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
 834        list_del(&mnt->mnt_instance);
 835        br_write_unlock(&vfsmount_lock);
 836        mntfree(mnt);
 837}
 838
 839void mntput(struct vfsmount *mnt)
 840{
 841        if (mnt) {
 842                struct mount *m = real_mount(mnt);
 843                /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
 844                if (unlikely(m->mnt_expiry_mark))
 845                        m->mnt_expiry_mark = 0;
 846                mntput_no_expire(m);
 847        }
 848}
 849EXPORT_SYMBOL(mntput);
 850
 851struct vfsmount *mntget(struct vfsmount *mnt)
 852{
 853        if (mnt)
 854                mnt_add_count(real_mount(mnt), 1);
 855        return mnt;
 856}
 857EXPORT_SYMBOL(mntget);
 858
 859void mnt_pin(struct vfsmount *mnt)
 860{
 861        br_write_lock(&vfsmount_lock);
 862        real_mount(mnt)->mnt_pinned++;
 863        br_write_unlock(&vfsmount_lock);
 864}
 865EXPORT_SYMBOL(mnt_pin);
 866
 867void mnt_unpin(struct vfsmount *m)
 868{
 869        struct mount *mnt = real_mount(m);
 870        br_write_lock(&vfsmount_lock);
 871        if (mnt->mnt_pinned) {
 872                mnt_add_count(mnt, 1);
 873                mnt->mnt_pinned--;
 874        }
 875        br_write_unlock(&vfsmount_lock);
 876}
 877EXPORT_SYMBOL(mnt_unpin);
 878
 879static inline void mangle(struct seq_file *m, const char *s)
 880{
 881        seq_escape(m, s, " \t\n\\");
 882}
 883
 884/*
 885 * Simple .show_options callback for filesystems which don't want to
 886 * implement more complex mount option showing.
 887 *
 888 * See also save_mount_options().
 889 */
 890int generic_show_options(struct seq_file *m, struct dentry *root)
 891{
 892        const char *options;
 893
 894        rcu_read_lock();
 895        options = rcu_dereference(root->d_sb->s_options);
 896
 897        if (options != NULL && options[0]) {
 898                seq_putc(m, ',');
 899                mangle(m, options);
 900        }
 901        rcu_read_unlock();
 902
 903        return 0;
 904}
 905EXPORT_SYMBOL(generic_show_options);
 906
 907/*
 908 * If filesystem uses generic_show_options(), this function should be
 909 * called from the fill_super() callback.
 910 *
 911 * The .remount_fs callback usually needs to be handled in a special
 912 * way, to make sure, that previous options are not overwritten if the
 913 * remount fails.
 914 *
 915 * Also note, that if the filesystem's .remount_fs function doesn't
 916 * reset all options to their default value, but changes only newly
 917 * given options, then the displayed options will not reflect reality
 918 * any more.
 919 */
 920void save_mount_options(struct super_block *sb, char *options)
 921{
 922        BUG_ON(sb->s_options);
 923        rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
 924}
 925EXPORT_SYMBOL(save_mount_options);
 926
 927void replace_mount_options(struct super_block *sb, char *options)
 928{
 929        char *old = sb->s_options;
 930        rcu_assign_pointer(sb->s_options, options);
 931        if (old) {
 932                synchronize_rcu();
 933                kfree(old);
 934        }
 935}
 936EXPORT_SYMBOL(replace_mount_options);
 937
 938#ifdef CONFIG_PROC_FS
 939/* iterator; we want it to have access to namespace_sem, thus here... */
 940static void *m_start(struct seq_file *m, loff_t *pos)
 941{
 942        struct proc_mounts *p = container_of(m, struct proc_mounts, m);
 943
 944        down_read(&namespace_sem);
 945        return seq_list_start(&p->ns->list, *pos);
 946}
 947
 948static void *m_next(struct seq_file *m, void *v, loff_t *pos)
 949{
 950        struct proc_mounts *p = container_of(m, struct proc_mounts, m);
 951
 952        return seq_list_next(v, &p->ns->list, pos);
 953}
 954
 955static void m_stop(struct seq_file *m, void *v)
 956{
 957        up_read(&namespace_sem);
 958}
 959
 960static int m_show(struct seq_file *m, void *v)
 961{
 962        struct proc_mounts *p = container_of(m, struct proc_mounts, m);
 963        struct mount *r = list_entry(v, struct mount, mnt_list);
 964        return p->show(m, &r->mnt);
 965}
 966
 967const struct seq_operations mounts_op = {
 968        .start  = m_start,
 969        .next   = m_next,
 970        .stop   = m_stop,
 971        .show   = m_show,
 972};
 973#endif  /* CONFIG_PROC_FS */
 974
 975/**
 976 * may_umount_tree - check if a mount tree is busy
 977 * @mnt: root of mount tree
 978 *
 979 * This is called to check if a tree of mounts has any
 980 * open files, pwds, chroots or sub mounts that are
 981 * busy.
 982 */
 983int may_umount_tree(struct vfsmount *m)
 984{
 985        struct mount *mnt = real_mount(m);
 986        int actual_refs = 0;
 987        int minimum_refs = 0;
 988        struct mount *p;
 989        BUG_ON(!m);
 990
 991        /* write lock needed for mnt_get_count */
 992        br_write_lock(&vfsmount_lock);
 993        for (p = mnt; p; p = next_mnt(p, mnt)) {
 994                actual_refs += mnt_get_count(p);
 995                minimum_refs += 2;
 996        }
 997        br_write_unlock(&vfsmount_lock);
 998
 999        if (actual_refs > minimum_refs)
1000                return 0;
1001
1002        return 1;
1003}
1004
1005EXPORT_SYMBOL(may_umount_tree);
1006
1007/**
1008 * may_umount - check if a mount point is busy
1009 * @mnt: root of mount
1010 *
1011 * This is called to check if a mount point has any
1012 * open files, pwds, chroots or sub mounts. If the
1013 * mount has sub mounts this will return busy
1014 * regardless of whether the sub mounts are busy.
1015 *
1016 * Doesn't take quota and stuff into account. IOW, in some cases it will
1017 * give false negatives. The main reason why it's here is that we need
1018 * a non-destructive way to look for easily umountable filesystems.
