~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/namespace.c

Version: ~ [ linux-5.8 ] ~ [ linux-5.7.14 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.57 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.138 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.193 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.232 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.232 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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/config.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/quotaops.h>
 17 #include <linux/acct.h>
 18 #include <linux/module.h>
 19 #include <linux/seq_file.h>
 20 #include <linux/namespace.h>
 21 #include <linux/namei.h>
 22 #include <linux/security.h>
 23 #include <linux/mount.h>
 24 #include <asm/uaccess.h>
 25 
 26 extern int __init init_rootfs(void);
 27 extern int __init sysfs_init(void);
 28 
 29 /* spinlock for vfsmount related operations, inplace of dcache_lock */
 30 spinlock_t vfsmount_lock __cacheline_aligned_in_smp = SPIN_LOCK_UNLOCKED;
 31 static struct list_head *mount_hashtable;
 32 static int hash_mask, hash_bits;
 33 static kmem_cache_t *mnt_cache; 
 34 
 35 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
 36 {
 37         unsigned long tmp = ((unsigned long) mnt / L1_CACHE_BYTES);
 38         tmp += ((unsigned long) dentry / L1_CACHE_BYTES);
 39         tmp = tmp + (tmp >> hash_bits);
 40         return tmp & hash_mask;
 41 }
 42 
 43 struct vfsmount *alloc_vfsmnt(const char *name)
 44 {
 45         struct vfsmount *mnt = kmem_cache_alloc(mnt_cache, GFP_KERNEL); 
 46         if (mnt) {
 47                 memset(mnt, 0, sizeof(struct vfsmount));
 48                 atomic_set(&mnt->mnt_count,1);
 49                 INIT_LIST_HEAD(&mnt->mnt_hash);
 50                 INIT_LIST_HEAD(&mnt->mnt_child);
 51                 INIT_LIST_HEAD(&mnt->mnt_mounts);
 52                 INIT_LIST_HEAD(&mnt->mnt_list);
 53                 if (name) {
 54                         int size = strlen(name)+1;
 55                         char *newname = kmalloc(size, GFP_KERNEL);
 56                         if (newname) {
 57                                 memcpy(newname, name, size);
 58                                 mnt->mnt_devname = newname;
 59                         }
 60                 }
 61         }
 62         return mnt;
 63 }
 64 
 65 void free_vfsmnt(struct vfsmount *mnt)
 66 {
 67         kfree(mnt->mnt_devname);
 68         kmem_cache_free(mnt_cache, mnt);
 69 }
 70 
 71 /*
 72  * Now, lookup_mnt increments the ref count before returning
 73  * the vfsmount struct.
 74  */
 75 struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
 76 {
 77         struct list_head * head = mount_hashtable + hash(mnt, dentry);
 78         struct list_head * tmp = head;
 79         struct vfsmount *p, *found = NULL;
 80 
 81         spin_lock(&vfsmount_lock);
 82         for (;;) {
 83                 tmp = tmp->next;
 84                 p = NULL;
 85                 if (tmp == head)
 86                         break;
 87                 p = list_entry(tmp, struct vfsmount, mnt_hash);
 88                 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
 89                         found = mntget(p);
 90                         break;
 91                 }
 92         }
 93         spin_unlock(&vfsmount_lock);
 94         return found;
 95 }
 96 
 97 EXPORT_SYMBOL(lookup_mnt);
 98 
 99 static int check_mnt(struct vfsmount *mnt)
100 {
101         spin_lock(&vfsmount_lock);
102         while (mnt->mnt_parent != mnt)
103                 mnt = mnt->mnt_parent;
104         spin_unlock(&vfsmount_lock);
105         return mnt == current->namespace->root;
106 }
107 
108 static void detach_mnt(struct vfsmount *mnt, struct nameidata *old_nd)
109 {
110         old_nd->dentry = mnt->mnt_mountpoint;
111         old_nd->mnt = mnt->mnt_parent;
112         mnt->mnt_parent = mnt;
113         mnt->mnt_mountpoint = mnt->mnt_root;
114         list_del_init(&mnt->mnt_child);
115         list_del_init(&mnt->mnt_hash);
116         old_nd->dentry->d_mounted--;
117 }
118 
119 static void attach_mnt(struct vfsmount *mnt, struct nameidata *nd)
120 {
121         mnt->mnt_parent = mntget(nd->mnt);
122         mnt->mnt_mountpoint = dget(nd->dentry);
123         list_add(&mnt->mnt_hash, mount_hashtable+hash(nd->mnt, nd->dentry));
124         list_add_tail(&mnt->mnt_child, &nd->mnt->mnt_mounts);
125         nd->dentry->d_mounted++;
126 }
127 
128 static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
129 {
130         struct list_head *next = p->mnt_mounts.next;
131         if (next == &p->mnt_mounts) {
132                 while (1) {
133                         if (p == root)
134                                 return NULL;
135                         next = p->mnt_child.next;
136                         if (next != &p->mnt_parent->mnt_mounts)
137                                 break;
138                         p = p->mnt_parent;
139                 }
140         }
141         return list_entry(next, struct vfsmount, mnt_child);
142 }
143 
144 static struct vfsmount *
145 clone_mnt(struct vfsmount *old, struct dentry *root)
146 {
147         struct super_block *sb = old->mnt_sb;
148         struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
149 
150         if (mnt) {
151                 mnt->mnt_flags = old->mnt_flags;
152                 atomic_inc(&sb->s_active);
