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 <asm/uaccess.h>
  31#include <asm/unistd.h>
  32#include "pnode.h"
  33#include "internal.h"
  34
  35#define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
  36#define HASH_SIZE (1UL << HASH_SHIFT)
  37
  38/* spinlock for vfsmount related operations, inplace of dcache_lock */
  39__cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
  40
  41static int event;
  42static DEFINE_IDA(mnt_id_ida);
  43static DEFINE_IDA(mnt_group_ida);
  44
  45static struct list_head *mount_hashtable __read_mostly;
  46static struct kmem_cache *mnt_cache __read_mostly;
  47static struct rw_semaphore namespace_sem;
  48
  49/* /sys/fs */
  50struct kobject *fs_kobj;
  51EXPORT_SYMBOL_GPL(fs_kobj);
  52
  53static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
  54{
  55        unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
  56        tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
  57        tmp = tmp + (tmp >> HASH_SHIFT);
  58        return tmp & (HASH_SIZE - 1);
  59}
  60
  61#define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
  62
  63/* allocation is serialized by namespace_sem */
  64static int mnt_alloc_id(struct vfsmount *mnt)
  65{
  66        int res;
  67
  68retry:
  69        ida_pre_get(&mnt_id_ida, GFP_KERNEL);
  70        spin_lock(&vfsmount_lock);
  71        res = ida_get_new(&mnt_id_ida, &mnt->mnt_id);
  72        spin_unlock(&vfsmount_lock);
  73        if (res == -EAGAIN)
  74                goto retry;
  75
  76        return res;
  77}
  78
  79static void mnt_free_id(struct vfsmount *mnt)
  80{
  81        spin_lock(&vfsmount_lock);
  82        ida_remove(&mnt_id_ida, mnt->mnt_id);
  83        spin_unlock(&vfsmount_lock);
  84}
  85
  86/*
  87 * Allocate a new peer group ID
  88 *
  89 * mnt_group_ida is protected by namespace_sem
  90 */
  91static int mnt_alloc_group_id(struct vfsmount *mnt)
  92{
  93        if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
  94                return -ENOMEM;
  95
  96        return ida_get_new_above(&mnt_group_ida, 1, &mnt->mnt_group_id);
  97}
  98
  99/*
 100 * Release a peer group ID
 101 */
 102void mnt_release_group_id(struct vfsmount *mnt)
 103{
 104        ida_remove(&mnt_group_ida, mnt->mnt_group_id);
 105        mnt->mnt_group_id = 0;
 106}
 107
 108struct vfsmount *alloc_vfsmnt(const char *name)
 109{
 110        struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
 111        if (mnt) {
 112                int err;
 113
 114                err = mnt_alloc_id(mnt);
 115                if (err)
 116                        goto out_free_cache;
 117
 118                if (name) {
 119                        mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
 120                        if (!mnt->mnt_devname)
 121                                goto out_free_id;
 122                }
 123
 124                atomic_set(&mnt->mnt_count, 1);
 125                INIT_LIST_HEAD(&mnt->mnt_hash);
 126                INIT_LIST_HEAD(&mnt->mnt_child);
 127                INIT_LIST_HEAD(&mnt->mnt_mounts);
 128                INIT_LIST_HEAD(&mnt->mnt_list);
 129                INIT_LIST_HEAD(&mnt->mnt_expire);
 130                INIT_LIST_HEAD(&mnt->mnt_share);
 131                INIT_LIST_HEAD(&mnt->mnt_slave_list);
 132                INIT_LIST_HEAD(&mnt->mnt_slave);
 133                atomic_set(&mnt->__mnt_writers, 0);
 134        }
 135        return mnt;
 136
 137out_free_id:
 138        mnt_free_id(mnt);
 139out_free_cache:
 140        kmem_cache_free(mnt_cache, mnt);
 141        return NULL;
 142}
 143
 144/*
 145 * Most r/o checks on a fs are for operations that take
 146 * discrete amounts of time, like a write() or unlink().
 147 * We must keep track of when those operations start
 148 * (for permission checks) and when they end, so that
 149 * we can determine when writes are able to occur to
 150 * a filesystem.
 151 */
 152/*
 153 * __mnt_is_readonly: check whether a mount is read-only
 154 * @mnt: the mount to check for its write status
 155 *
 156 * This shouldn't be used directly ouside of the VFS.
 157 * It does not guarantee that the filesystem will stay
 158 * r/w, just that it is right *now*.  This can not and
 159 * should not be used in place of IS_RDONLY(inode).
 160 * mnt_want/drop_write() will _keep_ the filesystem
 161 * r/w.
 162 */
 163int __mnt_is_readonly(struct vfsmount *mnt)
 164{
 165        if (mnt->mnt_flags & MNT_READONLY)
 166                return 1;
 167        if (mnt->mnt_sb->s_flags & MS_RDONLY)
 168                return 1;
 169        return 0;
 170}
 171EXPORT_SYMBOL_GPL(__mnt_is_readonly);
 172
 173struct mnt_writer {
 174        /*
 175         * If holding multiple instances of this lock, they
 176         * must be ordered by cpu number.
 177         */
 178        spinlock_t lock;
 179        struct lock_class_key lock_class; /* compiles out with !lockdep */
 180        unsigned long count;
 181        struct vfsmount *mnt;
 182} ____cacheline_aligned_in_smp;
 183static DEFINE_PER_CPU(struct mnt_writer, mnt_writers);
 184
 185static int __init init_mnt_writers(void)
 186{
 187        int cpu;
 188        for_each_possible_cpu(cpu) {
 189                struct mnt_writer *writer = &per_cpu(mnt_writers, cpu);
 190                spin_lock_init(&writer->lock);
 191                lockdep_set_class(&writer->lock, &writer->lock_class);
 192                writer->count = 0;
 193        }
 194        return 0;
 195}
 196fs_initcall(init_mnt_writers);
 197
 198static void unlock_mnt_writers(void)
 199{
 200        int cpu;
 201        struct mnt_writer *cpu_writer;
 202
 203        for_each_possible_cpu(cpu) {
 204                cpu_writer = &per_cpu(mnt_writers, cpu);
 205                spin_unlock(&cpu_writer->lock);
 206        }
 207}
 208
 209static inline void __clear_mnt_count(struct mnt_writer *cpu_writer)
 210{
 211        if (!cpu_writer->mnt)
 212                return;
 213        /*
 214         * This is in case anyone ever leaves an invalid,
 215         * old ->mnt and a count of 0.
 216         */
 217        if (!cpu_writer->count)
 218                return;
 219        atomic_add(cpu_writer->count, &cpu_writer->mnt->__mnt_writers);
 220        cpu_writer->count = 0;
 221}
 222 /*
 223 * must hold cpu_writer->lock
 224 */
 225static inline void use_cpu_writer_for_mount(struct mnt_writer *cpu_writer,
 226                                          struct vfsmount *mnt)
 227{
 228        if (cpu_writer->mnt == mnt)
 229                return;
 230        __clear_mnt_count(cpu_writer);
 231        cpu_writer->mnt = mnt;
 232}
 233
 234/*
 235 * Most r/o checks on a fs are for operations that take
 236 * discrete amounts of time, like a write() or unlink().
 237 * We must keep track of when those operations start
 238 * (for permission checks) and when they end, so that
 239 * we can determine when writes are able to occur to
 240 * a filesystem.
 241 */
 242/**
 243 * mnt_want_write - get write access to a mount
 244 * @mnt: the mount on which to take a write
 245 *
 246 * This tells the low-level filesystem that a write is
 247 * about to be performed to it, and makes sure that
 248 * writes are allowed before returning success.  When
 249 * the write operation is finished, mnt_drop_write()
 250 * must be called.  This is effectively a refcount.
 251 */
 252int mnt_want_write(struct vfsmount *mnt)
 253{
 254        int ret = 0;
 255        struct mnt_writer *cpu_writer;
 256
 257        cpu_writer = &get_cpu_var(mnt_writers);
 258        spin_lock(&cpu_writer->lock);
 259        if (__mnt_is_readonly(mnt)) {
 260                ret = -EROFS;
 261                goto out;
 262        }
 263        use_cpu_writer_for_mount(cpu_writer, mnt);
 264        cpu_writer->count++;
 265out:
 266        spin_unlock(&cpu_writer->lock);
 267        put_cpu_var(mnt_writers);
 268        return ret;
 269}
 270EXPORT_SYMBOL_GPL(mnt_want_write);
 271
 272static void lock_mnt_writers(void)
 273{
 274        int cpu;
 275        struct mnt_writer *cpu_writer;
 276
 277        for_each_possible_cpu(cpu) {
 278                cpu_writer = &per_cpu(mnt_writers, cpu);
 279                spin_lock(&cpu_writer->lock);
 280                __clear_mnt_count(cpu_writer);
 281                cpu_writer->mnt = NULL;
 282        }
 283}
 284
 285/*
 286 * These per-cpu write counts are not guaranteed to have
 287 * matched increments and decrements on any given cpu.
