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

TOMOYO Linux Cross Reference
Linux/fs/namespace.c

Version: ~ [ linux-5.3-rc5 ] ~ [ linux-5.2.9 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.67 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.139 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.189 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.189 ] ~ [ 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-3.16.72 ] ~ [ 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-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ 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/syscalls.h>
 12 #include <linux/export.h>
 13 #include <linux/capability.h>
 14 #include <linux/mnt_namespace.h>
 15 #include <linux/user_namespace.h>
 16 #include <linux/namei.h>
 17 #include <linux/security.h>
 18 #include <linux/idr.h>
 19 #include <linux/acct.h>         /* acct_auto_close_mnt */
 20 #include <linux/ramfs.h>        /* init_rootfs */
 21 #include <linux/fs_struct.h>    /* get_fs_root et.al. */
 22 #include <linux/fsnotify.h>     /* fsnotify_vfsmount_delete */
 23 #include <linux/uaccess.h>
 24 #include <linux/proc_fs.h>
 25 #include "pnode.h"
 26 #include "internal.h"
 27 
 28 #define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
 29 #define HASH_SIZE (1UL << HASH_SHIFT)
 30 
 31 static int event;
 32 static DEFINE_IDA(mnt_id_ida);
 33 static DEFINE_IDA(mnt_group_ida);
 34 static DEFINE_SPINLOCK(mnt_id_lock);
 35 static int mnt_id_start = 0;
 36 static int mnt_group_start = 1;
 37 
 38 static struct list_head *mount_hashtable __read_mostly;
 39 static struct kmem_cache *mnt_cache __read_mostly;
 40 static struct rw_semaphore namespace_sem;
 41 
 42 /* /sys/fs */
 43 struct kobject *fs_kobj;
 44 EXPORT_SYMBOL_GPL(fs_kobj);
 45 
 46 /*
 47  * vfsmount lock may be taken for read to prevent changes to the
 48  * vfsmount hash, ie. during mountpoint lookups or walking back
 49  * up the tree.
 50  *
 51  * It should be taken for write in all cases where the vfsmount
 52  * tree or hash is modified or when a vfsmount structure is modified.
 53  */
 54 DEFINE_BRLOCK(vfsmount_lock);
 55 
 56 static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
 57 {
 58         unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
 59         tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
 60         tmp = tmp + (tmp >> HASH_SHIFT);
 61         return tmp & (HASH_SIZE - 1);
 62 }
 63 
 64 #define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
 65 
 66 /*
 67  * allocation is serialized by namespace_sem, but we need the spinlock to
 68  * serialize with freeing.
 69  */
 70 static int mnt_alloc_id(struct mount *mnt)
 71 {
 72         int res;
 73 
 74 retry:
 75         ida_pre_get(&mnt_id_ida, GFP_KERNEL);
 76         spin_lock(&mnt_id_lock);
 77         res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
 78         if (!res)
 79                 mnt_id_start = mnt->mnt_id + 1;
 80         spin_unlock(&mnt_id_lock);
 81         if (res == -EAGAIN)
 82                 goto retry;
 83 
 84         return res;
 85 }
 86 
 87 static void mnt_free_id(struct mount *mnt)
 88 {
 89         int id = mnt->mnt_id;
 90         spin_lock(&mnt_id_lock);
 91         ida_remove(&mnt_id_ida, id);
 92         if (mnt_id_start > id)
 93                 mnt_id_start = id;
 94         spin_unlock(&mnt_id_lock);
 95 }
 96 
 97 /*
 98  * Allocate a new peer group ID
 99  *
100  * mnt_group_ida is protected by namespace_sem
101  */
102 static int mnt_alloc_group_id(struct mount *mnt)
103 {
104         int res;
105 
106         if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
107                 return -ENOMEM;
108 
109         res = ida_get_new_above(&mnt_group_ida,
110                                 mnt_group_start,
111                                 &mnt->mnt_group_id);
112         if (!res)
113                 mnt_group_start = mnt->mnt_group_id + 1;
114 
115         return res;
116 }
117 
118 /*
119  * Release a peer group ID
120  */
121 void mnt_release_group_id(struct mount *mnt)
122 {
123         int id = mnt->mnt_group_id;
124         ida_remove(&mnt_group_ida, id);
125         if (mnt_group_start > id)
126                 mnt_group_start = id;
127         mnt->mnt_group_id = 0;
128 }
129 
130 /*
131  * vfsmount lock must be held for read
132  */
133 static inline void mnt_add_count(struct mount *mnt, int n)
134 {
135 #ifdef CONFIG_SMP
136         this_cpu_add(mnt->mnt_pcp->mnt_count, n);
137 #else
138         preempt_disable();
139         mnt->mnt_count += n;
140         preempt_enable();
141 #endif
142 }
143 
144 /*
145  * vfsmount lock must be held for write
146  */
147 unsigned int mnt_get_count(struct mount *mnt)
148 {
149 #ifdef CONFIG_SMP
150         unsigned int count = 0;
151         int cpu;
152 
153         for_each_possible_cpu(cpu) {
154                 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_count;
155         }
156 
157         return count;
158 #else
159         return mnt->mnt_count;
160 #endif
161 }
162 
163 static struct mount *alloc_vfsmnt(const char *name)
164 {
165         struct mount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
166         if (mnt) {
167                 int err;
168 
169                 err = mnt_alloc_id(mnt);
170                 if (err)
171                         goto out_free_cache;
172 
173                 if (name) {
174                         mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
175                         if (!mnt->mnt_devname)
176                                 goto out_free_id;
177                 }
178 
179 #ifdef CONFIG_SMP
180                 mnt->mnt_pcp = alloc_percpu(struct mnt_pcp);
181                 if (!mnt->mnt_pcp)
182                         goto out_free_devname;
183 
184                 this_cpu_add(mnt->mnt_pcp->mnt_count, 1);
185 #else
186                 mnt->mnt_count = 1;
187                 mnt->mnt_writers = 0;
188 #endif
189 
190                 INIT_LIST_HEAD(&mnt->mnt_hash);
191                 INIT_LIST_HEAD(&mnt->mnt_child);
192                 INIT_LIST_HEAD(&mnt->mnt_mounts);
193                 INIT_LIST_HEAD(&mnt->mnt_list);
194                 INIT_LIST_HEAD(&mnt->mnt_expire);
195                 INIT_LIST_HEAD(&mnt->mnt_share);
196                 INIT_LIST_HEAD(&mnt->mnt_slave_list);
197                 INIT_LIST_HEAD(&mnt->mnt_slave);
198 #ifdef CONFIG_FSNOTIFY
199                 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
200 #endif
201         }
202         return mnt;
203 
204 #ifdef CONFIG_SMP
205 out_free_devname:
206         kfree(mnt->mnt_devname);
207 #endif
208 out_free_id:
209         mnt_free_id(mnt);
210 out_free_cache:
211         kmem_cache_free(mnt_cache, mnt);
212         return NULL;
213 }
214 
215 /*
216  * Most r/o checks on a fs are for operations that take
217  * discrete amounts of time, like a write() or unlink().
218  * We must keep track of when those operations start
219  * (for permission checks) and when they end, so that
220  * we can determine when writes are able to occur to
221  * a filesystem.
222  */
223 /*
224  * __mnt_is_readonly: check whether a mount is read-only
225  * @mnt: the mount to check for its write status
226  *
227  * This shouldn't be used directly ouside of the VFS.
228  * It does not guarantee that the filesystem will stay
229  * r/w, just that it is right *now*.  This can not and
230  * should not be used in place of IS_RDONLY(inode).
231  * mnt_want/drop_write() will _keep_ the filesystem
232  * r/w.
233  */
234 int __mnt_is_readonly(struct vfsmount *mnt)
235 {
236         if (mnt->mnt_flags & MNT_READONLY)
237                 return 1;
238         if (mnt->mnt_sb->s_flags & MS_RDONLY)
239                 return 1;
240         return 0;
241 }
242 EXPORT_SYMBOL_GPL(__mnt_is_readonly);
243 
244 static inline void mnt_inc_writers(struct mount *mnt)
245 {
246 #ifdef CONFIG_SMP
247         this_cpu_inc(mnt->mnt_pcp->mnt_writers);
248 #else
249         mnt->mnt_writers++;
250 #endif
251 }
252 
253 static inline void mnt_dec_writers(struct mount *mnt)
254 {
255 #ifdef CONFIG_SMP
256         this_cpu_dec(mnt->mnt_pcp->mnt_writers);
257 #else
258         mnt->mnt_writers--;
259 #endif
260 }
261 
262 static unsigned int mnt_get_writers(struct mount *mnt)
263 {
264 #ifdef CONFIG_SMP
265         unsigned int count = 0;
266         int cpu;
267 
268         for_each_possible_cpu(cpu) {
269                 count += per_cpu_ptr(mnt->mnt_pcp, cpu)->mnt_writers;
270         }
271 
272         return count;
273 #else
274         return mnt->mnt_writers;
275 #endif
276 }
277 
278 static int mnt_is_readonly(struct vfsmount *mnt)
279 {
280         if (mnt->mnt_sb->s_readonly_remount)
281                 return 1;
282         /* Order wrt setting s_flags/s_readonly_remount in do_remount() */
283         smp_rmb();
284         return __mnt_is_readonly(mnt);
285 }
286 
287 /*
288  * Most r/o & frozen checks on a fs are for operations that take discrete
289  * amounts of time, like a write() or unlink().  We must keep track of when
290  * those operations start (for permission checks) and when they end, so that we
291  * can determine when writes are able to occur to a filesystem.
292  */
293 /**
294  * __mnt_want_write - get write access to a mount without freeze protection
295  * @m: the mount on which to take a write
296  *
297  * This tells the low-level filesystem that a write is about to be performed to
298  * it, and makes sure that writes are allowed (mnt it read-write) before
299  * returning success. This operation does not protect against filesystem being
300  * frozen. When the write operation is finished, __mnt_drop_write() must be
301  * called. This is effectively a refcount.
302  */
303 int __mnt_want_write(struct vfsmount *m)
304 {
305         struct mount *mnt = real_mount(m);
306         int ret = 0;
307 
308         preempt_disable();
309         mnt_inc_writers(mnt);
310         /*
311          * The store to mnt_inc_writers must be visible before we pass
312          * MNT_WRITE_HOLD loop below, so that the slowpath can see our
313          * incremented count after it has set MNT_WRITE_HOLD.
314          */
315         smp_mb();
316         while (ACCESS_ONCE(mnt->mnt.mnt_flags) & MNT_WRITE_HOLD)
317                 cpu_relax();
318         /*
319          * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
320          * be set to match its requirements. So we must not load that until
321          * MNT_WRITE_HOLD is cleared.
322          */
323         smp_rmb();
324         if (mnt_is_readonly(m)) {
325                 mnt_dec_writers(mnt);
326                 ret = -EROFS;
327         }
328         preempt_enable();
329 
330         return ret;
331 }
332 
333 /**
334  * mnt_want_write - get write access to a mount
335  * @m: the mount on which to take a write
336  *
337  * This tells the low-level filesystem that a write is about to be performed to
338  * it, and makes sure that writes are allowed (mount is read-write, filesystem
339  * is not frozen) before returning success.  When the write operation is
340  * finished, mnt_drop_write() must be called.  This is effectively a refcount.
341  */
342 int mnt_want_write(struct vfsmount *m)
343 {
344         int ret;
345 
346         sb_start_write(m->mnt_sb);
347         ret = __mnt_want_write(m);
348         if (ret)
349                 sb_end_write(m->mnt_sb);
350         return ret;
351 }
352 EXPORT_SYMBOL_GPL(mnt_want_write);
353 
354 /**
355  * mnt_clone_write - get write access to a mount
356  * @mnt: the mount on which to take a write
357  *
358  * This is effectively like mnt_want_write, except
359  * it must only be used to take an extra write reference
360  * on a mountpoint that we already know has a write reference
361  * on it. This allows some optimisation.
362  *
363  * After finished, mnt_drop_write must be called as usual to
364  * drop the reference.
