blob: de402eb6eafbad3df3957ca7b74ee92256c1e8c4 [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>
Nick Piggin99b7db72010-08-18 04:37:39 +100014#include <linux/spinlock.h>
15#include <linux/percpu.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070016#include <linux/smp_lock.h>
17#include <linux/init.h>
Randy Dunlap15a67dd2006-09-29 01:58:57 -070018#include <linux/kernel.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/acct.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080020#include <linux/capability.h>
Dave Hansen3d733632008-02-15 14:37:59 -080021#include <linux/cpumask.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/module.h>
Andrew Mortonf20a9ea2006-08-14 22:43:23 -070023#include <linux/sysfs.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/seq_file.h>
Kirill Korotaev6b3286e2006-12-08 02:37:56 -080025#include <linux/mnt_namespace.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026#include <linux/namei.h>
Alexey Dobriyanb43f3cb2009-07-08 01:54:37 +040027#include <linux/nsproxy.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/security.h>
29#include <linux/mount.h>
David Howells07f3f052006-09-30 20:52:18 +020030#include <linux/ramfs.h>
Eric Dumazet13f14b42008-02-06 01:37:57 -080031#include <linux/log2.h>
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010032#include <linux/idr.h>
Al Viro5ad4e532009-03-29 19:50:06 -040033#include <linux/fs_struct.h>
Andreas Gruenbacher2504c5d2009-12-17 21:24:27 -050034#include <linux/fsnotify.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <asm/uaccess.h>
36#include <asm/unistd.h>
Ram Pai07b20882005-11-07 17:19:07 -050037#include "pnode.h"
Adrian Bunk948730b2007-07-15 23:41:25 -070038#include "internal.h"
Linus Torvalds1da177e2005-04-16 15:20:36 -070039
Eric Dumazet13f14b42008-02-06 01:37:57 -080040#define HASH_SHIFT ilog2(PAGE_SIZE / sizeof(struct list_head))
41#define HASH_SIZE (1UL << HASH_SHIFT)
42
Al Viro5addc5d2005-11-07 17:15:49 -050043static int event;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010044static DEFINE_IDA(mnt_id_ida);
Miklos Szeredi719f5d72008-03-27 13:06:23 +010045static DEFINE_IDA(mnt_group_ida);
Nick Piggin99b7db72010-08-18 04:37:39 +100046static DEFINE_SPINLOCK(mnt_id_lock);
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
Nick Piggin99b7db72010-08-18 04:37:39 +100058/*
59 * vfsmount lock may be taken for read to prevent changes to the
60 * vfsmount hash, ie. during mountpoint lookups or walking back
61 * up the tree.
62 *
63 * It should be taken for write in all cases where the vfsmount
64 * tree or hash is modified or when a vfsmount structure is modified.
65 */
66DEFINE_BRLOCK(vfsmount_lock);
67
Linus Torvalds1da177e2005-04-16 15:20:36 -070068static inline unsigned long hash(struct vfsmount *mnt, struct dentry *dentry)
69{
Ram Paib58fed82005-11-07 17:16:09 -050070 unsigned long tmp = ((unsigned long)mnt / L1_CACHE_BYTES);
71 tmp += ((unsigned long)dentry / L1_CACHE_BYTES);
Eric Dumazet13f14b42008-02-06 01:37:57 -080072 tmp = tmp + (tmp >> HASH_SHIFT);
73 return tmp & (HASH_SIZE - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -070074}
75
Dave Hansen3d733632008-02-15 14:37:59 -080076#define MNT_WRITER_UNDERFLOW_LIMIT -(1<<16)
77
Nick Piggin99b7db72010-08-18 04:37:39 +100078/*
79 * allocation is serialized by namespace_sem, but we need the spinlock to
80 * serialize with freeing.
81 */
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010082static int mnt_alloc_id(struct vfsmount *mnt)
83{
84 int res;
85
86retry:
87 ida_pre_get(&mnt_id_ida, GFP_KERNEL);
Nick Piggin99b7db72010-08-18 04:37:39 +100088 spin_lock(&mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -040089 res = ida_get_new_above(&mnt_id_ida, mnt_id_start, &mnt->mnt_id);
90 if (!res)
91 mnt_id_start = mnt->mnt_id + 1;
Nick Piggin99b7db72010-08-18 04:37:39 +100092 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +010093 if (res == -EAGAIN)
94 goto retry;
95
96 return res;
97}
98
99static void mnt_free_id(struct vfsmount *mnt)
100{
Al Virof21f6222009-06-24 03:12:00 -0400101 int id = mnt->mnt_id;
Nick Piggin99b7db72010-08-18 04:37:39 +1000102 spin_lock(&mnt_id_lock);
Al Virof21f6222009-06-24 03:12:00 -0400103 ida_remove(&mnt_id_ida, id);
104 if (mnt_id_start > id)
105 mnt_id_start = id;
Nick Piggin99b7db72010-08-18 04:37:39 +1000106 spin_unlock(&mnt_id_lock);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100107}
108
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100109/*
110 * Allocate a new peer group ID
111 *
112 * mnt_group_ida is protected by namespace_sem
113 */
114static int mnt_alloc_group_id(struct vfsmount *mnt)
115{
Al Virof21f6222009-06-24 03:12:00 -0400116 int res;
117
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100118 if (!ida_pre_get(&mnt_group_ida, GFP_KERNEL))
119 return -ENOMEM;
120
Al Virof21f6222009-06-24 03:12:00 -0400121 res = ida_get_new_above(&mnt_group_ida,
122 mnt_group_start,
123 &mnt->mnt_group_id);
124 if (!res)
125 mnt_group_start = mnt->mnt_group_id + 1;
126
127 return res;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100128}
129
130/*
131 * Release a peer group ID
132 */
133void mnt_release_group_id(struct vfsmount *mnt)
134{
Al Virof21f6222009-06-24 03:12:00 -0400135 int id = mnt->mnt_group_id;
136 ida_remove(&mnt_group_ida, id);
137 if (mnt_group_start > id)
138 mnt_group_start = id;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100139 mnt->mnt_group_id = 0;
140}
141
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142struct vfsmount *alloc_vfsmnt(const char *name)
143{
Robert P. J. Dayc3762222007-02-10 01:45:03 -0800144 struct vfsmount *mnt = kmem_cache_zalloc(mnt_cache, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145 if (mnt) {
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100146 int err;
147
148 err = mnt_alloc_id(mnt);
Li Zefan88b38782008-07-21 18:06:36 +0800149 if (err)
150 goto out_free_cache;
151
152 if (name) {
153 mnt->mnt_devname = kstrdup(name, GFP_KERNEL);
154 if (!mnt->mnt_devname)
155 goto out_free_id;
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100156 }
157
Ram Paib58fed82005-11-07 17:16:09 -0500158 atomic_set(&mnt->mnt_count, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 INIT_LIST_HEAD(&mnt->mnt_hash);
160 INIT_LIST_HEAD(&mnt->mnt_child);
161 INIT_LIST_HEAD(&mnt->mnt_mounts);
162 INIT_LIST_HEAD(&mnt->mnt_list);
Miklos Szeredi55e700b2005-07-07 17:57:30 -0700163 INIT_LIST_HEAD(&mnt->mnt_expire);
Ram Pai03e06e62005-11-07 17:19:33 -0500164 INIT_LIST_HEAD(&mnt->mnt_share);
Ram Paia58b0eb2005-11-07 17:20:48 -0500165 INIT_LIST_HEAD(&mnt->mnt_slave_list);
166 INIT_LIST_HEAD(&mnt->mnt_slave);
Andreas Gruenbacher2504c5d2009-12-17 21:24:27 -0500167#ifdef CONFIG_FSNOTIFY
168 INIT_HLIST_HEAD(&mnt->mnt_fsnotify_marks);
169#endif
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000170#ifdef CONFIG_SMP
171 mnt->mnt_writers = alloc_percpu(int);
172 if (!mnt->mnt_writers)
173 goto out_free_devname;
174#else
175 mnt->mnt_writers = 0;
176#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 }
178 return mnt;
Li Zefan88b38782008-07-21 18:06:36 +0800179
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000180#ifdef CONFIG_SMP
181out_free_devname:
182 kfree(mnt->mnt_devname);
183#endif
Li Zefan88b38782008-07-21 18:06:36 +0800184out_free_id:
185 mnt_free_id(mnt);
186out_free_cache:
187 kmem_cache_free(mnt_cache, mnt);
188 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189}
190
Dave Hansen83660252008-02-15 14:37:30 -0800191/*
192 * Most r/o checks on a fs are for operations that take
193 * discrete amounts of time, like a write() or unlink().
194 * We must keep track of when those operations start
195 * (for permission checks) and when they end, so that
196 * we can determine when writes are able to occur to
197 * a filesystem.
198 */
Dave Hansen3d733632008-02-15 14:37:59 -0800199/*
200 * __mnt_is_readonly: check whether a mount is read-only
201 * @mnt: the mount to check for its write status
202 *
203 * This shouldn't be used directly ouside of the VFS.
204 * It does not guarantee that the filesystem will stay
205 * r/w, just that it is right *now*. This can not and
206 * should not be used in place of IS_RDONLY(inode).
207 * mnt_want/drop_write() will _keep_ the filesystem
208 * r/w.
209 */
210int __mnt_is_readonly(struct vfsmount *mnt)
211{
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800212 if (mnt->mnt_flags & MNT_READONLY)
213 return 1;
214 if (mnt->mnt_sb->s_flags & MS_RDONLY)
215 return 1;
216 return 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800217}
218EXPORT_SYMBOL_GPL(__mnt_is_readonly);
219
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000220static inline void inc_mnt_writers(struct vfsmount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800221{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000222#ifdef CONFIG_SMP
223 (*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))++;
224#else
225 mnt->mnt_writers++;
226#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800227}
Dave Hansen3d733632008-02-15 14:37:59 -0800228
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000229static inline void dec_mnt_writers(struct vfsmount *mnt)
Dave Hansen3d733632008-02-15 14:37:59 -0800230{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000231#ifdef CONFIG_SMP
232 (*per_cpu_ptr(mnt->mnt_writers, smp_processor_id()))--;
233#else
234 mnt->mnt_writers--;
235#endif
236}
237
238static unsigned int count_mnt_writers(struct vfsmount *mnt)
239{
240#ifdef CONFIG_SMP
241 unsigned int count = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800242 int cpu;
Dave Hansen3d733632008-02-15 14:37:59 -0800243
244 for_each_possible_cpu(cpu) {
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000245 count += *per_cpu_ptr(mnt->mnt_writers, cpu);
Dave Hansen3d733632008-02-15 14:37:59 -0800246 }
Dave Hansen3d733632008-02-15 14:37:59 -0800247
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000248 return count;
249#else
250 return mnt->mnt_writers;
251#endif
Dave Hansen3d733632008-02-15 14:37:59 -0800252}
253
254/*
255 * Most r/o checks on a fs are for operations that take
256 * discrete amounts of time, like a write() or unlink().
257 * We must keep track of when those operations start
258 * (for permission checks) and when they end, so that
259 * we can determine when writes are able to occur to
260 * a filesystem.
261 */
Dave Hansen83660252008-02-15 14:37:30 -0800262/**
263 * mnt_want_write - get write access to a mount
264 * @mnt: the mount on which to take a write
265 *
266 * This tells the low-level filesystem that a write is
267 * about to be performed to it, and makes sure that
268 * writes are allowed before returning success. When
269 * the write operation is finished, mnt_drop_write()
270 * must be called. This is effectively a refcount.
271 */
272int mnt_want_write(struct vfsmount *mnt)
273{
Dave Hansen3d733632008-02-15 14:37:59 -0800274 int ret = 0;
Dave Hansen3d733632008-02-15 14:37:59 -0800275
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000276 preempt_disable();
277 inc_mnt_writers(mnt);
278 /*
279 * The store to inc_mnt_writers must be visible before we pass
280 * MNT_WRITE_HOLD loop below, so that the slowpath can see our
281 * incremented count after it has set MNT_WRITE_HOLD.
282 */
283 smp_mb();
284 while (mnt->mnt_flags & MNT_WRITE_HOLD)
285 cpu_relax();
286 /*
287 * After the slowpath clears MNT_WRITE_HOLD, mnt_is_readonly will
288 * be set to match its requirements. So we must not load that until
289 * MNT_WRITE_HOLD is cleared.
290 */
291 smp_rmb();
Dave Hansen3d733632008-02-15 14:37:59 -0800292 if (__mnt_is_readonly(mnt)) {
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000293 dec_mnt_writers(mnt);
Dave Hansen3d733632008-02-15 14:37:59 -0800294 ret = -EROFS;
295 goto out;
296 }
Dave Hansen3d733632008-02-15 14:37:59 -0800297out:
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000298 preempt_enable();
Dave Hansen3d733632008-02-15 14:37:59 -0800299 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800300}
301EXPORT_SYMBOL_GPL(mnt_want_write);
302
303/**
npiggin@suse.de96029c42009-04-26 20:25:55 +1000304 * mnt_clone_write - get write access to a mount
305 * @mnt: the mount on which to take a write
306 *
307 * This is effectively like mnt_want_write, except
308 * it must only be used to take an extra write reference
309 * on a mountpoint that we already know has a write reference
310 * on it. This allows some optimisation.
311 *
312 * After finished, mnt_drop_write must be called as usual to
313 * drop the reference.
314 */
315int mnt_clone_write(struct vfsmount *mnt)
316{
317 /* superblock may be r/o */
318 if (__mnt_is_readonly(mnt))
319 return -EROFS;
320 preempt_disable();
321 inc_mnt_writers(mnt);
322 preempt_enable();
323 return 0;
324}
325EXPORT_SYMBOL_GPL(mnt_clone_write);
326
327/**
328 * mnt_want_write_file - get write access to a file's mount
329 * @file: the file who's mount on which to take a write
330 *
331 * This is like mnt_want_write, but it takes a file and can
332 * do some optimisations if the file is open for write already
333 */
334int mnt_want_write_file(struct file *file)
335{
OGAWA Hirofumi2d8dd382009-08-06 15:07:39 -0700336 struct inode *inode = file->f_dentry->d_inode;
337 if (!(file->f_mode & FMODE_WRITE) || special_file(inode->i_mode))
npiggin@suse.de96029c42009-04-26 20:25:55 +1000338 return mnt_want_write(file->f_path.mnt);
339 else
340 return mnt_clone_write(file->f_path.mnt);
341}
342EXPORT_SYMBOL_GPL(mnt_want_write_file);
343
344/**
Dave Hansen83660252008-02-15 14:37:30 -0800345 * mnt_drop_write - give up write access to a mount
346 * @mnt: the mount on which to give up write access
347 *
348 * Tells the low-level filesystem that we are done
349 * performing writes to it. Must be matched with
350 * mnt_want_write() call above.