153                 mnt->mnt_sb = sb;
154                 mnt->mnt_root = dget(root);
155                 mnt->mnt_mountpoint = mnt->mnt_root;
156                 mnt->mnt_parent = mnt;
157         }
158         return mnt;
159 }
160 
161 void __mntput(struct vfsmount *mnt)
162 {
163         struct super_block *sb = mnt->mnt_sb;
164         dput(mnt->mnt_root);
165         free_vfsmnt(mnt);
166         deactivate_super(sb);
167 }
168 
169 EXPORT_SYMBOL(__mntput);
170 
171 /* iterator */
172 static void *m_start(struct seq_file *m, loff_t *pos)
173 {
174         struct namespace *n = m->private;
175         struct list_head *p;
176         loff_t l = *pos;
177 
178         down_read(&n->sem);
179         list_for_each(p, &n->list)
180                 if (!l--)
181                         return list_entry(p, struct vfsmount, mnt_list);
182         return NULL;
183 }
184 
185 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
186 {
187         struct namespace *n = m->private;
188         struct list_head *p = ((struct vfsmount *)v)->mnt_list.next;
189         (*pos)++;
190         return p==&n->list ? NULL : list_entry(p, struct vfsmount, mnt_list);
191 }
192 
193 static void m_stop(struct seq_file *m, void *v)
194 {
195         struct namespace *n = m->private;
196         up_read(&n->sem);
197 }
198 
199 static inline void mangle(struct seq_file *m, const char *s)
200 {
201         seq_escape(m, s, " \t\n\\");
202 }
203 
204 static int show_vfsmnt(struct seq_file *m, void *v)
205 {
206         struct vfsmount *mnt = v;
207         int err = 0;
208         static struct proc_fs_info {
209                 int flag;
210                 char *str;
211         } fs_info[] = {
212                 { MS_SYNCHRONOUS, ",sync" },
213                 { MS_DIRSYNC, ",dirsync" },
214                 { MS_MANDLOCK, ",mand" },
215                 { MS_NOATIME, ",noatime" },
216                 { MS_NODIRATIME, ",nodiratime" },
217                 { 0, NULL }
218         };
219         static struct proc_fs_info mnt_info[] = {
220                 { MNT_NOSUID, ",nosuid" },
221                 { MNT_NODEV, ",nodev" },
222                 { MNT_NOEXEC, ",noexec" },
223                 { 0, NULL }
224         };
225         struct proc_fs_info *fs_infop;
226 
227         mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
228         seq_putc(m, ' ');
229         seq_path(m, mnt, mnt->mnt_root, " \t\n\\");
230         seq_putc(m, ' ');
231         mangle(m, mnt->mnt_sb->s_type->name);
232         seq_puts(m, mnt->mnt_sb->s_flags & MS_RDONLY ? " ro" : " rw");
233         for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
234                 if (mnt->mnt_sb->s_flags & fs_infop->flag)
235                         seq_puts(m, fs_infop->str);
236         }
237         for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
238                 if (mnt->mnt_flags & fs_infop->flag)
239                         seq_puts(m, fs_infop->str);
240         }
241         if (mnt->mnt_sb->s_op->show_options)
242                 err = mnt->mnt_sb->s_op->show_options(m, mnt);
243         seq_puts(m, " 0 0\n");
244         return err;
245 }
246 
247 struct seq_operations mounts_op = {
248         .start  = m_start,
249         .next   = m_next,
250         .stop   = m_stop,
251         .show   = show_vfsmnt
252 };
253 
254 /*
255  * Doesn't take quota and stuff into account. IOW, in some cases it will
256  * give false negatives. The main reason why it's here is that we need
257  * a non-destructive way to look for easily umountable filesystems.
258  */
259 int may_umount(struct vfsmount *mnt)
260 {
261         if (atomic_read(&mnt->mnt_count) > 2)
262                 return -EBUSY;
263         return 0;
264 }
265 
266 EXPORT_SYMBOL(may_umount);
267 
268 void umount_tree(struct vfsmount *mnt)
269 {
270         struct vfsmount *p;
271         LIST_HEAD(kill);
272 
273         for (p = mnt; p; p = next_mnt(p, mnt)) {
274                 list_del(&p->mnt_list);
275                 list_add(&p->mnt_list, &kill);
276         }
277 
278         while (!list_empty(&kill)) {
279                 mnt = list_entry(kill.next, struct vfsmount, mnt_list);
280                 list_del_init(&mnt->mnt_list);
281                 if (mnt->mnt_parent == mnt) {
282                         spin_unlock(&vfsmount_lock);
283                 } else {
284                         struct nameidata old_nd;
285                         detach_mnt(mnt, &old_nd);
286                         spin_unlock(&vfsmount_lock);
287                         path_release(&old_nd);
288                 }
289                 mntput(mnt);
290                 spin_lock(&vfsmount_lock);
291         }
292 }
293 
294 static int do_umount(struct vfsmount *mnt, int flags)
295 {
296         struct super_block * sb = mnt->mnt_sb;
297         int retval;
298 
299         retval = security_sb_umount(mnt, flags);
300         if (retval)
301                 return retval;
302 
303         /*
304          * If we may have to abort operations to get out of this
305          * mount, and they will themselves hold resources we must
306          * allow the fs to do things. In the Unix tradition of
307          * 'Gee thats tricky lets do it in userspace' the umount_begin
308          * might fail to complete on the first run through as other tasks
309          * must return, and the like. Thats for the mount program to worry
310          * about for the moment.
