[PATCH] firmware/efivars: handle error
[linux-2.6.22.y-op.git] / fs / namespace.c
blob55442a6cf22133d2c9687636b3e3ca788eacac53
1 /*
2 * linux/fs/namespace.c
4 * (C) Copyright Al Viro 2000, 2001
5 * Released under GPL v2.
7 * Based on code from fs/super.c, copyright Linus Torvalds and others.
8 * Heavily rewritten.
9 */
11 #include <linux/syscalls.h>
12 #include <linux/slab.h>
13 #include <linux/sched.h>
14 #include <linux/smp_lock.h>
15 #include <linux/init.h>
16 #include <linux/kernel.h>
17 #include <linux/quotaops.h>
18 #include <linux/acct.h>
19 #include <linux/capability.h>
20 #include <linux/module.h>
21 #include <linux/sysfs.h>
22 #include <linux/seq_file.h>
23 #include <linux/namespace.h>
24 #include <linux/namei.h>
25 #include <linux/security.h>
26 #include <linux/mount.h>
27 #include <linux/ramfs.h>
28 #include <asm/uaccess.h>
29 #include <asm/unistd.h>
30 #include "pnode.h"
32 /* spinlock for vfsmount related operations, inplace of dcache_lock */
33 __cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
35 static int event;
37 static struct list_head *mount_hashtable __read_mostly;
38 static int hash_mask __read_mostly, hash_bits __read_mostly;
39 static kmem_cache_t *mnt_cache __read_mostly;
40 static struct rw_semaphore namespace_sem;
42 /* /sys/fs */
43 decl_subsys(fs, NULL, NULL);
44 EXPORT_SYMBOL_GPL(fs_subsys);
46 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
48 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
49 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
50 tmp = tmp + (tmp >> hash_bits);
51 return tmp & hash_mask;
54 struct vfsmount *alloc_vfsmnt(const char *name)
56 struct vfsmount *mnt = kmem_cache_alloc(mnt_cache, GFP_KERNEL);
57 if (mnt) {
58 memset(mnt, 0, sizeof(struct vfsmount));
59 atomic_set(&mnt->mnt_count, 1);
60 INIT_LIST_HEAD(&mnt->mnt_hash);
61 INIT_LIST_HEAD(&mnt->mnt_child);
62 INIT_LIST_HEAD(&mnt->mnt_mounts);
63 INIT_LIST_HEAD(&mnt->mnt_list);
64 INIT_LIST_HEAD(&mnt->mnt_expire);
65 INIT_LIST_HEAD(&mnt->mnt_share);
66 INIT_LIST_HEAD(&mnt->mnt_slave_list);
67 INIT_LIST_HEAD(&mnt->mnt_slave);
68 if (name) {
69 int size = strlen(name) + 1;
70 char *newname = kmalloc(size, GFP_KERNEL);
71 if (newname) {
72 memcpy(newname, name, size);
73 mnt->mnt_devname = newname;
77 return mnt;
80 int simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
82 mnt->mnt_sb = sb;
83 mnt->mnt_root = dget(sb->s_root);
84 return 0;
87 EXPORT_SYMBOL(simple_set_mnt);
89 void free_vfsmnt(struct vfsmount *mnt)
91 kfree(mnt->mnt_devname);
92 kmem_cache_free(mnt_cache, mnt);
96 * find the first or last mount at @dentry on vfsmount @mnt depending on
97 * @dir. If @dir is set return the first mount else return the last mount.
99 struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
100 int dir)
102 struct list_head *head = mount_hashtable + hash(mnt, dentry);
103 struct list_head *tmp = head;
104 struct vfsmount *p, *found = NULL;
106 for (;;) {
107 tmp = dir ? tmp->next : tmp->prev;
108 p = NULL;
109 if (tmp == head)
110 break;
111 p = list_entry(tmp, struct vfsmount, mnt_hash);
112 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
113 found = p;
114 break;
117 return found;
121 * lookup_mnt increments the ref count before returning
122 * the vfsmount struct.
124 struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
126 struct vfsmount *child_mnt;
127 spin_lock(&vfsmount_lock);
128 if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
129 mntget(child_mnt);
130 spin_unlock(&vfsmount_lock);
131 return child_mnt;
134 static inline int check_mnt(struct vfsmount *mnt)
136 return mnt->mnt_namespace == current->nsproxy->namespace;
139 static void touch_namespace(struct namespace *ns)
141 if (ns) {
142 ns->event = ++event;
143 wake_up_interruptible(&ns->poll);
147 static void __touch_namespace(struct namespace *ns)
149 if (ns && ns->event != event) {
150 ns->event = event;
151 wake_up_interruptible(&ns->poll);
155 static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
157 old_nd->dentry = mnt->mnt_mountpoint;
158 old_nd->mnt = mnt->mnt_parent;
159 mnt->mnt_parent = mnt;
160 mnt->mnt_mountpoint = mnt->mnt_root;
161 list_del_init(&mnt->mnt_child);
162 list_del_init(&mnt->mnt_hash);
163 old_nd->dentry->d_mounted--;
166 void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
167 struct vfsmount *child_mnt)
169 child_mnt->mnt_parent = mntget(mnt);
170 child_mnt->mnt_mountpoint = dget(dentry);
171 dentry->d_mounted++;
174 static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
176 mnt_set_mountpoint(nd->mnt, nd->dentry, mnt);
177 list_add_tail(&mnt->mnt_hash, mount_hashtable +
178 hash(nd->mnt, nd->dentry));
179 list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts);
183 * the caller must hold vfsmount_lock
185 static void commit_tree(struct vfsmount *mnt)
187 struct vfsmount *parent = mnt->mnt_parent;
188 struct vfsmount *m;
189 LIST_HEAD(head);
190 struct namespace *n = parent->mnt_namespace;
192 BUG_ON(parent == mnt);
194 list_add_tail(&head, &mnt->mnt_list);
195 list_for_each_entry(m, &head, mnt_list)
196 m->mnt_namespace = n;
197 list_splice(&head, n->list.prev);
199 list_add_tail(&mnt->mnt_hash, mount_hashtable +
200 hash(parent, mnt->mnt_mountpoint));
201 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
202 touch_namespace(n);
205 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
207 struct list_head *next = p->mnt_mounts.next;
208 if (next == &p->mnt_mounts) {
209 while (1) {
210 if (p == root)
211 return NULL;
212 next = p->mnt_child.next;
213 if (next != &p->mnt_parent->mnt_mounts)
214 break;
215 p = p->mnt_parent;
218 return list_entry(next, struct vfsmount, mnt_child);
221 static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
223 struct list_head *prev = p->mnt_mounts.prev;
224 while (prev != &p->mnt_mounts) {
225 p = list_entry(prev, struct vfsmount, mnt_child);
226 prev = p->mnt_mounts.prev;
228 return p;