351 */
352void mnt_drop_write(struct vfsmount *mnt)
353{
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000354 preempt_disable();
355 dec_mnt_writers(mnt);
356 preempt_enable();
Dave Hansen83660252008-02-15 14:37:30 -0800357}
358EXPORT_SYMBOL_GPL(mnt_drop_write);
359
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800360static int mnt_make_readonly(struct vfsmount *mnt)
Dave Hansen83660252008-02-15 14:37:30 -0800361{
Dave Hansen3d733632008-02-15 14:37:59 -0800362 int ret = 0;
363
Nick Piggin99b7db72010-08-18 04:37:39 +1000364 br_write_lock(vfsmount_lock);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000365 mnt->mnt_flags |= MNT_WRITE_HOLD;
366 /*
367 * After storing MNT_WRITE_HOLD, we'll read the counters. This store
368 * should be visible before we do.
369 */
370 smp_mb();
371
372 /*
373 * With writers on hold, if this value is zero, then there are
374 * definitely no active writers (although held writers may subsequently
375 * increment the count, they'll have to wait, and decrement it after
376 * seeing MNT_READONLY).
377 *
378 * It is OK to have counter incremented on one CPU and decremented on
379 * another: the sum will add up correctly. The danger would be when we
380 * sum up each counter, if we read a counter before it is incremented,
381 * but then read another CPU's count which it has been subsequently
382 * decremented from -- we would see more decrements than we should.
383 * MNT_WRITE_HOLD protects against this scenario, because
384 * mnt_want_write first increments count, then smp_mb, then spins on
385 * MNT_WRITE_HOLD, so it can't be decremented by another CPU while
386 * we're counting up here.
387 */
388 if (count_mnt_writers(mnt) > 0)
389 ret = -EBUSY;
390 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800391 mnt->mnt_flags |= MNT_READONLY;
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000392 /*
393 * MNT_READONLY must become visible before ~MNT_WRITE_HOLD, so writers
394 * that become unheld will see MNT_READONLY.
395 */
396 smp_wmb();
397 mnt->mnt_flags &= ~MNT_WRITE_HOLD;
Nick Piggin99b7db72010-08-18 04:37:39 +1000398 br_write_unlock(vfsmount_lock);
Dave Hansen3d733632008-02-15 14:37:59 -0800399 return ret;
Dave Hansen83660252008-02-15 14:37:30 -0800400}
Dave Hansen83660252008-02-15 14:37:30 -0800401
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800402static void __mnt_unmake_readonly(struct vfsmount *mnt)
403{
Nick Piggin99b7db72010-08-18 04:37:39 +1000404 br_write_lock(vfsmount_lock);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800405 mnt->mnt_flags &= ~MNT_READONLY;
Nick Piggin99b7db72010-08-18 04:37:39 +1000406 br_write_unlock(vfsmount_lock);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800407}
408
Sukadev Bhattiprolua3ec9472009-03-04 12:06:34 -0800409void simple_set_mnt(struct vfsmount *mnt, struct super_block *sb)
David Howells454e2392006-06-23 02:02:57 -0700410{
411 mnt->mnt_sb = sb;
412 mnt->mnt_root = dget(sb->s_root);
David Howells454e2392006-06-23 02:02:57 -0700413}
414
415EXPORT_SYMBOL(simple_set_mnt);
416
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417void free_vfsmnt(struct vfsmount *mnt)
418{
419 kfree(mnt->mnt_devname);
Miklos Szeredi73cd49e2008-03-26 22:11:34 +0100420 mnt_free_id(mnt);
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000421#ifdef CONFIG_SMP
422 free_percpu(mnt->mnt_writers);
423#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424 kmem_cache_free(mnt_cache, mnt);
425}
426
427/*
Ram Paia05964f2005-11-07 17:20:17 -0500428 * find the first or last mount at @dentry on vfsmount @mnt depending on
429 * @dir. If @dir is set return the first mount else return the last mount.
Nick Piggin99b7db72010-08-18 04:37:39 +1000430 * vfsmount_lock must be held for read or write.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 */
Ram Paia05964f2005-11-07 17:20:17 -0500432struct vfsmount *__lookup_mnt(struct vfsmount *mnt, struct dentry *dentry,
433 int dir)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434{
Ram Paib58fed82005-11-07 17:16:09 -0500435 struct list_head *head = mount_hashtable + hash(mnt, dentry);
436 struct list_head *tmp = head;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 struct vfsmount *p, *found = NULL;
438
Linus Torvalds1da177e2005-04-16 15:20:36 -0700439 for (;;) {
Ram Paia05964f2005-11-07 17:20:17 -0500440 tmp = dir ? tmp->next : tmp->prev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 p = NULL;
442 if (tmp == head)
443 break;
444 p = list_entry(tmp, struct vfsmount, mnt_hash);
445 if (p->mnt_parent == mnt && p->mnt_mountpoint == dentry) {
Ram Paia05964f2005-11-07 17:20:17 -0500446 found = p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 break;
448 }
449 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 return found;
451}
452
Ram Paia05964f2005-11-07 17:20:17 -0500453/*
454 * lookup_mnt increments the ref count before returning
455 * the vfsmount struct.
456 */
Al Viro1c755af2009-04-18 14:06:57 -0400457struct vfsmount *lookup_mnt(struct path *path)
Ram Paia05964f2005-11-07 17:20:17 -0500458{
459 struct vfsmount *child_mnt;
Nick Piggin99b7db72010-08-18 04:37:39 +1000460
461 br_read_lock(vfsmount_lock);
Al Viro1c755af2009-04-18 14:06:57 -0400462 if ((child_mnt = __lookup_mnt(path->mnt, path->dentry, 1)))
Ram Paia05964f2005-11-07 17:20:17 -0500463 mntget(child_mnt);
Nick Piggin99b7db72010-08-18 04:37:39 +1000464 br_read_unlock(vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -0500465 return child_mnt;
466}
467
Linus Torvalds1da177e2005-04-16 15:20:36 -0700468static inline int check_mnt(struct vfsmount *mnt)
469{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800470 return mnt->mnt_ns == current->nsproxy->mnt_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471}
472
Nick Piggin99b7db72010-08-18 04:37:39 +1000473/*
474 * vfsmount lock must be held for write
475 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800476static void touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500477{
478 if (ns) {
479 ns->event = ++event;
480 wake_up_interruptible(&ns->poll);
481 }
482}
483
Nick Piggin99b7db72010-08-18 04:37:39 +1000484/*
485 * vfsmount lock must be held for write
486 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800487static void __touch_mnt_namespace(struct mnt_namespace *ns)
Al Viro5addc5d2005-11-07 17:15:49 -0500488{
489 if (ns && ns->event != event) {
490 ns->event = event;
491 wake_up_interruptible(&ns->poll);
492 }
493}
494
Nick Piggin99b7db72010-08-18 04:37:39 +1000495/*
496 * vfsmount lock must be held for write
497 */
Al Viro1a390682008-03-21 20:48:19 -0400498static void detach_mnt(struct vfsmount *mnt, struct path *old_path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499{
Al Viro1a390682008-03-21 20:48:19 -0400500 old_path->dentry = mnt->mnt_mountpoint;
501 old_path->mnt = mnt->mnt_parent;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 mnt->mnt_parent = mnt;
503 mnt->mnt_mountpoint = mnt->mnt_root;
504 list_del_init(&mnt->mnt_child);
505 list_del_init(&mnt->mnt_hash);
Al Viro1a390682008-03-21 20:48:19 -0400506 old_path->dentry->d_mounted--;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507}
508
Nick Piggin99b7db72010-08-18 04:37:39 +1000509/*
510 * vfsmount lock must be held for write
511 */
Ram Paib90fa9a2005-11-07 17:19:50 -0500512void mnt_set_mountpoint(struct vfsmount *mnt, struct dentry *dentry,
513 struct vfsmount *child_mnt)
514{
515 child_mnt->mnt_parent = mntget(mnt);
516 child_mnt->mnt_mountpoint = dget(dentry);
517 dentry->d_mounted++;
518}
519
Nick Piggin99b7db72010-08-18 04:37:39 +1000520/*
521 * vfsmount lock must be held for write
522 */
Al Viro1a390682008-03-21 20:48:19 -0400523static void attach_mnt(struct vfsmount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524{
Al Viro1a390682008-03-21 20:48:19 -0400525 mnt_set_mountpoint(path->mnt, path->dentry, mnt);
Ram Paib90fa9a2005-11-07 17:19:50 -0500526 list_add_tail(&mnt->mnt_hash, mount_hashtable +
Al Viro1a390682008-03-21 20:48:19 -0400527 hash(path->mnt, path->dentry));
528 list_add_tail(&mnt->mnt_child, &path->mnt->mnt_mounts);
Ram Paib90fa9a2005-11-07 17:19:50 -0500529}
530
531/*
Nick Piggin99b7db72010-08-18 04:37:39 +1000532 * vfsmount lock must be held for write
Ram Paib90fa9a2005-11-07 17:19:50 -0500533 */
534static void commit_tree(struct vfsmount *mnt)
535{
536 struct vfsmount *parent = mnt->mnt_parent;
537 struct vfsmount *m;
538 LIST_HEAD(head);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800539 struct mnt_namespace *n = parent->mnt_ns;
Ram Paib90fa9a2005-11-07 17:19:50 -0500540
541 BUG_ON(parent == mnt);
542
543 list_add_tail(&head, &mnt->mnt_list);
544 list_for_each_entry(m, &head, mnt_list)
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800545 m->mnt_ns = n;
Ram Paib90fa9a2005-11-07 17:19:50 -0500546 list_splice(&head, n->list.prev);
547
548 list_add_tail(&mnt->mnt_hash, mount_hashtable +
549 hash(parent, mnt->mnt_mountpoint));
550 list_add_tail(&mnt->mnt_child, &parent->mnt_mounts);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -0800551 touch_mnt_namespace(n);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552}
553
554static struct vfsmount *next_mnt(struct vfsmount *p, struct vfsmount *root)
555{
556 struct list_head *next = p->mnt_mounts.next;
557 if (next == &p->mnt_mounts) {
558 while (1) {
559 if (p == root)
560 return NULL;
561 next = p->mnt_child.next;
562 if (next != &p->mnt_parent->mnt_mounts)
563 break;
564 p = p->mnt_parent;
565 }
566 }
567 return list_entry(next, struct vfsmount, mnt_child);
568}
569
Ram Pai9676f0c2005-11-07 17:21:20 -0500570static struct vfsmount *skip_mnt_tree(struct vfsmount *p)
571{
572 struct list_head *prev = p->mnt_mounts.prev;
573 while (prev != &p->mnt_mounts) {
574 p = list_entry(prev, struct vfsmount, mnt_child);
575 prev = p->mnt_mounts.prev;
576 }
577 return p;
578}
579
Ram Pai36341f62005-11-07 17:17:22 -0500580static struct vfsmount *clone_mnt(struct vfsmount *old, struct dentry *root,
581 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582{
583 struct super_block *sb = old->mnt_sb;
584 struct vfsmount *mnt = alloc_vfsmnt(old->mnt_devname);
585
586 if (mnt) {
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100587 if (flag & (CL_SLAVE | CL_PRIVATE))
588 mnt->mnt_group_id = 0; /* not a peer of original */
589 else
590 mnt->mnt_group_id = old->mnt_group_id;
591
592 if ((flag & CL_MAKE_SHARED) && !mnt->mnt_group_id) {
593 int err = mnt_alloc_group_id(mnt);
594 if (err)
595 goto out_free;
596 }
597
Linus Torvalds1da177e2005-04-16 15:20:36 -0700598 mnt->mnt_flags = old->mnt_flags;
599 atomic_inc(&sb->s_active);
600 mnt->mnt_sb = sb;
601 mnt->mnt_root = dget(root);
602 mnt->mnt_mountpoint = mnt->mnt_root;
603 mnt->mnt_parent = mnt;
Ram Paib90fa9a2005-11-07 17:19:50 -0500604
Ram Pai5afe0022005-11-07 17:21:01 -0500605 if (flag & CL_SLAVE) {
606 list_add(&mnt->mnt_slave, &old->mnt_slave_list);
607 mnt->mnt_master = old;
608 CLEAR_MNT_SHARED(mnt);
Al Viro8aec0802007-06-07 12:20:32 -0400609 } else if (!(flag & CL_PRIVATE)) {
Al Viro796a6b52010-01-16 13:28:47 -0500610 if ((flag & CL_MAKE_SHARED) || IS_MNT_SHARED(old))
Ram Pai5afe0022005-11-07 17:21:01 -0500611 list_add(&mnt->mnt_share, &old->mnt_share);
612 if (IS_MNT_SLAVE(old))
613 list_add(&mnt->mnt_slave, &old->mnt_slave);
614 mnt->mnt_master = old->mnt_master;
615 }
Ram Paib90fa9a2005-11-07 17:19:50 -0500616 if (flag & CL_MAKE_SHARED)
617 set_mnt_shared(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700618
619 /* stick the duplicate mount on the same expiry list
620 * as the original if that was on one */
Ram Pai36341f62005-11-07 17:17:22 -0500621 if (flag & CL_EXPIRE) {
Ram Pai36341f62005-11-07 17:17:22 -0500622 if (!list_empty(&old->mnt_expire))
623 list_add(&mnt->mnt_expire, &old->mnt_expire);
Ram Pai36341f62005-11-07 17:17:22 -0500624 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700625 }
626 return mnt;
Miklos Szeredi719f5d72008-03-27 13:06:23 +0100627
628 out_free:
629 free_vfsmnt(mnt);
630 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631}
632
Al Viro7b7b1ac2005-11-07 17:13:39 -0500633static inline void __mntput(struct vfsmount *mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700634{
635 struct super_block *sb = mnt->mnt_sb;
Dave Hansen3d733632008-02-15 14:37:59 -0800636 /*
Dave Hansen3d733632008-02-15 14:37:59 -0800637 * This probably indicates that somebody messed
638 * up a mnt_want/drop_write() pair. If this
639 * happens, the filesystem was probably unable
640 * to make r/w->r/o transitions.