 288 * A file open()ed for write on one cpu and close()d on
 289 * another cpu will imbalance this count.  Make sure it
 290 * does not get too far out of whack.
 291 */
 292static void handle_write_count_underflow(struct vfsmount *mnt)
 293{
 294        if (atomic_read(&mnt->__mnt_writers) >=
 295            MNT_WRITER_UNDERFLOW_LIMIT)
 296                return;
 297        /*
 298         * It isn't necessary to hold all of the locks
 299         * at the same time, but doing it this way makes
 300         * us share a lot more code.
 301         */
 302        lock_mnt_writers();
 303        /*
 304         * vfsmount_lock is for mnt_flags.
 305         */
 306        spin_lock(&vfsmount_lock);
 307        /*
 308         * If coalescing the per-cpu writer counts did not
 309         * get us back to a positive writer count, we have
 310         * a bug.
 311         */
 312        if ((atomic_read(&mnt->__mnt_writers) < 0) &&
 313            !(mnt->mnt_flags & MNT_IMBALANCED_WRITE_COUNT)) {
 314                WARN(1, KERN_DEBUG "leak detected on mount(%p) writers "
 315                                "count: %d\n",
 316                        mnt, atomic_read(&mnt->__mnt_writers));
 317                /* use the flag to keep the dmesg spam down */
 318                mnt->mnt_flags |= MNT_IMBALANCED_WRITE_COUNT;
 319        }
 320        spin_unlock(&vfsmount_lock);
 321        unlock_mnt_writers();
 322}
 323
 324/**
 325 * mnt_drop_write - give up write access to a mount
 326 * @mnt: the mount on which to give up write access
 327 *
 328 * Tells the low-level filesystem that we are done
 329 * performing writes to it.  Must be matched with
 330 * mnt_want_write() call above.
 331 */
 332void mnt_drop_write(struct vfsmount *mnt)
 333{
 334        int must_check_underflow = 0;
 335        struct mnt_writer *cpu_writer;
 336
 337        cpu_writer = &get_cpu_var(mnt_writers);
 338        spin_lock(&cpu_writer->lock);
 339
 340        use_cpu_writer_for_mount(cpu_writer, mnt);
 341        if (cpu_writer->count > 0) {
 342                cpu_writer->count--;
 343        } else {
 344                must_check_underflow = 1;
 345                atomic_dec(&mnt->__mnt_writers);
 346        }
 347
 348        spin_unlock(&cpu_writer->lock);
 349        /*
 350         * Logically, we could call this each time,
 351         * but the __mnt_writers cacheline tends to
 352         * be cold, and makes this expensive.
 353         */
 354        if (must_check_underflow)
 355                handle_write_count_underflow(mnt);
 356        /*
 357         * This could be done right after the spinlock
 358         * is taken because the spinlock keeps us on
 359         * the cpu, and disables preemption.  However,
 360         * putting it here bounds the amount that
 361         * __mnt_writers can underflow.  Without it,
 362         * we could theoretically wrap __mnt_writers.
 363         */
 364        put_cpu_var(mnt_writers);
 365}
 366EXPORT_SYMBOL_GPL(mnt_drop_write);
 367
 368static int mnt_make_readonly(struct vfsmount *mnt)
 369{
 370        int ret = 0;
 371
 372        lock_mnt_writers();
 373        /*
 374         * With all the locks held, this value is stable
 375         */
 376        if (atomic_read(&mnt->__mnt_writers) > 0) {
 377                ret = -EBUSY;
 378                goto out;
 379        }
 380        /*
 381         * nobody can do a successful mnt_want_write() with all
 382         * of the counts in MNT_DENIED_WRITE and the locks held.
 383         */
 384        spin_lock(&vfsmount_lock);
 385        if (!ret)
 386                mnt->mnt_flags |= MNT_READONLY;
 387        spin_unlock(&vfsmount_lock);
 388out:
 389        unlock_mnt_writers();
 390        return ret;
 391}
 392
 393static void __mnt_unmake_readonly(struct vfsmount *mnt)
 394{
 395        spin_lock(&vfsmount_lock);
 396        mnt->mnt_flags &= ~MNT_READONLY;
 397        spin_unlock(&vfsmount_lock);
 398}
 399
 400int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
 401{
 402        mnt->mnt_sb = sb;
 403        mnt->mnt_root = dget(sb->s_root);
 404        return 0;
 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                if (cpu_writer->mnt != mnt)
 618                        continue;
 619                spin_lock(&cpu_writer->lock);
 620                atomic_add(cpu_writer->count, &mnt->__mnt_writers);
 621                cpu_writer->count = 0;
 622                /*
 623                 * Might as well do this so that no one
 624                 * ever sees the pointer and expects
 625                 * it to be valid.
 626                 */
 627                cpu_writer->mnt = NULL;
 628                spin_unlock(&cpu_writer->lock);
 629        }
 630        /*
 631         * This probably indicates that somebody messed
 632         * up a mnt_want/drop_write() pair.  If this
 633         * happens, the filesystem was probably unable
 634         * to make r/w->r/o transitions.
 635         */
 636        WARN_ON(atomic_read(&mnt->__mnt_writers));
 637        dput(mnt->mnt_root);
 638        free_vfsmnt(mnt);
 639        deactivate_super(sb);
 640}
 641
 642void mntput_no_expire(struct vfsmount *mnt)
 643{
 644repeat:
 645        if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
 646                if (likely(!mnt->mnt_pinned)) {
 647                        spin_unlock(&vfsmount_lock);
 648                        __mntput(mnt);
 649                        return;
 650                }
 651                atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
 652                mnt->mnt_pinned = 0;
 653                spin_unlock(&vfsmount_lock);
 654                acct_auto_close_mnt(mnt);
 655                security_sb_umount_close(mnt);
 656                goto repeat;
 657        }
 658}
 659
 660EXPORT_SYMBOL(mntput_no_expire);
 661
 662void mnt_pin(struct vfsmount *mnt)
 663{
 664        spin_lock(&vfsmount_lock);
 665        mnt->mnt_pinned++;
 666        spin_unlock(&vfsmount_lock);
 667}
 668
 669EXPORT_SYMBOL(mnt_pin);
 670
 671void mnt_unpin(struct vfsmount *mnt)
 672{
 673        spin_lock(&vfsmount_lock);
 674        if (mnt->mnt_pinned) {
 675                atomic_inc(&mnt->mnt_count);
 676                mnt->mnt_pinned--;
 677        }
 678        spin_unlock(&vfsmount_lock);
 679}
 680
 681EXPORT_SYMBOL(mnt_unpin);
 682
 683static inline void mangle(struct seq_file *m, const char *s)
 684{
 685        seq_escape(m, s, " \t\n\\");
 686}
 687
 688/*
 689 * Simple .show_options callback for filesystems which don't want to
 690 * implement more complex mount option showing.
 691 *
 692 * See also save_mount_options().
 693 */
 694int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
 695{
 696        const char *options = mnt->mnt_sb->s_options;
 697
 698        if (options != NULL && options[0]) {
 699                seq_putc(m, ',');
 700                mangle(m, options);
 701        }
 702
 703        return 0;
 704}
 705EXPORT_SYMBOL(generic_show_options);
 706
 707/*
 708 * If filesystem uses generic_show_options(), this function should be
 709 * called from the fill_super() callback.
 710 *
 711 * The .remount_fs callback usually needs to be handled in a special
 712 * way, to make sure, that previous options are not overwritten if the
 713 * remount fails.
 714 *
 715 * Also note, that if the filesystem's .remount_fs function doesn't
 716 * reset all options to their default value, but changes only newly
 717 * given options, then the displayed options will not reflect reality
 718 * any more.
