blob: 2e10cb19c5b02983e159bfe5f8039f3d08f3035d [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>
Alexey Dobriyanb43f3cb2009-07-08 01:54:37 +040025#include <linux/nsproxy.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/security.h>
27#include <linux/mount.h>
David Howells07f3f052006-09-30 20:52:18 +020028#include <linux/ramfs.h>
Eric Dumazet13f14b42008-02-06 01:37:57 -080029#include <linux/log2.h>
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010030#include <linux/idr.h>
Al Viro5ad4e532009-03-29 19:50:06 -040031#include <linux/fs_struct.h>
Andreas Gruenbacher2504c5d2009-12-17 21:24:27 -050032#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include <asm/uaccess.h>
34#include <asm/unistd.h>
Ram Pai07b20882005-11-07 17:19:07 -050035#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070036#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
Eric Dumazet13f14b42008-02-06 01:37:57 -080038#define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
39#define HASH_SIZE (1UL << HASH_SHIFT)
40
Linus Torvalds1da177e2005-04-16 15:20:36 -070041/* spinlock for vfsmount related operations, inplace of dcache_lock */
Al Viro5addc5d2005-11-07 17:15:49 -050042__cacheline_aligned_in_smp DEFINE_SPINLOCK(vfsmount_lock);
43
44static int event;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010045static DEFINE_IDA(mnt_id_ida);
Miklos Szeredi719f5d72008-03-27 13:06:23 +010046static DEFINE_IDA(mnt_group_ida);
Al Virof21f6222009-06-24 03:12:00 -040047static int mnt_id_start = 0;
48static int mnt_group_start = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -070049
Eric Dumazetfa3536c2006-03-26 01:37:24 -080050static struct list_head *mount_hashtable __read_mostly;
Christoph Lametere18b8902006-12-06 20:33:20 -080051static struct kmem_cache *mnt_cache __read_mostly;
Ram Pai390c6842005-11-07 17:17:51 -050052static struct rw_semaphore namespace_sem;
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080054/* /sys/fs */
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -060055struct kobject *fs_kobj;
56EXPORT_SYMBOL_GPL(fs_kobj);
Miklos Szeredif87fd4c2006-01-16 22:14:23 -080057
Linus Torvalds1da177e2005-04-16 15:20:36 -070058static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
59{
Ram Paib58fed82005-11-07 17:16:09 -050060 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
61 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Eric Dumazet13f14b42008-02-06 01:37:57 -080062 tmp = tmp + (tmp >> HASH_SHIFT);
63 return tmp & (HASH_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070064}
65
Dave Hansen3d733632008-02-15 14:37:59 -080066#define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
67
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010068/* allocation is serialized by namespace_sem */
69static int mnt_alloc_id(struct vfsmount *mnt)
70{
71 int res;
72
73retry:
74 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
75 spin_lock(&vfsmount_lock);
Al Virof21f6222009-06-24 03:12:00 -040076 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
77 if (!res)
78 mnt_id_start = mnt->mnt_id + 1;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010079 spin_unlock(&vfsmount_lock);
80 if (res == -EAGAIN)
81 goto retry;
82
83 return res;
84}
85
86static void mnt_free_id(struct vfsmount *mnt)
87{
Al Virof21f6222009-06-24 03:12:00 -040088 int id = mnt->mnt_id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010089 spin_lock(&vfsmount_lock);
Al Virof21f6222009-06-24 03:12:00 -040090 ida_remove(&mnt_id_ida, id);
91 if (mnt_id_start > id)
92 mnt_id_start = id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010093 spin_unlock(&vfsmount_lock);
94}
95
Miklos Szeredi719f5d72008-03-27 13:06:23 +010096/*
97 * Allocate a new peer group ID
98 *
99 * mnt_group_ida is protected by namespace_sem
100 */
101static int mnt_alloc_group_id(struct vfsmount *mnt)
102{
Al Virof21f6222009-06-24 03:12:00 -0400103 int res;
104
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100105 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
106 return -ENOMEM;
107
Al Virof21f6222009-06-24 03:12:00 -0400108 res = ida_get_new_above(&mnt_group_ida,
109 mnt_group_start,
110 &mnt->mnt_group_id);
111 if (!res)
112 mnt_group_start = mnt->mnt_group_id + 1;
113
114 return res;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100115}
116
117/*
118 * Release a peer group ID
119 */
120void mnt_release_group_id(struct vfsmount *mnt)
121{
Al Virof21f6222009-06-24 03:12:00 -0400122 int id = mnt->mnt_group_id;
123 ida_remove(&mnt_group_ida, id);
124 if (mnt_group_start > id)
125 mnt_group_start = id;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100126 mnt->mnt_group_id = 0;
127}
128
Linus Torvalds1da177e2005-04-16 15:20:36 -0700129struct vfsmount *alloc_vfsmnt(const char *name)
130{
Robert P. J. Dayc3762222007-02-10 01:45:03 -0800131 struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 if (mnt) {
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100133 int err;
134
135 err = mnt_alloc_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800136 if (err)
137 goto out_free_cache;
138
139 if (name) {
140 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
141 if (!mnt->mnt_devname)
142 goto out_free_id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100143 }
144
Ram Paib58fed82005-11-07 17:16:09 -0500145 atomic_set(&mnt->mnt_count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700146 INIT_LIST_HEAD(&mnt->mnt_hash);
147 INIT_LIST_HEAD(&mnt->mnt_child);
148 INIT_LIST_HEAD(&mnt->mnt_mounts);
149 INIT_LIST_HEAD(&mnt->mnt_list);
Miklos Szeredi55e700b2005-07-07 17:57:30 -0700150 INIT_LIST_HEAD(&mnt->mnt_expire);
Ram Pai03e06e62005-11-07 17:19:33 -0500151 INIT_LIST_HEAD(&mnt->mnt_share);
Ram Paia58b0eb2005-11-07 17:20:48 -0500152 INIT_LIST_HEAD(&mnt->mnt_slave_list);
153 INIT_LIST_HEAD(&mnt->mnt_slave);
Andreas Gruenbacher2504c5d2009-12-17 21:24:27 -0500154#ifdef CONFIG_FSNOTIFY
155 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
156#endif
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000157#ifdef CONFIG_SMP
158 mnt->mnt_writers = alloc_percpu(int);
159 if (!mnt->mnt_writers)
160 goto out_free_devname;
161#else
162 mnt->mnt_writers = 0;
163#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164 }
165 return mnt;
Li Zefan88b38782008-07-21 18:06:36 +0800166
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000167#ifdef CONFIG_SMP
168out_free_devname:
169 kfree(mnt->mnt_devname);
170#endif
Li Zefan88b38782008-07-21 18:06:36 +0800171out_free_id:
172 mnt_free_id(mnt);
173out_free_cache:
174 kmem_cache_free(mnt_cache, mnt);
175 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176}
177
Dave Hansen83660252008-02-15 14:37:30 -0800178/*
179 * Most r/o checks on a fs are for operations that take
180 * discrete amounts of time, like a write() or unlink().
181 * We must keep track of when those operations start
182 * (for permission checks) and when they end, so that
183 * we can determine when writes are able to occur to
184 * a filesystem.
185 */
Dave Hansen3d733632008-02-15 14:37:59 -0800186/*
187 * __mnt_is_readonly: check whether a mount is read-only
188 * @mnt: the mount to check for its write status
189 *
190 * This shouldn't be used directly ouside of the VFS.
191 * It does not guarantee that the filesystem will stay
192 * r/w, just that it is right *now*. This can not and
193 * should not be used in place of IS_RDONLY(inode).
194 * mnt_want/drop_write() will _keep_ the filesystem
195 * r/w.
196 */
197int __mnt_is_readonly(struct vfsmount *mnt)
198{
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800199 if (mnt->mnt_flags & MNT_READONLY)
200 return 1;
201 if (mnt->mnt_sb->s_flags & MS_RDONLY)
202 return 1;
203 return 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800204}
205EXPORT_SYMBOL_GPL(__mnt_is_readonly);
206
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000207static inline void inc_mnt_writers(struct vfsmount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800208{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000209#ifdef CONFIG_SMP
210 (*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))++;
211#else
212 mnt->mnt_writers++;
213#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800214}
Dave Hansen3d733632008-02-15 14:37:59 -0800215
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000216static inline void dec_mnt_writers(struct vfsmount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800217{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000218#ifdef CONFIG_SMP
219 (*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))--;
220#else
221 mnt->mnt_writers--;
222#endif
223}
224
225static unsigned int count_mnt_writers(struct vfsmount *mnt)
226{
227#ifdef CONFIG_SMP
228 unsigned int count = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800229 int cpu;
Dave Hansen3d733632008-02-15 14:37:59 -0800230
231 for_each_possible_cpu(cpu) {
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000232 count += *per_cpu_ptr(mnt->mnt_writers, cpu);
Dave Hansen3d733632008-02-15 14:37:59 -0800233 }
Dave Hansen3d733632008-02-15 14:37:59 -0800234
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000235 return count;
236#else
237 return mnt->mnt_writers;
238#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800239}
240
241/*
242 * Most r/o checks on a fs are for operations that take
243 * discrete amounts of time, like a write() or unlink().
244 * We must keep track of when those operations start
245 * (for permission checks) and when they end, so that
246 * we can determine when writes are able to occur to
247 * a filesystem.
248 */
Dave Hansen83660252008-02-15 14:37:30 -0800249/**
250 * mnt_want_write - get write access to a mount
251 * @mnt: the mount on which to take a write
252 *
253 * This tells the low-level filesystem that a write is
254 * about to be performed to it, and makes sure that
255 * writes are allowed before returning success. When
256 * the write operation is finished, mnt_drop_write()
257 * must be called. This is effectively a refcount.
258 */
259int mnt_want_write(struct vfsmount *mnt)
260{
Dave Hansen3d733632008-02-15 14:37:59 -0800261 int ret = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800262
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000263 preempt_disable();
264 inc_mnt_writers(mnt);
265 /*
266 * The store to inc_mnt_writers must be visible before we pass
267 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
268 * incremented count after it has set MNT_WRITE_HOLD.
269 */
270 smp_mb();
271 while (mnt->mnt_flags & MNT_WRITE_HOLD)
272 cpu_relax();
273 /*
274 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
275 * be set to match its requirements. So we must not load that until
276 * MNT_WRITE_HOLD is cleared.
277 */
278 smp_rmb();
Dave Hansen3d733632008-02-15 14:37:59 -0800279 if (__mnt_is_readonly(mnt)) {
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000280 dec_mnt_writers(mnt);
Dave Hansen3d733632008-02-15 14:37:59 -0800281 ret = -EROFS;
282 goto out;
283 }
Dave Hansen3d733632008-02-15 14:37:59 -0800284out:
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000285 preempt_enable();
Dave Hansen3d733632008-02-15 14:37:59 -0800286 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800287}
288EXPORT_SYMBOL_GPL(mnt_want_write);
289
290/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000291 * mnt_clone_write - get write access to a mount
292 * @mnt: the mount on which to take a write
293 *
294 * This is effectively like mnt_want_write, except
295 * it must only be used to take an extra write reference
296 * on a mountpoint that we already know has a write reference
297 * on it. This allows some optimisation.
