blob: d9138f81ec10727a0807e8f15a9e798ee93a20dc [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Linus Torvalds1da177e2005-04-16 15:20:36 -070011#include <linux/syscalls.h>
12#include <linux/slab.h>
13#include <linux/sched.h>
14#include <linux/smp_lock.h>
15#include <linux/init.h>
Randy Dunlap15a67dd2006-09-29 01:58:57 -070016#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/acct.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080018#include <linux/capability.h>
Dave Hansen3d733632008-02-15 14:37:59 -080019#include <linux/cpumask.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070020#include <linux/module.h>
Andrew Mortonf20a9ea2006-08-14 22:43:23 -070021#include <linux/sysfs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/seq_file.h>
Kirill Korotaev6b3286e2006-12-08 02:37:56 -080023#include <linux/mnt_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/namei.h>
25#include <linux/security.h>
26#include <linux/mount.h>
David Howells07f3f052006-09-30 20:52:18 +020027#include <linux/ramfs.h>
Eric Dumazet13f14b42008-02-06 01:37:57 -080028#include <linux/log2.h>
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010029#include <linux/idr.h>
Al Viro5ad4e532009-03-29 19:50:06 -040030#include <linux/fs_struct.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <asm/uaccess.h>
32#include <asm/unistd.h>
Ram Pai07b20882005-11-07 17:19:07 -050033#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070034#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070035
Eric Dumazet13f14b42008-02-06 01:37:57 -080036#define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
37#define HASH_SIZE (1UL << HASH_SHIFT)
38
Linus Torvalds1da177e2005-04-16 15:20:36 -070039/* spinlock for vfsmount related operations, inplace of dcache_lock */
Al Viro5addc5d2005-11-07 17:15:49 -050040__cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
41
42static int event;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010043static DEFINE_IDA(mnt_id_ida);
Miklos Szeredi719f5d72008-03-27 13:06:23 +010044static DEFINE_IDA(mnt_group_ida);
Linus Torvalds1da177e2005-04-16 15:20:36 -070045
Eric Dumazetfa3536c2006-03-26 01:37:24 -080046static struct list_head *mount_hashtable __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080047static struct kmem_cache *mnt_cache __read_mostly;
Ram Pai390c6842005-11-07 17:17:51 -050048static struct rw_semaphore namespace_sem;
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080050/* /sys/fs */
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -060051struct kobject *fs_kobj;
52EXPORT_SYMBOL_GPL(fs_kobj);
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080053
Linus Torvalds1da177e2005-04-16 15:20:36 -070054static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
55{
Ram Paib58fed82005-11-07 17:16:09 -050056 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
57 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Eric Dumazet13f14b42008-02-06 01:37:57 -080058 tmp = tmp + (tmp >> HASH_SHIFT);
59 return tmp & (HASH_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070060}
61
Dave Hansen3d733632008-02-15 14:37:59 -080062#define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
63
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010064/* allocation is serialized by namespace_sem */
65static int mnt_alloc_id(struct vfsmount *mnt)
66{
67 int res;
68
69retry:
70 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
71 spin_lock(&vfsmount_lock);
72 res = ida_get_new(&mnt_id_ida, &mnt->mnt_id);
73 spin_unlock(&vfsmount_lock);
74 if (res == -EAGAIN)
75 goto retry;
76
77 return res;
78}
79
80static void mnt_free_id(struct vfsmount *mnt)
81{
82 spin_lock(&vfsmount_lock);
83 ida_remove(&mnt_id_ida, mnt->mnt_id);
84 spin_unlock(&vfsmount_lock);
85}
86
Miklos Szeredi719f5d72008-03-27 13:06:23 +010087/*
88 * Allocate a new peer group ID
89 *
90 * mnt_group_ida is protected by namespace_sem
91 */
92static int mnt_alloc_group_id(struct vfsmount *mnt)
93{
94 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
95 return -ENOMEM;
96
97 return ida_get_new_above(&mnt_group_ida, 1, &mnt->mnt_group_id);
98}
99
100/*
101 * Release a peer group ID
102 */
103void mnt_release_group_id(struct vfsmount *mnt)
104{
105 ida_remove(&mnt_group_ida, mnt->mnt_group_id);
106 mnt->mnt_group_id = 0;
107}
108
Linus Torvalds1da177e2005-04-16 15:20:36 -0700109struct vfsmount *alloc_vfsmnt(const char *name)
110{
Robert P. J. Dayc3762222007-02-10 01:45:03 -0800111 struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700112 if (mnt) {
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100113 int err;
114
115 err = mnt_alloc_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800116 if (err)
117 goto out_free_cache;
118
119 if (name) {
120 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
121 if (!mnt->mnt_devname)
122 goto out_free_id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100123 }
124
Ram Paib58fed82005-11-07 17:16:09 -0500125 atomic_set(&mnt->mnt_count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126 INIT_LIST_HEAD(&mnt->mnt_hash);
127 INIT_LIST_HEAD(&mnt->mnt_child);
128 INIT_LIST_HEAD(&mnt->mnt_mounts);
129 INIT_LIST_HEAD(&mnt->mnt_list);
Miklos Szeredi55e700b2005-07-07 17:57:30 -0700130 INIT_LIST_HEAD(&mnt->mnt_expire);
Ram Pai03e06e62005-11-07 17:19:33 -0500131 INIT_LIST_HEAD(&mnt->mnt_share);
Ram Paia58b0eb2005-11-07 17:20:48 -0500132 INIT_LIST_HEAD(&mnt->mnt_slave_list);
133 INIT_LIST_HEAD(&mnt->mnt_slave);
Dave Hansen3d733632008-02-15 14:37:59 -0800134 atomic_set(&mnt->__mnt_writers, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135 }
136 return mnt;
Li Zefan88b38782008-07-21 18:06:36 +0800137
138out_free_id:
139 mnt_free_id(mnt);
140out_free_cache:
141 kmem_cache_free(mnt_cache, mnt);
142 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143}
144
Dave Hansen83660252008-02-15 14:37:30 -0800145/*
146 * Most r/o checks on a fs are for operations that take
147 * discrete amounts of time, like a write() or unlink().
148 * We must keep track of when those operations start
149 * (for permission checks) and when they end, so that
150 * we can determine when writes are able to occur to
151 * a filesystem.
152 */
Dave Hansen3d733632008-02-15 14:37:59 -0800153/*
154 * __mnt_is_readonly: check whether a mount is read-only
155 * @mnt: the mount to check for its write status
156 *
157 * This shouldn't be used directly ouside of the VFS.
158 * It does not guarantee that the filesystem will stay
159 * r/w, just that it is right *now*. This can not and
160 * should not be used in place of IS_RDONLY(inode).
161 * mnt_want/drop_write() will _keep_ the filesystem
162 * r/w.
163 */
164int __mnt_is_readonly(struct vfsmount *mnt)
165{
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800166 if (mnt->mnt_flags & MNT_READONLY)
167 return 1;
168 if (mnt->mnt_sb->s_flags & MS_RDONLY)
169 return 1;
170 return 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800171}
172EXPORT_SYMBOL_GPL(__mnt_is_readonly);
173
174struct mnt_writer {
175 /*
176 * If holding multiple instances of this lock, they
177 * must be ordered by cpu number.
178 */
179 spinlock_t lock;
180 struct lock_class_key lock_class; /* compiles out with !lockdep */
181 unsigned long count;
182 struct vfsmount *mnt;
183} ____cacheline_aligned_in_smp;
184static DEFINE_PER_CPU(struct mnt_writer, mnt_writers);
185
186static int __init init_mnt_writers(void)
187{
188 int cpu;
189 for_each_possible_cpu(cpu) {
190 struct mnt_writer *writer = &per_cpu(mnt_writers, cpu);
191 spin_lock_init(&writer->lock);
192 lockdep_set_class(&writer->lock, &writer->lock_class);
193 writer->count = 0;
194 }
195 return 0;
196}
197fs_initcall(init_mnt_writers);
198
199static void unlock_mnt_writers(void)
200{
201 int cpu;
202 struct mnt_writer *cpu_writer;
203
204 for_each_possible_cpu(cpu) {
205 cpu_writer = &per_cpu(mnt_writers, cpu);
206 spin_unlock(&cpu_writer->lock);
207 }
208}
209
210static inline void __clear_mnt_count(struct mnt_writer *cpu_writer)
211{
212 if (!cpu_writer->mnt)
213 return;
214 /*
215 * This is in case anyone ever leaves an invalid,
216 * old ->mnt and a count of 0.
217 */
218 if (!cpu_writer->count)
219 return;
220 atomic_add(cpu_writer->count, &cpu_writer->mnt->__mnt_writers);
221 cpu_writer->count = 0;
222}
223 /*
224 * must hold cpu_writer->lock
225 */
226static inline void use_cpu_writer_for_mount(struct mnt_writer *cpu_writer,
227 struct vfsmount *mnt)
228{
229 if (cpu_writer->mnt == mnt)
230 return;
231 __clear_mnt_count(cpu_writer);
232 cpu_writer->mnt = mnt;
233}
234
235/*
236 * Most r/o checks on a fs are for operations that take
237 * discrete amounts of time, like a write() or unlink().
238 * We must keep track of when those operations start
239 * (for permission checks) and when they end, so that
240 * we can determine when writes are able to occur to
241 * a filesystem.
242 */
Dave Hansen83660252008-02-15 14:37:30 -0800243/**
244 * mnt_want_write - get write access to a mount
245 * @mnt: the mount on which to take a write
246 *
247 * This tells the low-level filesystem that a write is
248 * about to be performed to it, and makes sure that
249 * writes are allowed before returning success. When
250 * the write operation is finished, mnt_drop_write()
251 * must be called. This is effectively a refcount.
252 */
253int mnt_want_write(struct vfsmount *mnt)
254{
Dave Hansen3d733632008-02-15 14:37:59 -0800255 int ret = 0;
256 struct mnt_writer *cpu_writer;
257
258 cpu_writer = &get_cpu_var(mnt_writers);
259 spin_lock(&cpu_writer->lock);
260 if (__mnt_is_readonly(mnt)) {
261 ret = -EROFS;
262 goto out;
263 }
264 use_cpu_writer_for_mount(cpu_writer, mnt);
265 cpu_writer->count++;
266out:
267 spin_unlock(&cpu_writer->lock);
268 put_cpu_var(mnt_writers);
269 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800270}
271EXPORT_SYMBOL_GPL(mnt_want_write);
272
Dave Hansen3d733632008-02-15 14:37:59 -0800273static void lock_mnt_writers(void)
274{
275 int cpu;
276 struct mnt_writer *cpu_writer;
277
278 for_each_possible_cpu(cpu) {
279 cpu_writer = &per_cpu(mnt_writers, cpu);
280 spin_lock(&cpu_writer->lock);
281 __clear_mnt_count(cpu_writer);
282 cpu_writer->mnt = NULL;
283 }
284}
285
286/*
287 * These per-cpu write counts are not guaranteed to have
288 * matched increments and decrements on any given cpu.