641 */
npiggin@suse.ded3ef3d72009-04-26 20:25:54 +1000642 /*
643 * atomic_dec_and_lock() used to deal with ->mnt_count decrements
644 * provides barriers, so count_mnt_writers() below is safe. AV
645 */
646 WARN_ON(count_mnt_writers(mnt));
Andreas Gruenbacherca9c7262009-12-17 21:24:27 -0500647 fsnotify_vfsmount_delete(mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 dput(mnt->mnt_root);
649 free_vfsmnt(mnt);
650 deactivate_super(sb);
651}
652
Al Viro7b7b1ac2005-11-07 17:13:39 -0500653void mntput_no_expire(struct vfsmount *mnt)
654{
655repeat:
Nick Piggin99b7db72010-08-18 04:37:39 +1000656 if (atomic_add_unless(&mnt->mnt_count, -1, 1))
657 return;
658 br_write_lock(vfsmount_lock);
659 if (!atomic_dec_and_test(&mnt->mnt_count)) {
660 br_write_unlock(vfsmount_lock);
661 return;
Al Viro7b7b1ac2005-11-07 17:13:39 -0500662 }
Nick Piggin99b7db72010-08-18 04:37:39 +1000663 if (likely(!mnt->mnt_pinned)) {
664 br_write_unlock(vfsmount_lock);
665 __mntput(mnt);
666 return;
667 }
668 atomic_add(mnt->mnt_pinned + 1, &mnt->mnt_count);
669 mnt->mnt_pinned = 0;
670 br_write_unlock(vfsmount_lock);
671 acct_auto_close_mnt(mnt);
672 goto repeat;
Al Viro7b7b1ac2005-11-07 17:13:39 -0500673}
Al Viro7b7b1ac2005-11-07 17:13:39 -0500674EXPORT_SYMBOL(mntput_no_expire);
675
676void mnt_pin(struct vfsmount *mnt)
677{
Nick Piggin99b7db72010-08-18 04:37:39 +1000678 br_write_lock(vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500679 mnt->mnt_pinned++;
Nick Piggin99b7db72010-08-18 04:37:39 +1000680 br_write_unlock(vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500681}
682
683EXPORT_SYMBOL(mnt_pin);
684
685void mnt_unpin(struct vfsmount *mnt)
686{
Nick Piggin99b7db72010-08-18 04:37:39 +1000687 br_write_lock(vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500688 if (mnt->mnt_pinned) {
689 atomic_inc(&mnt->mnt_count);
690 mnt->mnt_pinned--;
691 }
Nick Piggin99b7db72010-08-18 04:37:39 +1000692 br_write_unlock(vfsmount_lock);
Al Viro7b7b1ac2005-11-07 17:13:39 -0500693}
694
695EXPORT_SYMBOL(mnt_unpin);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700696
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800697static inline void mangle(struct seq_file *m, const char *s)
698{
699 seq_escape(m, s, " \t\n\\");
700}
701
702/*
703 * Simple .show_options callback for filesystems which don't want to
704 * implement more complex mount option showing.
705 *
706 * See also save_mount_options().
707 */
708int generic_show_options(struct seq_file *m, struct vfsmount *mnt)
709{
Al Viro2a32ceb2009-05-08 16:05:57 -0400710 const char *options;
711
712 rcu_read_lock();
713 options = rcu_dereference(mnt->mnt_sb->s_options);
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800714
715 if (options != NULL && options[0]) {
716 seq_putc(m, ',');
717 mangle(m, options);
718 }
Al Viro2a32ceb2009-05-08 16:05:57 -0400719 rcu_read_unlock();
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800720
721 return 0;
722}
723EXPORT_SYMBOL(generic_show_options);
724
725/*
726 * If filesystem uses generic_show_options(), this function should be
727 * called from the fill_super() callback.
728 *
729 * The .remount_fs callback usually needs to be handled in a special
730 * way, to make sure, that previous options are not overwritten if the
731 * remount fails.
732 *
733 * Also note, that if the filesystem's .remount_fs function doesn't
734 * reset all options to their default value, but changes only newly
735 * given options, then the displayed options will not reflect reality
736 * any more.
737 */
738void save_mount_options(struct super_block *sb, char *options)
739{
Al Viro2a32ceb2009-05-08 16:05:57 -0400740 BUG_ON(sb->s_options);
741 rcu_assign_pointer(sb->s_options, kstrdup(options, GFP_KERNEL));
Miklos Szeredib3b304a2008-02-08 04:21:35 -0800742}
743EXPORT_SYMBOL(save_mount_options);
744
Al Viro2a32ceb2009-05-08 16:05:57 -0400745void replace_mount_options(struct super_block *sb, char *options)
746{
747 char *old = sb->s_options;
748 rcu_assign_pointer(sb->s_options, options);
749 if (old) {
750 synchronize_rcu();
751 kfree(old);
752 }
753}
754EXPORT_SYMBOL(replace_mount_options);
755
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100756#ifdef CONFIG_PROC_FS
Linus Torvalds1da177e2005-04-16 15:20:36 -0700757/* iterator */
758static void *m_start(struct seq_file *m, loff_t *pos)
759{
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100760 struct proc_mounts *p = m->private;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700761
Ram Pai390c6842005-11-07 17:17:51 -0500762 down_read(&namespace_sem);
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100763 return seq_list_start(&p->ns->list, *pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764}
765
766static void *m_next(struct seq_file *m, void *v, loff_t *pos)
767{
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100768 struct proc_mounts *p = m->private;
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700769
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100770 return seq_list_next(v, &p->ns->list, pos);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771}
772
773static void m_stop(struct seq_file *m, void *v)
774{
Ram Pai390c6842005-11-07 17:17:51 -0500775 up_read(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776}
777
Al Viro9f5596a2010-02-05 00:40:25 -0500778int mnt_had_events(struct proc_mounts *p)
779{
780 struct mnt_namespace *ns = p->ns;
781 int res = 0;
782
Nick Piggin99b7db72010-08-18 04:37:39 +1000783 br_read_lock(vfsmount_lock);
Al Viro9f5596a2010-02-05 00:40:25 -0500784 if (p->event != ns->event) {
785 p->event = ns->event;
786 res = 1;
787 }
Nick Piggin99b7db72010-08-18 04:37:39 +1000788 br_read_unlock(vfsmount_lock);
Al Viro9f5596a2010-02-05 00:40:25 -0500789
790 return res;
791}
792
Ram Pai2d4d4862008-03-27 13:06:25 +0100793struct proc_fs_info {
794 int flag;
795 const char *str;
796};
797
Eric Paris2069f452008-07-04 09:47:13 +1000798static int show_sb_opts(struct seq_file *m, struct super_block *sb)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799{
Ram Pai2d4d4862008-03-27 13:06:25 +0100800 static const struct proc_fs_info fs_info[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801 { MS_SYNCHRONOUS, ",sync" },
802 { MS_DIRSYNC, ",dirsync" },
803 { MS_MANDLOCK, ",mand" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804 { 0, NULL }
805 };
Ram Pai2d4d4862008-03-27 13:06:25 +0100806 const struct proc_fs_info *fs_infop;
807
808 for (fs_infop = fs_info; fs_infop->flag; fs_infop++) {
809 if (sb->s_flags & fs_infop->flag)
810 seq_puts(m, fs_infop->str);
811 }
Eric Paris2069f452008-07-04 09:47:13 +1000812
813 return security_sb_show_options(m, sb);
Ram Pai2d4d4862008-03-27 13:06:25 +0100814}
815
816static void show_mnt_opts(struct seq_file *m, struct vfsmount *mnt)
817{
818 static const struct proc_fs_info mnt_info[] = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819 { MNT_NOSUID, ",nosuid" },
820 { MNT_NODEV, ",nodev" },
821 { MNT_NOEXEC, ",noexec" },
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -0800822 { MNT_NOATIME, ",noatime" },
823 { MNT_NODIRATIME, ",nodiratime" },
Valerie Henson47ae32d2006-12-13 00:34:34 -0800824 { MNT_RELATIME, ",relatime" },
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825 { 0, NULL }
826 };
Ram Pai2d4d4862008-03-27 13:06:25 +0100827 const struct proc_fs_info *fs_infop;
828
829 for (fs_infop = mnt_info; fs_infop->flag; fs_infop++) {
830 if (mnt->mnt_flags & fs_infop->flag)
831 seq_puts(m, fs_infop->str);
832 }
833}
834
835static void show_type(struct seq_file *m, struct super_block *sb)
836{
837 mangle(m, sb->s_type->name);
838 if (sb->s_subtype && sb->s_subtype[0]) {
839 seq_putc(m, '.');
840 mangle(m, sb->s_subtype);
841 }
842}
843
844static int show_vfsmnt(struct seq_file *m, void *v)
845{
846 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
847 int err = 0;
Jan Blunckc32c2f62008-02-14 19:38:43 -0800848 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700849
850 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
851 seq_putc(m, ' ');
Jan Blunckc32c2f62008-02-14 19:38:43 -0800852 seq_path(m, &mnt_path, " \t\n\\");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700853 seq_putc(m, ' ');
Ram Pai2d4d4862008-03-27 13:06:25 +0100854 show_type(m, mnt->mnt_sb);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -0800855 seq_puts(m, __mnt_is_readonly(mnt) ? " ro" : " rw");
Eric Paris2069f452008-07-04 09:47:13 +1000856 err = show_sb_opts(m, mnt->mnt_sb);
857 if (err)
858 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +0100859 show_mnt_opts(m, mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700860 if (mnt->mnt_sb->s_op->show_options)
861 err = mnt->mnt_sb->s_op->show_options(m, mnt);
862 seq_puts(m, " 0 0\n");
Eric Paris2069f452008-07-04 09:47:13 +1000863out:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700864 return err;
865}
866
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100867const struct seq_operations mounts_op = {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700868 .start = m_start,
869 .next = m_next,
870 .stop = m_stop,
871 .show = show_vfsmnt
872};
873
Ram Pai2d4d4862008-03-27 13:06:25 +0100874static int show_mountinfo(struct seq_file *m, void *v)
875{
876 struct proc_mounts *p = m->private;
877 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
878 struct super_block *sb = mnt->mnt_sb;
879 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
880 struct path root = p->root;
881 int err = 0;
882
883 seq_printf(m, "%i %i %u:%u ", mnt->mnt_id, mnt->mnt_parent->mnt_id,
884 MAJOR(sb->s_dev), MINOR(sb->s_dev));
885 seq_dentry(m, mnt->mnt_root, " \t\n\\");
886 seq_putc(m, ' ');
887 seq_path_root(m, &mnt_path, &root, " \t\n\\");
888 if (root.mnt != p->root.mnt || root.dentry != p->root.dentry) {
889 /*
890 * Mountpoint is outside root, discard that one. Ugly,
891 * but less so than trying to do that in iterator in a
892 * race-free way (due to renames).
893 */
894 return SEQ_SKIP;
895 }
896 seq_puts(m, mnt->mnt_flags & MNT_READONLY ? " ro" : " rw");
897 show_mnt_opts(m, mnt);
898
899 /* Tagged fields ("foo:X" or "bar") */
900 if (IS_MNT_SHARED(mnt))
901 seq_printf(m, " shared:%i", mnt->mnt_group_id);
Miklos Szeredi97e7e0f2008-03-27 13:06:26 +0100902 if (IS_MNT_SLAVE(mnt)) {
903 int master = mnt->mnt_master->mnt_group_id;
904 int dom = get_dominating_id(mnt, &p->root);
905 seq_printf(m, " master:%i", master);
906 if (dom && dom != master)
907 seq_printf(m, " propagate_from:%i", dom);
908 }
Ram Pai2d4d4862008-03-27 13:06:25 +0100909 if (IS_MNT_UNBINDABLE(mnt))
910 seq_puts(m, " unbindable");
911
912 /* Filesystem specific data */
913 seq_puts(m, " - ");
914 show_type(m, sb);
915 seq_putc(m, ' ');
916 mangle(m, mnt->mnt_devname ? mnt->mnt_devname : "none");
917 seq_puts(m, sb->s_flags & MS_RDONLY ? " ro" : " rw");
Eric Paris2069f452008-07-04 09:47:13 +1000918 err = show_sb_opts(m, sb);
919 if (err)
920 goto out;
Ram Pai2d4d4862008-03-27 13:06:25 +0100921 if (sb->s_op->show_options)
922 err = sb->s_op->show_options(m, mnt);
923 seq_putc(m, '\n');
Eric Paris2069f452008-07-04 09:47:13 +1000924out:
Ram Pai2d4d4862008-03-27 13:06:25 +0100925 return err;
926}
927
928const struct seq_operations mountinfo_op = {
929 .start = m_start,
930 .next = m_next,
931 .stop = m_stop,
932 .show = show_mountinfo,
933};
934
Chuck Leverb4629fe2006-03-20 13:44:12 -0500935static int show_vfsstat(struct seq_file *m, void *v)
936{
Pavel Emelianovb0765fb2007-07-15 23:39:55 -0700937 struct vfsmount *mnt = list_entry(v, struct vfsmount, mnt_list);
Jan Blunckc32c2f62008-02-14 19:38:43 -0800938 struct path mnt_path = { .dentry = mnt->mnt_root, .mnt = mnt };
Chuck Leverb4629fe2006-03-20 13:44:12 -0500939 int err = 0;
940
941 /* device */
942 if (mnt->mnt_devname) {
943 seq_puts(m, "device ");
944 mangle(m, mnt->mnt_devname);
945 } else
946 seq_puts(m, "no device");
947
948 /* mount point */
949 seq_puts(m, " mounted on ");
Jan Blunckc32c2f62008-02-14 19:38:43 -0800950 seq_path(m, &mnt_path, " \t\n\\");
Chuck Leverb4629fe2006-03-20 13:44:12 -0500951 seq_putc(m, ' ');
952
953 /* file system type */
954 seq_puts(m, "with fstype ");
Ram Pai2d4d4862008-03-27 13:06:25 +0100955 show_type(m, mnt->mnt_sb);
Chuck Leverb4629fe2006-03-20 13:44:12 -0500956
957 /* optional statistics */
958 if (mnt->mnt_sb->s_op->show_stats) {
959 seq_putc(m, ' ');
960 err = mnt->mnt_sb->s_op->show_stats(m, mnt);
961 }
962
963 seq_putc(m, '\n');
964 return err;
965}
966
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100967const struct seq_operations mountstats_op = {
Chuck Leverb4629fe2006-03-20 13:44:12 -0500968 .start = m_start,
969 .next = m_next,
970 .stop = m_stop,
971 .show = show_vfsstat,
972};
Miklos Szeredia1a2c402008-03-27 13:06:24 +0100973#endif /* CONFIG_PROC_FS */
Chuck Leverb4629fe2006-03-20 13:44:12 -0500974
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975/**
976 * may_umount_tree - check if a mount tree is busy
977 * @mnt: root of mount tree
978 *
979 * This is called to check if a tree of mounts has any
980 * open files, pwds, chroots or sub mounts that are
981 * busy.
