capabilities: remain source compatible with 32-bit raw legacy capability support.
[linux-2.6.git] / 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  * Version 2 capabilities worked fine, but the linux/capability.h file
57  * that accompanied their introduction encouraged their use without
58  * the necessary user-space source code changes. As such, we have
59  * created a version 3 with equivalent functionality to version 2, but
60  * with a header change to protect legacy source code from using
61  * version 2 when it wanted to use version 1. If your system has code
62  * that trips the following warning, it is using version 2 specific
63  * capabilities and may be doing so insecurely.
64  *
65  * The remedy is to either upgrade your version of libcap (to 2.10+,
66  * if the application is linked against it), or recompile your
67  * application with modern kernel headers and this warning will go
68  * away.
69  */
70
71 static void warn_deprecated_v2(void)
72 {
73         static int warned;
74
75         if (!warned) {
76                 char name[sizeof(current->comm)];
77
78                 printk(KERN_INFO "warning: `%s' uses deprecated v2"
79                        " capabilities in a way that may be insecure.\n",
80                        get_task_comm(name, current));
81                 warned = 1;
82         }
83 }
84
85 /*
86  * Version check. Return the number of u32s in each capability flag
87  * array, or a negative value on error.
88  */
89 static int cap_validate_magic(cap_user_header_t header, unsigned *tocopy)
90 {
91         __u32 version;
92
93         if (get_user(version, &header->version))
94                 return -EFAULT;
95
96         switch (version) {
97         case _LINUX_CAPABILITY_VERSION_1:
98                 warn_legacy_capability_use();
99                 *tocopy = _LINUX_CAPABILITY_U32S_1;
100                 break;
101         case _LINUX_CAPABILITY_VERSION_2:
102                 warn_deprecated_v2();
103                 /*
104                  * fall through - v3 is otherwise equivalent to v2.
105                  */
106         case _LINUX_CAPABILITY_VERSION_3:
107                 *tocopy = _LINUX_CAPABILITY_U32S_3;
108                 break;
109         default:
110                 if (put_user((u32)_KERNEL_CAPABILITY_VERSION, &header->version))
111                         return -EFAULT;
112                 return -EINVAL;
113         }
114
115         return 0;
116 }
117
118 /*
119  * For sys_getproccap() and sys_setproccap(), any of the three
120  * capability set pointers may be NULL -- indicating that that set is
121  * uninteresting and/or not to be changed.
122  */
123
124 /**
125  * sys_capget - get the capabilities of a given process.
126  * @header: pointer to struct that contains capability version and
127  *      target pid data
128  * @dataptr: pointer to struct that contains the effective, permitted,
129  *      and inheritable capabilities that are returned
130  *
131  * Returns 0 on success and < 0 on error.
132  */
133 asmlinkage long sys_capget(cap_user_header_t header, cap_user_data_t dataptr)
134 {
135         int ret = 0;
136         pid_t pid;
137         struct task_struct *target;
138         unsigned tocopy;
139         kernel_cap_t pE, pI, pP;
140
141         ret = cap_validate_magic(header, &tocopy);
142         if (ret != 0)
143                 return ret;
144
145         if (get_user(pid, &header->pid))
146                 return -EFAULT;
147
148         if (pid < 0)
149                 return -EINVAL;
150
151         spin_lock(&task_capability_lock);
152         read_lock(&tasklist_lock);
153
154         if (pid && pid != task_pid_vnr(current)) {
155                 target = find_task_by_vpid(pid);
156                 if (!target) {
157                         ret = -ESRCH;
158                         goto out;
159                 }
160         } else
161                 target = current;
162
163         ret = security_capget(target, &pE, &pI, &pP);
164
165 out:
166         read_unlock(&tasklist_lock);
167         spin_unlock(&task_capability_lock);
168
169         if (!ret) {
170                 struct __user_cap_data_struct kdata[_KERNEL_CAPABILITY_U32S];
171                 unsigned i;
172
173                 for (i = 0; i < tocopy; i++) {
174                         kdata[i].effective = pE.cap[i];
175                         kdata[i].permitted = pP.cap[i];
176                         kdata[i].inheritable = pI.cap[i];
177                 }
178
179                 /*
180                  * Note, in the case, tocopy < _KERNEL_CAPABILITY_U32S,
181                  * we silently drop the upper capabilities here. This
182                  * has the effect of making older libcap
183                  * implementations implicitly drop upper capability
184                  * bits when they perform a: capget/modify/capset
185                  * sequence.
186                  *
187                  * This behavior is considered fail-safe
188                  * behavior. Upgrading the application to a newer
189                  * version of libcap will enable access to the newer
190                  * capabilities.
191                  *
192                  * An alternative would be to return an error here
193                  * (-ERANGE), but that causes legacy applications to
194                  * unexpectidly fail; the capget/modify/capset aborts
195                  * before modification is attempted and the application
196                  * fails.
197                  */
198
199                 if (copy_to_user(dataptr, kdata, tocopy
200                                  * sizeof(struct __user_cap_data_struct))) {
201                         return -EFAULT;
202                 }
203         }
204
205         return ret;
206 }
207
208 /*
209  * cap_set_pg - set capabilities for all processes in a given process
210  * group.  We call this holding task_capability_lock and tasklist_lock.