1019 */
1020int may_umount(struct vfsmount *mnt)
1021{
1022        int ret = 1;
1023        down_read(&namespace_sem);
1024        br_write_lock(&vfsmount_lock);
1025        if (propagate_mount_busy(real_mount(mnt), 2))
1026                ret = 0;
1027        br_write_unlock(&vfsmount_lock);
1028        up_read(&namespace_sem);
1029        return ret;
1030}
1031
1032EXPORT_SYMBOL(may_umount);
1033
1034void release_mounts(struct list_head *head)
1035{
1036        struct mount *mnt;
1037        while (!list_empty(head)) {
1038                mnt = list_first_entry(head, struct mount, mnt_hash);
1039                list_del_init(&mnt->mnt_hash);
1040                if (mnt_has_parent(mnt)) {
1041                        struct dentry *dentry;
1042                        struct mount *m;
1043
1044                        br_write_lock(&vfsmount_lock);
1045                        dentry = mnt->mnt_mountpoint;
1046                        m = mnt->mnt_parent;
1047                        mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1048                        mnt->mnt_parent = mnt;
1049                        m->mnt_ghosts--;
1050                        br_write_unlock(&vfsmount_lock);
1051                        dput(dentry);
1052                        mntput(&m->mnt);
1053                }
1054                mntput(&mnt->mnt);
1055        }
1056}
1057
1058/*
1059 * vfsmount lock must be held for write
1060 * namespace_sem must be held for write
1061 */
1062void umount_tree(struct mount *mnt, int propagate, struct list_head *kill)
1063{
1064        LIST_HEAD(tmp_list);
1065        struct mount *p;
1066
1067        for (p = mnt; p; p = next_mnt(p, mnt))
1068                list_move(&p->mnt_hash, &tmp_list);
1069
1070        if (propagate)
1071                propagate_umount(&tmp_list);
1072
1073        list_for_each_entry(p, &tmp_list, mnt_hash) {
1074                list_del_init(&p->mnt_expire);
1075                list_del_init(&p->mnt_list);
1076                __touch_mnt_namespace(p->mnt_ns);
1077                if (p->mnt_ns)
1078                        __mnt_make_shortterm(p);
1079                p->mnt_ns = NULL;
1080                list_del_init(&p->mnt_child);
1081                if (mnt_has_parent(p)) {
1082                        p->mnt_parent->mnt_ghosts++;
1083                        dentry_reset_mounted(p->mnt_mountpoint);
1084                }
1085                change_mnt_propagation(p, MS_PRIVATE);
1086        }
1087        list_splice(&tmp_list, kill);
1088}
1089
1090static void shrink_submounts(struct mount *mnt, struct list_head *umounts);
1091
1092static int do_umount(struct mount *mnt, int flags)
1093{
1094        struct super_block *sb = mnt->mnt.mnt_sb;
1095        int retval;
1096        LIST_HEAD(umount_list);
1097
1098        retval = security_sb_umount(&mnt->mnt, flags);
1099        if (retval)
1100                return retval;
1101
1102        /*
1103         * Allow userspace to request a mountpoint be expired rather than
1104         * unmounting unconditionally. Unmount only happens if:
1105         *  (1) the mark is already set (the mark is cleared by mntput())
1106         *  (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1107         */
1108        if (flags & MNT_EXPIRE) {
1109                if (&mnt->mnt == current->fs->root.mnt ||
1110                    flags & (MNT_FORCE | MNT_DETACH))
1111                        return -EINVAL;
1112
1113                /*
1114                 * probably don't strictly need the lock here if we examined
1115                 * all race cases, but it's a slowpath.
1116                 */
1117                br_write_lock(&vfsmount_lock);
1118                if (mnt_get_count(mnt) != 2) {
1119                        br_write_unlock(&vfsmount_lock);
1120                        return -EBUSY;
1121                }
1122                br_write_unlock(&vfsmount_lock);
1123
1124                if (!xchg(&mnt->mnt_expiry_mark, 1))
1125                        return -EAGAIN;
1126        }
1127
1128        /*
1129         * If we may have to abort operations to get out of this
1130         * mount, and they will themselves hold resources we must
1131         * allow the fs to do things. In the Unix tradition of
1132         * 'Gee thats tricky lets do it in userspace' the umount_begin
1133         * might fail to complete on the first run through as other tasks
1134         * must return, and the like. Thats for the mount program to worry
1135         * about for the moment.
1136         */
1137
1138        if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1139                sb->s_op->umount_begin(sb);
1140        }
1141
1142        /*
1143         * No sense to grab the lock for this test, but test itself looks
1144         * somewhat bogus. Suggestions for better replacement?
1145         * Ho-hum... In principle, we might treat that as umount + switch
1146         * to rootfs. GC would eventually take care of the old vfsmount.
1147         * Actually it makes sense, especially if rootfs would contain a
1148         * /reboot - static binary that would close all descriptors and
1149         * call reboot(9). Then init(8) could umount root and exec /reboot.
1150         */
1151        if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1152                /*
1153                 * Special case for "unmounting" root ...
1154                 * we just try to remount it readonly.
1155                 */
1156                down_write(&sb->s_umount);
1157                if (!(sb->s_flags & MS_RDONLY))
1158                        retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1159                up_write(&sb->s_umount);
1160                return retval;
1161        }
1162
1163        down_write(&namespace_sem);
1164        br_write_lock(&vfsmount_lock);
1165        event++;
1166
1167        if (!(flags & MNT_DETACH))
1168                shrink_submounts(mnt, &umount_list);
1169
1170        retval = -EBUSY;
1171        if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1172                if (!list_empty(&mnt->mnt_list))
1173                        umount_tree(mnt, 1, &umount_list);
1174                retval = 0;
1175        }
1176        br_write_unlock(&vfsmount_lock);
1177        up_write(&namespace_sem);
1178        release_mounts(&umount_list);
1179        return retval;
1180}
1181
1182/*
1183 * Now umount can handle mount points as well as block devices.
1184 * This is important for filesystems which use unnamed block devices.
1185 *
1186 * We now support a flag for forced unmount like the other 'big iron'
1187 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1188 */
1189
1190SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1191{
1192        struct path path;
1193        struct mount *mnt;
1194        int retval;
1195        int lookup_flags = 0;
1196
1197        if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1198                return -EINVAL;
1199
1200        if (!(flags & UMOUNT_NOFOLLOW))
1201                lookup_flags |= LOOKUP_FOLLOW;
1202
1203        retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1204        if (retval)
1205                goto out;
1206        mnt = real_mount(path.mnt);
1207        retval = -EINVAL;
1208        if (path.dentry != path.mnt->mnt_root)
1209                goto dput_and_out;
1210        if (!check_mnt(mnt))
1211                goto dput_and_out;
1212
1213        retval = -EPERM;
1214        if (!capable(CAP_SYS_ADMIN))
1215                goto dput_and_out;
1216
1217        retval = do_umount(mnt, flags);
1218dput_and_out:
1219        /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1220        dput(path.dentry);
1221        mntput_no_expire(mnt);
1222out:
1223        return retval;
1224}
1225
1226#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1227
1228/*
1229 *      The 2.0 compatible umount. No flags.