 719 */
 720void save_mount_options(struct super_block *sb, char *options)
 721{
 722        kfree(sb->s_options);
 723        sb->s_options = kstrdup(options, GFP_KERNEL);
 724}
 725EXPORT_SYMBOL(save_mount_options);
 726
 727#ifdef CONFIG_PROC_FS
 728/* iterator */
 729static void *m_start(struct seq_file *m, loff_t *pos)
 730{
 731        struct proc_mounts *p = m->private;
 732
 733        down_read(&namespace_sem);
 734        return seq_list_start(&p->ns->list, *pos);
 735}
 736
 737static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 738{
 739        struct proc_mounts *p = m->private;
 740
 741        return seq_list_next(v, &p->ns->list, pos);
 742}
 743
 744static void m_stop(struct seq_file *m, void *v)
 745{
 746        up_read(&namespace_sem);
 747}
 748
 749struct proc_fs_info {
 750        int flag;
 751        const char *str;
 752};
 753
 754static int show_sb_opts(struct seq_file *m, struct super_block *sb)
 755{
 756        static const struct proc_fs_info fs_info[] = {
 757                { MS_SYNCHRONOUS, ",sync" },
 758                { MS_DIRSYNC, ",dirsync" },
 759                { MS_MANDLOCK, ",mand" },
 760                { 0, NULL }
 761        };
 762        const struct proc_fs_info *fs_infop;
 763
 764        for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
 765                if (sb->s_flags & fs_infop->flag)
 766                        seq_puts(m, fs_infop->str);
 767        }
 768
 769        return security_sb_show_options(m, sb);
 770}
 771
 772static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
 773{
 774        static const struct proc_fs_info mnt_info[] = {
 775                { MNT_NOSUID, ",nosuid" },
 776                { MNT_NODEV, ",nodev" },
 777                { MNT_NOEXEC, ",noexec" },
 778                { MNT_NOATIME, ",noatime" },
 779                { MNT_NODIRATIME, ",nodiratime" },
 780                { MNT_RELATIME, ",relatime" },
 781                { 0, NULL }
 782        };
 783        const struct proc_fs_info *fs_infop;
 784
 785        for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
 786                if (mnt->mnt_flags & fs_infop->flag)
 787                        seq_puts(m, fs_infop->str);
 788        }
 789}
 790
 791static void show_type(struct seq_file *m, struct super_block *sb)
 792{
 793        mangle(m, sb->s_type->name);
 794        if (sb->s_subtype && sb->s_subtype[0]) {
 795                seq_putc(m, '.');
 796                mangle(m, sb->s_subtype);
 797        }
 798}
 799
 800static int show_vfsmnt(struct seq_file *m, void *v)
 801{
 802        struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
 803        int err = 0;
 804        struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
 805
 806        mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
 807        seq_putc(m, ' ');
 808        seq_path(m, &mnt_path, " \t\n\\");
 809        seq_putc(m, ' ');
 810        show_type(m, mnt->mnt_sb);
 811        seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
 812        err = show_sb_opts(m, mnt->mnt_sb);
 813        if (err)
 814                goto out;
 815        show_mnt_opts(m, mnt);
 816        if (mnt->mnt_sb->s_op->show_options)
 817                err = mnt->mnt_sb->s_op->show_options(m, mnt);
 818        seq_puts(m, " 0 0\n");
 819out:
 820        return err;
 821}
 822
 823const struct seq_operations mounts_op = {
 824        .start  = m_start,
 825        .next   = m_next,
 826        .stop   = m_stop,
 827        .show   = show_vfsmnt
 828};
 829
 830static int show_mountinfo(struct seq_file *m, void *v)
 831{
 832        struct proc_mounts *p = m->private;
 833        struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
 834        struct super_block *sb = mnt->mnt_sb;
 835        struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
 836        struct path root = p->root;
 837        int err = 0;
 838
 839        seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
 840                   MAJOR(sb->s_dev), MINOR(sb->s_dev));
 841        seq_dentry(m, mnt->mnt_root, " \t\n\\");
 842        seq_putc(m, ' ');
 843        seq_path_root(m, &mnt_path, &root, " \t\n\\");
 844        if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
 845                /*
 846                 * Mountpoint is outside root, discard that one.  Ugly,
 847                 * but less so than trying to do that in iterator in a
 848                 * race-free way (due to renames).
 849                 */
 850                return SEQ_SKIP;
 851        }
 852        seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
 853        show_mnt_opts(m, mnt);
 854
 855        /* Tagged fields ("foo:X" or "bar") */
 856        if (IS_MNT_SHARED(mnt))
 857                seq_printf(m, " shared:%i", mnt->mnt_group_id);
 858        if (IS_MNT_SLAVE(mnt)) {
 859                int master = mnt->mnt_master->mnt_group_id;
 860                int dom = get_dominating_id(mnt, &p->root);
 861                seq_printf(m, " master:%i", master);
 862                if (dom && dom != master)
 863                        seq_printf(m, " propagate_from:%i", dom);
 864        }
 865        if (IS_MNT_UNBINDABLE(mnt))
 866                seq_puts(m, " unbindable");
 867
 868        /* Filesystem specific data */
 869        seq_puts(m, " - ");
 870        show_type(m, sb);
 871        seq_putc(m, ' ');
 872        mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
 873        seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
 874        err = show_sb_opts(m, sb);
 875        if (err)
 876                goto out;
 877        if (sb->s_op->show_options)
 878                err = sb->s_op->show_options(m, mnt);
 879        seq_putc(m, '\n');
 880out:
 881        return err;
 882}
 883
 884const struct seq_operations mountinfo_op = {
 885        .start  = m_start,
 886        .next   = m_next,
 887        .stop   = m_stop,
 888        .show   = show_mountinfo,
 889};
 890
 891static int show_vfsstat(struct seq_file *m, void *v)
 892{
 893        struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
 894        struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
 895        int err = 0;
 896
 897        /* device */
 898        if (mnt->mnt_devname) {
 899                seq_puts(m, "device ");
 900                mangle(m, mnt->mnt_devname);
 901        } else
 902                seq_puts(m, "no device");
 903
 904        /* mount point */
 905        seq_puts(m, " mounted on ");
 906        seq_path(m, &mnt_path, " \t\n\\");
 907        seq_putc(m, ' ');
 908
 909        /* file system type */
 910        seq_puts(m, "with fstype ");
 911        show_type(m, mnt->mnt_sb);
 912
 913        /* optional statistics */
 914        if (mnt->mnt_sb->s_op->show_stats) {
 915                seq_putc(m, ' ');
 916                err = mnt->mnt_sb->s_op->show_stats(m, mnt);
 917        }
 918
 919        seq_putc(m, '\n');
 920        return err;
 921}
 922
 923const struct seq_operations mountstats_op = {
 924        .start  = m_start,
 925        .next   = m_next,
 926        .stop   = m_stop,
 927        .show   = show_vfsstat,
 928};
 929#endif  /* CONFIG_PROC_FS */
 930
 931/**
 932 * may_umount_tree - check if a mount tree is busy
 933 * @mnt: root of mount tree
 934 *
 935 * This is called to check if a tree of mounts has any
 936 * open files, pwds, chroots or sub mounts that are
 937 * busy.
 938 */
 939int may_umount_tree(struct vfsmount *mnt)
 940{
 941        int actual_refs = 0;
 942        int minimum_refs = 0;
 943        struct vfsmount *p;
 944
 945        spin_lock(&vfsmount_lock);
 946        for (p = mnt; p; p = next_mnt(p, mnt)) {
 947                actual_refs += atomic_read(&p->mnt_count);
 948                minimum_refs += 2;
 949        }
 950        spin_unlock(&vfsmount_lock);
 951
 952        if (actual_refs > minimum_refs)
 953                return 0;
 954
 955        return 1;
 956}
 957
 958EXPORT_SYMBOL(may_umount_tree);
 959
 960/**
 961 * may_umount - check if a mount point is busy
 962 * @mnt: root of mount
 963 *
 964 * This is called to check if a mount point has any
 965 * open files, pwds, chroots or sub mounts. If the
 966 * mount has sub mounts this will return busy
 967 * regardless of whether the sub mounts are busy.
 968 *
 969 * Doesn't take quota and stuff into account. IOW, in some cases it will
 970 * give false negatives. The main reason why it's here is that we need
 971 * a non-destructive way to look for easily umountable filesystems.