365  */
366 int mnt_clone_write(struct vfsmount *mnt)
367 {
368         /* superblock may be r/o */
369         if (__mnt_is_readonly(mnt))
370                 return -EROFS;
371         preempt_disable();
372         mnt_inc_writers(real_mount(mnt));
373         preempt_enable();
374         return 0;
375 }
376 EXPORT_SYMBOL_GPL(mnt_clone_write);
377 
378 /**
379  * __mnt_want_write_file - get write access to a file's mount
380  * @file: the file who's mount on which to take a write
381  *
382  * This is like __mnt_want_write, but it takes a file and can
383  * do some optimisations if the file is open for write already
384  */
385 int __mnt_want_write_file(struct file *file)
386 {
387         struct inode *inode = file->f_dentry->d_inode;
388 
389         if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
390                 return __mnt_want_write(file->f_path.mnt);
391         else
392                 return mnt_clone_write(file->f_path.mnt);
393 }
394 
395 /**
396  * mnt_want_write_file - get write access to a file's mount
397  * @file: the file who's mount on which to take a write
398  *
399  * This is like mnt_want_write, but it takes a file and can
400  * do some optimisations if the file is open for write already
401  */
402 int mnt_want_write_file(struct file *file)
403 {
404         int ret;
405 
406         sb_start_write(file->f_path.mnt->mnt_sb);
407         ret = __mnt_want_write_file(file);
408         if (ret)
409                 sb_end_write(file->f_path.mnt->mnt_sb);
410         return ret;
411 }
412 EXPORT_SYMBOL_GPL(mnt_want_write_file);
413 
414 /**
415  * __mnt_drop_write - give up write access to a mount
416  * @mnt: the mount on which to give up write access
417  *
418  * Tells the low-level filesystem that we are done
419  * performing writes to it.  Must be matched with
420  * __mnt_want_write() call above.
421  */
422 void __mnt_drop_write(struct vfsmount *mnt)
423 {
424         preempt_disable();
425         mnt_dec_writers(real_mount(mnt));
426         preempt_enable();
427 }
428 
429 /**
430  * mnt_drop_write - give up write access to a mount
431  * @mnt: the mount on which to give up write access
432  *
433  * Tells the low-level filesystem that we are done performing writes to it and
434  * also allows filesystem to be frozen again.  Must be matched with
435  * mnt_want_write() call above.
436  */
437 void mnt_drop_write(struct vfsmount *mnt)
438 {
439         __mnt_drop_write(mnt);
440         sb_end_write(mnt->mnt_sb);
441 }
442 EXPORT_SYMBOL_GPL(mnt_drop_write);
443 
444 void __mnt_drop_write_file(struct file *file)
445 {
446         __mnt_drop_write(file->f_path.mnt);
447 }
448 
449 void mnt_drop_write_file(struct file *file)
450 {
451         mnt_drop_write(file->f_path.mnt);
452 }
453 EXPORT_SYMBOL(mnt_drop_write_file);
454 
455 static int mnt_make_readonly(struct mount *mnt)
456 {
457         int ret = 0;
458 
459         br_write_lock(&vfsmount_lock);
460         mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
461         /*
462          * After storing MNT_WRITE_HOLD, we'll read the counters. This store
463          * should be visible before we do.
464          */
465         smp_mb();
466 
467         /*
468          * With writers on hold, if this value is zero, then there are
469          * definitely no active writers (although held writers may subsequently
470          * increment the count, they'll have to wait, and decrement it after
471          * seeing MNT_READONLY).
472          *
473          * It is OK to have counter incremented on one CPU and decremented on
474          * another: the sum will add up correctly. The danger would be when we
475          * sum up each counter, if we read a counter before it is incremented,
476          * but then read another CPU's count which it has been subsequently
477          * decremented from -- we would see more decrements than we should.
478          * MNT_WRITE_HOLD protects against this scenario, because
479          * mnt_want_write first increments count, then smp_mb, then spins on
480          * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
481          * we're counting up here.
482          */
483         if (mnt_get_writers(mnt) > 0)
484                 ret = -EBUSY;
485         else
486                 mnt->mnt.mnt_flags |= MNT_READONLY;
487         /*
488          * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
489          * that become unheld will see MNT_READONLY.
490          */
491         smp_wmb();
492         mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
493         br_write_unlock(&vfsmount_lock);
494         return ret;
495 }
496 
497 static void __mnt_unmake_readonly(struct mount *mnt)
498 {
499         br_write_lock(&vfsmount_lock);
500         mnt->mnt.mnt_flags &= ~MNT_READONLY;
501         br_write_unlock(&vfsmount_lock);
502 }
503 
504 int sb_prepare_remount_readonly(struct super_block *sb)
505 {
506         struct mount *mnt;
507         int err = 0;
508 
509         /* Racy optimization.  Recheck the counter under MNT_WRITE_HOLD */
510         if (atomic_long_read(&sb->s_remove_count))
511                 return -EBUSY;
512 
513         br_write_lock(&vfsmount_lock);
514         list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
515                 if (!(mnt->mnt.mnt_flags & MNT_READONLY)) {
516                         mnt->mnt.mnt_flags |= MNT_WRITE_HOLD;
517                         smp_mb();
518                         if (mnt_get_writers(mnt) > 0) {
519                                 err = -EBUSY;
520                                 break;
521                         }
522                 }
523         }
524         if (!err && atomic_long_read(&sb->s_remove_count))
525                 err = -EBUSY;
526 
527         if (!err) {
528                 sb->s_readonly_remount = 1;
529                 smp_wmb();
530         }
531         list_for_each_entry(mnt, &sb->s_mounts, mnt_instance) {
532                 if (mnt->mnt.mnt_flags & MNT_WRITE_HOLD)
533                         mnt->mnt.mnt_flags &= ~MNT_WRITE_HOLD;
534         }
535         br_write_unlock(&vfsmount_lock);
536 
537         return err;
538 }
539 
540 static void free_vfsmnt(struct mount *mnt)
541 {
542         kfree(mnt->mnt_devname);
543         mnt_free_id(mnt);
544 #ifdef CONFIG_SMP
545         free_percpu(mnt->mnt_pcp);
546 #endif
547         kmem_cache_free(mnt_cache, mnt);
548 }
549 
550 /*
551  * find the first or last mount at @dentry on vfsmount @mnt depending on
552  * @dir. If @dir is set return the first mount else return the last mount.
553  * vfsmount_lock must be held for read or write.
554  */
555 struct mount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
556                               int dir)
557 {
558         struct list_head *head = mount_hashtable + hash(mnt, dentry);
559         struct list_head *tmp = head;
560         struct mount *p, *found = NULL;
561 
562         for (;;) {
563                 tmp = dir ? tmp->next : tmp->prev;
564                 p = NULL;
565                 if (tmp == head)
566                         break;
567                 p = list_entry(tmp, struct mount, mnt_hash);
568                 if (&p->mnt_parent->mnt == mnt && p->mnt_mountpoint == dentry) {
569                         found = p;
570                         break;
571                 }
572         }
573         return found;
574 }
575 
576 /*
577  * lookup_mnt - Return the first child mount mounted at path
578  *
579  * "First" means first mounted chronologically.  If you create the
580  * following mounts:
581  *
582  * mount /dev/sda1 /mnt
583  * mount /dev/sda2 /mnt
584  * mount /dev/sda3 /mnt
585  *
586  * Then lookup_mnt() on the base /mnt dentry in the root mount will
587  * return successively the root dentry and vfsmount of /dev/sda1, then
588  * /dev/sda2, then /dev/sda3, then NULL.
589  *
590  * lookup_mnt takes a reference to the found vfsmount.
591  */
592 struct vfsmount *lookup_mnt(struct path *path)
593 {
594         struct mount *child_mnt;
595 
596         br_read_lock(&vfsmount_lock);
597         child_mnt = __lookup_mnt(path->mnt, path->dentry, 1);
598         if (child_mnt) {
599                 mnt_add_count(child_mnt, 1);
600                 br_read_unlock(&vfsmount_lock);
601                 return &child_mnt->mnt;
602         } else {
603                 br_read_unlock(&vfsmount_lock);
604                 return NULL;
605         }
606 }
607 
608 static inline int check_mnt(struct mount *mnt)
609 {
610         return mnt->mnt_ns == current->nsproxy->mnt_ns;
611 }
612 
613 /*
614  * vfsmount lock must be held for write
615  */
616 static void touch_mnt_namespace(struct mnt_namespace *ns)
617 {
618         if (ns) {
619                 ns->event = ++event;
620                 wake_up_interruptible(&ns->poll);
621         }
622 }
623 
624 /*
625  * vfsmount lock must be held for write
626  */
627 static void __touch_mnt_namespace(struct mnt_namespace *ns)
628 {
629         if (ns && ns->event != event) {
630                 ns->event = event;
631                 wake_up_interruptible(&ns->poll);
632         }
633 }
634 
635 /*
636  * Clear dentry's mounted state if it has no remaining mounts.
637  * vfsmount_lock must be held for write.
638  */
639 static void dentry_reset_mounted(struct dentry *dentry)
640 {
641         unsigned u;
642 
643         for (u = 0; u < HASH_SIZE; u++) {
644                 struct mount *p;
645 
646                 list_for_each_entry(p, &mount_hashtable[u], mnt_hash) {
647                         if (p->mnt_mountpoint == dentry)
648                                 return;
649                 }
650         }
651         spin_lock(&dentry->d_lock);
652         dentry->d_flags &= ~DCACHE_MOUNTED;
653         spin_unlock(&dentry->d_lock);
654 }
655 
656 /*
657  * vfsmount lock must be held for write
658  */
659 static void detach_mnt(struct mount *mnt, struct path *old_path)
660 {
661         old_path->dentry = mnt->mnt_mountpoint;
662         old_path->mnt = &mnt->mnt_parent->mnt;
663         mnt->mnt_parent = mnt;
664         mnt->mnt_mountpoint = mnt->mnt.mnt_root;
665         list_del_init(&mnt->mnt_child);
666         list_del_init(&mnt->mnt_hash);
667         dentry_reset_mounted(old_path->dentry);
668 }
669 
670 /*
671  * vfsmount lock must be held for write
672  */
673 void mnt_set_mountpoint(struct mount *mnt, struct dentry *dentry,
674                         struct mount *child_mnt)
675 {
676         mnt_add_count(mnt, 1);  /* essentially, that's mntget */
677         child_mnt->mnt_mountpoint = dget(dentry);
678         child_mnt->mnt_parent = mnt;
679         spin_lock(&dentry->d_lock);
680         dentry->d_flags |= DCACHE_MOUNTED;
681         spin_unlock(&dentry->d_lock);
682 }
683 
684 /*
685  * vfsmount lock must be held for write
686  */
687 static void attach_mnt(struct mount *mnt, struct path *path)
688 {
689         mnt_set_mountpoint(real_mount(path->mnt), path->dentry, mnt);
690         list_add_tail(&mnt->mnt_hash, mount_hashtable +
691                         hash(path->mnt, path->dentry));
692         list_add_tail(&mnt->mnt_child, &real_mount(path->mnt)->mnt_mounts);
693 }
694 
695 /*
696  * vfsmount lock must be held for write
697  */
698 static void commit_tree(struct mount *mnt)
699 {
700         struct mount *parent = mnt->mnt_parent;
701         struct mount *m;
702         LIST_HEAD(head);
703         struct mnt_namespace *n = parent->mnt_ns;
704 
705         BUG_ON(parent == mnt);
706 
707         list_add_tail(&head, &mnt->mnt_list);
708         list_for_each_entry(m, &head, mnt_list)
709                 m->mnt_ns = n;
710 
711         list_splice(&head, n->list.prev);
712 
713         list_add_tail(&mnt->mnt_hash, mount_hashtable +
714                                 hash(&parent->mnt, mnt->mnt_mountpoint));
715         list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
716         touch_mnt_namespace(n);
717 }
718 
719 static struct mount *next_mnt(struct mount *p, struct mount *root)
720 {
721         struct list_head *next = p->mnt_mounts.next;