298 *
299 * After finished, mnt_drop_write must be called as usual to
300 * drop the reference.
301 */
302int mnt_clone_write(struct vfsmount *mnt)
303{
304 /* superblock may be r/o */
305 if (__mnt_is_readonly(mnt))
306 return -EROFS;
307 preempt_disable();
308 inc_mnt_writers(mnt);
309 preempt_enable();
310 return 0;
311}
312EXPORT_SYMBOL_GPL(mnt_clone_write);
313
314/**
315 * mnt_want_write_file - get write access to a file's mount
316 * @file: the file who's mount on which to take a write
317 *
318 * This is like mnt_want_write, but it takes a file and can
319 * do some optimisations if the file is open for write already
320 */
321int mnt_want_write_file(struct file *file)
322{
OGAWA Hirofumi2d8dd382009-08-06 15:07:39 -0700323 struct inode *inode = file->f_dentry->d_inode;
324 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
npiggin@suse.de96029c42009-04-26 20:25:55 +1000325 return mnt_want_write(file->f_path.mnt);
326 else
327 return mnt_clone_write(file->f_path.mnt);
328}
329EXPORT_SYMBOL_GPL(mnt_want_write_file);
330
331/**
Dave Hansen83660252008-02-15 14:37:30 -0800332 * mnt_drop_write - give up write access to a mount
333 * @mnt: the mount on which to give up write access
334 *
335 * Tells the low-level filesystem that we are done
336 * performing writes to it. Must be matched with
337 * mnt_want_write() call above.
338 */
339void mnt_drop_write(struct vfsmount *mnt)
340{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000341 preempt_disable();
342 dec_mnt_writers(mnt);
343 preempt_enable();
Dave Hansen83660252008-02-15 14:37:30 -0800344}
345EXPORT_SYMBOL_GPL(mnt_drop_write);
346
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800347static int mnt_make_readonly(struct vfsmount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800348{
Dave Hansen3d733632008-02-15 14:37:59 -0800349 int ret = 0;
350
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800351 spin_lock(&vfsmount_lock);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000352 mnt->mnt_flags |= MNT_WRITE_HOLD;
353 /*
354 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
355 * should be visible before we do.
356 */
357 smp_mb();
358
359 /*
360 * With writers on hold, if this value is zero, then there are
361 * definitely no active writers (although held writers may subsequently
362 * increment the count, they'll have to wait, and decrement it after
363 * seeing MNT_READONLY).
364 *
365 * It is OK to have counter incremented on one CPU and decremented on
366 * another: the sum will add up correctly. The danger would be when we
367 * sum up each counter, if we read a counter before it is incremented,
368 * but then read another CPU's count which it has been subsequently
369 * decremented from -- we would see more decrements than we should.
370 * MNT_WRITE_HOLD protects against this scenario, because
371 * mnt_want_write first increments count, then smp_mb, then spins on
372 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
373 * we're counting up here.
374 */
375 if (count_mnt_writers(mnt) > 0)
376 ret = -EBUSY;
377 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800378 mnt->mnt_flags |= MNT_READONLY;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000379 /*
380 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
381 * that become unheld will see MNT_READONLY.
382 */
383 smp_wmb();
384 mnt->mnt_flags &= ~MNT_WRITE_HOLD;
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800385 spin_unlock(&vfsmount_lock);
Dave Hansen3d733632008-02-15 14:37:59 -0800386 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800387}
Dave Hansen83660252008-02-15 14:37:30 -0800388
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800389static void __mnt_unmake_readonly(struct vfsmount *mnt)
390{
391 spin_lock(&vfsmount_lock);
392 mnt->mnt_flags &= ~MNT_READONLY;
393 spin_unlock(&vfsmount_lock);
394}
395
Sukadev Bhattiprolua3ec9472009-03-04 12:06:34 -0800396void simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
David Howells454e2392006-06-23 02:02:57 -0700397{
398 mnt->mnt_sb = sb;
399 mnt->mnt_root = dget(sb->s_root);
David Howells454e2392006-06-23 02:02:57 -0700400}
401
402EXPORT_SYMBOL(simple_set_mnt);
403
Linus Torvalds1da177e2005-04-16 15:20:36 -0700404void free_vfsmnt(struct vfsmount *mnt)
405{
406 kfree(mnt->mnt_devname);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100407 mnt_free_id(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000408#ifdef CONFIG_SMP
409 free_percpu(mnt->mnt_writers);
410#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700411 kmem_cache_free(mnt_cache, mnt);
412}
413
414/*
Ram Paia05964f2005-11-07 17:20:17 -0500415 * find the first or last mount at @dentry on vfsmount @mnt depending on
416 * @dir. If @dir is set return the first mount else return the last mount.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 */
Ram Paia05964f2005-11-07 17:20:17 -0500418struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
419 int dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420{
Ram Paib58fed82005-11-07 17:16:09 -0500421 struct list_head *head = mount_hashtable + hash(mnt, dentry);
422 struct list_head *tmp = head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423 struct vfsmount *p, *found = NULL;
424
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425 for (;;) {
Ram Paia05964f2005-11-07 17:20:17 -0500426 tmp = dir ? tmp->next : tmp->prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 p = NULL;
428 if (tmp == head)
429 break;
430 p = list_entry(tmp, struct vfsmount, mnt_hash);
431 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
Ram Paia05964f2005-11-07 17:20:17 -0500432 found = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700433 break;
434 }
435 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436 return found;
437}
438
Ram Paia05964f2005-11-07 17:20:17 -0500439/*
440 * lookup_mnt increments the ref count before returning
441 * the vfsmount struct.
442 */
Al Viro1c755af2009-04-18 14:06:57 -0400443struct vfsmount *lookup_mnt(struct path *path)
Ram Paia05964f2005-11-07 17:20:17 -0500444{
445 struct vfsmount *child_mnt;
446 spin_lock(&vfsmount_lock);
Al Viro1c755af2009-04-18 14:06:57 -0400447 if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1)))
Ram Paia05964f2005-11-07 17:20:17 -0500448 mntget(child_mnt);
449 spin_unlock(&vfsmount_lock);
450 return child_mnt;
451}
452
Linus Torvalds1da177e2005-04-16 15:20:36 -0700453static inline int check_mnt(struct vfsmount *mnt)
454{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800455 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700456}
457
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800458static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500459{
460 if (ns) {
461 ns->event = ++event;
462 wake_up_interruptible(&ns->poll);
463 }
464}
465
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800466static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500467{
468 if (ns && ns->event != event) {
469 ns->event = event;
470 wake_up_interruptible(&ns->poll);
471 }
472}
473
Al Viro1a390682008-03-21 20:48:19 -0400474static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475{
Al Viro1a390682008-03-21 20:48:19 -0400476 old_path->dentry = mnt->mnt_mountpoint;
477 old_path->mnt = mnt->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 mnt->mnt_parent = mnt;
479 mnt->mnt_mountpoint = mnt->mnt_root;
480 list_del_init(&mnt->mnt_child);
481 list_del_init(&mnt->mnt_hash);
Al Viro1a390682008-03-21 20:48:19 -0400482 old_path->dentry->d_mounted--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483}
484
Ram Paib90fa9a2005-11-07 17:19:50 -0500485void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
486 struct vfsmount *child_mnt)
487{
488 child_mnt->mnt_parent = mntget(mnt);
489 child_mnt->mnt_mountpoint = dget(dentry);
490 dentry->d_mounted++;
491}
492
Al Viro1a390682008-03-21 20:48:19 -0400493static void attach_mnt(struct vfsmount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700494{
Al Viro1a390682008-03-21 20:48:19 -0400495 mnt_set_mountpoint(path->mnt, path->dentry, mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500496 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viro1a390682008-03-21 20:48:19 -0400497 hash(path->mnt, path->dentry));
498 list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500499}
500
501/*
502 * the caller must hold vfsmount_lock
503 */
504static void commit_tree(struct vfsmount *mnt)
505{
506 struct vfsmount *parent = mnt->mnt_parent;
507 struct vfsmount *m;
508 LIST_HEAD(head);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800509 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500510
511 BUG_ON(parent == mnt);
512
513 list_add_tail(&head, &mnt->mnt_list);
514 list_for_each_entry(m, &head, mnt_list)
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800515 m->mnt_ns = n;
Ram Paib90fa9a2005-11-07 17:19:50 -0500516 list_splice(&head, n->list.prev);
517
518 list_add_tail(&mnt->mnt_hash, mount_hashtable +
519 hash(parent, mnt->mnt_mountpoint));
520 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800521 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522}
523
524static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
525{
526 struct list_head *next = p->mnt_mounts.next;
527 if (next == &p->mnt_mounts) {
528 while (1) {
529 if (p == root)
530 return NULL;
531 next = p->mnt_child.next;
532 if (next != &p->mnt_parent->mnt_mounts)
533 break;
534 p = p->mnt_parent;
535 }
536 }
537 return list_entry(next, struct vfsmount, mnt_child);
538}
539
Ram Pai9676f0c2005-11-07 17:21:20 -0500540static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
541{
542 struct list_head *prev = p->mnt_mounts.prev;
543 while (prev != &p->mnt_mounts) {
544 p = list_entry(prev, struct vfsmount, mnt_child);
545 prev = p->mnt_mounts.prev;
546 }
547 return p;
548}
549
Ram Pai36341f62005-11-07 17:17:22 -0500550static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
551 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552{
553 struct super_block *sb = old->mnt_sb;
554 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
555
556 if (mnt) {
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100557 if (flag & (CL_SLAVE | CL_PRIVATE))
558 mnt->mnt_group_id = 0; /* not a peer of original */
559 else
560 mnt->mnt_group_id = old->mnt_group_id;
561
562 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
563 int err = mnt_alloc_group_id(mnt);
564 if (err)
565 goto out_free;
566 }
567
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 mnt->mnt_flags = old->mnt_flags;
569 atomic_inc(&sb->s_active);
570 mnt->mnt_sb = sb;
571 mnt->mnt_root = dget(root);
572 mnt->mnt_mountpoint = mnt->mnt_root;
573 mnt->mnt_parent = mnt;
Ram Paib90fa9a2005-11-07 17:19:50 -0500574
Ram Pai5afe0022005-11-07 17:21:01 -0500575 if (flag & CL_SLAVE) {
576 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
577 mnt->mnt_master = old;
578 CLEAR_MNT_SHARED(mnt);
Al Viro8aec0802007-06-07 12:20:32 -0400579 } else if (!(flag & CL_PRIVATE)) {
Al Viro796a6b52010-01-16 13:28:47 -0500580 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
Ram Pai5afe0022005-11-07 17:21:01 -0500581 list_add(&mnt->mnt_share, &old->mnt_share);
582 if (IS_MNT_SLAVE(old))
583 list_add(&mnt->mnt_slave, &old->mnt_slave);
584 mnt->mnt_master = old->mnt_master;
585 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500586 if (flag & CL_MAKE_SHARED)
587 set_mnt_shared(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700588
589 /* stick the duplicate mount on the same expiry list
590 * as the original if that was on one */
Ram Pai36341f62005-11-07 17:17:22 -0500591 if (flag & CL_EXPIRE) {
Ram Pai36341f62005-11-07 17:17:22 -0500592 if (!list_empty(&old->mnt_expire))
593 list_add(&mnt->mnt_expire, &old->mnt_expire);
Ram Pai36341f62005-11-07 17:17:22 -0500594 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 }
596 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100597
598 out_free:
599 free_vfsmnt(mnt);
600 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700601}
602
Al Viro7b7b1ac2005-11-07 17:13:39 -0500603static inline void __mntput(struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604{
605 struct super_block *sb = mnt->mnt_sb;
Dave Hansen3d733632008-02-15 14:37:59 -0800606 /*
Dave Hansen3d733632008-02-15 14:37:59 -0800607 * This probably indicates that somebody messed
608 * up a mnt_want/drop_write() pair. If this
609 * happens, the filesystem was probably unable
610 * to make r/w->r/o transitions.
