cgroup: fix and update documentation
[linux-2.6] / kernel / capability.c
1 /*
2  * linux/kernel/capability.c
3  *
4  * Copyright (C) 1997  Andrew Main <zefram@fysh.org>
5  *
6  * Integrated into 2.1.97+,  Andrew G. Morgan <morgan@kernel.org>
7  * 30 May 2002: Cleanup, Robert M. Love <rml@tech9.net>
8  */
9
10 #include <linux/capability.h>
11 #include <linux/mm.h>
12 #include <linux/module.h>
13 #include <linux/security.h>
14 #include <linux/syscalls.h>
15 #include <linux/pid_namespace.h>
16 #include <asm/uaccess.h>
17
18 /*
19  * This lock protects task->cap_* for all tasks including current.
20  * Locking rule: acquire this prior to tasklist_lock.
21  */
22 static DEFINE_SPINLOCK(task_capability_lock);
23
24 /*
25  * Leveraged for setting/resetting capabilities
26  */
27
28 const kernel_cap_t __cap_empty_set = CAP_EMPTY_SET;
29 const kernel_cap_t __cap_full_set = CAP_FULL_SET;
30 const kernel_cap_t __cap_init_eff_set = CAP_INIT_EFF_SET;
31
32 EXPORT_SYMBOL(__cap_empty_set);
33 EXPORT_SYMBOL(__cap_full_set);
34 EXPORT_SYMBOL(__cap_init_eff_set);
35
36 /*
37  * More recent versions of libcap are available from:
38  *
39  *   http://www.kernel.org/pub/linux/libs/security/linux-privs/
40  */
41
42 static void warn_legacy_capability_use(void)
43 {
44         static int warned;
45         if (!warned) {
46                 char name[sizeof(current->comm)];
47
48                 printk(KERN_INFO "warning: `%s' uses 32-bit capabilities"
49                        " (legacy support in use)\n",
50                        get_task_comm(name, current));
51                 warned = 1;
52         }
53 }
54
55 /*
56  * For sys_getproccap() and sys_setproccap(), any of the three
57  * capability set pointers may be NULL -- indicating that that set is
58  * uninteresting and/or not to be changed.
59  */
60
61 /**
62  * sys_capget - get the capabilities of a given process.
63  * @header: pointer to struct that contains capability version and
64  *      target pid data
65  * @dataptr: pointer to struct that contains the effective, permitted,
66  *      and inheritable capabilities that are returned
67  *
68  * Returns 0 on success and < 0 on error.
69  */
70 asmlinkage long sys_capget(cap_user_header_t header, cap_user_data_t dataptr)
71 {
72         int ret = 0;
73         pid_t pid;
74         __u32 version;
75         struct task_struct *target;
76         unsigned tocopy;
77         kernel_cap_t pE, pI, pP;
78
79         if (get_user(version, &header->version))
80                 return -EFAULT;
81
82         switch (version) {
83         case _LINUX_CAPABILITY_VERSION_1:
84                 warn_legacy_capability_use();
85                 tocopy = _LINUX_CAPABILITY_U32S_1;
86                 break;
87         case _LINUX_CAPABILITY_VERSION_2:
88                 tocopy = _LINUX_CAPABILITY_U32S_2;
89                 break;
90         default:
91                 if (put_user(_LINUX_CAPABILITY_VERSION, &header->version))
92                         return -EFAULT;
93                 return -EINVAL;
94         }
95
96         if (get_user(pid, &header->pid))
97                 return -EFAULT;
98
99         if (pid < 0)
100                 return -EINVAL;
101
102         spin_lock(&task_capability_lock);
103         read_lock(&tasklist_lock);
104
105         if (pid && pid != task_pid_vnr(current)) {
106                 target = find_task_by_vpid(pid);
107                 if (!target) {
108                         ret = -ESRCH;
109                         goto out;
110                 }
111         } else
112                 target = current;
113
114         ret = security_capget(target, &pE, &pI, &pP);
115
116 out:
117         read_unlock(&tasklist_lock);
118         spin_unlock(&task_capability_lock);
119
120         if (!ret) {
121                 struct __user_cap_data_struct kdata[_LINUX_CAPABILITY_U32S];
122                 unsigned i;
123
124                 for (i = 0; i < tocopy; i++) {
125                         kdata[i].effective = pE.cap[i];
126                         kdata[i].permitted = pP.cap[i];
127                         kdata[i].inheritable = pI.cap[i];
128                 }
129
130                 /*
131                  * Note, in the case, tocopy < _LINUX_CAPABILITY_U32S,
132                  * we silently drop the upper capabilities here. This
133                  * has the effect of making older libcap
134                  * implementations implicitly drop upper capability
135                  * bits when they perform a: capget/modify/capset
136                  * sequence.
137                  *
138                  * This behavior is considered fail-safe
139                  * behavior. Upgrading the application to a newer
140                  * version of libcap will enable access to the newer
141                  * capabilities.
142                  *
143                  * An alternative would be to return an error here
144                  * (-ERANGE), but that causes legacy applications to
145                  * unexpectidly fail; the capget/modify/capset aborts
146                  * before modification is attempted and the application
147                  * fails.
148                  */
149
150                 if (copy_to_user(dataptr, kdata, tocopy
151                                  * sizeof(struct __user_cap_data_struct))) {
152                         return -EFAULT;
153                 }
154         }
155
156         return ret;
157 }
158
159 /*
160  * cap_set_pg - set capabilities for all processes in a given process
161  * group.  We call this holding task_capability_lock and tasklist_lock.