 972 */
 973int may_umount(struct vfsmount *mnt)
 974{
 975        int ret = 1;
 976        spin_lock(&vfsmount_lock);
 977        if (propagate_mount_busy(mnt, 2))
 978                ret = 0;
 979        spin_unlock(&vfsmount_lock);
 980        return ret;
 981}
 982
 983EXPORT_SYMBOL(may_umount);
 984
 985void release_mounts(struct list_head *head)
 986{
 987        struct vfsmount *mnt;
 988        while (!list_empty(head)) {
 989                mnt = list_first_entry(head, struct vfsmount, mnt_hash);
 990                list_del_init(&mnt->mnt_hash);
 991                if (mnt->mnt_parent != mnt) {
 992                        struct dentry *dentry;
 993                        struct vfsmount *m;
 994                        spin_lock(&vfsmount_lock);
 995                        dentry = mnt->mnt_mountpoint;
 996                        m = mnt->mnt_parent;
 997                        mnt->mnt_mountpoint = mnt->mnt_root;
 998                        mnt->mnt_parent = mnt;
 999                        m->mnt_ghosts--;
1000                        spin_unlock(&vfsmount_lock);
1001                        dput(dentry);
1002                        mntput(m);
1003                }
1004                mntput(mnt);
1005        }
1006}
1007
1008void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
1009{
1010        struct vfsmount *p;
1011
1012        for (p = mnt; p; p = next_mnt(p, mnt))
1013                list_move(&p->mnt_hash, kill);
1014
1015        if (propagate)
1016                propagate_umount(kill);
1017
1018        list_for_each_entry(p, kill, mnt_hash) {
1019                list_del_init(&p->mnt_expire);
1020                list_del_init(&p->mnt_list);
1021                __touch_mnt_namespace(p->mnt_ns);
1022                p->mnt_ns = NULL;
1023                list_del_init(&p->mnt_child);
1024                if (p->mnt_parent != p) {
1025                        p->mnt_parent->mnt_ghosts++;
1026                        p->mnt_mountpoint->d_mounted--;
1027                }
1028                change_mnt_propagation(p, MS_PRIVATE);
1029        }
1030}
1031
1032static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1033
1034static int do_umount(struct vfsmount *mnt, int flags)
1035{
1036        struct super_block *sb = mnt->mnt_sb;
1037        int retval;
1038        LIST_HEAD(umount_list);
1039
1040        retval = security_sb_umount(mnt, flags);
1041        if (retval)
1042                return retval;
1043
1044        /*
1045         * Allow userspace to request a mountpoint be expired rather than
1046         * unmounting unconditionally. Unmount only happens if:
1047         *  (1) the mark is already set (the mark is cleared by mntput())
1048         *  (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1049         */
1050        if (flags & MNT_EXPIRE) {
1051                if (mnt == current->fs->root.mnt ||
1052                    flags & (MNT_FORCE | MNT_DETACH))
1053                        return -EINVAL;
1054
1055                if (atomic_read(&mnt->mnt_count) != 2)
1056                        return -EBUSY;
1057
1058                if (!xchg(&mnt->mnt_expiry_mark, 1))
1059                        return -EAGAIN;
1060        }
1061
1062        /*
1063         * If we may have to abort operations to get out of this
1064         * mount, and they will themselves hold resources we must
1065         * allow the fs to do things. In the Unix tradition of
1066         * 'Gee thats tricky lets do it in userspace' the umount_begin
1067         * might fail to complete on the first run through as other tasks
1068         * must return, and the like. Thats for the mount program to worry
1069         * about for the moment.
1070         */
1071
1072        if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1073                lock_kernel();
1074                sb->s_op->umount_begin(sb);
1075                unlock_kernel();
1076        }
1077
1078        /*
1079         * No sense to grab the lock for this test, but test itself looks
1080         * somewhat bogus. Suggestions for better replacement?
1081         * Ho-hum... In principle, we might treat that as umount + switch
1082         * to rootfs. GC would eventually take care of the old vfsmount.
1083         * Actually it makes sense, especially if rootfs would contain a
1084         * /reboot - static binary that would close all descriptors and
1085         * call reboot(9). Then init(8) could umount root and exec /reboot.
1086         */
1087        if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1088                /*
1089                 * Special case for "unmounting" root ...
1090                 * we just try to remount it readonly.
1091                 */
1092                down_write(&sb->s_umount);
1093                if (!(sb->s_flags & MS_RDONLY)) {
1094                        lock_kernel();
1095                        retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1096                        unlock_kernel();
1097                }
1098                up_write(&sb->s_umount);
1099                return retval;
1100        }
1101
1102        down_write(&namespace_sem);
1103        spin_lock(&vfsmount_lock);
1104        event++;
1105
1106        if (!(flags & MNT_DETACH))
1107                shrink_submounts(mnt, &umount_list);
1108
1109        retval = -EBUSY;
1110        if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1111                if (!list_empty(&mnt->mnt_list))
1112                        umount_tree(mnt, 1, &umount_list);
1113                retval = 0;
1114        }
1115        spin_unlock(&vfsmount_lock);
1116        if (retval)
1117                security_sb_umount_busy(mnt);
1118        up_write(&namespace_sem);
1119        release_mounts(&umount_list);
1120        return retval;
1121}
1122
1123/*
1124 * Now umount can handle mount points as well as block devices.
1125 * This is important for filesystems which use unnamed block devices.
1126 *
1127 * We now support a flag for forced unmount like the other 'big iron'
1128 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1129 */
1130
1131SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1132{
1133        struct path path;
1134        int retval;
1135
1136        retval = user_path(name, &path);
1137        if (retval)
1138                goto out;
1139        retval = -EINVAL;
1140        if (path.dentry != path.mnt->mnt_root)
1141                goto dput_and_out;
1142        if (!check_mnt(path.mnt))
1143                goto dput_and_out;
1144
1145        retval = -EPERM;
1146        if (!capable(CAP_SYS_ADMIN))
1147                goto dput_and_out;
1148
1149        retval = do_umount(path.mnt, flags);
1150dput_and_out:
1151        /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1152        dput(path.dentry);
1153        mntput_no_expire(path.mnt);
1154out:
1155        return retval;
1156}
1157
1158#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1159
1160/*
1161 *      The 2.0 compatible umount. No flags.
1162 */
1163SYSCALL_DEFINE1(oldumount, char __user *, name)
1164{
1165        return sys_umount(name, 0);
1166}
1167
1168#endif
1169
1170static int mount_is_safe(struct path *path)
1171{
1172        if (capable(CAP_SYS_ADMIN))
1173                return 0;
1174        return -EPERM;
1175#ifdef notyet
1176        if (S_ISLNK(path->dentry->d_inode->i_mode))
1177                return -EPERM;
1178        if (path->dentry->d_inode->i_mode & S_ISVTX) {
1179                if (current->uid != path->dentry->d_inode->i_uid)
1180                        return -EPERM;
1181        }
1182        if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1183                return -EPERM;
1184        return 0;
1185#endif
1186}
1187
1188struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
1189                                        int flag)
1190{
1191        struct vfsmount *res, *p, *q, *r, *s;
1192        struct path path;
1193
1194        if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1195                return NULL;
1196
1197        res = q = clone_mnt(mnt, dentry, flag);
1198        if (!q)
1199                goto Enomem;
1200        q->mnt_mountpoint = mnt->mnt_mountpoint;
1201
1202        p = mnt;
1203        list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1204                if (!is_subdir(r->mnt_mountpoint, dentry))
1205                        continue;
1206
1207                for (s = r; s; s = next_mnt(s, r)) {
1208                        if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1209                                s = skip_mnt_tree(s);
1210                                continue;
1211                        }
1212                        while (p != s->mnt_parent) {
1213                                p = p->mnt_parent;
1214                                q = q->mnt_parent;
1215                        }
1216                        p = s;
1217                        path.mnt = q;
1218                        path.dentry = p->mnt_mountpoint;
1219                        q = clone_mnt(p, p->mnt_root, flag);
1220                        if (!q)
1221                                goto Enomem;
1222                        spin_lock(&vfsmount_lock);
1223                        list_add_tail(&q->mnt_list, &res->mnt_list);
1224                        attach_mnt(q, &path);
1225                        spin_unlock(&vfsmount_lock);
1226                }
1227        }
1228        return res;
1229Enomem:
1230        if (res) {
1231                LIST_HEAD(umount_list);
1232                spin_lock(&vfsmount_lock);
1233                umount_tree(res, 0, &umount_list);
1234                spin_unlock(&vfsmount_lock);
1235                release_mounts(&umount_list);
1236        }
1237        return NULL;
1238}
1239
1240struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry)
1241{
1242        struct vfsmount *tree;
1243        down_write(&namespace_sem);
1244        tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE);
1245        up_write(&namespace_sem);
1246        return tree;
1247}
1248
1249void drop_collected_mounts(struct vfsmount *mnt)
1250{
1251        LIST_HEAD(umount_list);
1252        down_write(&namespace_sem);
1253        spin_lock(&vfsmount_lock);
1254        umount_tree(mnt, 0, &umount_list);
1255        spin_unlock(&vfsmount_lock);
1256        up_write(&namespace_sem);
1257        release_mounts(&umount_list);
1258}
1259
1260static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1261{
1262        struct vfsmount *p;
1263
1264        for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1265                if (p->mnt_group_id && !IS_MNT_SHARED(p))
1266                        mnt_release_group_id(p);
1267        }
1268}
1269
1270static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1271{
1272        struct vfsmount *p;
1273
1274        for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1275                if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1276                        int err = mnt_alloc_group_id(p);
1277                        if (err) {
1278                                cleanup_group_ids(mnt, p);
1279                                return err;
1280                        }
1281                }
1282        }
1283
1284        return 0;
1285}
1286
1287/*
1288 *  @source_mnt : mount tree to be attached
1289 *  @nd         : place the mount tree @source_mnt is attached
1290 *  @parent_nd  : if non-null, detach the source_mnt from its parent and
1291 *                 store the parent mount and mountpoint dentry.