611 */
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000612 /*
613 * atomic_dec_and_lock() used to deal with ->mnt_count decrements
614 * provides barriers, so count_mnt_writers() below is safe. AV
615 */
616 WARN_ON(count_mnt_writers(mnt));
Andreas Gruenbacherca9c7262009-12-17 21:24:27 -0500617 fsnotify_vfsmount_delete(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618 dput(mnt->mnt_root);
619 free_vfsmnt(mnt);
620 deactivate_super(sb);
621}
622
Al Viro7b7b1ac2005-11-07 17:13:39 -0500623void mntput_no_expire(struct vfsmount *mnt)
624{
625repeat:
626 if (atomic_dec_and_lock(&mnt->mnt_count, &vfsmount_lock)) {
627 if (likely(!mnt->mnt_pinned)) {
628 spin_unlock(&vfsmount_lock);
629 __mntput(mnt);
630 return;
631 }
632 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
633 mnt->mnt_pinned = 0;
634 spin_unlock(&vfsmount_lock);
635 acct_auto_close_mnt(mnt);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500636 goto repeat;
637 }
638}
639
640EXPORT_SYMBOL(mntput_no_expire);
641
642void mnt_pin(struct vfsmount *mnt)
643{
644 spin_lock(&vfsmount_lock);
645 mnt->mnt_pinned++;
646 spin_unlock(&vfsmount_lock);
647}
648
649EXPORT_SYMBOL(mnt_pin);
650
651void mnt_unpin(struct vfsmount *mnt)
652{
653 spin_lock(&vfsmount_lock);
654 if (mnt->mnt_pinned) {
655 atomic_inc(&mnt->mnt_count);
656 mnt->mnt_pinned--;
657 }
658 spin_unlock(&vfsmount_lock);
659}
660
661EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800663static inline void mangle(struct seq_file *m, const char *s)
664{
665 seq_escape(m, s, " \t\n\\");
666}
667
668/*
669 * Simple .show_options callback for filesystems which don't want to
670 * implement more complex mount option showing.
671 *
672 * See also save_mount_options().
673 */
674int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
675{
Al Viro2a32ceb2009-05-08 16:05:57 -0400676 const char *options;
677
678 rcu_read_lock();
679 options = rcu_dereference(mnt->mnt_sb->s_options);
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800680
681 if (options != NULL && options[0]) {
682 seq_putc(m, ',');
683 mangle(m, options);
684 }
Al Viro2a32ceb2009-05-08 16:05:57 -0400685 rcu_read_unlock();
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800686
687 return 0;
688}
689EXPORT_SYMBOL(generic_show_options);
690
691/*
692 * If filesystem uses generic_show_options(), this function should be
693 * called from the fill_super() callback.
694 *
695 * The .remount_fs callback usually needs to be handled in a special
696 * way, to make sure, that previous options are not overwritten if the
697 * remount fails.
698 *
699 * Also note, that if the filesystem's .remount_fs function doesn't
700 * reset all options to their default value, but changes only newly
701 * given options, then the displayed options will not reflect reality
702 * any more.
703 */
704void save_mount_options(struct super_block *sb, char *options)
705{
Al Viro2a32ceb2009-05-08 16:05:57 -0400706 BUG_ON(sb->s_options);
707 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800708}
709EXPORT_SYMBOL(save_mount_options);
710
Al Viro2a32ceb2009-05-08 16:05:57 -0400711void replace_mount_options(struct super_block *sb, char *options)
712{
713 char *old = sb->s_options;
714 rcu_assign_pointer(sb->s_options, options);
715 if (old) {
716 synchronize_rcu();
717 kfree(old);
718 }
719}
720EXPORT_SYMBOL(replace_mount_options);
721
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100722#ifdef CONFIG_PROC_FS
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723/* iterator */
724static void *m_start(struct seq_file *m, loff_t *pos)
725{
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100726 struct proc_mounts *p = m->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727
Ram Pai390c6842005-11-07 17:17:51 -0500728 down_read(&namespace_sem);
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100729 return seq_list_start(&p->ns->list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730}
731
732static void *m_next(struct seq_file *m, void *v, loff_t *pos)
733{
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100734 struct proc_mounts *p = m->private;
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700735
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100736 return seq_list_next(v, &p->ns->list, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737}
738
739static void m_stop(struct seq_file *m, void *v)
740{
Ram Pai390c6842005-11-07 17:17:51 -0500741 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742}
743
Al Viro9f5596a2010-02-05 00:40:25 -0500744int mnt_had_events(struct proc_mounts *p)
745{
746 struct mnt_namespace *ns = p->ns;
747 int res = 0;
748
749 spin_lock(&vfsmount_lock);
750 if (p->event != ns->event) {
751 p->event = ns->event;
752 res = 1;
753 }
754 spin_unlock(&vfsmount_lock);
755
756 return res;
757}
758
Ram Pai2d4d4862008-03-27 13:06:25 +0100759struct proc_fs_info {
760 int flag;
761 const char *str;
762};
763
Eric Paris2069f452008-07-04 09:47:13 +1000764static int show_sb_opts(struct seq_file *m, struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765{
Ram Pai2d4d4862008-03-27 13:06:25 +0100766 static const struct proc_fs_info fs_info[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 { MS_SYNCHRONOUS, ",sync" },
768 { MS_DIRSYNC, ",dirsync" },
769 { MS_MANDLOCK, ",mand" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770 { 0, NULL }
771 };
Ram Pai2d4d4862008-03-27 13:06:25 +0100772 const struct proc_fs_info *fs_infop;
773
774 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
775 if (sb->s_flags & fs_infop->flag)
776 seq_puts(m, fs_infop->str);
777 }
Eric Paris2069f452008-07-04 09:47:13 +1000778
779 return security_sb_show_options(m, sb);
Ram Pai2d4d4862008-03-27 13:06:25 +0100780}
781
782static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
783{
784 static const struct proc_fs_info mnt_info[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700785 { MNT_NOSUID, ",nosuid" },
786 { MNT_NODEV, ",nodev" },
787 { MNT_NOEXEC, ",noexec" },
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -0800788 { MNT_NOATIME, ",noatime" },
789 { MNT_NODIRATIME, ",nodiratime" },
Valerie Henson47ae32d2006-12-13 00:34:34 -0800790 { MNT_RELATIME, ",relatime" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 { 0, NULL }
792 };
Ram Pai2d4d4862008-03-27 13:06:25 +0100793 const struct proc_fs_info *fs_infop;
794
795 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
796 if (mnt->mnt_flags & fs_infop->flag)
797 seq_puts(m, fs_infop->str);
798 }
799}
800
801static void show_type(struct seq_file *m, struct super_block *sb)
802{
803 mangle(m, sb->s_type->name);
804 if (sb->s_subtype && sb->s_subtype[0]) {
805 seq_putc(m, '.');
806 mangle(m, sb->s_subtype);
807 }
808}
809
810static int show_vfsmnt(struct seq_file *m, void *v)
811{
812 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
813 int err = 0;
Jan Blunckc32c2f62008-02-14 19:38:43 -0800814 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700815
816 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
817 seq_putc(m, ' ');
Jan Blunckc32c2f62008-02-14 19:38:43 -0800818 seq_path(m, &mnt_path, " \t\n\\");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 seq_putc(m, ' ');
Ram Pai2d4d4862008-03-27 13:06:25 +0100820 show_type(m, mnt->mnt_sb);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800821 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
Eric Paris2069f452008-07-04 09:47:13 +1000822 err = show_sb_opts(m, mnt->mnt_sb);
823 if (err)
824 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +0100825 show_mnt_opts(m, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 if (mnt->mnt_sb->s_op->show_options)
827 err = mnt->mnt_sb->s_op->show_options(m, mnt);
828 seq_puts(m, " 0 0\n");
Eric Paris2069f452008-07-04 09:47:13 +1000829out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 return err;
831}
832
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100833const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700834 .start = m_start,
835 .next = m_next,
836 .stop = m_stop,
837 .show = show_vfsmnt
838};
839
Ram Pai2d4d4862008-03-27 13:06:25 +0100840static int show_mountinfo(struct seq_file *m, void *v)
841{
842 struct proc_mounts *p = m->private;
843 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
844 struct super_block *sb = mnt->mnt_sb;
845 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
846 struct path root = p->root;
847 int err = 0;
848
849 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
850 MAJOR(sb->s_dev), MINOR(sb->s_dev));
851 seq_dentry(m, mnt->mnt_root, " \t\n\\");
852 seq_putc(m, ' ');
853 seq_path_root(m, &mnt_path, &root, " \t\n\\");
854 if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
855 /*
856 * Mountpoint is outside root, discard that one. Ugly,
857 * but less so than trying to do that in iterator in a
858 * race-free way (due to renames).