982 */
983int may_umount_tree(struct vfsmount *mnt)
984{
Ram Pai36341f62005-11-07 17:17:22 -0500985 int actual_refs = 0;
986 int minimum_refs = 0;
987 struct vfsmount *p;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988
Nick Piggin99b7db72010-08-18 04:37:39 +1000989 br_read_lock(vfsmount_lock);
Ram Pai36341f62005-11-07 17:17:22 -0500990 for (p = mnt; p; p = next_mnt(p, mnt)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991 actual_refs += atomic_read(&p->mnt_count);
992 minimum_refs += 2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993 }
Nick Piggin99b7db72010-08-18 04:37:39 +1000994 br_read_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995
996 if (actual_refs > minimum_refs)
Ian Kente3474a82006-03-27 01:14:51 -0800997 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700998
Ian Kente3474a82006-03-27 01:14:51 -0800999 return 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001000}
1001
1002EXPORT_SYMBOL(may_umount_tree);
1003
1004/**
1005 * may_umount - check if a mount point is busy
1006 * @mnt: root of mount
1007 *
1008 * This is called to check if a mount point has any
1009 * open files, pwds, chroots or sub mounts. If the
1010 * mount has sub mounts this will return busy
1011 * regardless of whether the sub mounts are busy.
1012 *
1013 * Doesn't take quota and stuff into account. IOW, in some cases it will
1014 * give false negatives. The main reason why it's here is that we need
1015 * a non-destructive way to look for easily umountable filesystems.
1016 */
1017int may_umount(struct vfsmount *mnt)
1018{
Ian Kente3474a82006-03-27 01:14:51 -08001019 int ret = 1;
Al Viro8ad08d82010-01-16 12:56:08 -05001020 down_read(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10001021 br_read_lock(vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001022 if (propagate_mount_busy(mnt, 2))
Ian Kente3474a82006-03-27 01:14:51 -08001023 ret = 0;
Nick Piggin99b7db72010-08-18 04:37:39 +10001024 br_read_unlock(vfsmount_lock);
Al Viro8ad08d82010-01-16 12:56:08 -05001025 up_read(&namespace_sem);
Ram Paia05964f2005-11-07 17:20:17 -05001026 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027}
1028
1029EXPORT_SYMBOL(may_umount);
1030
Ram Paib90fa9a2005-11-07 17:19:50 -05001031void release_mounts(struct list_head *head)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001032{
Ram Pai70fbcdf2005-11-07 17:17:04 -05001033 struct vfsmount *mnt;
Miklos Szeredibf066c72006-01-08 01:03:19 -08001034 while (!list_empty(head)) {
Pavel Emelianovb5e61812007-05-08 00:30:19 -07001035 mnt = list_first_entry(head, struct vfsmount, mnt_hash);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001036 list_del_init(&mnt->mnt_hash);
1037 if (mnt->mnt_parent != mnt) {
1038 struct dentry *dentry;
1039 struct vfsmount *m;
Nick Piggin99b7db72010-08-18 04:37:39 +10001040
1041 br_write_lock(vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001042 dentry = mnt->mnt_mountpoint;
1043 m = mnt->mnt_parent;
1044 mnt->mnt_mountpoint = mnt->mnt_root;
1045 mnt->mnt_parent = mnt;
Al Viro7c4b93d2008-03-21 23:59:49 -04001046 m->mnt_ghosts--;
Nick Piggin99b7db72010-08-18 04:37:39 +10001047 br_write_unlock(vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001048 dput(dentry);
1049 mntput(m);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 }
1051 mntput(mnt);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001052 }
1053}
1054
Nick Piggin99b7db72010-08-18 04:37:39 +10001055/*
1056 * vfsmount lock must be held for write
1057 * namespace_sem must be held for write
1058 */
Ram Paia05964f2005-11-07 17:20:17 -05001059void umount_tree(struct vfsmount *mnt, int propagate, struct list_head *kill)
Ram Pai70fbcdf2005-11-07 17:17:04 -05001060{
1061 struct vfsmount *p;
1062
Akinobu Mita1bfba4e2006-06-26 00:24:40 -07001063 for (p = mnt; p; p = next_mnt(p, mnt))
1064 list_move(&p->mnt_hash, kill);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001065
Ram Paia05964f2005-11-07 17:20:17 -05001066 if (propagate)
1067 propagate_umount(kill);
1068
Ram Pai70fbcdf2005-11-07 17:17:04 -05001069 list_for_each_entry(p, kill, mnt_hash) {
1070 list_del_init(&p->mnt_expire);
1071 list_del_init(&p->mnt_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08001072 __touch_mnt_namespace(p->mnt_ns);
1073 p->mnt_ns = NULL;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001074 list_del_init(&p->mnt_child);
Al Viro7c4b93d2008-03-21 23:59:49 -04001075 if (p->mnt_parent != p) {
1076 p->mnt_parent->mnt_ghosts++;
Al Virof30ac312006-02-01 07:53:21 -05001077 p->mnt_mountpoint->d_mounted--;
Al Viro7c4b93d2008-03-21 23:59:49 -04001078 }
Ram Paia05964f2005-11-07 17:20:17 -05001079 change_mnt_propagation(p, MS_PRIVATE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080 }
1081}
1082
Al Viroc35038b2008-03-22 00:46:23 -04001083static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts);
1084
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085static int do_umount(struct vfsmount *mnt, int flags)
1086{
Ram Paib58fed82005-11-07 17:16:09 -05001087 struct super_block *sb = mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 int retval;
Ram Pai70fbcdf2005-11-07 17:17:04 -05001089 LIST_HEAD(umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001090
1091 retval = security_sb_umount(mnt, flags);
1092 if (retval)
1093 return retval;
1094
1095 /*
1096 * Allow userspace to request a mountpoint be expired rather than
1097 * unmounting unconditionally. Unmount only happens if:
1098 * (1) the mark is already set (the mark is cleared by mntput())
1099 * (2) the usage count == 1 [parent vfsmount] + 1 [sys_umount]
1100 */
1101 if (flags & MNT_EXPIRE) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08001102 if (mnt == current->fs->root.mnt ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 flags & (MNT_FORCE | MNT_DETACH))
1104 return -EINVAL;
1105
1106 if (atomic_read(&mnt->mnt_count) != 2)
1107 return -EBUSY;
1108
1109 if (!xchg(&mnt->mnt_expiry_mark, 1))
1110 return -EAGAIN;
1111 }
1112
1113 /*
1114 * If we may have to abort operations to get out of this
1115 * mount, and they will themselves hold resources we must
1116 * allow the fs to do things. In the Unix tradition of
1117 * 'Gee thats tricky lets do it in userspace' the umount_begin
1118 * might fail to complete on the first run through as other tasks
1119 * must return, and the like. Thats for the mount program to worry
1120 * about for the moment.
1121 */
1122
Al Viro42faad92008-04-24 07:21:56 -04001123 if (flags & MNT_FORCE && sb->s_op->umount_begin) {
Al Viro42faad92008-04-24 07:21:56 -04001124 sb->s_op->umount_begin(sb);
Al Viro42faad92008-04-24 07:21:56 -04001125 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126
1127 /*
1128 * No sense to grab the lock for this test, but test itself looks
1129 * somewhat bogus. Suggestions for better replacement?
1130 * Ho-hum... In principle, we might treat that as umount + switch
1131 * to rootfs. GC would eventually take care of the old vfsmount.
1132 * Actually it makes sense, especially if rootfs would contain a
1133 * /reboot - static binary that would close all descriptors and
1134 * call reboot(9). Then init(8) could umount root and exec /reboot.
1135 */
Jan Blunck6ac08c32008-02-14 19:34:38 -08001136 if (mnt == current->fs->root.mnt && !(flags & MNT_DETACH)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001137 /*
1138 * Special case for "unmounting" root ...
1139 * we just try to remount it readonly.
1140 */
1141 down_write(&sb->s_umount);
Al Viro4aa98cf2009-05-08 13:36:58 -04001142 if (!(sb->s_flags & MS_RDONLY))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001143 retval = do_remount_sb(sb, MS_RDONLY, NULL, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001144 up_write(&sb->s_umount);
1145 return retval;
1146 }
1147
Ram Pai390c6842005-11-07 17:17:51 -05001148 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10001149 br_write_lock(vfsmount_lock);
Al Viro5addc5d2005-11-07 17:15:49 -05001150 event++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001151
Al Viroc35038b2008-03-22 00:46:23 -04001152 if (!(flags & MNT_DETACH))
1153 shrink_submounts(mnt, &umount_list);
1154
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155 retval = -EBUSY;
Ram Paia05964f2005-11-07 17:20:17 -05001156 if (flags & MNT_DETACH || !propagate_mount_busy(mnt, 2)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 if (!list_empty(&mnt->mnt_list))
Ram Paia05964f2005-11-07 17:20:17 -05001158 umount_tree(mnt, 1, &umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 retval = 0;
1160 }
Nick Piggin99b7db72010-08-18 04:37:39 +10001161 br_write_unlock(vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05001162 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001163 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 return retval;
1165}
1166
1167/*
1168 * Now umount can handle mount points as well as block devices.
1169 * This is important for filesystems which use unnamed block devices.
1170 *
1171 * We now support a flag for forced unmount like the other 'big iron'
1172 * unixes. Our API is identical to OSF/1 to avoid making a mess of AMD
1173 */
1174
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001175SYSCALL_DEFINE2(umount, char __user *, name, int, flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001176{
Al Viro2d8f3032008-07-22 09:59:21 -04001177 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 int retval;
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001179 int lookup_flags = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180
Miklos Szeredidb1f05b2010-02-10 12:15:53 +01001181 if (flags & ~(MNT_FORCE | MNT_DETACH | MNT_EXPIRE | UMOUNT_NOFOLLOW))
1182 return -EINVAL;
1183
1184 if (!(flags & UMOUNT_NOFOLLOW))
1185 lookup_flags |= LOOKUP_FOLLOW;
1186
1187 retval = user_path_at(AT_FDCWD, name, lookup_flags, &path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188 if (retval)
1189 goto out;
1190 retval = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04001191 if (path.dentry != path.mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192 goto dput_and_out;
Al Viro2d8f3032008-07-22 09:59:21 -04001193 if (!check_mnt(path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194 goto dput_and_out;
1195
1196 retval = -EPERM;
1197 if (!capable(CAP_SYS_ADMIN))
1198 goto dput_and_out;
1199
Al Viro2d8f3032008-07-22 09:59:21 -04001200 retval = do_umount(path.mnt, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201dput_and_out:
Jan Blunck429731b2008-02-14 19:34:31 -08001202 /* we mustn't call path_put() as that would clear mnt_expiry_mark */
Al Viro2d8f3032008-07-22 09:59:21 -04001203 dput(path.dentry);
1204 mntput_no_expire(path.mnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205out:
1206 return retval;
1207}
1208
1209#ifdef __ARCH_WANT_SYS_OLDUMOUNT
1210
1211/*
Ram Paib58fed82005-11-07 17:16:09 -05001212 * The 2.0 compatible umount. No flags.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01001214SYSCALL_DEFINE1(oldumount, char __user *, name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215{
Ram Paib58fed82005-11-07 17:16:09 -05001216 return sys_umount(name, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001217}
1218
1219#endif
1220
Al Viro2d92ab32008-08-02 00:51:11 -04001221static int mount_is_safe(struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001222{
1223 if (capable(CAP_SYS_ADMIN))
1224 return 0;
1225 return -EPERM;
1226#ifdef notyet
Al Viro2d92ab32008-08-02 00:51:11 -04001227 if (S_ISLNK(path->dentry->d_inode->i_mode))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 return -EPERM;
Al Viro2d92ab32008-08-02 00:51:11 -04001229 if (path->dentry->d_inode->i_mode & S_ISVTX) {
David Howellsda9592e2008-11-14 10:39:05 +11001230 if (current_uid() != path->dentry->d_inode->i_uid)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001231 return -EPERM;
1232 }
Al Viro2d92ab32008-08-02 00:51:11 -04001233 if (inode_permission(path->dentry->d_inode, MAY_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001234 return -EPERM;
1235 return 0;
1236#endif
1237}
1238
Ram Paib90fa9a2005-11-07 17:19:50 -05001239struct vfsmount *copy_tree(struct vfsmount *mnt, struct dentry *dentry,
Ram Pai36341f62005-11-07 17:17:22 -05001240 int flag)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001241{
1242 struct vfsmount *res, *p, *q, *r, *s;
Al Viro1a390682008-03-21 20:48:19 -04001243 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001244
Ram Pai9676f0c2005-11-07 17:21:20 -05001245 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(mnt))
1246 return NULL;
1247
Ram Pai36341f62005-11-07 17:17:22 -05001248 res = q = clone_mnt(mnt, dentry, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249 if (!q)
1250 goto Enomem;
1251 q->mnt_mountpoint = mnt->mnt_mountpoint;
1252
1253 p = mnt;
Domen Puncerfdadd652005-09-10 00:27:07 -07001254 list_for_each_entry(r, &mnt->mnt_mounts, mnt_child) {
Jan Blunck7ec02ef2008-04-29 00:59:40 -07001255 if (!is_subdir(r->mnt_mountpoint, dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256 continue;
1257
1258 for (s = r; s; s = next_mnt(s, r)) {
Ram Pai9676f0c2005-11-07 17:21:20 -05001259 if (!(flag & CL_COPY_ALL) && IS_MNT_UNBINDABLE(s)) {
1260 s = skip_mnt_tree(s);
1261 continue;
1262 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263 while (p != s->mnt_parent) {
1264 p = p->mnt_parent;
1265 q = q->mnt_parent;
1266 }
1267 p = s;
Al Viro1a390682008-03-21 20:48:19 -04001268 path.mnt = q;
1269 path.dentry = p->mnt_mountpoint;
Ram Pai36341f62005-11-07 17:17:22 -05001270 q = clone_mnt(p, p->mnt_root, flag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 if (!q)
1272 goto Enomem;
Nick Piggin99b7db72010-08-18 04:37:39 +10001273 br_write_lock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274 list_add_tail(&q->mnt_list, &res->mnt_list);
Al Viro1a390682008-03-21 20:48:19 -04001275 attach_mnt(q, &path);
Nick Piggin99b7db72010-08-18 04:37:39 +10001276 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277 }
1278 }
1279 return res;
Ram Paib58fed82005-11-07 17:16:09 -05001280Enomem:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281 if (res) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001282 LIST_HEAD(umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001283 br_write_lock(vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001284 umount_tree(res, 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001285 br_write_unlock(vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001286 release_mounts(&umount_list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001287 }
1288 return NULL;
1289}
1290
Al Viro589ff872009-04-18 03:28:19 -04001291struct vfsmount *collect_mounts(struct path *path)
Al Viro8aec0802007-06-07 12:20:32 -04001292{
1293 struct vfsmount *tree;
Al Viro1a60a282008-03-22 16:19:49 -04001294 down_write(&namespace_sem);
Al Viro589ff872009-04-18 03:28:19 -04001295 tree = copy_tree(path->mnt, path->dentry, CL_COPY_ALL | CL_PRIVATE);
Al Viro1a60a282008-03-22 16:19:49 -04001296 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001297 return tree;
1298}
1299
1300void drop_collected_mounts(struct vfsmount *mnt)
1301{
1302 LIST_HEAD(umount_list);
Al Viro1a60a282008-03-22 16:19:49 -04001303 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10001304 br_write_lock(vfsmount_lock);
Al Viro8aec0802007-06-07 12:20:32 -04001305 umount_tree(mnt, 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001306 br_write_unlock(vfsmount_lock);
Al Viro1a60a282008-03-22 16:19:49 -04001307 up_write(&namespace_sem);
Al Viro8aec0802007-06-07 12:20:32 -04001308 release_mounts(&umount_list);
1309}
1310
Al Viro1f707132010-01-30 22:51:25 -05001311int iterate_mounts(int (*f)(struct vfsmount *, void *), void *arg,
1312 struct vfsmount *root)
1313{
1314 struct vfsmount *mnt;
1315 int res = f(root, arg);
1316 if (res)
1317 return res;
1318 list_for_each_entry(mnt, &root->mnt_list, mnt_list) {
1319 res = f(mnt, arg);
1320 if (res)
1321 return res;
1322 }
1323 return 0;
1324}
1325
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001326static void cleanup_group_ids(struct vfsmount *mnt, struct vfsmount *end)
1327{
1328 struct vfsmount *p;
1329
1330 for (p = mnt; p != end; p = next_mnt(p, mnt)) {
1331 if (p->mnt_group_id && !IS_MNT_SHARED(p))
1332 mnt_release_group_id(p);
1333 }
1334}
1335
1336static int invent_group_ids(struct vfsmount *mnt, bool recurse)
1337{
1338 struct vfsmount *p;
1339
1340 for (p = mnt; p; p = recurse ? next_mnt(p, mnt) : NULL) {
1341 if (!p->mnt_group_id && !IS_MNT_SHARED(p)) {
1342 int err = mnt_alloc_group_id(p);
1343 if (err) {
1344 cleanup_group_ids(mnt, p);
1345 return err;
1346 }
1347 }
1348 }
1349
1350 return 0;
1351}
1352
Ram Paib90fa9a2005-11-07 17:19:50 -05001353/*
1354 * @source_mnt : mount tree to be attached
Ram Pai21444402005-11-07 17:20:03 -05001355 * @nd : place the mount tree @source_mnt is attached
1356 * @parent_nd : if non-null, detach the source_mnt from its parent and
1357 * store the parent mount and mountpoint dentry.
