linux/fs/namespace.c
<<
>>
Prefs
   1/*
   2 *  linux/fs/namespace.c
   3 *
   4 * (C) Copyright Al Viro 2000, 2001
   5 *      Released under GPL v2.
   6 *
   7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
   8 * Heavily rewritten.
   9 */
  10
  11#include <linux/syscalls.h>
  12#include <linux/slab.h>
  13#include <linux/sched.h>
  14#include <linux/smp_lock.h>
  15#include <linux/init.h>
  16#include <linux/kernel.h>
  17#include <linux/acct.h>
  18#include <linux/capability.h>
  19#include <linux/cpumask.h>
  20#include <linux/module.h>
  21#include <linux/sysfs.h>
  22#include <linux/seq_file.h>
  23#include <linux/mnt_namespace.h>
  24#include <linux/namei.h>
  25#include <linux/security.h>
  26#include <linux/mount.h>
  27#include <linux/ramfs.h>
  28#include <linux/log2.h>
  29#include <linux/idr.h>
  30#include <linux/fs_struct.h>
  31#include <asm/uaccess.h>
  32#include <asm/unistd.h>
  33#include "pnode.h"
  34#include "internal.h"
  35
  36#define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
  37#define HASH_SIZE (1UL << HASH_SHIFT)
  38
  39/* spinlock for vfsmount related operations, inplace of dcache_lock */
  40__cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
  41
  42static int event;
  43static DEFINE_IDA(mnt_id_ida);
  44static DEFINE_IDA(mnt_group_ida);
  45
  46static struct list_head *mount_hashtable __read_mostly;
  47static struct kmem_cache *mnt_cache __read_mostly;
  48static struct rw_semaphore namespace_sem;
  49
  50/* /sys/fs */
  51struct kobject *fs_kobj;
  52EXPORT_SYMBOL_GPL(fs_kobj);
  53
  54static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
  55{
  56        unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
  57        tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
  58        tmp = tmp + (tmp >> HASH_SHIFT);
  59        return tmp & (HASH_SIZE - 1);
  60}
  61
  62#define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
  63
  64/* allocation is serialized by namespace_sem */
  65static int mnt_alloc_id(struct vfsmount *mnt)
  66{
  67        int res;
  68
  69retry:
  70        ida_pre_get(&mnt_id_ida, GFP_KERNEL);
  71        spin_lock(&vfsmount_lock);
  72        res = ida_get_new(&mnt_id_ida, &mnt->mnt_id);
  73        spin_unlock(&vfsmount_lock);
  74        if (res == -EAGAIN)
  75                goto retry;
  76
  77        return res;
  78}
  79
  80static void mnt_free_id(struct vfsmount *mnt)
  81{
  82        spin_lock(&vfsmount_lock);
  83        ida_remove(&mnt_id_ida, mnt->mnt_id);
  84        spin_unlock(&vfsmount_lock);
  85}
  86
  87/*
  88 * Allocate a new peer group ID
  89 *
  90 * mnt_group_ida is protected by namespace_sem
  91 */
  92static int mnt_alloc_group_id(struct vfsmount *mnt)
  93{
  94        if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
  95                return -ENOMEM;
  96
  97        return ida_get_new_above(&mnt_group_ida, 1, &mnt->mnt_group_id);
  98}
  99
 100/*
 101 * Release a peer group ID
 102 */
 103void mnt_release_group_id(struct vfsmount *mnt)
 104{
 105        ida_remove(&mnt_group_ida, mnt->mnt_group_id);
 106        mnt->mnt_group_id = 0;
 107}
 108
 109struct vfsmount *alloc_vfsmnt(const char *name)
 110{
 111        struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
 112        if (mnt) {
 113                int err;
 114
 115                err = mnt_alloc_id(mnt);
 116                if (err)
 117                        goto out_free_cache;
 118
 119                if (name) {
 120                        mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
 121                        if (!mnt->mnt_devname)
 122                                goto out_free_id;
 123                }
 124
 125                atomic_set(&mnt->mnt_count, 1);
 126                INIT_LIST_HEAD(&mnt->mnt_hash);
 127                INIT_LIST_HEAD(&mnt->mnt_child);
 128                INIT_LIST_HEAD(&mnt->mnt_mounts);
 129                INIT_LIST_HEAD(&mnt->mnt_list);
 130                INIT_LIST_HEAD(&mnt->mnt_expire);
 131                INIT_LIST_HEAD(&mnt->mnt_share);
 132                INIT_LIST_HEAD(&mnt->mnt_slave_list);
 133                INIT_LIST_HEAD(&mnt->mnt_slave);
 134                atomic_set(&mnt->__mnt_writers, 0);
 135        }
 136        return mnt;
 137
 138out_free_id:
 139        mnt_free_id(mnt);
 140out_free_cache:
 141        kmem_cache_free(mnt_cache, mnt);
 142        return NULL;
 143}
 144
 145/*
 146 * Most r/o checks on a fs are for operations that take
 147 * discrete amounts of time, like a write() or unlink().
 148 * We must keep track of when those operations start
 149 * (for permission checks) and when they end, so that
 150 * we can determine when writes are able to occur to
 151 * a filesystem.
 152 */
 153/*
 154 * __mnt_is_readonly: check whether a mount is read-only
 155 * @mnt: the mount to check for its write status
 156 *
 157 * This shouldn't be used directly ouside of the VFS.
 158 * It does not guarantee that the filesystem will stay
 159 * r/w, just that it is right *now*.  This can not and
 160 * should not be used in place of IS_RDONLY(inode).
 161 * mnt_want/drop_write() will _keep_ the filesystem
 162 * r/w.
 163 */
 164int __mnt_is_readonly(struct vfsmount *mnt)
 165{
 166        if (mnt->mnt_flags & MNT_READONLY)
 167                return 1;
 168        if (mnt->mnt_sb->s_flags & MS_RDONLY)
 169                return 1;
 170        return 0;
 171}
 172EXPORT_SYMBOL_GPL(__mnt_is_readonly);
 173
 174struct mnt_writer {
 175        /*
 176         * If holding multiple instances of this lock, they
 177         * must be ordered by cpu number.
 178         */
 179        spinlock_t lock;
 180        struct lock_class_key lock_class; /* compiles out with !lockdep */
 181        unsigned long count;
 182        struct vfsmount *mnt;
 183} ____cacheline_aligned_in_smp;
 184static DEFINE_PER_CPU(struct mnt_writer, mnt_writers);
 185
 186static int __init init_mnt_writers(void)
 187{
 188        int cpu;
 189        for_each_possible_cpu(cpu) {
 190                struct mnt_writer *writer = &per_cpu(mnt_writers, cpu);
 191                spin_lock_init(&writer->lock);
 192                lockdep_set_class(&writer->lock, &writer->lock_class);
 193                writer->count = 0;
 194        }
 195        return 0;
 196}
 197fs_initcall(init_mnt_writers);
 198
 199static void unlock_mnt_writers(void)
 200{
 201        int cpu;
 202        struct mnt_writer *cpu_writer;
 203
 204        for_each_possible_cpu(cpu) {
 205                cpu_writer = &per_cpu(mnt_writers, cpu);
 206                spin_unlock(&cpu_writer->lock);
 207        }
 208}
 209
 210static inline void __clear_mnt_count(struct mnt_writer *cpu_writer)
 211{
 212        if (!cpu_writer->mnt)
 213                return;
 214        /*
 215         * This is in case anyone ever leaves an invalid,
 216         * old ->mnt and a count of 0.
 217         */
 218        if (!cpu_writer->count)
 219                return;
 220        atomic_add(cpu_writer->count, &cpu_writer->mnt->__mnt_writers);
 221        cpu_writer->count = 0;
 222}
 223 /*
 224 * must hold cpu_writer->lock
 225 */
 226static inline void use_cpu_writer_for_mount(struct mnt_writer *cpu_writer,
 227                                          struct vfsmount *mnt)
 228{
 229        if (cpu_writer->mnt == mnt)
 230                return;
 231        __clear_mnt_count(cpu_writer);
 232        cpu_writer->mnt = mnt;
 233}
 234
 235/*
 236 * Most r/o checks on a fs are for operations that take
 237 * discrete amounts of time, like a write() or unlink().
 238 * We must keep track of when those operations start
 239 * (for permission checks) and when they end, so that
 240 * we can determine when writes are able to occur to
 241 * a filesystem.
 242 */
 243/**
 244 * mnt_want_write - get write access to a mount
 245 * @mnt: the mount on which to take a write
 246 *
 247 * This tells the low-level filesystem that a write is
 248 * about to be performed to it, and makes sure that
 249 * writes are allowed before returning success.  When
 250 * the write operation is finished, mnt_drop_write()
 251 * must be called.  This is effectively a refcount.
 252 */
 253int mnt_want_write(struct vfsmount *mnt)
 254{
 255        int ret = 0;
 256        struct mnt_writer *cpu_writer;
 257
 258        cpu_writer = &get_cpu_var(mnt_writers);
 259        spin_lock(&cpu_writer->lock);
 260        if (__mnt_is_readonly(mnt)) {
 261                ret = -EROFS;
 262                goto out;
 263        }
 264        use_cpu_writer_for_mount(cpu_writer, mnt);
 265        cpu_writer->count++;
 266out:
 267        spin_unlock(&cpu_writer->lock);
 268        put_cpu_var(mnt_writers);
 269        return ret;
 270}
 271EXPORT_SYMBOL_GPL(mnt_want_write);
 272
 273static void lock_mnt_writers(void)
 274{
 275        int cpu;
 276        struct mnt_writer *cpu_writer;
 277
 278        for_each_possible_cpu(cpu) {
 279                cpu_writer = &per_cpu(mnt_writers, cpu);
 280                spin_lock(&cpu_writer->lock);
 281                __clear_mnt_count(cpu_writer);
 282                cpu_writer->mnt = NULL;
 283        }
 284}
 285
 286/*
 287 * These per-cpu write counts are not guaranteed to have
 288 * matched increments and decrements on any given cpu.