859 */
860 return SEQ_SKIP;
861 }
862 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
863 show_mnt_opts(m, mnt);
864
865 /* Tagged fields ("foo:X" or "bar") */
866 if (IS_MNT_SHARED(mnt))
867 seq_printf(m, " shared:%i", mnt->mnt_group_id);
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +0100868 if (IS_MNT_SLAVE(mnt)) {
869 int master = mnt->mnt_master->mnt_group_id;
870 int dom = get_dominating_id(mnt, &p->root);
871 seq_printf(m, " master:%i", master);
872 if (dom && dom != master)
873 seq_printf(m, " propagate_from:%i", dom);
874 }
Ram Pai2d4d4862008-03-27 13:06:25 +0100875 if (IS_MNT_UNBINDABLE(mnt))
876 seq_puts(m, " unbindable");
877
878 /* Filesystem specific data */
879 seq_puts(m, " - ");
880 show_type(m, sb);
881 seq_putc(m, ' ');
882 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
883 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
Eric Paris2069f452008-07-04 09:47:13 +1000884 err = show_sb_opts(m, sb);
885 if (err)
886 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +0100887 if (sb->s_op->show_options)
888 err = sb->s_op->show_options(m, mnt);
889 seq_putc(m, '\n');
Eric Paris2069f452008-07-04 09:47:13 +1000890out:
Ram Pai2d4d4862008-03-27 13:06:25 +0100891 return err;
892}
893
894const struct seq_operations mountinfo_op = {
895 .start = m_start,
896 .next = m_next,
897 .stop = m_stop,
898 .show = show_mountinfo,
899};
900
Chuck Leverb4629fe2006-03-20 13:44:12 -0500901static int show_vfsstat(struct seq_file *m, void *v)
902{
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700903 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
Jan Blunckc32c2f62008-02-14 19:38:43 -0800904 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Chuck Leverb4629fe2006-03-20 13:44:12 -0500905 int err = 0;
906
907 /* device */
908 if (mnt->mnt_devname) {
909 seq_puts(m, "device ");
910 mangle(m, mnt->mnt_devname);
911 } else
912 seq_puts(m, "no device");
913
914 /* mount point */
915 seq_puts(m, " mounted on ");
Jan Blunckc32c2f62008-02-14 19:38:43 -0800916 seq_path(m, &mnt_path, " \t\n\\");
Chuck Leverb4629fe2006-03-20 13:44:12 -0500917 seq_putc(m, ' ');
918
919 /* file system type */
920 seq_puts(m, "with fstype ");
Ram Pai2d4d4862008-03-27 13:06:25 +0100921 show_type(m, mnt->mnt_sb);
Chuck Leverb4629fe2006-03-20 13:44:12 -0500922
923 /* optional statistics */
924 if (mnt->mnt_sb->s_op->show_stats) {
925 seq_putc(m, ' ');
926 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
927 }
928
929 seq_putc(m, '\n');
930 return err;
931}
932
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100933const struct seq_operations mountstats_op = {
Chuck Leverb4629fe2006-03-20 13:44:12 -0500934 .start = m_start,
935 .next = m_next,
936 .stop = m_stop,
937 .show = show_vfsstat,
938};
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100939#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -0500940
Linus Torvalds1da177e2005-04-16 15:20:36 -0700941/**
942 * may_umount_tree - check if a mount tree is busy
943 * @mnt: root of mount tree
944 *
945 * This is called to check if a tree of mounts has any
946 * open files, pwds, chroots or sub mounts that are
947 * busy.
948 */
949int may_umount_tree(struct vfsmount *mnt)
950{
Ram Pai36341f62005-11-07 17:17:22 -0500951 int actual_refs = 0;
952 int minimum_refs = 0;
953 struct vfsmount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954
955 spin_lock(&vfsmount_lock);
Ram Pai36341f62005-11-07 17:17:22 -0500956 for (p = mnt; p; p = next_mnt(p, mnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 actual_refs += atomic_read(&p->mnt_count);
958 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959 }
960 spin_unlock(&vfsmount_lock);
961
962 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -0800963 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700964
Ian Kente3474a82006-03-27 01:14:51 -0800965 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966}
967
968EXPORT_SYMBOL(may_umount_tree);
969
970/**
971 * may_umount - check if a mount point is busy
972 * @mnt: root of mount
973 *
974 * This is called to check if a mount point has any
975 * open files, pwds, chroots or sub mounts. If the
976 * mount has sub mounts this will return busy
977 * regardless of whether the sub mounts are busy.
978 *
979 * Doesn't take quota and stuff into account. IOW, in some cases it will
980 * give false negatives. The main reason why it's here is that we need
981 * a non-destructive way to look for easily umountable filesystems.
982 */
983int may_umount(struct vfsmount *mnt)
984{
Ian Kente3474a82006-03-27 01:14:51 -0800985 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -0500986 down_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -0500987 spin_lock(&vfsmount_lock);
988 if (propagate_mount_busy(mnt, 2))
Ian Kente3474a82006-03-27 01:14:51 -0800989 ret = 0;
Ram Paia05964f2005-11-07 17:20:17 -0500990 spin_unlock(&vfsmount_lock);
Al Viro8ad08d82010-01-16 12:56:08 -0500991 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -0500992 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993}
994
995EXPORT_SYMBOL(may_umount);
996
Ram Paib90fa9a2005-11-07 17:19:50 -0500997void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998{
Ram Pai70fbcdf2005-11-07 17:17:04 -0500999 struct vfsmount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -08001000 while (!list_empty(head)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -07001001 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001002 list_del_init(&mnt->mnt_hash);
1003 if (mnt->mnt_parent != mnt) {
1004 struct dentry *dentry;
1005 struct vfsmount *m;
1006 spin_lock(&vfsmount_lock);
1007 dentry = mnt->mnt_mountpoint;
1008 m = mnt->mnt_parent;
1009 mnt->mnt_mountpoint = mnt->mnt_root;
1010 mnt->mnt_parent = mnt;
Al Viro7c4b93d2008-03-21 23:59:49 -04001011 m->mnt_ghosts--;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001013 dput(dentry);
1014 mntput(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015 }
1016 mntput(mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001017 }
1018}
1019
Ram Paia05964f2005-11-07 17:20:17 -05001020void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001021{
1022 struct vfsmount *p;
1023
Akinobu Mita1bfba4e2006-06-26 00:24:40 -07001024 for (p = mnt; p; p = next_mnt(p, mnt))
1025 list_move(&p->mnt_hash, kill);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001026
Ram Paia05964f2005-11-07 17:20:17 -05001027 if (propagate)
1028 propagate_umount(kill);
1029
Ram Pai70fbcdf2005-11-07 17:17:04 -05001030 list_for_each_entry(p, kill, mnt_hash) {
1031 list_del_init(&p->mnt_expire);
1032 list_del_init(&p->mnt_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001033 __touch_mnt_namespace(p->mnt_ns);
1034 p->mnt_ns = NULL;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001035 list_del_init(&p->mnt_child);
Al Viro7c4b93d2008-03-21 23:59:49 -04001036 if (p->mnt_parent != p) {
1037 p->mnt_parent->mnt_ghosts++;
Al Virof30ac312006-02-01 07:53:21 -05001038 p->mnt_mountpoint->d_mounted--;
Al Viro7c4b93d2008-03-21 23:59:49 -04001039 }
Ram Paia05964f2005-11-07 17:20:17 -05001040 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 }
1042}
1043
Al Viroc35038b2008-03-22 00:46:23 -04001044static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1045
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046static int do_umount(struct vfsmount *mnt, int flags)
1047{
Ram Paib58fed82005-11-07 17:16:09 -05001048 struct super_block *sb = mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001050 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051
1052 retval = security_sb_umount(mnt, flags);
1053 if (retval)
1054 return retval;
1055
1056 /*
1057 * Allow userspace to request a mountpoint be expired rather than
1058 * unmounting unconditionally. Unmount only happens if:
1059 * (1) the mark is already set (the mark is cleared by mntput())
1060 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1061 */
1062 if (flags & MNT_EXPIRE) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08001063 if (mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064 flags & (MNT_FORCE | MNT_DETACH))
1065 return -EINVAL;
1066
1067 if (atomic_read(&mnt->mnt_count) != 2)
1068 return -EBUSY;
1069
1070 if (!xchg(&mnt->mnt_expiry_mark, 1))
1071 return -EAGAIN;
1072 }
1073
1074 /*
1075 * If we may have to abort operations to get out of this
1076 * mount, and they will themselves hold resources we must
1077 * allow the fs to do things. In the Unix tradition of
1078 * 'Gee thats tricky lets do it in userspace' the umount_begin
1079 * might fail to complete on the first run through as other tasks
1080 * must return, and the like. Thats for the mount program to worry
1081 * about for the moment.
1082 */
1083
Al Viro42faad92008-04-24 07:21:56 -04001084 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001085 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001086 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087
1088 /*
1089 * No sense to grab the lock for this test, but test itself looks
1090 * somewhat bogus. Suggestions for better replacement?
1091 * Ho-hum... In principle, we might treat that as umount + switch
1092 * to rootfs. GC would eventually take care of the old vfsmount.
1093 * Actually it makes sense, especially if rootfs would contain a
1094 * /reboot - static binary that would close all descriptors and
1095 * call reboot(9). Then init(8) could umount root and exec /reboot.
1096 */
Jan Blunck6ac08c32008-02-14 19:34:38 -08001097 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001098 /*
1099 * Special case for "unmounting" root ...
1100 * we just try to remount it readonly.
1101 */
1102 down_write(&sb->s_umount);
Al Viro4aa98cf2009-05-08 13:36:58 -04001103 if (!(sb->s_flags & MS_RDONLY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001104 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001105 up_write(&sb->s_umount);
1106 return retval;
1107 }
1108
Ram Pai390c6842005-11-07 17:17:51 -05001109 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 spin_lock(&vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -05001111 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112
Al Viroc35038b2008-03-22 00:46:23 -04001113 if (!(flags & MNT_DETACH))
1114 shrink_submounts(mnt, &umount_list);
1115
Linus Torvalds1da177e2005-04-16 15:20:36 -07001116 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -05001117 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 if (!list_empty(&mnt->mnt_list))
Ram Paia05964f2005-11-07 17:20:17 -05001119 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001120 retval = 0;
1121 }
1122 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001123 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001124 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125 return retval;
1126}
1127
1128/*
1129 * Now umount can handle mount points as well as block devices.
1130 * This is important for filesystems which use unnamed block devices.