722         if (next == &p->mnt_mounts) {
723                 while (1) {
724                         if (p == root)
725                                 return NULL;
726                         next = p->mnt_child.next;
727                         if (next != &p->mnt_parent->mnt_mounts)
728                                 break;
729                         p = p->mnt_parent;
730                 }
731         }
732         return list_entry(next, struct mount, mnt_child);
733 }
734 
735 static struct mount *skip_mnt_tree(struct mount *p)
736 {
737         struct list_head *prev = p->mnt_mounts.prev;
738         while (prev != &p->mnt_mounts) {
739                 p = list_entry(prev, struct mount, mnt_child);
740                 prev = p->mnt_mounts.prev;
741         }
742         return p;
743 }
744 
745 struct vfsmount *
746 vfs_kern_mount(struct file_system_type *type, int flags, const char *name, void *data)
747 {
748         struct mount *mnt;
749         struct dentry *root;
750 
751         if (!type)
752                 return ERR_PTR(-ENODEV);
753 
754         mnt = alloc_vfsmnt(name);
755         if (!mnt)
756                 return ERR_PTR(-ENOMEM);
757 
758         if (flags & MS_KERNMOUNT)
759                 mnt->mnt.mnt_flags = MNT_INTERNAL;
760 
761         root = mount_fs(type, flags, name, data);
762         if (IS_ERR(root)) {
763                 free_vfsmnt(mnt);
764                 return ERR_CAST(root);
765         }
766 
767         mnt->mnt.mnt_root = root;
768         mnt->mnt.mnt_sb = root->d_sb;
769         mnt->mnt_mountpoint = mnt->mnt.mnt_root;
770         mnt->mnt_parent = mnt;
771         br_write_lock(&vfsmount_lock);
772         list_add_tail(&mnt->mnt_instance, &root->d_sb->s_mounts);
773         br_write_unlock(&vfsmount_lock);
774         return &mnt->mnt;
775 }
776 EXPORT_SYMBOL_GPL(vfs_kern_mount);
777 
778 static struct mount *clone_mnt(struct mount *old, struct dentry *root,
779                                         int flag)
780 {
781         struct super_block *sb = old->mnt.mnt_sb;
782         struct mount *mnt;
783         int err;
784 
785         mnt = alloc_vfsmnt(old->mnt_devname);
786         if (!mnt)
787                 return ERR_PTR(-ENOMEM);
788 
789         if (flag & (CL_SLAVE | CL_PRIVATE | CL_SHARED_TO_SLAVE))
790                 mnt->mnt_group_id = 0; /* not a peer of original */
791         else
792                 mnt->mnt_group_id = old->mnt_group_id;
793 
794         if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
795                 err = mnt_alloc_group_id(mnt);
796                 if (err)
797                         goto out_free;
798         }
799 
800         mnt->mnt.mnt_flags = old->mnt.mnt_flags & ~MNT_WRITE_HOLD;
801         /* Don't allow unprivileged users to change mount flags */
802         if ((flag & CL_UNPRIVILEGED) && (mnt->mnt.mnt_flags & MNT_READONLY))
803                 mnt->mnt.mnt_flags |= MNT_LOCK_READONLY;
804 
805         atomic_inc(&sb->s_active);
806         mnt->mnt.mnt_sb = sb;
807         mnt->mnt.mnt_root = dget(root);
808         mnt->mnt_mountpoint = mnt->mnt.mnt_root;
809         mnt->mnt_parent = mnt;
810         br_write_lock(&vfsmount_lock);
811         list_add_tail(&mnt->mnt_instance, &sb->s_mounts);
812         br_write_unlock(&vfsmount_lock);
813 
814         if ((flag & CL_SLAVE) ||
815             ((flag & CL_SHARED_TO_SLAVE) && IS_MNT_SHARED(old))) {
816                 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
817                 mnt->mnt_master = old;
818                 CLEAR_MNT_SHARED(mnt);
819         } else if (!(flag & CL_PRIVATE)) {
820                 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
821                         list_add(&mnt->mnt_share, &old->mnt_share);
822                 if (IS_MNT_SLAVE(old))
823                         list_add(&mnt->mnt_slave, &old->mnt_slave);
824                 mnt->mnt_master = old->mnt_master;
825         }
826         if (flag & CL_MAKE_SHARED)
827                 set_mnt_shared(mnt);
828 
829         /* stick the duplicate mount on the same expiry list
830          * as the original if that was on one */
831         if (flag & CL_EXPIRE) {
832                 if (!list_empty(&old->mnt_expire))
833                         list_add(&mnt->mnt_expire, &old->mnt_expire);
834         }
835 
836         return mnt;
837 
838  out_free:
839         free_vfsmnt(mnt);
840         return ERR_PTR(err);
841 }
842 
843 static inline void mntfree(struct mount *mnt)
844 {
845         struct vfsmount *m = &mnt->mnt;
846         struct super_block *sb = m->mnt_sb;
847 
848         /*
849          * This probably indicates that somebody messed
850          * up a mnt_want/drop_write() pair.  If this
851          * happens, the filesystem was probably unable
852          * to make r/w->r/o transitions.
853          */
854         /*
855          * The locking used to deal with mnt_count decrement provides barriers,
856          * so mnt_get_writers() below is safe.
857          */
858         WARN_ON(mnt_get_writers(mnt));
859         fsnotify_vfsmount_delete(m);
860         dput(m->mnt_root);
861         free_vfsmnt(mnt);
862         deactivate_super(sb);
863 }
864 
865 static void mntput_no_expire(struct mount *mnt)
866 {
867 put_again:
868 #ifdef CONFIG_SMP
869         br_read_lock(&vfsmount_lock);
870         if (likely(mnt->mnt_ns)) {
871                 /* shouldn't be the last one */
872                 mnt_add_count(mnt, -1);
873                 br_read_unlock(&vfsmount_lock);
874                 return;
875         }
876         br_read_unlock(&vfsmount_lock);
877 
878         br_write_lock(&vfsmount_lock);
879         mnt_add_count(mnt, -1);
880         if (mnt_get_count(mnt)) {
881                 br_write_unlock(&vfsmount_lock);
882                 return;
883         }
884 #else
885         mnt_add_count(mnt, -1);
886         if (likely(mnt_get_count(mnt)))
887                 return;
888         br_write_lock(&vfsmount_lock);
889 #endif
890         if (unlikely(mnt->mnt_pinned)) {
891                 mnt_add_count(mnt, mnt->mnt_pinned + 1);
892                 mnt->mnt_pinned = 0;
893                 br_write_unlock(&vfsmount_lock);
894                 acct_auto_close_mnt(&mnt->mnt);
895                 goto put_again;
896         }
897 
898         list_del(&mnt->mnt_instance);
899         br_write_unlock(&vfsmount_lock);
900         mntfree(mnt);
901 }
902 
903 void mntput(struct vfsmount *mnt)
904 {
905         if (mnt) {
906                 struct mount *m = real_mount(mnt);
907                 /* avoid cacheline pingpong, hope gcc doesn't get "smart" */
908                 if (unlikely(m->mnt_expiry_mark))
909                         m->mnt_expiry_mark = 0;
910                 mntput_no_expire(m);
911         }
912 }
913 EXPORT_SYMBOL(mntput);
914 
915 struct vfsmount *mntget(struct vfsmount *mnt)
916 {
917         if (mnt)
918                 mnt_add_count(real_mount(mnt), 1);
919         return mnt;
920 }
921 EXPORT_SYMBOL(mntget);
922 
923 void mnt_pin(struct vfsmount *mnt)
924 {
925         br_write_lock(&vfsmount_lock);
926         real_mount(mnt)->mnt_pinned++;
927         br_write_unlock(&vfsmount_lock);
928 }
929 EXPORT_SYMBOL(mnt_pin);
930 
931 void mnt_unpin(struct vfsmount *m)
932 {
933         struct mount *mnt = real_mount(m);
934         br_write_lock(&vfsmount_lock);
935         if (mnt->mnt_pinned) {
936                 mnt_add_count(mnt, 1);
937                 mnt->mnt_pinned--;
938         }
939         br_write_unlock(&vfsmount_lock);
940 }
941 EXPORT_SYMBOL(mnt_unpin);
942 
943 static inline void mangle(struct seq_file *m, const char *s)
944 {
945         seq_escape(m, s, " \t\n\\");
946 }
947 
948 /*
949  * Simple .show_options callback for filesystems which don't want to
950  * implement more complex mount option showing.
951  *
952  * See also save_mount_options().
953  */
954 int generic_show_options(struct seq_file *m, struct dentry *root)
955 {
956         const char *options;
957 
958         rcu_read_lock();
959         options = rcu_dereference(root->d_sb->s_options);
960 
961         if (options != NULL && options[0]) {
962                 seq_putc(m, ',');
963                 mangle(m, options);
964         }
965         rcu_read_unlock();
966 
967         return 0;
968 }
969 EXPORT_SYMBOL(generic_show_options);
970 
971 /*
972  * If filesystem uses generic_show_options(), this function should be
973  * called from the fill_super() callback.
974  *
975  * The .remount_fs callback usually needs to be handled in a special
976  * way, to make sure, that previous options are not overwritten if the
977  * remount fails.
978  *
979  * Also note, that if the filesystem's .remount_fs function doesn't
980  * reset all options to their default value, but changes only newly
981  * given options, then the displayed options will not reflect reality
982  * any more.
983  */
984 void save_mount_options(struct super_block *sb, char *options)
985 {
986         BUG_ON(sb->s_options);
987         rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
988 }
989 EXPORT_SYMBOL(save_mount_options);
990 
991 void replace_mount_options(struct super_block *sb, char *options)
992 {
993         char *old = sb->s_options;
994         rcu_assign_pointer(sb->s_options, options);
995         if (old) {
996                 synchronize_rcu();
997                 kfree(old);
998         }
999 }
1000 EXPORT_SYMBOL(replace_mount_options);
1001 
1002 #ifdef CONFIG_PROC_FS
1003 /* iterator; we want it to have access to namespace_sem, thus here... */
1004 static void *m_start(struct seq_file *m, loff_t *pos)
1005 {
1006         struct proc_mounts *p = proc_mounts(m);
1007 
1008         down_read(&namespace_sem);
1009         return seq_list_start(&p->ns->list, *pos);
1010 }
1011 
1012 static void *m_next(struct seq_file *m, void *v, loff_t *pos)
1013 {
1014         struct proc_mounts *p = proc_mounts(m);
1015 
1016         return seq_list_next(v, &p->ns->list, pos);
1017 }
1018 
1019 static void m_stop(struct seq_file *m, void *v)
1020 {
1021         up_read(&namespace_sem);
1022 }
1023 
1024 static int m_show(struct seq_file *m, void *v)
1025 {
1026         struct proc_mounts *p = proc_mounts(m);
1027         struct mount *r = list_entry(v, struct mount, mnt_list);
1028         return p->show(m, &r->mnt);
1029 }
1030 
1031 const struct seq_operations mounts_op = {
1032         .start  = m_start,
1033         .next   = m_next,
1034         .stop   = m_stop,
1035         .show   = m_show,
1036 };
1037 #endif  /* CONFIG_PROC_FS */
1038 
1039 /**
1040  * may_umount_tree - check if a mount tree is busy
1041  * @mnt: root of mount tree
1042  *
1043  * This is called to check if a tree of mounts has any
1044  * open files, pwds, chroots or sub mounts that are
1045  * busy.
1046  */
1047 int may_umount_tree(struct vfsmount *m)
1048 {
1049         struct mount *mnt = real_mount(m);
1050         int actual_refs = 0;
1051         int minimum_refs = 0;
1052         struct mount *p;
1053         BUG_ON(!m);
1054 
1055         /* write lock needed for mnt_get_count */
1056         br_write_lock(&vfsmount_lock);
1057         for (p = mnt; p; p = next_mnt(p, mnt)) {
1058                 actual_refs += mnt_get_count(p);
1059                 minimum_refs += 2;
1060         }
1061         br_write_unlock(&vfsmount_lock);
1062 
1063         if (actual_refs > minimum_refs)
1064                 return 0;
1065 
1066         return 1;
1067 }
1068 
1069 EXPORT_SYMBOL(may_umount_tree);
1070 
1071 /**
1072  * may_umount - check if a mount point is busy
1073  * @mnt: root of mount
1074  *
1075  * This is called to check if a mount point has any
1076  * open files, pwds, chroots or sub mounts. If the
1077  * mount has sub mounts this will return busy
1078  * regardless of whether the sub mounts are busy.
1079  *
1080  * Doesn't take quota and stuff into account. IOW, in some cases it will
1081  * give false negatives. The main reason why it's here is that we need
1082  * a non-destructive way to look for easily umountable filesystems.