231 static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
232 int flag)
234 struct super_block *sb = old->mnt_sb;
235 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
237 if (mnt) {
238 mnt->mnt_flags = old->mnt_flags;
239 atomic_inc(&sb->s_active);
240 mnt->mnt_sb = sb;
241 mnt->mnt_root = dget(root);
242 mnt->mnt_mountpoint = mnt->mnt_root;
243 mnt->mnt_parent = mnt;
245 if (flag & CL_SLAVE) {
246 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
247 mnt->mnt_master = old;
248 CLEAR_MNT_SHARED(mnt);
249 } else {
250 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
251 list_add(&mnt->mnt_share, &old->mnt_share);
252 if (IS_MNT_SLAVE(old))
253 list_add(&mnt->mnt_slave, &old->mnt_slave);
254 mnt->mnt_master = old->mnt_master;
256 if (flag & CL_MAKE_SHARED)
257 set_mnt_shared(mnt);
259 /* stick the duplicate mount on the same expiry list
260 * as the original if that was on one */
261 if (flag & CL_EXPIRE) {
262 spin_lock(&vfsmount_lock);
263 if (!list_empty(&old->mnt_expire))
264 list_add(&mnt->mnt_expire, &old->mnt_expire);
265 spin_unlock(&vfsmount_lock);
268 return mnt;
271 static inline void __mntput(struct vfsmount *mnt)
273 struct super_block *sb = mnt->mnt_sb;
274 dput(mnt->mnt_root);
275 free_vfsmnt(mnt);
276 deactivate_super(sb);
279 void mntput_no_expire(struct vfsmount *mnt)
281 repeat:
282 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
283 if (likely(!mnt->mnt_pinned)) {
284 spin_unlock(&vfsmount_lock);
285 __mntput(mnt);
286 return;
288 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
289 mnt->mnt_pinned = 0;
290 spin_unlock(&vfsmount_lock);
291 acct_auto_close_mnt(mnt);
292 security_sb_umount_close(mnt);
293 goto repeat;
297 EXPORT_SYMBOL(mntput_no_expire);
299 void mnt_pin(struct vfsmount *mnt)
301 spin_lock(&vfsmount_lock);
302 mnt->mnt_pinned++;
303 spin_unlock(&vfsmount_lock);
306 EXPORT_SYMBOL(mnt_pin);
308 void mnt_unpin(struct vfsmount *mnt)
310 spin_lock(&vfsmount_lock);
311 if (mnt->mnt_pinned) {
312 atomic_inc(&mnt->mnt_count);
313 mnt->mnt_pinned--;
315 spin_unlock(&vfsmount_lock);
318 EXPORT_SYMBOL(mnt_unpin);
320 /* iterator */
321 static void *m_start(struct seq_file *m, loff_t *pos)
323 struct namespace *n = m->private;
324 struct list_head *p;
325 loff_t l = *pos;
327 down_read(&namespace_sem);
328 list_for_each(p, &n->list)
329 if (!l--)
330 return list_entry(p, struct vfsmount, mnt_list);
331 return NULL;
334 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
336 struct namespace *n = m->private;
337 struct list_head *p = ((struct vfsmount *)v)->mnt_list.next;
338 (*pos)++;
339 return p == &n->list ? NULL : list_entry(p, struct vfsmount, mnt_list);
342 static void m_stop(struct seq_file *m, void *v)
344 up_read(&namespace_sem);
347 static inline void mangle(struct seq_file *m, const char *s)
349 seq_escape(m, s, " \t\n\\");
352 static int show_vfsmnt(struct seq_file *m, void *v)
354 struct vfsmount *mnt = v;
355 int err = 0;
356 static struct proc_fs_info {
357 int flag;
358 char *str;
359 } fs_info[] = {
360 { MS_SYNCHRONOUS, ",sync" },
361 { MS_DIRSYNC, ",dirsync" },
362 { MS_MANDLOCK, ",mand" },
363 { 0, NULL }
365 static struct proc_fs_info mnt_info[] = {
366 { MNT_NOSUID, ",nosuid" },
367 { MNT_NODEV, ",nodev" },
368 { MNT_NOEXEC, ",noexec" },
369 { MNT_NOATIME, ",noatime" },
370 { MNT_NODIRATIME, ",nodiratime" },
371 { 0, NULL }
373 struct proc_fs_info *fs_infop;
375 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
376 seq_putc(m, ' ');
377 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
378 seq_putc(m, ' ');
379 mangle(m, mnt->mnt_sb->s_type->name);
380 seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw");
381 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
382 if (mnt->mnt_sb->s_flags & fs_infop->flag)
383 seq_puts(m, fs_infop->str);
385 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
386 if (mnt->mnt_flags & fs_infop->flag)
387 seq_puts(m, fs_infop->str);
389 if (mnt->mnt_sb->s_op->show_options)
390 err = mnt->mnt_sb->s_op->show_options(m, mnt);
391 seq_puts(m, " 0 0\n");
392 return err;
395 struct seq_operations mounts_op = {
396 .start = m_start,
397 .next = m_next,
398 .stop = m_stop,
399 .show = show_vfsmnt
402 static int show_vfsstat(struct seq_file *m, void *v)
404 struct vfsmount *mnt = v;
405 int err = 0;
407 /* device */
408 if (mnt->mnt_devname) {
409 seq_puts(m, "device ");
410 mangle(m, mnt->mnt_devname);
411 } else
412 seq_puts(m, "no device");
414 /* mount point */
415 seq_puts(m, " mounted on ");
416 seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
417 seq_putc(m, ' ');
419 /* file system type */
420 seq_puts(m, "with fstype ");
421 mangle(m, mnt->mnt_sb->s_type->name);
423 /* optional statistics */
424 if (mnt->mnt_sb->s_op->show_stats) {
425 seq_putc(m, ' ');
426 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
429 seq_putc(m, '\n');
430 return err;
433 struct seq_operations mountstats_op = {
434 .start = m_start,
435 .next = m_next,
436 .stop = m_stop,
437 .show = show_vfsstat,
441 * may_umount_tree - check if a mount tree is busy
442 * @mnt: root of mount tree
444 * This is called to check if a tree of mounts has any
445 * open files, pwds, chroots or sub mounts that are
446 * busy.
448 int may_umount_tree(struct vfsmount *mnt)
450 int actual_refs = 0;
451 int minimum_refs = 0;
452 struct vfsmount *p;
454 spin_lock(&vfsmount_lock);
455 for (p = mnt; p; p = next_mnt(p, mnt)) {
456 actual_refs += atomic_read(&p->mnt_count);
457 minimum_refs += 2;
459 spin_unlock(&vfsmount_lock);
461 if (actual_refs > minimum_refs)
462 return 0;
464 return 1;
467 EXPORT_SYMBOL(may_umount_tree);
470 * may_umount - check if a mount point is busy
471 * @mnt: root of mount
473 * This is called to check if a mount point has any
474 * open files, pwds, chroots or sub mounts. If the
475 * mount has sub mounts this will return busy
476 * regardless of whether the sub mounts are busy.