1230 */
1231SYSCALL_DEFINE1(oldumount, char __user *, name)
1232{
1233        return sys_umount(name, 0);
1234}
1235
1236#endif
1237
1238static int mount_is_safe(struct path *path)
1239{
1240        if (capable(CAP_SYS_ADMIN))
1241                return 0;
1242        return -EPERM;
1243#ifdef notyet
1244        if (S_ISLNK(path->dentry->d_inode->i_mode))
1245                return -EPERM;
1246        if (path->dentry->d_inode->i_mode & S_ISVTX) {
1247                if (current_uid() != path->dentry->d_inode->i_uid)
1248                        return -EPERM;
1249        }
1250        if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1251                return -EPERM;
1252        return 0;
1253#endif
1254}
1255
1256struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1257                                        int flag)
1258{
1259        struct mount *res, *p, *q, *r;
1260        struct path path;
1261
1262        if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1263                return NULL;
1264
1265        res = q = clone_mnt(mnt, dentry, flag);
1266        if (!q)
1267                goto Enomem;
1268        q->mnt_mountpoint = mnt->mnt_mountpoint;
1269
1270        p = mnt;
1271        list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1272                struct mount *s;
1273                if (!is_subdir(r->mnt_mountpoint, dentry))
1274                        continue;
1275
1276                for (s = r; s; s = next_mnt(s, r)) {
1277                        if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1278                                s = skip_mnt_tree(s);
1279                                continue;
1280                        }
1281                        while (p != s->mnt_parent) {
1282                                p = p->mnt_parent;
1283                                q = q->mnt_parent;
1284                        }
1285                        p = s;
1286                        path.mnt = &q->mnt;
1287                        path.dentry = p->mnt_mountpoint;
1288                        q = clone_mnt(p, p->mnt.mnt_root, flag);
1289                        if (!q)
1290                                goto Enomem;
1291                        br_write_lock(&vfsmount_lock);
1292                        list_add_tail(&q->mnt_list, &res->mnt_list);
1293                        attach_mnt(q, &path);
1294                        br_write_unlock(&vfsmount_lock);
1295                }
1296        }
1297        return res;
1298Enomem:
1299        if (res) {
1300                LIST_HEAD(umount_list);
1301                br_write_lock(&vfsmount_lock);
1302                umount_tree(res, 0, &umount_list);
1303                br_write_unlock(&vfsmount_lock);
1304                release_mounts(&umount_list);
1305        }
1306        return NULL;
1307}
1308
1309struct vfsmount *collect_mounts(struct path *path)
1310{
1311        struct mount *tree;
1312        down_write(&namespace_sem);
1313        tree = copy_tree(real_mount(path->mnt), path->dentry,
1314                         CL_COPY_ALL | CL_PRIVATE);
1315        up_write(&namespace_sem);
1316        return tree ? &tree->mnt : NULL;
1317}
1318
1319void drop_collected_mounts(struct vfsmount *mnt)
1320{
1321        LIST_HEAD(umount_list);
1322        down_write(&namespace_sem);
1323        br_write_lock(&vfsmount_lock);
1324        umount_tree(real_mount(mnt), 0, &umount_list);
1325        br_write_unlock(&vfsmount_lock);
1326        up_write(&namespace_sem);
1327        release_mounts(&umount_list);
1328}
1329
1330int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1331                   struct vfsmount *root)
1332{
1333        struct mount *mnt;
1334        int res = f(root, arg);
1335        if (res)
1336                return res;
1337        list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1338                res = f(&mnt->mnt, arg);
1339                if (res)
1340                        return res;
1341        }
1342        return 0;
1343}
1344
1345static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1346{
1347        struct mount *p;
1348
1349        for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1350                if (p->mnt_group_id && !IS_MNT_SHARED(p))
1351                        mnt_release_group_id(p);
1352        }
1353}
1354
1355static int invent_group_ids(struct mount *mnt, bool recurse)
1356{
1357        struct mount *p;
1358
1359        for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1360                if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1361                        int err = mnt_alloc_group_id(p);
1362                        if (err) {
1363                                cleanup_group_ids(mnt, p);
1364                                return err;
1365                        }
1366                }
1367        }
1368
1369        return 0;
1370}
1371
1372/*
1373 *  @source_mnt : mount tree to be attached
1374 *  @nd         : place the mount tree @source_mnt is attached
1375 *  @parent_nd  : if non-null, detach the source_mnt from its parent and
1376 *                 store the parent mount and mountpoint dentry.
1377 *                 (done when source_mnt is moved)
1378 *
1379 *  NOTE: in the table below explains the semantics when a source mount
1380 *  of a given type is attached to a destination mount of a given type.
1381 * ---------------------------------------------------------------------------
1382 * |         BIND MOUNT OPERATION                                            |
1383 * |**************************************************************************
1384 * | source-->| shared        |       private  |       slave    | unbindable |
1385 * | dest     |               |                |                |            |
1386 * |   |      |               |                |                |            |
1387 * |   v      |               |                |                |            |
1388 * |**************************************************************************
1389 * |  shared  | shared (++)   |     shared (+) |     shared(+++)|  invalid   |
1390 * |          |               |                |                |            |
1391 * |non-shared| shared (+)    |      private   |      slave (*) |  invalid   |
1392 * ***************************************************************************
1393 * A bind operation clones the source mount and mounts the clone on the
1394 * destination mount.
1395 *
1396 * (++)  the cloned mount is propagated to all the mounts in the propagation
1397 *       tree of the destination mount and the cloned mount is added to
1398 *       the peer group of the source mount.
1399 * (+)   the cloned mount is created under the destination mount and is marked
1400 *       as shared. The cloned mount is added to the peer group of the source
1401 *       mount.
1402 * (+++) the mount is propagated to all the mounts in the propagation tree
1403 *       of the destination mount and the cloned mount is made slave
1404 *       of the same master as that of the source mount. The cloned mount
1405 *       is marked as 'shared and slave'.
1406 * (*)   the cloned mount is made a slave of the same master as that of the
1407 *       source mount.
1408 *
1409 * ---------------------------------------------------------------------------
1410 * |                    MOVE MOUNT OPERATION                                 |
1411 * |**************************************************************************
1412 * | source-->| shared        |       private  |       slave    | unbindable |
1413 * | dest     |               |                |                |            |
1414 * |   |      |               |                |                |            |
1415 * |   v      |               |                |                |            |
1416 * |**************************************************************************
1417 * |  shared  | shared (+)    |     shared (+) |    shared(+++) |  invalid   |
1418 * |          |               |                |                |            |
1419 * |non-shared| shared (+*)   |      private   |    slave (*)   | unbindable |
1420 * ***************************************************************************
1421 *
1422 * (+)  the mount is moved to the destination. And is then propagated to
1423 *      all the mounts in the propagation tree of the destination mount.
1424 * (+*)  the mount is moved to the destination.
1425 * (+++)  the mount is moved to the destination and is then propagated to
1426 *      all the mounts belonging to the destination mount's propagation tree.
1427 *      the mount is marked as 'shared and slave'.
1428 * (*)  the mount continues to be a slave at the new location.
1429 *
1430 * if the source mount is a tree, the operations explained above is
1431 * applied to each mount in the tree.
1432 * Must be called without spinlocks held, since this function can sleep
1433 * in allocations.
