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/export.h>
  13#include <linux/capability.h>
  14#include <linux/mnt_namespace.h>
  15#include <linux/namei.h>
  16#include <linux/security.h>
  17#include <linux/idr.h>
  18#include <linux/acct.h>         /* acct_auto_close_mnt */
  19#include <linux/ramfs.h>        /* init_rootfs */
  20#include <linux/fs_struct.h>    /* get_fs_root et.al. */
  21#include <linux/fsnotify.h>     /* fsnotify_vfsmount_delete */
  22#include <linux/uaccess.h>
  23#include "pnode.h"
  24#include "internal.h"
  25
  26#define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
  27#define HASH_SIZE (1UL << HASH_SHIFT)
  28
  29static int event;
  30static DEFINE_IDA(mnt_id_ida);
  31static DEFINE_IDA(mnt_group_ida);
  32static DEFINE_SPINLOCK(mnt_id_lock);
  33static int mnt_id_start = 0;
  34static int mnt_group_start = 1;
  35
  36static struct list_head *mount_hashtable __read_mostly;
  37static struct kmem_cache *mnt_cache __read_mostly;
  38static struct rw_semaphore namespace_sem;
  39
  40/* /sys/fs */
  41struct kobject *fs_kobj;
  42EXPORT_SYMBOL_GPL(fs_kobj);
  43
  44/*
  45 * vfsmount lock may be taken for read to prevent changes to the
  46 * vfsmount hash, ie. during mountpoint lookups or walking back
  47 * up the tree.
  48 *
  49 * It should be taken for write in all cases where the vfsmount
  50 * tree or hash is modified or when a vfsmount structure is modified.
  51 */
  52DEFINE_BRLOCK(vfsmount_lock);
  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/*
  65 * allocation is serialized by namespace_sem, but we need the spinlock to
  66 * serialize with freeing.
  67 */
  68static int mnt_alloc_id(struct mount *mnt)
  69{
  70        int res;
  71
  72retry:
  73        ida_pre_get(&mnt_id_ida, GFP_KERNEL);
  74        spin_lock(&mnt_id_lock);
  75        res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
  76        if (!res)
  77                mnt_id_start = mnt->mnt_id + 1;
  78        spin_unlock(&mnt_id_lock);
  79        if (res == -EAGAIN)
  80                goto retry;
  81
  82        return res;
  83}
  84
  85static void mnt_free_id(struct mount *mnt)
  86{
  87        int id = mnt->mnt_id;
  88        spin_lock(&mnt_id_lock);
  89        ida_remove(&mnt_id_ida, id);
  90        if (mnt_id_start > id)
  91                mnt_id_start = id;
  92        spin_unlock(&mnt_id_lock);
  93}
  94
  95/*
  96 * Allocate a new peer group ID
  97 *
  98 * mnt_group_ida is protected by namespace_sem
  99 */
 100static int mnt_alloc_group_id(struct mount *mnt)
 101{
 102        int res;
 103
 104        if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
 105                return -ENOMEM;
 106
 107        res = ida_get_new_above(&mnt_group_ida,
 108                                mnt_group_start,
 109                                &mnt->mnt_group_id);
 110        if (!res)
 111                mnt_group_start = mnt->mnt_group_id + 1;
 112
 113        return res;
 114}
 115
 116/*
 117 * Release a peer group ID
 118 */
 119void mnt_release_group_id(struct mount *mnt)
 120{
 121        int id = mnt->mnt_group_id;
 122        ida_remove(&mnt_group_ida, id);
 123        if (mnt_group_start > id)
 124                mnt_group_start = id;
 125        mnt->mnt_group_id = 0;
 126}
 127
 128/*
 129 * vfsmount lock must be held for read
 130 */
 131static inline void mnt_add_count(struct mount *mnt, int n)
 132{
 133#ifdef CONFIG_SMP
 134        this_cpu_add(mnt->mnt_pcp->mnt_count, n);
 135#else
 136        preempt_disable();
 137        mnt->mnt_count += n;
 138        preempt_enable();
 139#endif
 140}
 141
 142/*
 143 * vfsmount lock must be held for write
 144 */
 145unsigned int mnt_get_count(struct mount *mnt)
 146{
 147#ifdef CONFIG_SMP
 148        unsigned int count = 0;
 149        int cpu;
 150
 151        for_each_possible_cpu(cpu) {
 152                count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
 153        }
 154
 155        return count;
 156#else
 157        return mnt->mnt_count;
 158#endif
 159}
 160
 161static struct mount *alloc_vfsmnt(const char *name)
 162{
 163        struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
 164        if (mnt) {
 165                int err;
 166
 167                err = mnt_alloc_id(mnt);
 168                if (err)
 169                        goto out_free_cache;
 170
 171                if (name) {
 172                        mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
 173                        if (!mnt->mnt_devname)
 174                                goto out_free_id;
 175                }
 176
 177#ifdef CONFIG_SMP
 178                mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
 179                if (!mnt->mnt_pcp)
 180                        goto out_free_devname;
 181
 182                this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
 183#else
 184                mnt->mnt_count = 1;
 185                mnt->mnt_writers = 0;
 186#endif
 187
 188                INIT_LIST_HEAD(&mnt->mnt_hash);
 189                INIT_LIST_HEAD(&mnt->mnt_child);
 190                INIT_LIST_HEAD(&mnt->mnt_mounts);
 191                INIT_LIST_HEAD(&mnt->mnt_list);
 192                INIT_LIST_HEAD(&mnt->mnt_expire);
 193                INIT_LIST_HEAD(&mnt->mnt_share);
 194                INIT_LIST_HEAD(&mnt->mnt_slave_list);
 195                INIT_LIST_HEAD(&mnt->mnt_slave);
 196#ifdef CONFIG_FSNOTIFY
 197                INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
 198#endif
 199        }
 200        return mnt;
 201
 202#ifdef CONFIG_SMP
 203out_free_devname:
 204        kfree(mnt->mnt_devname);
 205#endif
 206out_free_id:
 207        mnt_free_id(mnt);
 208out_free_cache:
 209        kmem_cache_free(mnt_cache, mnt);
 210        return NULL;
 211}
 212
 213/*
 214 * Most r/o checks on a fs are for operations that take
 215 * discrete amounts of time, like a write() or unlink().
 216 * We must keep track of when those operations start
 217 * (for permission checks) and when they end, so that
 218 * we can determine when writes are able to occur to
 219 * a filesystem.
 220 */
 221/*
 222 * __mnt_is_readonly: check whether a mount is read-only
 223 * @mnt: the mount to check for its write status
 224 *
 225 * This shouldn't be used directly ouside of the VFS.
 226 * It does not guarantee that the filesystem will stay
 227 * r/w, just that it is right *now*.  This can not and
 228 * should not be used in place of IS_RDONLY(inode).
 229 * mnt_want/drop_write() will _keep_ the filesystem
 230 * r/w.
 231 */
 232int __mnt_is_readonly(struct vfsmount *mnt)
 233{
 234        if (mnt->mnt_flags & MNT_READONLY)
 235                return 1;
 236        if (mnt->mnt_sb->s_flags & MS_RDONLY)
 237                return 1;
 238        return 0;
 239}
 240EXPORT_SYMBOL_GPL(__mnt_is_readonly);
 241
 242static inline void mnt_inc_writers(struct mount *mnt)
 243{
 244#ifdef CONFIG_SMP
 245        this_cpu_inc(mnt->mnt_pcp->mnt_writers);
 246#else
 247        mnt->mnt_writers++;
 248#endif
 249}
 250
 251static inline void mnt_dec_writers(struct mount *mnt)
 252{
 253#ifdef CONFIG_SMP
 254        this_cpu_dec(mnt->mnt_pcp->mnt_writers);
 255#else
 256        mnt->mnt_writers--;
 257#endif
 258}
 259
 260static unsigned int mnt_get_writers(struct mount *mnt)
 261{
 262#ifdef CONFIG_SMP
 263        unsigned int count = 0;
 264        int cpu;
 265
 266        for_each_possible_cpu(cpu) {
 267                count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
 268        }
 269
 270        return count;
 271#else
 272        return mnt->mnt_writers;
 273#endif
 274}
 275
 276static int mnt_is_readonly(struct vfsmount *mnt)
 277{
 278        if (mnt->mnt_sb->s_readonly_remount)
 279                return 1;
 280        /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
 281        smp_rmb();
 282        return __mnt_is_readonly(mnt);
 283}
 284
 285/*
 286 * Most r/o & frozen checks on a fs are for operations that take discrete
 287 * amounts of time, like a write() or unlink().  We must keep track of when
 288 * those operations start (for permission checks) and when they end, so that we
 289 * can determine when writes are able to occur to a filesystem.
 290 */
 291/**
 292 * __mnt_want_write - get write access to a mount without freeze protection
 293 * @m: the mount on which to take a write
 294 *
 295 * This tells the low-level filesystem that a write is about to be performed to
 296 * it, and makes sure that writes are allowed (mnt it read-write) before
 297 * returning success. This operation does not protect against filesystem being
 298 * frozen. When the write operation is finished, __mnt_drop_write() must be
 299 * called. This is effectively a refcount.
 300 */
 301int __mnt_want_write(struct vfsmount *m)
 302{
 303        struct mount *mnt = real_mount(m);
 304        int ret = 0;
 305
 306        preempt_disable();
 307        mnt_inc_writers(mnt);
 308        /*
 309         * The store to mnt_inc_writers must be visible before we pass
 310         * MNT_WRITE_HOLD loop below, so that the slowpath can see our
 311         * incremented count after it has set MNT_WRITE_HOLD.
 312         */
 313        smp_mb();
 314        while (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
 315                cpu_relax();
 316        /*
 317         * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
 318         * be set to match its requirements. So we must not load that until
 319         * MNT_WRITE_HOLD is cleared.
 320         */
 321        smp_rmb();
 322        if (mnt_is_readonly(m)) {
 323                mnt_dec_writers(mnt);
 324                ret = -EROFS;
 325        }
 326        preempt_enable();
 327
 328        return ret;
 329}
 330
 331/**
 332 * mnt_want_write - get write access to a mount
 333 * @m: the mount on which to take a write
 334 *
 335 * This tells the low-level filesystem that a write is about to be performed to
 336 * it, and makes sure that writes are allowed (mount is read-write, filesystem
 337 * is not frozen) before returning success.  When the write operation is
 338 * finished, mnt_drop_write() must be called.  This is effectively a refcount.