1292 *                 (done when source_mnt is moved)
1293 *
1294 *  NOTE: in the table below explains the semantics when a source mount
1295 *  of a given type is attached to a destination mount of a given type.
1296 * ---------------------------------------------------------------------------
1297 * |         BIND MOUNT OPERATION                                            |
1298 * |**************************************************************************
1299 * | source-->| shared        |       private  |       slave    | unbindable |
1300 * | dest     |               |                |                |            |
1301 * |   |      |               |                |                |            |
1302 * |   v      |               |                |                |            |
1303 * |**************************************************************************
1304 * |  shared  | shared (++)   |     shared (+) |     shared(+++)|  invalid   |
1305 * |          |               |                |                |            |
1306 * |non-shared| shared (+)    |      private   |      slave (*) |  invalid   |
1307 * ***************************************************************************
1308 * A bind operation clones the source mount and mounts the clone on the
1309 * destination mount.
1310 *
1311 * (++)  the cloned mount is propagated to all the mounts in the propagation
1312 *       tree of the destination mount and the cloned mount is added to
1313 *       the peer group of the source mount.
1314 * (+)   the cloned mount is created under the destination mount and is marked
1315 *       as shared. The cloned mount is added to the peer group of the source
1316 *       mount.
1317 * (+++) the mount is propagated to all the mounts in the propagation tree
1318 *       of the destination mount and the cloned mount is made slave
1319 *       of the same master as that of the source mount. The cloned mount
1320 *       is marked as 'shared and slave'.
1321 * (*)   the cloned mount is made a slave of the same master as that of the
1322 *       source mount.
1323 *
1324 * ---------------------------------------------------------------------------
1325 * |                    MOVE MOUNT OPERATION                                 |
1326 * |**************************************************************************
1327 * | source-->| shared        |       private  |       slave    | unbindable |
1328 * | dest     |               |                |                |            |
1329 * |   |      |               |                |                |            |
1330 * |   v      |               |                |                |            |
1331 * |**************************************************************************
1332 * |  shared  | shared (+)    |     shared (+) |    shared(+++) |  invalid   |
1333 * |          |               |                |                |            |
1334 * |non-shared| shared (+*)   |      private   |    slave (*)   | unbindable |
1335 * ***************************************************************************
1336 *
1337 * (+)  the mount is moved to the destination. And is then propagated to
1338 *      all the mounts in the propagation tree of the destination mount.
1339 * (+*)  the mount is moved to the destination.
1340 * (+++)  the mount is moved to the destination and is then propagated to
1341 *      all the mounts belonging to the destination mount's propagation tree.
1342 *      the mount is marked as 'shared and slave'.
1343 * (*)  the mount continues to be a slave at the new location.
1344 *
1345 * if the source mount is a tree, the operations explained above is
1346 * applied to each mount in the tree.
1347 * Must be called without spinlocks held, since this function can sleep
1348 * in allocations.
1349 */
1350static int attach_recursive_mnt(struct vfsmount *source_mnt,
1351                        struct path *path, struct path *parent_path)
1352{
1353        LIST_HEAD(tree_list);
1354        struct vfsmount *dest_mnt = path->mnt;
1355        struct dentry *dest_dentry = path->dentry;
1356        struct vfsmount *child, *p;
1357        int err;
1358
1359        if (IS_MNT_SHARED(dest_mnt)) {
1360                err = invent_group_ids(source_mnt, true);
1361                if (err)
1362                        goto out;
1363        }
1364        err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1365        if (err)
1366                goto out_cleanup_ids;
1367
1368        if (IS_MNT_SHARED(dest_mnt)) {
1369                for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1370                        set_mnt_shared(p);
1371        }
1372
1373        spin_lock(&vfsmount_lock);
1374        if (parent_path) {
1375                detach_mnt(source_mnt, parent_path);
1376                attach_mnt(source_mnt, path);
1377                touch_mnt_namespace(current->nsproxy->mnt_ns);
1378        } else {
1379                mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1380                commit_tree(source_mnt);
1381        }
1382
1383        list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1384                list_del_init(&child->mnt_hash);
1385                commit_tree(child);
1386        }
1387        spin_unlock(&vfsmount_lock);
1388        return 0;
1389
1390 out_cleanup_ids:
1391        if (IS_MNT_SHARED(dest_mnt))
1392                cleanup_group_ids(source_mnt, NULL);
1393 out:
1394        return err;
1395}
1396
1397static int graft_tree(struct vfsmount *mnt, struct path *path)
1398{
1399        int err;
1400        if (mnt->mnt_sb->s_flags & MS_NOUSER)
1401                return -EINVAL;
1402
1403        if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1404              S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1405                return -ENOTDIR;
1406
1407        err = -ENOENT;
1408        mutex_lock(&path->dentry->d_inode->i_mutex);
1409        if (IS_DEADDIR(path->dentry->d_inode))
1410                goto out_unlock;
1411
1412        err = security_sb_check_sb(mnt, path);
1413        if (err)
1414                goto out_unlock;
1415
1416        err = -ENOENT;
1417        if (IS_ROOT(path->dentry) || !d_unhashed(path->dentry))
1418                err = attach_recursive_mnt(mnt, path, NULL);
1419out_unlock:
1420        mutex_unlock(&path->dentry->d_inode->i_mutex);
1421        if (!err)
1422                security_sb_post_addmount(mnt, path);
1423        return err;
1424}
1425
1426/*
1427 * recursively change the type of the mountpoint.
1428 */
1429static int do_change_type(struct path *path, int flag)
1430{
1431        struct vfsmount *m, *mnt = path->mnt;
1432        int recurse = flag & MS_REC;
1433        int type = flag & ~MS_REC;
1434        int err = 0;
1435
1436        if (!capable(CAP_SYS_ADMIN))
1437                return -EPERM;
1438
1439        if (path->dentry != path->mnt->mnt_root)
1440                return -EINVAL;
1441
1442        down_write(&namespace_sem);
1443        if (type == MS_SHARED) {
1444                err = invent_group_ids(mnt, recurse);
1445                if (err)
1446                        goto out_unlock;
1447        }
1448
1449        spin_lock(&vfsmount_lock);
1450        for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1451                change_mnt_propagation(m, type);
1452        spin_unlock(&vfsmount_lock);
1453
1454 out_unlock:
1455        up_write(&namespace_sem);
1456        return err;
1457}
1458
1459/*
1460 * do loopback mount.
1461 */
1462static int do_loopback(struct path *path, char *old_name,
1463                                int recurse)
1464{
1465        struct path old_path;
1466        struct vfsmount *mnt = NULL;
1467        int err = mount_is_safe(path);
1468        if (err)
1469                return err;
1470        if (!old_name || !*old_name)
1471                return -EINVAL;
1472        err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1473        if (err)
1474                return err;
1475
1476        down_write(&namespace_sem);
1477        err = -EINVAL;
1478        if (IS_MNT_UNBINDABLE(old_path.mnt))
1479                goto out;
1480
1481        if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1482                goto out;
1483
1484        err = -ENOMEM;
1485        if (recurse)
1486                mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
1487        else
1488                mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
1489
1490        if (!mnt)
1491                goto out;
1492
1493        err = graft_tree(mnt, path);
1494        if (err) {
1495                LIST_HEAD(umount_list);
1496                spin_lock(&vfsmount_lock);
1497                umount_tree(mnt, 0, &umount_list);
1498                spin_unlock(&vfsmount_lock);
1499                release_mounts(&umount_list);
1500        }
1501
1502out:
1503        up_write(&namespace_sem);
1504        path_put(&old_path);
1505        return err;
1506}
1507
1508static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1509{
1510        int error = 0;
1511        int readonly_request = 0;
1512
1513        if (ms_flags & MS_RDONLY)
1514                readonly_request = 1;
1515        if (readonly_request == __mnt_is_readonly(mnt))
1516                return 0;
1517
1518        if (readonly_request)
1519                error = mnt_make_readonly(mnt);
1520        else
1521                __mnt_unmake_readonly(mnt);
1522        return error;
1523}
1524
1525/*
1526 * change filesystem flags. dir should be a physical root of filesystem.
1527 * If you've mounted a non-root directory somewhere and want to do remount
1528 * on it - tough luck.