311          */
312 
313         lock_kernel();
314         if( (flags&MNT_FORCE) && sb->s_op->umount_begin)
315                 sb->s_op->umount_begin(sb);
316         unlock_kernel();
317 
318         /*
319          * No sense to grab the lock for this test, but test itself looks
320          * somewhat bogus. Suggestions for better replacement?
321          * Ho-hum... In principle, we might treat that as umount + switch
322          * to rootfs. GC would eventually take care of the old vfsmount.
323          * Actually it makes sense, especially if rootfs would contain a
324          * /reboot - static binary that would close all descriptors and
325          * call reboot(9). Then init(8) could umount root and exec /reboot.
326          */
327         if (mnt == current->fs->rootmnt && !(flags & MNT_DETACH)) {
328                 /*
329                  * Special case for "unmounting" root ...
330                  * we just try to remount it readonly.
331                  */
332                 down_write(&sb->s_umount);
333                 if (!(sb->s_flags & MS_RDONLY)) {
334                         lock_kernel();
335                         retval = do_remount_sb(sb, MS_RDONLY, 0, 0);
336                         unlock_kernel();
337                 }
338                 up_write(&sb->s_umount);
339                 return retval;
340         }
341 
342         down_write(&current->namespace->sem);
343         spin_lock(&vfsmount_lock);
344 
345         if (atomic_read(&sb->s_active) == 1) {
346                 /* last instance - try to be smart */
347                 spin_unlock(&vfsmount_lock);
348                 lock_kernel();
349                 DQUOT_OFF(sb);
350                 acct_auto_close(sb);
351                 unlock_kernel();
352                 security_sb_umount_close(mnt);
353                 spin_lock(&vfsmount_lock);
354         }
355         retval = -EBUSY;
356         if (atomic_read(&mnt->mnt_count) == 2 || flags & MNT_DETACH) {
357                 if (!list_empty(&mnt->mnt_list))
358                         umount_tree(mnt);
359                 retval = 0;
360         }
361         spin_unlock(&vfsmount_lock);
362         if (retval)
363                 security_sb_umount_busy(mnt);
364         up_write(&current->namespace->sem);
365         return retval;
366 }
367 
368 /*
369  * Now umount can handle mount points as well as block devices.
370  * This is important for filesystems which use unnamed block devices.
371  *
372  * We now support a flag for forced unmount like the other 'big iron'
373  * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
374  */
375 
376 asmlinkage long sys_umount(char __user * name, int flags)
377 {
378         struct nameidata nd;
379         int retval;
380 
381         retval = __user_walk(name, LOOKUP_FOLLOW, &nd);
382         if (retval)
383                 goto out;
384         retval = -EINVAL;
385         if (nd.dentry != nd.mnt->mnt_root)
386                 goto dput_and_out;
387         if (!check_mnt(nd.mnt))
388                 goto dput_and_out;
389 
390         retval = -EPERM;
391         if (!capable(CAP_SYS_ADMIN))
392                 goto dput_and_out;
393 
394         retval = do_umount(nd.mnt, flags);
395 dput_and_out:
396         path_release(&nd);
397 out:
398         return retval;
399 }
400 
401 /*
402  *      The 2.0 compatible umount. No flags. 