 289 * A file open()ed for write on one cpu and close()d on
 290 * another cpu will imbalance this count.  Make sure it
 291 * does not get too far out of whack.
 292 */
 293static void handle_write_count_underflow(struct vfsmount *mnt)
 294{
 295        if (atomic_read(&mnt->__mnt_writers) >=
 296            MNT_WRITER_UNDERFLOW_LIMIT)
 297                return;
 298        /*
 299         * It isn't necessary to hold all of the locks
 300         * at the same time, but doing it this way makes
 301         * us share a lot more code.
 302         */
 303        lock_mnt_writers();
 304        /*
 305         * vfsmount_lock is for mnt_flags.
 306         */
 307        spin_lock(&vfsmount_lock);
 308        /*
 309         * If coalescing the per-cpu writer counts did not
 310         * get us back to a positive writer count, we have
 311         * a bug.
 312         */
 313        if ((atomic_read(&mnt->__mnt_writers) < 0) &&
 314            !(mnt->mnt_flags & MNT_IMBALANCED_WRITE_COUNT)) {
 315                WARN(1, KERN_DEBUG "leak detected on mount(%p) writers "
 316                                "count: %d\n",
 317                        mnt, atomic_read(&mnt->__mnt_writers));
 318                /* use the flag to keep the dmesg spam down */
 319                mnt->mnt_flags |= MNT_IMBALANCED_WRITE_COUNT;
 320        }
 321        spin_unlock(&vfsmount_lock);
 322        unlock_mnt_writers();
 323}
 324
 325/**
 326 * mnt_drop_write - give up write access to a mount
 327 * @mnt: the mount on which to give up write access
 328 *
 329 * Tells the low-level filesystem that we are done
 330 * performing writes to it.  Must be matched with
 331 * mnt_want_write() call above.
 332 */
 333void mnt_drop_write(struct vfsmount *mnt)
 334{
 335        int must_check_underflow = 0;
 336        struct mnt_writer *cpu_writer;
 337
 338        cpu_writer = &get_cpu_var(mnt_writers);
 339        spin_lock(&cpu_writer->lock);
 340
 341        use_cpu_writer_for_mount(cpu_writer, mnt);
 342        if (cpu_writer->count > 0) {
 343                cpu_writer->count--;
 344        } else {
 345                must_check_underflow = 1;
 346                atomic_dec(&mnt->__mnt_writers);
 347        }
 348
 349        spin_unlock(&cpu_writer->lock);
 350        /*
 351         * Logically, we could call this each time,
 352         * but the __mnt_writers cacheline tends to
 353         * be cold, and makes this expensive.
 354         */
 355        if (must_check_underflow)
 356                handle_write_count_underflow(mnt);
 357        /*
 358         * This could be done right after the spinlock
 359         * is taken because the spinlock keeps us on
 360         * the cpu, and disables preemption.  However,
 361         * putting it here bounds the amount that
 362         * __mnt_writers can underflow.  Without it,
 363         * we could theoretically wrap __mnt_writers.
 364         */
 365        put_cpu_var(mnt_writers);
 366}
 367EXPORT_SYMBOL_GPL(mnt_drop_write);
 368
 369static int mnt_make_readonly(struct vfsmount *mnt)
 370{
 371        int ret = 0;
 372
 373        lock_mnt_writers();
 374        /*
 375         * With all the locks held, this value is stable
 376         */
 377        if (atomic_read(&mnt->__mnt_writers) > 0) {
 378                ret = -EBUSY;
 379                goto out;
 380        }
 381        /*
 382         * nobody can do a successful mnt_want_write() with all
 383         * of the counts in MNT_DENIED_WRITE and the locks held.
 384         */
 385        spin_lock(&vfsmount_lock);
 386        if (!ret)
 387                mnt->mnt_flags |= MNT_READONLY;
 388        spin_unlock(&vfsmount_lock);
 389out:
 390        unlock_mnt_writers();
 391        return ret;
 392}
 393
 394static void __mnt_unmake_readonly(struct vfsmount *mnt)
 395{
 396        spin_lock(&vfsmount_lock);
 397        mnt->mnt_flags &= ~MNT_READONLY;
 398        spin_unlock(&vfsmount_lock);
 399}
 400
 401void simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
 402{
 403        mnt->mnt_sb = sb;
 404        mnt->mnt_root = dget(sb->s_root);
 405}
 406
 407EXPORT_SYMBOL(simple_set_mnt);
 408
 409void free_vfsmnt(struct vfsmount *mnt)
 410{
 411        kfree(mnt->mnt_devname);
 412        mnt_free_id(mnt);
 413        kmem_cache_free(mnt_cache, mnt);
 414}
 415
 416/*
 417 * find the first or last mount at @dentry on vfsmount @mnt depending on
 418 * @dir. If @dir is set return the first mount else return the last mount.
 419 */
 420struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
 421                              int dir)
 422{
 423        struct list_head *head = mount_hashtable + hash(mnt, dentry);
 424        struct list_head *tmp = head;
 425        struct vfsmount *p, *found = NULL;
 426
 427        for (;;) {
 428                tmp = dir ? tmp->next : tmp->prev;
 429                p = NULL;
 430                if (tmp == head)
 431                        break;
 432                p = list_entry(tmp, struct vfsmount, mnt_hash);
 433                if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
 434                        found = p;
 435                        break;
 436                }
 437        }
 438        return found;
 439}
 440
 441/*
 442 * lookup_mnt increments the ref count before returning
 443 * the vfsmount struct.
 444 */
 445struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
 446{
 447        struct vfsmount *child_mnt;
 448        spin_lock(&vfsmount_lock);
 449        if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
 450                mntget(child_mnt);
 451        spin_unlock(&vfsmount_lock);
 452        return child_mnt;
 453}
 454
 455static inline int check_mnt(struct vfsmount *mnt)
 456{
 457        return mnt->mnt_ns == current->nsproxy->mnt_ns;
 458}
 459
 460static void touch_mnt_namespace(struct mnt_namespace *ns)
 461{
 462        if (ns) {
 463                ns->event = ++event;
 464                wake_up_interruptible(&ns->poll);
 465        }
 466}
 467
 468static void __touch_mnt_namespace(struct mnt_namespace *ns)
 469{
 470        if (ns && ns->event != event) {
 471                ns->event = event;
 472                wake_up_interruptible(&ns->poll);
 473        }
 474}
 475
 476static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
 477{
 478        old_path->dentry = mnt->mnt_mountpoint;
 479        old_path->mnt = mnt->mnt_parent;
 480        mnt->mnt_parent = mnt;
 481        mnt->mnt_mountpoint = mnt->mnt_root;
 482        list_del_init(&mnt->mnt_child);
 483        list_del_init(&mnt->mnt_hash);
 484        old_path->dentry->d_mounted--;
 485}
 486
 487void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
 488                        struct vfsmount *child_mnt)
 489{
 490        child_mnt->mnt_parent = mntget(mnt);
 491        child_mnt->mnt_mountpoint = dget(dentry);
 492        dentry->d_mounted++;
 493}
 494
 495static void attach_mnt(struct vfsmount *mnt, struct path *path)
 496{
 497        mnt_set_mountpoint(path->mnt, path->dentry, mnt);
 498        list_add_tail(&mnt->mnt_hash, mount_hashtable +
 499                        hash(path->mnt, path->dentry));
 500        list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
 501}
 502
 503/*
 504 * the caller must hold vfsmount_lock
 505 */
 506static void commit_tree(struct vfsmount *mnt)
 507{
 508        struct vfsmount *parent = mnt->mnt_parent;
 509        struct vfsmount *m;
 510        LIST_HEAD(head);
 511        struct mnt_namespace *n = parent->mnt_ns;
 512
 513        BUG_ON(parent == mnt);
 514
 515        list_add_tail(&head, &mnt->mnt_list);
 516        list_for_each_entry(m, &head, mnt_list)
 517                m->mnt_ns = n;
 518        list_splice(&head, n->list.prev);
 519
 520        list_add_tail(&mnt->mnt_hash, mount_hashtable +
 521                                hash(parent, mnt->mnt_mountpoint));
 522        list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
 523        touch_mnt_namespace(n);
 524}
 525
 526static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
 527{
 528        struct list_head *next = p->mnt_mounts.next;
 529        if (next == &p->mnt_mounts) {
 530                while (1) {
 531                        if (p == root)
 532                                return NULL;
 533                        next = p->mnt_child.next;
 534                        if (next != &p->mnt_parent->mnt_mounts)
 535                                break;
 536                        p = p->mnt_parent;
 537                }
 538        }
 539        return list_entry(next, struct vfsmount, mnt_child);
 540}
 541
 542static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
 543{
 544        struct list_head *prev = p->mnt_mounts.prev;
 545        while (prev != &p->mnt_mounts) {
 546                p = list_entry(prev, struct vfsmount, mnt_child);
 547                prev = p->mnt_mounts.prev;
 548        }
 549        return p;
 550}
 551
 552static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
 553                                        int flag)
 554{
 555        struct super_block *sb = old->mnt_sb;
 556        struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
 557
 558        if (mnt) {
 559                if (flag & (CL_SLAVE | CL_PRIVATE))
 560                        mnt->mnt_group_id = 0; /* not a peer of original */
 561                else
 562                        mnt->mnt_group_id = old->mnt_group_id;
 563
 564                if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
 565                        int err = mnt_alloc_group_id(mnt);
 566                        if (err)
 567                                goto out_free;
 568                }
 569
 570                mnt->mnt_flags = old->mnt_flags;
 571                atomic_inc(&sb->s_active);
 572                mnt->mnt_sb = sb;
 573                mnt->mnt_root = dget(root);
 574                mnt->mnt_mountpoint = mnt->mnt_root;
 575                mnt->mnt_parent = mnt;
 576
 577                if (flag & CL_SLAVE) {
 578                        list_add(&mnt->mnt_slave, &old->mnt_slave_list);
 579                        mnt->mnt_master = old;
 580                        CLEAR_MNT_SHARED(mnt);
 581                } else if (!(flag & CL_PRIVATE)) {
 582                        if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
 583                                list_add(&mnt->mnt_share, &old->mnt_share);
 584                        if (IS_MNT_SLAVE(old))
 585                                list_add(&mnt->mnt_slave, &old->mnt_slave);
 586                        mnt->mnt_master = old->mnt_master;
 587                }
 588                if (flag & CL_MAKE_SHARED)
 589                        set_mnt_shared(mnt);
 590
 591                /* stick the duplicate mount on the same expiry list
 592                 * as the original if that was on one */
 593                if (flag & CL_EXPIRE) {
 594                        if (!list_empty(&old->mnt_expire))
 595                                list_add(&mnt->mnt_expire, &old->mnt_expire);
 596                }
 597        }
 598        return mnt;
 599
 600 out_free:
 601        free_vfsmnt(mnt);
 602        return NULL;
 603}
 604
 605static inline void __mntput(struct vfsmount *mnt)
 606{
 607        int cpu;
 608        struct super_block *sb = mnt->mnt_sb;
 609        /*
 610         * We don't have to hold all of the locks at the
 611         * same time here because we know that we're the
 612         * last reference to mnt and that no new writers
 613         * can come in.