289 * A file open()ed for write on one cpu and close()d on
290 * another cpu will imbalance this count. Make sure it
291 * does not get too far out of whack.
292 */
293static void handle_write_count_underflow(struct vfsmount *mnt)
294{
295 if (atomic_read(&mnt->__mnt_writers) >=
296 MNT_WRITER_UNDERFLOW_LIMIT)
297 return;
298 /*
299 * It isn't necessary to hold all of the locks
300 * at the same time, but doing it this way makes
301 * us share a lot more code.
302 */
303 lock_mnt_writers();
304 /*
305 * vfsmount_lock is for mnt_flags.
306 */
307 spin_lock(&vfsmount_lock);
308 /*
309 * If coalescing the per-cpu writer counts did not
310 * get us back to a positive writer count, we have
311 * a bug.
312 */
313 if ((atomic_read(&mnt->__mnt_writers) < 0) &&
314 !(mnt->mnt_flags & MNT_IMBALANCED_WRITE_COUNT)) {
Arjan van de Ven5c752ad2008-07-25 19:45:40 -0700315 WARN(1, KERN_DEBUG "leak detected on mount(%p) writers "
Dave Hansen3d733632008-02-15 14:37:59 -0800316 "count: %d\n",
317 mnt, atomic_read(&mnt->__mnt_writers));
Dave Hansen3d733632008-02-15 14:37:59 -0800318 /* use the flag to keep the dmesg spam down */
319 mnt->mnt_flags |= MNT_IMBALANCED_WRITE_COUNT;
320 }
321 spin_unlock(&vfsmount_lock);
322 unlock_mnt_writers();
323}
324
Dave Hansen83660252008-02-15 14:37:30 -0800325/**
326 * mnt_drop_write - give up write access to a mount
327 * @mnt: the mount on which to give up write access
328 *
329 * Tells the low-level filesystem that we are done
330 * performing writes to it. Must be matched with
331 * mnt_want_write() call above.
332 */
333void mnt_drop_write(struct vfsmount *mnt)
334{
Dave Hansen3d733632008-02-15 14:37:59 -0800335 int must_check_underflow = 0;
336 struct mnt_writer *cpu_writer;
337
338 cpu_writer = &get_cpu_var(mnt_writers);
339 spin_lock(&cpu_writer->lock);
340
341 use_cpu_writer_for_mount(cpu_writer, mnt);
342 if (cpu_writer->count > 0) {
343 cpu_writer->count--;
344 } else {
345 must_check_underflow = 1;
346 atomic_dec(&mnt->__mnt_writers);
347 }
348
349 spin_unlock(&cpu_writer->lock);
350 /*
351 * Logically, we could call this each time,
352 * but the __mnt_writers cacheline tends to
353 * be cold, and makes this expensive.
354 */
355 if (must_check_underflow)
356 handle_write_count_underflow(mnt);
357 /*
358 * This could be done right after the spinlock
359 * is taken because the spinlock keeps us on
360 * the cpu, and disables preemption. However,
361 * putting it here bounds the amount that
362 * __mnt_writers can underflow. Without it,
363 * we could theoretically wrap __mnt_writers.
364 */
365 put_cpu_var(mnt_writers);
Dave Hansen83660252008-02-15 14:37:30 -0800366}
367EXPORT_SYMBOL_GPL(mnt_drop_write);
368
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800369static int mnt_make_readonly(struct vfsmount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800370{
Dave Hansen3d733632008-02-15 14:37:59 -0800371 int ret = 0;
372
373 lock_mnt_writers();
374 /*
375 * With all the locks held, this value is stable
376 */
377 if (atomic_read(&mnt->__mnt_writers) > 0) {
378 ret = -EBUSY;
379 goto out;
380 }
381 /*
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800382 * nobody can do a successful mnt_want_write() with all
383 * of the counts in MNT_DENIED_WRITE and the locks held.
Dave Hansen3d733632008-02-15 14:37:59 -0800384 */
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800385 spin_lock(&vfsmount_lock);
386 if (!ret)
387 mnt->mnt_flags |= MNT_READONLY;
388 spin_unlock(&vfsmount_lock);
Dave Hansen3d733632008-02-15 14:37:59 -0800389out:
390 unlock_mnt_writers();
391 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800392}
Dave Hansen83660252008-02-15 14:37:30 -0800393
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800394static void __mnt_unmake_readonly(struct vfsmount *mnt)
395{
396 spin_lock(&vfsmount_lock);
397 mnt->mnt_flags &= ~MNT_READONLY;
398 spin_unlock(&vfsmount_lock);
399}
400
Sukadev Bhattiprolua3ec9472009-03-04 12:06:34 -0800401void simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
David Howells454e2392006-06-23 02:02:57 -0700402{
403 mnt->mnt_sb = sb;
404 mnt->mnt_root = dget(sb->s_root);
David Howells454e2392006-06-23 02:02:57 -0700405}
406
407EXPORT_SYMBOL(simple_set_mnt);
408
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409void free_vfsmnt(struct vfsmount *mnt)
410{
411 kfree(mnt->mnt_devname);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100412 mnt_free_id(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700413 kmem_cache_free(mnt_cache, mnt);
414}
415
416/*
Ram Paia05964f2005-11-07 17:20:17 -0500417 * find the first or last mount at @dentry on vfsmount @mnt depending on
418 * @dir. If @dir is set return the first mount else return the last mount.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 */
Ram Paia05964f2005-11-07 17:20:17 -0500420struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
421 int dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422{
Ram Paib58fed82005-11-07 17:16:09 -0500423 struct list_head *head = mount_hashtable + hash(mnt, dentry);
424 struct list_head *tmp = head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 struct vfsmount *p, *found = NULL;
426
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 for (;;) {
Ram Paia05964f2005-11-07 17:20:17 -0500428 tmp = dir ? tmp->next : tmp->prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 p = NULL;
430 if (tmp == head)
431 break;
432 p = list_entry(tmp, struct vfsmount, mnt_hash);
433 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
Ram Paia05964f2005-11-07 17:20:17 -0500434 found = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435 break;
436 }
437 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700438 return found;
439}
440
Ram Paia05964f2005-11-07 17:20:17 -0500441/*
442 * lookup_mnt increments the ref count before returning
443 * the vfsmount struct.
444 */
445struct vfsmount *lookup_mnt(struct vfsmount *mnt, struct dentry *dentry)
446{
447 struct vfsmount *child_mnt;
448 spin_lock(&vfsmount_lock);
449 if ((child_mnt = __lookup_mnt(mnt, dentry, 1)))
450 mntget(child_mnt);
451 spin_unlock(&vfsmount_lock);
452 return child_mnt;
453}
454
Linus Torvalds1da177e2005-04-16 15:20:36 -0700455static inline int check_mnt(struct vfsmount *mnt)
456{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800457 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700458}
459
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800460static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500461{
462 if (ns) {
463 ns->event = ++event;
464 wake_up_interruptible(&ns->poll);
465 }
466}
467
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800468static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500469{
470 if (ns && ns->event != event) {
471 ns->event = event;
472 wake_up_interruptible(&ns->poll);
473 }
474}
475
Al Viro1a390682008-03-21 20:48:19 -0400476static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700477{
Al Viro1a390682008-03-21 20:48:19 -0400478 old_path->dentry = mnt->mnt_mountpoint;
479 old_path->mnt = mnt->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480 mnt->mnt_parent = mnt;
481 mnt->mnt_mountpoint = mnt->mnt_root;
482 list_del_init(&mnt->mnt_child);
483 list_del_init(&mnt->mnt_hash);
Al Viro1a390682008-03-21 20:48:19 -0400484 old_path->dentry->d_mounted--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485}
486
Ram Paib90fa9a2005-11-07 17:19:50 -0500487void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
488 struct vfsmount *child_mnt)
489{
490 child_mnt->mnt_parent = mntget(mnt);
491 child_mnt->mnt_mountpoint = dget(dentry);
492 dentry->d_mounted++;
493}
494
Al Viro1a390682008-03-21 20:48:19 -0400495static void attach_mnt(struct vfsmount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496{
Al Viro1a390682008-03-21 20:48:19 -0400497 mnt_set_mountpoint(path->mnt, path->dentry, mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500498 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viro1a390682008-03-21 20:48:19 -0400499 hash(path->mnt, path->dentry));
500 list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500501}
502
503/*
504 * the caller must hold vfsmount_lock
505 */
506static void commit_tree(struct vfsmount *mnt)
507{
508 struct vfsmount *parent = mnt->mnt_parent;
509 struct vfsmount *m;
510 LIST_HEAD(head);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800511 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500512
513 BUG_ON(parent == mnt);
514
515 list_add_tail(&head, &mnt->mnt_list);
516 list_for_each_entry(m, &head, mnt_list)
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800517 m->mnt_ns = n;
Ram Paib90fa9a2005-11-07 17:19:50 -0500518 list_splice(&head, n->list.prev);
519
520 list_add_tail(&mnt->mnt_hash, mount_hashtable +
521 hash(parent, mnt->mnt_mountpoint));
522 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800523 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524}
525
526static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
527{
528 struct list_head *next = p->mnt_mounts.next;
529 if (next == &p->mnt_mounts) {
530 while (1) {
531 if (p == root)
532 return NULL;
533 next = p->mnt_child.next;
534 if (next != &p->mnt_parent->mnt_mounts)
535 break;
536 p = p->mnt_parent;
537 }
538 }
539 return list_entry(next, struct vfsmount, mnt_child);
540}
541
Ram Pai9676f0c2005-11-07 17:21:20 -0500542static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
543{
544 struct list_head *prev = p->mnt_mounts.prev;
545 while (prev != &p->mnt_mounts) {
546 p = list_entry(prev, struct vfsmount, mnt_child);
547 prev = p->mnt_mounts.prev;
548 }
549 return p;
550}
551
Ram Pai36341f62005-11-07 17:17:22 -0500552static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
553 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554{
555 struct super_block *sb = old->mnt_sb;
556 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
557
558 if (mnt) {
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100559 if (flag & (CL_SLAVE | CL_PRIVATE))
560 mnt->mnt_group_id = 0; /* not a peer of original */
561 else
562 mnt->mnt_group_id = old->mnt_group_id;
563
564 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
565 int err = mnt_alloc_group_id(mnt);
566 if (err)
567 goto out_free;
568 }
569
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 mnt->mnt_flags = old->mnt_flags;
571 atomic_inc(&sb->s_active);
572 mnt->mnt_sb = sb;
573 mnt->mnt_root = dget(root);
574 mnt->mnt_mountpoint = mnt->mnt_root;
575 mnt->mnt_parent = mnt;
Ram Paib90fa9a2005-11-07 17:19:50 -0500576
Ram Pai5afe0022005-11-07 17:21:01 -0500577 if (flag & CL_SLAVE) {
578 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
579 mnt->mnt_master = old;
580 CLEAR_MNT_SHARED(mnt);
Al Viro8aec0802007-06-07 12:20:32 -0400581 } else if (!(flag & CL_PRIVATE)) {
Ram Pai5afe0022005-11-07 17:21:01 -0500582 if ((flag & CL_PROPAGATION) || IS_MNT_SHARED(old))
583 list_add(&mnt->mnt_share, &old->mnt_share);
584 if (IS_MNT_SLAVE(old))
585 list_add(&mnt->mnt_slave, &old->mnt_slave);
586 mnt->mnt_master = old->mnt_master;
587 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500588 if (flag & CL_MAKE_SHARED)
589 set_mnt_shared(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590
591 /* stick the duplicate mount on the same expiry list
592 * as the original if that was on one */
Ram Pai36341f62005-11-07 17:17:22 -0500593 if (flag & CL_EXPIRE) {
Ram Pai36341f62005-11-07 17:17:22 -0500594 if (!list_empty(&old->mnt_expire))
595 list_add(&mnt->mnt_expire, &old->mnt_expire);
Ram Pai36341f62005-11-07 17:17:22 -0500596 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 }
598 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100599
600 out_free:
601 free_vfsmnt(mnt);
602 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700603}
604
Al Viro7b7b1ac2005-11-07 17:13:39 -0500605static inline void __mntput(struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606{
Dave Hansen3d733632008-02-15 14:37:59 -0800607 int cpu;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 struct super_block *sb = mnt->mnt_sb;
Dave Hansen3d733632008-02-15 14:37:59 -0800609 /*
610 * We don't have to hold all of the locks at the
611 * same time here because we know that we're the
612 * last reference to mnt and that no new writers
613 * can come in.