478 * Doesn't take quota and stuff into account. IOW, in some cases it will
479 * give false negatives. The main reason why it's here is that we need
480 * a non-destructive way to look for easily umountable filesystems.
482 int may_umount(struct vfsmount *mnt)
484 int ret = 1;
485 spin_lock(&vfsmount_lock);
486 if (propagate_mount_busy(mnt, 2))
487 ret = 0;
488 spin_unlock(&vfsmount_lock);
489 return ret;
492 EXPORT_SYMBOL(may_umount);
494 void release_mounts(struct list_head *head)
496 struct vfsmount *mnt;
497 while (!list_empty(head)) {
498 mnt = list_entry(head->next, struct vfsmount, mnt_hash);
499 list_del_init(&mnt->mnt_hash);
500 if (mnt->mnt_parent != mnt) {
501 struct dentry *dentry;
502 struct vfsmount *m;
503 spin_lock(&vfsmount_lock);
504 dentry = mnt->mnt_mountpoint;
505 m = mnt->mnt_parent;
506 mnt->mnt_mountpoint = mnt->mnt_root;
507 mnt->mnt_parent = mnt;
508 spin_unlock(&vfsmount_lock);
509 dput(dentry);
510 mntput(m);
512 mntput(mnt);
516 void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
518 struct vfsmount *p;
520 for (p = mnt; p; p = next_mnt(p, mnt))
521 list_move(&p->mnt_hash, kill);
523 if (propagate)
524 propagate_umount(kill);
526 list_for_each_entry(p, kill, mnt_hash) {
527 list_del_init(&p->mnt_expire);
528 list_del_init(&p->mnt_list);
529 __touch_namespace(p->mnt_namespace);
530 p->mnt_namespace = NULL;
531 list_del_init(&p->mnt_child);
532 if (p->mnt_parent != p)
533 p->mnt_mountpoint->d_mounted--;
534 change_mnt_propagation(p, MS_PRIVATE);
538 static int do_umount(struct vfsmount *mnt, int flags)
540 struct super_block *sb = mnt->mnt_sb;
541 int retval;
542 LIST_HEAD(umount_list);
544 retval = security_sb_umount(mnt, flags);
545 if (retval)
546 return retval;
549 * Allow userspace to request a mountpoint be expired rather than
550 * unmounting unconditionally. Unmount only happens if:
551 * (1) the mark is already set (the mark is cleared by mntput())
552 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
554 if (flags & MNT_EXPIRE) {
555 if (mnt == current->fs->rootmnt ||
556 flags & (MNT_FORCE | MNT_DETACH))
557 return -EINVAL;
559 if (atomic_read(&mnt->mnt_count) != 2)
560 return -EBUSY;
562 if (!xchg(&mnt->mnt_expiry_mark, 1))
563 return -EAGAIN;
567 * If we may have to abort operations to get out of this
568 * mount, and they will themselves hold resources we must
569 * allow the fs to do things. In the Unix tradition of
570 * 'Gee thats tricky lets do it in userspace' the umount_begin
571 * might fail to complete on the first run through as other tasks
572 * must return, and the like. Thats for the mount program to worry
573 * about for the moment.
576 lock_kernel();
577 if (sb->s_op->umount_begin)
578 sb->s_op->umount_begin(mnt, flags);
579 unlock_kernel();
582 * No sense to grab the lock for this test, but test itself looks
583 * somewhat bogus. Suggestions for better replacement?
584 * Ho-hum... In principle, we might treat that as umount + switch
585 * to rootfs. GC would eventually take care of the old vfsmount.
586 * Actually it makes sense, especially if rootfs would contain a
587 * /reboot - static binary that would close all descriptors and
588 * call reboot(9). Then init(8) could umount root and exec /reboot.
590 if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
592 * Special case for "unmounting" root ...
593 * we just try to remount it readonly.
595 down_write(&sb->s_umount);
596 if (!(sb->s_flags & MS_RDONLY)) {
597 lock_kernel();
598 DQUOT_OFF(sb);
599 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
600 unlock_kernel();
602 up_write(&sb->s_umount);
603 return retval;
606 down_write(&namespace_sem);
607 spin_lock(&vfsmount_lock);
608 event++;
610 retval = -EBUSY;
611 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
612 if (!list_empty(&mnt->mnt_list))
613 umount_tree(mnt, 1, &umount_list);
614 retval = 0;
616 spin_unlock(&vfsmount_lock);
617 if (retval)
618 security_sb_umount_busy(mnt);
619 up_write(&namespace_sem);
620 release_mounts(&umount_list);
621 return retval;
625 * Now umount can handle mount points as well as block devices.
626 * This is important for filesystems which use unnamed block devices.
628 * We now support a flag for forced unmount like the other 'big iron'
629 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
632 asmlinkage long sys_umount(char __user * name, int flags)
634 struct nameidata nd;
635 int retval;
637 retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
638 if (retval)
639 goto out;
640 retval = -EINVAL;
641 if (nd.dentry != nd.mnt->mnt_root)
642 goto dput_and_out;
643 if (!check_mnt(nd.mnt))
644 goto dput_and_out;
646 retval = -EPERM;
647 if (!capable(CAP_SYS_ADMIN))
648 goto dput_and_out;
650 retval = do_umount(nd.mnt, flags);
651 dput_and_out:
652 path_release_on_umount(&nd);
653 out:
654 return retval;
657 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
660 * The 2.0 compatible umount. No flags.
662 asmlinkage long sys_oldumount(char __user * name)
664 return sys_umount(name, 0);
667 #endif
669 static int mount_is_safe(struct nameidata *nd)
671 if (capable(CAP_SYS_ADMIN))
672 return 0;
673 return -EPERM;
674 #ifdef notyet
675 if (S_ISLNK(nd->dentry->d_inode->i_mode))
676 return -EPERM;
677 if (nd->dentry->d_inode->i_mode & S_ISVTX) {
678 if (current->uid != nd->dentry->d_inode->i_uid)
679 return -EPERM;
681 if (vfs_permission(nd, MAY_WRITE))
682 return -EPERM;
683 return 0;
684 #endif
687 static int lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
689 while (1) {
690 if (d == dentry)
691 return 1;
692 if (d == NULL || d == d->d_parent)
693 return 0;
694 d = d->d_parent;
698 struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
699 int flag)
701 struct vfsmount *res, *p, *q, *r, *s;
702 struct nameidata nd;
704 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
705 return NULL;
707 res = q = clone_mnt(mnt, dentry, flag);
708 if (!q)
709 goto Enomem;
710 q->mnt_mountpoint = mnt->mnt_mountpoint;
712 p = mnt;
713 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
714 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
715 continue;
717 for (s = r; s; s = next_mnt(s, r)) {
718 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
719 s = skip_mnt_tree(s);
720 continue;
722 while (p != s->mnt_parent) {
723 p = p->mnt_parent;
724 q = q->mnt_parent;
726 p = s;
727 nd.mnt = q;
728 nd.dentry = p->mnt_mountpoint;
729 q = clone_mnt(p, p->mnt_root, flag);
730 if (!q)
731 goto Enomem;
732 spin_lock(&vfsmount_lock);
733 list_add_tail(&q->mnt_list, &res->mnt_list);
734 attach_mnt(q, &nd);
735 spin_unlock(&vfsmount_lock);
738 return res;
739 Enomem:
740 if (res) {
741 LIST_HEAD(umount_list);
742 spin_lock(&vfsmount_lock);
743 umount_tree(res, 0, &umount_list);
744 spin_unlock(&vfsmount_lock);
745 release_mounts(&umount_list);
747 return NULL;
751 * @source_mnt : mount tree to be attached
752 * @nd : place the mount tree @source_mnt is attached
753 * @parent_nd : if non-null, detach the source_mnt from its parent and
754 * store the parent mount and mountpoint dentry.