 614         */
 615        for_each_possible_cpu(cpu) {
 616                struct mnt_writer *cpu_writer = &per_cpu(mnt_writers, cpu);
 617                spin_lock(&cpu_writer->lock);
 618                if (cpu_writer->mnt != mnt) {
 619                        spin_unlock(&cpu_writer->lock);
 620                        continue;
 621                }
 622                atomic_add(cpu_writer->count, &mnt->__mnt_writers);
 623                cpu_writer->count = 0;
 624                /*
 625                 * Might as well do this so that no one
 626                 * ever sees the pointer and expects
 627                 * it to be valid.
 628                 */
 629                cpu_writer->mnt = NULL;
 630                spin_unlock(&cpu_writer->lock);
 631        }
 632        /*
 633         * This probably indicates that somebody messed
 634         * up a mnt_want/drop_write() pair.  If this
 635         * happens, the filesystem was probably unable
 636         * to make r/w->r/o transitions.
 637         */
 638        WARN_ON(atomic_read(&mnt->__mnt_writers));
 639        dput(mnt->mnt_root);
 640        free_vfsmnt(mnt);
 641        deactivate_super(sb);
 642}
 643
 644void mntput_no_expire(struct vfsmount *mnt)
 645{
 646repeat:
 647        if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
 648                if (likely(!mnt->mnt_pinned)) {
 649                        spin_unlock(&vfsmount_lock);
 650                        __mntput(mnt);
 651                        return;
 652                }
 653                atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
 654                mnt->mnt_pinned = 0;
 655                spin_unlock(&vfsmount_lock);
 656                acct_auto_close_mnt(mnt);
 657                security_sb_umount_close(mnt);
 658                goto repeat;
 659        }
 660}
 661
 662EXPORT_SYMBOL(mntput_no_expire);
 663
 664void mnt_pin(struct vfsmount *mnt)
 665{
 666        spin_lock(&vfsmount_lock);
 667        mnt->mnt_pinned++;
 668        spin_unlock(&vfsmount_lock);
 669}
 670
 671EXPORT_SYMBOL(mnt_pin);
 672
 673void mnt_unpin(struct vfsmount *mnt)
 674{
 675        spin_lock(&vfsmount_lock);
 676        if (mnt->mnt_pinned) {
 677                atomic_inc(&mnt->mnt_count);
 678                mnt->mnt_pinned--;
 679        }
 680        spin_unlock(&vfsmount_lock);
 681}
 682
 683EXPORT_SYMBOL(mnt_unpin);
 684
 685static inline void mangle(struct seq_file *m, const char *s)
 686{
 687        seq_escape(m, s, " \t\n\\");
 688}
 689
 690/*
 691 * Simple .show_options callback for filesystems which don't want to
 692 * implement more complex mount option showing.
 693 *
 694 * See also save_mount_options().
 695 */
 696int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
 697{
 698        const char *options;
 699
 700        rcu_read_lock();
 701        options = rcu_dereference(mnt->mnt_sb->s_options);
 702
 703        if (options != NULL && options[0]) {
 704                seq_putc(m, ',');
 705                mangle(m, options);
 706        }
 707        rcu_read_unlock();
 708
 709        return 0;
 710}
 711EXPORT_SYMBOL(generic_show_options);
 712
 713/*
 714 * If filesystem uses generic_show_options(), this function should be
 715 * called from the fill_super() callback.
 716 *
 717 * The .remount_fs callback usually needs to be handled in a special
 718 * way, to make sure, that previous options are not overwritten if the
 719 * remount fails.
 720 *
 721 * Also note, that if the filesystem's .remount_fs function doesn't
 722 * reset all options to their default value, but changes only newly
 723 * given options, then the displayed options will not reflect reality
 724 * any more.
 725 */
 726void save_mount_options(struct super_block *sb, char *options)
 727{
 728        BUG_ON(sb->s_options);
 729        rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
 730}
 731EXPORT_SYMBOL(save_mount_options);
 732
 733void replace_mount_options(struct super_block *sb, char *options)
 734{
 735        char *old = sb->s_options;
 736        rcu_assign_pointer(sb->s_options, options);
 737        if (old) {
 738                synchronize_rcu();
 739                kfree(old);
 740        }
 741}
 742EXPORT_SYMBOL(replace_mount_options);
 743
 744#ifdef CONFIG_PROC_FS
 745/* iterator */
 746static void *m_start(struct seq_file *m, loff_t *pos)
 747{
 748        struct proc_mounts *p = m->private;
 749
 750        down_read(&namespace_sem);
 751        return seq_list_start(&p->ns->list, *pos);
 752}
 753
 754static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 755{
 756        struct proc_mounts *p = m->private;
 757
 758        return seq_list_next(v, &p->ns->list, pos);
 759}
 760
 761static void m_stop(struct seq_file *m, void *v)
 762{
 763        up_read(&namespace_sem);
 764}
 765
 766struct proc_fs_info {
 767        int flag;
 768        const char *str;
 769};
 770
 771static int show_sb_opts(struct seq_file *m, struct super_block *sb)
 772{
 773        static const struct proc_fs_info fs_info[] = {
 774                { MS_SYNCHRONOUS, ",sync" },
 775                { MS_DIRSYNC, ",dirsync" },
 776                { MS_MANDLOCK, ",mand" },
 777                { 0, NULL }
 778        };
 779        const struct proc_fs_info *fs_infop;
 780
 781        for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
 782                if (sb->s_flags & fs_infop->flag)
 783                        seq_puts(m, fs_infop->str);
 784        }
 785
 786        return security_sb_show_options(m, sb);
 787}
 788
 789static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
 790{
 791        static const struct proc_fs_info mnt_info[] = {
 792                { MNT_NOSUID, ",nosuid" },
 793                { MNT_NODEV, ",nodev" },
 794                { MNT_NOEXEC, ",noexec" },
 795                { MNT_NOATIME, ",noatime" },
 796                { MNT_NODIRATIME, ",nodiratime" },
 797                { MNT_RELATIME, ",relatime" },
 798                { MNT_STRICTATIME, ",strictatime" },
 799                { 0, NULL }
 800        };
 801        const struct proc_fs_info *fs_infop;
 802
 803        for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
 804                if (mnt->mnt_flags & fs_infop->flag)
 805                        seq_puts(m, fs_infop->str);
 806        }
 807}
 808
 809static void show_type(struct seq_file *m, struct super_block *sb)
 810{
 811        mangle(m, sb->s_type->name);
 812        if (sb->s_subtype && sb->s_subtype[0]) {
 813                seq_putc(m, '.');
 814                mangle(m, sb->s_subtype);
 815        }
 816}
 817
 818static int show_vfsmnt(struct seq_file *m, void *v)
 819{
 820        struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
 821        int err = 0;
 822        struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
 823
 824        mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
 825        seq_putc(m, ' ');
 826        seq_path(m, &mnt_path, " \t\n\\");
 827        seq_putc(m, ' ');
 828        show_type(m, mnt->mnt_sb);
 829        seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
 830        err = show_sb_opts(m, mnt->mnt_sb);
 831        if (err)
 832                goto out;
 833        show_mnt_opts(m, mnt);
 834        if (mnt->mnt_sb->s_op->show_options)
 835                err = mnt->mnt_sb->s_op->show_options(m, mnt);
 836        seq_puts(m, " 0 0\n");
 837out:
 838        return err;
 839}
 840
 841const struct seq_operations mounts_op = {
 842        .start  = m_start,
 843        .next   = m_next,
 844        .stop   = m_stop,
 845        .show   = show_vfsmnt
 846};
 847
 848static int show_mountinfo(struct seq_file *m, void *v)
 849{
 850        struct proc_mounts *p = m->private;
 851        struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
 852        struct super_block *sb = mnt->mnt_sb;
 853        struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
 854        struct path root = p->root;
 855        int err = 0;
 856
 857        seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
 858                   MAJOR(sb->s_dev), MINOR(sb->s_dev));
 859        seq_dentry(m, mnt->mnt_root, " \t\n\\");
 860        seq_putc(m, ' ');
 861        seq_path_root(m, &mnt_path, &root, " \t\n\\");
 862        if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
 863                /*
 864                 * Mountpoint is outside root, discard that one.  Ugly,
 865                 * but less so than trying to do that in iterator in a
 866                 * race-free way (due to renames).