1083  */
1084 int may_umount(struct vfsmount *mnt)
1085 {
1086         int ret = 1;
1087         down_read(&namespace_sem);
1088         br_write_lock(&vfsmount_lock);
1089         if (propagate_mount_busy(real_mount(mnt), 2))
1090                 ret = 0;
1091         br_write_unlock(&vfsmount_lock);
1092         up_read(&namespace_sem);
1093         return ret;
1094 }
1095 
1096 EXPORT_SYMBOL(may_umount);
1097 
1098 void release_mounts(struct list_head *head)
1099 {
1100         struct mount *mnt;
1101         while (!list_empty(head)) {
1102                 mnt = list_first_entry(head, struct mount, mnt_hash);
1103                 list_del_init(&mnt->mnt_hash);
1104                 if (mnt_has_parent(mnt)) {
1105                         struct dentry *dentry;
1106                         struct mount *m;
1107 
1108                         br_write_lock(&vfsmount_lock);
1109                         dentry = mnt->mnt_mountpoint;
1110                         m = mnt->mnt_parent;
1111                         mnt->mnt_mountpoint = mnt->mnt.mnt_root;
1112                         mnt->mnt_parent = mnt;
1113                         m->mnt_ghosts--;
1114                         br_write_unlock(&vfsmount_lock);
1115                         dput(dentry);
1116                         mntput(&m->mnt);
1117                 }
1118                 mntput(&mnt->mnt);
1119         }
1120 }
1121 
1122 /*
1123  * vfsmount lock must be held for write
1124  * namespace_sem must be held for write
1125  */
1126 void umount_tree(struct mount *mnt, int propagate, struct list_head *kill)
1127 {
1128         LIST_HEAD(tmp_list);
1129         struct mount *p;
1130 
1131         for (p = mnt; p; p = next_mnt(p, mnt))
1132                 list_move(&p->mnt_hash, &tmp_list);
1133 
1134         if (propagate)
1135                 propagate_umount(&tmp_list);
1136 
1137         list_for_each_entry(p, &tmp_list, mnt_hash) {
1138                 list_del_init(&p->mnt_expire);
1139                 list_del_init(&p->mnt_list);
1140                 __touch_mnt_namespace(p->mnt_ns);
1141                 p->mnt_ns = NULL;
1142                 list_del_init(&p->mnt_child);
1143                 if (mnt_has_parent(p)) {
1144                         p->mnt_parent->mnt_ghosts++;
1145                         dentry_reset_mounted(p->mnt_mountpoint);
1146                 }
1147                 change_mnt_propagation(p, MS_PRIVATE);
1148         }
1149         list_splice(&tmp_list, kill);
1150 }
1151 
1152 static void shrink_submounts(struct mount *mnt, struct list_head *umounts);
1153 
1154 static int do_umount(struct mount *mnt, int flags)
1155 {
1156         struct super_block *sb = mnt->mnt.mnt_sb;
1157         int retval;
1158         LIST_HEAD(umount_list);
1159 
1160         retval = security_sb_umount(&mnt->mnt, flags);
1161         if (retval)
1162                 return retval;
1163 
1164         /*
1165          * Allow userspace to request a mountpoint be expired rather than
1166          * unmounting unconditionally. Unmount only happens if:
1167          *  (1) the mark is already set (the mark is cleared by mntput())
1168          *  (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1169          */
1170         if (flags & MNT_EXPIRE) {
1171                 if (&mnt->mnt == current->fs->root.mnt ||
1172                     flags & (MNT_FORCE | MNT_DETACH))
1173                         return -EINVAL;
1174 
1175                 /*
1176                  * probably don't strictly need the lock here if we examined
1177                  * all race cases, but it's a slowpath.
1178                  */
1179                 br_write_lock(&vfsmount_lock);
1180                 if (mnt_get_count(mnt) != 2) {
1181                         br_write_unlock(&vfsmount_lock);
1182                         return -EBUSY;
1183                 }
1184                 br_write_unlock(&vfsmount_lock);
1185 
1186                 if (!xchg(&mnt->mnt_expiry_mark, 1))
1187                         return -EAGAIN;
1188         }
1189 
1190         /*
1191          * If we may have to abort operations to get out of this
1192          * mount, and they will themselves hold resources we must
1193          * allow the fs to do things. In the Unix tradition of
1194          * 'Gee thats tricky lets do it in userspace' the umount_begin
1195          * might fail to complete on the first run through as other tasks
1196          * must return, and the like. Thats for the mount program to worry
1197          * about for the moment.
1198          */
1199 
1200         if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1201                 sb->s_op->umount_begin(sb);
1202         }
1203 
1204         /*
1205          * No sense to grab the lock for this test, but test itself looks
1206          * somewhat bogus. Suggestions for better replacement?
1207          * Ho-hum... In principle, we might treat that as umount + switch
1208          * to rootfs. GC would eventually take care of the old vfsmount.
1209          * Actually it makes sense, especially if rootfs would contain a
1210          * /reboot - static binary that would close all descriptors and
1211          * call reboot(9). Then init(8) could umount root and exec /reboot.
1212          */
1213         if (&mnt->mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
1214                 /*
1215                  * Special case for "unmounting" root ...
1216                  * we just try to remount it readonly.
1217                  */
1218                 down_write(&sb->s_umount);
1219                 if (!(sb->s_flags & MS_RDONLY))
1220                         retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1221                 up_write(&sb->s_umount);
1222                 return retval;
1223         }
1224 
1225         down_write(&namespace_sem);
1226         br_write_lock(&vfsmount_lock);
1227         event++;
1228 
1229         if (!(flags & MNT_DETACH))
1230                 shrink_submounts(mnt, &umount_list);
1231 
1232         retval = -EBUSY;
1233         if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
1234                 if (!list_empty(&mnt->mnt_list))
1235                         umount_tree(mnt, 1, &umount_list);
1236                 retval = 0;
1237         }
1238         br_write_unlock(&vfsmount_lock);
1239         up_write(&namespace_sem);
1240         release_mounts(&umount_list);
1241         return retval;
1242 }
1243 
1244 /*
1245  * Is the caller allowed to modify his namespace?
1246  */
1247 static inline bool may_mount(void)
1248 {
1249         return ns_capable(current->nsproxy->mnt_ns->user_ns, CAP_SYS_ADMIN);
1250 }
1251 
1252 /*
1253  * Now umount can handle mount points as well as block devices.
1254  * This is important for filesystems which use unnamed block devices.
1255  *
1256  * We now support a flag for forced unmount like the other 'big iron'
1257  * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1258  */
1259 
1260 SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
1261 {
1262         struct path path;
1263         struct mount *mnt;
1264         int retval;
1265         int lookup_flags = 0;
1266 
1267         if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1268                 return -EINVAL;
1269 
1270         if (!may_mount())
1271                 return -EPERM;
1272 
1273         if (!(flags & UMOUNT_NOFOLLOW))
1274                 lookup_flags |= LOOKUP_FOLLOW;
1275 
1276         retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
1277         if (retval)
1278                 goto out;
1279         mnt = real_mount(path.mnt);
1280         retval = -EINVAL;
1281         if (path.dentry != path.mnt->mnt_root)
1282                 goto dput_and_out;
1283         if (!check_mnt(mnt))
1284                 goto dput_and_out;
1285 
1286         retval = do_umount(mnt, flags);
1287 dput_and_out:
1288         /* we mustn't call path_put() as that would clear mnt_expiry_mark */
1289         dput(path.dentry);
1290         mntput_no_expire(mnt);
1291 out:
1292         return retval;
1293 }
1294 
1295 #ifdef __ARCH_WANT_SYS_OLDUMOUNT
1296 
1297 /*
1298  *      The 2.0 compatible umount. No flags.
1299  */
1300 SYSCALL_DEFINE1(oldumount, char __user *, name)
1301 {
1302         return sys_umount(name, 0);
1303 }
1304 
1305 #endif
1306 
1307 static int mount_is_safe(struct path *path)
1308 {
1309         if (may_mount())
1310                 return 0;
1311         return -EPERM;
1312 #ifdef notyet
1313         if (S_ISLNK(path->dentry->d_inode->i_mode))
1314                 return -EPERM;
1315         if (path->dentry->d_inode->i_mode & S_ISVTX) {
1316                 if (current_uid() != path->dentry->d_inode->i_uid)
1317                         return -EPERM;
1318         }
1319         if (inode_permission(path->dentry->d_inode, MAY_WRITE))
1320                 return -EPERM;
1321         return 0;
1322 #endif
1323 }
1324 
1325 static bool mnt_ns_loop(struct path *path)
1326 {
1327         /* Could bind mounting the mount namespace inode cause a
1328          * mount namespace loop?
1329          */
1330         struct inode *inode = path->dentry->d_inode;
1331         struct proc_inode *ei;
1332         struct mnt_namespace *mnt_ns;
1333 
1334         if (!proc_ns_inode(inode))
1335                 return false;
1336 
1337         ei = PROC_I(inode);
1338         if (ei->ns_ops != &mntns_operations)
1339                 return false;
1340 
1341         mnt_ns = ei->ns;
1342         return current->nsproxy->mnt_ns->seq >= mnt_ns->seq;
1343 }
1344 
1345 struct mount *copy_tree(struct mount *mnt, struct dentry *dentry,
1346                                         int flag)
1347 {
1348         struct mount *res, *p, *q, *r;
1349         struct path path;
1350 
1351         if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1352                 return ERR_PTR(-EINVAL);
1353 
1354         res = q = clone_mnt(mnt, dentry, flag);
1355         if (IS_ERR(q))
1356                 return q;
1357 
1358         q->mnt_mountpoint = mnt->mnt_mountpoint;
1359 
1360         p = mnt;
1361         list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
1362                 struct mount *s;
1363                 if (!is_subdir(r->mnt_mountpoint, dentry))
1364                         continue;
1365 
1366                 for (s = r; s; s = next_mnt(s, r)) {
1367                         if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1368                                 s = skip_mnt_tree(s);
1369                                 continue;
1370                         }
1371                         while (p != s->mnt_parent) {
1372                                 p = p->mnt_parent;
1373                                 q = q->mnt_parent;
1374                         }
1375                         p = s;
1376                         path.mnt = &q->mnt;
1377                         path.dentry = p->mnt_mountpoint;
1378                         q = clone_mnt(p, p->mnt.mnt_root, flag);
1379                         if (IS_ERR(q))
1380                                 goto out;
1381                         br_write_lock(&vfsmount_lock);
1382                         list_add_tail(&q->mnt_list, &res->mnt_list);
1383                         attach_mnt(q, &path);
1384                         br_write_unlock(&vfsmount_lock);
1385                 }
1386         }
1387         return res;
1388 out:
1389         if (res) {
1390                 LIST_HEAD(umount_list);
1391                 br_write_lock(&vfsmount_lock);
1392                 umount_tree(res, 0, &umount_list);
1393                 br_write_unlock(&vfsmount_lock);
1394                 release_mounts(&umount_list);
1395         }
1396         return q;
1397 }
1398 
1399 /* Caller should check returned pointer for errors */
1400 
1401 struct vfsmount *collect_mounts(struct path *path)
1402 {
1403         struct mount *tree;
1404         down_write(&namespace_sem);
1405         tree = copy_tree(real_mount(path->mnt), path->dentry,
1406                          CL_COPY_ALL | CL_PRIVATE);
1407         up_write(&namespace_sem);
1408         if (IS_ERR(tree))
1409                 return NULL;
1410         return &tree->mnt;
1411 }
1412 
1413 void drop_collected_mounts(struct vfsmount *mnt)
1414 {
1415         LIST_HEAD(umount_list);
1416         down_write(&namespace_sem);
1417         br_write_lock(&vfsmount_lock);
1418         umount_tree(real_mount(mnt), 0, &umount_list);
1419         br_write_unlock(&vfsmount_lock);
1420         up_write(&namespace_sem);
1421         release_mounts(&umount_list);
1422 }
1423 
1424 int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1425                    struct vfsmount *root)
1426 {
1427         struct mount *mnt;
1428         int res = f(root, arg);
1429         if (res)
1430                 return res;
1431         list_for_each_entry(mnt, &real_mount(root)->mnt_list, mnt_list) {
1432                 res = f(&mnt->mnt, arg);
1433                 if (res)
1434                         return res;
1435         }
1436         return 0;
1437 }
1438 
1439 static void cleanup_group_ids(struct mount *mnt, struct mount *end)
1440 {
1441         struct mount *p;
1442 
1443         for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1444                 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1445                         mnt_release_group_id(p);
1446         }
1447 }
1448 
1449 static int invent_group_ids(struct mount *mnt, bool recurse)
1450 {
1451         struct mount *p;
1452 
1453         for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1454                 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1455                         int err = mnt_alloc_group_id(p);
1456                         if (err) {
1457                                 cleanup_group_ids(mnt, p);
1458                                 return err;
1459                         }
1460                 }
1461         }
1462 
1463         return 0;
1464 }
1465 
1466 /*
1467  *  @source_mnt : mount tree to be attached
1468  *  @nd         : place the mount tree @source_mnt is attached
1469  *  @parent_nd  : if non-null, detach the source_mnt from its parent and
1470  *                 store the parent mount and mountpoint dentry.