755 * (done when source_mnt is moved)
757 * NOTE: in the table below explains the semantics when a source mount
758 * of a given type is attached to a destination mount of a given type.
759 * ---------------------------------------------------------------------------
760 * | BIND MOUNT OPERATION |
761 * |**************************************************************************
762 * | source-->| shared | private | slave | unbindable |
763 * | dest | | | | |
764 * | | | | | | |
765 * | v | | | | |
766 * |**************************************************************************
767 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
768 * | | | | | |
769 * |non-shared| shared (+) | private | slave (*) | invalid |
770 * ***************************************************************************
771 * A bind operation clones the source mount and mounts the clone on the
772 * destination mount.
774 * (++) the cloned mount is propagated to all the mounts in the propagation
775 * tree of the destination mount and the cloned mount is added to
776 * the peer group of the source mount.
777 * (+) the cloned mount is created under the destination mount and is marked
778 * as shared. The cloned mount is added to the peer group of the source
779 * mount.
780 * (+++) the mount is propagated to all the mounts in the propagation tree
781 * of the destination mount and the cloned mount is made slave
782 * of the same master as that of the source mount. The cloned mount
783 * is marked as 'shared and slave'.
784 * (*) the cloned mount is made a slave of the same master as that of the
785 * source mount.
787 * ---------------------------------------------------------------------------
788 * | MOVE MOUNT OPERATION |
789 * |**************************************************************************
790 * | source-->| shared | private | slave | unbindable |
791 * | dest | | | | |
792 * | | | | | | |
793 * | v | | | | |
794 * |**************************************************************************
795 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
796 * | | | | | |
797 * |non-shared| shared (+*) | private | slave (*) | unbindable |
798 * ***************************************************************************
800 * (+) the mount is moved to the destination. And is then propagated to
801 * all the mounts in the propagation tree of the destination mount.
802 * (+*) the mount is moved to the destination.
803 * (+++) the mount is moved to the destination and is then propagated to
804 * all the mounts belonging to the destination mount's propagation tree.
805 * the mount is marked as 'shared and slave'.
806 * (*) the mount continues to be a slave at the new location.
808 * if the source mount is a tree, the operations explained above is
809 * applied to each mount in the tree.
810 * Must be called without spinlocks held, since this function can sleep
811 * in allocations.
813 static int attach_recursive_mnt(struct vfsmount *source_mnt,
814 struct nameidata *nd, struct nameidata *parent_nd)
816 LIST_HEAD(tree_list);
817 struct vfsmount *dest_mnt = nd->mnt;
818 struct dentry *dest_dentry = nd->dentry;
819 struct vfsmount *child, *p;
821 if (propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list))
822 return -EINVAL;
824 if (IS_MNT_SHARED(dest_mnt)) {
825 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
826 set_mnt_shared(p);
829 spin_lock(&vfsmount_lock);
830 if (parent_nd) {
831 detach_mnt(source_mnt, parent_nd);
832 attach_mnt(source_mnt, nd);
833 touch_namespace(current->nsproxy->namespace);
834 } else {
835 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
836 commit_tree(source_mnt);
839 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
840 list_del_init(&child->mnt_hash);
841 commit_tree(child);
843 spin_unlock(&vfsmount_lock);
844 return 0;
847 static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
849 int err;
850 if (mnt->mnt_sb->s_flags & MS_NOUSER)
851 return -EINVAL;
853 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
854 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
855 return -ENOTDIR;
857 err = -ENOENT;
858 mutex_lock(&nd->dentry->d_inode->i_mutex);
859 if (IS_DEADDIR(nd->dentry->d_inode))
860 goto out_unlock;
862 err = security_sb_check_sb(mnt, nd);
863 if (err)
864 goto out_unlock;
866 err = -ENOENT;
867 if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry))
868 err = attach_recursive_mnt(mnt, nd, NULL);
869 out_unlock:
870 mutex_unlock(&nd->dentry->d_inode->i_mutex);
871 if (!err)
872 security_sb_post_addmount(mnt, nd);
873 return err;
877 * recursively change the type of the mountpoint.
879 static int do_change_type(struct nameidata *nd, int flag)
881 struct vfsmount *m, *mnt = nd->mnt;
882 int recurse = flag & MS_REC;
883 int type = flag & ~MS_REC;
885 if (nd->dentry != nd->mnt->mnt_root)
886 return -EINVAL;
888 down_write(&namespace_sem);
889 spin_lock(&vfsmount_lock);
890 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
891 change_mnt_propagation(m, type);
892 spin_unlock(&vfsmount_lock);
893 up_write(&namespace_sem);
894 return 0;
898 * do loopback mount.
900 static int do_loopback(struct nameidata *nd, char *old_name, int recurse)
902 struct nameidata old_nd;
903 struct vfsmount *mnt = NULL;
904 int err = mount_is_safe(nd);
905 if (err)
906 return err;
907 if (!old_name || !*old_name)
908 return -EINVAL;
909 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
910 if (err)
911 return err;
913 down_write(&namespace_sem);
914 err = -EINVAL;
915 if (IS_MNT_UNBINDABLE(old_nd.mnt))
916 goto out;
918 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
919 goto out;
921 err = -ENOMEM;
922 if (recurse)
923 mnt = copy_tree(old_nd.mnt, old_nd.dentry, 0);
924 else
925 mnt = clone_mnt(old_nd.mnt, old_nd.dentry, 0);
927 if (!mnt)
928 goto out;
930 err = graft_tree(mnt, nd);
931 if (err) {
932 LIST_HEAD(umount_list);
933 spin_lock(&vfsmount_lock);
934 umount_tree(mnt, 0, &umount_list);
935 spin_unlock(&vfsmount_lock);
936 release_mounts(&umount_list);
939 out:
940 up_write(&namespace_sem);
941 path_release(&old_nd);
942 return err;
946 * change filesystem flags. dir should be a physical root of filesystem.
947 * If you've mounted a non-root directory somewhere and want to do remount
948 * on it - tough luck.