1434 */
1435static int attach_recursive_mnt(struct mount *source_mnt,
1436                        struct path *path, struct path *parent_path)
1437{
1438        LIST_HEAD(tree_list);
1439        struct mount *dest_mnt = real_mount(path->mnt);
1440        struct dentry *dest_dentry = path->dentry;
1441        struct mount *child, *p;
1442        int err;
1443
1444        if (IS_MNT_SHARED(dest_mnt)) {
1445                err = invent_group_ids(source_mnt, true);
1446                if (err)
1447                        goto out;
1448        }
1449        err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1450        if (err)
1451                goto out_cleanup_ids;
1452
1453        br_write_lock(&vfsmount_lock);
1454
1455        if (IS_MNT_SHARED(dest_mnt)) {
1456                for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1457                        set_mnt_shared(p);
1458        }
1459        if (parent_path) {
1460                detach_mnt(source_mnt, parent_path);
1461                attach_mnt(source_mnt, path);
1462                touch_mnt_namespace(source_mnt->mnt_ns);
1463        } else {
1464                mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1465                commit_tree(source_mnt);
1466        }
1467
1468        list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1469                list_del_init(&child->mnt_hash);
1470                commit_tree(child);
1471        }
1472        br_write_unlock(&vfsmount_lock);
1473
1474        return 0;
1475
1476 out_cleanup_ids:
1477        if (IS_MNT_SHARED(dest_mnt))
1478                cleanup_group_ids(source_mnt, NULL);
1479 out:
1480        return err;
1481}
1482
1483static int lock_mount(struct path *path)
1484{
1485        struct vfsmount *mnt;
1486retry:
1487        mutex_lock(&path->dentry->d_inode->i_mutex);
1488        if (unlikely(cant_mount(path->dentry))) {
1489                mutex_unlock(&path->dentry->d_inode->i_mutex);
1490                return -ENOENT;
1491        }
1492        down_write(&namespace_sem);
1493        mnt = lookup_mnt(path);
1494        if (likely(!mnt))
1495                return 0;
1496        up_write(&namespace_sem);
1497        mutex_unlock(&path->dentry->d_inode->i_mutex);
1498        path_put(path);
1499        path->mnt = mnt;
1500        path->dentry = dget(mnt->mnt_root);
1501        goto retry;
1502}
1503
1504static void unlock_mount(struct path *path)
1505{
1506        up_write(&namespace_sem);
1507        mutex_unlock(&path->dentry->d_inode->i_mutex);
1508}
1509
1510static int graft_tree(struct mount *mnt, struct path *path)
1511{
1512        if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1513                return -EINVAL;
1514
1515        if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1516              S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
1517                return -ENOTDIR;
1518
1519        if (d_unlinked(path->dentry))
1520                return -ENOENT;
1521
1522        return attach_recursive_mnt(mnt, path, NULL);
1523}
1524
1525/*
1526 * Sanity check the flags to change_mnt_propagation.
1527 */
1528
1529static int flags_to_propagation_type(int flags)
1530{
1531        int type = flags & ~(MS_REC | MS_SILENT);
1532
1533        /* Fail if any non-propagation flags are set */
1534        if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1535                return 0;
1536        /* Only one propagation flag should be set */
1537        if (!is_power_of_2(type))
1538                return 0;
1539        return type;
1540}
1541
1542/*
1543 * recursively change the type of the mountpoint.
1544 */
1545static int do_change_type(struct path *path, int flag)
1546{
1547        struct mount *m;
1548        struct mount *mnt = real_mount(path->mnt);
1549        int recurse = flag & MS_REC;
1550        int type;
1551        int err = 0;
1552
1553        if (!capable(CAP_SYS_ADMIN))
1554                return -EPERM;
1555
1556        if (path->dentry != path->mnt->mnt_root)
1557                return -EINVAL;
1558
1559        type = flags_to_propagation_type(flag);
1560        if (!type)
1561                return -EINVAL;
1562
1563        down_write(&namespace_sem);
1564        if (type == MS_SHARED) {
1565                err = invent_group_ids(mnt, recurse);
1566                if (err)
1567                        goto out_unlock;
1568        }
1569
1570        br_write_lock(&vfsmount_lock);
1571        for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1572                change_mnt_propagation(m, type);
1573        br_write_unlock(&vfsmount_lock);
1574
1575 out_unlock:
1576        up_write(&namespace_sem);
1577        return err;
1578}
1579
1580/*
1581 * do loopback mount.
1582 */
1583static int do_loopback(struct path *path, char *old_name,
1584                                int recurse)
1585{
1586        LIST_HEAD(umount_list);
1587        struct path old_path;
1588        struct mount *mnt = NULL, *old;
1589        int err = mount_is_safe(path);
1590        if (err)
1591                return err;
1592        if (!old_name || !*old_name)
1593                return -EINVAL;
1594        err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1595        if (err)
1596                return err;
1597
1598        err = lock_mount(path);
1599        if (err)
1600                goto out;
1601
1602        old = real_mount(old_path.mnt);
1603
1604        err = -EINVAL;
1605        if (IS_MNT_UNBINDABLE(old))
1606                goto out2;
1607
1608        if (!check_mnt(real_mount(path->mnt)) || !check_mnt(old))
1609                goto out2;
1610
1611        err = -ENOMEM;
1612        if (recurse)
1613                mnt = copy_tree(old, old_path.dentry, 0);
1614        else
1615                mnt = clone_mnt(old, old_path.dentry, 0);
1616
1617        if (!mnt)
1618                goto out2;
1619
1620        err = graft_tree(mnt, path);
1621        if (err) {
1622                br_write_lock(&vfsmount_lock);
1623                umount_tree(mnt, 0, &umount_list);
1624                br_write_unlock(&vfsmount_lock);
1625        }
1626out2:
1627        unlock_mount(path);
1628        release_mounts(&umount_list);
1629out:
1630        path_put(&old_path);
1631        return err;
1632}
1633
1634static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1635{
1636        int error = 0;
1637        int readonly_request = 0;
1638
1639        if (ms_flags & MS_RDONLY)
1640                readonly_request = 1;
1641        if (readonly_request == __mnt_is_readonly(mnt))
1642                return 0;
1643
1644        if (readonly_request)
1645                error = mnt_make_readonly(real_mount(mnt));
1646        else
1647                __mnt_unmake_readonly(real_mount(mnt));
1648        return error;
1649}
1650
1651/*
1652 * change filesystem flags. dir should be a physical root of filesystem.
1653 * If you've mounted a non-root directory somewhere and want to do remount
1654 * on it - tough luck.
1655 */
1656static int do_remount(struct path *path, int flags, int mnt_flags,
1657                      void *data)
1658{
1659        int err;
1660        struct super_block *sb = path->mnt->mnt_sb;
1661        struct mount *mnt = real_mount(path->mnt);
1662
1663        if (!capable(CAP_SYS_ADMIN))
1664                return -EPERM;
1665
1666        if (!check_mnt(mnt))
1667                return -EINVAL;
1668
1669        if (path->dentry != path->mnt->mnt_root)
1670                return -EINVAL;
1671
1672        err = security_sb_remount(sb, data);
1673        if (err)
1674                return err;
1675
1676        down_write(&sb->s_umount);
1677        if (flags & MS_BIND)
1678                err = change_mount_flags(path->mnt, flags);
1679        else
1680                err = do_remount_sb(sb, flags, data, 0);
1681        if (!err) {
1682                br_write_lock(&vfsmount_lock);
1683                mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK;
1684                mnt->mnt.mnt_flags = mnt_flags;
1685                br_write_unlock(&vfsmount_lock);
1686        }
1687        up_write(&sb->s_umount);
1688        if (!err) {
1689                br_write_lock(&vfsmount_lock);
1690                touch_mnt_namespace(mnt->mnt_ns);
1691                br_write_unlock(&vfsmount_lock);
1692        }
1693        return err;
1694}
1695
1696static inline int tree_contains_unbindable(struct mount *mnt)
1697{
1698        struct mount *p;
1699        for (p = mnt; p; p = next_mnt(p, mnt)) {
1700                if (IS_MNT_UNBINDABLE(p))
1701                        return 1;
1702        }
1703        return 0;
1704}
1705
1706static int do_move_mount(struct path *path, char *old_name)
1707{
1708        struct path old_path, parent_path;
1709        struct mount *p;
1710        struct mount *old;
1711        int err = 0;
1712        if (!capable(CAP_SYS_ADMIN))
1713                return -EPERM;
1714        if (!old_name || !*old_name)
1715                return -EINVAL;
1716        err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1717        if (err)
1718                return err;
1719
1720        err = lock_mount(path);
1721        if (err < 0)
1722                goto out;
1723
1724        old = real_mount(old_path.mnt);
1725        p = real_mount(path->mnt);
1726
1727        err = -EINVAL;
1728        if (!check_mnt(p) || !check_mnt(old))
1729                goto out1;
1730
1731        if (d_unlinked(path->dentry))
1732                goto out1;
1733
1734        err = -EINVAL;
1735        if (old_path.dentry != old_path.mnt->mnt_root)
1736                goto out1;
1737
1738        if (!mnt_has_parent(old))
1739                goto out1;
1740
1741        if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1742              S_ISDIR(old_path.dentry->d_inode->i_mode))
1743                goto out1;
1744        /*
1745         * Don't move a mount residing in a shared parent.