1471  *                 (done when source_mnt is moved)
1472  *
1473  *  NOTE: in the table below explains the semantics when a source mount
1474  *  of a given type is attached to a destination mount of a given type.
1475  * ---------------------------------------------------------------------------
1476  * |         BIND MOUNT OPERATION                                            |
1477  * |**************************************************************************
1478  * | source-->| shared        |       private  |       slave    | unbindable |
1479  * | dest     |               |                |                |            |
1480  * |   |      |               |                |                |            |
1481  * |   v      |               |                |                |            |
1482  * |**************************************************************************
1483  * |  shared  | shared (++)   |     shared (+) |     shared(+++)|  invalid   |
1484  * |          |               |                |                |            |
1485  * |non-shared| shared (+)    |      private   |      slave (*) |  invalid   |
1486  * ***************************************************************************
1487  * A bind operation clones the source mount and mounts the clone on the
1488  * destination mount.
1489  *
1490  * (++)  the cloned mount is propagated to all the mounts in the propagation
1491  *       tree of the destination mount and the cloned mount is added to
1492  *       the peer group of the source mount.
1493  * (+)   the cloned mount is created under the destination mount and is marked
1494  *       as shared. The cloned mount is added to the peer group of the source
1495  *       mount.
1496  * (+++) the mount is propagated to all the mounts in the propagation tree
1497  *       of the destination mount and the cloned mount is made slave
1498  *       of the same master as that of the source mount. The cloned mount
1499  *       is marked as 'shared and slave'.
1500  * (*)   the cloned mount is made a slave of the same master as that of the
1501  *       source mount.
1502  *
1503  * ---------------------------------------------------------------------------
1504  * |                    MOVE MOUNT OPERATION                                 |
1505  * |**************************************************************************
1506  * | source-->| shared        |       private  |       slave    | unbindable |
1507  * | dest     |               |                |                |            |
1508  * |   |      |               |                |                |            |
1509  * |   v      |               |                |                |            |
1510  * |**************************************************************************
1511  * |  shared  | shared (+)    |     shared (+) |    shared(+++) |  invalid   |
1512  * |          |               |                |                |            |
1513  * |non-shared| shared (+*)   |      private   |    slave (*)   | unbindable |
1514  * ***************************************************************************
1515  *
1516  * (+)  the mount is moved to the destination. And is then propagated to
1517  *      all the mounts in the propagation tree of the destination mount.
1518  * (+*)  the mount is moved to the destination.
1519  * (+++)  the mount is moved to the destination and is then propagated to
1520  *      all the mounts belonging to the destination mount's propagation tree.
1521  *      the mount is marked as 'shared and slave'.
1522  * (*)  the mount continues to be a slave at the new location.
1523  *
1524  * if the source mount is a tree, the operations explained above is
1525  * applied to each mount in the tree.
1526  * Must be called without spinlocks held, since this function can sleep
1527  * in allocations.
1528  */
1529 static int attach_recursive_mnt(struct mount *source_mnt,
1530                         struct path *path, struct path *parent_path)
1531 {
1532         LIST_HEAD(tree_list);
1533         struct mount *dest_mnt = real_mount(path->mnt);
1534         struct dentry *dest_dentry = path->dentry;
1535         struct mount *child, *p;
1536         int err;
1537 
1538         if (IS_MNT_SHARED(dest_mnt)) {
1539                 err = invent_group_ids(source_mnt, true);
1540                 if (err)
1541                         goto out;
1542         }
1543         err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1544         if (err)
1545                 goto out_cleanup_ids;
1546 
1547         br_write_lock(&vfsmount_lock);
1548 
1549         if (IS_MNT_SHARED(dest_mnt)) {
1550                 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1551                         set_mnt_shared(p);
1552         }
1553         if (parent_path) {
1554                 detach_mnt(source_mnt, parent_path);
1555                 attach_mnt(source_mnt, path);
1556                 touch_mnt_namespace(source_mnt->mnt_ns);
1557         } else {
1558                 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1559                 commit_tree(source_mnt);
1560         }
1561 
1562         list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1563                 list_del_init(&child->mnt_hash);
1564                 commit_tree(child);
1565         }
1566         br_write_unlock(&vfsmount_lock);
1567 
1568         return 0;
1569 
1570  out_cleanup_ids:
1571         if (IS_MNT_SHARED(dest_mnt))
1572                 cleanup_group_ids(source_mnt, NULL);
1573  out:
1574         return err;
1575 }
1576 
1577 static int lock_mount(struct path *path)
1578 {
1579         struct vfsmount *mnt;
1580 retry:
1581         mutex_lock(&path->dentry->d_inode->i_mutex);
1582         if (unlikely(cant_mount(path->dentry))) {
1583                 mutex_unlock(&path->dentry->d_inode->i_mutex);
1584                 return -ENOENT;
1585         }
1586         down_write(&namespace_sem);
1587         mnt = lookup_mnt(path);
1588         if (likely(!mnt))
1589                 return 0;
1590         up_write(&namespace_sem);
1591         mutex_unlock(&path->dentry->d_inode->i_mutex);
1592         path_put(path);
1593         path->mnt = mnt;
1594         path->dentry = dget(mnt->mnt_root);
1595         goto retry;
1596 }
1597 
1598 static void unlock_mount(struct path *path)
1599 {
1600         up_write(&namespace_sem);
1601         mutex_unlock(&path->dentry->d_inode->i_mutex);
1602 }
1603 
1604 static int graft_tree(struct mount *mnt, struct path *path)
1605 {
1606         if (mnt->mnt.mnt_sb->s_flags & MS_NOUSER)
1607                 return -EINVAL;
1608 
1609         if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1610               S_ISDIR(mnt->mnt.mnt_root->d_inode->i_mode))
1611                 return -ENOTDIR;
1612 
1613         if (d_unlinked(path->dentry))
1614                 return -ENOENT;
1615 
1616         return attach_recursive_mnt(mnt, path, NULL);
1617 }
1618 
1619 /*
1620  * Sanity check the flags to change_mnt_propagation.
1621  */
1622 
1623 static int flags_to_propagation_type(int flags)
1624 {
1625         int type = flags & ~(MS_REC | MS_SILENT);
1626 
1627         /* Fail if any non-propagation flags are set */
1628         if (type & ~(MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
1629                 return 0;
1630         /* Only one propagation flag should be set */
1631         if (!is_power_of_2(type))
1632                 return 0;
1633         return type;
1634 }
1635 
1636 /*
1637  * recursively change the type of the mountpoint.
1638  */
1639 static int do_change_type(struct path *path, int flag)
1640 {
1641         struct mount *m;
1642         struct mount *mnt = real_mount(path->mnt);
1643         int recurse = flag & MS_REC;
1644         int type;
1645         int err = 0;
1646 
1647         if (!may_mount())
1648                 return -EPERM;
1649 
1650         if (path->dentry != path->mnt->mnt_root)
1651                 return -EINVAL;
1652 
1653         type = flags_to_propagation_type(flag);
1654         if (!type)
1655                 return -EINVAL;
1656 
1657         down_write(&namespace_sem);
1658         if (type == MS_SHARED) {
1659                 err = invent_group_ids(mnt, recurse);
1660                 if (err)
1661                         goto out_unlock;
1662         }
1663 
1664         br_write_lock(&vfsmount_lock);
1665         for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1666                 change_mnt_propagation(m, type);
1667         br_write_unlock(&vfsmount_lock);
1668 
1669  out_unlock:
1670         up_write(&namespace_sem);
1671         return err;
1672 }
1673 
1674 /*
1675  * do loopback mount.
1676  */
1677 static int do_loopback(struct path *path, const char *old_name,
1678                                 int recurse)
1679 {
1680         LIST_HEAD(umount_list);
1681         struct path old_path;
1682         struct mount *mnt = NULL, *old;
1683         int err = mount_is_safe(path);
1684         if (err)
1685                 return err;
1686         if (!old_name || !*old_name)
1687                 return -EINVAL;
1688         err = kern_path(old_name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &old_path);
1689         if (err)
1690                 return err;
1691 
1692         err = -EINVAL;
1693         if (mnt_ns_loop(&old_path))
1694                 goto out; 
1695 
1696         err = lock_mount(path);
1697         if (err)
1698                 goto out;
1699 
1700         old = real_mount(old_path.mnt);
1701 
1702         err = -EINVAL;
1703         if (IS_MNT_UNBINDABLE(old))
1704                 goto out2;
1705 
1706         if (!check_mnt(real_mount(path->mnt)) || !check_mnt(old))
1707                 goto out2;
1708 
1709         if (recurse)
1710                 mnt = copy_tree(old, old_path.dentry, 0);
1711         else
1712                 mnt = clone_mnt(old, old_path.dentry, 0);
1713 
1714         if (IS_ERR(mnt)) {
1715                 err = PTR_ERR(mnt);
1716                 goto out;
1717         }
1718 
1719         err = graft_tree(mnt, path);
1720         if (err) {
1721                 br_write_lock(&vfsmount_lock);
1722                 umount_tree(mnt, 0, &umount_list);
1723                 br_write_unlock(&vfsmount_lock);
1724         }
1725 out2:
1726         unlock_mount(path);
1727         release_mounts(&umount_list);
1728 out:
1729         path_put(&old_path);
1730         return err;
1731 }
1732 
1733 static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1734 {
1735         int error = 0;
1736         int readonly_request = 0;
1737 
1738         if (ms_flags & MS_RDONLY)
1739                 readonly_request = 1;
1740         if (readonly_request == __mnt_is_readonly(mnt))
1741                 return 0;
1742 
1743         if (mnt->mnt_flags & MNT_LOCK_READONLY)
1744                 return -EPERM;
1745 
1746         if (readonly_request)
1747                 error = mnt_make_readonly(real_mount(mnt));
1748         else
1749                 __mnt_unmake_readonly(real_mount(mnt));
1750         return error;
1751 }
1752 
1753 /*
1754  * change filesystem flags. dir should be a physical root of filesystem.
1755  * If you've mounted a non-root directory somewhere and want to do remount
1756  * on it - tough luck.
1757  */
1758 static int do_remount(struct path *path, int flags, int mnt_flags,
1759                       void *data)
1760 {
1761         int err;
1762         struct super_block *sb = path->mnt->mnt_sb;
1763         struct mount *mnt = real_mount(path->mnt);
1764 
1765         if (!capable(CAP_SYS_ADMIN))
1766                 return -EPERM;
1767 
1768         if (!check_mnt(mnt))
1769                 return -EINVAL;
1770 
1771         if (path->dentry != path->mnt->mnt_root)
1772                 return -EINVAL;
1773 
1774         err = security_sb_remount(sb, data);
1775         if (err)
1776                 return err;
1777 
1778         down_write(&sb->s_umount);
1779         if (flags & MS_BIND)
1780                 err = change_mount_flags(path->mnt, flags);
1781         else
1782                 err = do_remount_sb(sb, flags, data, 0);
1783         if (!err) {
1784                 br_write_lock(&vfsmount_lock);
1785                 mnt_flags |= mnt->mnt.mnt_flags & MNT_PROPAGATION_MASK;
1786                 mnt->mnt.mnt_flags = mnt_flags;
1787                 br_write_unlock(&vfsmount_lock);
1788         }
1789         up_write(&sb->s_umount);
1790         if (!err) {
1791                 br_write_lock(&vfsmount_lock);
1792                 touch_mnt_namespace(mnt->mnt_ns);
1793                 br_write_unlock(&vfsmount_lock);
1794         }
1795         return err;
1796 }
1797 
1798 static inline int tree_contains_unbindable(struct mount *mnt)
1799 {
1800         struct mount *p;
1801         for (p = mnt; p; p = next_mnt(p, mnt)) {
1802                 if (IS_MNT_UNBINDABLE(p))
1803                         return 1;
1804         }
1805         return 0;
1806 }
1807 
1808 static int do_move_mount(struct path *path, const char *old_name)
1809 {
1810         struct path old_path, parent_path;
1811         struct mount *p;
1812         struct mount *old;
1813         int err = 0;
1814         if (!may_mount())
1815                 return -EPERM;
1816         if (!old_name || !*old_name)
1817                 return -EINVAL;
1818         err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
1819         if (err)
1820                 return err;
1821 
1822         err = lock_mount(path);
1823         if (err < 0)
1824                 goto out;
1825 
1826         old = real_mount(old_path.mnt);
1827         p = real_mount(path->mnt);
1828 
1829         err = -EINVAL;
1830         if (!check_mnt(p) || !check_mnt(old))
1831                 goto out1;
1832 
1833         if (d_unlinked(path->dentry))
1834                 goto out1;
1835 
1836         err = -EINVAL;
1837         if (old_path.dentry != old_path.mnt->mnt_root)
1838                 goto out1;
1839 
1840         if (!mnt_has_parent(old))
1841                 goto out1;
1842 
1843         if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1844               S_ISDIR(old_path.dentry->d_inode->i_mode))
1845                 goto out1;
1846         /*
1847          * Don't move a mount residing in a shared parent.