1358 * (done when source_mnt is moved)
Ram Paib90fa9a2005-11-07 17:19:50 -05001359 *
1360 * NOTE: in the table below explains the semantics when a source mount
1361 * of a given type is attached to a destination mount of a given type.
Ram Pai9676f0c2005-11-07 17:21:20 -05001362 * ---------------------------------------------------------------------------
1363 * | BIND MOUNT OPERATION |
1364 * |**************************************************************************
1365 * | source-->| shared | private | slave | unbindable |
1366 * | dest | | | | |
1367 * | | | | | | |
1368 * | v | | | | |
1369 * |**************************************************************************
1370 * | shared | shared (++) | shared (+) | shared(+++)| invalid |
1371 * | | | | | |
1372 * |non-shared| shared (+) | private | slave (*) | invalid |
1373 * ***************************************************************************
Ram Paib90fa9a2005-11-07 17:19:50 -05001374 * A bind operation clones the source mount and mounts the clone on the
1375 * destination mount.
1376 *
1377 * (++) the cloned mount is propagated to all the mounts in the propagation
1378 * tree of the destination mount and the cloned mount is added to
1379 * the peer group of the source mount.
1380 * (+) the cloned mount is created under the destination mount and is marked
1381 * as shared. The cloned mount is added to the peer group of the source
1382 * mount.
Ram Pai5afe0022005-11-07 17:21:01 -05001383 * (+++) the mount is propagated to all the mounts in the propagation tree
1384 * of the destination mount and the cloned mount is made slave
1385 * of the same master as that of the source mount. The cloned mount
1386 * is marked as 'shared and slave'.
1387 * (*) the cloned mount is made a slave of the same master as that of the
1388 * source mount.
1389 *
Ram Pai9676f0c2005-11-07 17:21:20 -05001390 * ---------------------------------------------------------------------------
1391 * | MOVE MOUNT OPERATION |
1392 * |**************************************************************************
1393 * | source-->| shared | private | slave | unbindable |
1394 * | dest | | | | |
1395 * | | | | | | |
1396 * | v | | | | |
1397 * |**************************************************************************
1398 * | shared | shared (+) | shared (+) | shared(+++) | invalid |
1399 * | | | | | |
1400 * |non-shared| shared (+*) | private | slave (*) | unbindable |
1401 * ***************************************************************************
Ram Pai5afe0022005-11-07 17:21:01 -05001402 *
1403 * (+) the mount is moved to the destination. And is then propagated to
1404 * all the mounts in the propagation tree of the destination mount.
Ram Pai21444402005-11-07 17:20:03 -05001405 * (+*) the mount is moved to the destination.
Ram Pai5afe0022005-11-07 17:21:01 -05001406 * (+++) the mount is moved to the destination and is then propagated to
1407 * all the mounts belonging to the destination mount's propagation tree.
1408 * the mount is marked as 'shared and slave'.
1409 * (*) the mount continues to be a slave at the new location.
Ram Paib90fa9a2005-11-07 17:19:50 -05001410 *
1411 * if the source mount is a tree, the operations explained above is
1412 * applied to each mount in the tree.
1413 * Must be called without spinlocks held, since this function can sleep
1414 * in allocations.
1415 */
1416static int attach_recursive_mnt(struct vfsmount *source_mnt,
Al Viro1a390682008-03-21 20:48:19 -04001417 struct path *path, struct path *parent_path)
Ram Paib90fa9a2005-11-07 17:19:50 -05001418{
1419 LIST_HEAD(tree_list);
Al Viro1a390682008-03-21 20:48:19 -04001420 struct vfsmount *dest_mnt = path->mnt;
1421 struct dentry *dest_dentry = path->dentry;
Ram Paib90fa9a2005-11-07 17:19:50 -05001422 struct vfsmount *child, *p;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001423 int err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001424
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001425 if (IS_MNT_SHARED(dest_mnt)) {
1426 err = invent_group_ids(source_mnt, true);
1427 if (err)
1428 goto out;
1429 }
1430 err = propagate_mnt(dest_mnt, dest_dentry, source_mnt, &tree_list);
1431 if (err)
1432 goto out_cleanup_ids;
Ram Paib90fa9a2005-11-07 17:19:50 -05001433
Nick Piggin99b7db72010-08-18 04:37:39 +10001434 br_write_lock(vfsmount_lock);
Al Virodf1a1ad2010-01-16 12:57:40 -05001435
Ram Paib90fa9a2005-11-07 17:19:50 -05001436 if (IS_MNT_SHARED(dest_mnt)) {
1437 for (p = source_mnt; p; p = next_mnt(p, source_mnt))
1438 set_mnt_shared(p);
1439 }
Al Viro1a390682008-03-21 20:48:19 -04001440 if (parent_path) {
1441 detach_mnt(source_mnt, parent_path);
1442 attach_mnt(source_mnt, path);
Al Viroe5d67f02009-04-07 12:15:39 -04001443 touch_mnt_namespace(parent_path->mnt->mnt_ns);
Ram Pai21444402005-11-07 17:20:03 -05001444 } else {
1445 mnt_set_mountpoint(dest_mnt, dest_dentry, source_mnt);
1446 commit_tree(source_mnt);
1447 }
Ram Paib90fa9a2005-11-07 17:19:50 -05001448
1449 list_for_each_entry_safe(child, p, &tree_list, mnt_hash) {
1450 list_del_init(&child->mnt_hash);
1451 commit_tree(child);
1452 }
Nick Piggin99b7db72010-08-18 04:37:39 +10001453 br_write_unlock(vfsmount_lock);
1454
Ram Paib90fa9a2005-11-07 17:19:50 -05001455 return 0;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001456
1457 out_cleanup_ids:
1458 if (IS_MNT_SHARED(dest_mnt))
1459 cleanup_group_ids(source_mnt, NULL);
1460 out:
1461 return err;
Ram Paib90fa9a2005-11-07 17:19:50 -05001462}
1463
Al Viro8c3ee422008-03-22 18:00:39 -04001464static int graft_tree(struct vfsmount *mnt, struct path *path)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465{
1466 int err;
1467 if (mnt->mnt_sb->s_flags & MS_NOUSER)
1468 return -EINVAL;
1469
Al Viro8c3ee422008-03-22 18:00:39 -04001470 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471 S_ISDIR(mnt->mnt_root->d_inode->i_mode))
1472 return -ENOTDIR;
1473
1474 err = -ENOENT;
Al Viro8c3ee422008-03-22 18:00:39 -04001475 mutex_lock(&path->dentry->d_inode->i_mutex);
Al Virod83c49f2010-04-30 17:17:09 -04001476 if (cant_mount(path->dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001477 goto out_unlock;
1478
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04001479 if (!d_unlinked(path->dentry))
Al Viro8c3ee422008-03-22 18:00:39 -04001480 err = attach_recursive_mnt(mnt, path, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481out_unlock:
Al Viro8c3ee422008-03-22 18:00:39 -04001482 mutex_unlock(&path->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483 return err;
1484}
1485
1486/*
Ram Pai07b20882005-11-07 17:19:07 -05001487 * recursively change the type of the mountpoint.
1488 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001489static int do_change_type(struct path *path, int flag)
Ram Pai07b20882005-11-07 17:19:07 -05001490{
Al Viro2d92ab32008-08-02 00:51:11 -04001491 struct vfsmount *m, *mnt = path->mnt;
Ram Pai07b20882005-11-07 17:19:07 -05001492 int recurse = flag & MS_REC;
1493 int type = flag & ~MS_REC;
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001494 int err = 0;
Ram Pai07b20882005-11-07 17:19:07 -05001495
Miklos Szerediee6f9582007-05-08 00:30:40 -07001496 if (!capable(CAP_SYS_ADMIN))
1497 return -EPERM;
1498
Al Viro2d92ab32008-08-02 00:51:11 -04001499 if (path->dentry != path->mnt->mnt_root)
Ram Pai07b20882005-11-07 17:19:07 -05001500 return -EINVAL;
1501
1502 down_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001503 if (type == MS_SHARED) {
1504 err = invent_group_ids(mnt, recurse);
1505 if (err)
1506 goto out_unlock;
1507 }
1508
Nick Piggin99b7db72010-08-18 04:37:39 +10001509 br_write_lock(vfsmount_lock);
Ram Pai07b20882005-11-07 17:19:07 -05001510 for (m = mnt; m; m = (recurse ? next_mnt(m, mnt) : NULL))
1511 change_mnt_propagation(m, type);
Nick Piggin99b7db72010-08-18 04:37:39 +10001512 br_write_unlock(vfsmount_lock);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001513
1514 out_unlock:
Ram Pai07b20882005-11-07 17:19:07 -05001515 up_write(&namespace_sem);
Miklos Szeredi719f5d72008-03-27 13:06:23 +01001516 return err;
Ram Pai07b20882005-11-07 17:19:07 -05001517}
1518
1519/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001520 * do loopback mount.
1521 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001522static int do_loopback(struct path *path, char *old_name,
Eric Sandeen2dafe1c2008-02-08 04:22:12 -08001523 int recurse)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524{
Al Viro2d92ab32008-08-02 00:51:11 -04001525 struct path old_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 struct vfsmount *mnt = NULL;
Al Viro2d92ab32008-08-02 00:51:11 -04001527 int err = mount_is_safe(path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 if (err)
1529 return err;
1530 if (!old_name || !*old_name)
1531 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001532 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001533 if (err)
1534 return err;
1535
Ram Pai390c6842005-11-07 17:17:51 -05001536 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001538 if (IS_MNT_UNBINDABLE(old_path.mnt))
Jan Blunck4ac91372008-02-14 19:34:32 -08001539 goto out;
Ram Pai9676f0c2005-11-07 17:21:20 -05001540
Al Viro2d92ab32008-08-02 00:51:11 -04001541 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
Al Viroccd48bc2005-11-07 17:15:04 -05001542 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543
Al Viroccd48bc2005-11-07 17:15:04 -05001544 err = -ENOMEM;
1545 if (recurse)
Al Viro2d92ab32008-08-02 00:51:11 -04001546 mnt = copy_tree(old_path.mnt, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001547 else
Al Viro2d92ab32008-08-02 00:51:11 -04001548 mnt = clone_mnt(old_path.mnt, old_path.dentry, 0);
Al Viroccd48bc2005-11-07 17:15:04 -05001549
1550 if (!mnt)
1551 goto out;
1552
Al Viro2d92ab32008-08-02 00:51:11 -04001553 err = graft_tree(mnt, path);
Al Viroccd48bc2005-11-07 17:15:04 -05001554 if (err) {
Ram Pai70fbcdf2005-11-07 17:17:04 -05001555 LIST_HEAD(umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001556
1557 br_write_lock(vfsmount_lock);
Ram Paia05964f2005-11-07 17:20:17 -05001558 umount_tree(mnt, 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10001559 br_write_unlock(vfsmount_lock);
Ram Pai70fbcdf2005-11-07 17:17:04 -05001560 release_mounts(&umount_list);
Ram Pai5b83d2c2005-11-07 17:16:29 -05001561 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562
Al Viroccd48bc2005-11-07 17:15:04 -05001563out:
Ram Pai390c6842005-11-07 17:17:51 -05001564 up_write(&namespace_sem);
Al Viro2d92ab32008-08-02 00:51:11 -04001565 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 return err;
1567}
1568
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001569static int change_mount_flags(struct vfsmount *mnt, int ms_flags)
1570{
1571 int error = 0;
1572 int readonly_request = 0;
1573
1574 if (ms_flags & MS_RDONLY)
1575 readonly_request = 1;
1576 if (readonly_request == __mnt_is_readonly(mnt))
1577 return 0;
1578
1579 if (readonly_request)
1580 error = mnt_make_readonly(mnt);
1581 else
1582 __mnt_unmake_readonly(mnt);
1583 return error;
1584}
1585
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586/*
1587 * change filesystem flags. dir should be a physical root of filesystem.