1746         */
1747        if (IS_MNT_SHARED(old->mnt_parent))
1748                goto out1;
1749        /*
1750         * Don't move a mount tree containing unbindable mounts to a destination
1751         * mount which is shared.
1752         */
1753        if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
1754                goto out1;
1755        err = -ELOOP;
1756        for (; mnt_has_parent(p); p = p->mnt_parent)
1757                if (p == old)
1758                        goto out1;
1759
1760        err = attach_recursive_mnt(old, path, &parent_path);
1761        if (err)
1762                goto out1;
1763
1764        /* if the mount is moved, it should no longer be expire
1765         * automatically */
1766        list_del_init(&old->mnt_expire);
1767out1:
1768        unlock_mount(path);
1769out:
1770        if (!err)
1771                path_put(&parent_path);
1772        path_put(&old_path);
1773        return err;
1774}
1775
1776static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1777{
1778        int err;
1779        const char *subtype = strchr(fstype, '.');
1780        if (subtype) {
1781                subtype++;
1782                err = -EINVAL;
1783                if (!subtype[0])
1784                        goto err;
1785        } else
1786                subtype = "";
1787
1788        mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1789        err = -ENOMEM;
1790        if (!mnt->mnt_sb->s_subtype)
1791                goto err;
1792        return mnt;
1793
1794 err:
1795        mntput(mnt);
1796        return ERR_PTR(err);
1797}
1798
1799static struct vfsmount *
1800do_kern_mount(const char *fstype, int flags, const char *name, void *data)
1801{
1802        struct file_system_type *type = get_fs_type(fstype);
1803        struct vfsmount *mnt;
1804        if (!type)
1805                return ERR_PTR(-ENODEV);
1806        mnt = vfs_kern_mount(type, flags, name, data);
1807        if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1808            !mnt->mnt_sb->s_subtype)
1809                mnt = fs_set_subtype(mnt, fstype);
1810        put_filesystem(type);
1811        return mnt;
1812}
1813
1814/*
1815 * add a mount into a namespace's mount tree
1816 */
1817static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
1818{
1819        int err;
1820
1821        mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1822
1823        err = lock_mount(path);
1824        if (err)
1825                return err;
1826
1827        err = -EINVAL;
1828        if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(real_mount(path->mnt)))
1829                goto unlock;
1830
1831        /* Refuse the same filesystem on the same mount point */
1832        err = -EBUSY;
1833        if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
1834            path->mnt->mnt_root == path->dentry)
1835                goto unlock;
1836
1837        err = -EINVAL;
1838        if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
1839                goto unlock;
1840
1841        newmnt->mnt.mnt_flags = mnt_flags;
1842        err = graft_tree(newmnt, path);
1843
1844unlock:
1845        unlock_mount(path);
1846        return err;
1847}
1848
1849/*
1850 * create a new mount for userspace and request it to be added into the
1851 * namespace's tree
1852 */
1853static int do_new_mount(struct path *path, char *type, int flags,
1854                        int mnt_flags, char *name, void *data)
1855{
1856        struct vfsmount *mnt;
1857        int err;
1858
1859        if (!type)
1860                return -EINVAL;
1861
1862        /* we need capabilities... */
1863        if (!capable(CAP_SYS_ADMIN))
1864                return -EPERM;
1865
1866        mnt = do_kern_mount(type, flags, name, data);
1867        if (IS_ERR(mnt))
1868                return PTR_ERR(mnt);
1869
1870        err = do_add_mount(real_mount(mnt), path, mnt_flags);
1871        if (err)
1872                mntput(mnt);
1873        return err;
1874}
1875
1876int finish_automount(struct vfsmount *m, struct path *path)
1877{
1878        struct mount *mnt = real_mount(m);
1879        int err;
1880        /* The new mount record should have at least 2 refs to prevent it being
1881         * expired before we get a chance to add it
1882         */
1883        BUG_ON(mnt_get_count(mnt) < 2);
1884
1885        if (m->mnt_sb == path->mnt->mnt_sb &&
1886            m->mnt_root == path->dentry) {
1887                err = -ELOOP;
1888                goto fail;
1889        }
1890
1891        err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
1892        if (!err)
1893                return 0;
1894fail:
1895        /* remove m from any expiration list it may be on */
1896        if (!list_empty(&mnt->mnt_expire)) {
1897                down_write(&namespace_sem);
1898                br_write_lock(&vfsmount_lock);
1899                list_del_init(&mnt->mnt_expire);
1900                br_write_unlock(&vfsmount_lock);
1901                up_write(&namespace_sem);
1902        }
1903        mntput(m);
1904        mntput(m);
1905        return err;
1906}
1907
1908/**
1909 * mnt_set_expiry - Put a mount on an expiration list
1910 * @mnt: The mount to list.
1911 * @expiry_list: The list to add the mount to.
1912 */
1913void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
1914{
1915        down_write(&namespace_sem);
1916        br_write_lock(&vfsmount_lock);
1917
1918        list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
1919
1920        br_write_unlock(&vfsmount_lock);
1921        up_write(&namespace_sem);
1922}
1923EXPORT_SYMBOL(mnt_set_expiry);
1924
1925/*
1926 * process a list of expirable mountpoints with the intent of discarding any
1927 * mountpoints that aren't in use and haven't been touched since last we came
1928 * here
1929 */
1930void mark_mounts_for_expiry(struct list_head *mounts)
1931{
1932        struct mount *mnt, *next;
1933        LIST_HEAD(graveyard);
1934        LIST_HEAD(umounts);
1935
1936        if (list_empty(mounts))
1937                return;
1938
1939        down_write(&namespace_sem);
1940        br_write_lock(&vfsmount_lock);
1941
1942        /* extract from the expiration list every vfsmount that matches the
1943         * following criteria:
1944         * - only referenced by its parent vfsmount
1945         * - still marked for expiry (marked on the last call here; marks are
1946         *   cleared by mntput())
1947         */
1948        list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1949                if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1950                        propagate_mount_busy(mnt, 1))
1951                        continue;
1952                list_move(&mnt->mnt_expire, &graveyard);
1953        }
1954        while (!list_empty(&graveyard)) {
1955                mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
1956                touch_mnt_namespace(mnt->mnt_ns);
1957                umount_tree(mnt, 1, &umounts);
1958        }
1959        br_write_unlock(&vfsmount_lock);
1960        up_write(&namespace_sem);
1961
1962        release_mounts(&umounts);
1963}
1964
1965EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1966
1967/*
1968 * Ripoff of 'select_parent()'
1969 *
1970 * search the list of submounts for a given mountpoint, and move any
1971 * shrinkable submounts to the 'graveyard' list.
