
4530                goto found;
4531        next:   ;
4532        }
4533found:
4534        unlock_ns_list(ns);
4535        up_read(&namespace_sem);
4536        return visible;
4537}
4538
4539static bool mount_too_revealing(const struct super_block *sb, int *new_mnt_flags)
4540{
4541        const unsigned long required_iflags = SB_I_NOEXEC | SB_I_NODEV;
4542        struct mnt_namespace *ns = current->nsproxy->mnt_ns;
4543        unsigned long s_iflags;
4544
4545        if (ns->user_ns == &init_user_ns)
4546                return false;
4547
4548        /* Can this filesystem be too revealing? */
4549        s_iflags = sb->s_iflags;
4550        if (!(s_iflags & SB_I_USERNS_VISIBLE))
4551                return false;
4552
4553        if ((s_iflags & required_iflags) != required_iflags) {
4554                WARN_ONCE(1, "Expected s_iflags to contain 0x%lx\n",
4555                          required_iflags);
4556                return true;
4557        }
4558
4559        return !mnt_already_visible(ns, sb, new_mnt_flags);
4560}
4561
4562bool mnt_may_suid(struct vfsmount *mnt)
4563{
4564        /*
4565         * Foreign mounts (accessed via fchdir or through /proc
4566         * symlinks) are always treated as if they are nosuid.  This
4567         * prevents namespaces from trusting potentially unsafe
4568         * suid/sgid bits, file caps, or security labels that originate
4569         * in other namespaces.
4570         */
4571        return !(mnt->mnt_flags & MNT_NOSUID) && check_mnt(real_mount(mnt)) &&
4572               current_in_userns(mnt->mnt_sb->s_user_ns);
4573}
4574
4575static struct ns_common *mntns_get(struct task_struct *task)
4576{
4577        struct ns_common *ns = NULL;
4578        struct nsproxy *nsproxy;
4579
4580        task_lock(task);
4581        nsproxy = task->nsproxy;
4582        if (nsproxy) {
4583                ns = &nsproxy->mnt_ns->ns;
4584                get_mnt_ns(to_mnt_ns(ns));
4585        }
4586        task_unlock(task);
4587
4588        return ns;
4589}
4590
4591static void mntns_put(struct ns_common *ns)
4592{
4593        put_mnt_ns(to_mnt_ns(ns));
4594}
4595
4596static int mntns_install(struct nsset *nsset, struct ns_common *ns)
4597{
4598        struct nsproxy *nsproxy = nsset->nsproxy;
4599        struct fs_struct *fs = nsset->fs;
4600        struct mnt_namespace *mnt_ns = to_mnt_ns(ns), *old_mnt_ns;
4601        struct user_namespace *user_ns = nsset->cred->user_ns;
4602        struct path root;
4603        int err;
4604
4605        if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
4606            !ns_capable(user_ns, CAP_SYS_CHROOT) ||
4607            !ns_capable(user_ns, CAP_SYS_ADMIN))
4608                return -EPERM;
4609
4610        if (is_anon_ns(mnt_ns))
4611                return -EINVAL;
4612
4613        if (fs->users != 1)
4614                return -EINVAL;
4615
4616        get_mnt_ns(mnt_ns);
4617        old_mnt_ns = nsproxy->mnt_ns;
4618        nsproxy->mnt_ns = mnt_ns;
4619
4620        /* Find the root */
4621        err = vfs_path_lookup(mnt_ns->root->mnt.mnt_root, &mnt_ns->root->mnt,
4622                                "/", LOOKUP_DOWN, &root);
4623        if (err) {
4624                /* revert to old namespace */
4625                nsproxy->mnt_ns = old_mnt_ns;
4626                put_mnt_ns(mnt_ns);
4627                return err;
4628        }
4629
4630        put_mnt_ns(old_mnt_ns);
4631
4632        /* Update the pwd and root */
4633        set_fs_pwd(fs, &root);
4634        set_fs_root(fs, &root);
4635
4636        path_put(&root);
4637        return 0;
4638}
4639
4640static struct user_namespace *mntns_owner(struct ns_common *ns)
4641{
4642        return to_mnt_ns(ns)->user_ns;
4643}
4644
4645const struct proc_ns_operations mntns_operations = {
4646        .name           = "mnt",
4647        .type           = CLONE_NEWNS,
4648        .get            = mntns_get,
4649        .put            = mntns_put,
4650        .install        = mntns_install,
4651        .owner          = mntns_owner,
4652};
4653
4654#ifdef CONFIG_SYSCTL
4655static struct ctl_table fs_namespace_sysctls[] = {
4656        {
4657                .procname       = "mount-max",
4658                .data           = &sysctl_mount_max,
4659                .maxlen         = sizeof(unsigned int),
4660                .mode           = 0644,
4661                .proc_handler   = proc_dointvec_minmax,
4662                .extra1         = SYSCTL_ONE,
4663        },
4664        { }
4665};
4666
4667static int __init init_fs_namespace_sysctls(void)
4668{
4669        register_sysctl_init("fs", fs_namespace_sysctls);
4670        return 0;
4671}
4672fs_initcall(init_fs_namespace_sysctls);
4673
4674#endif /* CONFIG_SYSCTL */
4675