1131 *
1132 * We now support a flag for forced unmount like the other 'big iron'
1133 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1134 */
1135
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001136SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137{
Al Viro2d8f3032008-07-22 09:59:21 -04001138 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 int retval;
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001140 int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001141
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001142 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1143 return -EINVAL;
1144
1145 if (!(flags & UMOUNT_NOFOLLOW))
1146 lookup_flags |= LOOKUP_FOLLOW;
1147
1148 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 if (retval)
1150 goto out;
1151 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001152 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153 goto dput_and_out;
Al Viro2d8f3032008-07-22 09:59:21 -04001154 if (!check_mnt(path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 goto dput_and_out;
1156
1157 retval = -EPERM;
1158 if (!capable(CAP_SYS_ADMIN))
1159 goto dput_and_out;
1160
Al Viro2d8f3032008-07-22 09:59:21 -04001161 retval = do_umount(path.mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001163 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001164 dput(path.dentry);
1165 mntput_no_expire(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001166out:
1167 return retval;
1168}
1169
1170#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1171
1172/*
Ram Paib58fed82005-11-07 17:16:09 -05001173 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001175SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176{
Ram Paib58fed82005-11-07 17:16:09 -05001177 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178}
1179
1180#endif
1181
Al Viro2d92ab32008-08-02 00:51:11 -04001182static int mount_is_safe(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183{
1184 if (capable(CAP_SYS_ADMIN))
1185 return 0;
1186 return -EPERM;
1187#ifdef notyet
Al Viro2d92ab32008-08-02 00:51:11 -04001188 if (S_ISLNK(path->dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189 return -EPERM;
Al Viro2d92ab32008-08-02 00:51:11 -04001190 if (path->dentry->d_inode->i_mode & S_ISVTX) {
David Howellsda9592e2008-11-14 10:39:05 +11001191 if (current_uid() != path->dentry->d_inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 return -EPERM;
1193 }
Al Viro2d92ab32008-08-02 00:51:11 -04001194 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001195 return -EPERM;
1196 return 0;
1197#endif
1198}
1199
Ram Paib90fa9a2005-11-07 17:19:50 -05001200struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001201 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202{
1203 struct vfsmount *res, *p, *q, *r, *s;
Al Viro1a390682008-03-21 20:48:19 -04001204 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205
Ram Pai9676f0c2005-11-07 17:21:20 -05001206 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1207 return NULL;
1208
Ram Pai36341f62005-11-07 17:17:22 -05001209 res = q = clone_mnt(mnt, dentry, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210 if (!q)
1211 goto Enomem;
1212 q->mnt_mountpoint = mnt->mnt_mountpoint;
1213
1214 p = mnt;
Domen Puncerfdadd652005-09-10 00:27:07 -07001215 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001216 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217 continue;
1218
1219 for (s = r; s; s = next_mnt(s, r)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001220 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1221 s = skip_mnt_tree(s);
1222 continue;
1223 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 while (p != s->mnt_parent) {
1225 p = p->mnt_parent;
1226 q = q->mnt_parent;
1227 }
1228 p = s;
Al Viro1a390682008-03-21 20:48:19 -04001229 path.mnt = q;
1230 path.dentry = p->mnt_mountpoint;
Ram Pai36341f62005-11-07 17:17:22 -05001231 q = clone_mnt(p, p->mnt_root, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232 if (!q)
1233 goto Enomem;
1234 spin_lock(&vfsmount_lock);
1235 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Viro1a390682008-03-21 20:48:19 -04001236 attach_mnt(q, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001237 spin_unlock(&vfsmount_lock);
1238 }
1239 }
1240 return res;
Ram Paib58fed82005-11-07 17:16:09 -05001241Enomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001242 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001243 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001245 umount_tree(res, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001247 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001248 }
1249 return NULL;
1250}
1251
Al Viro589ff872009-04-18 03:28:19 -04001252struct vfsmount *collect_mounts(struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001253{
1254 struct vfsmount *tree;
Al Viro1a60a282008-03-22 16:19:49 -04001255 down_write(&namespace_sem);
Al Viro589ff872009-04-18 03:28:19 -04001256 tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
Al Viro1a60a282008-03-22 16:19:49 -04001257 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001258 return tree;
1259}
1260
1261void drop_collected_mounts(struct vfsmount *mnt)
1262{
1263 LIST_HEAD(umount_list);
Al Viro1a60a282008-03-22 16:19:49 -04001264 down_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001265 spin_lock(&vfsmount_lock);
1266 umount_tree(mnt, 0, &umount_list);
1267 spin_unlock(&vfsmount_lock);
Al Viro1a60a282008-03-22 16:19:49 -04001268 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001269 release_mounts(&umount_list);
1270}
1271
Al Viro1f707132010-01-30 22:51:25 -05001272int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1273 struct vfsmount *root)
1274{
1275 struct vfsmount *mnt;
1276 int res = f(root, arg);
1277 if (res)
1278 return res;
1279 list_for_each_entry(mnt, &root->mnt_list, mnt_list) {
1280 res = f(mnt, arg);
1281 if (res)
1282 return res;
1283 }
1284 return 0;
1285}
1286
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001287static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1288{
1289 struct vfsmount *p;
1290
1291 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1292 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1293 mnt_release_group_id(p);
1294 }
1295}
1296
1297static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1298{
1299 struct vfsmount *p;
1300
1301 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1302 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1303 int err = mnt_alloc_group_id(p);
1304 if (err) {
1305 cleanup_group_ids(mnt, p);
1306 return err;
1307 }
1308 }
1309 }
1310
1311 return 0;
1312}
1313
Ram Paib90fa9a2005-11-07 17:19:50 -05001314/*
1315 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001316 * @nd : place the mount tree @source_mnt is attached
1317 * @parent_nd : if non-null, detach the source_mnt from its parent and
1318 * store the parent mount and mountpoint dentry.
1319 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001320 *
1321 * NOTE: in the table below explains the semantics when a source mount
1322 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001323 * ---------------------------------------------------------------------------
1324 * | BIND MOUNT OPERATION |
1325 * |**************************************************************************
1326 * | source-->| shared | private | slave | unbindable |
1327 * | dest | | | | |
1328 * | | | | | | |
1329 * | v | | | | |
1330 * |**************************************************************************
1331 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1332 * | | | | | |
1333 * |non-shared| shared (+) | private | slave (*) | invalid |
1334 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001335 * A bind operation clones the source mount and mounts the clone on the
1336 * destination mount.
1337 *
1338 * (++) the cloned mount is propagated to all the mounts in the propagation
1339 * tree of the destination mount and the cloned mount is added to
1340 * the peer group of the source mount.
1341 * (+) the cloned mount is created under the destination mount and is marked
1342 * as shared. The cloned mount is added to the peer group of the source
1343 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001344 * (+++) the mount is propagated to all the mounts in the propagation tree
1345 * of the destination mount and the cloned mount is made slave
1346 * of the same master as that of the source mount. The cloned mount
1347 * is marked as 'shared and slave'.
1348 * (*) the cloned mount is made a slave of the same master as that of the
1349 * source mount.
1350 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001351 * ---------------------------------------------------------------------------
1352 * | MOVE MOUNT OPERATION |
1353 * |**************************************************************************
1354 * | source-->| shared | private | slave | unbindable |
1355 * | dest | | | | |
1356 * | | | | | | |
1357 * | v | | | | |
1358 * |**************************************************************************
1359 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1360 * | | | | | |
1361 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1362 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001363 *
1364 * (+) the mount is moved to the destination. And is then propagated to
1365 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001366 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001367 * (+++) the mount is moved to the destination and is then propagated to
1368 * all the mounts belonging to the destination mount's propagation tree.
1369 * the mount is marked as 'shared and slave'.
1370 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001371 *
1372 * if the source mount is a tree, the operations explained above is
1373 * applied to each mount in the tree.
1374 * Must be called without spinlocks held, since this function can sleep
1375 * in allocations.
1376 */
1377static int attach_recursive_mnt(struct vfsmount *source_mnt,
Al Viro1a390682008-03-21 20:48:19 -04001378 struct path *path, struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001379{
1380 LIST_HEAD(tree_list);
Al Viro1a390682008-03-21 20:48:19 -04001381 struct vfsmount *dest_mnt = path->mnt;
1382 struct dentry *dest_dentry = path->dentry;
Ram Paib90fa9a2005-11-07 17:19:50 -05001383 struct vfsmount *child, *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001384 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001385
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001386 if (IS_MNT_SHARED(dest_mnt)) {
1387 err = invent_group_ids(source_mnt, true);
1388 if (err)
1389 goto out;
1390 }
1391 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1392 if (err)
1393 goto out_cleanup_ids;
Ram Paib90fa9a2005-11-07 17:19:50 -05001394
Al Virodf1a1ad2010-01-16 12:57:40 -05001395 spin_lock(&vfsmount_lock);
1396
Ram Paib90fa9a2005-11-07 17:19:50 -05001397 if (IS_MNT_SHARED(dest_mnt)) {
1398 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1399 set_mnt_shared(p);
1400 }
Al Viro1a390682008-03-21 20:48:19 -04001401 if (parent_path) {
1402 detach_mnt(source_mnt, parent_path);
1403 attach_mnt(source_mnt, path);
Al Viroe5d67f02009-04-07 12:15:39 -04001404 touch_mnt_namespace(parent_path->mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001405 } else {
1406 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1407 commit_tree(source_mnt);
1408 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001409
1410 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1411 list_del_init(&child->mnt_hash);
1412 commit_tree(child);
1413 }
1414 spin_unlock(&vfsmount_lock);
1415 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001416
1417 out_cleanup_ids:
1418 if (IS_MNT_SHARED(dest_mnt))
1419 cleanup_group_ids(source_mnt, NULL);
1420 out:
1421 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001422}
1423
Al Viro8c3ee422008-03-22 18:00:39 -04001424static int graft_tree(struct vfsmount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425{
1426 int err;
1427 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1428 return -EINVAL;
1429
Al Viro8c3ee422008-03-22 18:00:39 -04001430 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1432 return -ENOTDIR;
1433
1434 err = -ENOENT;
Al Viro8c3ee422008-03-22 18:00:39 -04001435 mutex_lock(&path->dentry->d_inode->i_mutex);
Al Virod83c49f2010-04-30 17:17:09 -04001436 if (cant_mount(path->dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 goto out_unlock;
1438
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04001439 if (!d_unlinked(path->dentry))
Al Viro8c3ee422008-03-22 18:00:39 -04001440 err = attach_recursive_mnt(mnt, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441out_unlock:
Al Viro8c3ee422008-03-22 18:00:39 -04001442 mutex_unlock(&path->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001443 return err;
1444}
1445
1446/*
Ram Pai07b20882005-11-07 17:19:07 -05001447 * recursively change the type of the mountpoint.
1448 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001449static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05001450{
Al Viro2d92ab32008-08-02 00:51:11 -04001451 struct vfsmount *m, *mnt = path->mnt;
Ram Pai07b20882005-11-07 17:19:07 -05001452 int recurse = flag & MS_REC;
1453 int type = flag & ~MS_REC;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001454 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05001455
Miklos Szerediee6f9582007-05-08 00:30:40 -07001456 if (!capable(CAP_SYS_ADMIN))
1457 return -EPERM;
1458
Al Viro2d92ab32008-08-02 00:51:11 -04001459 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05001460 return -EINVAL;
1461
1462 down_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001463 if (type == MS_SHARED) {
1464 err = invent_group_ids(mnt, recurse);
1465 if (err)
1466 goto out_unlock;
1467 }
1468
Ram Pai07b20882005-11-07 17:19:07 -05001469 spin_lock(&vfsmount_lock);
1470 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1471 change_mnt_propagation(m, type);
1472 spin_unlock(&vfsmount_lock);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001473
1474 out_unlock:
Ram Pai07b20882005-11-07 17:19:07 -05001475 up_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001476 return err;
Ram Pai07b20882005-11-07 17:19:07 -05001477}
1478
1479/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 * do loopback mount.