 867                 */
 868                return SEQ_SKIP;
 869        }
 870        seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
 871        show_mnt_opts(m, mnt);
 872
 873        /* Tagged fields ("foo:X" or "bar") */
 874        if (IS_MNT_SHARED(mnt))
 875                seq_printf(m, " shared:%i", mnt->mnt_group_id);
 876        if (IS_MNT_SLAVE(mnt)) {
 877                int master = mnt->mnt_master->mnt_group_id;
 878                int dom = get_dominating_id(mnt, &p->root);
 879                seq_printf(m, " master:%i", master);
 880                if (dom && dom != master)
 881                        seq_printf(m, " propagate_from:%i", dom);
 882        }
 883        if (IS_MNT_UNBINDABLE(mnt))
 884                seq_puts(m, " unbindable");
 885
 886        /* Filesystem specific data */
 887        seq_puts(m, " - ");
 888        show_type(m, sb);
 889        seq_putc(m, ' ');
 890        mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
 891        seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
 892        err = show_sb_opts(m, sb);
 893        if (err)
 894                goto out;
 895        if (sb->s_op->show_options)
 896                err = sb->s_op->show_options(m, mnt);
 897        seq_putc(m, '\n');
 898out:
 899        return err;
 900}
 901
 902const struct seq_operations mountinfo_op = {
 903        .start  = m_start,
 904        .next   = m_next,
 905        .stop   = m_stop,
 906        .show   = show_mountinfo,
 907};
 908
 909static int show_vfsstat(struct seq_file *m, void *v)
 910{
 911        struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
 912        struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
 913        int err = 0;
 914
 915        /* device */
 916        if (mnt->mnt_devname) {
 917                seq_puts(m, "device ");
 918                mangle(m, mnt->mnt_devname);
 919        } else
 920                seq_puts(m, "no device");
 921
 922        /* mount point */
 923        seq_puts(m, " mounted on ");
 924        seq_path(m, &mnt_path, " \t\n\\");
 925        seq_putc(m, ' ');
 926
 927        /* file system type */
 928        seq_puts(m, "with fstype ");
 929        show_type(m, mnt->mnt_sb);
 930
 931        /* optional statistics */
 932        if (mnt->mnt_sb->s_op->show_stats) {
 933                seq_putc(m, ' ');
 934                err = mnt->mnt_sb->s_op->show_stats(m, mnt);
 935        }
 936
 937        seq_putc(m, '\n');
 938        return err;
 939}
 940
 941const struct seq_operations mountstats_op = {
 942        .start  = m_start,
 943        .next   = m_next,
 944        .stop   = m_stop,
 945        .show   = show_vfsstat,
 946};
 947#endif  /* CONFIG_PROC_FS */
 948
 949/**
 950 * may_umount_tree - check if a mount tree is busy
 951 * @mnt: root of mount tree
 952 *
 953 * This is called to check if a tree of mounts has any
 954 * open files, pwds, chroots or sub mounts that are
 955 * busy.
 956 */
 957int may_umount_tree(struct vfsmount *mnt)
 958{
 959        int actual_refs = 0;
 960        int minimum_refs = 0;
 961        struct vfsmount *p;
 962
 963        spin_lock(&vfsmount_lock);
 964        for (p = mnt; p; p = next_mnt(p, mnt)) {
 965                actual_refs += atomic_read(&p->mnt_count);
 966                minimum_refs += 2;
 967        }
 968        spin_unlock(&vfsmount_lock);
 969
 970        if (actual_refs > minimum_refs)
 971                return 0;
 972
 973        return 1;
 974}
 975
 976EXPORT_SYMBOL(may_umount_tree);
 977
 978/**
 979 * may_umount - check if a mount point is busy
 980 * @mnt: root of mount
 981 *
 982 * This is called to check if a mount point has any
 983 * open files, pwds, chroots or sub mounts. If the
 984 * mount has sub mounts this will return busy
 985 * regardless of whether the sub mounts are busy.
 986 *
 987 * Doesn't take quota and stuff into account. IOW, in some cases it will
 988 * give false negatives. The main reason why it's here is that we need
 989 * a non-destructive way to look for easily umountable filesystems.
 990 */
 991int may_umount(struct vfsmount *mnt)
 992{
 993        int ret = 1;
 994        spin_lock(&vfsmount_lock);
 995        if (propagate_mount_busy(mnt, 2))
 996                ret = 0;
 997        spin_unlock(&vfsmount_lock);
 998        return ret;
 999}
1000
1001EXPORT_SYMBOL(may_umount);
1002
1003void release_mounts(struct list_head *head)
1004{
1005        struct vfsmount *mnt;
1006        while (!list_empty(head)) {
1007                mnt = list_first_entry(head, struct vfsmount, mnt_hash);
1008                list_del_init(&mnt->mnt_hash);
1009                if (mnt->mnt_parent != mnt) {
1010                        struct dentry *dentry;
1011                        struct vfsmount *m;
1012                        spin_lock(&vfsmount_lock);
1013                        dentry = mnt->mnt_mountpoint;
1014                        m = mnt->mnt_parent;
1015                        mnt->mnt_mountpoint = mnt->mnt_root;
1016                        mnt->mnt_parent = mnt;
1017                        m->mnt_ghosts--;
1018                        spin_unlock(&vfsmount_lock);
1019                        dput(dentry);
1020                        mntput(m);
1021                }
1022                mntput(mnt);
1023        }
1024}
1025
1026void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
1027{
1028        struct vfsmount *p;
1029
1030        for (p = mnt; p; p = next_mnt(p, mnt))
1031                list_move(&p->mnt_hash, kill);
1032
1033        if (propagate)
1034                propagate_umount(kill);
1035
1036        list_for_each_entry(p, kill, mnt_hash) {
1037                list_del_init(&p->mnt_expire);
1038                list_del_init(&p->mnt_list);
1039                __touch_mnt_namespace(p->mnt_ns);
1040                p->mnt_ns = NULL;
1041                list_del_init(&p->mnt_child);
1042                if (p->mnt_parent != p) {
1043                        p->mnt_parent->mnt_ghosts++;
1044                        p->mnt_mountpoint->d_mounted--;
1045                }
1046                change_mnt_propagation(p, MS_PRIVATE);
1047        }
1048}
1049
1050static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1051
1052static int do_umount(struct vfsmount *mnt, int flags)
1053{
1054        struct super_block *sb = mnt->mnt_sb;
1055        int retval;
1056        LIST_HEAD(umount_list);
1057
1058        retval = security_sb_umount(mnt, flags);
1059        if (retval)
1060                return retval;
1061
1062        /*
1063         * Allow userspace to request a mountpoint be expired rather than
1064         * unmounting unconditionally. Unmount only happens if:
1065         *  (1) the mark is already set (the mark is cleared by mntput())
1066         *  (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1067         */
1068        if (flags & MNT_EXPIRE) {
1069                if (mnt == current->fs->root.mnt ||
1070                    flags & (MNT_FORCE | MNT_DETACH))
1071                        return -EINVAL;
1072
1073                if (atomic_read(&mnt->mnt_count) != 2)
1074                        return -EBUSY;
1075
1076                if (!xchg(&mnt->mnt_expiry_mark, 1))
1077                        return -EAGAIN;
1078        }
1079
1080        /*
1081         * If we may have to abort operations to get out of this
1082         * mount, and they will themselves hold resources we must
1083         * allow the fs to do things. In the Unix tradition of
1084         * 'Gee thats tricky lets do it in userspace' the umount_begin
1085         * might fail to complete on the first run through as other tasks
1086         * must return, and the like. Thats for the mount program to worry
1087         * about for the moment.
1088         */
1089
1090        if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1091                sb->s_op->umount_begin(sb);
1092        }
1093
1094        /*
1095         * No sense to grab the lock for this test, but test itself looks
1096         * somewhat bogus. Suggestions for better replacement?
1097         * Ho-hum... In principle, we might treat that as umount + switch
1098         * to rootfs. GC would eventually take care of the old vfsmount.
1099         * Actually it makes sense, especially if rootfs would contain a
1100         * /reboot - static binary that would close all descriptors and
1101         * call reboot(9). Then init(8) could umount root and exec /reboot.
1102         */
1103        if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1104                /*
1105                 * Special case for "unmounting" root ...
1106                 * we just try to remount it readonly.