1588 * If you've mounted a non-root directory somewhere and want to do remount
1589 * on it - tough luck.
1590 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001591static int do_remount(struct path *path, int flags, int mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 void *data)
1593{
1594 int err;
Al Viro2d92ab32008-08-02 00:51:11 -04001595 struct super_block *sb = path->mnt->mnt_sb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001596
1597 if (!capable(CAP_SYS_ADMIN))
1598 return -EPERM;
1599
Al Viro2d92ab32008-08-02 00:51:11 -04001600 if (!check_mnt(path->mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 return -EINVAL;
1602
Al Viro2d92ab32008-08-02 00:51:11 -04001603 if (path->dentry != path->mnt->mnt_root)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604 return -EINVAL;
1605
1606 down_write(&sb->s_umount);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001607 if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04001608 err = change_mount_flags(path->mnt, flags);
Al Viro4aa98cf2009-05-08 13:36:58 -04001609 else
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08001610 err = do_remount_sb(sb, flags, data, 0);
Al Viro7b43a792010-01-16 13:01:26 -05001611 if (!err) {
Nick Piggin99b7db72010-08-18 04:37:39 +10001612 br_write_lock(vfsmount_lock);
Valerie Aurora495d6c92010-01-26 14:20:47 -05001613 mnt_flags |= path->mnt->mnt_flags & MNT_PROPAGATION_MASK;
Al Viro2d92ab32008-08-02 00:51:11 -04001614 path->mnt->mnt_flags = mnt_flags;
Nick Piggin99b7db72010-08-18 04:37:39 +10001615 br_write_unlock(vfsmount_lock);
Al Viro7b43a792010-01-16 13:01:26 -05001616 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617 up_write(&sb->s_umount);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001618 if (!err) {
Nick Piggin99b7db72010-08-18 04:37:39 +10001619 br_write_lock(vfsmount_lock);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001620 touch_mnt_namespace(path->mnt->mnt_ns);
Nick Piggin99b7db72010-08-18 04:37:39 +10001621 br_write_unlock(vfsmount_lock);
Dan Williams0e55a7c2008-09-26 19:01:20 -07001622 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 return err;
1624}
1625
Ram Pai9676f0c2005-11-07 17:21:20 -05001626static inline int tree_contains_unbindable(struct vfsmount *mnt)
1627{
1628 struct vfsmount *p;
1629 for (p = mnt; p; p = next_mnt(p, mnt)) {
1630 if (IS_MNT_UNBINDABLE(p))
1631 return 1;
1632 }
1633 return 0;
1634}
1635
Al Viro0a0d8a42008-08-02 00:55:27 -04001636static int do_move_mount(struct path *path, char *old_name)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001637{
Al Viro2d92ab32008-08-02 00:51:11 -04001638 struct path old_path, parent_path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001639 struct vfsmount *p;
1640 int err = 0;
1641 if (!capable(CAP_SYS_ADMIN))
1642 return -EPERM;
1643 if (!old_name || !*old_name)
1644 return -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001645 err = kern_path(old_name, LOOKUP_FOLLOW, &old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 if (err)
1647 return err;
1648
Ram Pai390c6842005-11-07 17:17:51 -05001649 down_write(&namespace_sem);
Al Viro2d92ab32008-08-02 00:51:11 -04001650 while (d_mountpoint(path->dentry) &&
Al Viro9393bd02009-04-18 13:58:15 -04001651 follow_down(path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652 ;
1653 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001654 if (!check_mnt(path->mnt) || !check_mnt(old_path.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655 goto out;
1656
1657 err = -ENOENT;
Al Viro2d92ab32008-08-02 00:51:11 -04001658 mutex_lock(&path->dentry->d_inode->i_mutex);
Al Virod83c49f2010-04-30 17:17:09 -04001659 if (cant_mount(path->dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660 goto out1;
1661
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04001662 if (d_unlinked(path->dentry))
Ram Pai21444402005-11-07 17:20:03 -05001663 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664
1665 err = -EINVAL;
Al Viro2d92ab32008-08-02 00:51:11 -04001666 if (old_path.dentry != old_path.mnt->mnt_root)
Ram Pai21444402005-11-07 17:20:03 -05001667 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668
Al Viro2d92ab32008-08-02 00:51:11 -04001669 if (old_path.mnt == old_path.mnt->mnt_parent)
Ram Pai21444402005-11-07 17:20:03 -05001670 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671
Al Viro2d92ab32008-08-02 00:51:11 -04001672 if (S_ISDIR(path->dentry->d_inode->i_mode) !=
1673 S_ISDIR(old_path.dentry->d_inode->i_mode))
Ram Pai21444402005-11-07 17:20:03 -05001674 goto out1;
1675 /*
1676 * Don't move a mount residing in a shared parent.
1677 */
Al Viro2d92ab32008-08-02 00:51:11 -04001678 if (old_path.mnt->mnt_parent &&
1679 IS_MNT_SHARED(old_path.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05001680 goto out1;
Ram Pai9676f0c2005-11-07 17:21:20 -05001681 /*
1682 * Don't move a mount tree containing unbindable mounts to a destination
1683 * mount which is shared.
1684 */
Al Viro2d92ab32008-08-02 00:51:11 -04001685 if (IS_MNT_SHARED(path->mnt) &&
1686 tree_contains_unbindable(old_path.mnt))
Ram Pai9676f0c2005-11-07 17:21:20 -05001687 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688 err = -ELOOP;
Al Viro2d92ab32008-08-02 00:51:11 -04001689 for (p = path->mnt; p->mnt_parent != p; p = p->mnt_parent)
1690 if (p == old_path.mnt)
Ram Pai21444402005-11-07 17:20:03 -05001691 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692
Al Viro2d92ab32008-08-02 00:51:11 -04001693 err = attach_recursive_mnt(old_path.mnt, path, &parent_path);
Jan Blunck4ac91372008-02-14 19:34:32 -08001694 if (err)
Ram Pai21444402005-11-07 17:20:03 -05001695 goto out1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696
1697 /* if the mount is moved, it should no longer be expire
1698 * automatically */
Al Viro2d92ab32008-08-02 00:51:11 -04001699 list_del_init(&old_path.mnt->mnt_expire);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700out1:
Al Viro2d92ab32008-08-02 00:51:11 -04001701 mutex_unlock(&path->dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702out:
Ram Pai390c6842005-11-07 17:17:51 -05001703 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 if (!err)
Al Viro1a390682008-03-21 20:48:19 -04001705 path_put(&parent_path);
Al Viro2d92ab32008-08-02 00:51:11 -04001706 path_put(&old_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 return err;
1708}
1709
1710/*
1711 * create a new mount for userspace and request it to be added into the
1712 * namespace's tree
1713 */
Al Viro0a0d8a42008-08-02 00:55:27 -04001714static int do_new_mount(struct path *path, char *type, int flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 int mnt_flags, char *name, void *data)
1716{
1717 struct vfsmount *mnt;
1718
Vegard Nossumeca6f532009-09-18 13:05:45 -07001719 if (!type)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 return -EINVAL;
1721
1722 /* we need capabilities... */
1723 if (!capable(CAP_SYS_ADMIN))
1724 return -EPERM;
1725
Al Viro7f78d4c2009-05-08 13:34:06 -04001726 lock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 mnt = do_kern_mount(type, flags, name, data);
Al Viro7f78d4c2009-05-08 13:34:06 -04001728 unlock_kernel();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 if (IS_ERR(mnt))
1730 return PTR_ERR(mnt);
1731
Al Viro2d92ab32008-08-02 00:51:11 -04001732 return do_add_mount(mnt, path, mnt_flags, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733}
1734
1735/*
1736 * add a mount into a namespace's mount tree
1737 * - provide the option of adding the new mount to an expiration list
1738 */
Al Viro8d66bf52008-08-01 09:05:54 -04001739int do_add_mount(struct vfsmount *newmnt, struct path *path,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740 int mnt_flags, struct list_head *fslist)
1741{
1742 int err;
1743
Al Viro80893522010-02-05 09:30:46 -05001744 mnt_flags &= ~(MNT_SHARED | MNT_WRITE_HOLD | MNT_INTERNAL);
Al Viro27d55f12010-01-16 13:07:36 -05001745
Ram Pai390c6842005-11-07 17:17:51 -05001746 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 /* Something was mounted here while we slept */
Al Viro8d66bf52008-08-01 09:05:54 -04001748 while (d_mountpoint(path->dentry) &&
Al Viro9393bd02009-04-18 13:58:15 -04001749 follow_down(path))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 ;
1751 err = -EINVAL;
Al Virodd5cae62009-04-07 12:21:18 -04001752 if (!(mnt_flags & MNT_SHRINKABLE) && !check_mnt(path->mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001753 goto unlock;
1754
1755 /* Refuse the same filesystem on the same mount point */
1756 err = -EBUSY;
Al Viro8d66bf52008-08-01 09:05:54 -04001757 if (path->mnt->mnt_sb == newmnt->mnt_sb &&
1758 path->mnt->mnt_root == path->dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 goto unlock;
1760
1761 err = -EINVAL;
1762 if (S_ISLNK(newmnt->mnt_root->d_inode->i_mode))
1763 goto unlock;
1764
1765 newmnt->mnt_flags = mnt_flags;
Al Viro8d66bf52008-08-01 09:05:54 -04001766 if ((err = graft_tree(newmnt, path)))
Ram Pai5b83d2c2005-11-07 17:16:29 -05001767 goto unlock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768
Al Viro6758f952008-03-22 16:14:30 -04001769 if (fslist) /* add to the specified expiration list */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001770 list_add_tail(&newmnt->mnt_expire, fslist);
Al Viro6758f952008-03-22 16:14:30 -04001771
Ram Pai390c6842005-11-07 17:17:51 -05001772 up_write(&namespace_sem);
Ram Pai5b83d2c2005-11-07 17:16:29 -05001773 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774
1775unlock:
Ram Pai390c6842005-11-07 17:17:51 -05001776 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001777 mntput(newmnt);
1778 return err;
1779}
1780
1781EXPORT_SYMBOL_GPL(do_add_mount);
1782
Trond Myklebust5528f9112006-06-09 09:34:17 -04001783/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784 * process a list of expirable mountpoints with the intent of discarding any
1785 * mountpoints that aren't in use and haven't been touched since last we came
1786 * here
1787 */
1788void mark_mounts_for_expiry(struct list_head *mounts)
1789{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 struct vfsmount *mnt, *next;
1791 LIST_HEAD(graveyard);
Al Virobcc5c7d2008-03-22 00:21:53 -04001792 LIST_HEAD(umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793
1794 if (list_empty(mounts))
1795 return;
1796
Al Virobcc5c7d2008-03-22 00:21:53 -04001797 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10001798 br_write_lock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799
1800 /* extract from the expiration list every vfsmount that matches the
1801 * following criteria:
1802 * - only referenced by its parent vfsmount
1803 * - still marked for expiry (marked on the last call here; marks are
1804 * cleared by mntput())
1805 */
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001806 list_for_each_entry_safe(mnt, next, mounts, mnt_expire) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 if (!xchg(&mnt->mnt_expiry_mark, 1) ||
Al Virobcc5c7d2008-03-22 00:21:53 -04001808 propagate_mount_busy(mnt, 1))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 continue;
Miklos Szeredi55e700b2005-07-07 17:57:30 -07001810 list_move(&mnt->mnt_expire, &graveyard);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 }
Al Virobcc5c7d2008-03-22 00:21:53 -04001812 while (!list_empty(&graveyard)) {
1813 mnt = list_first_entry(&graveyard, struct vfsmount, mnt_expire);
1814 touch_mnt_namespace(mnt->mnt_ns);
1815 umount_tree(mnt, 1, &umounts);
1816 }
Nick Piggin99b7db72010-08-18 04:37:39 +10001817 br_write_unlock(vfsmount_lock);
Al Virobcc5c7d2008-03-22 00:21:53 -04001818 up_write(&namespace_sem);
1819
1820 release_mounts(&umounts);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821}
1822
1823EXPORT_SYMBOL_GPL(mark_mounts_for_expiry);
1824
1825/*
Trond Myklebust5528f9112006-06-09 09:34:17 -04001826 * Ripoff of 'select_parent()'
1827 *
1828 * search the list of submounts for a given mountpoint, and move any
1829 * shrinkable submounts to the 'graveyard' list.
1830 */
1831static int select_submounts(struct vfsmount *parent, struct list_head *graveyard)
1832{
1833 struct vfsmount *this_parent = parent;
1834 struct list_head *next;
1835 int found = 0;
1836
1837repeat:
1838 next = this_parent->mnt_mounts.next;
1839resume:
1840 while (next != &this_parent->mnt_mounts) {
1841 struct list_head *tmp = next;
1842 struct vfsmount *mnt = list_entry(tmp, struct vfsmount, mnt_child);
1843
1844 next = tmp->next;
1845 if (!(mnt->mnt_flags & MNT_SHRINKABLE))
1846 continue;
1847 /*
1848 * Descend a level if the d_mounts list is non-empty.