1481 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001482static int do_loopback(struct path *path, char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001483 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484{
Al Viro2d92ab32008-08-02 00:51:11 -04001485 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001486 struct vfsmount *mnt = NULL;
Al Viro2d92ab32008-08-02 00:51:11 -04001487 int err = mount_is_safe(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 if (err)
1489 return err;
1490 if (!old_name || !*old_name)
1491 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001492 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 if (err)
1494 return err;
1495
Ram Pai390c6842005-11-07 17:17:51 -05001496 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001497 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001498 if (IS_MNT_UNBINDABLE(old_path.mnt))
Jan Blunck4ac91372008-02-14 19:34:32 -08001499 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05001500
Al Viro2d92ab32008-08-02 00:51:11 -04001501 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
Al Viroccd48bc2005-11-07 17:15:04 -05001502 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503
Al Viroccd48bc2005-11-07 17:15:04 -05001504 err = -ENOMEM;
1505 if (recurse)
Al Viro2d92ab32008-08-02 00:51:11 -04001506 mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001507 else
Al Viro2d92ab32008-08-02 00:51:11 -04001508 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001509
1510 if (!mnt)
1511 goto out;
1512
Al Viro2d92ab32008-08-02 00:51:11 -04001513 err = graft_tree(mnt, path);
Al Viroccd48bc2005-11-07 17:15:04 -05001514 if (err) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001515 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 spin_lock(&vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001517 umount_tree(mnt, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 spin_unlock(&vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001519 release_mounts(&umount_list);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001520 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521
Al Viroccd48bc2005-11-07 17:15:04 -05001522out:
Ram Pai390c6842005-11-07 17:17:51 -05001523 up_write(&namespace_sem);
Al Viro2d92ab32008-08-02 00:51:11 -04001524 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525 return err;
1526}
1527
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001528static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1529{
1530 int error = 0;
1531 int readonly_request = 0;
1532
1533 if (ms_flags & MS_RDONLY)
1534 readonly_request = 1;
1535 if (readonly_request == __mnt_is_readonly(mnt))
1536 return 0;
1537
1538 if (readonly_request)
1539 error = mnt_make_readonly(mnt);
1540 else
1541 __mnt_unmake_readonly(mnt);
1542 return error;
1543}
1544
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545/*
1546 * change filesystem flags. dir should be a physical root of filesystem.
1547 * If you've mounted a non-root directory somewhere and want to do remount
1548 * on it - tough luck.
1549 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001550static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 void *data)
1552{
1553 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04001554 struct super_block *sb = path->mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555
1556 if (!capable(CAP_SYS_ADMIN))
1557 return -EPERM;
1558
Al Viro2d92ab32008-08-02 00:51:11 -04001559 if (!check_mnt(path->mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001560 return -EINVAL;
1561
Al Viro2d92ab32008-08-02 00:51:11 -04001562 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563 return -EINVAL;
1564
1565 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001566 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001567 err = change_mount_flags(path->mnt, flags);
Al Viro4aa98cf2009-05-08 13:36:58 -04001568 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001569 err = do_remount_sb(sb, flags, data, 0);
Al Viro7b43a792010-01-16 13:01:26 -05001570 if (!err) {
1571 spin_lock(&vfsmount_lock);
Valerie Aurora495d6c92010-01-26 14:20:47 -05001572 mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK;
Al Viro2d92ab32008-08-02 00:51:11 -04001573 path->mnt->mnt_flags = mnt_flags;
Al Viro7b43a792010-01-16 13:01:26 -05001574 spin_unlock(&vfsmount_lock);
1575 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576 up_write(&sb->s_umount);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001577 if (!err) {
Dan Williams0e55a7c2008-09-26 19:01:20 -07001578 spin_lock(&vfsmount_lock);
1579 touch_mnt_namespace(path->mnt->mnt_ns);
1580 spin_unlock(&vfsmount_lock);
1581 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001582 return err;
1583}
1584
Ram Pai9676f0c2005-11-07 17:21:20 -05001585static inline int tree_contains_unbindable(struct vfsmount *mnt)
1586{
1587 struct vfsmount *p;
1588 for (p = mnt; p; p = next_mnt(p, mnt)) {
1589 if (IS_MNT_UNBINDABLE(p))
1590 return 1;
1591 }
1592 return 0;
1593}
1594
Al Viro0a0d8a42008-08-02 00:55:27 -04001595static int do_move_mount(struct path *path, char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596{
Al Viro2d92ab32008-08-02 00:51:11 -04001597 struct path old_path, parent_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 struct vfsmount *p;
1599 int err = 0;
1600 if (!capable(CAP_SYS_ADMIN))
1601 return -EPERM;
1602 if (!old_name || !*old_name)
1603 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001604 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001605 if (err)
1606 return err;
1607
Ram Pai390c6842005-11-07 17:17:51 -05001608 down_write(&namespace_sem);
Al Viro2d92ab32008-08-02 00:51:11 -04001609 while (d_mountpoint(path->dentry) &&
Al Viro9393bd02009-04-18 13:58:15 -04001610 follow_down(path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611 ;
1612 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001613 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001614 goto out;
1615
1616 err = -ENOENT;
Al Viro2d92ab32008-08-02 00:51:11 -04001617 mutex_lock(&path->dentry->d_inode->i_mutex);
Al Virod83c49f2010-04-30 17:17:09 -04001618 if (cant_mount(path->dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 goto out1;
1620
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04001621 if (d_unlinked(path->dentry))
Ram Pai21444402005-11-07 17:20:03 -05001622 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623
1624 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001625 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001626 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627
Al Viro2d92ab32008-08-02 00:51:11 -04001628 if (old_path.mnt == old_path.mnt->mnt_parent)
Ram Pai21444402005-11-07 17:20:03 -05001629 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630
Al Viro2d92ab32008-08-02 00:51:11 -04001631 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1632 S_ISDIR(old_path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001633 goto out1;
1634 /*
1635 * Don't move a mount residing in a shared parent.
1636 */
Al Viro2d92ab32008-08-02 00:51:11 -04001637 if (old_path.mnt->mnt_parent &&
1638 IS_MNT_SHARED(old_path.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05001639 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001640 /*
1641 * Don't move a mount tree containing unbindable mounts to a destination
1642 * mount which is shared.
1643 */
Al Viro2d92ab32008-08-02 00:51:11 -04001644 if (IS_MNT_SHARED(path->mnt) &&
1645 tree_contains_unbindable(old_path.mnt))
Ram Pai9676f0c2005-11-07 17:21:20 -05001646 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001647 err = -ELOOP;
Al Viro2d92ab32008-08-02 00:51:11 -04001648 for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1649 if (p == old_path.mnt)
Ram Pai21444402005-11-07 17:20:03 -05001650 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651
Al Viro2d92ab32008-08-02 00:51:11 -04001652 err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08001653 if (err)
Ram Pai21444402005-11-07 17:20:03 -05001654 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655
1656 /* if the mount is moved, it should no longer be expire
1657 * automatically */
Al Viro2d92ab32008-08-02 00:51:11 -04001658 list_del_init(&old_path.mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659out1:
Al Viro2d92ab32008-08-02 00:51:11 -04001660 mutex_unlock(&path->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661out:
Ram Pai390c6842005-11-07 17:17:51 -05001662 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04001664 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04001665 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 return err;
1667}
1668
1669/*
1670 * create a new mount for userspace and request it to be added into the
1671 * namespace's tree
1672 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001673static int do_new_mount(struct path *path, char *type, int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 int mnt_flags, char *name, void *data)
1675{
1676 struct vfsmount *mnt;
1677
Vegard Nossumeca6f532009-09-18 13:05:45 -07001678 if (!type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 return -EINVAL;
1680
1681 /* we need capabilities... */
1682 if (!capable(CAP_SYS_ADMIN))
1683 return -EPERM;
1684
Al Viro7f78d4c2009-05-08 13:34:06 -04001685 lock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 mnt = do_kern_mount(type, flags, name, data);
Al Viro7f78d4c2009-05-08 13:34:06 -04001687 unlock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 if (IS_ERR(mnt))
1689 return PTR_ERR(mnt);
1690
Al Viro2d92ab32008-08-02 00:51:11 -04001691 return do_add_mount(mnt, path, mnt_flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692}
1693
1694/*
1695 * add a mount into a namespace's mount tree
1696 * - provide the option of adding the new mount to an expiration list
1697 */
Al Viro8d66bf52008-08-01 09:05:54 -04001698int do_add_mount(struct vfsmount *newmnt, struct path *path,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 int mnt_flags, struct list_head *fslist)
1700{
1701 int err;
1702
Al Viro80893522010-02-05 09:30:46 -05001703 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
Al Viro27d55f12010-01-16 13:07:36 -05001704
Ram Pai390c6842005-11-07 17:17:51 -05001705 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 /* Something was mounted here while we slept */
Al Viro8d66bf52008-08-01 09:05:54 -04001707 while (d_mountpoint(path->dentry) &&
Al Viro9393bd02009-04-18 13:58:15 -04001708 follow_down(path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 ;
1710 err = -EINVAL;
Al Virodd5cae62009-04-07 12:21:18 -04001711 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712 goto unlock;
1713
1714 /* Refuse the same filesystem on the same mount point */
1715 err = -EBUSY;
Al Viro8d66bf52008-08-01 09:05:54 -04001716 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1717 path->mnt->mnt_root == path->dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 goto unlock;
1719
1720 err = -EINVAL;
1721 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1722 goto unlock;
1723
1724 newmnt->mnt_flags = mnt_flags;
Al Viro8d66bf52008-08-01 09:05:54 -04001725 if ((err = graft_tree(newmnt, path)))
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001726 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727
Al Viro6758f952008-03-22 16:14:30 -04001728 if (fslist) /* add to the specified expiration list */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001729 list_add_tail(&newmnt->mnt_expire, fslist);
Al Viro6758f952008-03-22 16:14:30 -04001730
Ram Pai390c6842005-11-07 17:17:51 -05001731 up_write(&namespace_sem);
Ram Pai5b83d2c5c2005-11-07 17:16:29 -05001732 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733
1734unlock:
Ram Pai390c6842005-11-07 17:17:51 -05001735 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001736 mntput(newmnt);
1737 return err;
1738}
1739
1740EXPORT_SYMBOL_GPL(do_add_mount);
1741
Trond Myklebust5528f9112006-06-09 09:34:17 -04001742/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 * process a list of expirable mountpoints with the intent of discarding any
1744 * mountpoints that aren't in use and haven't been touched since last we came
1745 * here
1746 */
1747void mark_mounts_for_expiry(struct list_head *mounts)
1748{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749 struct vfsmount *mnt, *next;
1750 LIST_HEAD(graveyard);
Al Virobcc5c7d2008-03-22 00:21:53 -04001751 LIST_HEAD(umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752
1753 if (list_empty(mounts))
1754 return;
1755
Al Virobcc5c7d2008-03-22 00:21:53 -04001756 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 spin_lock(&vfsmount_lock);
1758
1759 /* extract from the expiration list every vfsmount that matches the
1760 * following criteria:
1761 * - only referenced by its parent vfsmount
1762 * - still marked for expiry (marked on the last call here; marks are
1763 * cleared by mntput())
1764 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001765 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Virobcc5c7d2008-03-22 00:21:53 -04001767 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 continue;
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001769 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 }
Al Virobcc5c7d2008-03-22 00:21:53 -04001771 while (!list_empty(&graveyard)) {
1772 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
1773 touch_mnt_namespace(mnt->mnt_ns);
1774 umount_tree(mnt, 1, &umounts);
1775 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 spin_unlock(&vfsmount_lock);
Al Virobcc5c7d2008-03-22 00:21:53 -04001777 up_write(&namespace_sem);
1778
1779 release_mounts(&umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780}
1781
1782EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1783
1784/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04001785 * Ripoff of 'select_parent()'
1786 *
1787 * search the list of submounts for a given mountpoint, and move any
1788 * shrinkable submounts to the 'graveyard' list.