162  */
163 static inline int cap_set_pg(int pgrp_nr, kernel_cap_t *effective,
164                               kernel_cap_t *inheritable,
165                               kernel_cap_t *permitted)
166 {
167         struct task_struct *g, *target;
168         int ret = -EPERM;
169         int found = 0;
170         struct pid *pgrp;
171
172         pgrp = find_vpid(pgrp_nr);
173         do_each_pid_task(pgrp, PIDTYPE_PGID, g) {
174                 target = g;
175                 while_each_thread(g, target) {
176                         if (!security_capset_check(target, effective,
177                                                         inheritable,
178                                                         permitted)) {
179                                 security_capset_set(target, effective,
180                                                         inheritable,
181                                                         permitted);
182                                 ret = 0;
183                         }
184                         found = 1;
185                 }
186         } while_each_pid_task(pgrp, PIDTYPE_PGID, g);
187
188         if (!found)
189                 ret = 0;
190         return ret;
191 }
192
193 /*
194  * cap_set_all - set capabilities for all processes other than init
195  * and self.  We call this holding task_capability_lock and tasklist_lock.
196  */
197 static inline int cap_set_all(kernel_cap_t *effective,
198                                kernel_cap_t *inheritable,
199                                kernel_cap_t *permitted)
200 {
201      struct task_struct *g, *target;
202      int ret = -EPERM;
203      int found = 0;
204
205      do_each_thread(g, target) {
206              if (target == current || is_container_init(target->group_leader))
207                      continue;
208              found = 1;
209              if (security_capset_check(target, effective, inheritable,
210                                                 permitted))
211                      continue;
212              ret = 0;
213              security_capset_set(target, effective, inheritable, permitted);
214      } while_each_thread(g, target);
215
216      if (!found)
217              ret = 0;
218      return ret;
219 }
220
221 /**
222  * sys_capset - set capabilities for a process or a group of processes
223  * @header: pointer to struct that contains capability version and
224  *      target pid data
225  * @data: pointer to struct that contains the effective, permitted,
226  *      and inheritable capabilities
227  *
228  * Set capabilities for a given process, all processes, or all
229  * processes in a given process group.
230  *
231  * The restrictions on setting capabilities are specified as:
232  *
233  * [pid is for the 'target' task.  'current' is the calling task.]
234  *
235  * I: any raised capabilities must be a subset of the (old current) permitted
236  * P: any raised capabilities must be a subset of the (old current) permitted
237  * E: must be set to a subset of (new target) permitted
238  *
239  * Returns 0 on success and < 0 on error.
240  */
241 asmlinkage long sys_capset(cap_user_header_t header, const cap_user_data_t data)
242 {
243         struct __user_cap_data_struct kdata[_LINUX_CAPABILITY_U32S];
244         unsigned i, tocopy;
245         kernel_cap_t inheritable, permitted, effective;
246         __u32 version;
247         struct task_struct *target;
248         int ret;
249         pid_t pid;
250
251         if (get_user(version, &header->version))
252                 return -EFAULT;
253
254         switch (version) {
255         case _LINUX_CAPABILITY_VERSION_1:
256                 warn_legacy_capability_use();
257                 tocopy = _LINUX_CAPABILITY_U32S_1;
258                 break;
259         case _LINUX_CAPABILITY_VERSION_2:
260                 tocopy = _LINUX_CAPABILITY_U32S_2;
261                 break;
262         default:
263                 if (put_user(_LINUX_CAPABILITY_VERSION, &header->version))
264                         return -EFAULT;
265                 return -EINVAL;
266         }
267
268         if (get_user(pid, &header->pid))
269                 return -EFAULT;
270
271         if (pid && pid != task_pid_vnr(current) && !capable(CAP_SETPCAP))
272                 return -EPERM;
273
274         if (copy_from_user(&kdata, data, tocopy
275                            * sizeof(struct __user_cap_data_struct))) {
276                 return -EFAULT;
277         }
278
279         for (i = 0; i < tocopy; i++) {
280                 effective.cap[i] = kdata[i].effective;
281                 permitted.cap[i] = kdata[i].permitted;
282                 inheritable.cap[i] = kdata[i].inheritable;
283         }
284         while (i < _LINUX_CAPABILITY_U32S) {
285                 effective.cap[i] = 0;
286                 permitted.cap[i] = 0;
287                 inheritable.cap[i] = 0;
288                 i++;
289         }
290
291         spin_lock(&task_capability_lock);
292         read_lock(&tasklist_lock);
293
294         if (pid > 0 && pid != task_pid_vnr(current)) {
295                 target = find_task_by_vpid(pid);
296                 if (!target) {
297                         ret = -ESRCH;
298                         goto out;
299                 }
300         } else
301                 target = current;
302
303         ret = 0;
304
305         /* having verified that the proposed changes are legal,
306            we now put them into effect. */
307         if (pid < 0) {
308                 if (pid == -1)  /* all procs other than current and init */
309                         ret = cap_set_all(&effective, &inheritable, &permitted);
310
311                 else            /* all procs in process group */
312                         ret = cap_set_pg(-pid, &effective, &inheritable,
313                                          &permitted);
314         } else {
315                 ret = security_capset_check(target, &effective, &inheritable,
316                                             &permitted);
317                 if (!ret)
318                         security_capset_set(target, &effective, &inheritable,
319                                             &permitted);
320         }
321
322 out:
323         read_unlock(&tasklist_lock);
324         spin_unlock(&task_capability_lock);
325
326         return ret;
327 }
328
329 int __capable(struct task_struct *t, int cap)
330 {
331         if (security_capable(t, cap) == 0) {
332                 t->flags |= PF_SUPERPRIV;
333                 return 1;
334         }
335         return 0;
336 }
337
338 int capable(int cap)
339 {
340         return __capable(current, cap);
341 }
342 EXPORT_SYMBOL(capable);