1848          */
1849         if (IS_MNT_SHARED(old->mnt_parent))
1850                 goto out1;
1851         /*
1852          * Don't move a mount tree containing unbindable mounts to a destination
1853          * mount which is shared.
1854          */
1855         if (IS_MNT_SHARED(p) && tree_contains_unbindable(old))
1856                 goto out1;
1857         err = -ELOOP;
1858         for (; mnt_has_parent(p); p = p->mnt_parent)
1859                 if (p == old)
1860                         goto out1;
1861 
1862         err = attach_recursive_mnt(old, path, &parent_path);
1863         if (err)
1864                 goto out1;
1865 
1866         /* if the mount is moved, it should no longer be expire
1867          * automatically */
1868         list_del_init(&old->mnt_expire);
1869 out1:
1870         unlock_mount(path);
1871 out:
1872         if (!err)
1873                 path_put(&parent_path);
1874         path_put(&old_path);
1875         return err;
1876 }
1877 
1878 static struct vfsmount *fs_set_subtype(struct vfsmount *mnt, const char *fstype)
1879 {
1880         int err;
1881         const char *subtype = strchr(fstype, '.');
1882         if (subtype) {
1883                 subtype++;
1884                 err = -EINVAL;
1885                 if (!subtype[0])
1886                         goto err;
1887         } else
1888                 subtype = "";
1889 
1890         mnt->mnt_sb->s_subtype = kstrdup(subtype, GFP_KERNEL);
1891         err = -ENOMEM;
1892         if (!mnt->mnt_sb->s_subtype)
1893                 goto err;
1894         return mnt;
1895 
1896  err:
1897         mntput(mnt);
1898         return ERR_PTR(err);
1899 }
1900 
1901 /*
1902  * add a mount into a namespace's mount tree
1903  */
1904 static int do_add_mount(struct mount *newmnt, struct path *path, int mnt_flags)
1905 {
1906         int err;
1907 
1908         mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
1909 
1910         err = lock_mount(path);
1911         if (err)
1912                 return err;
1913 
1914         err = -EINVAL;
1915         if (unlikely(!check_mnt(real_mount(path->mnt)))) {
1916                 /* that's acceptable only for automounts done in private ns */
1917                 if (!(mnt_flags & MNT_SHRINKABLE))
1918                         goto unlock;
1919                 /* ... and for those we'd better have mountpoint still alive */
1920                 if (!real_mount(path->mnt)->mnt_ns)
1921                         goto unlock;
1922         }
1923 
1924         /* Refuse the same filesystem on the same mount point */
1925         err = -EBUSY;
1926         if (path->mnt->mnt_sb == newmnt->mnt.mnt_sb &&
1927             path->mnt->mnt_root == path->dentry)
1928                 goto unlock;
1929 
1930         err = -EINVAL;
1931         if (S_ISLNK(newmnt->mnt.mnt_root->d_inode->i_mode))
1932                 goto unlock;
1933 
1934         newmnt->mnt.mnt_flags = mnt_flags;
1935         err = graft_tree(newmnt, path);
1936 
1937 unlock:
1938         unlock_mount(path);
1939         return err;
1940 }
1941 
1942 /*
1943  * create a new mount for userspace and request it to be added into the
1944  * namespace's tree
1945  */
1946 static int do_new_mount(struct path *path, const char *fstype, int flags,
1947                         int mnt_flags, const char *name, void *data)
1948 {
1949         struct file_system_type *type;
1950         struct user_namespace *user_ns = current->nsproxy->mnt_ns->user_ns;
1951         struct vfsmount *mnt;
1952         int err;
1953 
1954         if (!fstype)
1955                 return -EINVAL;
1956 
1957         if (!may_mount())
1958                 return -EPERM;
1959 
1960         type = get_fs_type(fstype);
1961         if (!type)
1962                 return -ENODEV;
1963 
1964         if (user_ns != &init_user_ns) {
1965                 if (!(type->fs_flags & FS_USERNS_MOUNT)) {
1966                         put_filesystem(type);
1967                         return -EPERM;
1968                 }
1969                 /* Only in special cases allow devices from mounts
1970                  * created outside the initial user namespace.
1971                  */
1972                 if (!(type->fs_flags & FS_USERNS_DEV_MOUNT)) {
1973                         flags |= MS_NODEV;
1974                         mnt_flags |= MNT_NODEV;
1975                 }
1976         }
1977 
1978         mnt = vfs_kern_mount(type, flags, name, data);
1979         if (!IS_ERR(mnt) && (type->fs_flags & FS_HAS_SUBTYPE) &&
1980             !mnt->mnt_sb->s_subtype)
1981                 mnt = fs_set_subtype(mnt, fstype);
1982 
1983         put_filesystem(type);
1984         if (IS_ERR(mnt))
1985                 return PTR_ERR(mnt);
1986 
1987         err = do_add_mount(real_mount(mnt), path, mnt_flags);
1988         if (err)
1989                 mntput(mnt);
1990         return err;
1991 }
1992 
1993 int finish_automount(struct vfsmount *m, struct path *path)
1994 {
1995         struct mount *mnt = real_mount(m);
1996         int err;
1997         /* The new mount record should have at least 2 refs to prevent it being
1998          * expired before we get a chance to add it
1999          */
2000         BUG_ON(mnt_get_count(mnt) < 2);
2001 
2002         if (m->mnt_sb == path->mnt->mnt_sb &&
2003             m->mnt_root == path->dentry) {
2004                 err = -ELOOP;
2005                 goto fail;
2006         }
2007 
2008         err = do_add_mount(mnt, path, path->mnt->mnt_flags | MNT_SHRINKABLE);
2009         if (!err)
2010                 return 0;
2011 fail:
2012         /* remove m from any expiration list it may be on */
2013         if (!list_empty(&mnt->mnt_expire)) {
2014                 down_write(&namespace_sem);
2015                 br_write_lock(&vfsmount_lock);
2016                 list_del_init(&mnt->mnt_expire);
2017                 br_write_unlock(&vfsmount_lock);
2018                 up_write(&namespace_sem);
2019         }
2020         mntput(m);
2021         mntput(m);
2022         return err;
2023 }
2024 
2025 /**
2026  * mnt_set_expiry - Put a mount on an expiration list
2027  * @mnt: The mount to list.
2028  * @expiry_list: The list to add the mount to.
2029  */
2030 void mnt_set_expiry(struct vfsmount *mnt, struct list_head *expiry_list)
2031 {
2032         down_write(&namespace_sem);
2033         br_write_lock(&vfsmount_lock);
2034 
2035         list_add_tail(&real_mount(mnt)->mnt_expire, expiry_list);
2036 
2037         br_write_unlock(&vfsmount_lock);
2038         up_write(&namespace_sem);
2039 }
2040 EXPORT_SYMBOL(mnt_set_expiry);
2041 
2042 /*
2043  * process a list of expirable mountpoints with the intent of discarding any
2044  * mountpoints that aren't in use and haven't been touched since last we came
2045  * here
2046  */
2047 void mark_mounts_for_expiry(struct list_head *mounts)
2048 {
2049         struct mount *mnt, *next;
2050         LIST_HEAD(graveyard);
2051         LIST_HEAD(umounts);
2052 
2053         if (list_empty(mounts))
2054                 return;
2055 
2056         down_write(&namespace_sem);
2057         br_write_lock(&vfsmount_lock);
2058 
2059         /* extract from the expiration list every vfsmount that matches the
2060          * following criteria:
2061          * - only referenced by its parent vfsmount
2062          * - still marked for expiry (marked on the last call here; marks are
2063          *   cleared by mntput())
2064          */
2065         list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
2066                 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
2067                         propagate_mount_busy(mnt, 1))
2068                         continue;
2069                 list_move(&mnt->mnt_expire, &graveyard);
2070         }
2071         while (!list_empty(&graveyard)) {
2072                 mnt = list_first_entry(&graveyard, struct mount, mnt_expire);
2073                 touch_mnt_namespace(mnt->mnt_ns);
2074                 umount_tree(mnt, 1, &umounts);
2075         }
2076         br_write_unlock(&vfsmount_lock);
2077         up_write(&namespace_sem);
2078 
2079         release_mounts(&umounts);
2080 }
2081 
2082 EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
2083 
2084 /*
2085  * Ripoff of 'select_parent()'
2086  *
2087  * search the list of submounts for a given mountpoint, and move any
2088  * shrinkable submounts to the 'graveyard' list.
2089  */
2090 static int select_submounts(struct mount *parent, struct list_head *graveyard)
2091 {
2092         struct mount *this_parent = parent;
2093         struct list_head *next;
2094         int found = 0;
2095 
2096 repeat:
2097         next = this_parent->mnt_mounts.next;
2098 resume:
2099         while (next != &this_parent->mnt_mounts) {
2100                 struct list_head *tmp = next;
2101                 struct mount *mnt = list_entry(tmp, struct mount, mnt_child);
2102 
2103                 next = tmp->next;
2104                 if (!(mnt->mnt.mnt_flags & MNT_SHRINKABLE))
2105                         continue;
2106                 /*
2107                  * Descend a level if the d_mounts list is non-empty.
2108                  */
2109                 if (!list_empty(&mnt->mnt_mounts)) {
2110                         this_parent = mnt;
2111                         goto repeat;
2112                 }
2113 
2114                 if (!propagate_mount_busy(mnt, 1)) {
2115                         list_move_tail(&mnt->mnt_expire, graveyard);
2116                         found++;
2117                 }
2118         }
2119         /*
2120          * All done at this level ... ascend and resume the search
2121          */
2122         if (this_parent != parent) {
2123                 next = this_parent->mnt_child.next;
2124                 this_parent = this_parent->mnt_parent;
2125                 goto resume;
2126         }
2127         return found;
2128 }
2129 
2130 /*
2131  * process a list of expirable mountpoints with the intent of discarding any
2132  * submounts of a specific parent mountpoint
2133  *
2134  * vfsmount_lock must be held for write
2135  */
2136 static void shrink_submounts(struct mount *mnt, struct list_head *umounts)
2137 {
2138         LIST_HEAD(graveyard);
2139         struct mount *m;
2140 
2141         /* extract submounts of 'mountpoint' from the expiration list */
2142         while (select_submounts(mnt, &graveyard)) {
2143                 while (!list_empty(&graveyard)) {
2144                         m = list_first_entry(&graveyard, struct mount,
2145                                                 mnt_expire);
2146                         touch_mnt_namespace(m->mnt_ns);
2147                         umount_tree(m, 1, umounts);
2148                 }
2149         }
2150 }
2151 
2152 /*
2153  * Some copy_from_user() implementations do not return the exact number of
2154  * bytes remaining to copy on a fault.  But copy_mount_options() requires that.
2155  * Note that this function differs from copy_from_user() in that it will oops
2156  * on bad values of `to', rather than returning a short copy.