1972 */
1973static int select_submounts(struct mount *parent, struct list_head *graveyard)
1974{
1975        struct mount *this_parent = parent;
1976        struct list_head *next;
1977        int found = 0;
1978
1979repeat:
1980        next = this_parent->mnt_mounts.next;
1981resume:
1982        while (next != &this_parent->mnt_mounts) {
1983                struct list_head *tmp = next;
1984                struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
1985
1986                next = tmp->next;
1987                if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
1988                        continue;
1989                /*
1990                 * Descend a level if the d_mounts list is non-empty.
1991                 */
1992                if (!list_empty(&mnt->mnt_mounts)) {
1993                        this_parent = mnt;
1994                        goto repeat;
1995                }
1996
1997                if (!propagate_mount_busy(mnt, 1)) {
1998                        list_move_tail(&mnt->mnt_expire, graveyard);
1999                        found++;
2000                }
2001        }
2002        /*
2003         * All done at this level ... ascend and resume the search
2004         */
2005        if (this_parent != parent) {
2006                next = this_parent->mnt_child.next;
2007                this_parent = this_parent->mnt_parent;
2008                goto resume;
2009        }
2010        return found;
2011}
2012
2013/*
2014 * process a list of expirable mountpoints with the intent of discarding any
2015 * submounts of a specific parent mountpoint
2016 *
2017 * vfsmount_lock must be held for write
2018 */
2019static void shrink_submounts(struct mount *mnt, struct list_head *umounts)
2020{
2021        LIST_HEAD(graveyard);
2022        struct mount *m;
2023
2024        /* extract submounts of 'mountpoint' from the expiration list */
2025        while (select_submounts(mnt, &graveyard)) {
2026                while (!list_empty(&graveyard)) {
2027                        m = list_first_entry(&graveyard, struct mount,
2028                                                mnt_expire);
2029                        touch_mnt_namespace(m->mnt_ns);
2030                        umount_tree(m, 1, umounts);
2031                }
2032        }
2033}
2034
2035/*
2036 * Some copy_from_user() implementations do not return the exact number of
2037 * bytes remaining to copy on a fault.  But copy_mount_options() requires that.
2038 * Note that this function differs from copy_from_user() in that it will oops
2039 * on bad values of `to', rather than returning a short copy.
2040 */
2041static long exact_copy_from_user(void *to, const void __user * from,
2042                                 unsigned long n)
2043{
2044        char *t = to;
2045        const char __user *f = from;
2046        char c;
2047
2048        if (!access_ok(VERIFY_READ, from, n))
2049                return n;
2050
2051        while (n) {
2052                if (__get_user(c, f)) {
2053                        memset(t, 0, n);
2054                        break;
2055                }
2056                *t++ = c;
2057                f++;
2058                n--;
2059        }
2060        return n;
2061}
2062
2063int copy_mount_options(const void __user * data, unsigned long *where)
2064{
2065        int i;
2066        unsigned long page;
2067        unsigned long size;
2068
2069        *where = 0;
2070        if (!data)
2071                return 0;
2072
2073        if (!(page = __get_free_page(GFP_KERNEL)))
2074                return -ENOMEM;
2075
2076        /* We only care that *some* data at the address the user
2077         * gave us is valid.  Just in case, we'll zero
2078         * the remainder of the page.
2079         */
2080        /* copy_from_user cannot cross TASK_SIZE ! */
2081        size = TASK_SIZE - (unsigned long)data;
2082        if (size > PAGE_SIZE)
2083                size = PAGE_SIZE;
2084
2085        i = size - exact_copy_from_user((void *)page, data, size);
2086        if (!i) {
2087                free_page(page);
2088                return -EFAULT;
2089        }
2090        if (i != PAGE_SIZE)
2091                memset((char *)page + i, 0, PAGE_SIZE - i);
2092        *where = page;
2093        return 0;
2094}
2095
2096int copy_mount_string(const void __user *data, char **where)
2097{
2098        char *tmp;
2099
2100        if (!data) {
2101                *where = NULL;
2102                return 0;
2103        }
2104
2105        tmp = strndup_user(data, PAGE_SIZE);
2106        if (IS_ERR(tmp))
2107                return PTR_ERR(tmp);
2108
2109        *where = tmp;
2110        return 0;
2111}
2112
2113/*
2114 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2115 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2116 *
2117 * data is a (void *) that can point to any structure up to
2118 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2119 * information (or be NULL).
2120 *
2121 * Pre-0.97 versions of mount() didn't have a flags word.
2122 * When the flags word was introduced its top half was required
2123 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2124 * Therefore, if this magic number is present, it carries no information
2125 * and must be discarded.
2126 */
2127long do_mount(char *dev_name, char *dir_name, char *type_page,
2128                  unsigned long flags, void *data_page)
2129{
2130        struct path path;
2131        int retval = 0;
2132        int mnt_flags = 0;
2133
2134        /* Discard magic */
2135        if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2136                flags &= ~MS_MGC_MSK;
2137
2138        /* Basic sanity checks */
2139
2140        if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2141                return -EINVAL;
2142
2143        if (data_page)
2144                ((char *)data_page)[PAGE_SIZE - 1] = 0;
2145
2146        /* ... and get the mountpoint */
2147        retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2148        if (retval)
2149                return retval;
2150
2151        retval = security_sb_mount(dev_name, &path,
2152                                   type_page, flags, data_page);
2153        if (retval)
2154                goto dput_out;
2155
2156        /* Default to relatime unless overriden */
2157        if (!(flags & MS_NOATIME))
2158                mnt_flags |= MNT_RELATIME;
2159
2160        /* Separate the per-mountpoint flags */
2161        if (flags & MS_NOSUID)
2162                mnt_flags |= MNT_NOSUID;
2163        if (flags & MS_NODEV)
2164                mnt_flags |= MNT_NODEV;
2165        if (flags & MS_NOEXEC)
2166                mnt_flags |= MNT_NOEXEC;
2167        if (flags & MS_NOATIME)
2168                mnt_flags |= MNT_NOATIME;
2169        if (flags & MS_NODIRATIME)
2170                mnt_flags |= MNT_NODIRATIME;
2171        if (flags & MS_STRICTATIME)
2172                mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2173        if (flags & MS_RDONLY)
2174                mnt_flags |= MNT_READONLY;
2175
2176        flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2177                   MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2178                   MS_STRICTATIME);
2179
2180        if (flags & MS_REMOUNT)
2181                retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2182                                    data_page);
2183        else if (flags & MS_BIND)
2184                retval = do_loopback(&path, dev_name, flags & MS_REC);
2185        else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2186                retval = do_change_type(&path, flags);
2187        else if (flags & MS_MOVE)
2188                retval = do_move_mount(&path, dev_name);
2189        else
2190                retval = do_new_mount(&path, type_page, flags, mnt_flags,
2191                                      dev_name, data_page);
2192dput_out:
2193        path_put(&path);
2194        return retval;
2195}
2196
2197static struct mnt_namespace *alloc_mnt_ns(void)
2198{
2199        struct mnt_namespace *new_ns;
2200
2201        new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2202        if (!new_ns)
2203                return ERR_PTR(-ENOMEM);
2204        atomic_set(&new_ns->count, 1);
2205        new_ns->root = NULL;
2206        INIT_LIST_HEAD(&new_ns->list);
2207        init_waitqueue_head(&new_ns->poll);
2208        new_ns->event = 0;
2209        return new_ns;
2210}
2211
2212void mnt_make_longterm(struct vfsmount *mnt)
2213{
2214        __mnt_make_longterm(real_mount(mnt));
2215}
2216
2217void mnt_make_shortterm(struct vfsmount *m)
2218{
2219#ifdef CONFIG_SMP
2220        struct mount *mnt = real_mount(m);
2221        if (atomic_add_unless(&mnt->mnt_longterm, -1, 1))
2222                return;
2223        br_write_lock(&vfsmount_lock);
2224        atomic_dec(&mnt->mnt_longterm);
2225        br_write_unlock(&vfsmount_lock);
2226#endif
2227}
2228
2229/*
2230 * Allocate a new namespace structure and populate it with contents
2231 * copied from the namespace of the passed in task structure.
