]>
Commit | Line | Data |
---|---|---|
1 | #include <linux/mount.h> | |
2 | #include <linux/file.h> | |
3 | #include <linux/fs.h> | |
4 | #include <linux/proc_ns.h> | |
5 | #include <linux/magic.h> | |
6 | #include <linux/ktime.h> | |
7 | #include <linux/seq_file.h> | |
8 | #include <linux/user_namespace.h> | |
9 | #include <linux/nsfs.h> | |
10 | #include <linux/uaccess.h> | |
11 | ||
12 | static struct vfsmount *nsfs_mnt; | |
13 | ||
14 | static long ns_ioctl(struct file *filp, unsigned int ioctl, | |
15 | unsigned long arg); | |
16 | static const struct file_operations ns_file_operations = { | |
17 | .llseek = no_llseek, | |
18 | .unlocked_ioctl = ns_ioctl, | |
19 | }; | |
20 | ||
21 | static char *ns_dname(struct dentry *dentry, char *buffer, int buflen) | |
22 | { | |
23 | struct inode *inode = d_inode(dentry); | |
24 | const struct proc_ns_operations *ns_ops = dentry->d_fsdata; | |
25 | ||
26 | return dynamic_dname(dentry, buffer, buflen, "%s:[%lu]", | |
27 | ns_ops->name, inode->i_ino); | |
28 | } | |
29 | ||
30 | static void ns_prune_dentry(struct dentry *dentry) | |
31 | { | |
32 | struct inode *inode = d_inode(dentry); | |
33 | if (inode) { | |
34 | struct ns_common *ns = inode->i_private; | |
35 | atomic_long_set(&ns->stashed, 0); | |
36 | } | |
37 | } | |
38 | ||
39 | const struct dentry_operations ns_dentry_operations = | |
40 | { | |
41 | .d_prune = ns_prune_dentry, | |
42 | .d_delete = always_delete_dentry, | |
43 | .d_dname = ns_dname, | |
44 | }; | |
45 | ||
46 | static void nsfs_evict(struct inode *inode) | |
47 | { | |
48 | struct ns_common *ns = inode->i_private; | |
49 | clear_inode(inode); | |
50 | ns->ops->put(ns); | |
51 | } | |
52 | ||
53 | static void *__ns_get_path(struct path *path, struct ns_common *ns) | |
54 | { | |
55 | struct vfsmount *mnt = nsfs_mnt; | |
56 | struct dentry *dentry; | |
57 | struct inode *inode; | |
58 | unsigned long d; | |
59 | ||
60 | rcu_read_lock(); | |
61 | d = atomic_long_read(&ns->stashed); | |
62 | if (!d) | |
63 | goto slow; | |
64 | dentry = (struct dentry *)d; | |
65 | if (!lockref_get_not_dead(&dentry->d_lockref)) | |
66 | goto slow; | |
67 | rcu_read_unlock(); | |
68 | ns->ops->put(ns); | |
69 | got_it: | |
70 | path->mnt = mntget(mnt); | |
71 | path->dentry = dentry; | |
72 | return NULL; | |
73 | slow: | |
74 | rcu_read_unlock(); | |
75 | inode = new_inode_pseudo(mnt->mnt_sb); | |
76 | if (!inode) { | |
77 | ns->ops->put(ns); | |
78 | return ERR_PTR(-ENOMEM); | |
79 | } | |
80 | inode->i_ino = ns->inum; | |
81 | inode->i_mtime = inode->i_atime = inode->i_ctime = current_time(inode); | |
82 | inode->i_flags |= S_IMMUTABLE; | |
83 | inode->i_mode = S_IFREG | S_IRUGO; | |
84 | inode->i_fop = &ns_file_operations; | |
85 | inode->i_private = ns; | |
86 | ||
87 | dentry = d_alloc_pseudo(mnt->mnt_sb, &empty_name); | |
88 | if (!dentry) { | |
89 | iput(inode); | |
90 | return ERR_PTR(-ENOMEM); | |
91 | } | |
92 | d_instantiate(dentry, inode); | |
93 | dentry->d_flags |= DCACHE_RCUACCESS; | |
94 | dentry->d_fsdata = (void *)ns->ops; | |
95 | d = atomic_long_cmpxchg(&ns->stashed, 0, (unsigned long)dentry); | |
96 | if (d) { | |
97 | d_delete(dentry); /* make sure ->d_prune() does nothing */ | |
98 | dput(dentry); | |
99 | cpu_relax(); | |
100 | return ERR_PTR(-EAGAIN); | |
101 | } | |
102 | goto got_it; | |
103 | } | |
104 | ||
105 | void *ns_get_path(struct path *path, struct task_struct *task, | |
106 | const struct proc_ns_operations *ns_ops) | |
107 | { | |
108 | struct ns_common *ns; | |
109 | void *ret; | |
110 | ||
111 | again: | |
112 | ns = ns_ops->get(task); | |
113 | if (!ns) | |
114 | return ERR_PTR(-ENOENT); | |
115 | ||
116 | ret = __ns_get_path(path, ns); | |
117 | if (IS_ERR(ret) && PTR_ERR(ret) == -EAGAIN) | |
118 | goto again; | |
119 | return ret; | |
120 | } | |
121 | ||
122 | int open_related_ns(struct ns_common *ns, | |
123 | struct ns_common *(*get_ns)(struct ns_common *ns)) | |
124 | { | |
125 | struct path path = {}; | |