2157  */
2158 static long exact_copy_from_user(void *to, const void __user * from,
2159                                  unsigned long n)
2160 {
2161         char *t = to;
2162         const char __user *f = from;
2163         char c;
2164 
2165         if (!access_ok(VERIFY_READ, from, n))
2166                 return n;
2167 
2168         while (n) {
2169                 if (__get_user(c, f)) {
2170                         memset(t, 0, n);
2171                         break;
2172                 }
2173                 *t++ = c;
2174                 f++;
2175                 n--;
2176         }
2177         return n;
2178 }
2179 
2180 int copy_mount_options(const void __user * data, unsigned long *where)
2181 {
2182         int i;
2183         unsigned long page;
2184         unsigned long size;
2185 
2186         *where = 0;
2187         if (!data)
2188                 return 0;
2189 
2190         if (!(page = __get_free_page(GFP_KERNEL)))
2191                 return -ENOMEM;
2192 
2193         /* We only care that *some* data at the address the user
2194          * gave us is valid.  Just in case, we'll zero
2195          * the remainder of the page.
2196          */
2197         /* copy_from_user cannot cross TASK_SIZE ! */
2198         size = TASK_SIZE - (unsigned long)data;
2199         if (size > PAGE_SIZE)
2200                 size = PAGE_SIZE;
2201 
2202         i = size - exact_copy_from_user((void *)page, data, size);
2203         if (!i) {
2204                 free_page(page);
2205                 return -EFAULT;
2206         }
2207         if (i != PAGE_SIZE)
2208                 memset((char *)page + i, 0, PAGE_SIZE - i);
2209         *where = page;
2210         return 0;
2211 }
2212 
2213 int copy_mount_string(const void __user *data, char **where)
2214 {
2215         char *tmp;
2216 
2217         if (!data) {
2218                 *where = NULL;
2219                 return 0;
2220         }
2221 
2222         tmp = strndup_user(data, PAGE_SIZE);
2223         if (IS_ERR(tmp))
2224                 return PTR_ERR(tmp);
2225 
2226         *where = tmp;
2227         return 0;
2228 }
2229 
2230 /*
2231  * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
2232  * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
2233  *
2234  * data is a (void *) that can point to any structure up to
2235  * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
2236  * information (or be NULL).
2237  *
2238  * Pre-0.97 versions of mount() didn't have a flags word.
2239  * When the flags word was introduced its top half was required
2240  * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
2241  * Therefore, if this magic number is present, it carries no information
2242  * and must be discarded.
2243  */
2244 long do_mount(const char *dev_name, const char *dir_name,
2245                 const char *type_page, unsigned long flags, void *data_page)
2246 {
2247         struct path path;
2248         int retval = 0;
2249         int mnt_flags = 0;
2250 
2251         /* Discard magic */
2252         if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
2253                 flags &= ~MS_MGC_MSK;
2254 
2255         /* Basic sanity checks */
2256 
2257         if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
2258                 return -EINVAL;
2259 
2260         if (data_page)
2261                 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2262 
2263         /* ... and get the mountpoint */
2264         retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2265         if (retval)
2266                 return retval;
2267 
2268         retval = security_sb_mount(dev_name, &path,
2269                                    type_page, flags, data_page);
2270         if (retval)
2271                 goto dput_out;
2272 
2273         /* Default to relatime unless overriden */
2274         if (!(flags & MS_NOATIME))
2275                 mnt_flags |= MNT_RELATIME;
2276 
2277         /* Separate the per-mountpoint flags */
2278         if (flags & MS_NOSUID)
2279                 mnt_flags |= MNT_NOSUID;
2280         if (flags & MS_NODEV)
2281                 mnt_flags |= MNT_NODEV;
2282         if (flags & MS_NOEXEC)
2283                 mnt_flags |= MNT_NOEXEC;
2284         if (flags & MS_NOATIME)
2285                 mnt_flags |= MNT_NOATIME;
2286         if (flags & MS_NODIRATIME)
2287                 mnt_flags |= MNT_NODIRATIME;
2288         if (flags & MS_STRICTATIME)
2289                 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
2290         if (flags & MS_RDONLY)
2291                 mnt_flags |= MNT_READONLY;
2292 
2293         flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
2294                    MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2295                    MS_STRICTATIME);
2296 
2297         if (flags & MS_REMOUNT)
2298                 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
2299                                     data_page);
2300         else if (flags & MS_BIND)
2301                 retval = do_loopback(&path, dev_name, flags & MS_REC);
2302         else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
2303                 retval = do_change_type(&path, flags);
2304         else if (flags & MS_MOVE)
2305                 retval = do_move_mount(&path, dev_name);
2306         else
2307                 retval = do_new_mount(&path, type_page, flags, mnt_flags,
2308                                       dev_name, data_page);
2309 dput_out:
2310         path_put(&path);
2311         return retval;
2312 }
2313 
2314 static void free_mnt_ns(struct mnt_namespace *ns)
2315 {
2316         proc_free_inum(ns->proc_inum);
2317         put_user_ns(ns->user_ns);
2318         kfree(ns);
2319 }
2320 
2321 /*
2322  * Assign a sequence number so we can detect when we attempt to bind
2323  * mount a reference to an older mount namespace into the current
2324  * mount namespace, preventing reference counting loops.  A 64bit
2325  * number incrementing at 10Ghz will take 12,427 years to wrap which
2326  * is effectively never, so we can ignore the possibility.
2327  */
2328 static atomic64_t mnt_ns_seq = ATOMIC64_INIT(1);
2329 
2330 static struct mnt_namespace *alloc_mnt_ns(struct user_namespace *user_ns)
2331 {
2332         struct mnt_namespace *new_ns;
2333         int ret;
2334 
2335         new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2336         if (!new_ns)
2337                 return ERR_PTR(-ENOMEM);
2338         ret = proc_alloc_inum(&new_ns->proc_inum);
2339         if (ret) {
2340                 kfree(new_ns);
2341                 return ERR_PTR(ret);
2342         }
2343         new_ns->seq = atomic64_add_return(1, &mnt_ns_seq);
2344         atomic_set(&new_ns->count, 1);
2345         new_ns->root = NULL;
2346         INIT_LIST_HEAD(&new_ns->list);
2347         init_waitqueue_head(&new_ns->poll);
2348         new_ns->event = 0;
2349         new_ns->user_ns = get_user_ns(user_ns);
2350         return new_ns;
2351 }
2352 
2353 /*
2354  * Allocate a new namespace structure and populate it with contents
2355  * copied from the namespace of the passed in task structure.
2356  */
2357 static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
2358                 struct user_namespace *user_ns, struct fs_struct *fs)
2359 {
2360         struct mnt_namespace *new_ns;
2361         struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
2362         struct mount *p, *q;
2363         struct mount *old = mnt_ns->root;
2364         struct mount *new;
2365         int copy_flags;
2366 
2367         new_ns = alloc_mnt_ns(user_ns);
2368         if (IS_ERR(new_ns))
2369                 return new_ns;
2370 
2371         down_write(&namespace_sem);
2372         /* First pass: copy the tree topology */
2373         copy_flags = CL_COPY_ALL | CL_EXPIRE;
2374         if (user_ns != mnt_ns->user_ns)
2375                 copy_flags |= CL_SHARED_TO_SLAVE | CL_UNPRIVILEGED;
2376         new = copy_tree(old, old->mnt.mnt_root, copy_flags);
2377         if (IS_ERR(new)) {
2378                 up_write(&namespace_sem);
2379                 free_mnt_ns(new_ns);
2380                 return ERR_CAST(new);
2381         }
2382         new_ns->root = new;
2383         br_write_lock(&vfsmount_lock);
2384         list_add_tail(&new_ns->list, &new->mnt_list);
2385         br_write_unlock(&vfsmount_lock);
2386 
2387         /*
2388          * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2389          * as belonging to new namespace.  We have already acquired a private
2390          * fs_struct, so tsk->fs->lock is not needed.
2391          */
2392         p = old;
2393         q = new;
2394         while (p) {
2395                 q->mnt_ns = new_ns;
2396                 if (fs) {
2397                         if (&p->mnt == fs->root.mnt) {
2398                                 fs->root.mnt = mntget(&q->mnt);
2399                                 rootmnt = &p->mnt;
2400                         }
2401                         if (&p->mnt == fs->pwd.mnt) {
2402                                 fs->pwd.mnt = mntget(&q->mnt);
2403                                 pwdmnt = &p->mnt;
2404                         }
2405                 }
2406                 p = next_mnt(p, old);
2407                 q = next_mnt(q, new);
2408         }
2409         up_write(&namespace_sem);
2410 
2411         if (rootmnt)
2412                 mntput(rootmnt);
2413         if (pwdmnt)
2414                 mntput(pwdmnt);
2415 
2416         return new_ns;
2417 }
2418 
2419 struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
2420                 struct user_namespace *user_ns, struct fs_struct *new_fs)
2421 {
2422         struct mnt_namespace *new_ns;
2423 
2424         BUG_ON(!ns);
2425         get_mnt_ns(ns);
2426 
2427         if (!(flags & CLONE_NEWNS))
2428                 return ns;
2429 
2430         new_ns = dup_mnt_ns(ns, user_ns, new_fs);
2431 
2432         put_mnt_ns(ns);
2433         return new_ns;
2434 }
2435 
2436 /**
2437  * create_mnt_ns - creates a private namespace and adds a root filesystem
2438  * @mnt: pointer to the new root filesystem mountpoint
2439  */
2440 static struct mnt_namespace *create_mnt_ns(struct vfsmount *m)
2441 {
2442         struct mnt_namespace *new_ns = alloc_mnt_ns(&init_user_ns);
2443         if (!IS_ERR(new_ns)) {
2444                 struct mount *mnt = real_mount(m);
2445                 mnt->mnt_ns = new_ns;
2446                 new_ns->root = mnt;
2447                 list_add(&new_ns->list, &mnt->mnt_list);
2448         } else {
2449                 mntput(m);
2450         }
2451         return new_ns;
2452 }
2453 
2454 struct dentry *mount_subtree(struct vfsmount *mnt, const char *name)
2455 {
2456         struct mnt_namespace *ns;
2457         struct super_block *s;
2458         struct path path;
2459         int err;
2460 
2461         ns = create_mnt_ns(mnt);
2462         if (IS_ERR(ns))
2463                 return ERR_CAST(ns);
2464 
2465         err = vfs_path_lookup(mnt->mnt_root, mnt,
2466                         name, LOOKUP_FOLLOW|LOOKUP_AUTOMOUNT, &path);
2467 
2468         put_mnt_ns(ns);
2469 
2470         if (err)
2471                 return ERR_PTR(err);
2472 
2473         /* trade a vfsmount reference for active sb one */
2474         s = path.mnt->mnt_sb;
2475         atomic_inc(&s->s_active);
2476         mntput(path.mnt);
2477         /* lock the sucker */
2478         down_write(&s->s_umount);
2479         /* ... and return the root of (sub)tree on it */
2480         return path.dentry;
2481 }
2482 EXPORT_SYMBOL(mount_subtree);
2483 
2484 SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2485                 char __user *, type, unsigned long, flags, void __user *, data)
2486 {
2487         int ret;
2488         char *kernel_type;
2489         struct filename *kernel_dir;
2490         char *kernel_dev;
2491         unsigned long data_page;
2492 
2493         ret = copy_mount_string(type, &kernel_type);
2494         if (ret < 0)
2495                 goto out_type;
2496 
2497         kernel_dir = getname(dir_name);
2498         if (IS_ERR(kernel_dir)) {
2499                 ret = PTR_ERR(kernel_dir);
2500                 goto out_dir;
2501         }
2502 
2503         ret = copy_mount_string(dev_name, &kernel_dev);
2504         if (ret < 0)
2505                 goto out_dev;
2506 
2507         ret = copy_mount_options(data, &data_page);
2508         if (ret < 0)
2509                 goto out_data;
2510 
2511         ret = do_mount(kernel_dev, kernel_dir->name, kernel_type, flags,
2512                 (void *) data_page);
2513 
2514         free_page(data_page);
2515 out_data:
2516         kfree(kernel_dev);
2517 out_dev:
2518         putname(kernel_dir);
2519 out_dir:
2520         kfree(kernel_type);
2521 out_type:
2522         return ret;
2523 }
2524 
2525 /*
2526  * Return true if path is reachable from root
2527  *
2528  * namespace_sem or vfsmount_lock is held
2529  */
2530 bool is_path_reachable(struct mount *mnt, struct dentry *dentry,
2531                          const struct path *root)
2532 {
2533         while (&mnt->mnt != root->mnt && mnt_has_parent(mnt)) {
2534                 dentry = mnt->mnt_mountpoint;
2535                 mnt = mnt->mnt_parent;
2536         }
2537         return &mnt->mnt == root->mnt && is_subdir(dentry, root->dentry);
2538 }
2539 
2540 int path_is_under(struct path *path1, struct path *path2)
2541 {
2542         int res;
2543         br_read_lock(&vfsmount_lock);
2544         res = is_path_reachable(real_mount(path1->mnt), path1->dentry, path2);
2545         br_read_unlock(&vfsmount_lock);
2546         return res;
2547 }
2548 EXPORT_SYMBOL(path_is_under);
2549 
2550 /*
2551  * pivot_root Semantics:
2552  * Moves the root file system of the current process to the directory put_old,
2553  * makes new_root as the new root file system of the current process, and sets
2554  * root/cwd of all processes which had them on the current root to new_root.