1107                 */
1108                down_write(&sb->s_umount);
1109                if (!(sb->s_flags & MS_RDONLY)) {
1110                        lock_kernel();
1111                        retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1112                        unlock_kernel();
1113                }
1114                up_write(&sb->s_umount);
1115                return retval;
1116        }
1117
1118        down_write(&namespace_sem);
1119        spin_lock(&vfsmount_lock);
1120        event++;
1121
1122        if (!(flags & MNT_DETACH))
1123                shrink_submounts(mnt, &umount_list);
1124
1125        retval = -EBUSY;
1126        if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1127                if (!list_empty(&mnt->mnt_list))
1128                        umount_tree(mnt, 1, &umount_list);
1129                retval = 0;
1130        }
1131        spin_unlock(&vfsmount_lock);
1132        if (retval)
1133                security_sb_umount_busy(mnt);
1134        up_write(&namespace_sem);
1135        release_mounts(&umount_list);
1136        return retval;
1137}
1138
1139/*
1140 * Now umount can handle mount points as well as block devices.
1141 * This is important for filesystems which use unnamed block devices.
1142 *
1143 * We now support a flag for forced unmount like the other 'big iron'
1144 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1145 */
1146
1147SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1148{
1149        struct path path;
1150        int retval;
1151
1152        retval = user_path(name, &path);
1153        if (retval)
1154                goto out;
1155        retval = -EINVAL;
1156        if (path.dentry != path.mnt->mnt_root)
1157                goto dput_and_out;
1158        if (!check_mnt(path.mnt))
1159                goto dput_and_out;
1160
1161        retval = -EPERM;
1162        if (!capable(CAP_SYS_ADMIN))
1163                goto dput_and_out;
1164
1165        retval = do_umount(path.mnt, flags);
1166dput_and_out:
1167        /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1168        dput(path.dentry);
1169        mntput_no_expire(path.mnt);
1170out:
1171        return retval;
1172}
1173
1174#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1175
1176/*
1177 *      The 2.0 compatible umount. No flags.
1178 */
1179SYSCALL_DEFINE1(oldumount, char __user *, name)
1180{
1181        return sys_umount(name, 0);
1182}
1183
1184#endif
1185
1186static int mount_is_safe(struct path *path)
1187{
1188        if (capable(CAP_SYS_ADMIN))
1189                return 0;
1190        return -EPERM;
1191#ifdef notyet
1192        if (S_ISLNK(path->dentry->d_inode->i_mode))
1193                return -EPERM;
1194        if (path->dentry->d_inode->i_mode & S_ISVTX) {
1195                if (current_uid() != path->dentry->d_inode->i_uid)
1196                        return -EPERM;
1197        }
1198        if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1199                return -EPERM;
1200        return 0;
1201#endif
1202}
1203
1204struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
1205                                        int flag)
1206{
1207        struct vfsmount *res, *p, *q, *r, *s;
1208        struct path path;
1209
1210        if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1211                return NULL;
1212
1213        res = q = clone_mnt(mnt, dentry, flag);
1214        if (!q)
1215                goto Enomem;
1216        q->mnt_mountpoint = mnt->mnt_mountpoint;
1217
1218        p = mnt;
1219        list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1220                if (!is_subdir(r->mnt_mountpoint, dentry))
1221                        continue;
1222
1223                for (s = r; s; s = next_mnt(s, r)) {
1224                        if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1225                                s = skip_mnt_tree(s);
1226                                continue;
1227                        }
1228                        while (p != s->mnt_parent) {
1229                                p = p->mnt_parent;
1230                                q = q->mnt_parent;
1231                        }
1232                        p = s;
1233                        path.mnt = q;
1234                        path.dentry = p->mnt_mountpoint;
1235                        q = clone_mnt(p, p->mnt_root, flag);
1236                        if (!q)
1237                                goto Enomem;
1238                        spin_lock(&vfsmount_lock);
1239                        list_add_tail(&q->mnt_list, &res->mnt_list);
1240                        attach_mnt(q, &path);
1241                        spin_unlock(&vfsmount_lock);
1242                }
1243        }
1244        return res;
1245Enomem:
1246        if (res) {
1247                LIST_HEAD(umount_list);
1248                spin_lock(&vfsmount_lock);
1249                umount_tree(res, 0, &umount_list);
1250                spin_unlock(&vfsmount_lock);
1251                release_mounts(&umount_list);
1252        }
1253        return NULL;
1254}
1255
1256struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry)
1257{
1258        struct vfsmount *tree;
1259        down_write(&namespace_sem);
1260        tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE);
1261        up_write(&namespace_sem);
1262        return tree;
1263}
1264
1265void drop_collected_mounts(struct vfsmount *mnt)
1266{
1267        LIST_HEAD(umount_list);
1268        down_write(&namespace_sem);
1269        spin_lock(&vfsmount_lock);
1270        umount_tree(mnt, 0, &umount_list);
1271        spin_unlock(&vfsmount_lock);
1272        up_write(&namespace_sem);
1273        release_mounts(&umount_list);
1274}
1275
1276static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1277{
1278        struct vfsmount *p;
1279
1280        for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1281                if (p->mnt_group_id && !IS_MNT_SHARED(p))
1282                        mnt_release_group_id(p);
1283        }
1284}
1285
1286static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1287{
1288        struct vfsmount *p;
1289
1290        for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1291                if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1292                        int err = mnt_alloc_group_id(p);
1293                        if (err) {
1294                                cleanup_group_ids(mnt, p);
1295                                return err;
1296                        }
1297                }
1298        }
1299
1300        return 0;
1301}
1302
1303/*
1304 *  @source_mnt : mount tree to be attached
1305 *  @nd         : place the mount tree @source_mnt is attached
1306 *  @parent_nd  : if non-null, detach the source_mnt from its parent and
1307 *                 store the parent mount and mountpoint dentry.
1308 *                 (done when source_mnt is moved)
1309 *
1310 *  NOTE: in the table below explains the semantics when a source mount
1311 *  of a given type is attached to a destination mount of a given type.
1312 * ---------------------------------------------------------------------------
1313 * |         BIND MOUNT OPERATION                                            |
1314 * |**************************************************************************
1315 * | source-->| shared        |       private  |       slave    | unbindable |
1316 * | dest     |               |                |                |            |
1317 * |   |      |               |                |                |            |
1318 * |   v      |               |                |                |            |
1319 * |**************************************************************************
1320 * |  shared  | shared (++)   |     shared (+) |     shared(+++)|  invalid   |
1321 * |          |               |                |                |            |
1322 * |non-shared| shared (+)    |      private   |      slave (*) |  invalid   |
1323 * ***************************************************************************
1324 * A bind operation clones the source mount and mounts the clone on the
1325 * destination mount.
1326 *
1327 * (++)  the cloned mount is propagated to all the mounts in the propagation
1328 *       tree of the destination mount and the cloned mount is added to
1329 *       the peer group of the source mount.
1330 * (+)   the cloned mount is created under the destination mount and is marked
1331 *       as shared. The cloned mount is added to the peer group of the source
1332 *       mount.
1333 * (+++) the mount is propagated to all the mounts in the propagation tree
1334 *       of the destination mount and the cloned mount is made slave
1335 *       of the same master as that of the source mount. The cloned mount
1336 *       is marked as 'shared and slave'.
1337 * (*)   the cloned mount is made a slave of the same master as that of the
1338 *       source mount.
1339 *
1340 * ---------------------------------------------------------------------------
1341 * |                    MOVE MOUNT OPERATION                                 |
1342 * |**************************************************************************
1343 * | source-->| shared        |       private  |       slave    | unbindable |
1344 * | dest     |               |                |                |            |
1345 * |   |      |               |                |                |            |
1346 * |   v      |               |                |                |            |
1347 * |**************************************************************************
1348 * |  shared  | shared (+)    |     shared (+) |    shared(+++) |  invalid   |
1349 * |          |               |                |                |            |
1350 * |non-shared| shared (+*)   |      private   |    slave (*)   | unbindable |
1351 * ***************************************************************************
1352 *
1353 * (+)  the mount is moved to the destination. And is then propagated to
1354 *      all the mounts in the propagation tree of the destination mount.
1355 * (+*)  the mount is moved to the destination.
1356 * (+++)  the mount is moved to the destination and is then propagated to
1357 *      all the mounts belonging to the destination mount's propagation tree.
1358 *      the mount is marked as 'shared and slave'.
1359 * (*)  the mount continues to be a slave at the new location.
1360 *
1361 * if the source mount is a tree, the operations explained above is
1362 * applied to each mount in the tree.
1363 * Must be called without spinlocks held, since this function can sleep
1364 * in allocations.
1365 */
1366static int attach_recursive_mnt(struct vfsmount *source_mnt,
1367                        struct path *path, struct path *parent_path)
1368{
1369        LIST_HEAD(tree_list);
1370        struct vfsmount *dest_mnt = path->mnt;
1371        struct dentry *dest_dentry = path->dentry;
1372        struct vfsmount *child, *p;
1373        int err;
1374
1375        if (IS_MNT_SHARED(dest_mnt)) {
1376                err = invent_group_ids(source_mnt, true);
1377                if (err)
1378                        goto out;
1379        }
1380        err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1381        if (err)
1382                goto out_cleanup_ids;
1383
1384        if (IS_MNT_SHARED(dest_mnt)) {
1385                for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1386                        set_mnt_shared(p);
1387        }
1388
1389        spin_lock(&vfsmount_lock);
1390        if (parent_path) {
1391                detach_mnt(source_mnt, parent_path);
1392                attach_mnt(source_mnt, path);
1393                touch_mnt_namespace(parent_path->mnt->mnt_ns);
1394        } else {
1395                mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1396                commit_tree(source_mnt);
1397        }
1398
1399        list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1400                list_del_init(&child->mnt_hash);
1401                commit_tree(child);
1402        }
1403        spin_unlock(&vfsmount_lock);
1404        return 0;
1405
1406 out_cleanup_ids:
1407        if (IS_MNT_SHARED(dest_mnt))
1408                cleanup_group_ids(source_mnt, NULL);
1409 out:
1410        return err;
1411}
1412
1413static int graft_tree(struct vfsmount *mnt, struct path *path)
1414{
1415        int err;
1416        if (mnt->mnt_sb->s_flags & MS_NOUSER)
1417                return -EINVAL;
1418
1419        if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1420              S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1421                return -ENOTDIR;
1422
1423        err = -ENOENT;
1424        mutex_lock(&path->dentry->d_inode->i_mutex);
1425        if (IS_DEADDIR(path->dentry->d_inode))
1426                goto out_unlock;
1427
1428        err = security_sb_check_sb(mnt, path);
1429        if (err)
1430                goto out_unlock;
1431
1432        err = -ENOENT;
1433        if (IS_ROOT(path->dentry) || !d_unhashed(path->dentry))
1434                err = attach_recursive_mnt(mnt, path, NULL);
1435out_unlock:
1436        mutex_unlock(&path->dentry->d_inode->i_mutex);
1437        if (!err)
1438                security_sb_post_addmount(mnt, path);
1439        return err;
1440}
1441
1442/*
1443 * recursively change the type of the mountpoint.