950 static int do_remount(struct nameidata *nd, int flags, int mnt_flags,
951 void *data)
953 int err;
954 struct super_block *sb = nd->mnt->mnt_sb;
956 if (!capable(CAP_SYS_ADMIN))
957 return -EPERM;
959 if (!check_mnt(nd->mnt))
960 return -EINVAL;
962 if (nd->dentry != nd->mnt->mnt_root)
963 return -EINVAL;
965 down_write(&sb->s_umount);
966 err = do_remount_sb(sb, flags, data, 0);
967 if (!err)
968 nd->mnt->mnt_flags = mnt_flags;
969 up_write(&sb->s_umount);
970 if (!err)
971 security_sb_post_remount(nd->mnt, flags, data);
972 return err;
975 static inline int tree_contains_unbindable(struct vfsmount *mnt)
977 struct vfsmount *p;
978 for (p = mnt; p; p = next_mnt(p, mnt)) {
979 if (IS_MNT_UNBINDABLE(p))
980 return 1;
982 return 0;
985 static int do_move_mount(struct nameidata *nd, char *old_name)
987 struct nameidata old_nd, parent_nd;
988 struct vfsmount *p;
989 int err = 0;
990 if (!capable(CAP_SYS_ADMIN))
991 return -EPERM;
992 if (!old_name || !*old_name)
993 return -EINVAL;
994 err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
995 if (err)
996 return err;
998 down_write(&namespace_sem);
999 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
1001 err = -EINVAL;
1002 if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
1003 goto out;
1005 err = -ENOENT;
1006 mutex_lock(&nd->dentry->d_inode->i_mutex);
1007 if (IS_DEADDIR(nd->dentry->d_inode))
1008 goto out1;
1010 if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
1011 goto out1;
1013 err = -EINVAL;
1014 if (old_nd.dentry != old_nd.mnt->mnt_root)
1015 goto out1;
1017 if (old_nd.mnt == old_nd.mnt->mnt_parent)
1018 goto out1;
1020 if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
1021 S_ISDIR(old_nd.dentry->d_inode->i_mode))
1022 goto out1;
1024 * Don't move a mount residing in a shared parent.
1026 if (old_nd.mnt->mnt_parent && IS_MNT_SHARED(old_nd.mnt->mnt_parent))
1027 goto out1;
1029 * Don't move a mount tree containing unbindable mounts to a destination
1030 * mount which is shared.
1032 if (IS_MNT_SHARED(nd->mnt) && tree_contains_unbindable(old_nd.mnt))
1033 goto out1;
1034 err = -ELOOP;
1035 for (p = nd->mnt; p->mnt_parent != p; p = p->mnt_parent)
1036 if (p == old_nd.mnt)
1037 goto out1;
1039 if ((err = attach_recursive_mnt(old_nd.mnt, nd, &parent_nd)))
1040 goto out1;
1042 spin_lock(&vfsmount_lock);
1043 /* if the mount is moved, it should no longer be expire
1044 * automatically */
1045 list_del_init(&old_nd.mnt->mnt_expire);
1046 spin_unlock(&vfsmount_lock);
1047 out1:
1048 mutex_unlock(&nd->dentry->d_inode->i_mutex);
1049 out:
1050 up_write(&namespace_sem);
1051 if (!err)
1052 path_release(&parent_nd);
1053 path_release(&old_nd);
1054 return err;
1058 * create a new mount for userspace and request it to be added into the
1059 * namespace's tree
1061 static int do_new_mount(struct nameidata *nd, char *type, int flags,
1062 int mnt_flags, char *name, void *data)
1064 struct vfsmount *mnt;
1066 if (!type || !memchr(type, 0, PAGE_SIZE))
1067 return -EINVAL;
1069 /* we need capabilities... */
1070 if (!capable(CAP_SYS_ADMIN))
1071 return -EPERM;
1073 mnt = do_kern_mount(type, flags, name, data);
1074 if (IS_ERR(mnt))
1075 return PTR_ERR(mnt);
1077 return do_add_mount(mnt, nd, mnt_flags, NULL);
1081 * add a mount into a namespace's mount tree
1082 * - provide the option of adding the new mount to an expiration list
1084 int do_add_mount(struct vfsmount *newmnt, struct nameidata *nd,
1085 int mnt_flags, struct list_head *fslist)
1087 int err;
1089 down_write(&namespace_sem);
1090 /* Something was mounted here while we slept */
1091 while (d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
1093 err = -EINVAL;
1094 if (!check_mnt(nd->mnt))
1095 goto unlock;
1097 /* Refuse the same filesystem on the same mount point */
1098 err = -EBUSY;
1099 if (nd->mnt->mnt_sb == newmnt->mnt_sb &&
1100 nd->mnt->mnt_root == nd->dentry)
1101 goto unlock;
1103 err = -EINVAL;
1104 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1105 goto unlock;
1107 newmnt->mnt_flags = mnt_flags;
1108 if ((err = graft_tree(newmnt, nd)))
1109 goto unlock;
1111 if (fslist) {
1112 /* add to the specified expiration list */
1113 spin_lock(&vfsmount_lock);
1114 list_add_tail(&newmnt->mnt_expire, fslist);
1115 spin_unlock(&vfsmount_lock);
1117 up_write(&namespace_sem);
1118 return 0;
1120 unlock:
1121 up_write(&namespace_sem);
1122 mntput(newmnt);
1123 return err;
1126 EXPORT_SYMBOL_GPL(do_add_mount);
1128 static void expire_mount(struct vfsmount *mnt, struct list_head *mounts,
1129 struct list_head *umounts)
1131 spin_lock(&vfsmount_lock);
1134 * Check if mount is still attached, if not, let whoever holds it deal
1135 * with the sucker
1137 if (mnt->mnt_parent == mnt) {
1138 spin_unlock(&vfsmount_lock);
1139 return;
1143 * Check that it is still dead: the count should now be 2 - as
1144 * contributed by the vfsmount parent and the mntget above
1146 if (!propagate_mount_busy(mnt, 2)) {
1147 /* delete from the namespace */
1148 touch_namespace(mnt->mnt_namespace);
1149 list_del_init(&mnt->mnt_list);
1150 mnt->mnt_namespace = NULL;
1151 umount_tree(mnt, 1, umounts);
1152 spin_unlock(&vfsmount_lock);
1153 } else {
1155 * Someone brought it back to life whilst we didn't have any
1156 * locks held so return it to the expiration list
1158 list_add_tail(&mnt->mnt_expire, mounts);
1159 spin_unlock(&vfsmount_lock);
1164 * go through the vfsmounts we've just consigned to the graveyard to
1165 * - check that they're still dead
1166 * - delete the vfsmount from the appropriate namespace under lock
1167 * - dispose of the corpse
1169 static void expire_mount_list(struct list_head *graveyard, struct list_head *mounts)
1171 struct namespace *namespace;
1172 struct vfsmount *mnt;
1174 while (!list_empty(graveyard)) {
1175 LIST_HEAD(umounts);
1176 mnt = list_entry(graveyard->next, struct vfsmount, mnt_expire);
1177 list_del_init(&mnt->mnt_expire);
1179 /* don't do anything if the namespace is dead - all the
1180 * vfsmounts from it are going away anyway */
1181 namespace = mnt->mnt_namespace;
1182 if (!namespace || !namespace->root)
1183 continue;
1184 get_namespace(namespace);
1186 spin_unlock(&vfsmount_lock);
1187 down_write(&namespace_sem);
1188 expire_mount(mnt, mounts, &umounts);
1189 up_write(&namespace_sem);
1190 release_mounts(&umounts);
1191 mntput(mnt);
1192 put_namespace(namespace);
1193 spin_lock(&vfsmount_lock);
1198 * process a list of expirable mountpoints with the intent of discarding any
1199 * mountpoints that aren't in use and haven't been touched since last we came
1200 * here
1202 void mark_mounts_for_expiry(struct list_head *mounts)
1204 struct vfsmount *mnt, *next;
1205 LIST_HEAD(graveyard);
1207 if (list_empty(mounts))
1208 return;
1210 spin_lock(&vfsmount_lock);
1212 /* extract from the expiration list every vfsmount that matches the
1213 * following criteria:
1214 * - only referenced by its parent vfsmount
1215 * - still marked for expiry (marked on the last call here; marks are
1216 * cleared by mntput())
1218 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
1219 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
1220 atomic_read(&mnt->mnt_count) != 1)
1221 continue;
1223 mntget(mnt);
1224 list_move(&mnt->mnt_expire, &graveyard);
1227 expire_mount_list(&graveyard, mounts);
1229 spin_unlock(&vfsmount_lock);
1232 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1235 * Ripoff of 'select_parent()'
1237 * search the list of submounts for a given mountpoint, and move any
1238 * shrinkable submounts to the 'graveyard' list.