2232 */
2233static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2234                struct fs_struct *fs)
2235{
2236        struct mnt_namespace *new_ns;
2237        struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2238        struct mount *p, *q;
2239        struct mount *old = mnt_ns->root;
2240        struct mount *new;
2241
2242        new_ns = alloc_mnt_ns();
2243        if (IS_ERR(new_ns))
2244                return new_ns;
2245
2246        down_write(&namespace_sem);
2247        /* First pass: copy the tree topology */
2248        new = copy_tree(old, old->mnt.mnt_root, CL_COPY_ALL | CL_EXPIRE);
2249        if (!new) {
2250                up_write(&namespace_sem);
2251                kfree(new_ns);
2252                return ERR_PTR(-ENOMEM);
2253        }
2254        new_ns->root = new;
2255        br_write_lock(&vfsmount_lock);
2256        list_add_tail(&new_ns->list, &new->mnt_list);
2257        br_write_unlock(&vfsmount_lock);
2258
2259        /*
2260         * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2261         * as belonging to new namespace.  We have already acquired a private
2262         * fs_struct, so tsk->fs->lock is not needed.
2263         */
2264        p = old;
2265        q = new;
2266        while (p) {
2267                q->mnt_ns = new_ns;
2268                __mnt_make_longterm(q);
2269                if (fs) {
2270                        if (&p->mnt == fs->root.mnt) {
2271                                fs->root.mnt = mntget(&q->mnt);
2272                                __mnt_make_longterm(q);
2273                                mnt_make_shortterm(&p->mnt);
2274                                rootmnt = &p->mnt;
2275                        }
2276                        if (&p->mnt == fs->pwd.mnt) {
2277                                fs->pwd.mnt = mntget(&q->mnt);
2278                                __mnt_make_longterm(q);
2279                                mnt_make_shortterm(&p->mnt);
2280                                pwdmnt = &p->mnt;
2281                        }
2282                }
2283                p = next_mnt(p, old);
2284                q = next_mnt(q, new);
2285        }
2286        up_write(&namespace_sem);
2287
2288        if (rootmnt)
2289                mntput(rootmnt);
2290        if (pwdmnt)
2291                mntput(pwdmnt);
2292
2293        return new_ns;
2294}
2295
2296struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2297                struct fs_struct *new_fs)
2298{
2299        struct mnt_namespace *new_ns;
2300
2301        BUG_ON(!ns);
2302        get_mnt_ns(ns);
2303
2304        if (!(flags & CLONE_NEWNS))
2305                return ns;
2306
2307        new_ns = dup_mnt_ns(ns, new_fs);
2308
2309        put_mnt_ns(ns);
2310        return new_ns;
2311}
2312
2313/**
2314 * create_mnt_ns - creates a private namespace and adds a root filesystem
2315 * @mnt: pointer to the new root filesystem mountpoint
2316 */
2317static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2318{
2319        struct mnt_namespace *new_ns = alloc_mnt_ns();
2320        if (!IS_ERR(new_ns)) {
2321                struct mount *mnt = real_mount(m);
2322                mnt->mnt_ns = new_ns;
2323                __mnt_make_longterm(mnt);
2324                new_ns->root = mnt;
2325                list_add(&new_ns->list, &mnt->mnt_list);
2326        } else {
2327                mntput(m);
2328        }
2329        return new_ns;
2330}
2331
2332struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2333{
2334        struct mnt_namespace *ns;
2335        struct super_block *s;
2336        struct path path;
2337        int err;
2338
2339        ns = create_mnt_ns(mnt);
2340        if (IS_ERR(ns))
2341                return ERR_CAST(ns);
2342
2343        err = vfs_path_lookup(mnt->mnt_root, mnt,
2344                        name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2345
2346        put_mnt_ns(ns);
2347
2348        if (err)
2349                return ERR_PTR(err);
2350
2351        /* trade a vfsmount reference for active sb one */
2352        s = path.mnt->mnt_sb;
2353        atomic_inc(&s->s_active);
2354        mntput(path.mnt);
2355        /* lock the sucker */
2356        down_write(&s->s_umount);
2357        /* ... and return the root of (sub)tree on it */
2358        return path.dentry;
2359}
2360EXPORT_SYMBOL(mount_subtree);
2361
2362SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2363                char __user *, type, unsigned long, flags, void __user *, data)
2364{
2365        int ret;
2366        char *kernel_type;
2367        char *kernel_dir;
2368        char *kernel_dev;
2369        unsigned long data_page;
2370
2371        ret = copy_mount_string(type, &kernel_type);
2372        if (ret < 0)
2373                goto out_type;
2374
2375        kernel_dir = getname(dir_name);
2376        if (IS_ERR(kernel_dir)) {
2377                ret = PTR_ERR(kernel_dir);
2378                goto out_dir;
2379        }
2380
2381        ret = copy_mount_string(dev_name, &kernel_dev);
2382        if (ret < 0)
2383                goto out_dev;
2384
2385        ret = copy_mount_options(data, &data_page);
2386        if (ret < 0)
2387                goto out_data;
2388
2389        ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2390                (void *) data_page);
2391
2392        free_page(data_page);
2393out_data:
2394        kfree(kernel_dev);
2395out_dev:
2396        putname(kernel_dir);
2397out_dir:
2398        kfree(kernel_type);
2399out_type:
2400        return ret;
2401}
2402
2403/*
2404 * Return true if path is reachable from root
2405 *
2406 * namespace_sem or vfsmount_lock is held
2407 */
2408bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2409                         const struct path *root)
2410{
2411        while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2412                dentry = mnt->mnt_mountpoint;
2413                mnt = mnt->mnt_parent;
2414        }
2415        return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2416}
2417
2418int path_is_under(struct path *path1, struct path *path2)
2419{
2420        int res;
2421        br_read_lock(&vfsmount_lock);
2422        res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2423        br_read_unlock(&vfsmount_lock);
2424        return res;
2425}
2426EXPORT_SYMBOL(path_is_under);
2427
2428/*
2429 * pivot_root Semantics:
2430 * Moves the root file system of the current process to the directory put_old,
2431 * makes new_root as the new root file system of the current process, and sets
2432 * root/cwd of all processes which had them on the current root to new_root.