614 */
615 for_each_possible_cpu(cpu) {
616 struct mnt_writer *cpu_writer = &per_cpu(mnt_writers, cpu);
Dave Hansen3d733632008-02-15 14:37:59 -0800617 spin_lock(&cpu_writer->lock);
Al Viro1a88b532009-02-16 02:38:12 +0000618 if (cpu_writer->mnt != mnt) {
619 spin_unlock(&cpu_writer->lock);
620 continue;
621 }
Dave Hansen3d733632008-02-15 14:37:59 -0800622 atomic_add(cpu_writer->count, &mnt->__mnt_writers);
623 cpu_writer->count = 0;
624 /*
625 * Might as well do this so that no one
626 * ever sees the pointer and expects
627 * it to be valid.
628 */
629 cpu_writer->mnt = NULL;
630 spin_unlock(&cpu_writer->lock);
631 }
632 /*
633 * This probably indicates that somebody messed
634 * up a mnt_want/drop_write() pair. If this
635 * happens, the filesystem was probably unable
636 * to make r/w->r/o transitions.
637 */
638 WARN_ON(atomic_read(&mnt->__mnt_writers));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639 dput(mnt->mnt_root);
640 free_vfsmnt(mnt);
641 deactivate_super(sb);
642}
643
Al Viro7b7b1ac2005-11-07 17:13:39 -0500644void mntput_no_expire(struct vfsmount *mnt)
645{
646repeat:
647 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
648 if (likely(!mnt->mnt_pinned)) {
649 spin_unlock(&vfsmount_lock);
650 __mntput(mnt);
651 return;
652 }
653 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
654 mnt->mnt_pinned = 0;
655 spin_unlock(&vfsmount_lock);
656 acct_auto_close_mnt(mnt);
657 security_sb_umount_close(mnt);
658 goto repeat;
659 }
660}
661
662EXPORT_SYMBOL(mntput_no_expire);
663
664void mnt_pin(struct vfsmount *mnt)
665{
666 spin_lock(&vfsmount_lock);
667 mnt->mnt_pinned++;
668 spin_unlock(&vfsmount_lock);
669}
670
671EXPORT_SYMBOL(mnt_pin);
672
673void mnt_unpin(struct vfsmount *mnt)
674{
675 spin_lock(&vfsmount_lock);
676 if (mnt->mnt_pinned) {
677 atomic_inc(&mnt->mnt_count);
678 mnt->mnt_pinned--;
679 }
680 spin_unlock(&vfsmount_lock);
681}
682
683EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800685static inline void mangle(struct seq_file *m, const char *s)
686{
687 seq_escape(m, s, " \t\n\\");
688}
689
690/*
691 * Simple .show_options callback for filesystems which don't want to
692 * implement more complex mount option showing.
693 *
694 * See also save_mount_options().
695 */
696int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
697{
698 const char *options = mnt->mnt_sb->s_options;
699
700 if (options != NULL && options[0]) {
701 seq_putc(m, ',');
702 mangle(m, options);
703 }
704
705 return 0;
706}
707EXPORT_SYMBOL(generic_show_options);
708
709/*
710 * If filesystem uses generic_show_options(), this function should be
711 * called from the fill_super() callback.
712 *
713 * The .remount_fs callback usually needs to be handled in a special
714 * way, to make sure, that previous options are not overwritten if the
715 * remount fails.
716 *
717 * Also note, that if the filesystem's .remount_fs function doesn't
718 * reset all options to their default value, but changes only newly
719 * given options, then the displayed options will not reflect reality
720 * any more.
721 */
722void save_mount_options(struct super_block *sb, char *options)
723{
724 kfree(sb->s_options);
725 sb->s_options = kstrdup(options, GFP_KERNEL);
726}
727EXPORT_SYMBOL(save_mount_options);
728
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100729#ifdef CONFIG_PROC_FS
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730/* iterator */
731static void *m_start(struct seq_file *m, loff_t *pos)
732{
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100733 struct proc_mounts *p = m->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734
Ram Pai390c6842005-11-07 17:17:51 -0500735 down_read(&namespace_sem);
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100736 return seq_list_start(&p->ns->list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737}
738
739static void *m_next(struct seq_file *m, void *v, loff_t *pos)
740{
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100741 struct proc_mounts *p = m->private;
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700742
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100743 return seq_list_next(v, &p->ns->list, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744}
745
746static void m_stop(struct seq_file *m, void *v)
747{
Ram Pai390c6842005-11-07 17:17:51 -0500748 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749}
750
Ram Pai2d4d4862008-03-27 13:06:25 +0100751struct proc_fs_info {
752 int flag;
753 const char *str;
754};
755
Eric Paris2069f452008-07-04 09:47:13 +1000756static int show_sb_opts(struct seq_file *m, struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757{
Ram Pai2d4d4862008-03-27 13:06:25 +0100758 static const struct proc_fs_info fs_info[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700759 { MS_SYNCHRONOUS, ",sync" },
760 { MS_DIRSYNC, ",dirsync" },
761 { MS_MANDLOCK, ",mand" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 { 0, NULL }
763 };
Ram Pai2d4d4862008-03-27 13:06:25 +0100764 const struct proc_fs_info *fs_infop;
765
766 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
767 if (sb->s_flags & fs_infop->flag)
768 seq_puts(m, fs_infop->str);
769 }
Eric Paris2069f452008-07-04 09:47:13 +1000770
771 return security_sb_show_options(m, sb);
Ram Pai2d4d4862008-03-27 13:06:25 +0100772}
773
774static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
775{
776 static const struct proc_fs_info mnt_info[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 { MNT_NOSUID, ",nosuid" },
778 { MNT_NODEV, ",nodev" },
779 { MNT_NOEXEC, ",noexec" },
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -0800780 { MNT_NOATIME, ",noatime" },
781 { MNT_NODIRATIME, ",nodiratime" },
Valerie Henson47ae32d2006-12-13 00:34:34 -0800782 { MNT_RELATIME, ",relatime" },
Matthew Garrettd0adde52009-03-26 17:49:56 +0000783 { MNT_STRICTATIME, ",strictatime" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 { 0, NULL }
785 };
Ram Pai2d4d4862008-03-27 13:06:25 +0100786 const struct proc_fs_info *fs_infop;
787
788 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
789 if (mnt->mnt_flags & fs_infop->flag)
790 seq_puts(m, fs_infop->str);
791 }
792}
793
794static void show_type(struct seq_file *m, struct super_block *sb)
795{
796 mangle(m, sb->s_type->name);
797 if (sb->s_subtype && sb->s_subtype[0]) {
798 seq_putc(m, '.');
799 mangle(m, sb->s_subtype);
800 }
801}
802
803static int show_vfsmnt(struct seq_file *m, void *v)
804{
805 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
806 int err = 0;
Jan Blunckc32c2f62008-02-14 19:38:43 -0800807 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700808
809 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
810 seq_putc(m, ' ');
Jan Blunckc32c2f62008-02-14 19:38:43 -0800811 seq_path(m, &mnt_path, " \t\n\\");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700812 seq_putc(m, ' ');
Ram Pai2d4d4862008-03-27 13:06:25 +0100813 show_type(m, mnt->mnt_sb);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800814 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
Eric Paris2069f452008-07-04 09:47:13 +1000815 err = show_sb_opts(m, mnt->mnt_sb);
816 if (err)
817 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +0100818 show_mnt_opts(m, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 if (mnt->mnt_sb->s_op->show_options)
820 err = mnt->mnt_sb->s_op->show_options(m, mnt);
821 seq_puts(m, " 0 0\n");
Eric Paris2069f452008-07-04 09:47:13 +1000822out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700823 return err;
824}
825
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100826const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827 .start = m_start,
828 .next = m_next,
829 .stop = m_stop,
830 .show = show_vfsmnt
831};
832
Ram Pai2d4d4862008-03-27 13:06:25 +0100833static int show_mountinfo(struct seq_file *m, void *v)
834{
835 struct proc_mounts *p = m->private;
836 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
837 struct super_block *sb = mnt->mnt_sb;
838 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
839 struct path root = p->root;
840 int err = 0;
841
842 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
843 MAJOR(sb->s_dev), MINOR(sb->s_dev));
844 seq_dentry(m, mnt->mnt_root, " \t\n\\");
845 seq_putc(m, ' ');
846 seq_path_root(m, &mnt_path, &root, " \t\n\\");
847 if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
848 /*
849 * Mountpoint is outside root, discard that one. Ugly,
850 * but less so than trying to do that in iterator in a
851 * race-free way (due to renames).