1789 */
1790static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1791{
1792 struct vfsmount *this_parent = parent;
1793 struct list_head *next;
1794 int found = 0;
1795
1796repeat:
1797 next = this_parent->mnt_mounts.next;
1798resume:
1799 while (next != &this_parent->mnt_mounts) {
1800 struct list_head *tmp = next;
1801 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1802
1803 next = tmp->next;
1804 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1805 continue;
1806 /*
1807 * Descend a level if the d_mounts list is non-empty.
1808 */
1809 if (!list_empty(&mnt->mnt_mounts)) {
1810 this_parent = mnt;
1811 goto repeat;
1812 }
1813
1814 if (!propagate_mount_busy(mnt, 1)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04001815 list_move_tail(&mnt->mnt_expire, graveyard);
1816 found++;
1817 }
1818 }
1819 /*
1820 * All done at this level ... ascend and resume the search
1821 */
1822 if (this_parent != parent) {
1823 next = this_parent->mnt_child.next;
1824 this_parent = this_parent->mnt_parent;
1825 goto resume;
1826 }
1827 return found;
1828}
1829
1830/*
1831 * process a list of expirable mountpoints with the intent of discarding any
1832 * submounts of a specific parent mountpoint
1833 */
Al Viroc35038b2008-03-22 00:46:23 -04001834static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
Trond Myklebust5528f9112006-06-09 09:34:17 -04001835{
1836 LIST_HEAD(graveyard);
Al Viroc35038b2008-03-22 00:46:23 -04001837 struct vfsmount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04001838
Trond Myklebust5528f9112006-06-09 09:34:17 -04001839 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04001840 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04001841 while (!list_empty(&graveyard)) {
Al Viroc35038b2008-03-22 00:46:23 -04001842 m = list_first_entry(&graveyard, struct vfsmount,
Al Virobcc5c7d2008-03-22 00:21:53 -04001843 mnt_expire);
Eric W. Biedermanafef80b2008-11-12 13:26:54 -08001844 touch_mnt_namespace(m->mnt_ns);
1845 umount_tree(m, 1, umounts);
Al Virobcc5c7d2008-03-22 00:21:53 -04001846 }
1847 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04001848}
1849
Trond Myklebust5528f9112006-06-09 09:34:17 -04001850/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851 * Some copy_from_user() implementations do not return the exact number of
1852 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1853 * Note that this function differs from copy_from_user() in that it will oops
1854 * on bad values of `to', rather than returning a short copy.
1855 */
Ram Paib58fed82005-11-07 17:16:09 -05001856static long exact_copy_from_user(void *to, const void __user * from,
1857 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858{
1859 char *t = to;
1860 const char __user *f = from;
1861 char c;
1862
1863 if (!access_ok(VERIFY_READ, from, n))
1864 return n;
1865
1866 while (n) {
1867 if (__get_user(c, f)) {
1868 memset(t, 0, n);
1869 break;
1870 }
1871 *t++ = c;
1872 f++;
1873 n--;
1874 }
1875 return n;
1876}
1877
Ram Paib58fed82005-11-07 17:16:09 -05001878int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001879{
1880 int i;
1881 unsigned long page;
1882 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05001883
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 *where = 0;
1885 if (!data)
1886 return 0;
1887
1888 if (!(page = __get_free_page(GFP_KERNEL)))
1889 return -ENOMEM;
1890
1891 /* We only care that *some* data at the address the user
1892 * gave us is valid. Just in case, we'll zero
1893 * the remainder of the page.
1894 */
1895 /* copy_from_user cannot cross TASK_SIZE ! */
1896 size = TASK_SIZE - (unsigned long)data;
1897 if (size > PAGE_SIZE)
1898 size = PAGE_SIZE;
1899
1900 i = size - exact_copy_from_user((void *)page, data, size);
1901 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05001902 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903 return -EFAULT;
1904 }
1905 if (i != PAGE_SIZE)
1906 memset((char *)page + i, 0, PAGE_SIZE - i);
1907 *where = page;
1908 return 0;
1909}
1910
Vegard Nossumeca6f532009-09-18 13:05:45 -07001911int copy_mount_string(const void __user *data, char **where)
1912{
1913 char *tmp;
1914
1915 if (!data) {
1916 *where = NULL;
1917 return 0;
1918 }
1919
1920 tmp = strndup_user(data, PAGE_SIZE);
1921 if (IS_ERR(tmp))
1922 return PTR_ERR(tmp);
1923
1924 *where = tmp;
1925 return 0;
1926}
1927
Linus Torvalds1da177e2005-04-16 15:20:36 -07001928/*
1929 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1930 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1931 *
1932 * data is a (void *) that can point to any structure up to
1933 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1934 * information (or be NULL).
1935 *
1936 * Pre-0.97 versions of mount() didn't have a flags word.
1937 * When the flags word was introduced its top half was required
1938 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1939 * Therefore, if this magic number is present, it carries no information
1940 * and must be discarded.
1941 */
Ram Paib58fed82005-11-07 17:16:09 -05001942long do_mount(char *dev_name, char *dir_name, char *type_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943 unsigned long flags, void *data_page)
1944{
Al Viro2d92ab32008-08-02 00:51:11 -04001945 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 int retval = 0;
1947 int mnt_flags = 0;
1948
1949 /* Discard magic */
1950 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1951 flags &= ~MS_MGC_MSK;
1952
1953 /* Basic sanity checks */
1954
1955 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1956 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957
1958 if (data_page)
1959 ((char *)data_page)[PAGE_SIZE - 1] = 0;
1960
Tetsuo Handaa27ab9f22009-10-04 21:49:49 +09001961 /* ... and get the mountpoint */
1962 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
1963 if (retval)
1964 return retval;
1965
1966 retval = security_sb_mount(dev_name, &path,
1967 type_page, flags, data_page);
1968 if (retval)
1969 goto dput_out;
1970
Andi Kleen613cbe32009-04-19 18:40:43 +02001971 /* Default to relatime unless overriden */
1972 if (!(flags & MS_NOATIME))
1973 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00001974
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 /* Separate the per-mountpoint flags */
1976 if (flags & MS_NOSUID)
1977 mnt_flags |= MNT_NOSUID;
1978 if (flags & MS_NODEV)
1979 mnt_flags |= MNT_NODEV;
1980 if (flags & MS_NOEXEC)
1981 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001982 if (flags & MS_NOATIME)
1983 mnt_flags |= MNT_NOATIME;
1984 if (flags & MS_NODIRATIME)
1985 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00001986 if (flags & MS_STRICTATIME)
1987 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001988 if (flags & MS_RDONLY)
1989 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08001990
Al Viro7a4dec52010-08-09 12:05:43 -04001991 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
Matthew Garrettd0adde52009-03-26 17:49:56 +00001992 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
1993 MS_STRICTATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04001996 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997 data_page);
1998 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001999 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05002000 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04002001 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04002003 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002004 else
Al Viro2d92ab32008-08-02 00:51:11 -04002005 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002006 dev_name, data_page);
2007dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04002008 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002009 return retval;
2010}
2011
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002012static struct mnt_namespace *alloc_mnt_ns(void)
2013{
2014 struct mnt_namespace *new_ns;
2015
2016 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2017 if (!new_ns)
2018 return ERR_PTR(-ENOMEM);
2019 atomic_set(&new_ns->count, 1);
2020 new_ns->root = NULL;
2021 INIT_LIST_HEAD(&new_ns->list);
2022 init_waitqueue_head(&new_ns->poll);
2023 new_ns->event = 0;
2024 return new_ns;
2025}
2026
JANAK DESAI741a2952006-02-07 12:59:00 -08002027/*
2028 * Allocate a new namespace structure and populate it with contents
2029 * copied from the namespace of the passed in task structure.
2030 */
Badari Pulavartye3222c42007-05-08 00:25:21 -07002031static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002032 struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002034 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04002035 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036 struct vfsmount *p, *q;
2037
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002038 new_ns = alloc_mnt_ns();
2039 if (IS_ERR(new_ns))
2040 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041
Ram Pai390c6842005-11-07 17:17:51 -05002042 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 /* First pass: copy the tree topology */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002044 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
Ram Pai9676f0c2005-11-07 17:21:20 -05002045 CL_COPY_ALL | CL_EXPIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046 if (!new_ns->root) {
Ram Pai390c6842005-11-07 17:17:51 -05002047 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002048 kfree(new_ns);
Julia Lawall5cc4a032008-12-01 14:34:51 -08002049 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 }
2051 spin_lock(&vfsmount_lock);
2052 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
2053 spin_unlock(&vfsmount_lock);
2054
2055 /*
2056 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2057 * as belonging to new namespace. We have already acquired a private
2058 * fs_struct, so tsk->fs->lock is not needed.