126 | struct file *f; | |
127 | void *err; | |
128 | int fd; | |
129 | ||
130 | fd = get_unused_fd_flags(O_CLOEXEC); | |
131 | if (fd < 0) | |
132 | return fd; | |
133 | ||
134 | while (1) { | |
135 | struct ns_common *relative; | |
136 | ||
137 | relative = get_ns(ns); | |
138 | if (IS_ERR(relative)) { | |
139 | put_unused_fd(fd); | |
140 | return PTR_ERR(relative); | |
141 | } | |
142 | ||
143 | err = __ns_get_path(&path, relative); | |
144 | if (IS_ERR(err) && PTR_ERR(err) == -EAGAIN) | |
145 | continue; | |
146 | break; | |
147 | } | |
148 | if (IS_ERR(err)) { | |
149 | put_unused_fd(fd); | |
150 | return PTR_ERR(err); | |
151 | } | |
152 | ||
153 | f = dentry_open(&path, O_RDONLY, current_cred()); | |
154 | path_put(&path); | |
155 | if (IS_ERR(f)) { | |
156 | put_unused_fd(fd); | |
157 | fd = PTR_ERR(f); | |
158 | } else | |
159 | fd_install(fd, f); | |
160 | ||
161 | return fd; | |
162 | } | |
163 | ||
164 | static long ns_ioctl(struct file *filp, unsigned int ioctl, | |
165 | unsigned long arg) | |
166 | { | |
167 | struct user_namespace *user_ns; | |
168 | struct ns_common *ns = get_proc_ns(file_inode(filp)); | |
169 | uid_t __user *argp; | |
170 | uid_t uid; | |
171 | ||
172 | switch (ioctl) { | |
173 | case NS_GET_USERNS: | |
174 | return open_related_ns(ns, ns_get_owner); | |
175 | case NS_GET_PARENT: | |
176 | if (!ns->ops->get_parent) | |
177 | return -EINVAL; | |
178 | return open_related_ns(ns, ns->ops->get_parent); | |
179 | case NS_GET_NSTYPE: | |
180 | return ns->ops->type; | |
181 | case NS_GET_OWNER_UID: | |
182 | if (ns->ops->type != CLONE_NEWUSER) | |
183 | return -EINVAL; | |
184 | user_ns = container_of(ns, struct user_namespace, ns); | |
185 | argp = (uid_t __user *) arg; | |
186 | uid = from_kuid_munged(current_user_ns(), user_ns->owner); | |
187 | return put_user(uid, argp); | |
188 | default: | |
189 | return -ENOTTY; | |
190 | } | |
191 | } | |
192 | ||
193 | int ns_get_name(char *buf, size_t size, struct task_struct *task, | |
194 | const struct proc_ns_operations *ns_ops) | |
195 | { | |
196 | struct ns_common *ns; | |
197 | int res = -ENOENT; | |
198 | const char *name; | |
199 | ns = ns_ops->get(task); | |
200 | if (ns) { | |
201 | name = ns_ops->real_ns_name ? : ns_ops->name; | |
202 | res = snprintf(buf, size, "%s:[%u]", name, ns->inum); | |
203 | ns_ops->put(ns); | |
204 | } | |
205 | return res; | |
206 | } | |
207 | ||
208 | struct file *proc_ns_fget(int fd) | |
209 | { | |
210 | struct file *file; | |
211 | ||
212 | file = fget(fd); | |
213 | if (!file) | |
214 | return ERR_PTR(-EBADF); | |
215 | ||
216 | if (file->f_op != &ns_file_operations) | |
217 | goto out_invalid; | |
218 | ||
219 | return file; | |
220 | ||
221 | out_invalid: | |
222 | fput(file); | |
223 | return ERR_PTR(-EINVAL); | |
224 | } | |
225 | ||
226 | static int nsfs_show_path(struct seq_file *seq, struct dentry *dentry) | |
227 | { | |
228 | struct inode *inode = d_inode(dentry); | |
229 | const struct proc_ns_operations *ns_ops = dentry->d_fsdata; | |
230 | ||
231 | seq_printf(seq, "%s:[%lu]", ns_ops->name, inode->i_ino); | |
232 | return 0; | |
233 | } | |
234 | ||
235 | static const struct super_operations nsfs_ops = { | |
236 | .statfs = simple_statfs, | |
237 | .evict_inode = nsfs_evict, | |
238 | .show_path = nsfs_show_path, | |
239 | }; | |
240 | static struct dentry *nsfs_mount(struct file_system_type *fs_type, | |
241 | int flags, const char *dev_name, void *data) | |
242 | { | |
243 | return mount_pseudo(fs_type, "nsfs:", &nsfs_ops, | |
244 | &ns_dentry_operations, NSFS_MAGIC); | |
245 | } | |
246 | static struct file_system_type nsfs = { | |
247 | .name = "nsfs", | |
248 | .mount = nsfs_mount, | |
249 | .kill_sb = kill_anon_super, | |
250 | }; | |
251 | ||
252 | void __init nsfs_init(void) | |
253 | { | |
254 | nsfs_mnt = kern_mount(&nsfs); | |
255 | if (IS_ERR(nsfs_mnt)) | |
256 | panic("can't set nsfs up\n"); | |
257 | nsfs_mnt->mnt_sb->s_flags &= ~MS_NOUSER; | |
258 | } |