1529 */
1530static int do_remount(struct path *path, int flags, int mnt_flags,
1531                      void *data)
1532{
1533        int err;
1534        struct super_block *sb = path->mnt->mnt_sb;
1535
1536        if (!capable(CAP_SYS_ADMIN))
1537                return -EPERM;
1538
1539        if (!check_mnt(path->mnt))
1540                return -EINVAL;
1541
1542        if (path->dentry != path->mnt->mnt_root)
1543                return -EINVAL;
1544
1545        down_write(&sb->s_umount);
1546        if (flags & MS_BIND)
1547                err = change_mount_flags(path->mnt, flags);
1548        else
1549                err = do_remount_sb(sb, flags, data, 0);
1550        if (!err)
1551                path->mnt->mnt_flags = mnt_flags;
1552        up_write(&sb->s_umount);
1553        if (!err) {
1554                security_sb_post_remount(path->mnt, flags, data);
1555
1556                spin_lock(&vfsmount_lock);
1557                touch_mnt_namespace(path->mnt->mnt_ns);
1558                spin_unlock(&vfsmount_lock);
1559        }
1560        return err;
1561}
1562
1563static inline int tree_contains_unbindable(struct vfsmount *mnt)
1564{
1565        struct vfsmount *p;
1566        for (p = mnt; p; p = next_mnt(p, mnt)) {
1567                if (IS_MNT_UNBINDABLE(p))
1568                        return 1;
1569        }
1570        return 0;
1571}
1572
1573static int do_move_mount(struct path *path, char *old_name)
1574{
1575        struct path old_path, parent_path;
1576        struct vfsmount *p;
1577        int err = 0;
1578        if (!capable(CAP_SYS_ADMIN))
1579                return -EPERM;
1580        if (!old_name || !*old_name)
1581                return -EINVAL;
1582        err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1583        if (err)
1584                return err;
1585
1586        down_write(&namespace_sem);
1587        while (d_mountpoint(path->dentry) &&
1588               follow_down(&path->mnt, &path->dentry))
1589                ;
1590        err = -EINVAL;
1591        if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
1592                goto out;
1593
1594        err = -ENOENT;
1595        mutex_lock(&path->dentry->d_inode->i_mutex);
1596        if (IS_DEADDIR(path->dentry->d_inode))
1597                goto out1;
1598
1599        if (!IS_ROOT(path->dentry) && d_unhashed(path->dentry))
1600                goto out1;
1601
1602        err = -EINVAL;
1603        if (old_path.dentry != old_path.mnt->mnt_root)
1604                goto out1;
1605
1606        if (old_path.mnt == old_path.mnt->mnt_parent)
1607                goto out1;
1608
1609        if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1610              S_ISDIR(old_path.dentry->d_inode->i_mode))
1611                goto out1;
1612        /*
1613         * Don't move a mount residing in a shared parent.
1614         */
1615        if (old_path.mnt->mnt_parent &&
1616            IS_MNT_SHARED(old_path.mnt->mnt_parent))
1617                goto out1;
1618        /*
1619         * Don't move a mount tree containing unbindable mounts to a destination
1620         * mount which is shared.
1621         */
1622        if (IS_MNT_SHARED(path->mnt) &&
1623            tree_contains_unbindable(old_path.mnt))
1624                goto out1;
1625        err = -ELOOP;
1626        for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1627                if (p == old_path.mnt)
1628                        goto out1;
1629
1630        err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
1631        if (err)
1632                goto out1;
1633
1634        /* if the mount is moved, it should no longer be expire
1635         * automatically */
1636        list_del_init(&old_path.mnt->mnt_expire);
1637out1:
1638        mutex_unlock(&path->dentry->d_inode->i_mutex);
1639out:
1640        up_write(&namespace_sem);
1641        if (!err)
1642                path_put(&parent_path);
1643        path_put(&old_path);
1644        return err;
1645}
1646
1647/*
1648 * create a new mount for userspace and request it to be added into the
1649 * namespace's tree
1650 */
1651static int do_new_mount(struct path *path, char *type, int flags,
1652                        int mnt_flags, char *name, void *data)
1653{
1654        struct vfsmount *mnt;
1655
1656        if (!type || !memchr(type, 0, PAGE_SIZE))
1657                return -EINVAL;
1658
1659        /* we need capabilities... */
1660        if (!capable(CAP_SYS_ADMIN))
1661                return -EPERM;
1662
1663        mnt = do_kern_mount(type, flags, name, data);
1664        if (IS_ERR(mnt))
1665                return PTR_ERR(mnt);
1666
1667        return do_add_mount(mnt, path, mnt_flags, NULL);
1668}
1669
1670/*
1671 * add a mount into a namespace's mount tree
1672 * - provide the option of adding the new mount to an expiration list
1673 */
1674int do_add_mount(struct vfsmount *newmnt, struct path *path,
1675                 int mnt_flags, struct list_head *fslist)
1676{
1677        int err;
1678
1679        down_write(&namespace_sem);
1680        /* Something was mounted here while we slept */
1681        while (d_mountpoint(path->dentry) &&
1682               follow_down(&path->mnt, &path->dentry))
1683                ;
1684        err = -EINVAL;
1685        if (!check_mnt(path->mnt))
1686                goto unlock;
1687
1688        /* Refuse the same filesystem on the same mount point */
1689        err = -EBUSY;
1690        if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1691            path->mnt->mnt_root == path->dentry)
1692                goto unlock;
1693
1694        err = -EINVAL;
1695        if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1696                goto unlock;
1697
1698        newmnt->mnt_flags = mnt_flags;
1699        if ((err = graft_tree(newmnt, path)))
1700                goto unlock;
1701
1702        if (fslist) /* add to the specified expiration list */
1703                list_add_tail(&newmnt->mnt_expire, fslist);
1704
1705        up_write(&namespace_sem);
1706        return 0;
1707
1708unlock:
1709        up_write(&namespace_sem);
1710        mntput(newmnt);
1711        return err;
1712}
1713
1714EXPORT_SYMBOL_GPL(do_add_mount);
1715
1716/*
1717 * process a list of expirable mountpoints with the intent of discarding any
1718 * mountpoints that aren't in use and haven't been touched since last we came
1719 * here
1720 */
1721void mark_mounts_for_expiry(struct list_head *mounts)
1722{
1723        struct vfsmount *mnt, *next;
1724        LIST_HEAD(graveyard);
1725        LIST_HEAD(umounts);
1726
1727        if (list_empty(mounts))
1728                return;
1729
1730        down_write(&namespace_sem);
1731        spin_lock(&vfsmount_lock);
1732
1733        /* extract from the expiration list every vfsmount that matches the
1734         * following criteria:
1735         * - only referenced by its parent vfsmount
1736         * - still marked for expiry (marked on the last call here; marks are
1737         *   cleared by mntput())
1738         */
1739        list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1740                if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1741                        propagate_mount_busy(mnt, 1))
1742                        continue;
1743                list_move(&mnt->mnt_expire, &graveyard);
1744        }
1745        while (!list_empty(&graveyard)) {
1746                mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
1747                touch_mnt_namespace(mnt->mnt_ns);
1748                umount_tree(mnt, 1, &umounts);
1749        }
1750        spin_unlock(&vfsmount_lock);
1751        up_write(&namespace_sem);
1752
1753        release_mounts(&umounts);
1754}
1755
1756EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1757
1758/*
1759 * Ripoff of 'select_parent()'
1760 *
1761 * search the list of submounts for a given mountpoint, and move any
1762 * shrinkable submounts to the 'graveyard' list.
1763 */
1764static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1765{
1766        struct vfsmount *this_parent = parent;
1767        struct list_head *next;
1768        int found = 0;
1769
1770repeat:
1771        next = this_parent->mnt_mounts.next;
1772resume:
1773        while (next != &this_parent->mnt_mounts) {
1774                struct list_head *tmp = next;
1775                struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1776
1777                next = tmp->next;
1778                if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1779                        continue;
1780                /*
1781                 * Descend a level if the d_mounts list is non-empty.
1782                 */
1783                if (!list_empty(&mnt->mnt_mounts)) {
1784                        this_parent = mnt;
1785                        goto repeat;
1786                }
1787
1788                if (!propagate_mount_busy(mnt, 1)) {
1789                        list_move_tail(&mnt->mnt_expire, graveyard);
1790                        found++;
1791                }
1792        }
1793        /*
1794         * All done at this level ... ascend and resume the search
1795         */
1796        if (this_parent != parent) {
1797                next = this_parent->mnt_child.next;
1798                this_parent = this_parent->mnt_parent;
1799                goto resume;
1800        }
1801        return found;
1802}
1803
1804/*
1805 * process a list of expirable mountpoints with the intent of discarding any
1806 * submounts of a specific parent mountpoint
1807 */
1808static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
1809{
1810        LIST_HEAD(graveyard);
1811        struct vfsmount *m;
1812
1813        /* extract submounts of 'mountpoint' from the expiration list */
1814        while (select_submounts(mnt, &graveyard)) {
1815                while (!list_empty(&graveyard)) {
1816                        m = list_first_entry(&graveyard, struct vfsmount,
1817                                                mnt_expire);
1818                        touch_mnt_namespace(m->mnt_ns);
1819                        umount_tree(m, 1, umounts);
1820                }
1821        }
1822}
1823
1824/*
1825 * Some copy_from_user() implementations do not return the exact number of
1826 * bytes remaining to copy on a fault.  But copy_mount_options() requires that.