1240 static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1242 struct vfsmount *this_parent = parent;
1243 struct list_head *next;
1244 int found = 0;
1246 repeat:
1247 next = this_parent->mnt_mounts.next;
1248 resume:
1249 while (next != &this_parent->mnt_mounts) {
1250 struct list_head *tmp = next;
1251 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1253 next = tmp->next;
1254 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1255 continue;
1257 * Descend a level if the d_mounts list is non-empty.
1259 if (!list_empty(&mnt->mnt_mounts)) {
1260 this_parent = mnt;
1261 goto repeat;
1264 if (!propagate_mount_busy(mnt, 1)) {
1265 mntget(mnt);
1266 list_move_tail(&mnt->mnt_expire, graveyard);
1267 found++;
1271 * All done at this level ... ascend and resume the search
1273 if (this_parent != parent) {
1274 next = this_parent->mnt_child.next;
1275 this_parent = this_parent->mnt_parent;
1276 goto resume;
1278 return found;
1282 * process a list of expirable mountpoints with the intent of discarding any
1283 * submounts of a specific parent mountpoint
1285 void shrink_submounts(struct vfsmount *mountpoint, struct list_head *mounts)
1287 LIST_HEAD(graveyard);
1288 int found;
1290 spin_lock(&vfsmount_lock);
1292 /* extract submounts of 'mountpoint' from the expiration list */
1293 while ((found = select_submounts(mountpoint, &graveyard)) != 0)
1294 expire_mount_list(&graveyard, mounts);
1296 spin_unlock(&vfsmount_lock);
1299 EXPORT_SYMBOL_GPL(shrink_submounts);
1302 * Some copy_from_user() implementations do not return the exact number of
1303 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1304 * Note that this function differs from copy_from_user() in that it will oops
1305 * on bad values of `to', rather than returning a short copy.
1307 static long exact_copy_from_user(void *to, const void __user * from,
1308 unsigned long n)
1310 char *t = to;
1311 const char __user *f = from;
1312 char c;
1314 if (!access_ok(VERIFY_READ, from, n))
1315 return n;
1317 while (n) {
1318 if (__get_user(c, f)) {
1319 memset(t, 0, n);
1320 break;
1322 *t++ = c;
1323 f++;
1324 n--;
1326 return n;
1329 int copy_mount_options(const void __user * data, unsigned long *where)
1331 int i;
1332 unsigned long page;
1333 unsigned long size;
1335 *where = 0;
1336 if (!data)
1337 return 0;
1339 if (!(page = __get_free_page(GFP_KERNEL)))
1340 return -ENOMEM;
1342 /* We only care that *some* data at the address the user
1343 * gave us is valid. Just in case, we'll zero
1344 * the remainder of the page.
1346 /* copy_from_user cannot cross TASK_SIZE ! */
1347 size = TASK_SIZE - (unsigned long)data;
1348 if (size > PAGE_SIZE)
1349 size = PAGE_SIZE;
1351 i = size - exact_copy_from_user((void *)page, data, size);
1352 if (!i) {
1353 free_page(page);
1354 return -EFAULT;
1356 if (i != PAGE_SIZE)
1357 memset((char *)page + i, 0, PAGE_SIZE - i);
1358 *where = page;
1359 return 0;
1363 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1364 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1366 * data is a (void *) that can point to any structure up to
1367 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1368 * information (or be NULL).
1370 * Pre-0.97 versions of mount() didn't have a flags word.
1371 * When the flags word was introduced its top half was required
1372 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1373 * Therefore, if this magic number is present, it carries no information
1374 * and must be discarded.
1376 long do_mount(char *dev_name, char *dir_name, char *type_page,
1377 unsigned long flags, void *data_page)
1379 struct nameidata nd;
1380 int retval = 0;
1381 int mnt_flags = 0;
1383 /* Discard magic */
1384 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1385 flags &= ~MS_MGC_MSK;
1387 /* Basic sanity checks */
1389 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1390 return -EINVAL;
1391 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1392 return -EINVAL;
1394 if (data_page)
1395 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1397 /* Separate the per-mountpoint flags */
1398 if (flags & MS_NOSUID)
1399 mnt_flags |= MNT_NOSUID;
1400 if (flags & MS_NODEV)
1401 mnt_flags |= MNT_NODEV;
1402 if (flags & MS_NOEXEC)
1403 mnt_flags |= MNT_NOEXEC;
1404 if (flags & MS_NOATIME)
1405 mnt_flags |= MNT_NOATIME;
1406 if (flags & MS_NODIRATIME)
1407 mnt_flags |= MNT_NODIRATIME;
1409 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
1410 MS_NOATIME | MS_NODIRATIME);
1412 /* ... and get the mountpoint */
1413 retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
1414 if (retval)
1415 return retval;
1417 retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
1418 if (retval)
1419 goto dput_out;
1421 if (flags & MS_REMOUNT)
1422 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
1423 data_page);
1424 else if (flags & MS_BIND)
1425 retval = do_loopback(&nd, dev_name, flags & MS_REC);
1426 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1427 retval = do_change_type(&nd, flags);
1428 else if (flags & MS_MOVE)
1429 retval = do_move_mount(&nd, dev_name);
1430 else
1431 retval = do_new_mount(&nd, type_page, flags, mnt_flags,
1432 dev_name, data_page);
1433 dput_out:
1434 path_release(&nd);
1435 return retval;
1439 * Allocate a new namespace structure and populate it with contents
1440 * copied from the namespace of the passed in task structure.