1444 */
1445static int do_change_type(struct path *path, int flag)
1446{
1447        struct vfsmount *m, *mnt = path->mnt;
1448        int recurse = flag & MS_REC;
1449        int type = flag & ~MS_REC;
1450        int err = 0;
1451
1452        if (!capable(CAP_SYS_ADMIN))
1453                return -EPERM;
1454
1455        if (path->dentry != path->mnt->mnt_root)
1456                return -EINVAL;
1457
1458        down_write(&namespace_sem);
1459        if (type == MS_SHARED) {
1460                err = invent_group_ids(mnt, recurse);
1461                if (err)
1462                        goto out_unlock;
1463        }
1464
1465        spin_lock(&vfsmount_lock);
1466        for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1467                change_mnt_propagation(m, type);
1468        spin_unlock(&vfsmount_lock);
1469
1470 out_unlock:
1471        up_write(&namespace_sem);
1472        return err;
1473}
1474
1475/*
1476 * do loopback mount.
1477 */
1478static int do_loopback(struct path *path, char *old_name,
1479                                int recurse)
1480{
1481        struct path old_path;
1482        struct vfsmount *mnt = NULL;
1483        int err = mount_is_safe(path);
1484        if (err)
1485                return err;
1486        if (!old_name || !*old_name)
1487                return -EINVAL;
1488        err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1489        if (err)
1490                return err;
1491
1492        down_write(&namespace_sem);
1493        err = -EINVAL;
1494        if (IS_MNT_UNBINDABLE(old_path.mnt))
1495                goto out;
1496
1497        if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1498                goto out;
1499
1500        err = -ENOMEM;
1501        if (recurse)
1502                mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
1503        else
1504                mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
1505
1506        if (!mnt)
1507                goto out;
1508
1509        err = graft_tree(mnt, path);
1510        if (err) {
1511                LIST_HEAD(umount_list);
1512                spin_lock(&vfsmount_lock);
1513                umount_tree(mnt, 0, &umount_list);
1514                spin_unlock(&vfsmount_lock);
1515                release_mounts(&umount_list);
1516        }
1517
1518out:
1519        up_write(&namespace_sem);
1520        path_put(&old_path);
1521        return err;
1522}
1523
1524static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1525{
1526        int error = 0;
1527        int readonly_request = 0;
1528
1529        if (ms_flags & MS_RDONLY)
1530                readonly_request = 1;
1531        if (readonly_request == __mnt_is_readonly(mnt))
1532                return 0;
1533
1534        if (readonly_request)
1535                error = mnt_make_readonly(mnt);
1536        else
1537                __mnt_unmake_readonly(mnt);
1538        return error;
1539}
1540
1541/*
1542 * change filesystem flags. dir should be a physical root of filesystem.
1543 * If you've mounted a non-root directory somewhere and want to do remount
1544 * on it - tough luck.
1545 */
1546static int do_remount(struct path *path, int flags, int mnt_flags,
1547                      void *data)
1548{
1549        int err;
1550        struct super_block *sb = path->mnt->mnt_sb;
1551
1552        if (!capable(CAP_SYS_ADMIN))
1553                return -EPERM;
1554
1555        if (!check_mnt(path->mnt))
1556                return -EINVAL;
1557
1558        if (path->dentry != path->mnt->mnt_root)
1559                return -EINVAL;
1560
1561        down_write(&sb->s_umount);
1562        if (flags & MS_BIND)
1563                err = change_mount_flags(path->mnt, flags);
1564        else
1565                err = do_remount_sb(sb, flags, data, 0);
1566        if (!err)
1567                path->mnt->mnt_flags = mnt_flags;
1568        up_write(&sb->s_umount);
1569        if (!err) {
1570                security_sb_post_remount(path->mnt, flags, data);
1571
1572                spin_lock(&vfsmount_lock);
1573                touch_mnt_namespace(path->mnt->mnt_ns);
1574                spin_unlock(&vfsmount_lock);
1575        }
1576        return err;
1577}
1578
1579static inline int tree_contains_unbindable(struct vfsmount *mnt)
1580{
1581        struct vfsmount *p;
1582        for (p = mnt; p; p = next_mnt(p, mnt)) {
1583                if (IS_MNT_UNBINDABLE(p))
1584                        return 1;
1585        }
1586        return 0;
1587}
1588
1589static int do_move_mount(struct path *path, char *old_name)
1590{
1591        struct path old_path, parent_path;
1592        struct vfsmount *p;
1593        int err = 0;
1594        if (!capable(CAP_SYS_ADMIN))
1595                return -EPERM;
1596        if (!old_name || !*old_name)
1597                return -EINVAL;
1598        err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1599        if (err)
1600                return err;
1601
1602        down_write(&namespace_sem);
1603        while (d_mountpoint(path->dentry) &&
1604               follow_down(&path->mnt, &path->dentry))
1605                ;
1606        err = -EINVAL;
1607        if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1608                goto out;
1609
1610        err = -ENOENT;
1611        mutex_lock(&path->dentry->d_inode->i_mutex);
1612        if (IS_DEADDIR(path->dentry->d_inode))
1613                goto out1;
1614
1615        if (!IS_ROOT(path->dentry) && d_unhashed(path->dentry))
1616                goto out1;
1617
1618        err = -EINVAL;
1619        if (old_path.dentry != old_path.mnt->mnt_root)
1620                goto out1;
1621
1622        if (old_path.mnt == old_path.mnt->mnt_parent)
1623                goto out1;
1624
1625        if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1626              S_ISDIR(old_path.dentry->d_inode->i_mode))
1627                goto out1;
1628        /*
1629         * Don't move a mount residing in a shared parent.
1630         */
1631        if (old_path.mnt->mnt_parent &&
1632            IS_MNT_SHARED(old_path.mnt->mnt_parent))
1633                goto out1;
1634        /*
1635         * Don't move a mount tree containing unbindable mounts to a destination
1636         * mount which is shared.
1637         */
1638        if (IS_MNT_SHARED(path->mnt) &&
1639            tree_contains_unbindable(old_path.mnt))
1640                goto out1;
1641        err = -ELOOP;
1642        for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1643                if (p == old_path.mnt)
1644                        goto out1;
1645
1646        err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
1647        if (err)
1648                goto out1;
1649
1650        /* if the mount is moved, it should no longer be expire
1651         * automatically */
1652        list_del_init(&old_path.mnt->mnt_expire);
1653out1:
1654        mutex_unlock(&path->dentry->d_inode->i_mutex);
1655out:
1656        up_write(&namespace_sem);
1657        if (!err)
1658                path_put(&parent_path);
1659        path_put(&old_path);
1660        return err;
1661}
1662
1663/*
1664 * create a new mount for userspace and request it to be added into the
1665 * namespace's tree
1666 */
1667static int do_new_mount(struct path *path, char *type, int flags,
1668                        int mnt_flags, char *name, void *data)
1669{
1670        struct vfsmount *mnt;
1671
1672        if (!type || !memchr(type, 0, PAGE_SIZE))
1673                return -EINVAL;
1674
1675        /* we need capabilities... */
1676        if (!capable(CAP_SYS_ADMIN))
1677                return -EPERM;
1678
1679        mnt = do_kern_mount(type, flags, name, data);
1680        if (IS_ERR(mnt))
1681                return PTR_ERR(mnt);
1682
1683        return do_add_mount(mnt, path, mnt_flags, NULL);
1684}
1685
1686/*
1687 * add a mount into a namespace's mount tree
1688 * - provide the option of adding the new mount to an expiration list
1689 */
1690int do_add_mount(struct vfsmount *newmnt, struct path *path,
1691                 int mnt_flags, struct list_head *fslist)
1692{
1693        int err;
1694
1695        down_write(&namespace_sem);
1696        /* Something was mounted here while we slept */
1697        while (d_mountpoint(path->dentry) &&
1698               follow_down(&path->mnt, &path->dentry))
1699                ;
1700        err = -EINVAL;
1701        if (!check_mnt(path->mnt))
1702                goto unlock;
1703
1704        /* Refuse the same filesystem on the same mount point */
1705        err = -EBUSY;
1706        if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1707            path->mnt->mnt_root == path->dentry)
1708                goto unlock;
1709
1710        err = -EINVAL;
1711        if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1712                goto unlock;
1713
1714        newmnt->mnt_flags = mnt_flags;
1715        if ((err = graft_tree(newmnt, path)))
1716                goto unlock;
1717
1718        if (fslist) /* add to the specified expiration list */
1719                list_add_tail(&newmnt->mnt_expire, fslist);
1720
1721        up_write(&namespace_sem);
1722        return 0;
1723
1724unlock:
1725        up_write(&namespace_sem);
1726        mntput(newmnt);
1727        return err;
1728}
1729
1730EXPORT_SYMBOL_GPL(do_add_mount);
1731
1732/*
1733 * process a list of expirable mountpoints with the intent of discarding any
1734 * mountpoints that aren't in use and haven't been touched since last we came
1735 * here
1736 */
1737void mark_mounts_for_expiry(struct list_head *mounts)
1738{
1739        struct vfsmount *mnt, *next;
1740        LIST_HEAD(graveyard);
1741        LIST_HEAD(umounts);
1742
1743        if (list_empty(mounts))
1744                return;
1745
1746        down_write(&namespace_sem);
1747        spin_lock(&vfsmount_lock);
1748
1749        /* extract from the expiration list every vfsmount that matches the
1750         * following criteria:
1751         * - only referenced by its parent vfsmount
1752         * - still marked for expiry (marked on the last call here; marks are
1753         *   cleared by mntput())
1754         */
1755        list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1756                if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1757                        propagate_mount_busy(mnt, 1))
1758                        continue;
1759                list_move(&mnt->mnt_expire, &graveyard);
1760        }
1761        while (!list_empty(&graveyard)) {
1762                mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
1763                touch_mnt_namespace(mnt->mnt_ns);
1764                umount_tree(mnt, 1, &umounts);
1765        }
1766        spin_unlock(&vfsmount_lock);
1767        up_write(&namespace_sem);
1768
1769        release_mounts(&umounts);
1770}
1771
1772EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1773
1774/*
1775 * Ripoff of 'select_parent()'
1776 *
1777 * search the list of submounts for a given mountpoint, and move any
1778 * shrinkable submounts to the 'graveyard' list.