1827 * Note that this function differs from copy_from_user() in that it will oops
1828 * on bad values of `to', rather than returning a short copy.
1829 */
1830static long exact_copy_from_user(void *to, const void __user * from,
1831                                 unsigned long n)
1832{
1833        char *t = to;
1834        const char __user *f = from;
1835        char c;
1836
1837        if (!access_ok(VERIFY_READ, from, n))
1838                return n;
1839
1840        while (n) {
1841                if (__get_user(c, f)) {
1842                        memset(t, 0, n);
1843                        break;
1844                }
1845                *t++ = c;
1846                f++;
1847                n--;
1848        }
1849        return n;
1850}
1851
1852int copy_mount_options(const void __user * data, unsigned long *where)
1853{
1854        int i;
1855        unsigned long page;
1856        unsigned long size;
1857
1858        *where = 0;
1859        if (!data)
1860                return 0;
1861
1862        if (!(page = __get_free_page(GFP_KERNEL)))
1863                return -ENOMEM;
1864
1865        /* We only care that *some* data at the address the user
1866         * gave us is valid.  Just in case, we'll zero
1867         * the remainder of the page.
1868         */
1869        /* copy_from_user cannot cross TASK_SIZE ! */
1870        size = TASK_SIZE - (unsigned long)data;
1871        if (size > PAGE_SIZE)
1872                size = PAGE_SIZE;
1873
1874        i = size - exact_copy_from_user((void *)page, data, size);
1875        if (!i) {
1876                free_page(page);
1877                return -EFAULT;
1878        }
1879        if (i != PAGE_SIZE)
1880                memset((char *)page + i, 0, PAGE_SIZE - i);
1881        *where = page;
1882        return 0;
1883}
1884
1885/*
1886 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1887 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1888 *
1889 * data is a (void *) that can point to any structure up to
1890 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1891 * information (or be NULL).
1892 *
1893 * Pre-0.97 versions of mount() didn't have a flags word.
1894 * When the flags word was introduced its top half was required
1895 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1896 * Therefore, if this magic number is present, it carries no information
1897 * and must be discarded.
1898 */
1899long do_mount(char *dev_name, char *dir_name, char *type_page,
1900                  unsigned long flags, void *data_page)
1901{
1902        struct path path;
1903        int retval = 0;
1904        int mnt_flags = 0;
1905
1906        /* Discard magic */
1907        if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1908                flags &= ~MS_MGC_MSK;
1909
1910        /* Basic sanity checks */
1911
1912        if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1913                return -EINVAL;
1914        if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1915                return -EINVAL;
1916
1917        if (data_page)
1918                ((char *)data_page)[PAGE_SIZE - 1] = 0;
1919
1920        /* Separate the per-mountpoint flags */
1921        if (flags & MS_NOSUID)
1922                mnt_flags |= MNT_NOSUID;
1923        if (flags & MS_NODEV)
1924                mnt_flags |= MNT_NODEV;
1925        if (flags & MS_NOEXEC)
1926                mnt_flags |= MNT_NOEXEC;
1927        if (flags & MS_NOATIME)
1928                mnt_flags |= MNT_NOATIME;
1929        if (flags & MS_NODIRATIME)
1930                mnt_flags |= MNT_NODIRATIME;
1931        if (flags & MS_RELATIME)
1932                mnt_flags |= MNT_RELATIME;
1933        if (flags & MS_RDONLY)
1934                mnt_flags |= MNT_READONLY;
1935
1936        flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
1937                   MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT);
1938
1939        /* ... and get the mountpoint */
1940        retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
1941        if (retval)
1942                return retval;
1943
1944        retval = security_sb_mount(dev_name, &path,
1945                                   type_page, flags, data_page);
1946        if (retval)
1947                goto dput_out;
1948
1949        if (flags & MS_REMOUNT)
1950                retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
1951                                    data_page);
1952        else if (flags & MS_BIND)
1953                retval = do_loopback(&path, dev_name, flags & MS_REC);
1954        else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1955                retval = do_change_type(&path, flags);
1956        else if (flags & MS_MOVE)
1957                retval = do_move_mount(&path, dev_name);
1958        else
1959                retval = do_new_mount(&path, type_page, flags, mnt_flags,
1960                                      dev_name, data_page);
1961dput_out:
1962        path_put(&path);
1963        return retval;
1964}
1965
1966/*
1967 * Allocate a new namespace structure and populate it with contents
1968 * copied from the namespace of the passed in task structure.
1969 */
1970static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
1971                struct fs_struct *fs)
1972{
1973        struct mnt_namespace *new_ns;
1974        struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
1975        struct vfsmount *p, *q;
1976
1977        new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
1978        if (!new_ns)
1979                return ERR_PTR(-ENOMEM);
1980
1981        atomic_set(&new_ns->count, 1);
1982        INIT_LIST_HEAD(&new_ns->list);
1983        init_waitqueue_head(&new_ns->poll);
1984        new_ns->event = 0;
1985
1986        down_write(&namespace_sem);
1987        /* First pass: copy the tree topology */
1988        new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
1989                                        CL_COPY_ALL | CL_EXPIRE);
1990        if (!new_ns->root) {
1991                up_write(&namespace_sem);
1992                kfree(new_ns);
1993                return ERR_PTR(-ENOMEM);;
1994        }
1995        spin_lock(&vfsmount_lock);
1996        list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1997        spin_unlock(&vfsmount_lock);
1998
1999        /*
2000         * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2001         * as belonging to new namespace.  We have already acquired a private
2002         * fs_struct, so tsk->fs->lock is not needed.
2003         */
2004        p = mnt_ns->root;
2005        q = new_ns->root;
2006        while (p) {
2007                q->mnt_ns = new_ns;
2008                if (fs) {
2009                        if (p == fs->root.mnt) {
2010                                rootmnt = p;
2011                                fs->root.mnt = mntget(q);
2012                        }
2013                        if (p == fs->pwd.mnt) {
2014                                pwdmnt = p;
2015                                fs->pwd.mnt = mntget(q);
2016                        }
2017                }
2018                p = next_mnt(p, mnt_ns->root);
2019                q = next_mnt(q, new_ns->root);
2020        }
2021        up_write(&namespace_sem);
2022
2023        if (rootmnt)
2024                mntput(rootmnt);
2025        if (pwdmnt)
2026                mntput(pwdmnt);
2027
2028        return new_ns;
2029}
2030
2031struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2032                struct fs_struct *new_fs)
2033{
2034        struct mnt_namespace *new_ns;
2035
2036        BUG_ON(!ns);
2037        get_mnt_ns(ns);
2038
2039        if (!(flags & CLONE_NEWNS))
2040                return ns;
2041
2042        new_ns = dup_mnt_ns(ns, new_fs);
2043
2044        put_mnt_ns(ns);
2045        return new_ns;
2046}
2047
2048SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2049                char __user *, type, unsigned long, flags, void __user *, data)
2050{
2051        int retval;
2052        unsigned long data_page;
2053        unsigned long type_page;
2054        unsigned long dev_page;
2055        char *dir_page;
2056
2057        retval = copy_mount_options(type, &type_page);
2058        if (retval < 0)
2059                return retval;
2060
2061        dir_page = getname(dir_name);
2062        retval = PTR_ERR(dir_page);
2063        if (IS_ERR(dir_page))
2064                goto out1;
2065
2066        retval = copy_mount_options(dev_name, &dev_page);
2067        if (retval < 0)
2068                goto out2;
2069
2070        retval = copy_mount_options(data, &data_page);
2071        if (retval < 0)
2072                goto out3;
2073
2074        lock_kernel();
2075        retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
2076                          flags, (void *)data_page);
2077        unlock_kernel();
2078        free_page(data_page);
2079
2080out3:
2081        free_page(dev_page);
2082out2:
2083        putname(dir_page);
2084out1:
2085        free_page(type_page);
2086        return retval;
2087}
2088
2089/*
2090 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
2091 * It can block. Requires the big lock held.
2092 */
2093void set_fs_root(struct fs_struct *fs, struct path *path)
2094{
2095        struct path old_root;
2096
2097        write_lock(&fs->lock);
2098        old_root = fs->root;
2099        fs->root = *path;
2100        path_get(path);
2101        write_unlock(&fs->lock);
2102        if (old_root.dentry)
2103                path_put(&old_root);
2104}
2105
2106/*
2107 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
2108 * It can block. Requires the big lock held.