211  */
212 static inline int cap_set_pg(int pgrp_nr, kernel_cap_t *effective,
213                               kernel_cap_t *inheritable,
214                               kernel_cap_t *permitted)
215 {
216         struct task_struct *g, *target;
217         int ret = -EPERM;
218         int found = 0;
219         struct pid *pgrp;
220
221         pgrp = find_vpid(pgrp_nr);
222         do_each_pid_task(pgrp, PIDTYPE_PGID, g) {
223                 target = g;
224                 while_each_thread(g, target) {
225                         if (!security_capset_check(target, effective,
226                                                         inheritable,
227                                                         permitted)) {
228                                 security_capset_set(target, effective,
229                                                         inheritable,
230                                                         permitted);
231                                 ret = 0;
232                         }
233                         found = 1;
234                 }
235         } while_each_pid_task(pgrp, PIDTYPE_PGID, g);
236
237         if (!found)
238                 ret = 0;
239         return ret;
240 }
241
242 /*
243  * cap_set_all - set capabilities for all processes other than init
244  * and self.  We call this holding task_capability_lock and tasklist_lock.
245  */
246 static inline int cap_set_all(kernel_cap_t *effective,
247                                kernel_cap_t *inheritable,
248                                kernel_cap_t *permitted)
249 {
250      struct task_struct *g, *target;
251      int ret = -EPERM;
252      int found = 0;
253
254      do_each_thread(g, target) {
255              if (target == current || is_container_init(target->group_leader))
256                      continue;
257              found = 1;
258              if (security_capset_check(target, effective, inheritable,
259                                                 permitted))
260                      continue;
261              ret = 0;
262              security_capset_set(target, effective, inheritable, permitted);
263      } while_each_thread(g, target);
264
265      if (!found)
266              ret = 0;
267      return ret;
268 }
269
270 /**
271  * sys_capset - set capabilities for a process or a group of processes
272  * @header: pointer to struct that contains capability version and
273  *      target pid data
274  * @data: pointer to struct that contains the effective, permitted,
275  *      and inheritable capabilities
276  *
277  * Set capabilities for a given process, all processes, or all
278  * processes in a given process group.
279  *
280  * The restrictions on setting capabilities are specified as:
281  *
282  * [pid is for the 'target' task.  'current' is the calling task.]
283  *
284  * I: any raised capabilities must be a subset of the (old current) permitted
285  * P: any raised capabilities must be a subset of the (old current) permitted
286  * E: must be set to a subset of (new target) permitted
287  *
288  * Returns 0 on success and < 0 on error.
289  */
290 asmlinkage long sys_capset(cap_user_header_t header, const cap_user_data_t data)
291 {
292         struct __user_cap_data_struct kdata[_KERNEL_CAPABILITY_U32S];
293         unsigned i, tocopy;
294         kernel_cap_t inheritable, permitted, effective;
295         struct task_struct *target;
296         int ret;
297         pid_t pid;
298
299         ret = cap_validate_magic(header, &tocopy);
300         if (ret != 0)
301                 return ret;
302
303         if (get_user(pid, &header->pid))
304                 return -EFAULT;
305
306         if (pid && pid != task_pid_vnr(current) && !capable(CAP_SETPCAP))
307                 return -EPERM;
308
309         if (copy_from_user(&kdata, data, tocopy
310                            * sizeof(struct __user_cap_data_struct))) {
311                 return -EFAULT;
312         }
313
314         for (i = 0; i < tocopy; i++) {
315                 effective.cap[i] = kdata[i].effective;
316                 permitted.cap[i] = kdata[i].permitted;
317                 inheritable.cap[i] = kdata[i].inheritable;
318         }
319         while (i < _KERNEL_CAPABILITY_U32S) {
320                 effective.cap[i] = 0;
321                 permitted.cap[i] = 0;
322                 inheritable.cap[i] = 0;
323                 i++;
324         }
325
326         spin_lock(&task_capability_lock);
327         read_lock(&tasklist_lock);
328
329         if (pid > 0 && pid != task_pid_vnr(current)) {
330                 target = find_task_by_vpid(pid);
331                 if (!target) {
332                         ret = -ESRCH;
333                         goto out;
334                 }
335         } else
336                 target = current;
337
338         ret = 0;
339
340         /* having verified that the proposed changes are legal,
341            we now put them into effect. */
342         if (pid < 0) {
343                 if (pid == -1)  /* all procs other than current and init */
344                         ret = cap_set_all(&effective, &inheritable, &permitted);
345
346                 else            /* all procs in process group */
347                         ret = cap_set_pg(-pid, &effective, &inheritable,
348                                          &permitted);
349         } else {
350                 ret = security_capset_check(target, &effective, &inheritable,
351                                             &permitted);
352                 if (!ret)
353                         security_capset_set(target, &effective, &inheritable,
354                                             &permitted);
355         }
356
357 out:
358         read_unlock(&tasklist_lock);
359         spin_unlock(&task_capability_lock);
360
361         return ret;
362 }
363
364 int __capable(struct task_struct *t, int cap)
365 {
366         if (security_capable(t, cap) == 0) {
367                 t->flags |= PF_SUPERPRIV;
368                 return 1;
369         }
370         return 0;
371 }
372
373 int capable(int cap)
374 {
375         return __capable(current, cap);
376 }
377 EXPORT_SYMBOL(capable);