852 */
853 return SEQ_SKIP;
854 }
855 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
856 show_mnt_opts(m, mnt);
857
858 /* Tagged fields ("foo:X" or "bar") */
859 if (IS_MNT_SHARED(mnt))
860 seq_printf(m, " shared:%i", mnt->mnt_group_id);
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +0100861 if (IS_MNT_SLAVE(mnt)) {
862 int master = mnt->mnt_master->mnt_group_id;
863 int dom = get_dominating_id(mnt, &p->root);
864 seq_printf(m, " master:%i", master);
865 if (dom && dom != master)
866 seq_printf(m, " propagate_from:%i", dom);
867 }
Ram Pai2d4d4862008-03-27 13:06:25 +0100868 if (IS_MNT_UNBINDABLE(mnt))
869 seq_puts(m, " unbindable");
870
871 /* Filesystem specific data */
872 seq_puts(m, " - ");
873 show_type(m, sb);
874 seq_putc(m, ' ');
875 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
876 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
Eric Paris2069f452008-07-04 09:47:13 +1000877 err = show_sb_opts(m, sb);
878 if (err)
879 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +0100880 if (sb->s_op->show_options)
881 err = sb->s_op->show_options(m, mnt);
882 seq_putc(m, '\n');
Eric Paris2069f452008-07-04 09:47:13 +1000883out:
Ram Pai2d4d4862008-03-27 13:06:25 +0100884 return err;
885}
886
887const struct seq_operations mountinfo_op = {
888 .start = m_start,
889 .next = m_next,
890 .stop = m_stop,
891 .show = show_mountinfo,
892};
893
Chuck Leverb4629fe2006-03-20 13:44:12 -0500894static int show_vfsstat(struct seq_file *m, void *v)
895{
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700896 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
Jan Blunckc32c2f62008-02-14 19:38:43 -0800897 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Chuck Leverb4629fe2006-03-20 13:44:12 -0500898 int err = 0;
899
900 /* device */
901 if (mnt->mnt_devname) {
902 seq_puts(m, "device ");
903 mangle(m, mnt->mnt_devname);
904 } else
905 seq_puts(m, "no device");
906
907 /* mount point */
908 seq_puts(m, " mounted on ");
Jan Blunckc32c2f62008-02-14 19:38:43 -0800909 seq_path(m, &mnt_path, " \t\n\\");
Chuck Leverb4629fe2006-03-20 13:44:12 -0500910 seq_putc(m, ' ');
911
912 /* file system type */
913 seq_puts(m, "with fstype ");
Ram Pai2d4d4862008-03-27 13:06:25 +0100914 show_type(m, mnt->mnt_sb);
Chuck Leverb4629fe2006-03-20 13:44:12 -0500915
916 /* optional statistics */
917 if (mnt->mnt_sb->s_op->show_stats) {
918 seq_putc(m, ' ');
919 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
920 }
921
922 seq_putc(m, '\n');
923 return err;
924}
925
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100926const struct seq_operations mountstats_op = {
Chuck Leverb4629fe2006-03-20 13:44:12 -0500927 .start = m_start,
928 .next = m_next,
929 .stop = m_stop,
930 .show = show_vfsstat,
931};
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100932#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -0500933
Linus Torvalds1da177e2005-04-16 15:20:36 -0700934/**
935 * may_umount_tree - check if a mount tree is busy
936 * @mnt: root of mount tree
937 *
938 * This is called to check if a tree of mounts has any
939 * open files, pwds, chroots or sub mounts that are
940 * busy.
941 */
942int may_umount_tree(struct vfsmount *mnt)
943{
Ram Pai36341f62005-11-07 17:17:22 -0500944 int actual_refs = 0;
945 int minimum_refs = 0;
946 struct vfsmount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947
948 spin_lock(&vfsmount_lock);
Ram Pai36341f62005-11-07 17:17:22 -0500949 for (p = mnt; p; p = next_mnt(p, mnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950 actual_refs += atomic_read(&p->mnt_count);
951 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952 }
953 spin_unlock(&vfsmount_lock);
954
955 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -0800956 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957
Ian Kente3474a82006-03-27 01:14:51 -0800958 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959}
960
961EXPORT_SYMBOL(may_umount_tree);
962
963/**
964 * may_umount - check if a mount point is busy
965 * @mnt: root of mount
966 *
967 * This is called to check if a mount point has any
968 * open files, pwds, chroots or sub mounts. If the
969 * mount has sub mounts this will return busy
970 * regardless of whether the sub mounts are busy.
971 *
972 * Doesn't take quota and stuff into account. IOW, in some cases it will
973 * give false negatives. The main reason why it's here is that we need
974 * a non-destructive way to look for easily umountable filesystems.
975 */
976int may_umount(struct vfsmount *mnt)
977{
Ian Kente3474a82006-03-27 01:14:51 -0800978 int ret = 1;
Ram Paia05964f2005-11-07 17:20:17 -0500979 spin_lock(&vfsmount_lock);
980 if (propagate_mount_busy(mnt, 2))
Ian Kente3474a82006-03-27 01:14:51 -0800981 ret = 0;
Ram Paia05964f2005-11-07 17:20:17 -0500982 spin_unlock(&vfsmount_lock);
983 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984}
985
986EXPORT_SYMBOL(may_umount);
987
Ram Paib90fa9a2005-11-07 17:19:50 -0500988void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700989{
Ram Pai70fbcdf2005-11-07 17:17:04 -0500990 struct vfsmount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -0800991 while (!list_empty(head)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -0700992 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
Ram Pai70fbcdf2005-11-07 17:17:04 -0500993 list_del_init(&mnt->mnt_hash);
994 if (mnt->mnt_parent != mnt) {
995 struct dentry *dentry;
996 struct vfsmount *m;
997 spin_lock(&vfsmount_lock);
998 dentry = mnt->mnt_mountpoint;
999 m = mnt->mnt_parent;
1000 mnt->mnt_mountpoint = mnt->mnt_root;
1001 mnt->mnt_parent = mnt;
Al Viro7c4b93d2008-03-21 23:59:49 -04001002 m->mnt_ghosts--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001003 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001004 dput(dentry);
1005 mntput(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 }
1007 mntput(mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001008 }
1009}
1010
Ram Paia05964f2005-11-07 17:20:17 -05001011void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001012{
1013 struct vfsmount *p;
1014
Akinobu Mita1bfba4e2006-06-26 00:24:40 -07001015 for (p = mnt; p; p = next_mnt(p, mnt))
1016 list_move(&p->mnt_hash, kill);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001017
Ram Paia05964f2005-11-07 17:20:17 -05001018 if (propagate)
1019 propagate_umount(kill);
1020
Ram Pai70fbcdf2005-11-07 17:17:04 -05001021 list_for_each_entry(p, kill, mnt_hash) {
1022 list_del_init(&p->mnt_expire);
1023 list_del_init(&p->mnt_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001024 __touch_mnt_namespace(p->mnt_ns);
1025 p->mnt_ns = NULL;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001026 list_del_init(&p->mnt_child);
Al Viro7c4b93d2008-03-21 23:59:49 -04001027 if (p->mnt_parent != p) {
1028 p->mnt_parent->mnt_ghosts++;
Al Virof30ac312006-02-01 07:53:21 -05001029 p->mnt_mountpoint->d_mounted--;
Al Viro7c4b93d2008-03-21 23:59:49 -04001030 }
Ram Paia05964f2005-11-07 17:20:17 -05001031 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032 }
1033}
1034
Al Viroc35038b2008-03-22 00:46:23 -04001035static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1036
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037static int do_umount(struct vfsmount *mnt, int flags)
1038{
Ram Paib58fed82005-11-07 17:16:09 -05001039 struct super_block *sb = mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001041 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042
1043 retval = security_sb_umount(mnt, flags);
1044 if (retval)
1045 return retval;
1046
1047 /*
1048 * Allow userspace to request a mountpoint be expired rather than
1049 * unmounting unconditionally. Unmount only happens if:
1050 * (1) the mark is already set (the mark is cleared by mntput())
1051 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1052 */
1053 if (flags & MNT_EXPIRE) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08001054 if (mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 flags & (MNT_FORCE | MNT_DETACH))
1056 return -EINVAL;
1057
1058 if (atomic_read(&mnt->mnt_count) != 2)
1059 return -EBUSY;
1060
1061 if (!xchg(&mnt->mnt_expiry_mark, 1))
1062 return -EAGAIN;
1063 }
1064
1065 /*
1066 * If we may have to abort operations to get out of this
1067 * mount, and they will themselves hold resources we must
1068 * allow the fs to do things. In the Unix tradition of
1069 * 'Gee thats tricky lets do it in userspace' the umount_begin
1070 * might fail to complete on the first run through as other tasks
1071 * must return, and the like. Thats for the mount program to worry
1072 * about for the moment.
1073 */
1074
Al Viro42faad92008-04-24 07:21:56 -04001075 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
1076 lock_kernel();
1077 sb->s_op->umount_begin(sb);
1078 unlock_kernel();
1079 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080
1081 /*
1082 * No sense to grab the lock for this test, but test itself looks
1083 * somewhat bogus. Suggestions for better replacement?
1084 * Ho-hum... In principle, we might treat that as umount + switch
1085 * to rootfs. GC would eventually take care of the old vfsmount.
1086 * Actually it makes sense, especially if rootfs would contain a
1087 * /reboot - static binary that would close all descriptors and
1088 * call reboot(9). Then init(8) could umount root and exec /reboot.
1089 */
Jan Blunck6ac08c32008-02-14 19:34:38 -08001090 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091 /*
1092 * Special case for "unmounting" root ...
1093 * we just try to remount it readonly.
1094 */
1095 down_write(&sb->s_umount);
1096 if (!(sb->s_flags & MS_RDONLY)) {
1097 lock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
1099 unlock_kernel();
1100 }
1101 up_write(&sb->s_umount);
1102 return retval;
1103 }
1104
Ram Pai390c6842005-11-07 17:17:51 -05001105 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106 spin_lock(&vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -05001107 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108
Al Viroc35038b2008-03-22 00:46:23 -04001109 if (!(flags & MNT_DETACH))
1110 shrink_submounts(mnt, &umount_list);
1111
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -05001113 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001114 if (!list_empty(&mnt->mnt_list))
Ram Paia05964f2005-11-07 17:20:17 -05001115 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 retval = 0;
1117 }
1118 spin_unlock(&vfsmount_lock);
1119 if (retval)
1120 security_sb_umount_busy(mnt);
Ram Pai390c6842005-11-07 17:17:51 -05001121 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001122 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 return retval;
1124}
1125
1126/*
1127 * Now umount can handle mount points as well as block devices.