2109 */
2110void set_fs_pwd(struct fs_struct *fs, struct path *path)
2111{
2112        struct path old_pwd;
2113
2114        write_lock(&fs->lock);
2115        old_pwd = fs->pwd;
2116        fs->pwd = *path;
2117        path_get(path);
2118        write_unlock(&fs->lock);
2119
2120        if (old_pwd.dentry)
2121                path_put(&old_pwd);
2122}
2123
2124static void chroot_fs_refs(struct path *old_root, struct path *new_root)
2125{
2126        struct task_struct *g, *p;
2127        struct fs_struct *fs;
2128
2129        read_lock(&tasklist_lock);
2130        do_each_thread(g, p) {
2131                task_lock(p);
2132                fs = p->fs;
2133                if (fs) {
2134                        atomic_inc(&fs->count);
2135                        task_unlock(p);
2136                        if (fs->root.dentry == old_root->dentry
2137                            && fs->root.mnt == old_root->mnt)
2138                                set_fs_root(fs, new_root);
2139                        if (fs->pwd.dentry == old_root->dentry
2140                            && fs->pwd.mnt == old_root->mnt)
2141                                set_fs_pwd(fs, new_root);
2142                        put_fs_struct(fs);
2143                } else
2144                        task_unlock(p);
2145        } while_each_thread(g, p);
2146        read_unlock(&tasklist_lock);
2147}
2148
2149/*
2150 * pivot_root Semantics:
2151 * Moves the root file system of the current process to the directory put_old,
2152 * makes new_root as the new root file system of the current process, and sets
2153 * root/cwd of all processes which had them on the current root to new_root.
2154 *
2155 * Restrictions:
2156 * The new_root and put_old must be directories, and  must not be on the
2157 * same file  system as the current process root. The put_old  must  be
2158 * underneath new_root,  i.e. adding a non-zero number of /.. to the string
2159 * pointed to by put_old must yield the same directory as new_root. No other
2160 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2161 *
2162 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2163 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2164 * in this situation.
2165 *
2166 * Notes:
2167 *  - we don't move root/cwd if they are not at the root (reason: if something
2168 *    cared enough to change them, it's probably wrong to force them elsewhere)
2169 *  - it's okay to pick a root that isn't the root of a file system, e.g.
2170 *    /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2171 *    though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2172 *    first.
2173 */
2174SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2175                const char __user *, put_old)
2176{
2177        struct vfsmount *tmp;
2178        struct path new, old, parent_path, root_parent, root;
2179        int error;
2180
2181        if (!capable(CAP_SYS_ADMIN))
2182                return -EPERM;
2183
2184        error = user_path_dir(new_root, &new);
2185        if (error)
2186                goto out0;
2187        error = -EINVAL;
2188        if (!check_mnt(new.mnt))
2189                goto out1;
2190
2191        error = user_path_dir(put_old, &old);
2192        if (error)
2193                goto out1;
2194
2195        error = security_sb_pivotroot(&old, &new);
2196        if (error) {
2197                path_put(&old);
2198                goto out1;
2199        }
2200
2201        read_lock(&current->fs->lock);
2202        root = current->fs->root;
2203        path_get(&current->fs->root);
2204        read_unlock(&current->fs->lock);
2205        down_write(&namespace_sem);
2206        mutex_lock(&old.dentry->d_inode->i_mutex);
2207        error = -EINVAL;
2208        if (IS_MNT_SHARED(old.mnt) ||
2209                IS_MNT_SHARED(new.mnt->mnt_parent) ||
2210                IS_MNT_SHARED(root.mnt->mnt_parent))
2211                goto out2;
2212        if (!check_mnt(root.mnt))
2213                goto out2;
2214        error = -ENOENT;
2215        if (IS_DEADDIR(new.dentry->d_inode))
2216                goto out2;
2217        if (d_unhashed(new.dentry) && !IS_ROOT(new.dentry))
2218                goto out2;
2219        if (d_unhashed(old.dentry) && !IS_ROOT(old.dentry))
2220                goto out2;
2221        error = -EBUSY;
2222        if (new.mnt == root.mnt ||
2223            old.mnt == root.mnt)
2224                goto out2; /* loop, on the same file system  */
2225        error = -EINVAL;
2226        if (root.mnt->mnt_root != root.dentry)
2227                goto out2; /* not a mountpoint */
2228        if (root.mnt->mnt_parent == root.mnt)
2229                goto out2; /* not attached */
2230        if (new.mnt->mnt_root != new.dentry)
2231                goto out2; /* not a mountpoint */
2232        if (new.mnt->mnt_parent == new.mnt)
2233                goto out2; /* not attached */
2234        /* make sure we can reach put_old from new_root */
2235        tmp = old.mnt;
2236        spin_lock(&vfsmount_lock);
2237        if (tmp != new.mnt) {
2238                for (;;) {
2239                        if (tmp->mnt_parent == tmp)
2240                                goto out3; /* already mounted on put_old */
2241                        if (tmp->mnt_parent == new.mnt)
2242                                break;
2243                        tmp = tmp->mnt_parent;
2244                }
2245                if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
2246                        goto out3;
2247        } else if (!is_subdir(old.dentry, new.dentry))
2248                goto out3;
2249        detach_mnt(new.mnt, &parent_path);
2250        detach_mnt(root.mnt, &root_parent);
2251        /* mount old root on put_old */
2252        attach_mnt(root.mnt, &old);
2253        /* mount new_root on / */
2254        attach_mnt(new.mnt, &root_parent);
2255        touch_mnt_namespace(current->nsproxy->mnt_ns);
2256        spin_unlock(&vfsmount_lock);
2257        chroot_fs_refs(&root, &new);
2258        security_sb_post_pivotroot(&root, &new);
2259        error = 0;
2260        path_put(&root_parent);
2261        path_put(&parent_path);
2262out2:
2263        mutex_unlock(&old.dentry->d_inode->i_mutex);
2264        up_write(&namespace_sem);
2265        path_put(&root);
2266        path_put(&old);
2267out1:
2268        path_put(&new);
2269out0:
2270        return error;
2271out3:
2272        spin_unlock(&vfsmount_lock);
2273        goto out2;
2274}
2275
2276static void __init init_mount_tree(void)
2277{
2278        struct vfsmount *mnt;
2279        struct mnt_namespace *ns;
2280        struct path root;
2281
2282        mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2283        if (IS_ERR(mnt))
2284                panic("Can't create rootfs");
2285        ns = kmalloc(sizeof(*ns), GFP_KERNEL);
2286        if (!ns)
2287                panic("Can't allocate initial namespace");
2288        atomic_set(&ns->count, 1);
2289        INIT_LIST_HEAD(&ns->list);
2290        init_waitqueue_head(&ns->poll);
2291        ns->event = 0;
2292        list_add(&mnt->mnt_list, &ns->list);
2293        ns->root = mnt;
2294        mnt->mnt_ns = ns;
2295
2296        init_task.nsproxy->mnt_ns = ns;
2297        get_mnt_ns(ns);
2298
2299        root.mnt = ns->root;
2300        root.dentry = ns->root->mnt_root;
2301
2302        set_fs_pwd(current->fs, &root);
2303        set_fs_root(current->fs, &root);
2304}
2305
2306void __init mnt_init(void)
2307{
2308        unsigned u;
2309        int err;
2310
2311        init_rwsem(&namespace_sem);
2312
2313        mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
2314                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2315
2316        mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2317
2318        if (!mount_hashtable)
2319                panic("Failed to allocate mount hash table\n");
2320
2321        printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
2322
2323        for (u = 0; u < HASH_SIZE; u++)
2324                INIT_LIST_HEAD(&mount_hashtable[u]);
2325
2326        err = sysfs_init();
2327        if (err)
2328                printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2329                        __func__, err);
2330        fs_kobj = kobject_create_and_add("fs", NULL);
2331        if (!fs_kobj)
2332                printk(KERN_WARNING "%s: kobj create error\n", __func__);
2333        init_rootfs();
2334        init_mount_tree();
2335}
2336
2337void __put_mnt_ns(struct mnt_namespace *ns)
2338{
2339        struct vfsmount *root = ns->root;
2340        LIST_HEAD(umount_list);
2341        ns->root = NULL;
2342        spin_unlock(&vfsmount_lock);
2343        down_write(&namespace_sem);
2344        spin_lock(&vfsmount_lock);
2345        umount_tree(root, 0, &umount_list);
2346        spin_unlock(&vfsmount_lock);
2347        up_write(&namespace_sem);
2348        release_mounts(&umount_list);
2349        kfree(ns);
2350}
2351
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.