2059 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002060 p = mnt_ns->root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 q = new_ns->root;
2062 while (p) {
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002063 q->mnt_ns = new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002064 if (fs) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08002065 if (p == fs->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002066 rootmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002067 fs->root.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002068 }
Jan Blunck6ac08c32008-02-14 19:34:38 -08002069 if (p == fs->pwd.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002070 pwdmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002071 fs->pwd.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002072 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002073 }
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002074 p = next_mnt(p, mnt_ns->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002075 q = next_mnt(q, new_ns->root);
2076 }
Ram Pai390c6842005-11-07 17:17:51 -05002077 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002078
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079 if (rootmnt)
2080 mntput(rootmnt);
2081 if (pwdmnt)
2082 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083
JANAK DESAI741a2952006-02-07 12:59:00 -08002084 return new_ns;
2085}
2086
Eric W. Biederman213dd262007-07-15 23:41:15 -07002087struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
Badari Pulavartye3222c42007-05-08 00:25:21 -07002088 struct fs_struct *new_fs)
JANAK DESAI741a2952006-02-07 12:59:00 -08002089{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002090 struct mnt_namespace *new_ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002091
Badari Pulavartye3222c42007-05-08 00:25:21 -07002092 BUG_ON(!ns);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002093 get_mnt_ns(ns);
JANAK DESAI741a2952006-02-07 12:59:00 -08002094
2095 if (!(flags & CLONE_NEWNS))
Badari Pulavartye3222c42007-05-08 00:25:21 -07002096 return ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002097
Badari Pulavartye3222c42007-05-08 00:25:21 -07002098 new_ns = dup_mnt_ns(ns, new_fs);
JANAK DESAI741a2952006-02-07 12:59:00 -08002099
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002100 put_mnt_ns(ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -07002101 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002102}
2103
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002104/**
2105 * create_mnt_ns - creates a private namespace and adds a root filesystem
2106 * @mnt: pointer to the new root filesystem mountpoint
2107 */
Linus Torvaldsa2770d82009-12-17 12:51:05 -08002108struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002109{
2110 struct mnt_namespace *new_ns;
2111
2112 new_ns = alloc_mnt_ns();
2113 if (!IS_ERR(new_ns)) {
2114 mnt->mnt_ns = new_ns;
2115 new_ns->root = mnt;
2116 list_add(&new_ns->list, &new_ns->root->mnt_list);
2117 }
2118 return new_ns;
2119}
Linus Torvaldsa2770d82009-12-17 12:51:05 -08002120EXPORT_SYMBOL(create_mnt_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002121
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002122SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2123 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124{
Vegard Nossumeca6f532009-09-18 13:05:45 -07002125 int ret;
2126 char *kernel_type;
2127 char *kernel_dir;
2128 char *kernel_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002129 unsigned long data_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130
Vegard Nossumeca6f532009-09-18 13:05:45 -07002131 ret = copy_mount_string(type, &kernel_type);
2132 if (ret < 0)
2133 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134
Vegard Nossumeca6f532009-09-18 13:05:45 -07002135 kernel_dir = getname(dir_name);
2136 if (IS_ERR(kernel_dir)) {
2137 ret = PTR_ERR(kernel_dir);
2138 goto out_dir;
2139 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140
Vegard Nossumeca6f532009-09-18 13:05:45 -07002141 ret = copy_mount_string(dev_name, &kernel_dev);
2142 if (ret < 0)
2143 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002144
Vegard Nossumeca6f532009-09-18 13:05:45 -07002145 ret = copy_mount_options(data, &data_page);
2146 if (ret < 0)
2147 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002148
Vegard Nossumeca6f532009-09-18 13:05:45 -07002149 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2150 (void *) data_page);
2151
Linus Torvalds1da177e2005-04-16 15:20:36 -07002152 free_page(data_page);
Vegard Nossumeca6f532009-09-18 13:05:45 -07002153out_data:
2154 kfree(kernel_dev);
2155out_dev:
2156 putname(kernel_dir);
2157out_dir:
2158 kfree(kernel_type);
2159out_type:
2160 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161}
2162
2163/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002164 * pivot_root Semantics:
2165 * Moves the root file system of the current process to the directory put_old,
2166 * makes new_root as the new root file system of the current process, and sets
2167 * root/cwd of all processes which had them on the current root to new_root.
2168 *
2169 * Restrictions:
2170 * The new_root and put_old must be directories, and must not be on the
2171 * same file system as the current process root. The put_old must be
2172 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2173 * pointed to by put_old must yield the same directory as new_root. No other
2174 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2175 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08002176 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2177 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2178 * in this situation.
2179 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 * Notes:
2181 * - we don't move root/cwd if they are not at the root (reason: if something
2182 * cared enough to change them, it's probably wrong to force them elsewhere)
2183 * - it's okay to pick a root that isn't the root of a file system, e.g.
2184 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2185 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2186 * first.
2187 */
Heiko Carstens3480b252009-01-14 14:14:16 +01002188SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2189 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002190{
2191 struct vfsmount *tmp;
Al Viro2d8f3032008-07-22 09:59:21 -04002192 struct path new, old, parent_path, root_parent, root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002193 int error;
2194
2195 if (!capable(CAP_SYS_ADMIN))
2196 return -EPERM;
2197
Al Viro2d8f3032008-07-22 09:59:21 -04002198 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199 if (error)
2200 goto out0;
2201 error = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04002202 if (!check_mnt(new.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203 goto out1;
2204
Al Viro2d8f3032008-07-22 09:59:21 -04002205 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002206 if (error)
2207 goto out1;
2208
Al Viro2d8f3032008-07-22 09:59:21 -04002209 error = security_sb_pivotroot(&old, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210 if (error) {
Al Viro2d8f3032008-07-22 09:59:21 -04002211 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002212 goto out1;
2213 }
2214
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002215 get_fs_root(current->fs, &root);
Ram Pai390c6842005-11-07 17:17:51 -05002216 down_write(&namespace_sem);
Al Viro2d8f3032008-07-22 09:59:21 -04002217 mutex_lock(&old.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002218 error = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04002219 if (IS_MNT_SHARED(old.mnt) ||
2220 IS_MNT_SHARED(new.mnt->mnt_parent) ||
Al Viro8c3ee422008-03-22 18:00:39 -04002221 IS_MNT_SHARED(root.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05002222 goto out2;
Al Viro8c3ee422008-03-22 18:00:39 -04002223 if (!check_mnt(root.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224 goto out2;
2225 error = -ENOENT;
Al Virod83c49f2010-04-30 17:17:09 -04002226 if (cant_mount(old.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002227 goto out2;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002228 if (d_unlinked(new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002229 goto out2;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002230 if (d_unlinked(old.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002231 goto out2;
2232 error = -EBUSY;
Al Viro2d8f3032008-07-22 09:59:21 -04002233 if (new.mnt == root.mnt ||
2234 old.mnt == root.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 goto out2; /* loop, on the same file system */
2236 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04002237 if (root.mnt->mnt_root != root.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002238 goto out2; /* not a mountpoint */
Al Viro8c3ee422008-03-22 18:00:39 -04002239 if (root.mnt->mnt_parent == root.mnt)
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07002240 goto out2; /* not attached */
Al Viro2d8f3032008-07-22 09:59:21 -04002241 if (new.mnt->mnt_root != new.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 goto out2; /* not a mountpoint */
Al Viro2d8f3032008-07-22 09:59:21 -04002243 if (new.mnt->mnt_parent == new.mnt)
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07002244 goto out2; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08002245 /* make sure we can reach put_old from new_root */
Al Viro2d8f3032008-07-22 09:59:21 -04002246 tmp = old.mnt;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 spin_lock(&vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002248 if (tmp != new.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 for (;;) {
2250 if (tmp->mnt_parent == tmp)
2251 goto out3; /* already mounted on put_old */
Al Viro2d8f3032008-07-22 09:59:21 -04002252 if (tmp->mnt_parent == new.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 break;
2254 tmp = tmp->mnt_parent;
2255 }
Al Viro2d8f3032008-07-22 09:59:21 -04002256 if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257 goto out3;
Al Viro2d8f3032008-07-22 09:59:21 -04002258 } else if (!is_subdir(old.dentry, new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259 goto out3;
Al Viro2d8f3032008-07-22 09:59:21 -04002260 detach_mnt(new.mnt, &parent_path);
Al Viro8c3ee422008-03-22 18:00:39 -04002261 detach_mnt(root.mnt, &root_parent);
Jan Blunck4ac91372008-02-14 19:34:32 -08002262 /* mount old root on put_old */
Al Viro2d8f3032008-07-22 09:59:21 -04002263 attach_mnt(root.mnt, &old);
Jan Blunck4ac91372008-02-14 19:34:32 -08002264 /* mount new_root on / */
Al Viro2d8f3032008-07-22 09:59:21 -04002265 attach_mnt(new.mnt, &root_parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002266 touch_mnt_namespace(current->nsproxy->mnt_ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 spin_unlock(&vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002268 chroot_fs_refs(&root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 error = 0;
Al Viro1a390682008-03-21 20:48:19 -04002270 path_put(&root_parent);
2271 path_put(&parent_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272out2:
Al Viro2d8f3032008-07-22 09:59:21 -04002273 mutex_unlock(&old.dentry->d_inode->i_mutex);
Ram Pai390c6842005-11-07 17:17:51 -05002274 up_write(&namespace_sem);
Al Viro8c3ee422008-03-22 18:00:39 -04002275 path_put(&root);
Al Viro2d8f3032008-07-22 09:59:21 -04002276 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277out1:
Al Viro2d8f3032008-07-22 09:59:21 -04002278 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 return error;
2281out3:
2282 spin_unlock(&vfsmount_lock);
2283 goto out2;
2284}
2285
2286static void __init init_mount_tree(void)
2287{
2288 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002289 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08002290 struct path root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291
2292 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2293 if (IS_ERR(mnt))
2294 panic("Can't create rootfs");
Trond Myklebust3b22edc2009-06-23 17:29:49 -04002295 ns = create_mnt_ns(mnt);
2296 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002297 panic("Can't allocate initial namespace");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002299 init_task.nsproxy->mnt_ns = ns;
2300 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002301
Jan Blunckac748a02008-02-14 19:34:39 -08002302 root.mnt = ns->root;
2303 root.dentry = ns->root->mnt_root;
2304
2305 set_fs_pwd(current->fs, &root);
2306 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307}
2308
Denis Cheng74bf17c2007-10-16 23:26:30 -07002309void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310{
Eric Dumazet13f14b42008-02-06 01:37:57 -08002311 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002312 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313
Ram Pai390c6842005-11-07 17:17:51 -05002314 init_rwsem(&namespace_sem);
2315
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
Paul Mundt20c2df82007-07-20 10:11:58 +09002317 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318
Ram Paib58fed82005-11-07 17:16:09 -05002319 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320
2321 if (!mount_hashtable)
2322 panic("Failed to allocate mount hash table\n");
2323
Eric Dumazet13f14b42008-02-06 01:37:57 -08002324 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325
Eric Dumazet13f14b42008-02-06 01:37:57 -08002326 for (u = 0; u < HASH_SIZE; u++)
2327 INIT_LIST_HEAD(&mount_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002328
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002329 err = sysfs_init();
2330 if (err)
2331 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002332 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06002333 fs_kobj = kobject_create_and_add("fs", NULL);
2334 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002335 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 init_rootfs();
2337 init_mount_tree();
2338}
2339
Trond Myklebust616511d2009-06-22 15:09:13 -04002340void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341{
Ram Pai70fbcdf2005-11-07 17:17:04 -05002342 LIST_HEAD(umount_list);
Trond Myklebust616511d2009-06-22 15:09:13 -04002343
Al Virod498b252010-02-05 02:21:06 -05002344 if (!atomic_dec_and_test(&ns->count))
Trond Myklebust616511d2009-06-22 15:09:13 -04002345 return;
Ram Pai390c6842005-11-07 17:17:51 -05002346 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002347 spin_lock(&vfsmount_lock);
Al Virod498b252010-02-05 02:21:06 -05002348 umount_tree(ns->root, 0, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 spin_unlock(&vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05002350 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05002351 release_mounts(&umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002352 kfree(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353}
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002354EXPORT_SYMBOL(put_mnt_ns);