1128 * This is important for filesystems which use unnamed block devices.
1129 *
1130 * We now support a flag for forced unmount like the other 'big iron'
1131 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1132 */
1133
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001134SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001135{
Al Viro2d8f3032008-07-22 09:59:21 -04001136 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 int retval;
1138
Al Viro2d8f3032008-07-22 09:59:21 -04001139 retval = user_path(name, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 if (retval)
1141 goto out;
1142 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001143 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 goto dput_and_out;
Al Viro2d8f3032008-07-22 09:59:21 -04001145 if (!check_mnt(path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146 goto dput_and_out;
1147
1148 retval = -EPERM;
1149 if (!capable(CAP_SYS_ADMIN))
1150 goto dput_and_out;
1151
Al Viro2d8f3032008-07-22 09:59:21 -04001152 retval = do_umount(path.mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001154 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001155 dput(path.dentry);
1156 mntput_no_expire(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157out:
1158 return retval;
1159}
1160
1161#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1162
1163/*
Ram Paib58fed82005-11-07 17:16:09 -05001164 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001165 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001166SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167{
Ram Paib58fed82005-11-07 17:16:09 -05001168 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169}
1170
1171#endif
1172
Al Viro2d92ab32008-08-02 00:51:11 -04001173static int mount_is_safe(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174{
1175 if (capable(CAP_SYS_ADMIN))
1176 return 0;
1177 return -EPERM;
1178#ifdef notyet
Al Viro2d92ab32008-08-02 00:51:11 -04001179 if (S_ISLNK(path->dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180 return -EPERM;
Al Viro2d92ab32008-08-02 00:51:11 -04001181 if (path->dentry->d_inode->i_mode & S_ISVTX) {
David Howellsda9592e2008-11-14 10:39:05 +11001182 if (current_uid() != path->dentry->d_inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183 return -EPERM;
1184 }
Al Viro2d92ab32008-08-02 00:51:11 -04001185 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186 return -EPERM;
1187 return 0;
1188#endif
1189}
1190
Ram Paib90fa9a2005-11-07 17:19:50 -05001191struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001192 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193{
1194 struct vfsmount *res, *p, *q, *r, *s;
Al Viro1a390682008-03-21 20:48:19 -04001195 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196
Ram Pai9676f0c2005-11-07 17:21:20 -05001197 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1198 return NULL;
1199
Ram Pai36341f62005-11-07 17:17:22 -05001200 res = q = clone_mnt(mnt, dentry, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201 if (!q)
1202 goto Enomem;
1203 q->mnt_mountpoint = mnt->mnt_mountpoint;
1204
1205 p = mnt;
Domen Puncerfdadd652005-09-10 00:27:07 -07001206 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001207 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 continue;
1209
1210 for (s = r; s; s = next_mnt(s, r)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001211 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1212 s = skip_mnt_tree(s);
1213 continue;
1214 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215 while (p != s->mnt_parent) {
1216 p = p->mnt_parent;
1217 q = q->mnt_parent;
1218 }
1219 p = s;
Al Viro1a390682008-03-21 20:48:19 -04001220 path.mnt = q;
1221 path.dentry = p->mnt_mountpoint;
Ram Pai36341f62005-11-07 17:17:22 -05001222 q = clone_mnt(p, p->mnt_root, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 if (!q)
1224 goto Enomem;
1225 spin_lock(&vfsmount_lock);
1226 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Viro1a390682008-03-21 20:48:19 -04001227 attach_mnt(q, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 spin_unlock(&vfsmount_lock);
1229 }
1230 }
1231 return res;
Ram Paib58fed82005-11-07 17:16:09 -05001232Enomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001234 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001236 umount_tree(res, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001238 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001239 }
1240 return NULL;
1241}
1242
Al Viro8aec0802007-06-07 12:20:32 -04001243struct vfsmount *collect_mounts(struct vfsmount *mnt, struct dentry *dentry)
1244{
1245 struct vfsmount *tree;
Al Viro1a60a282008-03-22 16:19:49 -04001246 down_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001247 tree = copy_tree(mnt, dentry, CL_COPY_ALL | CL_PRIVATE);
Al Viro1a60a282008-03-22 16:19:49 -04001248 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001249 return tree;
1250}
1251
1252void drop_collected_mounts(struct vfsmount *mnt)
1253{
1254 LIST_HEAD(umount_list);
Al Viro1a60a282008-03-22 16:19:49 -04001255 down_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001256 spin_lock(&vfsmount_lock);
1257 umount_tree(mnt, 0, &umount_list);
1258 spin_unlock(&vfsmount_lock);
Al Viro1a60a282008-03-22 16:19:49 -04001259 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001260 release_mounts(&umount_list);
1261}
1262
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001263static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1264{
1265 struct vfsmount *p;
1266
1267 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1268 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1269 mnt_release_group_id(p);
1270 }
1271}
1272
1273static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1274{
1275 struct vfsmount *p;
1276
1277 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1278 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1279 int err = mnt_alloc_group_id(p);
1280 if (err) {
1281 cleanup_group_ids(mnt, p);
1282 return err;
1283 }
1284 }
1285 }
1286
1287 return 0;
1288}
1289
Ram Paib90fa9a2005-11-07 17:19:50 -05001290/*
1291 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001292 * @nd : place the mount tree @source_mnt is attached
1293 * @parent_nd : if non-null, detach the source_mnt from its parent and
1294 * store the parent mount and mountpoint dentry.
1295 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001296 *
1297 * NOTE: in the table below explains the semantics when a source mount
1298 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001299 * ---------------------------------------------------------------------------
1300 * | BIND MOUNT OPERATION |
1301 * |**************************************************************************
1302 * | source-->| shared | private | slave | unbindable |
1303 * | dest | | | | |
1304 * | | | | | | |
1305 * | v | | | | |
1306 * |**************************************************************************
1307 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1308 * | | | | | |
1309 * |non-shared| shared (+) | private | slave (*) | invalid |
1310 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001311 * A bind operation clones the source mount and mounts the clone on the
1312 * destination mount.
1313 *
1314 * (++) the cloned mount is propagated to all the mounts in the propagation
1315 * tree of the destination mount and the cloned mount is added to
1316 * the peer group of the source mount.
1317 * (+) the cloned mount is created under the destination mount and is marked
1318 * as shared. The cloned mount is added to the peer group of the source
1319 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001320 * (+++) the mount is propagated to all the mounts in the propagation tree
1321 * of the destination mount and the cloned mount is made slave
1322 * of the same master as that of the source mount. The cloned mount
1323 * is marked as 'shared and slave'.
1324 * (*) the cloned mount is made a slave of the same master as that of the
1325 * source mount.
1326 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001327 * ---------------------------------------------------------------------------
1328 * | MOVE MOUNT OPERATION |
1329 * |**************************************************************************
1330 * | source-->| shared | private | slave | unbindable |
1331 * | dest | | | | |
1332 * | | | | | | |
1333 * | v | | | | |
1334 * |**************************************************************************
1335 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1336 * | | | | | |
1337 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1338 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001339 *
1340 * (+) the mount is moved to the destination. And is then propagated to
1341 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001342 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001343 * (+++) the mount is moved to the destination and is then propagated to
1344 * all the mounts belonging to the destination mount's propagation tree.
1345 * the mount is marked as 'shared and slave'.
1346 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001347 *
1348 * if the source mount is a tree, the operations explained above is
1349 * applied to each mount in the tree.
1350 * Must be called without spinlocks held, since this function can sleep
1351 * in allocations.
1352 */
1353static int attach_recursive_mnt(struct vfsmount *source_mnt,
Al Viro1a390682008-03-21 20:48:19 -04001354 struct path *path, struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001355{
1356 LIST_HEAD(tree_list);
Al Viro1a390682008-03-21 20:48:19 -04001357 struct vfsmount *dest_mnt = path->mnt;
1358 struct dentry *dest_dentry = path->dentry;
Ram Paib90fa9a2005-11-07 17:19:50 -05001359 struct vfsmount *child, *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001360 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001361
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001362 if (IS_MNT_SHARED(dest_mnt)) {
1363 err = invent_group_ids(source_mnt, true);
1364 if (err)
1365 goto out;
1366 }
1367 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1368 if (err)
1369 goto out_cleanup_ids;
Ram Paib90fa9a2005-11-07 17:19:50 -05001370
1371 if (IS_MNT_SHARED(dest_mnt)) {
1372 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1373 set_mnt_shared(p);
1374 }
1375
1376 spin_lock(&vfsmount_lock);
Al Viro1a390682008-03-21 20:48:19 -04001377 if (parent_path) {
1378 detach_mnt(source_mnt, parent_path);
1379 attach_mnt(source_mnt, path);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001380 touch_mnt_namespace(current->nsproxy->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001381 } else {
1382 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1383 commit_tree(source_mnt);
1384 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001385
1386 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1387 list_del_init(&child->mnt_hash);
1388 commit_tree(child);
1389 }
1390 spin_unlock(&vfsmount_lock);
1391 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001392
1393 out_cleanup_ids:
1394 if (IS_MNT_SHARED(dest_mnt))
1395 cleanup_group_ids(source_mnt, NULL);
1396 out:
1397 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001398}
1399
Al Viro8c3ee422008-03-22 18:00:39 -04001400static int graft_tree(struct vfsmount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401{
1402 int err;
1403 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1404 return -EINVAL;
1405
Al Viro8c3ee422008-03-22 18:00:39 -04001406 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1408 return -ENOTDIR;
1409
1410 err = -ENOENT;
Al Viro8c3ee422008-03-22 18:00:39 -04001411 mutex_lock(&path->dentry->d_inode->i_mutex);
1412 if (IS_DEADDIR(path->dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413 goto out_unlock;
1414
Al Viro8c3ee422008-03-22 18:00:39 -04001415 err = security_sb_check_sb(mnt, path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 if (err)
1417 goto out_unlock;
1418
1419 err = -ENOENT;
Al Viro8c3ee422008-03-22 18:00:39 -04001420 if (IS_ROOT(path->dentry) || !d_unhashed(path->dentry))
1421 err = attach_recursive_mnt(mnt, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422out_unlock:
Al Viro8c3ee422008-03-22 18:00:39 -04001423 mutex_unlock(&path->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 if (!err)
Al Viro8c3ee422008-03-22 18:00:39 -04001425 security_sb_post_addmount(mnt, path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 return err;
1427}
1428
1429/*
Ram Pai07b20882005-11-07 17:19:07 -05001430 * recursively change the type of the mountpoint.