403  */
404  
405 asmlinkage long sys_oldumount(char __user * name)
406 {
407         return sys_umount(name,0);
408 }
409 
410 static int mount_is_safe(struct nameidata *nd)
411 {
412         if (capable(CAP_SYS_ADMIN))
413                 return 0;
414         return -EPERM;
415 #ifdef notyet
416         if (S_ISLNK(nd->dentry->d_inode->i_mode))
417                 return -EPERM;
418         if (nd->dentry->d_inode->i_mode & S_ISVTX) {
419                 if (current->uid != nd->dentry->d_inode->i_uid)
420                         return -EPERM;
421         }
422         if (permission(nd->dentry->d_inode, MAY_WRITE, nd))
423                 return -EPERM;
424         return 0;
425 #endif
426 }
427 
428 static int
429 lives_below_in_same_fs(struct dentry *d, struct dentry *dentry)
430 {
431         while (1) {
432                 if (d == dentry)
433                         return 1;
434                 if (d == NULL || d == d->d_parent)
435                         return 0;
436                 d = d->d_parent;
437         }
438 }
439 
440 static struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry)
441 {
442         struct vfsmount *res, *p, *q, *r, *s;
443         struct list_head *h;
444         struct nameidata nd;
445 
446         res = q = clone_mnt(mnt, dentry);
447         if (!q)
448                 goto Enomem;
449         q->mnt_mountpoint = mnt->mnt_mountpoint;
450 
451         p = mnt;
452         for (h = mnt->mnt_mounts.next; h != &mnt->mnt_mounts; h = h->next) {
453                 r = list_entry(h, struct vfsmount, mnt_child);
454                 if (!lives_below_in_same_fs(r->mnt_mountpoint, dentry))
455                         continue;
456 
457                 for (s = r; s; s = next_mnt(s, r)) {
458                         while (p != s->mnt_parent) {
459                                 p = p->mnt_parent;
460                                 q = q->mnt_parent;
461                         }
462                         p = s;
463                         nd.mnt = q;
464                         nd.dentry = p->mnt_mountpoint;
465                         q = clone_mnt(p, p->mnt_root);
466                         if (!q)
467                                 goto Enomem;
468                         spin_lock(&vfsmount_lock);
469                         list_add_tail(&q->mnt_list, &res->mnt_list);
470                         attach_mnt(q, &nd);
471                         spin_unlock(&vfsmount_lock);
472                 }
473         }
474         return res;
475  Enomem:
476         if (res) {
477                 spin_lock(&vfsmount_lock);
478                 umount_tree(res);
479                 spin_unlock(&vfsmount_lock);
480         }
481         return NULL;
482 }
483 
484 static int graft_tree(struct vfsmount *mnt, struct nameidata *nd)
485 {
486         int err;
487         if (mnt->mnt_sb->s_flags & MS_NOUSER)
488                 return -EINVAL;
489 
490         if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
491               S_ISDIR(mnt->mnt_root->d_inode->i_mode))
492                 return -ENOTDIR;
493 
494         err = -ENOENT;
495         down(&nd->dentry->d_inode->i_sem);
496         if (IS_DEADDIR(nd->dentry->d_inode))
497                 goto out_unlock;
498 
499         err = security_sb_check_sb(mnt, nd);
500         if (err)
501                 goto out_unlock;
502 
503         err = -ENOENT;
504         spin_lock(&vfsmount_lock);
505         if (IS_ROOT(nd->dentry) || !d_unhashed(nd->dentry)) {
506                 struct list_head head;
507 
508                 attach_mnt(mnt, nd);
509                 list_add_tail(&head, &mnt->mnt_list);
510                 list_splice(&head, current->namespace->list.prev);
511                 mntget(mnt);
512                 err = 0;
513         }
514         spin_unlock(&vfsmount_lock);
515 out_unlock:
516         up(&nd->dentry->d_inode->i_sem);
517         if (!err)
518                 security_sb_post_addmount(mnt, nd);
519         return err;
520 }
521 
522 /*
523  * do loopback mount.
524  */
525 static int do_loopback(struct nameidata *nd, char *old_name, int recurse)
526 {
527         struct nameidata old_nd;
528         struct vfsmount *mnt = NULL;
529         int err = mount_is_safe(nd);
530         if (err)
531                 return err;
532         if (!old_name || !*old_name)
533                 return -EINVAL;
534         err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
535         if (err)
536                 return err;
537 
538         down_write(&current->namespace->sem);
539         err = -EINVAL;
540         if (check_mnt(nd->mnt) && (!recurse || check_mnt(old_nd.mnt))) {
541                 err = -ENOMEM;
542                 if (recurse)
543                         mnt = copy_tree(old_nd.mnt, old_nd.dentry);
544                 else
545                         mnt = clone_mnt(old_nd.mnt, old_nd.dentry);
546         }
547 
548         if (mnt) {
549                 err = graft_tree(mnt, nd);
550                 if (err) {
551                         spin_lock(&vfsmount_lock);
552                         umount_tree(mnt);
553                         spin_unlock(&vfsmount_lock);
554                 } else
555                         mntput(mnt);
556         }
557 
558         up_write(&current->namespace->sem);
559         path_release(&old_nd);
560         return err;
561 }
562 
563 /*
564  * change filesystem flags. dir should be a physical root of filesystem.
565  * If you've mounted a non-root directory somewhere and want to do remount
566  * on it - tough luck.