2555  *
2556  * Restrictions:
2557  * The new_root and put_old must be directories, and  must not be on the
2558  * same file  system as the current process root. The put_old  must  be
2559  * underneath new_root,  i.e. adding a non-zero number of /.. to the string
2560  * pointed to by put_old must yield the same directory as new_root. No other
2561  * file system may be mounted on put_old. After all, new_root is a mountpoint.
2562  *
2563  * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2564  * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2565  * in this situation.
2566  *
2567  * Notes:
2568  *  - we don't move root/cwd if they are not at the root (reason: if something
2569  *    cared enough to change them, it's probably wrong to force them elsewhere)
2570  *  - it's okay to pick a root that isn't the root of a file system, e.g.
2571  *    /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2572  *    though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2573  *    first.
2574  */
2575 SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2576                 const char __user *, put_old)
2577 {
2578         struct path new, old, parent_path, root_parent, root;
2579         struct mount *new_mnt, *root_mnt;
2580         int error;
2581 
2582         if (!may_mount())
2583                 return -EPERM;
2584 
2585         error = user_path_dir(new_root, &new);
2586         if (error)
2587                 goto out0;
2588 
2589         error = user_path_dir(put_old, &old);
2590         if (error)
2591                 goto out1;
2592 
2593         error = security_sb_pivotroot(&old, &new);
2594         if (error)
2595                 goto out2;
2596 
2597         get_fs_root(current->fs, &root);
2598         error = lock_mount(&old);
2599         if (error)
2600                 goto out3;
2601 
2602         error = -EINVAL;
2603         new_mnt = real_mount(new.mnt);
2604         root_mnt = real_mount(root.mnt);
2605         if (IS_MNT_SHARED(real_mount(old.mnt)) ||
2606                 IS_MNT_SHARED(new_mnt->mnt_parent) ||
2607                 IS_MNT_SHARED(root_mnt->mnt_parent))
2608                 goto out4;
2609         if (!check_mnt(root_mnt) || !check_mnt(new_mnt))
2610                 goto out4;
2611         error = -ENOENT;
2612         if (d_unlinked(new.dentry))
2613                 goto out4;
2614         if (d_unlinked(old.dentry))
2615                 goto out4;
2616         error = -EBUSY;
2617         if (new.mnt == root.mnt ||
2618             old.mnt == root.mnt)
2619                 goto out4; /* loop, on the same file system  */
2620         error = -EINVAL;
2621         if (root.mnt->mnt_root != root.dentry)
2622                 goto out4; /* not a mountpoint */
2623         if (!mnt_has_parent(root_mnt))
2624                 goto out4; /* not attached */
2625         if (new.mnt->mnt_root != new.dentry)
2626                 goto out4; /* not a mountpoint */
2627         if (!mnt_has_parent(new_mnt))
2628                 goto out4; /* not attached */
2629         /* make sure we can reach put_old from new_root */
2630         if (!is_path_reachable(real_mount(old.mnt), old.dentry, &new))
2631                 goto out4;
2632         br_write_lock(&vfsmount_lock);
2633         detach_mnt(new_mnt, &parent_path);
2634         detach_mnt(root_mnt, &root_parent);
2635         /* mount old root on put_old */
2636         attach_mnt(root_mnt, &old);
2637         /* mount new_root on / */
2638         attach_mnt(new_mnt, &root_parent);
2639         touch_mnt_namespace(current->nsproxy->mnt_ns);
2640         br_write_unlock(&vfsmount_lock);
2641         chroot_fs_refs(&root, &new);
2642         error = 0;
2643 out4:
2644         unlock_mount(&old);
2645         if (!error) {
2646                 path_put(&root_parent);
2647                 path_put(&parent_path);
2648         }
2649 out3:
2650         path_put(&root);
2651 out2:
2652         path_put(&old);
2653 out1:
2654         path_put(&new);
2655 out0:
2656         return error;
2657 }
2658 
2659 static void __init init_mount_tree(void)
2660 {
2661         struct vfsmount *mnt;
2662         struct mnt_namespace *ns;
2663         struct path root;
2664         struct file_system_type *type;
2665 
2666         type = get_fs_type("rootfs");
2667         if (!type)
2668                 panic("Can't find rootfs type");
2669         mnt = vfs_kern_mount(type, 0, "rootfs", NULL);
2670         put_filesystem(type);
2671         if (IS_ERR(mnt))
2672                 panic("Can't create rootfs");
2673 
2674         ns = create_mnt_ns(mnt);
2675         if (IS_ERR(ns))
2676                 panic("Can't allocate initial namespace");
2677 
2678         init_task.nsproxy->mnt_ns = ns;
2679         get_mnt_ns(ns);
2680 
2681         root.mnt = mnt;
2682         root.dentry = mnt->mnt_root;
2683 
2684         set_fs_pwd(current->fs, &root);
2685         set_fs_root(current->fs, &root);
2686 }
2687 
2688 void __init mnt_init(void)
2689 {
2690         unsigned u;
2691         int err;
2692 
2693         init_rwsem(&namespace_sem);
2694 
2695         mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct mount),
2696                         0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
2697 
2698         mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
2699 
2700         if (!mount_hashtable)
2701                 panic("Failed to allocate mount hash table\n");
2702 
2703         printk(KERN_INFO "Mount-cache hash table entries: %lu\n", HASH_SIZE);
2704 
2705         for (u = 0; u < HASH_SIZE; u++)
2706                 INIT_LIST_HEAD(&mount_hashtable[u]);
2707 
2708         br_lock_init(&vfsmount_lock);
2709 
2710         err = sysfs_init();
2711         if (err)
2712                 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
2713                         __func__, err);
2714         fs_kobj = kobject_create_and_add("fs", NULL);
2715         if (!fs_kobj)
2716                 printk(KERN_WARNING "%s: kobj create error\n", __func__);
2717         init_rootfs();
2718         init_mount_tree();
2719 }
2720 
2721 void put_mnt_ns(struct mnt_namespace *ns)
2722 {
2723         LIST_HEAD(umount_list);
2724 
2725         if (!atomic_dec_and_test(&ns->count))
2726                 return;
2727         down_write(&namespace_sem);
2728         br_write_lock(&vfsmount_lock);
2729         umount_tree(ns->root, 0, &umount_list);
2730         br_write_unlock(&vfsmount_lock);
2731         up_write(&namespace_sem);
2732         release_mounts(&umount_list);
2733         free_mnt_ns(ns);
2734 }
2735 
2736 struct vfsmount *kern_mount_data(struct file_system_type *type, void *data)
2737 {
2738         struct vfsmount *mnt;
2739         mnt = vfs_kern_mount(type, MS_KERNMOUNT, type->name, data);
2740         if (!IS_ERR(mnt)) {
2741                 /*
2742                  * it is a longterm mount, don't release mnt until
2743                  * we unmount before file sys is unregistered
2744                 */
2745                 real_mount(mnt)->mnt_ns = MNT_NS_INTERNAL;
2746         }
2747         return mnt;
2748 }
2749 EXPORT_SYMBOL_GPL(kern_mount_data);
2750 
2751 void kern_unmount(struct vfsmount *mnt)
2752 {
2753         /* release long term mount so mount point can be released */
2754         if (!IS_ERR_OR_NULL(mnt)) {
2755                 br_write_lock(&vfsmount_lock);
2756                 real_mount(mnt)->mnt_ns = NULL;
2757                 br_write_unlock(&vfsmount_lock);
2758                 mntput(mnt);
2759         }
2760 }
2761 EXPORT_SYMBOL(kern_unmount);
2762 
2763 bool our_mnt(struct vfsmount *mnt)
2764 {
2765         return check_mnt(real_mount(mnt));
2766 }
2767 
2768 bool current_chrooted(void)
2769 {
2770         /* Does the current process have a non-standard root */
2771         struct path ns_root;
2772         struct path fs_root;
2773         bool chrooted;
2774 
2775         /* Find the namespace root */
2776         ns_root.mnt = &current->nsproxy->mnt_ns->root->mnt;
2777         ns_root.dentry = ns_root.mnt->mnt_root;
2778         path_get(&ns_root);
2779         while (d_mountpoint(ns_root.dentry) && follow_down_one(&ns_root))
2780                 ;
2781 
2782         get_fs_root(current->fs, &fs_root);
2783 
2784         chrooted = !path_equal(&fs_root, &ns_root);
2785 
2786         path_put(&fs_root);
2787         path_put(&ns_root);
2788 
2789         return chrooted;
2790 }
2791 
2792 void update_mnt_policy(struct user_namespace *userns)
2793 {
2794         struct mnt_namespace *ns = current->nsproxy->mnt_ns;
2795         struct mount *mnt;
2796 
2797         down_read(&namespace_sem);
2798         list_for_each_entry(mnt, &ns->list, mnt_list) {
2799                 switch (mnt->mnt.mnt_sb->s_magic) {
2800                 case SYSFS_MAGIC:
2801                         userns->may_mount_sysfs = true;
2802                         break;
2803                 case PROC_SUPER_MAGIC:
2804                         userns->may_mount_proc = true;
2805                         break;
2806                 }
2807                 if (userns->may_mount_sysfs && userns->may_mount_proc)
2808                         break;
2809         }
2810         up_read(&namespace_sem);
2811 }
2812 
2813 static void *mntns_get(struct task_struct *task)
2814 {
2815         struct mnt_namespace *ns = NULL;
2816         struct nsproxy *nsproxy;
2817 
2818         rcu_read_lock();
2819         nsproxy = task_nsproxy(task);
2820         if (nsproxy) {
2821                 ns = nsproxy->mnt_ns;
2822                 get_mnt_ns(ns);
2823         }
2824         rcu_read_unlock();
2825 
2826         return ns;
2827 }
2828 
2829 static void mntns_put(void *ns)
2830 {
2831         put_mnt_ns(ns);
2832 }
2833 
2834 static int mntns_install(struct nsproxy *nsproxy, void *ns)
2835 {
2836         struct fs_struct *fs = current->fs;
2837         struct mnt_namespace *mnt_ns = ns;
2838         struct path root;
2839 
2840         if (!ns_capable(mnt_ns->user_ns, CAP_SYS_ADMIN) ||
2841             !nsown_capable(CAP_SYS_CHROOT) ||
2842             !nsown_capable(CAP_SYS_ADMIN))
2843                 return -EPERM;
2844 
2845         if (fs->users != 1)
2846                 return -EINVAL;
2847 
2848         get_mnt_ns(mnt_ns);
2849         put_mnt_ns(nsproxy->mnt_ns);
2850         nsproxy->mnt_ns = mnt_ns;
2851 
2852         /* Find the root */
2853         root.mnt    = &mnt_ns->root->mnt;
2854         root.dentry = mnt_ns->root->mnt.mnt_root;
2855         path_get(&root);
2856         while(d_mountpoint(root.dentry) && follow_down_one(&root))
2857                 ;
2858 
2859         /* Update the pwd and root */
2860         set_fs_pwd(fs, &root);
2861         set_fs_root(fs, &root);
2862 
2863         path_put(&root);
2864         return 0;
2865 }
2866 
2867 static unsigned int mntns_inum(void *ns)
2868 {
2869         struct mnt_namespace *mnt_ns = ns;
2870         return mnt_ns->proc_inum;
2871 }
2872 
2873 const struct proc_ns_operations mntns_operations = {
2874         .name           = "mnt",
2875         .type           = CLONE_NEWNS,
2876         .get            = mntns_get,
2877         .put            = mntns_put,
2878         .install        = mntns_install,
2879         .inum           = mntns_inum,
2880 };
2881 

~ [ 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