1431 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001432static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05001433{
Al Viro2d92ab32008-08-02 00:51:11 -04001434 struct vfsmount *m, *mnt = path->mnt;
Ram Pai07b20882005-11-07 17:19:07 -05001435 int recurse = flag & MS_REC;
1436 int type = flag & ~MS_REC;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001437 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05001438
Miklos Szerediee6f9582007-05-08 00:30:40 -07001439 if (!capable(CAP_SYS_ADMIN))
1440 return -EPERM;
1441
Al Viro2d92ab32008-08-02 00:51:11 -04001442 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05001443 return -EINVAL;
1444
1445 down_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001446 if (type == MS_SHARED) {
1447 err = invent_group_ids(mnt, recurse);
1448 if (err)
1449 goto out_unlock;
1450 }
1451
Ram Pai07b20882005-11-07 17:19:07 -05001452 spin_lock(&vfsmount_lock);
1453 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1454 change_mnt_propagation(m, type);
1455 spin_unlock(&vfsmount_lock);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001456
1457 out_unlock:
Ram Pai07b20882005-11-07 17:19:07 -05001458 up_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001459 return err;
Ram Pai07b20882005-11-07 17:19:07 -05001460}
1461
1462/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 * do loopback mount.
1464 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001465static int do_loopback(struct path *path, char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001466 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467{
Al Viro2d92ab32008-08-02 00:51:11 -04001468 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 struct vfsmount *mnt = NULL;
Al Viro2d92ab32008-08-02 00:51:11 -04001470 int err = mount_is_safe(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 if (err)
1472 return err;
1473 if (!old_name || !*old_name)
1474 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001475 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001476 if (err)
1477 return err;
1478
Ram Pai390c6842005-11-07 17:17:51 -05001479 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001481 if (IS_MNT_UNBINDABLE(old_path.mnt))
Jan Blunck4ac91372008-02-14 19:34:32 -08001482 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05001483
Al Viro2d92ab32008-08-02 00:51:11 -04001484 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
Al Viroccd48bc2005-11-07 17:15:04 -05001485 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486
Al Viroccd48bc2005-11-07 17:15:04 -05001487 err = -ENOMEM;
1488 if (recurse)
Al Viro2d92ab32008-08-02 00:51:11 -04001489 mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001490 else
Al Viro2d92ab32008-08-02 00:51:11 -04001491 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001492
1493 if (!mnt)
1494 goto out;
1495
Al Viro2d92ab32008-08-02 00:51:11 -04001496 err = graft_tree(mnt, path);
Al Viroccd48bc2005-11-07 17:15:04 -05001497 if (err) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001498 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001500 umount_tree(mnt, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001502 release_mounts(&umount_list);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001503 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504
Al Viroccd48bc2005-11-07 17:15:04 -05001505out:
Ram Pai390c6842005-11-07 17:17:51 -05001506 up_write(&namespace_sem);
Al Viro2d92ab32008-08-02 00:51:11 -04001507 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 return err;
1509}
1510
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001511static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1512{
1513 int error = 0;
1514 int readonly_request = 0;
1515
1516 if (ms_flags & MS_RDONLY)
1517 readonly_request = 1;
1518 if (readonly_request == __mnt_is_readonly(mnt))
1519 return 0;
1520
1521 if (readonly_request)
1522 error = mnt_make_readonly(mnt);
1523 else
1524 __mnt_unmake_readonly(mnt);
1525 return error;
1526}
1527
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528/*
1529 * change filesystem flags. dir should be a physical root of filesystem.
1530 * If you've mounted a non-root directory somewhere and want to do remount
1531 * on it - tough luck.
1532 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001533static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534 void *data)
1535{
1536 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04001537 struct super_block *sb = path->mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538
1539 if (!capable(CAP_SYS_ADMIN))
1540 return -EPERM;
1541
Al Viro2d92ab32008-08-02 00:51:11 -04001542 if (!check_mnt(path->mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 return -EINVAL;
1544
Al Viro2d92ab32008-08-02 00:51:11 -04001545 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546 return -EINVAL;
1547
1548 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001549 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001550 err = change_mount_flags(path->mnt, flags);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001551 else
1552 err = do_remount_sb(sb, flags, data, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 if (!err)
Al Viro2d92ab32008-08-02 00:51:11 -04001554 path->mnt->mnt_flags = mnt_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555 up_write(&sb->s_umount);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001556 if (!err) {
Al Viro2d92ab32008-08-02 00:51:11 -04001557 security_sb_post_remount(path->mnt, flags, data);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001558
1559 spin_lock(&vfsmount_lock);
1560 touch_mnt_namespace(path->mnt->mnt_ns);
1561 spin_unlock(&vfsmount_lock);
1562 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 return err;
1564}
1565
Ram Pai9676f0c2005-11-07 17:21:20 -05001566static inline int tree_contains_unbindable(struct vfsmount *mnt)
1567{
1568 struct vfsmount *p;
1569 for (p = mnt; p; p = next_mnt(p, mnt)) {
1570 if (IS_MNT_UNBINDABLE(p))
1571 return 1;
1572 }
1573 return 0;
1574}
1575
Al Viro0a0d8a42008-08-02 00:55:27 -04001576static int do_move_mount(struct path *path, char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577{
Al Viro2d92ab32008-08-02 00:51:11 -04001578 struct path old_path, parent_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 struct vfsmount *p;
1580 int err = 0;
1581 if (!capable(CAP_SYS_ADMIN))
1582 return -EPERM;
1583 if (!old_name || !*old_name)
1584 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001585 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 if (err)
1587 return err;
1588
Ram Pai390c6842005-11-07 17:17:51 -05001589 down_write(&namespace_sem);
Al Viro2d92ab32008-08-02 00:51:11 -04001590 while (d_mountpoint(path->dentry) &&
1591 follow_down(&path->mnt, &path->dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 ;
1593 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001594 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595 goto out;
1596
1597 err = -ENOENT;
Al Viro2d92ab32008-08-02 00:51:11 -04001598 mutex_lock(&path->dentry->d_inode->i_mutex);
1599 if (IS_DEADDIR(path->dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001600 goto out1;
1601
Al Viro2d92ab32008-08-02 00:51:11 -04001602 if (!IS_ROOT(path->dentry) && d_unhashed(path->dentry))
Ram Pai21444402005-11-07 17:20:03 -05001603 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604
1605 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001606 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001607 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608
Al Viro2d92ab32008-08-02 00:51:11 -04001609 if (old_path.mnt == old_path.mnt->mnt_parent)
Ram Pai21444402005-11-07 17:20:03 -05001610 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611
Al Viro2d92ab32008-08-02 00:51:11 -04001612 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1613 S_ISDIR(old_path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001614 goto out1;
1615 /*
1616 * Don't move a mount residing in a shared parent.
1617 */
Al Viro2d92ab32008-08-02 00:51:11 -04001618 if (old_path.mnt->mnt_parent &&
1619 IS_MNT_SHARED(old_path.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05001620 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001621 /*
1622 * Don't move a mount tree containing unbindable mounts to a destination
1623 * mount which is shared.
1624 */
Al Viro2d92ab32008-08-02 00:51:11 -04001625 if (IS_MNT_SHARED(path->mnt) &&
1626 tree_contains_unbindable(old_path.mnt))
Ram Pai9676f0c2005-11-07 17:21:20 -05001627 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 err = -ELOOP;
Al Viro2d92ab32008-08-02 00:51:11 -04001629 for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1630 if (p == old_path.mnt)
Ram Pai21444402005-11-07 17:20:03 -05001631 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632
Al Viro2d92ab32008-08-02 00:51:11 -04001633 err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08001634 if (err)
Ram Pai21444402005-11-07 17:20:03 -05001635 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636
1637 /* if the mount is moved, it should no longer be expire
1638 * automatically */
Al Viro2d92ab32008-08-02 00:51:11 -04001639 list_del_init(&old_path.mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640out1:
Al Viro2d92ab32008-08-02 00:51:11 -04001641 mutex_unlock(&path->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001642out:
Ram Pai390c6842005-11-07 17:17:51 -05001643 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04001645 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04001646 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 return err;
1648}
1649
1650/*
1651 * create a new mount for userspace and request it to be added into the
1652 * namespace's tree
1653 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001654static int do_new_mount(struct path *path, char *type, int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 int mnt_flags, char *name, void *data)
1656{
1657 struct vfsmount *mnt;
1658
1659 if (!type || !memchr(type, 0, PAGE_SIZE))
1660 return -EINVAL;
1661
1662 /* we need capabilities... */
1663 if (!capable(CAP_SYS_ADMIN))
1664 return -EPERM;
1665
1666 mnt = do_kern_mount(type, flags, name, data);
1667 if (IS_ERR(mnt))
1668 return PTR_ERR(mnt);
1669
Al Viro2d92ab32008-08-02 00:51:11 -04001670 return do_add_mount(mnt, path, mnt_flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671}
1672
1673/*
1674 * add a mount into a namespace's mount tree
1675 * - provide the option of adding the new mount to an expiration list
1676 */
Al Viro8d66bf52008-08-01 09:05:54 -04001677int do_add_mount(struct vfsmount *newmnt, struct path *path,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678 int mnt_flags, struct list_head *fslist)
1679{
1680 int err;
1681
Ram Pai390c6842005-11-07 17:17:51 -05001682 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 /* Something was mounted here while we slept */
Al Viro8d66bf52008-08-01 09:05:54 -04001684 while (d_mountpoint(path->dentry) &&
1685 follow_down(&path->mnt, &path->dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 ;
1687 err = -EINVAL;
Al Viro8d66bf52008-08-01 09:05:54 -04001688 if (!check_mnt(path->mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 goto unlock;
1690
1691 /* Refuse the same filesystem on the same mount point */
1692 err = -EBUSY;
Al Viro8d66bf52008-08-01 09:05:54 -04001693 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1694 path->mnt->mnt_root == path->dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 goto unlock;
1696
1697 err = -EINVAL;
1698 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1699 goto unlock;
1700
1701 newmnt->mnt_flags = mnt_flags;
Al Viro8d66bf52008-08-01 09:05:54 -04001702 if ((err = graft_tree(newmnt, path)))
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001703 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704
Al Viro6758f952008-03-22 16:14:30 -04001705 if (fslist) /* add to the specified expiration list */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001706 list_add_tail(&newmnt->mnt_expire, fslist);
Al Viro6758f952008-03-22 16:14:30 -04001707
Ram Pai390c6842005-11-07 17:17:51 -05001708 up_write(&namespace_sem);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001709 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710
1711unlock:
Ram Pai390c6842005-11-07 17:17:51 -05001712 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 mntput(newmnt);
1714 return err;
1715}
1716
1717EXPORT_SYMBOL_GPL(do_add_mount);
1718
Trond Myklebust5528f9112006-06-09 09:34:17 -04001719/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 * process a list of expirable mountpoints with the intent of discarding any
1721 * mountpoints that aren't in use and haven't been touched since last we came
1722 * here
1723 */
1724void mark_mounts_for_expiry(struct list_head *mounts)
1725{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 struct vfsmount *mnt, *next;
1727 LIST_HEAD(graveyard);
Al Virobcc5c7d2008-03-22 00:21:53 -04001728 LIST_HEAD(umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729
1730 if (list_empty(mounts))
1731 return;
1732
Al Virobcc5c7d2008-03-22 00:21:53 -04001733 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 spin_lock(&vfsmount_lock);
1735
1736 /* extract from the expiration list every vfsmount that matches the
1737 * following criteria:
1738 * - only referenced by its parent vfsmount
1739 * - still marked for expiry (marked on the last call here; marks are
1740 * cleared by mntput())
1741 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001742 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Virobcc5c7d2008-03-22 00:21:53 -04001744 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 continue;
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001746 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 }
Al Virobcc5c7d2008-03-22 00:21:53 -04001748 while (!list_empty(&graveyard)) {
1749 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
1750 touch_mnt_namespace(mnt->mnt_ns);
1751 umount_tree(mnt, 1, &umounts);
1752 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 spin_unlock(&vfsmount_lock);
Al Virobcc5c7d2008-03-22 00:21:53 -04001754 up_write(&namespace_sem);
1755
1756 release_mounts(&umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757}
1758
1759EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1760
1761/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04001762 * Ripoff of 'select_parent()'
1763 *
1764 * search the list of submounts for a given mountpoint, and move any
1765 * shrinkable submounts to the 'graveyard' list.