1779 */
1780static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1781{
1782        struct vfsmount *this_parent = parent;
1783        struct list_head *next;
1784        int found = 0;
1785
1786repeat:
1787        next = this_parent->mnt_mounts.next;
1788resume:
1789        while (next != &this_parent->mnt_mounts) {
1790                struct list_head *tmp = next;
1791                struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1792
1793                next = tmp->next;
1794                if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1795                        continue;
1796                /*
1797                 * Descend a level if the d_mounts list is non-empty.
1798                 */
1799                if (!list_empty(&mnt->mnt_mounts)) {
1800                        this_parent = mnt;
1801                        goto repeat;
1802                }
1803
1804                if (!propagate_mount_busy(mnt, 1)) {
1805                        list_move_tail(&mnt->mnt_expire, graveyard);
1806                        found++;
1807                }
1808        }
1809        /*
1810         * All done at this level ... ascend and resume the search
1811         */
1812        if (this_parent != parent) {
1813                next = this_parent->mnt_child.next;
1814                this_parent = this_parent->mnt_parent;
1815                goto resume;
1816        }
1817        return found;
1818}
1819
1820/*
1821 * process a list of expirable mountpoints with the intent of discarding any
1822 * submounts of a specific parent mountpoint
1823 */
1824static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
1825{
1826        LIST_HEAD(graveyard);
1827        struct vfsmount *m;
1828
1829        /* extract submounts of 'mountpoint' from the expiration list */
1830        while (select_submounts(mnt, &graveyard)) {
1831                while (!list_empty(&graveyard)) {
1832                        m = list_first_entry(&graveyard, struct vfsmount,
1833                                                mnt_expire);
1834                        touch_mnt_namespace(m->mnt_ns);
1835                        umount_tree(m, 1, umounts);
1836                }
1837        }
1838}
1839
1840/*
1841 * Some copy_from_user() implementations do not return the exact number of
1842 * bytes remaining to copy on a fault.  But copy_mount_options() requires that.
1843 * Note that this function differs from copy_from_user() in that it will oops
1844 * on bad values of `to', rather than returning a short copy.
1845 */
1846static long exact_copy_from_user(void *to, const void __user * from,
1847                                 unsigned long n)
1848{
1849        char *t = to;
1850        const char __user *f = from;
1851        char c;
1852
1853        if (!access_ok(VERIFY_READ, from, n))
1854                return n;
1855
1856        while (n) {
1857                if (__get_user(c, f)) {
1858                        memset(t, 0, n);
1859                        break;
1860                }
1861                *t++ = c;
1862                f++;
1863                n--;
1864        }
1865        return n;
1866}
1867
1868int copy_mount_options(const void __user * data, unsigned long *where)
1869{
1870        int i;
1871        unsigned long page;
1872        unsigned long size;
1873
1874        *where = 0;
1875        if (!data)
1876                return 0;
1877
1878        if (!(page = __get_free_page(GFP_KERNEL)))
1879                return -ENOMEM;
1880
1881        /* We only care that *some* data at the address the user
1882         * gave us is valid.  Just in case, we'll zero
1883         * the remainder of the page.
1884         */
1885        /* copy_from_user cannot cross TASK_SIZE ! */
1886        size = TASK_SIZE - (unsigned long)data;
1887        if (size > PAGE_SIZE)
1888                size = PAGE_SIZE;
1889
1890        i = size - exact_copy_from_user((void *)page, data, size);
1891        if (!i) {
1892                free_page(page);
1893                return -EFAULT;
1894        }
1895        if (i != PAGE_SIZE)
1896                memset((char *)page + i, 0, PAGE_SIZE - i);
1897        *where = page;
1898        return 0;
1899}
1900
1901/*
1902 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1903 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1904 *
1905 * data is a (void *) that can point to any structure up to
1906 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1907 * information (or be NULL).
1908 *
1909 * Pre-0.97 versions of mount() didn't have a flags word.
1910 * When the flags word was introduced its top half was required
1911 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1912 * Therefore, if this magic number is present, it carries no information
1913 * and must be discarded.
1914 */
1915long do_mount(char *dev_name, char *dir_name, char *type_page,
1916                  unsigned long flags, void *data_page)
1917{
1918        struct path path;
1919        int retval = 0;
1920        int mnt_flags = 0;
1921
1922        /* Discard magic */
1923        if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1924                flags &= ~MS_MGC_MSK;
1925
1926        /* Basic sanity checks */
1927
1928        if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1929                return -EINVAL;
1930        if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1931                return -EINVAL;
1932
1933        if (data_page)
1934                ((char *)data_page)[PAGE_SIZE - 1] = 0;
1935
1936        /* Default to relatime unless overriden */
1937        if (!(flags & MS_NOATIME))
1938                mnt_flags |= MNT_RELATIME;
1939
1940        /* Separate the per-mountpoint flags */
1941        if (flags & MS_NOSUID)
1942                mnt_flags |= MNT_NOSUID;
1943        if (flags & MS_NODEV)
1944                mnt_flags |= MNT_NODEV;
1945        if (flags & MS_NOEXEC)
1946                mnt_flags |= MNT_NOEXEC;
1947        if (flags & MS_NOATIME)
1948                mnt_flags |= MNT_NOATIME;
1949        if (flags & MS_NODIRATIME)
1950                mnt_flags |= MNT_NODIRATIME;
1951        if (flags & MS_STRICTATIME)
1952                mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
1953        if (flags & MS_RDONLY)
1954                mnt_flags |= MNT_READONLY;
1955
1956        flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
1957                   MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
1958                   MS_STRICTATIME);
1959
1960        /* ... and get the mountpoint */
1961        retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
1962        if (retval)
1963                return retval;
1964
1965        retval = security_sb_mount(dev_name, &path,
1966                                   type_page, flags, data_page);
1967        if (retval)
1968                goto dput_out;
1969
1970        if (flags & MS_REMOUNT)
1971                retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
1972                                    data_page);
1973        else if (flags & MS_BIND)
1974                retval = do_loopback(&path, dev_name, flags & MS_REC);
1975        else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1976                retval = do_change_type(&path, flags);
1977        else if (flags & MS_MOVE)
1978                retval = do_move_mount(&path, dev_name);
1979        else
1980                retval = do_new_mount(&path, type_page, flags, mnt_flags,
1981                                      dev_name, data_page);
1982dput_out:
1983        path_put(&path);
1984        return retval;
1985}
1986
1987/*
1988 * Allocate a new namespace structure and populate it with contents
1989 * copied from the namespace of the passed in task structure.
1990 */
1991static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
1992                struct fs_struct *fs)
1993{
1994        struct mnt_namespace *new_ns;
1995        struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
1996        struct vfsmount *p, *q;
1997
1998        new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
1999        if (!new_ns)
2000                return ERR_PTR(-ENOMEM);
2001
2002        atomic_set(&new_ns->count, 1);
2003        INIT_LIST_HEAD(&new_ns->list);
2004        init_waitqueue_head(&new_ns->poll);
2005        new_ns->event = 0;
2006
2007        down_write(&namespace_sem);
2008        /* First pass: copy the tree topology */
2009        new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
2010                                        CL_COPY_ALL | CL_EXPIRE);
2011        if (!new_ns->root) {
2012                up_write(&namespace_sem);
2013                kfree(new_ns);
2014                return ERR_PTR(-ENOMEM);
2015        }
2016        spin_lock(&vfsmount_lock);
2017        list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
2018        spin_unlock(&vfsmount_lock);
2019
2020        /*
2021         * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2022         * as belonging to new namespace.  We have already acquired a private
2023         * fs_struct, so tsk->fs->lock is not needed.
2024         */
2025        p = mnt_ns->root;
2026        q = new_ns->root;
2027        while (p) {
2028                q->mnt_ns = new_ns;
2029                if (fs) {
2030                        if (p == fs->root.mnt) {
2031                                rootmnt = p;
2032                                fs->root.mnt = mntget(q);
2033                        }
2034                        if (p == fs->pwd.mnt) {
2035                                pwdmnt = p;
2036                                fs->pwd.mnt = mntget(q);
2037                        }
2038                }
2039                p = next_mnt(p, mnt_ns->root);
2040                q = next_mnt(q, new_ns->root);
2041        }
2042        up_write(&namespace_sem);
2043
2044        if (rootmnt)
2045                mntput(rootmnt);
2046        if (pwdmnt)
2047                mntput(pwdmnt);
2048
2049        return new_ns;
2050}
2051
2052struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2053                struct fs_struct *new_fs)
2054{
2055        struct mnt_namespace *new_ns;
2056
2057        BUG_ON(!ns);
2058        get_mnt_ns(ns);
2059
2060        if (!(flags & CLONE_NEWNS))
2061                return ns;
2062
2063        new_ns = dup_mnt_ns(ns, new_fs);
2064
2065        put_mnt_ns(ns);
2066        return new_ns;
2067}
2068
2069SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2070                char __user *, type, unsigned long, flags, void __user *, data)
2071{
2072        int retval;
2073        unsigned long data_page;
2074        unsigned long type_page;
2075        unsigned long dev_page;
2076        char *dir_page;
2077
2078        retval = copy_mount_options(type, &type_page);
2079        if (retval < 0)
2080                return retval;
2081
2082        dir_page = getname(dir_name);
2083        retval = PTR_ERR(dir_page);
2084        if (IS_ERR(dir_page))
2085                goto out1;
2086
2087        retval = copy_mount_options(dev_name, &dev_page);
2088        if (retval < 0)
2089                goto out2;
2090
2091        retval = copy_mount_options(data, &data_page);
2092        if (retval < 0)
2093                goto out3;
2094
2095        lock_kernel();
2096        retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
2097                          flags, (void *)data_page);
2098        unlock_kernel();
2099        free_page(data_page);
2100
2101out3:
2102        free_page(dev_page);
2103out2:
2104        putname(dir_page);
2105out1:
2106        free_page(type_page);
2107        return retval;
2108}
2109
2110/*
2111 * pivot_root Semantics:
2112 * Moves the root file system of the current process to the directory put_old,
2113 * makes new_root as the new root file system of the current process, and sets
2114 * root/cwd of all processes which had them on the current root to new_root.
2115 *
2116 * Restrictions:
2117 * The new_root and put_old must be directories, and  must not be on the
2118 * same file  system as the current process root. The put_old  must  be
2119 * underneath new_root,  i.e. adding a non-zero number of /.. to the string
2120 * pointed to by put_old must yield the same directory as new_root. No other
2121 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2122 *
2123 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2124 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2125 * in this situation.
2126 *
2127 * Notes:
2128 *  - we don't move root/cwd if they are not at the root (reason: if something
2129 *    cared enough to change them, it's probably wrong to force them elsewhere)
2130 *  - it's okay to pick a root that isn't the root of a file system, e.g.
2131 *    /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2132 *    though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2133 *    first.
2134 */
2135SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2136                const char __user *, put_old)
2137{
2138        struct vfsmount *tmp;
2139        struct path new, old, parent_path, root_parent, root;
2140        int error;
2141
2142        if (!capable(CAP_SYS_ADMIN))
2143                return -EPERM;
2144
2145        error = user_path_dir(new_root, &new);
2146        if (error)
2147                goto out0;
2148        error = -EINVAL;
2149        if (!check_mnt(new.mnt))
2150                goto out1;
2151
2152        error = user_path_dir(put_old, &old);
2153        if (error)
2154                goto out1;
2155
2156        error = security_sb_pivotroot(&old, &new);
2157        if (error) {
2158                path_put(&old);
2159                goto out1;
2160        }
2161
2162        read_lock(&current->fs->lock);
2163        root = current->fs->root;
2164        path_get(&current->fs->root);
2165        read_unlock(&current->fs->lock);
2166        down_write(&namespace_sem);
2167        mutex_lock(&old.dentry->d_inode->i_mutex);
2168        error = -EINVAL;
2169        if (IS_MNT_SHARED(old.mnt) ||
2170                IS_MNT_SHARED(new.mnt->mnt_parent) ||
2171                IS_MNT_SHARED(root.mnt->mnt_parent))
2172                goto out2;
2173        if (!check_mnt(root.mnt))
2174                goto out2;
2175        error = -ENOENT;
2176        if (IS_DEADDIR(new.dentry->d_inode))
2177                goto out2;
2178        if (d_unhashed(new.dentry) && !IS_ROOT(new.dentry))
2179                goto out2;
2180        if (d_unhashed(old.dentry) && !IS_ROOT(old.dentry))
2181                goto out2;
2182        error = -EBUSY;
2183        if (new.mnt == root.mnt ||
2184            old.mnt == root.mnt)
2185                goto out2; /* loop, on the same file system  */
2186        error = -EINVAL;
2187        if (root.mnt->mnt_root != root.dentry)
2188                goto out2; /* not a mountpoint */
2189        if (root.mnt->mnt_parent == root.mnt)
2190                goto out2; /* not attached */
2191        if (new.mnt->mnt_root != new.dentry)
2192                goto out2; /* not a mountpoint */
2193        if (new.mnt->mnt_parent == new.mnt)
2194                goto out2; /* not attached */
2195        /* make sure we can reach put_old from new_root */
2196        tmp = old.mnt;
2197        spin_lock(&vfsmount_lock);
2198        if (tmp != new.mnt) {
2199                for (;;) {
2200                        if (tmp->mnt_parent == tmp)
2201                                goto out3; /* already mounted on put_old */
2202                        if (tmp->mnt_parent == new.mnt)
2203                                break;
2204                        tmp = tmp->mnt_parent;
2205                }
2206                if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
2207                        goto out3;
2208        } else if (!is_subdir(old.dentry, new.dentry))
2209                goto out3;
2210        detach_mnt(new.mnt, &parent_path);
2211        detach_mnt(root.mnt, &root_parent);
2212        /* mount old root on put_old */
2213        attach_mnt(root.mnt, &old);
2214        /* mount new_root on / */
2215        attach_mnt(new.mnt, &root_parent);
2216        touch_mnt_namespace(current->nsproxy->mnt_ns);
2217        spin_unlock(&vfsmount_lock);
2218        chroot_fs_refs(&root, &new);
2219        security_sb_post_pivotroot(&root, &new);
2220        error = 0;
2221        path_put(&root_parent);
2222        path_put(&parent_path);
2223out2:
2224        mutex_unlock(&old.dentry->d_inode->i_mutex);
2225        up_write(&namespace_sem);
2226        path_put(&root);
2227        path_put(&old);
2228out1:
2229        path_put(&new);
2230out0:
2231        return error;
2232out3:
2233        spin_unlock(&vfsmount_lock);
2234        goto out2;
2235}
2236
2237static void __init init_mount_tree(void)
2238{
2239        struct vfsmount *mnt;
2240        struct mnt_namespace *ns;
2241        struct path root;
2242
2243        mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2244        if (IS_ERR(mnt))
2245                panic("Can't create rootfs");
2246        ns = kmalloc(sizeof(*ns), GFP_KERNEL);
2247        if (!ns)
2248                panic("Can't allocate initial namespace");
2249        atomic_set(&ns->count, 1);
2250        INIT_LIST_HEAD(&ns->list);
2251        init_waitqueue_head(&ns->poll);
2252        ns->event = 0;
2253        list_add(&mnt->mnt_list, &ns->list);
2254        ns->root = mnt;
2255        mnt->mnt_ns = ns;
2256
2257        init_task.nsproxy->mnt_ns = ns;
2258        get_mnt_ns(ns);
2259
2260        root.mnt = ns->root;
2261        root.dentry = ns->root->mnt_root;
2262
2263        set_fs_pwd(current->fs, &root);
2264        set_fs_root(current->fs, &root);
2265}
2266
2267void __init mnt_init(void)
2268{
2269        unsigned u;
2270        int err;
2271
2272        init_rwsem(&namespace_sem);
2273
2274        mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
2275                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2276
2277        mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2278
2279        if (!mount_hashtable)
2280                panic("Failed to allocate mount hash table\n");
2281
2282        printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
2283
2284        for (u = 0; u < HASH_SIZE; u++)
2285                INIT_LIST_HEAD(&mount_hashtable[u]);
2286
2287        err = sysfs_init();
2288        if (err)
2289                printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2290                        __func__, err);
2291        fs_kobj = kobject_create_and_add("fs", NULL);
2292        if (!fs_kobj)
2293                printk(KERN_WARNING "%s: kobj create error\n", __func__);
2294        init_rootfs();
2295        init_mount_tree();
2296}
2297
2298void __put_mnt_ns(struct mnt_namespace *ns)
2299{
2300        struct vfsmount *root = ns->root;
2301        LIST_HEAD(umount_list);
2302        ns->root = NULL;
2303        spin_unlock(&vfsmount_lock);
2304        down_write(&namespace_sem);
2305        spin_lock(&vfsmount_lock);
2306        umount_tree(root, 0, &umount_list);
2307        spin_unlock(&vfsmount_lock);
2308        up_write(&namespace_sem);
2309        release_mounts(&umount_list);
2310        kfree(ns);
2311}
2312