1849 */
1850 if (!list_empty(&mnt->mnt_mounts)) {
1851 this_parent = mnt;
1852 goto repeat;
1853 }
1854
1855 if (!propagate_mount_busy(mnt, 1)) {
Trond Myklebust5528f9112006-06-09 09:34:17 -04001856 list_move_tail(&mnt->mnt_expire, graveyard);
1857 found++;
1858 }
1859 }
1860 /*
1861 * All done at this level ... ascend and resume the search
1862 */
1863 if (this_parent != parent) {
1864 next = this_parent->mnt_child.next;
1865 this_parent = this_parent->mnt_parent;
1866 goto resume;
1867 }
1868 return found;
1869}
1870
1871/*
1872 * process a list of expirable mountpoints with the intent of discarding any
1873 * submounts of a specific parent mountpoint
Nick Piggin99b7db72010-08-18 04:37:39 +10001874 *
1875 * vfsmount_lock must be held for write
Trond Myklebust5528f9112006-06-09 09:34:17 -04001876 */
Al Viroc35038b2008-03-22 00:46:23 -04001877static void shrink_submounts(struct vfsmount *mnt, struct list_head *umounts)
Trond Myklebust5528f9112006-06-09 09:34:17 -04001878{
1879 LIST_HEAD(graveyard);
Al Viroc35038b2008-03-22 00:46:23 -04001880 struct vfsmount *m;
Trond Myklebust5528f9112006-06-09 09:34:17 -04001881
Trond Myklebust5528f9112006-06-09 09:34:17 -04001882 /* extract submounts of 'mountpoint' from the expiration list */
Al Viroc35038b2008-03-22 00:46:23 -04001883 while (select_submounts(mnt, &graveyard)) {
Al Virobcc5c7d2008-03-22 00:21:53 -04001884 while (!list_empty(&graveyard)) {
Al Viroc35038b2008-03-22 00:46:23 -04001885 m = list_first_entry(&graveyard, struct vfsmount,
Al Virobcc5c7d2008-03-22 00:21:53 -04001886 mnt_expire);
Eric W. Biedermanafef80b2008-11-12 13:26:54 -08001887 touch_mnt_namespace(m->mnt_ns);
1888 umount_tree(m, 1, umounts);
Al Virobcc5c7d2008-03-22 00:21:53 -04001889 }
1890 }
Trond Myklebust5528f9112006-06-09 09:34:17 -04001891}
1892
Trond Myklebust5528f9112006-06-09 09:34:17 -04001893/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 * Some copy_from_user() implementations do not return the exact number of
1895 * bytes remaining to copy on a fault. But copy_mount_options() requires that.
1896 * Note that this function differs from copy_from_user() in that it will oops
1897 * on bad values of `to', rather than returning a short copy.
1898 */
Ram Paib58fed82005-11-07 17:16:09 -05001899static long exact_copy_from_user(void *to, const void __user * from,
1900 unsigned long n)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901{
1902 char *t = to;
1903 const char __user *f = from;
1904 char c;
1905
1906 if (!access_ok(VERIFY_READ, from, n))
1907 return n;
1908
1909 while (n) {
1910 if (__get_user(c, f)) {
1911 memset(t, 0, n);
1912 break;
1913 }
1914 *t++ = c;
1915 f++;
1916 n--;
1917 }
1918 return n;
1919}
1920
Ram Paib58fed82005-11-07 17:16:09 -05001921int copy_mount_options(const void __user * data, unsigned long *where)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001922{
1923 int i;
1924 unsigned long page;
1925 unsigned long size;
Ram Paib58fed82005-11-07 17:16:09 -05001926
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927 *where = 0;
1928 if (!data)
1929 return 0;
1930
1931 if (!(page = __get_free_page(GFP_KERNEL)))
1932 return -ENOMEM;
1933
1934 /* We only care that *some* data at the address the user
1935 * gave us is valid. Just in case, we'll zero
1936 * the remainder of the page.
1937 */
1938 /* copy_from_user cannot cross TASK_SIZE ! */
1939 size = TASK_SIZE - (unsigned long)data;
1940 if (size > PAGE_SIZE)
1941 size = PAGE_SIZE;
1942
1943 i = size - exact_copy_from_user((void *)page, data, size);
1944 if (!i) {
Ram Paib58fed82005-11-07 17:16:09 -05001945 free_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 return -EFAULT;
1947 }
1948 if (i != PAGE_SIZE)
1949 memset((char *)page + i, 0, PAGE_SIZE - i);
1950 *where = page;
1951 return 0;
1952}
1953
Vegard Nossumeca6f532009-09-18 13:05:45 -07001954int copy_mount_string(const void __user *data, char **where)
1955{
1956 char *tmp;
1957
1958 if (!data) {
1959 *where = NULL;
1960 return 0;
1961 }
1962
1963 tmp = strndup_user(data, PAGE_SIZE);
1964 if (IS_ERR(tmp))
1965 return PTR_ERR(tmp);
1966
1967 *where = tmp;
1968 return 0;
1969}
1970
Linus Torvalds1da177e2005-04-16 15:20:36 -07001971/*
1972 * Flags is a 32-bit value that allows up to 31 non-fs dependent flags to
1973 * be given to the mount() call (ie: read-only, no-dev, no-suid etc).
1974 *
1975 * data is a (void *) that can point to any structure up to
1976 * PAGE_SIZE-1 bytes, which can contain arbitrary fs-dependent
1977 * information (or be NULL).
1978 *
1979 * Pre-0.97 versions of mount() didn't have a flags word.
1980 * When the flags word was introduced its top half was required
1981 * to have the magic value 0xC0ED, and this remained so until 2.4.0-test9.
1982 * Therefore, if this magic number is present, it carries no information
1983 * and must be discarded.
1984 */
Ram Paib58fed82005-11-07 17:16:09 -05001985long do_mount(char *dev_name, char *dir_name, char *type_page,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 unsigned long flags, void *data_page)
1987{
Al Viro2d92ab32008-08-02 00:51:11 -04001988 struct path path;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 int retval = 0;
1990 int mnt_flags = 0;
1991
1992 /* Discard magic */
1993 if ((flags & MS_MGC_MSK) == MS_MGC_VAL)
1994 flags &= ~MS_MGC_MSK;
1995
1996 /* Basic sanity checks */
1997
1998 if (!dir_name || !*dir_name || !memchr(dir_name, 0, PAGE_SIZE))
1999 return -EINVAL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000
2001 if (data_page)
2002 ((char *)data_page)[PAGE_SIZE - 1] = 0;
2003
Tetsuo Handaa27ab9f2009-10-04 21:49:49 +09002004 /* ... and get the mountpoint */
2005 retval = kern_path(dir_name, LOOKUP_FOLLOW, &path);
2006 if (retval)
2007 return retval;
2008
2009 retval = security_sb_mount(dev_name, &path,
2010 type_page, flags, data_page);
2011 if (retval)
2012 goto dput_out;
2013
Andi Kleen613cbe32009-04-19 18:40:43 +02002014 /* Default to relatime unless overriden */
2015 if (!(flags & MS_NOATIME))
2016 mnt_flags |= MNT_RELATIME;
Matthew Garrett0a1c01c2009-03-26 17:53:14 +00002017
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 /* Separate the per-mountpoint flags */
2019 if (flags & MS_NOSUID)
2020 mnt_flags |= MNT_NOSUID;
2021 if (flags & MS_NODEV)
2022 mnt_flags |= MNT_NODEV;
2023 if (flags & MS_NOEXEC)
2024 mnt_flags |= MNT_NOEXEC;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002025 if (flags & MS_NOATIME)
2026 mnt_flags |= MNT_NOATIME;
2027 if (flags & MS_NODIRATIME)
2028 mnt_flags |= MNT_NODIRATIME;
Matthew Garrettd0adde52009-03-26 17:49:56 +00002029 if (flags & MS_STRICTATIME)
2030 mnt_flags &= ~(MNT_RELATIME | MNT_NOATIME);
Dave Hansen2e4b7fc2008-02-15 14:38:00 -08002031 if (flags & MS_RDONLY)
2032 mnt_flags |= MNT_READONLY;
Christoph Hellwigfc33a7b2006-01-09 20:52:17 -08002033
Al Viro7a4dec52010-08-09 12:05:43 -04002034 flags &= ~(MS_NOSUID | MS_NOEXEC | MS_NODEV | MS_ACTIVE | MS_BORN |
Matthew Garrettd0adde52009-03-26 17:49:56 +00002035 MS_NOATIME | MS_NODIRATIME | MS_RELATIME| MS_KERNMOUNT |
2036 MS_STRICTATIME);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037
Linus Torvalds1da177e2005-04-16 15:20:36 -07002038 if (flags & MS_REMOUNT)
Al Viro2d92ab32008-08-02 00:51:11 -04002039 retval = do_remount(&path, flags & ~MS_REMOUNT, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002040 data_page);
2041 else if (flags & MS_BIND)
Al Viro2d92ab32008-08-02 00:51:11 -04002042 retval = do_loopback(&path, dev_name, flags & MS_REC);
Ram Pai9676f0c2005-11-07 17:21:20 -05002043 else if (flags & (MS_SHARED | MS_PRIVATE | MS_SLAVE | MS_UNBINDABLE))
Al Viro2d92ab32008-08-02 00:51:11 -04002044 retval = do_change_type(&path, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002045 else if (flags & MS_MOVE)
Al Viro2d92ab32008-08-02 00:51:11 -04002046 retval = do_move_mount(&path, dev_name);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 else
Al Viro2d92ab32008-08-02 00:51:11 -04002048 retval = do_new_mount(&path, type_page, flags, mnt_flags,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 dev_name, data_page);
2050dput_out:
Al Viro2d92ab32008-08-02 00:51:11 -04002051 path_put(&path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002052 return retval;
2053}
2054
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002055static struct mnt_namespace *alloc_mnt_ns(void)
2056{
2057 struct mnt_namespace *new_ns;
2058
2059 new_ns = kmalloc(sizeof(struct mnt_namespace), GFP_KERNEL);
2060 if (!new_ns)
2061 return ERR_PTR(-ENOMEM);
2062 atomic_set(&new_ns->count, 1);
2063 new_ns->root = NULL;
2064 INIT_LIST_HEAD(&new_ns->list);
2065 init_waitqueue_head(&new_ns->poll);
2066 new_ns->event = 0;
2067 return new_ns;
2068}
2069
JANAK DESAI741a2952006-02-07 12:59:00 -08002070/*
2071 * Allocate a new namespace structure and populate it with contents
2072 * copied from the namespace of the passed in task structure.
2073 */
Badari Pulavartye3222c42007-05-08 00:25:21 -07002074static struct mnt_namespace *dup_mnt_ns(struct mnt_namespace *mnt_ns,
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002075 struct fs_struct *fs)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002077 struct mnt_namespace *new_ns;
Al Viro7f2da1e2008-05-10 20:44:54 -04002078 struct vfsmount *rootmnt = NULL, *pwdmnt = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079 struct vfsmount *p, *q;
2080
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002081 new_ns = alloc_mnt_ns();
2082 if (IS_ERR(new_ns))
2083 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002084
Ram Pai390c6842005-11-07 17:17:51 -05002085 down_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 /* First pass: copy the tree topology */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002087 new_ns->root = copy_tree(mnt_ns->root, mnt_ns->root->mnt_root,
Ram Pai9676f0c2005-11-07 17:21:20 -05002088 CL_COPY_ALL | CL_EXPIRE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 if (!new_ns->root) {
Ram Pai390c6842005-11-07 17:17:51 -05002090 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 kfree(new_ns);
Julia Lawall5cc4a032008-12-01 14:34:51 -08002092 return ERR_PTR(-ENOMEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093 }
Nick Piggin99b7db72010-08-18 04:37:39 +10002094 br_write_lock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095 list_add_tail(&new_ns->list, &new_ns->root->mnt_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10002096 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097
2098 /*
2099 * Second pass: switch the tsk->fs->* elements and mark new vfsmounts
2100 * as belonging to new namespace. We have already acquired a private
2101 * fs_struct, so tsk->fs->lock is not needed.