1766 */
1767static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1768{
1769 struct vfsmount *this_parent = parent;
1770 struct list_head *next;
1771 int found = 0;
1772
1773repeat:
1774 next = this_parent->mnt_mounts.next;
1775resume:
1776 while (next != &this_parent->mnt_mounts) {
1777 struct list_head *tmp = next;
1778 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1779
1780 next = tmp->next;
1781 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1782 continue;
1783 /*
1784 * Descend a level if the d_mounts list is non-empty.
1785 */
1786 if (!list_empty(&mnt->mnt_mounts)) {
1787 this_parent = mnt;
1788 goto repeat;
1789 }
1790
1791 if (!propagate_mount_busy(mnt, 1)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04001792 list_move_tail(&mnt->mnt_expire, graveyard);
1793 found++;
1794 }
1795 }
1796 /*
1797 * All done at this level ... ascend and resume the search
1798 */
1799 if (this_parent != parent) {
1800 next = this_parent->mnt_child.next;
1801 this_parent = this_parent->mnt_parent;
1802 goto resume;
1803 }
1804 return found;
1805}
1806
1807/*
1808 * process a list of expirable mountpoints with the intent of discarding any
1809 * submounts of a specific parent mountpoint
1810 */
Al Viroc35038b2008-03-22 00:46:23 -04001811static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
Trond Myklebust5528f9112006-06-09 09:34:17 -04001812{
1813 LIST_HEAD(graveyard);
Al Viroc35038b2008-03-22 00:46:23 -04001814 struct vfsmount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04001815
Trond Myklebust5528f9112006-06-09 09:34:17 -04001816 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04001817 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04001818 while (!list_empty(&graveyard)) {
Al Viroc35038b2008-03-22 00:46:23 -04001819 m = list_first_entry(&graveyard, struct vfsmount,
Al Virobcc5c7d2008-03-22 00:21:53 -04001820 mnt_expire);
Eric W. Biedermanafef80b2008-11-12 13:26:54 -08001821 touch_mnt_namespace(m->mnt_ns);
1822 umount_tree(m, 1, umounts);
Al Virobcc5c7d2008-03-22 00:21:53 -04001823 }
1824 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04001825}
1826
Trond Myklebust5528f9112006-06-09 09:34:17 -04001827/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 * Some copy_from_user() implementations do not return the exact number of
1829 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1830 * Note that this function differs from copy_from_user() in that it will oops
1831 * on bad values of `to', rather than returning a short copy.
1832 */
Ram Paib58fed82005-11-07 17:16:09 -05001833static long exact_copy_from_user(void *to, const void __user * from,
1834 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835{
1836 char *t = to;
1837 const char __user *f = from;
1838 char c;
1839
1840 if (!access_ok(VERIFY_READ, from, n))
1841 return n;
1842
1843 while (n) {
1844 if (__get_user(c, f)) {
1845 memset(t, 0, n);
1846 break;
1847 }
1848 *t++ = c;
1849 f++;
1850 n--;
1851 }
1852 return n;
1853}
1854
Ram Paib58fed82005-11-07 17:16:09 -05001855int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856{
1857 int i;
1858 unsigned long page;
1859 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05001860
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861 *where = 0;
1862 if (!data)
1863 return 0;
1864
1865 if (!(page = __get_free_page(GFP_KERNEL)))
1866 return -ENOMEM;
1867
1868 /* We only care that *some* data at the address the user
1869 * gave us is valid. Just in case, we'll zero
1870 * the remainder of the page.
1871 */
1872 /* copy_from_user cannot cross TASK_SIZE ! */
1873 size = TASK_SIZE - (unsigned long)data;
1874 if (size > PAGE_SIZE)
1875 size = PAGE_SIZE;
1876
1877 i = size - exact_copy_from_user((void *)page, data, size);
1878 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05001879 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880 return -EFAULT;
1881 }
1882 if (i != PAGE_SIZE)
1883 memset((char *)page + i, 0, PAGE_SIZE - i);
1884 *where = page;
1885 return 0;
1886}
1887
1888/*
1889 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1890 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1891 *
1892 * data is a (void *) that can point to any structure up to
1893 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1894 * information (or be NULL).
1895 *
1896 * Pre-0.97 versions of mount() didn't have a flags word.
1897 * When the flags word was introduced its top half was required
1898 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1899 * Therefore, if this magic number is present, it carries no information
1900 * and must be discarded.
1901 */
Ram Paib58fed82005-11-07 17:16:09 -05001902long do_mount(char *dev_name, char *dir_name, char *type_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 unsigned long flags, void *data_page)
1904{
Al Viro2d92ab32008-08-02 00:51:11 -04001905 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906 int retval = 0;
1907 int mnt_flags = 0;
1908
1909 /* Discard magic */
1910 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1911 flags &= ~MS_MGC_MSK;
1912
1913 /* Basic sanity checks */
1914
1915 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1916 return -EINVAL;
1917 if (dev_name && !memchr(dev_name, 0, PAGE_SIZE))
1918 return -EINVAL;
1919
1920 if (data_page)
1921 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1922
Andi Kleen613cbe32009-04-19 18:40:43 +02001923 /* Default to relatime unless overriden */
1924 if (!(flags & MS_NOATIME))
1925 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00001926
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927 /* Separate the per-mountpoint flags */
1928 if (flags & MS_NOSUID)
1929 mnt_flags |= MNT_NOSUID;
1930 if (flags & MS_NODEV)
1931 mnt_flags |= MNT_NODEV;
1932 if (flags & MS_NOEXEC)
1933 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001934 if (flags & MS_NOATIME)
1935 mnt_flags |= MNT_NOATIME;
1936 if (flags & MS_NODIRATIME)
1937 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00001938 if (flags & MS_STRICTATIME)
1939 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001940 if (flags & MS_RDONLY)
1941 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001942
1943 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE |
Matthew Garrettd0adde52009-03-26 17:49:56 +00001944 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
1945 MS_STRICTATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946
1947 /* ... and get the mountpoint */
Al Viro2d92ab32008-08-02 00:51:11 -04001948 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001949 if (retval)
1950 return retval;
1951
Al Viro2d92ab32008-08-02 00:51:11 -04001952 retval = security_sb_mount(dev_name, &path,
Al Virob5266eb2008-03-22 17:48:24 -04001953 type_page, flags, data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001954 if (retval)
1955 goto dput_out;
1956
1957 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04001958 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 data_page);
1960 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001961 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05001962 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04001963 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001964 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04001965 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001966 else
Al Viro2d92ab32008-08-02 00:51:11 -04001967 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 dev_name, data_page);
1969dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04001970 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971 return retval;
1972}
1973
JANAK DESAI741a2952006-02-07 12:59:00 -08001974/*
1975 * Allocate a new namespace structure and populate it with contents
1976 * copied from the namespace of the passed in task structure.
1977 */
Badari Pulavartye3222c42007-05-08 00:25:21 -07001978static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001979 struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001981 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04001982 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 struct vfsmount *p, *q;
1984
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001985 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 if (!new_ns)
Cedric Le Goater467e9f42007-07-15 23:41:06 -07001987 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988
1989 atomic_set(&new_ns->count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990 INIT_LIST_HEAD(&new_ns->list);
Al Viro5addc5d2005-11-07 17:15:49 -05001991 init_waitqueue_head(&new_ns->poll);
1992 new_ns->event = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001993
Ram Pai390c6842005-11-07 17:17:51 -05001994 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 /* First pass: copy the tree topology */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001996 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
Ram Pai9676f0c2005-11-07 17:21:20 -05001997 CL_COPY_ALL | CL_EXPIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 if (!new_ns->root) {
Ram Pai390c6842005-11-07 17:17:51 -05001999 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 kfree(new_ns);
Julia Lawall5cc4a032008-12-01 14:34:51 -08002001 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 }
2003 spin_lock(&vfsmount_lock);
2004 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
2005 spin_unlock(&vfsmount_lock);
2006
2007 /*
2008 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2009 * as belonging to new namespace. We have already acquired a private
2010 * fs_struct, so tsk->fs->lock is not needed.