1442 struct namespace *dup_namespace(struct task_struct *tsk, struct fs_struct *fs)
1444 struct namespace *namespace = tsk->nsproxy->namespace;
1445 struct namespace *new_ns;
1446 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
1447 struct vfsmount *p, *q;
1449 new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL);
1450 if (!new_ns)
1451 return NULL;
1453 atomic_set(&new_ns->count, 1);
1454 INIT_LIST_HEAD(&new_ns->list);
1455 init_waitqueue_head(&new_ns->poll);
1456 new_ns->event = 0;
1458 down_write(&namespace_sem);
1459 /* First pass: copy the tree topology */
1460 new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root,
1461 CL_COPY_ALL | CL_EXPIRE);
1462 if (!new_ns->root) {
1463 up_write(&namespace_sem);
1464 kfree(new_ns);
1465 return NULL;
1467 spin_lock(&vfsmount_lock);
1468 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
1469 spin_unlock(&vfsmount_lock);
1472 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
1473 * as belonging to new namespace. We have already acquired a private
1474 * fs_struct, so tsk->fs->lock is not needed.
1476 p = namespace->root;
1477 q = new_ns->root;
1478 while (p) {
1479 q->mnt_namespace = new_ns;
1480 if (fs) {
1481 if (p == fs->rootmnt) {
1482 rootmnt = p;
1483 fs->rootmnt = mntget(q);
1485 if (p == fs->pwdmnt) {
1486 pwdmnt = p;
1487 fs->pwdmnt = mntget(q);
1489 if (p == fs->altrootmnt) {
1490 altrootmnt = p;
1491 fs->altrootmnt = mntget(q);
1494 p = next_mnt(p, namespace->root);
1495 q = next_mnt(q, new_ns->root);
1497 up_write(&namespace_sem);
1499 if (rootmnt)
1500 mntput(rootmnt);
1501 if (pwdmnt)
1502 mntput(pwdmnt);
1503 if (altrootmnt)
1504 mntput(altrootmnt);
1506 return new_ns;
1509 int copy_namespace(int flags, struct task_struct *tsk)
1511 struct namespace *namespace = tsk->nsproxy->namespace;
1512 struct namespace *new_ns;
1513 int err = 0;
1515 if (!namespace)
1516 return 0;
1518 get_namespace(namespace);
1520 if (!(flags & CLONE_NEWNS))
1521 return 0;
1523 if (!capable(CAP_SYS_ADMIN)) {
1524 err = -EPERM;
1525 goto out;
1528 new_ns = dup_namespace(tsk, tsk->fs);
1529 if (!new_ns) {
1530 err = -ENOMEM;
1531 goto out;
1534 tsk->nsproxy->namespace = new_ns;
1536 out:
1537 put_namespace(namespace);
1538 return err;
1541 asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
1542 char __user * type, unsigned long flags,
1543 void __user * data)
1545 int retval;
1546 unsigned long data_page;
1547 unsigned long type_page;
1548 unsigned long dev_page;
1549 char *dir_page;
1551 retval = copy_mount_options(type, &type_page);
1552 if (retval < 0)
1553 return retval;
1555 dir_page = getname(dir_name);
1556 retval = PTR_ERR(dir_page);
1557 if (IS_ERR(dir_page))
1558 goto out1;
1560 retval = copy_mount_options(dev_name, &dev_page);
1561 if (retval < 0)
1562 goto out2;
1564 retval = copy_mount_options(data, &data_page);
1565 if (retval < 0)
1566 goto out3;
1568 lock_kernel();
1569 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
1570 flags, (void *)data_page);
1571 unlock_kernel();
1572 free_page(data_page);
1574 out3:
1575 free_page(dev_page);
1576 out2:
1577 putname(dir_page);
1578 out1:
1579 free_page(type_page);
1580 return retval;
1584 * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
1585 * It can block. Requires the big lock held.
1587 void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
1588 struct dentry *dentry)
1590 struct dentry *old_root;
1591 struct vfsmount *old_rootmnt;
1592 write_lock(&fs->lock);
1593 old_root = fs->root;
1594 old_rootmnt = fs->rootmnt;
1595 fs->rootmnt = mntget(mnt);
1596 fs->root = dget(dentry);
1597 write_unlock(&fs->lock);
1598 if (old_root) {
1599 dput(old_root);
1600 mntput(old_rootmnt);
1605 * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
1606 * It can block. Requires the big lock held.
1608 void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
1609 struct dentry *dentry)
1611 struct dentry *old_pwd;
1612 struct vfsmount *old_pwdmnt;
1614 write_lock(&fs->lock);
1615 old_pwd = fs->pwd;
1616 old_pwdmnt = fs->pwdmnt;
1617 fs->pwdmnt = mntget(mnt);
1618 fs->pwd = dget(dentry);
1619 write_unlock(&fs->lock);
1621 if (old_pwd) {
1622 dput(old_pwd);
1623 mntput(old_pwdmnt);
1627 static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
1629 struct task_struct *g, *p;
1630 struct fs_struct *fs;
1632 read_lock(&tasklist_lock);
1633 do_each_thread(g, p) {
1634 task_lock(p);
1635 fs = p->fs;
1636 if (fs) {
1637 atomic_inc(&fs->count);
1638 task_unlock(p);
1639 if (fs->root == old_nd->dentry
1640 && fs->rootmnt == old_nd->mnt)
1641 set_fs_root(fs, new_nd->mnt, new_nd->dentry);
1642 if (fs->pwd == old_nd->dentry
1643 && fs->pwdmnt == old_nd->mnt)
1644 set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
1645 put_fs_struct(fs);
1646 } else
1647 task_unlock(p);
1648 } while_each_thread(g, p);
1649 read_unlock(&tasklist_lock);
1653 * pivot_root Semantics:
1654 * Moves the root file system of the current process to the directory put_old,
1655 * makes new_root as the new root file system of the current process, and sets
1656 * root/cwd of all processes which had them on the current root to new_root.
1658 * Restrictions:
1659 * The new_root and put_old must be directories, and must not be on the
1660 * same file system as the current process root. The put_old must be
1661 * underneath new_root, i.e. adding a non-zero number of /.. to the string
1662 * pointed to by put_old must yield the same directory as new_root. No other
1663 * file system may be mounted on put_old. After all, new_root is a mountpoint.
1665 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
1666 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
1667 * in this situation.
1669 * Notes:
1670 * - we don't move root/cwd if they are not at the root (reason: if something
1671 * cared enough to change them, it's probably wrong to force them elsewhere)
1672 * - it's okay to pick a root that isn't the root of a file system, e.g.
1673 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
1674 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
1675 * first.