567  */
568 
569 static int do_remount(struct nameidata *nd,int flags,int mnt_flags,void *data)
570 {
571         int err;
572         struct super_block * sb = nd->mnt->mnt_sb;
573 
574         if (!capable(CAP_SYS_ADMIN))
575                 return -EPERM;
576 
577         if (!check_mnt(nd->mnt))
578                 return -EINVAL;
579 
580         if (nd->dentry != nd->mnt->mnt_root)
581                 return -EINVAL;
582 
583         down_write(&sb->s_umount);
584         err = do_remount_sb(sb, flags, data, 0);
585         if (!err)
586                 nd->mnt->mnt_flags=mnt_flags;
587         up_write(&sb->s_umount);
588         if (!err)
589                 security_sb_post_remount(nd->mnt, flags, data);
590         return err;
591 }
592 
593 static int do_move_mount(struct nameidata *nd, char *old_name)
594 {
595         struct nameidata old_nd, parent_nd;
596         struct vfsmount *p;
597         int err = 0;
598         if (!capable(CAP_SYS_ADMIN))
599                 return -EPERM;
600         if (!old_name || !*old_name)
601                 return -EINVAL;
602         err = path_lookup(old_name, LOOKUP_FOLLOW, &old_nd);
603         if (err)
604                 return err;
605 
606         down_write(&current->namespace->sem);
607         while(d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
608                 ;
609         err = -EINVAL;
610         if (!check_mnt(nd->mnt) || !check_mnt(old_nd.mnt))
611                 goto out;
612 
613         err = -ENOENT;
614         down(&nd->dentry->d_inode->i_sem);
615         if (IS_DEADDIR(nd->dentry->d_inode))
616                 goto out1;
617 
618         spin_lock(&vfsmount_lock);
619         if (!IS_ROOT(nd->dentry) && d_unhashed(nd->dentry))
620                 goto out2;
621 
622         err = -EINVAL;
623         if (old_nd.dentry != old_nd.mnt->mnt_root)
624                 goto out2;
625 
626         if (old_nd.mnt == old_nd.mnt->mnt_parent)
627                 goto out2;
628 
629         if (S_ISDIR(nd->dentry->d_inode->i_mode) !=
630               S_ISDIR(old_nd.dentry->d_inode->i_mode))
631                 goto out2;
632 
633         err = -ELOOP;
634         for (p = nd->mnt; p->mnt_parent!=p; p = p->mnt_parent)
635                 if (p == old_nd.mnt)
636                         goto out2;
637         err = 0;
638 
639         detach_mnt(old_nd.mnt, &parent_nd);
640         attach_mnt(old_nd.mnt, nd);
641 out2:
642         spin_unlock(&vfsmount_lock);
643 out1:
644         up(&nd->dentry->d_inode->i_sem);
645 out:
646         up_write(&current->namespace->sem);
647         if (!err)
648                 path_release(&parent_nd);
649         path_release(&old_nd);
650         return err;
651 }
652 
653 static int do_add_mount(struct nameidata *nd, char *type, int flags,
654                         int mnt_flags, char *name, void *data)
655 {
656         struct vfsmount *mnt;
657         int err;
658 
659         if (!type || !memchr(type, 0, PAGE_SIZE))
660                 return -EINVAL;
661 
662         /* we need capabilities... */
663         if (!capable(CAP_SYS_ADMIN))
664                 return -EPERM;
665 
666         mnt = do_kern_mount(type, flags, name, data);
667         err = PTR_ERR(mnt);
668         if (IS_ERR(mnt))
669                 goto out;
670 
671         down_write(&current->namespace->sem);
672         /* Something was mounted here while we slept */
673         while(d_mountpoint(nd->dentry) && follow_down(&nd->mnt, &nd->dentry))
674                 ;
675         err = -EINVAL;
676         if (!check_mnt(nd->mnt))
677                 goto unlock;
678 
679         /* Refuse the same filesystem on the same mount point */
680         err = -EBUSY;
681         if (nd->mnt->mnt_sb == mnt->mnt_sb && nd->mnt->mnt_root == nd->dentry)
682                 goto unlock;
683 
684         mnt->mnt_flags = mnt_flags;
685         err = graft_tree(mnt, nd);
686 unlock:
687         up_write(&current->namespace->sem);
688         mntput(mnt);
689 out:
690         return err;
691 }
692 
693 static int copy_mount_options (const void __user *data, unsigned long *where)
694 {
695         int i;
696         unsigned long page;
697         unsigned long size;
698         
699         *where = 0;
700         if (!data)
701                 return 0;
702 
703         if (!(page = __get_free_page(GFP_KERNEL)))
704                 return -ENOMEM;
705 
706         /* We only care that *some* data at the address the user
707          * gave us is valid.  Just in case, we'll zero
708          * the remainder of the page.
709          */
710         /* copy_from_user cannot cross TASK_SIZE ! */
711         size = TASK_SIZE - (unsigned long)data;
712         if (size > PAGE_SIZE)
713                 size = PAGE_SIZE;
714 
715         i = size - copy_from_user((void *)page, data, size);
716         if (!i) {
717                 free_page(page); 
718                 return -EFAULT;
719         }
720         if (i != PAGE_SIZE)
721                 memset((char *)page + i, 0, PAGE_SIZE - i);
722         *where = page;
723         return 0;
724 }
725 
726 /*
727  * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
728  * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
729  *
730  * data is a (void *) that can point to any structure up to
731  * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
732  * information (or be NULL).
733  *
734  * Pre-0.97 versions of mount() didn't have a flags word.
735  * When the flags word was introduced its top half was required
736  * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
737  * Therefore, if this magic number is present, it carries no information
738  * and must be discarded.