2011 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002012 p = mnt_ns->root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 q = new_ns->root;
2014 while (p) {
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002015 q->mnt_ns = new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016 if (fs) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08002017 if (p == fs->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 rootmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002019 fs->root.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 }
Jan Blunck6ac08c32008-02-14 19:34:38 -08002021 if (p == fs->pwd.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002022 pwdmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002023 fs->pwd.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002024 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 }
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002026 p = next_mnt(p, mnt_ns->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002027 q = next_mnt(q, new_ns->root);
2028 }
Ram Pai390c6842005-11-07 17:17:51 -05002029 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030
Linus Torvalds1da177e2005-04-16 15:20:36 -07002031 if (rootmnt)
2032 mntput(rootmnt);
2033 if (pwdmnt)
2034 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035
JANAK DESAI741a2952006-02-07 12:59:00 -08002036 return new_ns;
2037}
2038
Eric W. Biederman213dd262007-07-15 23:41:15 -07002039struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
Badari Pulavartye3222c42007-05-08 00:25:21 -07002040 struct fs_struct *new_fs)
JANAK DESAI741a2952006-02-07 12:59:00 -08002041{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002042 struct mnt_namespace *new_ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002043
Badari Pulavartye3222c42007-05-08 00:25:21 -07002044 BUG_ON(!ns);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002045 get_mnt_ns(ns);
JANAK DESAI741a2952006-02-07 12:59:00 -08002046
2047 if (!(flags & CLONE_NEWNS))
Badari Pulavartye3222c42007-05-08 00:25:21 -07002048 return ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002049
Badari Pulavartye3222c42007-05-08 00:25:21 -07002050 new_ns = dup_mnt_ns(ns, new_fs);
JANAK DESAI741a2952006-02-07 12:59:00 -08002051
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002052 put_mnt_ns(ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -07002053 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054}
2055
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002056SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2057 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002058{
2059 int retval;
2060 unsigned long data_page;
2061 unsigned long type_page;
2062 unsigned long dev_page;
2063 char *dir_page;
2064
Ram Paib58fed82005-11-07 17:16:09 -05002065 retval = copy_mount_options(type, &type_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 if (retval < 0)
2067 return retval;
2068
2069 dir_page = getname(dir_name);
2070 retval = PTR_ERR(dir_page);
2071 if (IS_ERR(dir_page))
2072 goto out1;
2073
Ram Paib58fed82005-11-07 17:16:09 -05002074 retval = copy_mount_options(dev_name, &dev_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 if (retval < 0)
2076 goto out2;
2077
Ram Paib58fed82005-11-07 17:16:09 -05002078 retval = copy_mount_options(data, &data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079 if (retval < 0)
2080 goto out3;
2081
2082 lock_kernel();
Ram Paib58fed82005-11-07 17:16:09 -05002083 retval = do_mount((char *)dev_page, dir_page, (char *)type_page,
2084 flags, (void *)data_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085 unlock_kernel();
2086 free_page(data_page);
2087
2088out3:
2089 free_page(dev_page);
2090out2:
2091 putname(dir_page);
2092out1:
2093 free_page(type_page);
2094 return retval;
2095}
2096
2097/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002098 * pivot_root Semantics:
2099 * Moves the root file system of the current process to the directory put_old,
2100 * makes new_root as the new root file system of the current process, and sets
2101 * root/cwd of all processes which had them on the current root to new_root.
2102 *
2103 * Restrictions:
2104 * The new_root and put_old must be directories, and must not be on the
2105 * same file system as the current process root. The put_old must be
2106 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2107 * pointed to by put_old must yield the same directory as new_root. No other
2108 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2109 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08002110 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2111 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2112 * in this situation.
2113 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 * Notes:
2115 * - we don't move root/cwd if they are not at the root (reason: if something
2116 * cared enough to change them, it's probably wrong to force them elsewhere)
2117 * - it's okay to pick a root that isn't the root of a file system, e.g.
2118 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2119 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2120 * first.
2121 */
Heiko Carstens3480b252009-01-14 14:14:16 +01002122SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2123 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124{
2125 struct vfsmount *tmp;
Al Viro2d8f3032008-07-22 09:59:21 -04002126 struct path new, old, parent_path, root_parent, root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127 int error;
2128
2129 if (!capable(CAP_SYS_ADMIN))
2130 return -EPERM;
2131
Al Viro2d8f3032008-07-22 09:59:21 -04002132 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133 if (error)
2134 goto out0;
2135 error = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04002136 if (!check_mnt(new.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137 goto out1;
2138
Al Viro2d8f3032008-07-22 09:59:21 -04002139 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140 if (error)
2141 goto out1;
2142
Al Viro2d8f3032008-07-22 09:59:21 -04002143 error = security_sb_pivotroot(&old, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144 if (error) {
Al Viro2d8f3032008-07-22 09:59:21 -04002145 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146 goto out1;
2147 }
2148
2149 read_lock(&current->fs->lock);
Al Viro8c3ee422008-03-22 18:00:39 -04002150 root = current->fs->root;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002151 path_get(&current->fs->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 read_unlock(&current->fs->lock);
Ram Pai390c6842005-11-07 17:17:51 -05002153 down_write(&namespace_sem);
Al Viro2d8f3032008-07-22 09:59:21 -04002154 mutex_lock(&old.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002155 error = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04002156 if (IS_MNT_SHARED(old.mnt) ||
2157 IS_MNT_SHARED(new.mnt->mnt_parent) ||
Al Viro8c3ee422008-03-22 18:00:39 -04002158 IS_MNT_SHARED(root.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05002159 goto out2;
Al Viro8c3ee422008-03-22 18:00:39 -04002160 if (!check_mnt(root.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161 goto out2;
2162 error = -ENOENT;
Al Viro2d8f3032008-07-22 09:59:21 -04002163 if (IS_DEADDIR(new.dentry->d_inode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 goto out2;
Al Viro2d8f3032008-07-22 09:59:21 -04002165 if (d_unhashed(new.dentry) && !IS_ROOT(new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002166 goto out2;
Al Viro2d8f3032008-07-22 09:59:21 -04002167 if (d_unhashed(old.dentry) && !IS_ROOT(old.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168 goto out2;
2169 error = -EBUSY;
Al Viro2d8f3032008-07-22 09:59:21 -04002170 if (new.mnt == root.mnt ||
2171 old.mnt == root.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 goto out2; /* loop, on the same file system */
2173 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04002174 if (root.mnt->mnt_root != root.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002175 goto out2; /* not a mountpoint */
Al Viro8c3ee422008-03-22 18:00:39 -04002176 if (root.mnt->mnt_parent == root.mnt)
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07002177 goto out2; /* not attached */
Al Viro2d8f3032008-07-22 09:59:21 -04002178 if (new.mnt->mnt_root != new.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002179 goto out2; /* not a mountpoint */
Al Viro2d8f3032008-07-22 09:59:21 -04002180 if (new.mnt->mnt_parent == new.mnt)
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07002181 goto out2; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08002182 /* make sure we can reach put_old from new_root */
Al Viro2d8f3032008-07-22 09:59:21 -04002183 tmp = old.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184 spin_lock(&vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002185 if (tmp != new.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186 for (;;) {
2187 if (tmp->mnt_parent == tmp)
2188 goto out3; /* already mounted on put_old */
Al Viro2d8f3032008-07-22 09:59:21 -04002189 if (tmp->mnt_parent == new.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190 break;
2191 tmp = tmp->mnt_parent;
2192 }
Al Viro2d8f3032008-07-22 09:59:21 -04002193 if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194 goto out3;
Al Viro2d8f3032008-07-22 09:59:21 -04002195 } else if (!is_subdir(old.dentry, new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002196 goto out3;
Al Viro2d8f3032008-07-22 09:59:21 -04002197 detach_mnt(new.mnt, &parent_path);
Al Viro8c3ee422008-03-22 18:00:39 -04002198 detach_mnt(root.mnt, &root_parent);
Jan Blunck4ac91372008-02-14 19:34:32 -08002199 /* mount old root on put_old */
Al Viro2d8f3032008-07-22 09:59:21 -04002200 attach_mnt(root.mnt, &old);
Jan Blunck4ac91372008-02-14 19:34:32 -08002201 /* mount new_root on / */
Al Viro2d8f3032008-07-22 09:59:21 -04002202 attach_mnt(new.mnt, &root_parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002203 touch_mnt_namespace(current->nsproxy->mnt_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204 spin_unlock(&vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002205 chroot_fs_refs(&root, &new);
2206 security_sb_post_pivotroot(&root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 error = 0;
Al Viro1a390682008-03-21 20:48:19 -04002208 path_put(&root_parent);
2209 path_put(&parent_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210out2:
Al Viro2d8f3032008-07-22 09:59:21 -04002211 mutex_unlock(&old.dentry->d_inode->i_mutex);
Ram Pai390c6842005-11-07 17:17:51 -05002212 up_write(&namespace_sem);
Al Viro8c3ee422008-03-22 18:00:39 -04002213 path_put(&root);
Al Viro2d8f3032008-07-22 09:59:21 -04002214 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002215out1:
Al Viro2d8f3032008-07-22 09:59:21 -04002216 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002217out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 return error;
2219out3:
2220 spin_unlock(&vfsmount_lock);
2221 goto out2;
2222}
2223
2224static void __init init_mount_tree(void)
2225{
2226 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002227 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08002228 struct path root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229
2230 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2231 if (IS_ERR(mnt))
2232 panic("Can't create rootfs");
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002233 ns = kmalloc(sizeof(*ns), GFP_KERNEL);
2234 if (!ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 panic("Can't allocate initial namespace");
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002236 atomic_set(&ns->count, 1);
2237 INIT_LIST_HEAD(&ns->list);
2238 init_waitqueue_head(&ns->poll);
2239 ns->event = 0;
2240 list_add(&mnt->mnt_list, &ns->list);
2241 ns->root = mnt;
2242 mnt->mnt_ns = ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002244 init_task.nsproxy->mnt_ns = ns;
2245 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246
Jan Blunckac748a02008-02-14 19:34:39 -08002247 root.mnt = ns->root;
2248 root.dentry = ns->root->mnt_root;
2249
2250 set_fs_pwd(current->fs, &root);
2251 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252}
2253
Denis Cheng74bf17c2007-10-16 23:26:30 -07002254void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255{
Eric Dumazet13f14b42008-02-06 01:37:57 -08002256 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002257 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002258
Ram Pai390c6842005-11-07 17:17:51 -05002259 init_rwsem(&namespace_sem);
2260
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
Paul Mundt20c2df82007-07-20 10:11:58 +09002262 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002263
Ram Paib58fed82005-11-07 17:16:09 -05002264 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002265
2266 if (!mount_hashtable)
2267 panic("Failed to allocate mount hash table\n");
2268
Eric Dumazet13f14b42008-02-06 01:37:57 -08002269 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270
Eric Dumazet13f14b42008-02-06 01:37:57 -08002271 for (u = 0; u < HASH_SIZE; u++)
2272 INIT_LIST_HEAD(&mount_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002274 err = sysfs_init();
2275 if (err)
2276 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002277 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06002278 fs_kobj = kobject_create_and_add("fs", NULL);
2279 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002280 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 init_rootfs();
2282 init_mount_tree();
2283}
2284
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002285void __put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002287 struct vfsmount *root = ns->root;
Ram Pai70fbcdf2005-11-07 17:17:04 -05002288 LIST_HEAD(umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002289 ns->root = NULL;
Miklos Szeredi1ce88cf2005-07-07 17:57:24 -07002290 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05002291 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05002293 umount_tree(root, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05002295 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05002296 release_mounts(&umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002297 kfree(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298}