1677 asmlinkage long sys_pivot_root(const char __user * new_root,
1678 const char __user * put_old)
1680 struct vfsmount *tmp;
1681 struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1682 int error;
1684 if (!capable(CAP_SYS_ADMIN))
1685 return -EPERM;
1687 lock_kernel();
1689 error = __user_walk(new_root, LOOKUP_FOLLOW | LOOKUP_DIRECTORY,
1690 &new_nd);
1691 if (error)
1692 goto out0;
1693 error = -EINVAL;
1694 if (!check_mnt(new_nd.mnt))
1695 goto out1;
1697 error = __user_walk(put_old, LOOKUP_FOLLOW | LOOKUP_DIRECTORY, &old_nd);
1698 if (error)
1699 goto out1;
1701 error = security_sb_pivotroot(&old_nd, &new_nd);
1702 if (error) {
1703 path_release(&old_nd);
1704 goto out1;
1707 read_lock(&current->fs->lock);
1708 user_nd.mnt = mntget(current->fs->rootmnt);
1709 user_nd.dentry = dget(current->fs->root);
1710 read_unlock(&current->fs->lock);
1711 down_write(&namespace_sem);
1712 mutex_lock(&old_nd.dentry->d_inode->i_mutex);
1713 error = -EINVAL;
1714 if (IS_MNT_SHARED(old_nd.mnt) ||
1715 IS_MNT_SHARED(new_nd.mnt->mnt_parent) ||
1716 IS_MNT_SHARED(user_nd.mnt->mnt_parent))
1717 goto out2;
1718 if (!check_mnt(user_nd.mnt))
1719 goto out2;
1720 error = -ENOENT;
1721 if (IS_DEADDIR(new_nd.dentry->d_inode))
1722 goto out2;
1723 if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1724 goto out2;
1725 if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1726 goto out2;
1727 error = -EBUSY;
1728 if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1729 goto out2; /* loop, on the same file system */
1730 error = -EINVAL;
1731 if (user_nd.mnt->mnt_root != user_nd.dentry)
1732 goto out2; /* not a mountpoint */
1733 if (user_nd.mnt->mnt_parent == user_nd.mnt)
1734 goto out2; /* not attached */
1735 if (new_nd.mnt->mnt_root != new_nd.dentry)
1736 goto out2; /* not a mountpoint */
1737 if (new_nd.mnt->mnt_parent == new_nd.mnt)
1738 goto out2; /* not attached */
1739 tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1740 spin_lock(&vfsmount_lock);
1741 if (tmp != new_nd.mnt) {
1742 for (;;) {
1743 if (tmp->mnt_parent == tmp)
1744 goto out3; /* already mounted on put_old */
1745 if (tmp->mnt_parent == new_nd.mnt)
1746 break;
1747 tmp = tmp->mnt_parent;
1749 if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1750 goto out3;
1751 } else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1752 goto out3;
1753 detach_mnt(new_nd.mnt, &parent_nd);
1754 detach_mnt(user_nd.mnt, &root_parent);
1755 attach_mnt(user_nd.mnt, &old_nd); /* mount old root on put_old */
1756 attach_mnt(new_nd.mnt, &root_parent); /* mount new_root on / */
1757 touch_namespace(current->nsproxy->namespace);
1758 spin_unlock(&vfsmount_lock);
1759 chroot_fs_refs(&user_nd, &new_nd);
1760 security_sb_post_pivotroot(&user_nd, &new_nd);
1761 error = 0;
1762 path_release(&root_parent);
1763 path_release(&parent_nd);
1764 out2:
1765 mutex_unlock(&old_nd.dentry->d_inode->i_mutex);
1766 up_write(&namespace_sem);
1767 path_release(&user_nd);
1768 path_release(&old_nd);
1769 out1:
1770 path_release(&new_nd);
1771 out0:
1772 unlock_kernel();
1773 return error;
1774 out3:
1775 spin_unlock(&vfsmount_lock);
1776 goto out2;
1779 static void __init init_mount_tree(void)
1781 struct vfsmount *mnt;
1782 struct namespace *namespace;
1784 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1785 if (IS_ERR(mnt))
1786 panic("Can't create rootfs");
1787 namespace = kmalloc(sizeof(*namespace), GFP_KERNEL);
1788 if (!namespace)
1789 panic("Can't allocate initial namespace");
1790 atomic_set(&namespace->count, 1);
1791 INIT_LIST_HEAD(&namespace->list);
1792 init_waitqueue_head(&namespace->poll);
1793 namespace->event = 0;
1794 list_add(&mnt->mnt_list, &namespace->list);
1795 namespace->root = mnt;
1796 mnt->mnt_namespace = namespace;
1798 init_task.nsproxy->namespace = namespace;
1799 get_namespace(namespace);
1801 set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root);
1802 set_fs_root(current->fs, namespace->root, namespace->root->mnt_root);
1805 void __init mnt_init(unsigned long mempages)
1807 struct list_head *d;
1808 unsigned int nr_hash;
1809 int i;
1810 int err;
1812 init_rwsem(&namespace_sem);
1814 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
1815 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL, NULL);
1817 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
1819 if (!mount_hashtable)
1820 panic("Failed to allocate mount hash table\n");
1823 * Find the power-of-two list-heads that can fit into the allocation..
1824 * We don't guarantee that "sizeof(struct list_head)" is necessarily
1825 * a power-of-two.
1827 nr_hash = PAGE_SIZE / sizeof(struct list_head);
1828 hash_bits = 0;
1829 do {
1830 hash_bits++;
1831 } while ((nr_hash >> hash_bits) != 0);
1832 hash_bits--;
1835 * Re-calculate the actual number of entries and the mask
1836 * from the number of bits we can fit.
1838 nr_hash = 1UL << hash_bits;
1839 hash_mask = nr_hash - 1;
1841 printk("Mount-cache hash table entries: %d\n", nr_hash);
1843 /* And initialize the newly allocated array */
1844 d = mount_hashtable;
1845 i = nr_hash;
1846 do {
1847 INIT_LIST_HEAD(d);
1848 d++;
1849 i--;
1850 } while (i);
1851 err = sysfs_init();
1852 if (err)
1853 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
1854 __FUNCTION__, err);
1855 err = subsystem_register(&fs_subsys);
1856 if (err)
1857 printk(KERN_WARNING "%s: subsystem_register error: %d\n",
1858 __FUNCTION__, err);
1859 init_rootfs();
1860 init_mount_tree();
1863 void __put_namespace(struct namespace *namespace)
1865 struct vfsmount *root = namespace->root;
1866 LIST_HEAD(umount_list);
1867 namespace->root = NULL;
1868 spin_unlock(&vfsmount_lock);
1869 down_write(&namespace_sem);
1870 spin_lock(&vfsmount_lock);
1871 umount_tree(root, 0, &umount_list);
1872 spin_unlock(&vfsmount_lock);
1873 up_write(&namespace_sem);
1874 release_mounts(&umount_list);
1875 kfree(namespace);