739  */
740 long do_mount(char * dev_name, char * dir_name, char *type_page,
741                   unsigned long flags, void *data_page)
742 {
743         struct nameidata nd;
744         int retval = 0;
745         int mnt_flags = 0;
746 
747         /* Discard magic */
748         if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
749                 flags &= ~MS_MGC_MSK;
750 
751         /* Basic sanity checks */
752 
753         if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
754                 return -EINVAL;
755         if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
756                 return -EINVAL;
757 
758         /* Separate the per-mountpoint flags */
759         if (flags & MS_NOSUID)
760                 mnt_flags |= MNT_NOSUID;
761         if (flags & MS_NODEV)
762                 mnt_flags |= MNT_NODEV;
763         if (flags & MS_NOEXEC)
764                 mnt_flags |= MNT_NOEXEC;
765         flags &= ~(MS_NOSUID|MS_NOEXEC|MS_NODEV);
766 
767         /* ... and get the mountpoint */
768         retval = path_lookup(dir_name, LOOKUP_FOLLOW, &nd);
769         if (retval)
770                 return retval;
771 
772         retval = security_sb_mount(dev_name, &nd, type_page, flags, data_page);
773         if (retval)
774                 goto dput_out;
775 
776         if (flags & MS_REMOUNT)
777                 retval = do_remount(&nd, flags & ~MS_REMOUNT, mnt_flags,
778                                     data_page);
779         else if (flags & MS_BIND)
780                 retval = do_loopback(&nd, dev_name, flags & MS_REC);
781         else if (flags & MS_MOVE)
782                 retval = do_move_mount(&nd, dev_name);
783         else
784                 retval = do_add_mount(&nd, type_page, flags, mnt_flags,
785                                       dev_name, data_page);
786 dput_out:
787         path_release(&nd);
788         return retval;
789 }
790 
791 int copy_namespace(int flags, struct task_struct *tsk)
792 {
793         struct namespace *namespace = tsk->namespace;
794         struct namespace *new_ns;
795         struct vfsmount *rootmnt = NULL, *pwdmnt = NULL, *altrootmnt = NULL;
796         struct fs_struct *fs = tsk->fs;
797 
798         if (!namespace)
799                 return 0;
800 
801         get_namespace(namespace);
802 
803         if (!(flags & CLONE_NEWNS))
804                 return 0;
805 
806         if (!capable(CAP_SYS_ADMIN)) {
807                 put_namespace(namespace);
808                 return -EPERM;
809         }
810 
811         new_ns = kmalloc(sizeof(struct namespace), GFP_KERNEL);
812         if (!new_ns)
813                 goto out;
814 
815         atomic_set(&new_ns->count, 1);
816         init_rwsem(&new_ns->sem);
817         new_ns->root = NULL;
818         INIT_LIST_HEAD(&new_ns->list);
819 
820         down_write(&tsk->namespace->sem);
821         /* First pass: copy the tree topology */
822         new_ns->root = copy_tree(namespace->root, namespace->root->mnt_root);
823         spin_lock(&vfsmount_lock);
824         list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
825         spin_unlock(&vfsmount_lock);
826 
827         /* Second pass: switch the tsk->fs->* elements */
828         if (fs) {
829                 struct vfsmount *p, *q;
830                 write_lock(&fs->lock);
831 
832                 p = namespace->root;
833                 q = new_ns->root;
834                 while (p) {
835                         if (p == fs->rootmnt) {
836                                 rootmnt = p;
837                                 fs->rootmnt = mntget(q);
838                         }
839                         if (p == fs->pwdmnt) {
840                                 pwdmnt = p;
841                                 fs->pwdmnt = mntget(q);
842                         }
843                         if (p == fs->altrootmnt) {
844                                 altrootmnt = p;
845                                 fs->altrootmnt = mntget(q);
846                         }
847                         p = next_mnt(p, namespace->root);
848                         q = next_mnt(q, new_ns->root);
849                 }
850                 write_unlock(&fs->lock);
851         }
852         up_write(&tsk->namespace->sem);
853 
854         tsk->namespace = new_ns;
855 
856         if (rootmnt)
857                 mntput(rootmnt);
858         if (pwdmnt)
859                 mntput(pwdmnt);
860         if (altrootmnt)
861                 mntput(altrootmnt);
862 
863         put_namespace(namespace);
864         return 0;
865 
866 out:
867         put_namespace(namespace);
868         return -ENOMEM;
869 }
870 
871 asmlinkage long sys_mount(char __user * dev_name, char __user * dir_name,
872                           char __user * type, unsigned long flags,
873                           void __user * data)
874 {
875         int retval;
876         unsigned long data_page;
877         unsigned long type_page;
878         unsigned long dev_page;
879         char *dir_page;
880 
881         retval = copy_mount_options (type, &type_page);
882         if (retval < 0)
883                 return retval;
884 
885         dir_page = getname(dir_name);
886         retval = PTR_ERR(dir_page);
887         if (IS_ERR(dir_page))
888                 goto out1;
889 
890         retval = copy_mount_options (dev_name, &dev_page);
891         if (retval < 0)
892                 goto out2;
893 
894         retval = copy_mount_options (data, &data_page);
895         if (retval < 0)
896                 goto out3;
897 
898         lock_kernel();
899         retval = do_mount((char*)dev_page, dir_page, (char*)type_page,
900                           flags, (void*)data_page);
901         unlock_kernel();
902         free_page(data_page);
903 
904 out3:
905         free_page(dev_page);
906 out2:
907         putname(dir_page);
908 out1:
909         free_page(type_page);
910         return retval;
911 }
912 
913 /*
914  * Replace the fs->{rootmnt,root} with {mnt,dentry}. Put the old values.