2102 */
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002103 p = mnt_ns->root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002104 q = new_ns->root;
2105 while (p) {
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002106 q->mnt_ns = new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107 if (fs) {
Jan Blunck6ac08c32008-02-14 19:34:38 -08002108 if (p == fs->root.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109 rootmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002110 fs->root.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002111 }
Jan Blunck6ac08c32008-02-14 19:34:38 -08002112 if (p == fs->pwd.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002113 pwdmnt = p;
Jan Blunck6ac08c32008-02-14 19:34:38 -08002114 fs->pwd.mnt = mntget(q);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002115 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116 }
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002117 p = next_mnt(p, mnt_ns->root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002118 q = next_mnt(q, new_ns->root);
2119 }
Ram Pai390c6842005-11-07 17:17:51 -05002120 up_write(&namespace_sem);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122 if (rootmnt)
2123 mntput(rootmnt);
2124 if (pwdmnt)
2125 mntput(pwdmnt);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126
JANAK DESAI741a2952006-02-07 12:59:00 -08002127 return new_ns;
2128}
2129
Eric W. Biederman213dd262007-07-15 23:41:15 -07002130struct mnt_namespace *copy_mnt_ns(unsigned long flags, struct mnt_namespace *ns,
Badari Pulavartye3222c42007-05-08 00:25:21 -07002131 struct fs_struct *new_fs)
JANAK DESAI741a2952006-02-07 12:59:00 -08002132{
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002133 struct mnt_namespace *new_ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002134
Badari Pulavartye3222c42007-05-08 00:25:21 -07002135 BUG_ON(!ns);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002136 get_mnt_ns(ns);
JANAK DESAI741a2952006-02-07 12:59:00 -08002137
2138 if (!(flags & CLONE_NEWNS))
Badari Pulavartye3222c42007-05-08 00:25:21 -07002139 return ns;
JANAK DESAI741a2952006-02-07 12:59:00 -08002140
Badari Pulavartye3222c42007-05-08 00:25:21 -07002141 new_ns = dup_mnt_ns(ns, new_fs);
JANAK DESAI741a2952006-02-07 12:59:00 -08002142
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002143 put_mnt_ns(ns);
Badari Pulavartye3222c42007-05-08 00:25:21 -07002144 return new_ns;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145}
2146
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002147/**
2148 * create_mnt_ns - creates a private namespace and adds a root filesystem
2149 * @mnt: pointer to the new root filesystem mountpoint
2150 */
Linus Torvaldsa2770d82009-12-17 12:51:05 -08002151struct mnt_namespace *create_mnt_ns(struct vfsmount *mnt)
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002152{
2153 struct mnt_namespace *new_ns;
2154
2155 new_ns = alloc_mnt_ns();
2156 if (!IS_ERR(new_ns)) {
2157 mnt->mnt_ns = new_ns;
2158 new_ns->root = mnt;
2159 list_add(&new_ns->list, &new_ns->root->mnt_list);
2160 }
2161 return new_ns;
2162}
Linus Torvaldsa2770d82009-12-17 12:51:05 -08002163EXPORT_SYMBOL(create_mnt_ns);
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002164
Heiko Carstensbdc480e2009-01-14 14:14:12 +01002165SYSCALL_DEFINE5(mount, char __user *, dev_name, char __user *, dir_name,
2166 char __user *, type, unsigned long, flags, void __user *, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167{
Vegard Nossumeca6f532009-09-18 13:05:45 -07002168 int ret;
2169 char *kernel_type;
2170 char *kernel_dir;
2171 char *kernel_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002172 unsigned long data_page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002173
Vegard Nossumeca6f532009-09-18 13:05:45 -07002174 ret = copy_mount_string(type, &kernel_type);
2175 if (ret < 0)
2176 goto out_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177
Vegard Nossumeca6f532009-09-18 13:05:45 -07002178 kernel_dir = getname(dir_name);
2179 if (IS_ERR(kernel_dir)) {
2180 ret = PTR_ERR(kernel_dir);
2181 goto out_dir;
2182 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183
Vegard Nossumeca6f532009-09-18 13:05:45 -07002184 ret = copy_mount_string(dev_name, &kernel_dev);
2185 if (ret < 0)
2186 goto out_dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002187
Vegard Nossumeca6f532009-09-18 13:05:45 -07002188 ret = copy_mount_options(data, &data_page);
2189 if (ret < 0)
2190 goto out_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191
Vegard Nossumeca6f532009-09-18 13:05:45 -07002192 ret = do_mount(kernel_dev, kernel_dir, kernel_type, flags,
2193 (void *) data_page);
2194
Linus Torvalds1da177e2005-04-16 15:20:36 -07002195 free_page(data_page);
Vegard Nossumeca6f532009-09-18 13:05:45 -07002196out_data:
2197 kfree(kernel_dev);
2198out_dev:
2199 putname(kernel_dir);
2200out_dir:
2201 kfree(kernel_type);
2202out_type:
2203 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002204}
2205
2206/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207 * pivot_root Semantics:
2208 * Moves the root file system of the current process to the directory put_old,
2209 * makes new_root as the new root file system of the current process, and sets
2210 * root/cwd of all processes which had them on the current root to new_root.
2211 *
2212 * Restrictions:
2213 * The new_root and put_old must be directories, and must not be on the
2214 * same file system as the current process root. The put_old must be
2215 * underneath new_root, i.e. adding a non-zero number of /.. to the string
2216 * pointed to by put_old must yield the same directory as new_root. No other
2217 * file system may be mounted on put_old. After all, new_root is a mountpoint.
2218 *
Neil Brown4a0d11f2006-01-08 01:03:18 -08002219 * Also, the current root cannot be on the 'rootfs' (initial ramfs) filesystem.
2220 * See Documentation/filesystems/ramfs-rootfs-initramfs.txt for alternatives
2221 * in this situation.
2222 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002223 * Notes:
2224 * - we don't move root/cwd if they are not at the root (reason: if something
2225 * cared enough to change them, it's probably wrong to force them elsewhere)
2226 * - it's okay to pick a root that isn't the root of a file system, e.g.
2227 * /nfs/my_root where /nfs is the mount point. It must be a mountpoint,
2228 * though, so you may need to say mount --bind /nfs/my_root /nfs/my_root
2229 * first.
2230 */
Heiko Carstens3480b252009-01-14 14:14:16 +01002231SYSCALL_DEFINE2(pivot_root, const char __user *, new_root,
2232 const char __user *, put_old)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002233{
2234 struct vfsmount *tmp;
Al Viro2d8f3032008-07-22 09:59:21 -04002235 struct path new, old, parent_path, root_parent, root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 int error;
2237
2238 if (!capable(CAP_SYS_ADMIN))
2239 return -EPERM;
2240
Al Viro2d8f3032008-07-22 09:59:21 -04002241 error = user_path_dir(new_root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 if (error)
2243 goto out0;
2244 error = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04002245 if (!check_mnt(new.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002246 goto out1;
2247
Al Viro2d8f3032008-07-22 09:59:21 -04002248 error = user_path_dir(put_old, &old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002249 if (error)
2250 goto out1;
2251
Al Viro2d8f3032008-07-22 09:59:21 -04002252 error = security_sb_pivotroot(&old, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 if (error) {
Al Viro2d8f3032008-07-22 09:59:21 -04002254 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002255 goto out1;
2256 }
2257
Miklos Szeredif7ad3c62010-08-10 11:41:36 +02002258 get_fs_root(current->fs, &root);
Ram Pai390c6842005-11-07 17:17:51 -05002259 down_write(&namespace_sem);
Al Viro2d8f3032008-07-22 09:59:21 -04002260 mutex_lock(&old.dentry->d_inode->i_mutex);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002261 error = -EINVAL;
Al Viro2d8f3032008-07-22 09:59:21 -04002262 if (IS_MNT_SHARED(old.mnt) ||
2263 IS_MNT_SHARED(new.mnt->mnt_parent) ||
Al Viro8c3ee422008-03-22 18:00:39 -04002264 IS_MNT_SHARED(root.mnt->mnt_parent))
Ram Pai21444402005-11-07 17:20:03 -05002265 goto out2;
Al Viro8c3ee422008-03-22 18:00:39 -04002266 if (!check_mnt(root.mnt))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267 goto out2;
2268 error = -ENOENT;
Al Virod83c49f2010-04-30 17:17:09 -04002269 if (cant_mount(old.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 goto out2;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002271 if (d_unlinked(new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002272 goto out2;
Alexey Dobriyanf3da3922009-05-04 03:32:03 +04002273 if (d_unlinked(old.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002274 goto out2;
2275 error = -EBUSY;
Al Viro2d8f3032008-07-22 09:59:21 -04002276 if (new.mnt == root.mnt ||
2277 old.mnt == root.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002278 goto out2; /* loop, on the same file system */
2279 error = -EINVAL;
Al Viro8c3ee422008-03-22 18:00:39 -04002280 if (root.mnt->mnt_root != root.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 goto out2; /* not a mountpoint */
Al Viro8c3ee422008-03-22 18:00:39 -04002282 if (root.mnt->mnt_parent == root.mnt)
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07002283 goto out2; /* not attached */
Al Viro2d8f3032008-07-22 09:59:21 -04002284 if (new.mnt->mnt_root != new.dentry)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 goto out2; /* not a mountpoint */
Al Viro2d8f3032008-07-22 09:59:21 -04002286 if (new.mnt->mnt_parent == new.mnt)
Miklos Szeredi0bb6fcc2005-09-06 15:19:36 -07002287 goto out2; /* not attached */
Jan Blunck4ac91372008-02-14 19:34:32 -08002288 /* make sure we can reach put_old from new_root */
Al Viro2d8f3032008-07-22 09:59:21 -04002289 tmp = old.mnt;
Nick Piggin99b7db72010-08-18 04:37:39 +10002290 br_write_lock(vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002291 if (tmp != new.mnt) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292 for (;;) {
2293 if (tmp->mnt_parent == tmp)
2294 goto out3; /* already mounted on put_old */
Al Viro2d8f3032008-07-22 09:59:21 -04002295 if (tmp->mnt_parent == new.mnt)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296 break;
2297 tmp = tmp->mnt_parent;
2298 }
Al Viro2d8f3032008-07-22 09:59:21 -04002299 if (!is_subdir(tmp->mnt_mountpoint, new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 goto out3;
Al Viro2d8f3032008-07-22 09:59:21 -04002301 } else if (!is_subdir(old.dentry, new.dentry))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302 goto out3;
Al Viro2d8f3032008-07-22 09:59:21 -04002303 detach_mnt(new.mnt, &parent_path);
Al Viro8c3ee422008-03-22 18:00:39 -04002304 detach_mnt(root.mnt, &root_parent);
Jan Blunck4ac91372008-02-14 19:34:32 -08002305 /* mount old root on put_old */
Al Viro2d8f3032008-07-22 09:59:21 -04002306 attach_mnt(root.mnt, &old);
Jan Blunck4ac91372008-02-14 19:34:32 -08002307 /* mount new_root on / */
Al Viro2d8f3032008-07-22 09:59:21 -04002308 attach_mnt(new.mnt, &root_parent);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002309 touch_mnt_namespace(current->nsproxy->mnt_ns);
Nick Piggin99b7db72010-08-18 04:37:39 +10002310 br_write_unlock(vfsmount_lock);
Al Viro2d8f3032008-07-22 09:59:21 -04002311 chroot_fs_refs(&root, &new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312 error = 0;
Al Viro1a390682008-03-21 20:48:19 -04002313 path_put(&root_parent);
2314 path_put(&parent_path);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315out2:
Al Viro2d8f3032008-07-22 09:59:21 -04002316 mutex_unlock(&old.dentry->d_inode->i_mutex);
Ram Pai390c6842005-11-07 17:17:51 -05002317 up_write(&namespace_sem);
Al Viro8c3ee422008-03-22 18:00:39 -04002318 path_put(&root);
Al Viro2d8f3032008-07-22 09:59:21 -04002319 path_put(&old);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002320out1:
Al Viro2d8f3032008-07-22 09:59:21 -04002321 path_put(&new);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002322out0:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 return error;
2324out3:
Nick Piggin99b7db72010-08-18 04:37:39 +10002325 br_write_unlock(vfsmount_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 goto out2;
2327}
2328
2329static void __init init_mount_tree(void)
2330{
2331 struct vfsmount *mnt;
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002332 struct mnt_namespace *ns;
Jan Blunckac748a02008-02-14 19:34:39 -08002333 struct path root;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002334
2335 mnt = do_kern_mount("rootfs", 0, "rootfs", NULL);
2336 if (IS_ERR(mnt))
2337 panic("Can't create rootfs");
Trond Myklebust3b22edc2009-06-23 17:29:49 -04002338 ns = create_mnt_ns(mnt);
2339 if (IS_ERR(ns))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002340 panic("Can't allocate initial namespace");
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002342 init_task.nsproxy->mnt_ns = ns;
2343 get_mnt_ns(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344
Jan Blunckac748a02008-02-14 19:34:39 -08002345 root.mnt = ns->root;
2346 root.dentry = ns->root->mnt_root;
2347
2348 set_fs_pwd(current->fs, &root);
2349 set_fs_root(current->fs, &root);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350}
2351
Denis Cheng74bf17c2007-10-16 23:26:30 -07002352void __init mnt_init(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353{
Eric Dumazet13f14b42008-02-06 01:37:57 -08002354 unsigned u;
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002355 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356
Ram Pai390c6842005-11-07 17:17:51 -05002357 init_rwsem(&namespace_sem);
2358
Linus Torvalds1da177e2005-04-16 15:20:36 -07002359 mnt_cache = kmem_cache_create("mnt_cache", sizeof(struct vfsmount),
Paul Mundt20c2df82007-07-20 10:11:58 +09002360 0, SLAB_HWCACHE_ALIGN | SLAB_PANIC, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361
Ram Paib58fed82005-11-07 17:16:09 -05002362 mount_hashtable = (struct list_head *)__get_free_page(GFP_ATOMIC);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002363
2364 if (!mount_hashtable)
2365 panic("Failed to allocate mount hash table\n");
2366
Eric Dumazet13f14b42008-02-06 01:37:57 -08002367 printk("Mount-cache hash table entries: %lu\n", HASH_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368
Eric Dumazet13f14b42008-02-06 01:37:57 -08002369 for (u = 0; u < HASH_SIZE; u++)
2370 INIT_LIST_HEAD(&mount_hashtable[u]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371
Nick Piggin99b7db72010-08-18 04:37:39 +10002372 br_lock_init(vfsmount_lock);
2373
Randy Dunlap15a67dd2006-09-29 01:58:57 -07002374 err = sysfs_init();
2375 if (err)
2376 printk(KERN_WARNING "%s: sysfs_init error: %d\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002377 __func__, err);
Greg Kroah-Hartman00d26662007-10-29 14:17:23 -06002378 fs_kobj = kobject_create_and_add("fs", NULL);
2379 if (!fs_kobj)
Harvey Harrison8e24eea2008-04-30 00:55:09 -07002380 printk(KERN_WARNING "%s: kobj create error\n", __func__);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 init_rootfs();
2382 init_mount_tree();
2383}
2384
Trond Myklebust616511d2009-06-22 15:09:13 -04002385void put_mnt_ns(struct mnt_namespace *ns)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386{
Ram Pai70fbcdf2005-11-07 17:17:04 -05002387 LIST_HEAD(umount_list);
Trond Myklebust616511d2009-06-22 15:09:13 -04002388
Al Virod498b252010-02-05 02:21:06 -05002389 if (!atomic_dec_and_test(&ns->count))
Trond Myklebust616511d2009-06-22 15:09:13 -04002390 return;
Ram Pai390c6842005-11-07 17:17:51 -05002391 down_write(&namespace_sem);
Nick Piggin99b7db72010-08-18 04:37:39 +10002392 br_write_lock(vfsmount_lock);
Al Virod498b252010-02-05 02:21:06 -05002393 umount_tree(ns->root, 0, &umount_list);
Nick Piggin99b7db72010-08-18 04:37:39 +10002394 br_write_unlock(vfsmount_lock);
Ram Pai390c6842005-11-07 17:17:51 -05002395 up_write(&namespace_sem);
Ram Pai70fbcdf2005-11-07 17:17:04 -05002396 release_mounts(&umount_list);
Kirill Korotaev6b3286e2006-12-08 02:37:56 -08002397 kfree(ns);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398}
Trond Myklebustcf8d2c12009-06-22 15:09:13 -04002399EXPORT_SYMBOL(put_mnt_ns);