2433 *
2434 * Restrictions:
2435 * The new_root and put_old must be directories, and  must not be on the
2436 * same file  system as the current process root. The put_old  must  be
2437 * underneath new_root,  i.e. adding a non-zero number of /.. to the string
2438 * pointed to by put_old must yield the same directory as new_root. No other
2439 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2440 *
2441 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2442 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2443 * in this situation.
2444 *
2445 * Notes:
2446 *  - we don't move root/cwd if they are not at the root (reason: if something
2447 *    cared enough to change them, it's probably wrong to force them elsewhere)
2448 *  - it's okay to pick a root that isn't the root of a file system, e.g.
2449 *    /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2450 *    though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2451 *    first.
2452 */
2453SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2454                const char __user *, put_old)
2455{
2456        struct path new, old, parent_path, root_parent, root;
2457        struct mount *new_mnt, *root_mnt;
2458        int error;
2459
2460        if (!capable(CAP_SYS_ADMIN))
2461                return -EPERM;
2462
2463        error = user_path_dir(new_root, &new);
2464        if (error)
2465                goto out0;
2466
2467        error = user_path_dir(put_old, &old);
2468        if (error)
2469                goto out1;
2470
2471        error = security_sb_pivotroot(&old, &new);
2472        if (error)
2473                goto out2;
2474
2475        get_fs_root(current->fs, &root);
2476        error = lock_mount(&old);
2477        if (error)
2478                goto out3;
2479
2480        error = -EINVAL;
2481        new_mnt = real_mount(new.mnt);
2482        root_mnt = real_mount(root.mnt);
2483        if (IS_MNT_SHARED(real_mount(old.mnt)) ||
2484                IS_MNT_SHARED(new_mnt->mnt_parent) ||
2485                IS_MNT_SHARED(root_mnt->mnt_parent))
2486                goto out4;
2487        if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
2488                goto out4;
2489        error = -ENOENT;
2490        if (d_unlinked(new.dentry))
2491                goto out4;
2492        if (d_unlinked(old.dentry))
2493                goto out4;
2494        error = -EBUSY;
2495        if (new.mnt == root.mnt ||
2496            old.mnt == root.mnt)
2497                goto out4; /* loop, on the same file system  */
2498        error = -EINVAL;
2499        if (root.mnt->mnt_root != root.dentry)
2500                goto out4; /* not a mountpoint */
2501        if (!mnt_has_parent(root_mnt))
2502                goto out4; /* not attached */
2503        if (new.mnt->mnt_root != new.dentry)
2504                goto out4; /* not a mountpoint */
2505        if (!mnt_has_parent(new_mnt))
2506                goto out4; /* not attached */
2507        /* make sure we can reach put_old from new_root */
2508        if (!is_path_reachable(real_mount(old.mnt), old.dentry, &new))
2509                goto out4;
2510        br_write_lock(&vfsmount_lock);
2511        detach_mnt(new_mnt, &parent_path);
2512        detach_mnt(root_mnt, &root_parent);
2513        /* mount old root on put_old */
2514        attach_mnt(root_mnt, &old);
2515        /* mount new_root on / */
2516        attach_mnt(new_mnt, &root_parent);
2517        touch_mnt_namespace(current->nsproxy->mnt_ns);
2518        br_write_unlock(&vfsmount_lock);
2519        chroot_fs_refs(&root, &new);
2520        error = 0;
2521out4:
2522        unlock_mount(&old);
2523        if (!error) {
2524                path_put(&root_parent);
2525                path_put(&parent_path);
2526        }
2527out3:
2528        path_put(&root);
2529out2:
2530        path_put(&old);
2531out1:
2532        path_put(&new);
2533out0:
2534        return error;
2535}
2536
2537static void __init init_mount_tree(void)
2538{
2539        struct vfsmount *mnt;
2540        struct mnt_namespace *ns;
2541        struct path root;
2542
2543        mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2544        if (IS_ERR(mnt))
2545                panic("Can't create rootfs");
2546
2547        ns = create_mnt_ns(mnt);
2548        if (IS_ERR(ns))
2549                panic("Can't allocate initial namespace");
2550
2551        init_task.nsproxy->mnt_ns = ns;
2552        get_mnt_ns(ns);
2553
2554        root.mnt = mnt;
2555        root.dentry = mnt->mnt_root;
2556
2557        set_fs_pwd(current->fs, &root);
2558        set_fs_root(current->fs, &root);
2559}
2560
2561void __init mnt_init(void)
2562{
2563        unsigned u;
2564        int err;
2565
2566        init_rwsem(&namespace_sem);
2567
2568        mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
2569                        0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2570
2571        mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2572
2573        if (!mount_hashtable)
2574                panic("Failed to allocate mount hash table\n");
2575
2576        printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
2577
2578        for (u = 0; u < HASH_SIZE; u++)
2579                INIT_LIST_HEAD(&mount_hashtable[u]);
2580
2581        br_lock_init(&vfsmount_lock);
2582
2583        err = sysfs_init();
2584        if (err)
2585                printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2586                        __func__, err);
2587        fs_kobj = kobject_create_and_add("fs", NULL);
2588        if (!fs_kobj)
2589                printk(KERN_WARNING "%s: kobj create error\n", __func__);
2590        init_rootfs();
2591        init_mount_tree();
2592}
2593
2594void put_mnt_ns(struct mnt_namespace *ns)
2595{
2596        LIST_HEAD(umount_list);
2597
2598        if (!atomic_dec_and_test(&ns->count))
2599                return;
2600        down_write(&namespace_sem);
2601        br_write_lock(&vfsmount_lock);
2602        umount_tree(ns->root, 0, &umount_list);
2603        br_write_unlock(&vfsmount_lock);
2604        up_write(&namespace_sem);
2605        release_mounts(&umount_list);
2606        kfree(ns);
2607}
2608
2609struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2610{
2611        struct vfsmount *mnt;
2612        mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2613        if (!IS_ERR(mnt)) {
2614                /*
2615                 * it is a longterm mount, don't release mnt until
2616                 * we unmount before file sys is unregistered
2617                */
2618                mnt_make_longterm(mnt);
2619        }
2620        return mnt;
2621}
2622EXPORT_SYMBOL_GPL(kern_mount_data);
2623
2624void kern_unmount(struct vfsmount *mnt)
2625{
2626        /* release long term mount so mount point can be released */
2627        if (!IS_ERR_OR_NULL(mnt)) {
2628                mnt_make_shortterm(mnt);
2629                mntput(mnt);
2630        }
2631}
2632EXPORT_SYMBOL(kern_unmount);
2633
2634bool our_mnt(struct vfsmount *mnt)
2635{
2636        return check_mnt(real_mount(mnt));
2637}
2638
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.