915  * It can block. Requires the big lock held.
916  */
917 void set_fs_root(struct fs_struct *fs, struct vfsmount *mnt,
918                  struct dentry *dentry)
919 {
920         struct dentry *old_root;
921         struct vfsmount *old_rootmnt;
922         write_lock(&fs->lock);
923         old_root = fs->root;
924         old_rootmnt = fs->rootmnt;
925         fs->rootmnt = mntget(mnt);
926         fs->root = dget(dentry);
927         write_unlock(&fs->lock);
928         if (old_root) {
929                 dput(old_root);
930                 mntput(old_rootmnt);
931         }
932 }
933 
934 EXPORT_SYMBOL(set_fs_root);
935 
936 /*
937  * Replace the fs->{pwdmnt,pwd} with {mnt,dentry}. Put the old values.
938  * It can block. Requires the big lock held.
939  */
940 void set_fs_pwd(struct fs_struct *fs, struct vfsmount *mnt,
941                 struct dentry *dentry)
942 {
943         struct dentry *old_pwd;
944         struct vfsmount *old_pwdmnt;
945 
946         write_lock(&fs->lock);
947         old_pwd = fs->pwd;
948         old_pwdmnt = fs->pwdmnt;
949         fs->pwdmnt = mntget(mnt);
950         fs->pwd = dget(dentry);
951         write_unlock(&fs->lock);
952 
953         if (old_pwd) {
954                 dput(old_pwd);
955                 mntput(old_pwdmnt);
956         }
957 }
958 
959 EXPORT_SYMBOL(set_fs_pwd);
960 
961 static void chroot_fs_refs(struct nameidata *old_nd, struct nameidata *new_nd)
962 {
963         struct task_struct *g, *p;
964         struct fs_struct *fs;
965 
966         read_lock(&tasklist_lock);
967         do_each_thread(g, p) {
968                 task_lock(p);
969                 fs = p->fs;
970                 if (fs) {
971                         atomic_inc(&fs->count);
972                         task_unlock(p);
973                         if (fs->root==old_nd->dentry&&fs->rootmnt==old_nd->mnt)
974                                 set_fs_root(fs, new_nd->mnt, new_nd->dentry);
975                         if (fs->pwd==old_nd->dentry&&fs->pwdmnt==old_nd->mnt)
976                                 set_fs_pwd(fs, new_nd->mnt, new_nd->dentry);
977                         put_fs_struct(fs);
978                 } else
979                         task_unlock(p);
980         } while_each_thread(g, p);
981         read_unlock(&tasklist_lock);
982 }
983 
984 /*
985  * Moves the current root to put_root, and sets root/cwd of all processes
986  * which had them on the old root to new_root.
987  *
988  * Note:
989  *  - we don't move root/cwd if they are not at the root (reason: if something
990  *    cared enough to change them, it's probably wrong to force them elsewhere)
991  *  - it's okay to pick a root that isn't the root of a file system, e.g.
992  *    /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
993  *    though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
994  *    first.
995  */
996 
997 asmlinkage long sys_pivot_root(const char __user *new_root, const char __user *put_old)
998 {
999         struct vfsmount *tmp;
1000         struct nameidata new_nd, old_nd, parent_nd, root_parent, user_nd;
1001         int error;
1002 
1003         if (!capable(CAP_SYS_ADMIN))
1004                 return -EPERM;
1005 
1006         lock_kernel();
1007 
1008         error = __user_walk(new_root, LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &new_nd);
1009         if (error)
1010                 goto out0;
1011         error = -EINVAL;
1012         if (!check_mnt(new_nd.mnt))
1013                 goto out1;
1014 
1015         error = __user_walk(put_old, LOOKUP_FOLLOW|LOOKUP_DIRECTORY, &old_nd);
1016         if (error)
1017                 goto out1;
1018 
1019         error = security_sb_pivotroot(&old_nd, &new_nd);
1020         if (error) {
1021                 path_release(&old_nd);
1022                 goto out1;
1023         }
1024 
1025         read_lock(&current->fs->lock);
1026         user_nd.mnt = mntget(current->fs->rootmnt);
1027         user_nd.dentry = dget(current->fs->root);
1028         read_unlock(&current->fs->lock);
1029         down_write(&current->namespace->sem);
1030         down(&old_nd.dentry->d_inode->i_sem);
1031         error = -EINVAL;
1032         if (!check_mnt(user_nd.mnt))
1033                 goto out2;
1034         error = -ENOENT;
1035         if (IS_DEADDIR(new_nd.dentry->d_inode))
1036                 goto out2;
1037         if (d_unhashed(new_nd.dentry) && !IS_ROOT(new_nd.dentry))
1038                 goto out2;
1039         if (d_unhashed(old_nd.dentry) && !IS_ROOT(old_nd.dentry))
1040                 goto out2;
1041         error = -EBUSY;
1042         if (new_nd.mnt == user_nd.mnt || old_nd.mnt == user_nd.mnt)
1043                 goto out2; /* loop */
1044         error = -EINVAL;
1045         if (user_nd.mnt->mnt_root != user_nd.dentry)
1046                 goto out2;
1047         if (new_nd.mnt->mnt_root != new_nd.dentry)
1048                 goto out2; /* not a mountpoint */