 339 */
 340int mnt_want_write(struct vfsmount *m)
 341{
 342        int ret;
 343
 344        sb_start_write(m->mnt_sb);
 345        ret = __mnt_want_write(m);
 346        if (ret)
 347                sb_end_write(m->mnt_sb);
 348        return ret;
 349}
 350EXPORT_SYMBOL_GPL(mnt_want_write);
 351
 352/**
 353 * mnt_clone_write - get write access to a mount
 354 * @mnt: the mount on which to take a write
 355 *
 356 * This is effectively like mnt_want_write, except
 357 * it must only be used to take an extra write reference
 358 * on a mountpoint that we already know has a write reference
 359 * on it. This allows some optimisation.
 360 *
 361 * After finished, mnt_drop_write must be called as usual to
 362 * drop the reference.
 363 */
 364int mnt_clone_write(struct vfsmount *mnt)
 365{
 366        /* superblock may be r/o */
 367        if (__mnt_is_readonly(mnt))
 368                return -EROFS;
 369        preempt_disable();
 370        mnt_inc_writers(real_mount(mnt));
 371        preempt_enable();
 372        return 0;
 373}
 374EXPORT_SYMBOL_GPL(mnt_clone_write);
 375
 376/**
 377 * __mnt_want_write_file - get write access to a file's mount
 378 * @file: the file who's mount on which to take a write
 379 *
 380 * This is like __mnt_want_write, but it takes a file and can
 381 * do some optimisations if the file is open for write already
 382 */
 383int __mnt_want_write_file(struct file *file)
 384{
 385        struct inode *inode = file->f_dentry->d_inode;
 386
 387        if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
 388                return __mnt_want_write(file->f_path.mnt);
 389        else
 390                return mnt_clone_write(file->f_path.mnt);
 391}
 392
 393/**
 394 * mnt_want_write_file - get write access to a file's mount
 395 * @file: the file who's mount on which to take a write
 396 *
 397 * This is like mnt_want_write, but it takes a file and can
 398 * do some optimisations if the file is open for write already
 399 */
 400int mnt_want_write_file(struct file *file)
 401{
 402        int ret;
 403
 404        sb_start_write(file->f_path.mnt->mnt_sb);
 405        ret = __mnt_want_write_file(file);
 406        if (ret)
 407                sb_end_write(file->f_path.mnt->mnt_sb);
 408        return ret;
 409}
 410EXPORT_SYMBOL_GPL(mnt_want_write_file);
 411
 412/**
 413 * __mnt_drop_write - give up write access to a mount
 414 * @mnt: the mount on which to give up write access
 415 *
 416 * Tells the low-level filesystem that we are done
 417 * performing writes to it.  Must be matched with
 418 * __mnt_want_write() call above.
 419 */
 420void __mnt_drop_write(struct vfsmount *mnt)
 421{
 422        preempt_disable();
 423        mnt_dec_writers(real_mount(mnt));
 424        preempt_enable();
 425}
 426
 427/**
 428 * mnt_drop_write - give up write access to a mount
 429 * @mnt: the mount on which to give up write access
 430 *
 431 * Tells the low-level filesystem that we are done performing writes to it and
 432 * also allows filesystem to be frozen again.  Must be matched with
 433 * mnt_want_write() call above.
 434 */
 435void mnt_drop_write(struct vfsmount *mnt)
 436{
 437        __mnt_drop_write(mnt);
 438        sb_end_write(mnt->mnt_sb);
 439}
 440EXPORT_SYMBOL_GPL(mnt_drop_write);
 441
 442void __mnt_drop_write_file(struct file *file)
 443{
 444        __mnt_drop_write(file->f_path.mnt);
 445}
 446
 447void mnt_drop_write_file(struct file *file)
 448{
 449        mnt_drop_write(file->f_path.mnt);
 450}
 451EXPORT_SYMBOL(mnt_drop_write_file);
 452
 453static int mnt_make_readonly(struct mount *mnt)
 454{
 455        int ret = 0;
 456
 457        br_write_lock(&vfsmount_lock);
 458        mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
 459        /*
 460         * After storing MNT_WRITE_HOLD, we'll read the counters. This store
 461         * should be visible before we do.
 462         */
 463        smp_mb();
 464
 465        /*
 466         * With writers on hold, if this value is zero, then there are
 467         * definitely no active writers (although held writers may subsequently
 468         * increment the count, they'll have to wait, and decrement it after
 469         * seeing MNT_READONLY).
 470         *
 471         * It is OK to have counter incremented on one CPU and decremented on
 472         * another: the sum will add up correctly. The danger would be when we
 473         * sum up each counter, if we read a counter before it is incremented,
 474         * but then read another CPU's count which it has been subsequently
 475         * decremented from -- we would see more decrements than we should.
 476         * MNT_WRITE_HOLD protects against this scenario, because
 477         * mnt_want_write first increments count, then smp_mb, then spins on
 478         * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
 479         * we're counting up here.
 480         */
 481        if (mnt_get_writers(mnt) > 0)
 482                ret = -EBUSY;
 483        else
 484                mnt->mnt.mnt_flags |= MNT_READONLY;
 485        /*
 486         * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
 487         * that become unheld will see MNT_READONLY.
 488         */
 489        smp_wmb();
 490        mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
 491        br_write_unlock(&vfsmount_lock);
 492        return ret;
 493}
 494
 495static void __mnt_unmake_readonly(struct mount *mnt)
 496{
 497        br_write_lock(&vfsmount_lock);
 498        mnt->mnt.mnt_flags &= ~MNT_READONLY;
 499        br_write_unlock(&vfsmount_lock);
 500}
 501
 502int sb_prepare_remount_readonly(struct super_block *sb)
 503{
 504        struct mount *mnt;
 505        int err = 0;
 506
 507        /* Racy optimization.  Recheck the counter under MNT_WRITE_HOLD */
 508        if (atomic_long_read(&sb->s_remove_count))
 509                return -EBUSY;
 510
 511        br_write_lock(&vfsmount_lock);
 512        list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
 513                if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
 514                        mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
 515                        smp_mb();
 516                        if (mnt_get_writers(mnt) > 0) {
 517                                err = -EBUSY;
 518                                break;
 519                        }
 520                }
 521        }
 522        if (!err && atomic_long_read(&sb->s_remove_count))
 523                err = -EBUSY;
 524
 525        if (!err) {
 526                sb->s_readonly_remount = 1;
 527                smp_wmb();
 528        }
 529        list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
 530                if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
 531                        mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
 532        }
 533        br_write_unlock(&vfsmount_lock);
 534
 535        return err;
 536}
 537
 538static void free_vfsmnt(struct mount *mnt)
 539{
 540        kfree(mnt->mnt_devname);
 541        mnt_free_id(mnt);
 542#ifdef CONFIG_SMP
 543        free_percpu(mnt->mnt_pcp);
 544#endif
 545        kmem_cache_free(mnt_cache, mnt);
 546}
 547
 548/*
 549 * find the first or last mount at @dentry on vfsmount @mnt depending on
 550 * @dir. If @dir is set return the first mount else return the last mount.
 551 * vfsmount_lock must be held for read or write.
 552 */
 553struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
 554                              int dir)
 555{
 556        struct list_head *head = mount_hashtable + hash(mnt, dentry);
 557        struct list_head *tmp = head;
 558        struct mount *p, *found = NULL;
 559
 560        for (;;) {
 561                tmp = dir ? tmp->next : tmp->prev;
 562                p = NULL;
 563                if (tmp == head)
 564                        break;
 565                p = list_entry(tmp, struct mount, mnt_hash);
 566                if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) {
 567                        found = p;
 568                        break;
 569                }
 570        }
 571        return found;
 572}
 573
 574/*
 575 * lookup_mnt - Return the first child mount mounted at path
 576 *
 577 * "First" means first mounted chronologically.  If you create the
 578 * following mounts:
 579 *
 580 * mount /dev/sda1 /mnt
 581 * mount /dev/sda2 /mnt
 582 * mount /dev/sda3 /mnt
 583 *
 584 * Then lookup_mnt() on the base /mnt dentry in the root mount will
 585 * return successively the root dentry and vfsmount of /dev/sda1, then
 586 * /dev/sda2, then /dev/sda3, then NULL.
 587 *
 588 * lookup_mnt takes a reference to the found vfsmount.
 589 */
 590struct vfsmount *lookup_mnt(struct path *path)
 591{
 592        struct mount *child_mnt;
 593
 594        br_read_lock(&vfsmount_lock);
 595        child_mnt = __lookup_mnt(path->mnt, path->dentry, 1);
 596        if (child_mnt) {
 597                mnt_add_count(child_mnt, 1);
 598                br_read_unlock(&vfsmount_lock);
 599                return &child_mnt->mnt;
 600        } else {
 601                br_read_unlock(&vfsmount_lock);
 602                return NULL;
 603        }
 604}
 605
 606static inline int check_mnt(struct mount *mnt)
 607{
 608        return mnt->mnt_ns == current->nsproxy->mnt_ns;
 609}
 610
 611/*
 612 * vfsmount lock must be held for write
 613 */
 614static void touch_mnt_namespace(struct mnt_namespace *ns)
 615{
 616        if (ns) {
 617                ns->event = ++event;
 618                wake_up_interruptible(&ns->poll);
 619        }
 620}
 621
 622/*
 623 * vfsmount lock must be held for write
 624 */
 625static void __touch_mnt_namespace(struct mnt_namespace *ns)
 626{
 627        if (ns && ns->event != event) {
 628                ns->event = event;
 629                wake_up_interruptible(&ns->poll);
 630        }
 631}
 632
 633/*
 634 * Clear dentry's mounted state if it has no remaining mounts.
 635 * vfsmount_lock must be held for write.
 636 */
 637static void dentry_reset_mounted(struct dentry *dentry)
 638{
 639        unsigned u;
 640
 641        for (u = 0; u < HASH_SIZE; u++) {
 642                struct mount *p;
 643
 644                list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
 645                        if (p->mnt_mountpoint == dentry)
 646                                return;
 647                }
 648        }
 649        spin_lock(&dentry->d_lock);
 650        dentry->d_flags &= ~DCACHE_MOUNTED;
 651        spin_unlock(&dentry->d_lock);
 652}
 653
 654/*
 655 * vfsmount lock must be held for write
 656 */
 657static void detach_mnt(struct mount *mnt, struct path *old_path)
 658{
 659        old_path->dentry = mnt->mnt_mountpoint;
 660        old_path->mnt = &mnt->mnt_parent->mnt;
 661        mnt->mnt_parent = mnt;
 662        mnt->mnt_mountpoint = mnt->mnt.mnt_root;
 663        list_del_init(&mnt->mnt_child);
 664        list_del_init(&mnt->mnt_hash);
 665        dentry_reset_mounted(old_path->dentry);
 666}
 667
 668/*
 669 * vfsmount lock must be held for write
 670 */
 671void mnt_set_mountpoint(struct mount *mnt, struct dentry *dentry,
 672                        struct mount *child_mnt)
 673{
 674        mnt_add_count(mnt, 1);  /* essentially, that's mntget */
 675        child_mnt->mnt_mountpoint = dget(dentry);
 676        child_mnt->mnt_parent = mnt;
 677        spin_lock(&dentry->d_lock);
 678        dentry->d_flags |= DCACHE_MOUNTED;
 679        spin_unlock(&dentry->d_lock);
 680}
 681
 682/*
 683 * vfsmount lock must be held for write
 684 */
 685static void attach_mnt(struct mount *mnt, struct path *path)
 686{
 687        mnt_set_mountpoint(real_mount(path->mnt), path->dentry, mnt);
 688        list_add_tail(&mnt->mnt_hash, mount_hashtable +
 689                        hash(path->mnt, path->dentry));
 690        list_add_tail(&mnt->mnt_child, &real_mount(path->mnt)->mnt_mounts);
 691}
 692
 693/*
 694 * vfsmount lock must be held for write
 695 */
 696static void commit_tree(struct mount *mnt)
 697{
 698        struct mount *parent = mnt->mnt_parent;
 699        struct mount *m;
 700        LIST_HEAD(head);
 701        struct mnt_namespace *n = parent->mnt_ns;
 702
 703        BUG_ON(parent == mnt);
 704
 705        list_add_tail(&head, &mnt->mnt_list);
 706        list_for_each_entry(m, &head, mnt_list)
 707                m->mnt_ns = n;
 708
 709        list_splice(&head, n->list.prev);
 710
 711        list_add_tail(&mnt->mnt_hash, mount_hashtable +
 712                                hash(&parent->mnt, mnt->mnt_mountpoint));
 713        list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
 714        touch_mnt_namespace(n);
 715}
 716
 717static struct mount *next_mnt(struct mount *p, struct mount *root)
 718{
 719        struct list_head *next = p->mnt_mounts.next;
 720        if (next == &p->mnt_mounts) {
 721                while (1) {
 722                        if (p == root)
 723                                return NULL;
 724                        next = p->mnt_child.next;
 725                        if (next != &p->mnt_parent->mnt_mounts)
 726                                break;
 727                        p = p->mnt_parent;
 728                }
 729        }
 730        return list_entry(next, struct mount, mnt_child);
 731}
 732
 733static struct mount *skip_mnt_tree(struct mount *p)
 734{
 735        struct list_head *prev = p->mnt_mounts.prev;
 736        while (prev != &p->mnt_mounts) {
 737                p = list_entry(prev, struct mount, mnt_child);
 738                prev = p->mnt_mounts.prev;
 739        }
 740        return p;
 741}
 742
 743struct vfsmount *
 744vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
 745{
 746        struct mount *mnt;
 747        struct dentry *root;
 748
 749        if (!type)
 750                return ERR_PTR(-ENODEV);
 751
 752        mnt = alloc_vfsmnt(name);
 753        if (!mnt)
 754                return ERR_PTR(-ENOMEM);
 755
 756        if (flags & MS_KERNMOUNT)
 757                mnt->mnt.mnt_flags = MNT_INTERNAL;
 758
 759        root = mount_fs(type, flags, name, data);
 760        if (IS_ERR(root)) {
 761                free_vfsmnt(mnt);
 762                return ERR_CAST(root);
 763        }
 764
 765        mnt->mnt.mnt_root = root;
 766        mnt->mnt.mnt_sb = root->d_sb;
 767        mnt->mnt_mountpoint = mnt->mnt.mnt_root;
 768        mnt->mnt_parent = mnt;
 769        br_write_lock(&vfsmount_lock);
 770        list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
 771        br_write_unlock(&vfsmount_lock);
 772        return &mnt->mnt;
 773}
 774EXPORT_SYMBOL_GPL(vfs_kern_mount);
 775
 776static struct mount *clone_mnt(struct mount *old, struct dentry *root,
 777                                        int flag)
 778{
 779        struct super_block *sb = old->mnt.mnt_sb;
 780        struct mount *mnt;
 781        int err;
 782
 783        mnt = alloc_vfsmnt(old->mnt_devname);
 784        if (!mnt)
 785                return ERR_PTR(-ENOMEM);
 786
 787        if (flag & (CL_SLAVE | CL_PRIVATE))
 788                mnt->mnt_group_id = 0; /* not a peer of original */
 789        else
 790                mnt->mnt_group_id = old->mnt_group_id;
 791
 792        if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
 793                err = mnt_alloc_group_id(mnt);
 794                if (err)
 795                        goto out_free;
 796        }
 797
 798        mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
 799        atomic_inc(&sb->s_active);
 800        mnt->mnt.mnt_sb = sb;
 801        mnt->mnt.mnt_root = dget(root);
 802        mnt->mnt_mountpoint = mnt->mnt.mnt_root;
 803        mnt->mnt_parent = mnt;
 804        br_write_lock(&vfsmount_lock);
 805        list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
 806        br_write_unlock(&vfsmount_lock);
 807
 808        if (flag & CL_SLAVE) {
 809                list_add(&mnt->mnt_slave, &old->mnt_slave_list);
 810                mnt->mnt_master = old;
 811                CLEAR_MNT_SHARED(mnt);
 812        } else if (!(flag & CL_PRIVATE)) {
 813                if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
 814                        list_add(&mnt->mnt_share, &old->mnt_share);
 815                if (IS_MNT_SLAVE(old))
 816                        list_add(&mnt->mnt_slave, &old->mnt_slave);
 817                mnt->mnt_master = old->mnt_master;
 818        }
 819        if (flag & CL_MAKE_SHARED)
 820                set_mnt_shared(mnt);
 821
 822        /* stick the duplicate mount on the same expiry list
 823         * as the original if that was on one */
 824        if (flag & CL_EXPIRE) {
 825                if (!list_empty(&old->mnt_expire))
 826                        list_add(&mnt->mnt_expire, &old->mnt_expire);
 827        }
 828
 829        return mnt;
 830
 831 out_free:
 832        free_vfsmnt(mnt);
 833        return ERR_PTR(err);
 834}
 835
 836static inline void mntfree(struct mount *mnt)
 837{
 838        struct vfsmount *m = &mnt->mnt;
 839        struct super_block *sb = m->mnt_sb;
 840
 841        /*
 842         * This probably indicates that somebody messed
 843         * up a mnt_want/drop_write() pair.  If this
 844         * happens, the filesystem was probably unable
 845         * to make r/w->r/o transitions.
 846         */
 847        /*
 848         * The locking used to deal with mnt_count decrement provides barriers,
 849         * so mnt_get_writers() below is safe.
 850         */
 851        WARN_ON(mnt_get_writers(mnt));
 852        fsnotify_vfsmount_delete(m);
 853        dput(m->mnt_root);
 854        free_vfsmnt(mnt);
 855        deactivate_super(sb);
 856}
 857
 858static void mntput_no_expire(struct mount *mnt)
 859{
 860put_again:
 861#ifdef CONFIG_SMP
 862        br_read_lock(&vfsmount_lock);
 863        if (likely(mnt->mnt_ns)) {
 864                /* shouldn't be the last one */
 865                mnt_add_count(mnt, -1);
 866                br_read_unlock(&vfsmount_lock);
 867                return;
 868        }
 869        br_read_unlock(&vfsmount_lock);
 870
 871        br_write_lock(&vfsmount_lock);
 872        mnt_add_count(mnt, -1);
 873        if (mnt_get_count(mnt)) {
 874                br_write_unlock(&vfsmount_lock);
 875                return;
 876        }
 877#else
 878        mnt_add_count(mnt, -1);
 879        if (likely(mnt_get_count(mnt)))
 880                return;
 881        br_write_lock(&vfsmount_lock);
 882#endif
 883        if (unlikely(mnt->mnt_pinned)) {
 884                mnt_add_count(mnt, mnt->mnt_pinned + 1);
 885                mnt->mnt_pinned = 0;
 886                br_write_unlock(&vfsmount_lock);
 887                acct_auto_close_mnt(&mnt->mnt);
 888                goto put_again;
 889        }
 890
 891        list_del(&mnt->mnt_instance);
 892        br_write_unlock(&vfsmount_lock);
 893        mntfree(mnt);
 894}
 895
 896void mntput(struct vfsmount *mnt)
 897{
 898        if (mnt) {
 899                struct mount *m = real_mount(mnt);
 900                /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
 901                if (unlikely(m->mnt_expiry_mark))
 902                        m->mnt_expiry_mark = 0;
 903                mntput_no_expire(m);
 904        }
 905}
 906EXPORT_SYMBOL(mntput);
 907
 908struct vfsmount *mntget(struct vfsmount *mnt)
 909{
 910        if (mnt)
 911                mnt_add_count(real_mount(mnt), 1);
 912        return mnt;
 913}
 914EXPORT_SYMBOL(mntget);
 915
 916void mnt_pin(struct vfsmount *mnt)
 917{
 918        br_write_lock(&vfsmount_lock);
 919        real_mount(mnt)->mnt_pinned++;
 920        br_write_unlock(&vfsmount_lock);
 921}
 922EXPORT_SYMBOL(mnt_pin);
 923
 924void mnt_unpin(struct vfsmount *m)
 925{
 926        struct mount *mnt = real_mount(m);
 927        br_write_lock(&vfsmount_lock);
 928        if (mnt->mnt_pinned) {
 929                mnt_add_count(mnt, 1);
 930                mnt->mnt_pinned--;
 931        }
 932        br_write_unlock(&vfsmount_lock);
 933}
 934EXPORT_SYMBOL(mnt_unpin);
 935
 936static inline void mangle(struct seq_file *m, const char *s)
 937{
 938        seq_escape(m, s, " \t\n\\");
 939}
 940
 941/*
 942 * Simple .show_options callback for filesystems which don't want to
 943 * implement more complex mount option showing.
 944 *
 945 * See also save_mount_options().
 946 */
 947int generic_show_options(struct seq_file *m, struct dentry *root)
 948{
 949        const char *options;
 950
 951        rcu_read_lock();
 952        options = rcu_dereference(root->d_sb->s_options);
 953
 954        if (options != NULL && options[0]) {
 955                seq_putc(m, ',');
 956                mangle(m, options);
 957        }
 958        rcu_read_unlock();
 959
 960        return 0;
 961}
 962EXPORT_SYMBOL(generic_show_options);
 963
 964/*
 965 * If filesystem uses generic_show_options(), this function should be
 966 * called from the fill_super() callback.
 967 *
 968 * The .remount_fs callback usually needs to be handled in a special
 969 * way, to make sure, that previous options are not overwritten if the
 970 * remount fails.
 971 *
 972 * Also note, that if the filesystem's .remount_fs function doesn't
 973 * reset all options to their default value, but changes only newly
 974 * given options, then the displayed options will not reflect reality
 975 * any more.
 976 */
 977void save_mount_options(struct super_block *sb, char *options)
 978{
 979        BUG_ON(sb->s_options);
 980        rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
 981}
 982EXPORT_SYMBOL(save_mount_options);
 983
 984void replace_mount_options(struct super_block *sb, char *options)
 985{
 986        char *old = sb->s_options;
 987        rcu_assign_pointer(sb->s_options, options);
 988        if (old) {
 989                synchronize_rcu();
 990                kfree(old);
 991        }
 992}
 993EXPORT_SYMBOL(replace_mount_options);
 994
 995#ifdef CONFIG_PROC_FS
 996/* iterator; we want it to have access to namespace_sem, thus here... */
 997static void *m_start(struct seq_file *m, loff_t *pos)
 998{
 999        struct proc_mounts *p = proc_mounts(m);
1000
1001        down_read(&namespace_sem);
1002        return seq_list_start(&p->ns->list, *pos);
1003}
1004
1005static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1006{
1007        struct proc_mounts *p = proc_mounts(m);
1008
1009        return seq_list_next(v, &p->ns->list, pos);
1010}
1011
1012static void m_stop(struct seq_file *m, void *v)
1013{
1014        up_read(&namespace_sem);
1015}
1016
1017static int m_show(struct seq_file *m, void *v)
1018{
1019        struct proc_mounts *p = proc_mounts(m);
1020        struct mount *r = list_entry(v, struct mount, mnt_list);
1021        return p->show(m, &r->mnt);
1022}
1023
1024const struct seq_operations mounts_op = {
1025        .start  = m_start,
1026        .next   = m_next,
1027        .stop   = m_stop,
1028        .show   = m_show,
1029};
1030#endif  /* CONFIG_PROC_FS */
1031
1032/**
1033 * may_umount_tree - check if a mount tree is busy
1034 * @mnt: root of mount tree
1035 *
1036 * This is called to check if a tree of mounts has any
1037 * open files, pwds, chroots or sub mounts that are
1038 * busy.
1039 */
1040int may_umount_tree(struct vfsmount *m)
1041{
1042        struct mount *mnt = real_mount(m);
1043        int actual_refs = 0;
1044        int minimum_refs = 0;
1045        struct mount *p;
1046        BUG_ON(!m);
1047
1048        /* write lock needed for mnt_get_count */
1049        br_write_lock(&vfsmount_lock);
1050        for (p = mnt; p; p = next_mnt(p, mnt)) {
1051                actual_refs += mnt_get_count(p);
1052                minimum_refs += 2;
1053        }
1054        br_write_unlock(&vfsmount_lock);
1055
1056        if (actual_refs > minimum_refs)
1057                return 0;
1058
1059        return 1;
1060}
1061
1062EXPORT_SYMBOL(may_umount_tree);
1063
1064/**
1065 * may_umount - check if a mount point is busy
1066 * @mnt: root of mount
1067 *
1068 * This is called to check if a mount point has any
1069 * open files, pwds, chroots or sub mounts. If the
1070 * mount has sub mounts this will return busy
1071 * regardless of whether the sub mounts are busy.
1072 *
1073 * Doesn't take quota and stuff into account. IOW, in some cases it will
1074 * give false negatives. The main reason why it's here is that we need
1075 * a non-destructive way to look for easily umountable filesystems.
1076 */
1077int may_umount(struct vfsmount *mnt)
1078{
1079        int ret = 1;
1080        down_read(&namespace_sem);
1081        br_write_lock(&vfsmount_lock);
1082        if (propagate_mount_busy(real_mount(mnt), 2))
1083                ret = 0;
1084        br_write_unlock(&vfsmount_lock);
1085        up_read(&namespace_sem);
1086        return ret;
1087}
1088
1089EXPORT_SYMBOL(may_umount);
1090
1091void release_mounts(struct list_head *head)
1092{
1093        struct mount *mnt;
1094        while (!list_empty(head)) {
1095                mnt = list_first_entry(head, struct mount, mnt_hash);
1096                list_del_init(&mnt->mnt_hash);
1097                if (mnt_has_parent(mnt)) {
1098                        struct dentry *dentry;
1099                        struct mount *m;
1100
1101                        br_write_lock(&vfsmount_lock);
1102                        dentry = mnt->mnt_mountpoint;
1103                        m = mnt->mnt_parent;
1104                        mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1105                        mnt->mnt_parent = mnt;
1106                        m->mnt_ghosts--;
1107                        br_write_unlock(&vfsmount_lock);
1108                        dput(dentry);
1109                        mntput(&m->mnt);
1110                }
1111                mntput(&mnt->mnt);
1112        }
1113}
1114
1115/*
1116 * vfsmount lock must be held for write
1117 * namespace_sem must be held for write
1118 */
1119void umount_tree(struct mount *mnt, int propagate, struct list_head *kill)
1120{
1121        LIST_HEAD(tmp_list);
1122        struct mount *p;
1123
1124        for (p = mnt; p; p = next_mnt(p, mnt))
1125                list_move(&p->mnt_hash, &tmp_list);
1126
1127        if (propagate)
1128                propagate_umount(&tmp_list);
1129
1130        list_for_each_entry(p, &tmp_list, mnt_hash) {
1131                list_del_init(&p->mnt_expire);
1132                list_del_init(&p->mnt_list);
1133                __touch_mnt_namespace(p->mnt_ns);
1134                p->mnt_ns = NULL;
1135                list_del_init(&p->mnt_child);
1136                if (mnt_has_parent(p)) {
1137                        p->mnt_parent->mnt_ghosts++;
1138                        dentry_reset_mounted(p->mnt_mountpoint);
1139                }
1140                change_mnt_propagation(p, MS_PRIVATE);
1141        }
1142        list_splice(&tmp_list, kill);
1143}
1144
1145static void shrink_submounts(struct mount *mnt, struct list_head *umounts);
1146
1147static int do_umount(struct mount *mnt, int flags)
1148{
1149        struct super_block *sb = mnt->mnt.mnt_sb;
1150        int retval;
1151        LIST_HEAD(umount_list);
1152
1153        retval = security_sb_umount(&mnt->mnt, flags);
1154        if (retval)
1155                return retval;
1156
1157        /*
1158         * Allow userspace to request a mountpoint be expired rather than
1159         * unmounting unconditionally. Unmount only happens if:
1160         *  (1) the mark is already set (the mark is cleared by mntput())
1161         *  (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1162         */
1163        if (flags & MNT_EXPIRE) {
1164                if (&mnt->mnt == current->fs->root.mnt ||
1165                    flags & (MNT_FORCE | MNT_DETACH))
1166                        return -EINVAL;
1167
1168                /*
1169                 * probably don't strictly need the lock here if we examined
1170                 * all race cases, but it's a slowpath.
1171                 */
1172                br_write_lock(&vfsmount_lock);
1173                if (mnt_get_count(mnt) != 2) {
1174                        br_write_unlock(&vfsmount_lock);
1175                        return -EBUSY;
1176                }
1177                br_write_unlock(&vfsmount_lock);
1178
1179                if (!xchg(&mnt->mnt_expiry_mark, 1))
1180                        return -EAGAIN;
1181        }
1182
1183        /*
1184         * If we may have to abort operations to get out of this
1185         * mount, and they will themselves hold resources we must
1186         * allow the fs to do things. In the Unix tradition of
1187         * 'Gee thats tricky lets do it in userspace' the umount_begin
1188         * might fail to complete on the first run through as other tasks
1189         * must return, and the like. Thats for the mount program to worry
1190         * about for the moment.
1191         */
1192
1193        if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1194                sb->s_op->umount_begin(sb);
1195        }
1196
1197        /*
1198         * No sense to grab the lock for this test, but test itself looks
1199         * somewhat bogus. Suggestions for better replacement?
1200         * Ho-hum... In principle, we might treat that as umount + switch
1201         * to rootfs. GC would eventually take care of the old vfsmount.
1202         * Actually it makes sense, especially if rootfs would contain a
1203         * /reboot - static binary that would close all descriptors and
1204         * call reboot(9). Then init(8) could umount root and exec /reboot.
1205         */
1206        if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1207                /*
1208                 * Special case for "unmounting" root ...
1209                 * we just try to remount it readonly.
1210                 */
1211                down_write(&sb->s_umount);
1212                if (!(sb->s_flags & MS_RDONLY))
1213                        retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1214                up_write(&sb->s_umount);
1215                return retval;
1216        }
1217
1218        down_write(&namespace_sem);
1219        br_write_lock(&vfsmount_lock);
1220        event++;
1221
1222        if (!(flags & MNT_DETACH))
1223                shrink_submounts(mnt, &umount_list);
1224
1225        retval = -EBUSY;
1226        if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1227                if (!list_empty(&mnt->mnt_list))
1228                        umount_tree(mnt, 1, &umount_list);
1229                retval = 0;
1230        }
1231        br_write_unlock(&vfsmount_lock);
1232        up_write(&namespace_sem);
1233        release_mounts(&umount_list);
1234        return retval;
1235}
1236
1237/*
1238 * Now umount can handle mount points as well as block devices.
1239 * This is important for filesystems which use unnamed block devices.
1240 *
1241 * We now support a flag for forced unmount like the other 'big iron'
1242 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1243 */
1244
1245SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1246{
1247        struct path path;
1248        struct mount *mnt;
1249        int retval;
1250        int lookup_flags = 0;
1251
1252        if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1253                return -EINVAL;
1254
1255        if (!(flags & UMOUNT_NOFOLLOW))
1256                lookup_flags |= LOOKUP_FOLLOW;
1257
1258        retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1259        if (retval)
1260                goto out;
1261        mnt = real_mount(path.mnt);
1262        retval = -EINVAL;
1263        if (path.dentry != path.mnt->mnt_root)
1264                goto dput_and_out;
1265        if (!check_mnt(mnt))
1266                goto dput_and_out;
1267
1268        retval = -EPERM;
1269        if (!capable(CAP_SYS_ADMIN))
1270                goto dput_and_out;
1271
1272        retval = do_umount(mnt, flags);
1273dput_and_out:
1274        /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1275        dput(path.dentry);
1276        mntput_no_expire(mnt);
1277out:
1278        return retval;
1279}
1280
1281#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1282
1283/*
1284 *      The 2.0 compatible umount. No flags.
1285 */
1286SYSCALL_DEFINE1(oldumount, char __user *, name)
1287{
1288        return sys_umount(name, 0);
1289}
1290
1291#endif
1292
1293static int mount_is_safe(struct path *path)
1294{
1295        if (capable(CAP_SYS_ADMIN))
1296                return 0;
1297        return -EPERM;
1298#ifdef notyet
1299        if (S_ISLNK(path->dentry->d_inode->i_mode))
1300                return -EPERM;
1301        if (path->dentry->d_inode->i_mode & S_ISVTX) {
1302                if (current_uid() != path->dentry->d_inode->i_uid)
1303                        return -EPERM;
1304        }
1305        if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1306                return -EPERM;
1307        return 0;
1308#endif
1309}
1310
1311struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1312                                        int flag)
1313{
1314        struct mount *res, *p, *q, *r;
1315        struct path path;
1316
1317        if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1318                return ERR_PTR(-EINVAL);
1319
1320        res = q = clone_mnt(mnt, dentry, flag);
1321        if (IS_ERR(q))
1322                return q;
1323
1324        q->mnt_mountpoint = mnt->mnt_mountpoint;
1325
1326        p = mnt;
1327        list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1328                struct mount *s;
1329                if (!is_subdir(r->mnt_mountpoint, dentry))
1330                        continue;
1331
1332                for (s = r; s; s = next_mnt(s, r)) {
1333                        if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1334                                s = skip_mnt_tree(s);
1335                                continue;
1336                        }
1337                        while (p != s->mnt_parent) {
1338                                p = p->mnt_parent;
1339                                q = q->mnt_parent;
1340                        }
1341                        p = s;
1342                        path.mnt = &q->mnt;
1343                        path.dentry = p->mnt_mountpoint;
1344                        q = clone_mnt(p, p->mnt.mnt_root, flag);
1345                        if (IS_ERR(q))
1346                                goto out;
1347                        br_write_lock(&vfsmount_lock);
1348                        list_add_tail(&q->mnt_list, &res->mnt_list);
1349                        attach_mnt(q, &path);
1350                        br_write_unlock(&vfsmount_lock);
1351                }
1352        }
1353        return res;
1354out:
1355        if (res) {
1356                LIST_HEAD(umount_list);
1357                br_write_lock(&vfsmount_lock);
1358                umount_tree(res, 0, &umount_list);
1359                br_write_unlock(&vfsmount_lock);
1360                release_mounts(&umount_list);
1361        }
1362        return q;
1363}
1364
1365/* Caller should check returned pointer for errors */
1366
1367struct vfsmount *collect_mounts(struct path *path)
1368{
1369        struct mount *tree;
1370        down_write(&namespace_sem);
1371        tree = copy_tree(real_mount(path->mnt), path->dentry,
1372                         CL_COPY_ALL | CL_PRIVATE);
1373        up_write(&namespace_sem);
1374        if (IS_ERR(tree))
1375                return NULL;
1376        return &tree->mnt;
1377}
1378
1379void drop_collected_mounts(struct vfsmount *mnt)
1380{
1381        LIST_HEAD(umount_list);
1382        down_write(&namespace_sem);
1383        br_write_lock(&vfsmount_lock);
1384        umount_tree(real_mount(mnt), 0, &umount_list);
1385        br_write_unlock(&vfsmount_lock);
1386        up_write(&namespace_sem);
1387        release_mounts(&umount_list);
1388}
1389
1390int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1391                   struct vfsmount *root)
1392{
1393        struct mount *mnt;
1394        int res = f(root, arg);
1395        if (res)
1396                return res;
1397        list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1398                res = f(&mnt->mnt, arg);
1399                if (res)
1400                        return res;
1401        }
1402        return 0;
1403}
1404
1405static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1406{
1407        struct mount *p;
1408
1409        for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1410                if (p->mnt_group_id && !IS_MNT_SHARED(p))
1411                        mnt_release_group_id(p);
1412        }
1413}
1414
1415static int invent_group_ids(struct mount *mnt, bool recurse)
1416{
1417        struct mount *p;
1418
1419        for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1420                if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1421                        int err = mnt_alloc_group_id(p);
1422                        if (err) {
1423                                cleanup_group_ids(mnt, p);
1424                                return err;
1425                        }
1426                }
1427        }
1428
1429        return 0;
1430}
1431
1432/*
1433 *  @source_mnt : mount tree to be attached
1434 *  @nd         : place the mount tree @source_mnt is attached
1435 *  @parent_nd  : if non-null, detach the source_mnt from its parent and
1436 *                 store the parent mount and mountpoint dentry.
1437 *                 (done when source_mnt is moved)
1438 *
1439 *  NOTE: in the table below explains the semantics when a source mount
1440 *  of a given type is attached to a destination mount of a given type.
1441 * ---------------------------------------------------------------------------
1442 * |         BIND MOUNT OPERATION                                            |
1443 * |**************************************************************************
1444 * | source-->| shared        |       private  |       slave    | unbindable |
1445 * | dest     |               |                |                |            |
1446 * |   |      |               |                |                |            |
1447 * |   v      |               |                |                |            |
1448 * |**************************************************************************
1449 * |  shared  | shared (++)   |     shared (+) |     shared(+++)|  invalid   |
1450 * |          |               |                |                |            |
1451 * |non-shared| shared (+)    |      private   |      slave (*) |  invalid   |
1452 * ***************************************************************************
1453 * A bind operation clones the source mount and mounts the clone on the
1454 * destination mount.
1455 *
1456 * (++)  the cloned mount is propagated to all the mounts in the propagation
1457 *       tree of the destination mount and the cloned mount is added to
1458 *       the peer group of the source mount.
1459 * (+)   the cloned mount is created under the destination mount and is marked
1460 *       as shared. The cloned mount is added to the peer group of the source
1461 *       mount.
1462 * (+++) the mount is propagated to all the mounts in the propagation tree
1463 *       of the destination mount and the cloned mount is made slave
1464 *       of the same master as that of the source mount. The cloned mount
1465 *       is marked as 'shared and slave'.
1466 * (*)   the cloned mount is made a slave of the same master as that of the
1467 *       source mount.
1468 *
1469 * ---------------------------------------------------------------------------
1470 * |                    MOVE MOUNT OPERATION                                 |
1471 * |**************************************************************************
1472 * | source-->| shared        |       private  |       slave    | unbindable |
1473 * | dest     |               |                |                |            |
1474 * |   |      |               |                |                |            |
1475 * |   v      |               |                |                |            |
1476 * |**************************************************************************
1477 * |  shared  | shared (+)    |     shared (+) |    shared(+++) |  invalid   |
1478 * |          |               |                |                |            |
1479 * |non-shared| shared (+*)   |      private   |    slave (*)   | unbindable |
1480 * ***************************************************************************
1481 *
1482 * (+)  the mount is moved to the destination. And is then propagated to
1483 *      all the mounts in the propagation tree of the destination mount.
1484 * (+*)  the mount is moved to the destination.
1485 * (+++)  the mount is moved to the destination and is then propagated to
1486 *      all the mounts belonging to the destination mount's propagation tree.
1487 *      the mount is marked as 'shared and slave'.
1488 * (*)  the mount continues to be a slave at the new location.
1489 *
1490 * if the source mount is a tree, the operations explained above is
1491 * applied to each mount in the tree.
1492 * Must be called without spinlocks held, since this function can sleep
1493 * in allocations.
1494 */
1495static int attach_recursive_mnt(struct mount *source_mnt,
1496                        struct path *path, struct path *parent_path)
1497{
1498        LIST_HEAD(tree_list);
1499        struct mount *dest_mnt = real_mount(path->mnt);
1500        struct dentry *dest_dentry = path->dentry;
1501        struct mount *child, *p;
1502        int err;
1503
1504        if (IS_MNT_SHARED(dest_mnt)) {
1505                err = invent_group_ids(source_mnt, true);
1506                if (err)
1507                        goto out;
1508        }
1509        err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1510        if (err)
1511                goto out_cleanup_ids;
1512
1513        br_write_lock(&vfsmount_lock);
1514
1515        if (IS_MNT_SHARED(dest_mnt)) {
1516                for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1517                        set_mnt_shared(p);
1518        }
1519        if (parent_path) {
1520                detach_mnt(source_mnt, parent_path);
1521                attach_mnt(source_mnt, path);
1522                touch_mnt_namespace(source_mnt->mnt_ns);
1523        } else {
1524                mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1525                commit_tree(source_mnt);
1526        }
1527
1528        list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1529                list_del_init(&child->mnt_hash);
1530                commit_tree(child);
1531        }
1532        br_write_unlock(&vfsmount_lock);
1533
1534        return 0;
1535
1536 out_cleanup_ids:
1537        if (IS_MNT_SHARED(dest_mnt))
1538                cleanup_group_ids(source_mnt, NULL);
1539 out:
1540        return err;
1541}
1542
1543static int lock_mount(struct path *path)
1544{
1545        struct vfsmount *mnt;
1546retry:
1547        mutex_lock(&path->dentry->d_inode->i_mutex);
1548        if (unlikely(cant_mount(path->dentry))) {
1549                mutex_unlock(&path->dentry->d_inode->i_mutex);
1550                return -ENOENT;
1551        }
1552        down_write(&namespace_sem);
1553        mnt = lookup_mnt(path);
1554        if (likely(!mnt))
1555                return 0;
1556        up_write(&namespace_sem);
1557        mutex_unlock(&path->dentry->d_inode->i_mutex);
1558        path_put(path);
1559        path->mnt = mnt;
1560        path->dentry = dget(mnt->mnt_root);
1561        goto retry;
1562}
1563
1564static void unlock_mount(struct path *path)
1565{
1566        up_write(&namespace_sem);
1567        mutex_unlock(&path->dentry->d_inode->i_mutex);
1568}
1569
1570static int graft_tree(struct mount *mnt, struct path *path)
1571{
1572        if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1573                return -EINVAL;
1574
1575        if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1576              S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
1577                return -ENOTDIR;
1578
1579        if (d_unlinked(path->dentry))
1580                return -ENOENT;
1581
1582        return attach_recursive_mnt(mnt, path, NULL);
1583}
1584
1585/*
1586 * Sanity check the flags to change_mnt_propagation.
1587 */
1588
1589static int flags_to_propagation_type(int flags)
1590{
1591        int type = flags & ~(MS_REC | MS_SILENT);
1592
1593        /* Fail if any non-propagation flags are set */
1594        if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1595                return 0;
1596        /* Only one propagation flag should be set */
1597        if (!is_power_of_2(type))
1598                return 0;
1599        return type;
1600}
1601
1602/*
1603 * recursively change the type of the mountpoint.
1604 */
1605static int do_change_type(struct path *path, int flag)
1606{
1607        struct mount *m;
1608        struct mount *mnt = real_mount(path->mnt);
1609        int recurse = flag & MS_REC;
1610        int type;
1611        int err = 0;
1612
1613        if (!capable(CAP_SYS_ADMIN))
1614                return -EPERM;
1615
1616        if (path->dentry != path->mnt->mnt_root)
1617                return -EINVAL;
1618
1619        type = flags_to_propagation_type(flag);
1620        if (!type)
1621                return -EINVAL;
1622
1623        down_write(&namespace_sem);
1624        if (type == MS_SHARED) {
1625                err = invent_group_ids(mnt, recurse);
1626                if (err)
1627                        goto out_unlock;
1628        }
1629
1630        br_write_lock(&vfsmount_lock);
1631        for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1632                change_mnt_propagation(m, type);
1633        br_write_unlock(&vfsmount_lock);
1634
1635 out_unlock:
1636        up_write(&namespace_sem);
1637        return err;
1638}
1639
1640/*
1641 * do loopback mount.
1642 */
1643static int do_loopback(struct path *path, char *old_name,
1644                                int recurse)
1645{
1646        LIST_HEAD(umount_list);
1647        struct path old_path;
1648        struct mount *mnt = NULL, *old;
1649        int err = mount_is_safe(path);
1650        if (err)
1651                return err;
1652        if (!old_name || !*old_name)
1653                return -EINVAL;
1654        err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1655        if (err)
1656                return err;
1657
1658        err = lock_mount(path);
1659        if (err)
1660                goto out;
1661
1662        old = real_mount(old_path.mnt);
1663
1664        err = -EINVAL;
1665        if (IS_MNT_UNBINDABLE(old))
1666                goto out2;
1667
1668        if (!check_mnt(real_mount(path->mnt)) || !check_mnt(old))
1669                goto out2;
1670
1671        if (recurse)
1672                mnt = copy_tree(old, old_path.dentry, 0);
1673        else
1674                mnt = clone_mnt(old, old_path.dentry, 0);
1675
1676        if (IS_ERR(mnt)) {
1677                err = PTR_ERR(mnt);
1678                goto out;
1679        }
1680
1681        err = graft_tree(mnt, path);
1682        if (err) {
1683                br_write_lock(&vfsmount_lock);
1684                umount_tree(mnt, 0, &umount_list);
1685                br_write_unlock(&vfsmount_lock);
1686        }
1687out2:
1688        unlock_mount(path);
1689        release_mounts(&umount_list);
1690out:
1691        path_put(&old_path);
1692        return err;
1693}
1694
1695static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1696{
1697        int error = 0;
1698        int readonly_request = 0;
1699
1700        if (ms_flags & MS_RDONLY)
1701                readonly_request = 1;
1702        if (readonly_request == __mnt_is_readonly(mnt))
1703                return 0;
1704
1705        if (readonly_request)
1706                error = mnt_make_readonly(real_mount(mnt));
1707        else
1708                __mnt_unmake_readonly(real_mount(mnt));
1709        return error;
1710}
1711
1712/*
1713 * change filesystem flags. dir should be a physical root of filesystem.
1714 * If you've mounted a non-root directory somewhere and want to do remount
1715 * on it - tough luck.
1716 */
1717static int do_remount(struct path *path, int flags, int mnt_flags,
1718                      void *data)
1719{
1720        int err;
1721        struct super_block *sb = path->mnt->mnt_sb;
1722        struct mount *mnt = real_mount(path->mnt);
1723
1724        if (!capable(CAP_SYS_ADMIN))
1725                return -EPERM;
1726
1727        if (!check_mnt(mnt))
1728                return -EINVAL;
1729
1730        if (path->dentry != path->mnt->mnt_root)
1731                return -EINVAL;
1732
1733        err = security_sb_remount(sb, data);
1734        if (err)
1735                return err;
1736
1737        down_write(&sb->s_umount);
1738        if (flags & MS_BIND)
1739                err = change_mount_flags(path->mnt, flags);
1740        else
1741                err = do_remount_sb(sb, flags, data, 0);
1742        if (!err) {
1743                br_write_lock(&vfsmount_lock);
1744                mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK;
1745                mnt->mnt.mnt_flags = mnt_flags;
1746                br_write_unlock(&vfsmount_lock);
1747        }
1748        up_write(&sb->s_umount);
1749        if (!err) {
1750                br_write_lock(&vfsmount_lock);
1751                touch_mnt_namespace(mnt->mnt_ns);
1752                br_write_unlock(&vfsmount_lock);
1753        }
1754        return err;
1755}
1756
1757static inline int tree_contains_unbindable(struct mount *mnt)
1758{
1759        struct mount *p;
1760        for (p = mnt; p; p = next_mnt(p, mnt)) {
1761                if (IS_MNT_UNBINDABLE(p))
1762                        return 1;
1763        }
1764        return 0;
1765}
1766
1767static int do_move_mount(struct path *path, char *old_name)
1768{
1769        struct path old_path, parent_path;
1770        struct mount *p;
1771        struct mount *old;
1772        int err = 0;
1773        if (!capable(CAP_SYS_ADMIN))
1774                return -EPERM;
1775        if (!old_name || !*old_name)
1776                return -EINVAL;
1777        err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1778        if (err)
1779                return err;
1780
1781        err = lock_mount(path);
1782        if (err < 0)
1783                goto out;
1784
1785        old = real_mount(old_path.mnt);
1786        p = real_mount(path->mnt);
1787
1788        err = -EINVAL;
1789        if (!check_mnt(p) || !check_mnt(old))
1790                goto out1;
1791
1792        if (d_unlinked(path->dentry))
1793                goto out1;
1794
1795        err = -EINVAL;
1796        if (old_path.dentry != old_path.mnt->mnt_root)
1797                goto out1;
1798
1799        if (!mnt_has_parent(old))
1800                goto out1;
1801
1802        if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1803              S_ISDIR(old_path.dentry->d_inode->i_mode))
1804                goto out1;
1805        /*
1806         * Don't move a mount residing in a shared parent.
1807         */
1808        if (IS_MNT_SHARED(old->mnt_parent))
1809                goto out1;
1810        /*
1811         * Don't move a mount tree containing unbindable mounts to a destination
1812         * mount which is shared.
1813         */
1814        if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
1815                goto out1;
1816        err = -ELOOP;
1817        for (; mnt_has_parent(p); p = p->mnt_parent)
1818                if (p == old)
1819                        goto out1;
1820
1821        err = attach_recursive_mnt(old, path, &parent_path);
1822        if (err)
1823                goto out1;
1824
1825        /* if the mount is moved, it should no longer be expire
1826         * automatically */
1827        list_del_init(&old->mnt_expire);
1828out1:
1829        unlock_mount(path);
1830out:
1831        if (!err)
1832                path_put(&parent_path);
1833        path_put(&old_path);
1834        return err;
1835}
1836
1837static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1838{
1839        int err;
1840        const char *subtype = strchr(fstype, '.');
1841        if (subtype) {
1842                subtype++;
1843                err = -EINVAL;
1844                if (!subtype[0])
1845                        goto err;
1846        } else
1847                subtype = "";
1848
1849        mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1850        err = -ENOMEM;
1851        if (!mnt->mnt_sb->s_subtype)
1852                goto err;
1853        return mnt;
1854
1855 err:
1856        mntput(mnt);
1857        return ERR_PTR(err);
1858}
1859
1860static struct vfsmount *
1861do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1862{
1863        struct file_system_type *type = get_fs_type(fstype);
1864        struct vfsmount *mnt;
1865        if (!type)
1866                return ERR_PTR(-ENODEV);
1867        mnt = vfs_kern_mount(type, flags, name, data);
1868        if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1869            !mnt->mnt_sb->s_subtype)
1870                mnt = fs_set_subtype(mnt, fstype);
1871        put_filesystem(type);
1872        return mnt;
1873}
1874
1875/*
1876 * add a mount into a namespace's mount tree
1877 */
1878static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
1879{
1880        int err;
1881
1882        mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1883
1884        err = lock_mount(path);
1885        if (err)
1886                return err;
1887
1888        err = -EINVAL;
1889        if (unlikely(!check_mnt(real_mount(path->mnt)))) {
1890                /* that's acceptable only for automounts done in private ns */
1891                if (!(mnt_flags & MNT_SHRINKABLE))
1892                        goto unlock;
1893                /* ... and for those we'd better have mountpoint still alive */
1894                if (!real_mount(path->mnt)->mnt_ns)
1895                        goto unlock;
1896        }
1897
1898        /* Refuse the same filesystem on the same mount point */
1899        err = -EBUSY;
1900        if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
1901            path->mnt->mnt_root == path->dentry)
1902                goto unlock;
1903
1904        err = -EINVAL;
1905        if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
1906                goto unlock;
1907
1908        newmnt->mnt.mnt_flags = mnt_flags;
1909        err = graft_tree(newmnt, path);
1910
1911unlock:
1912        unlock_mount(path);
1913        return err;
1914}
1915
1916/*
1917 * create a new mount for userspace and request it to be added into the
1918 * namespace's tree
1919 */
1920static int do_new_mount(struct path *path, char *type, int flags,
1921                        int mnt_flags, char *name, void *data)
1922{
1923        struct vfsmount *mnt;
1924        int err;
1925
1926        if (!type)
1927                return -EINVAL;
1928
1929        /* we need capabilities... */
1930        if (!capable(CAP_SYS_ADMIN))
1931                return -EPERM;
1932
1933        mnt = do_kern_mount(type, flags, name, data);
1934        if (IS_ERR(mnt))
1935                return PTR_ERR(mnt);
1936
1937        err = do_add_mount(real_mount(mnt), path, mnt_flags);
1938        if (err)
1939                mntput(mnt);
1940        return err;
1941}
1942
1943int finish_automount(struct vfsmount *m, struct path *path)
1944{
1945        struct mount *mnt = real_mount(m);
1946        int err;
1947        /* The new mount record should have at least 2 refs to prevent it being
1948         * expired before we get a chance to add it
1949         */
1950        BUG_ON(mnt_get_count(mnt) < 2);
1951
1952        if (m->mnt_sb == path->mnt->mnt_sb &&
1953            m->mnt_root == path->dentry) {
1954                err = -ELOOP;
1955                goto fail;
1956        }
1957
1958        err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
1959        if (!err)
1960                return 0;
1961fail:
1962        /* remove m from any expiration list it may be on */
1963        if (!list_empty(&mnt->mnt_expire)) {
1964                down_write(&namespace_sem);
1965                br_write_lock(&vfsmount_lock);
1966                list_del_init(&mnt->mnt_expire);
1967                br_write_unlock(&vfsmount_lock);
1968                up_write(&namespace_sem);
1969        }
1970        mntput(m);
1971        mntput(m);
1972        return err;
1973}
1974
1975/**
1976 * mnt_set_expiry - Put a mount on an expiration list
1977 * @mnt: The mount to list.
1978 * @expiry_list: The list to add the mount to.
1979 */
1980void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
1981{
1982        down_write(&namespace_sem);
1983        br_write_lock(&vfsmount_lock);
1984
1985        list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
1986
1987        br_write_unlock(&vfsmount_lock);
1988        up_write(&namespace_sem);
1989}
1990EXPORT_SYMBOL(mnt_set_expiry);
1991
1992/*
1993 * process a list of expirable mountpoints with the intent of discarding any
1994 * mountpoints that aren't in use and haven't been touched since last we came
1995 * here
1996 */
1997void mark_mounts_for_expiry(struct list_head *mounts)
1998{
1999        struct mount *mnt, *next;
2000        LIST_HEAD(graveyard);
2001        LIST_HEAD(umounts);
2002
2003        if (list_empty(mounts))
2004                return;
2005
2006        down_write(&namespace_sem);
2007        br_write_lock(&vfsmount_lock);
2008
2009        /* extract from the expiration list every vfsmount that matches the
2010         * following criteria:
2011         * - only referenced by its parent vfsmount
2012         * - still marked for expiry (marked on the last call here; marks are
2013         *   cleared by mntput())
2014         */
2015        list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2016                if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2017                        propagate_mount_busy(mnt, 1))
2018                        continue;
2019                list_move(&mnt->mnt_expire, &graveyard);
2020        }
2021        while (!list_empty(&graveyard)) {
2022                mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2023                touch_mnt_namespace(mnt->mnt_ns);
2024                umount_tree(mnt, 1, &umounts);
2025        }
2026        br_write_unlock(&vfsmount_lock);
2027        up_write(&namespace_sem);
2028
2029        release_mounts(&umounts);
2030}
2031
2032EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2033
2034/*
2035 * Ripoff of 'select_parent()'
2036 *
2037 * search the list of submounts for a given mountpoint, and move any
2038 * shrinkable submounts to the 'graveyard' list.
2039 */
2040static int select_submounts(struct mount *parent, struct list_head *graveyard)
2041{
2042        struct mount *this_parent = parent;
2043        struct list_head *next;
2044        int found = 0;
2045
2046repeat:
2047        next = this_parent->mnt_mounts.next;
2048resume:
2049        while (next != &this_parent->mnt_mounts) {
2050                struct list_head *tmp = next;
2051                struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2052
2053                next = tmp->next;
2054                if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2055                        continue;
2056                /*
2057                 * Descend a level if the d_mounts list is non-empty.
2058                 */
2059                if (!list_empty(&mnt->mnt_mounts)) {
2060                        this_parent = mnt;
2061                        goto repeat;
2062                }
2063
2064                if (!propagate_mount_busy(mnt, 1)) {
2065                        list_move_tail(&mnt->mnt_expire, graveyard);
2066                        found++;
2067                }
2068        }
2069        /*
2070         * All done at this level ... ascend and resume the search
2071         */
2072        if (this_parent != parent) {
2073                next = this_parent->mnt_child.next;
2074                this_parent = this_parent->mnt_parent;
2075                goto resume;
2076        }
2077        return found;
2078}
2079
2080/*
2081 * process a list of expirable mountpoints with the intent of discarding any
2082 * submounts of a specific parent mountpoint
2083 *
2084 * vfsmount_lock must be held for write
2085 */
2086static void shrink_submounts(struct mount *mnt, struct list_head *umounts)
2087{
2088        LIST_HEAD(graveyard);
2089        struct mount *m;
2090
2091        /* extract submounts of 'mountpoint' from the expiration list */
2092        while (select_submounts(mnt, &graveyard)) {
2093                while (!list_empty(&graveyard)) {
2094                        m = list_first_entry(&graveyard, struct mount,
2095                                                mnt_expire);
2096                        touch_mnt_namespace(m->mnt_ns);
2097                        umount_tree(m, 1, umounts);
2098                }
2099        }
2100}
2101
2102/*
2103 * Some copy_from_user() implementations do not return the exact number of
2104 * bytes remaining to copy on a fault.  But copy_mount_options() requires that.
2105 * Note that this function differs from copy_from_user() in that it will oops
2106 * on bad values of `to', rather than returning a short copy.
2107 */
2108static long exact_copy_from_user(void *to, const void __user * from,
2109                                 unsigned long n)
2110{
2111        char *t = to;
2112        const char __user *f = from;
2113        char c;
2114
2115        if (!access_ok(VERIFY_READ, from, n))
2116                return n;
2117
2118        while (n) {
2119                if (__get_user(c, f)) {
2120                        memset(t, 0, n);
2121                        break;
2122                }
2123                *t++ = c;
2124                f++;
2125                n--;
2126        }
2127        return n;
2128}
2129
2130int copy_mount_options(const void __user * data, unsigned long *where)
2131{
2132        int i;
2133        unsigned long page;
2134        unsigned long size;
2135
2136        *where = 0;
2137        if (!data)
2138                return 0;
2139
2140        if (!(page = __get_free_page(GFP_KERNEL)))
2141                return -ENOMEM;
2142
2143        /* We only care that *some* data at the address the user
2144         * gave us is valid.  Just in case, we'll zero
2145         * the remainder of the page.
2146         */
2147        /* copy_from_user cannot cross TASK_SIZE ! */
2148        size = TASK_SIZE - (unsigned long)data;
2149        if (size > PAGE_SIZE)
2150                size = PAGE_SIZE;
2151
2152        i = size - exact_copy_from_user((void *)page, data, size);
2153        if (!i) {
2154                free_page(page);
2155                return -EFAULT;
2156        }
2157        if (i != PAGE_SIZE)
2158                memset((char *)page + i, 0, PAGE_SIZE - i);
2159        *where = page;
2160        return 0;
2161}
2162
2163int copy_mount_string(const void __user *data, char **where)
2164{
2165        char *tmp;
2166
2167        if (!data) {
2168                *where = NULL;
2169                return 0;
2170        }
2171
2172        tmp = strndup_user(data, PAGE_SIZE);
2173        if (IS_ERR(tmp))
2174                return PTR_ERR(tmp);
2175
2176        *where = tmp;
2177        return 0;
2178}
2179
2180/*
2181 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2182 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2183 *
2184 * data is a (void *) that can point to any structure up to
2185 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2186 * information (or be NULL).
2187 *
2188 * Pre-0.97 versions of mount() didn't have a flags word.
2189 * When the flags word was introduced its top half was required
2190 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2191 * Therefore, if this magic number is present, it carries no information
2192 * and must be discarded.
2193 */
2194long do_mount(char *dev_name, char *dir_name, char *type_page,
2195                  unsigned long flags, void *data_page)
2196{
2197        struct path path;
2198        int retval = 0;
2199        int mnt_flags = 0;
2200
2201        /* Discard magic */
2202        if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2203                flags &= ~MS_MGC_MSK;
2204
2205        /* Basic sanity checks */
2206
2207        if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2208                return -EINVAL;
2209
2210        if (data_page)
2211                ((char *)data_page)[PAGE_SIZE - 1] = 0;
2212
2213        /* ... and get the mountpoint */
2214        retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2215        if (retval)
2216                return retval;
2217
2218        retval = security_sb_mount(dev_name, &path,
2219                                   type_page, flags, data_page);
2220        if (retval)
2221                goto dput_out;
2222
2223        /* Default to relatime unless overriden */
2224        if (!(flags & MS_NOATIME))
2225                mnt_flags |= MNT_RELATIME;
2226
2227        /* Separate the per-mountpoint flags */
2228        if (flags & MS_NOSUID)
2229                mnt_flags |= MNT_NOSUID;
2230        if (flags & MS_NODEV)
2231                mnt_flags |= MNT_NODEV;
2232        if (flags & MS_NOEXEC)
2233                mnt_flags |= MNT_NOEXEC;
2234        if (flags & MS_NOATIME)
2235                mnt_flags |= MNT_NOATIME;
2236        if (flags & MS_NODIRATIME)
2237                mnt_flags |= MNT_NODIRATIME;
2238        if (flags & MS_STRICTATIME)
2239                mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2240        if (flags & MS_RDONLY)
2241                mnt_flags |= MNT_READONLY;
2242
2243        flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2244                   MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2245                   MS_STRICTATIME);
2246
2247        if (flags & MS_REMOUNT)
2248                retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2249                                    data_page);
2250        else if (flags & MS_BIND)
2251                retval = do_loopback(&path, dev_name, flags & MS_REC);
2252        else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2253                retval = do_change_type(&path, flags);
2254        else if (flags & MS_MOVE)
2255                retval = do_move_mount(&path, dev_name);
2256        else
2257                retval = do_new_mount(&path, type_page, flags, mnt_flags,
2258                                      dev_name, data_page);
2259dput_out:
2260        path_put(&path);
2261        return retval;
2262}
2263
2264static struct mnt_namespace *alloc_mnt_ns(void)
2265{
2266        struct mnt_namespace *new_ns;
2267
2268        new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2269        if (!new_ns)
2270                return ERR_PTR(-ENOMEM);
2271        atomic_set(&new_ns->count, 1);
2272        new_ns->root = NULL;
2273        INIT_LIST_HEAD(&new_ns->list);
2274        init_waitqueue_head(&new_ns->poll);
2275        new_ns->event = 0;
2276        return new_ns;
2277}
2278
2279/*
2280 * Allocate a new namespace structure and populate it with contents
2281 * copied from the namespace of the passed in task structure.
2282 */
2283static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2284                struct fs_struct *fs)
2285{
2286        struct mnt_namespace *new_ns;
2287        struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2288        struct mount *p, *q;
2289        struct mount *old = mnt_ns->root;
2290        struct mount *new;
2291
2292        new_ns = alloc_mnt_ns();
2293        if (IS_ERR(new_ns))
2294                return new_ns;
2295
2296        down_write(&namespace_sem);
2297        /* First pass: copy the tree topology */
2298        new = copy_tree(old, old->mnt.mnt_root, CL_COPY_ALL | CL_EXPIRE);
2299        if (IS_ERR(new)) {
2300                up_write(&namespace_sem);
2301                kfree(new_ns);
2302                return ERR_CAST(new);
2303        }
2304        new_ns->root = new;
2305        br_write_lock(&vfsmount_lock);
2306        list_add_tail(&new_ns->list, &new->mnt_list);
2307        br_write_unlock(&vfsmount_lock);
2308
2309        /*
2310         * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2311         * as belonging to new namespace.  We have already acquired a private
2312         * fs_struct, so tsk->fs->lock is not needed.
2313         */
2314        p = old;
2315        q = new;
2316        while (p) {
2317                q->mnt_ns = new_ns;
2318                if (fs) {
2319                        if (&p->mnt == fs->root.mnt) {
2320                                fs->root.mnt = mntget(&q->mnt);
2321                                rootmnt = &p->mnt;
2322                        }
2323                        if (&p->mnt == fs->pwd.mnt) {
2324                                fs->pwd.mnt = mntget(&q->mnt);
2325                                pwdmnt = &p->mnt;
2326                        }
2327                }
2328                p = next_mnt(p, old);
2329                q = next_mnt(q, new);
2330        }
2331        up_write(&namespace_sem);
2332
2333        if (rootmnt)
2334                mntput(rootmnt);
2335        if (pwdmnt)
2336                mntput(pwdmnt);
2337
2338        return new_ns;
2339}
2340
2341struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2342                struct fs_struct *new_fs)
2343{
2344        struct mnt_namespace *new_ns;
2345
2346        BUG_ON(!ns);
2347        get_mnt_ns(ns);
2348
2349        if (!(flags & CLONE_NEWNS))
2350                return ns;
2351
2352        new_ns = dup_mnt_ns(ns, new_fs);
2353
2354        put_mnt_ns(ns);
2355        return new_ns;
2356}
2357
2358/**
2359 * create_mnt_ns - creates a private namespace and adds a root filesystem
2360 * @mnt: pointer to the new root filesystem mountpoint
2361 */
2362static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2363{
2364        struct mnt_namespace *new_ns = alloc_mnt_ns();
2365        if (!IS_ERR(new_ns)) {
2366                struct mount *mnt = real_mount(m);
2367                mnt->mnt_ns = new_ns;
2368                new_ns->root = mnt;
2369                list_add(&new_ns->list, &mnt->mnt_list);
2370        } else {
2371                mntput(m);
2372        }
2373        return new_ns;
2374}
2375
2376struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2377{
2378        struct mnt_namespace *ns;
2379        struct super_block *s;
2380        struct path path;
2381        int err;
2382
2383        ns = create_mnt_ns(mnt);
2384        if (IS_ERR(ns))
2385                return ERR_CAST(ns);
2386
2387        err = vfs_path_lookup(mnt->mnt_root, mnt,
2388                        name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2389
2390        put_mnt_ns(ns);
2391
2392        if (err)
2393                return ERR_PTR(err);
2394
2395        /* trade a vfsmount reference for active sb one */
2396        s = path.mnt->mnt_sb;
2397        atomic_inc(&s->s_active);
2398        mntput(path.mnt);
2399        /* lock the sucker */
2400        down_write(&s->s_umount);
2401        /* ... and return the root of (sub)tree on it */
2402        return path.dentry;
2403}
2404EXPORT_SYMBOL(mount_subtree);
2405
2406SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2407                char __user *, type, unsigned long, flags, void __user *, data)
2408{
2409        int ret;
2410        char *kernel_type;
2411        char *kernel_dir;
2412        char *kernel_dev;
2413        unsigned long data_page;
2414
2415        ret = copy_mount_string(type, &kernel_type);
2416        if (ret < 0)
2417                goto out_type;
2418
2419        kernel_dir = getname(dir_name);
2420        if (IS_ERR(kernel_dir)) {
2421                ret = PTR_ERR(kernel_dir);
2422                goto out_dir;
2423        }
2424
2425        ret = copy_mount_string(dev_name, &kernel_dev);
2426        if (ret < 0)
2427                goto out_dev;
2428
2429        ret = copy_mount_options(data, &data_page);
2430        if (ret < 0)
2431                goto out_data;
2432
2433        ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2434                (void *) data_page);
2435
2436        free_page(data_page);
2437out_data:
2438        kfree(kernel_dev);
2439out_dev:
2440        putname(kernel_dir);
2441out_dir:
2442        kfree(kernel_type);
2443out_type:
2444        return ret;
2445}
2446
2447/*
2448 * Return true if path is reachable from root
2449 *
2450 * namespace_sem or vfsmount_lock is held
2451 */
2452bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2453                         const struct path *root)
2454{
2455        while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2456                dentry = mnt->mnt_mountpoint;
2457                mnt = mnt->mnt_parent;
2458        }
2459        return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2460}
2461
2462int path_is_under(struct path *path1, struct path *path2)
2463{
2464        int res;
2465        br_read_lock(&vfsmount_lock);
2466        res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2467        br_read_unlock(&vfsmount_lock);
2468        return res;
2469}
2470EXPORT_SYMBOL(path_is_under);
2471
2472/*
2473 * pivot_root Semantics:
2474 * Moves the root file system of the current process to the directory put_old,
2475 * makes new_root as the new root file system of the current process, and sets
2476 * root/cwd of all processes which had them on the current root to new_root.
2477 *
2478 * Restrictions:
2479 * The new_root and put_old must be directories, and  must not be on the
2480 * same file  system as the current process root. The put_old  must  be
2481 * underneath new_root,  i.e. adding a non-zero number of /.. to the string
2482 * pointed to by put_old must yield the same directory as new_root. No other
2483 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2484 *
2485 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2486 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2487 * in this situation.
2488 *
2489 * Notes:
2490 *  - we don't move root/cwd if they are not at the root (reason: if something
2491 *    cared enough to change them, it's probably wrong to force them elsewhere)
2492 *  - it's okay to pick a root that isn't the root of a file system, e.g.
2493 *    /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2494 *    though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2495 *    first.
2496 */
2497SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2498                const char __user *, put_old)
2499{
2500        struct path new, old, parent_path, root_parent, root;
2501        struct mount *new_mnt, *root_mnt;
2502        int error;
2503
2504        if (!capable(CAP_SYS_ADMIN))
2505                return -EPERM;
2506
2507        error = user_path_dir(new_root, &new);
2508        if (error)
2509                goto out0;
2510
2511        error = user_path_dir(put_old, &old);
2512        if (error)
2513                goto out1;
2514
2515        error = security_sb_pivotroot(&old, &new);
2516        if (error)
2517                goto out2;
2518
2519        get_fs_root(current->fs, &root);
2520        error = lock_mount(&old);
2521        if (error)
2522                goto out3;
2523
2524        error = -EINVAL;
2525        new_mnt = real_mount(new.mnt);
2526        root_mnt = real_mount(root.mnt);
2527        if (IS_MNT_SHARED(real_mount(old.mnt)) ||
2528                IS_MNT_SHARED(new_mnt->mnt_parent) ||
2529                IS_MNT_SHARED(root_mnt->mnt_parent))
2530                goto out4;
2531        if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
2532                goto out4;
2533        error = -ENOENT;
2534        if (d_unlinked(new.dentry))
2535                goto out4;
2536        if (d_unlinked(old.dentry))
2537                goto out4;
2538        error = -EBUSY;
2539        if (new.mnt == root.mnt ||
2540            old.mnt == root.mnt)
2541                goto out4; /* loop, on the same file system  */
2542        error = -EINVAL;
2543        if (root.mnt->mnt_root != root.dentry)
2544                goto out4; /* not a mountpoint */
2545        if (!mnt_has_parent(root_mnt))
2546                goto out4; /* not attached */
2547        if (new.mnt->mnt_root != new.dentry)
2548                goto out4; /* not a mountpoint */
2549        if (!mnt_has_parent(new_mnt))
2550                goto out4; /* not attached */
2551        /* make sure we can reach put_old from new_root */
2552        if (!is_path_reachable(real_mount(old.mnt), old.dentry, &new))
2553                goto out4;
2554        br_write_lock(&vfsmount_lock);
2555        detach_mnt(new_mnt, &parent_path);
2556        detach_mnt(root_mnt, &root_parent);
2557        /* mount old root on put_old */
2558        attach_mnt(root_mnt, &old);
2559        /* mount new_root on / */
2560        attach_mnt(new_mnt, &root_parent);
2561        touch_mnt_namespace(current->nsproxy->mnt_ns);
2562        br_write_unlock(&vfsmount_lock);
2563        chroot_fs_refs(&root, &new);
2564        error = 0;
2565out4:
2566        unlock_mount(&old);
2567        if (!error) {
2568                path_put(&root_parent);
2569                path_put(&parent_path);
2570        }
2571out3:
2572        path_put(&root);
2573out2:
2574        path_put(&old);
2575out1:
2576        path_put(&new);
2577out0:
2578        return error;
2579}
2580
2581static void __init init_mount_tree(void)
2582{
2583        struct vfsmount *mnt;
2584        struct mnt_namespace *ns;
2585        struct path root;
2586
2587        mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2588        if (IS_ERR(mnt))
2589                panic("Can't create rootfs");
2590
2591        ns = create_mnt_ns(mnt);
2592        if (IS_ERR(ns))
2593                panic("Can't allocate initial namespace");
2594
2595        init_task.nsproxy->mnt_ns = ns;
2596        get_mnt_ns(ns);
2597
2598        root.mnt = mnt;
2599        root.dentry = mnt->mnt_root;
2600
2601        set_fs_pwd(current->fs, &root);
2602        set_fs_root(current->fs, &root);
2603}
2604
2605void __init mnt_init(void)
2606{
2607        unsigned u;
2608        int err;
2609
2610        init_rwsem(&namespace_sem);
2611
2612        mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
2613                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2614
2615        mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2616
2617        if (!mount_hashtable)
2618                panic("Failed to allocate mount hash table\n");
2619
2620        printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
2621
2622        for (u = 0; u < HASH_SIZE; u++)
2623                INIT_LIST_HEAD(&mount_hashtable[u]);
2624
2625        br_lock_init(&vfsmount_lock);
2626
2627        err = sysfs_init();
2628        if (err)
2629                printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2630                        __func__, err);
2631        fs_kobj = kobject_create_and_add("fs", NULL);
2632        if (!fs_kobj)
2633                printk(KERN_WARNING "%s: kobj create error\n", __func__);
2634        init_rootfs();
2635        init_mount_tree();
2636}
2637
2638void put_mnt_ns(struct mnt_namespace *ns)
2639{
2640        LIST_HEAD(umount_list);
2641
2642        if (!atomic_dec_and_test(&ns->count))
2643                return;
2644        down_write(&namespace_sem);
2645        br_write_lock(&vfsmount_lock);
2646        umount_tree(ns->root, 0, &umount_list);
2647        br_write_unlock(&vfsmount_lock);
2648        up_write(&namespace_sem);
2649        release_mounts(&umount_list);
2650        kfree(ns);
2651}
2652
2653struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2654{
2655        struct vfsmount *mnt;
2656        mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2657        if (!IS_ERR(mnt)) {
2658                /*
2659                 * it is a longterm mount, don't release mnt until
2660                 * we unmount before file sys is unregistered
2661                */
2662                real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
2663        }
2664        return mnt;
2665}
2666EXPORT_SYMBOL_GPL(kern_mount_data);
2667
2668void kern_unmount(struct vfsmount *mnt)
2669{
2670        /* release long term mount so mount point can be released */
2671        if (!IS_ERR_OR_NULL(mnt)) {
2672                br_write_lock(&vfsmount_lock);
2673                real_mount(mnt)->mnt_ns = NULL;
2674                br_write_unlock(&vfsmount_lock);
2675                mntput(mnt);
2676        }
2677}
2678EXPORT_SYMBOL(kern_unmount);
2679
2680bool our_mnt(struct vfsmount *mnt)
2681{
2682        return check_mnt(real_mount(mnt));
2683}
2684
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.