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