1049         tmp = old_nd.mnt; /* make sure we can reach put_old from new_root */
1050         spin_lock(&vfsmount_lock);
1051         if (tmp != new_nd.mnt) {
1052                 for (;;) {
1053                         if (tmp->mnt_parent == tmp)
1054                                 goto out3;
1055                         if (tmp->mnt_parent == new_nd.mnt)
1056                                 break;
1057                         tmp = tmp->mnt_parent;
1058                 }
1059                 if (!is_subdir(tmp->mnt_mountpoint, new_nd.dentry))
1060                         goto out3;
1061         } else if (!is_subdir(old_nd.dentry, new_nd.dentry))
1062                 goto out3;
1063         detach_mnt(new_nd.mnt, &parent_nd);
1064         detach_mnt(user_nd.mnt, &root_parent);
1065         attach_mnt(user_nd.mnt, &old_nd);
1066         attach_mnt(new_nd.mnt, &root_parent);
1067         spin_unlock(&vfsmount_lock);
1068         chroot_fs_refs(&user_nd, &new_nd);
1069         security_sb_post_pivotroot(&user_nd, &new_nd);
1070         error = 0;
1071         path_release(&root_parent);
1072         path_release(&parent_nd);
1073 out2:
1074         up(&old_nd.dentry->d_inode->i_sem);
1075         up_write(&current->namespace->sem);
1076         path_release(&user_nd);
1077         path_release(&old_nd);
1078 out1:
1079         path_release(&new_nd);
1080 out0:
1081         unlock_kernel();
1082         return error;
1083 out3:
1084         spin_unlock(&vfsmount_lock);
1085         goto out2;
1086 }
1087 
1088 static void __init init_mount_tree(void)
1089 {
1090         struct vfsmount *mnt;
1091         struct namespace *namespace;
1092         struct task_struct *g, *p;
1093 
1094         mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
1095         if (IS_ERR(mnt))
1096                 panic("Can't create rootfs");
1097         namespace = kmalloc(sizeof(*namespace), GFP_KERNEL);
1098         if (!namespace)
1099                 panic("Can't allocate initial namespace");
1100         atomic_set(&namespace->count, 1);
1101         INIT_LIST_HEAD(&namespace->list);
1102         init_rwsem(&namespace->sem);
1103         list_add(&mnt->mnt_list, &namespace->list);
1104         namespace->root = mnt;
1105 
1106         init_task.namespace = namespace;
1107         read_lock(&tasklist_lock);
1108         do_each_thread(g, p) {
1109                 get_namespace(namespace);
1110                 p->namespace = namespace;
1111         } while_each_thread(g, p);
1112         read_unlock(&tasklist_lock);
1113 
1114         set_fs_pwd(current->fs, namespace->root, namespace->root->mnt_root);
1115         set_fs_root(current->fs, namespace->root, namespace->root->mnt_root);
1116 }
1117 
1118 void __init mnt_init(unsigned long mempages)
1119 {
1120         struct list_head *d;
1121         unsigned long order;
1122         unsigned int nr_hash;
1123         int i;
1124 
1125         mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
1126                                         0, SLAB_HWCACHE_ALIGN, NULL, NULL);
1127         if (!mnt_cache)
1128                 panic("Cannot create vfsmount cache");
1129 
1130         order = 0; 
1131         mount_hashtable = (struct list_head *)
1132                 __get_free_pages(GFP_ATOMIC, order);
1133 
1134         if (!mount_hashtable)
1135                 panic("Failed to allocate mount hash table\n");
1136 
1137         /*
1138          * Find the power-of-two list-heads that can fit into the allocation..
1139          * We don't guarantee that "sizeof(struct list_head)" is necessarily
1140          * a power-of-two.
1141          */
1142         nr_hash = (1UL << order) * PAGE_SIZE / sizeof(struct list_head);
1143         hash_bits = 0;
1144         do {
1145                 hash_bits++;
1146         } while ((nr_hash >> hash_bits) != 0);
1147         hash_bits--;
1148 
1149         /*
1150          * Re-calculate the actual number of entries and the mask
1151          * from the number of bits we can fit.
1152          */
1153         nr_hash = 1UL << hash_bits;
1154         hash_mask = nr_hash-1;
1155 
1156         printk("Mount-cache hash table entries: %d (order: %ld, %ld bytes)\n",
1157                         nr_hash, order, (PAGE_SIZE << order));
1158 
1159         /* And initialize the newly allocated array */
1160         d = mount_hashtable;
1161         i = nr_hash;
1162         do {
1163                 INIT_LIST_HEAD(d);
1164                 d++;
1165                 i--;
1166         } while (i);
1167         sysfs_init();
1168         init_rootfs();
1169         init_mount_tree();
1170 }
1171 
1172 void __put_namespace(struct namespace *namespace)
1173 {
1174         down_write(&namespace->sem);
1175         spin_lock(&vfsmount_lock);
1176         umount_tree(namespace->root);
1177         spin_unlock(&vfsmount_lock);
1178         up_write(&namespace->sem);
1179         kfree(namespace);
1180 }
1181 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp