security: filesystem capabilities refactor kernel code
[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 #ifndef CONFIG_SECURITY_FILE_CAPABILITIES
119
120 /*
121  * Without filesystem capability support, we nominally support one process
122  * setting the capabilities of another
123  */
124 static inline int cap_get_target_pid(pid_t pid, kernel_cap_t *pEp,
125                                      kernel_cap_t *pIp, kernel_cap_t *pPp)
126 {
127         struct task_struct *target;
128         int ret;
129
130         spin_lock(&task_capability_lock);
131         read_lock(&tasklist_lock);
132
133         if (pid && pid != task_pid_vnr(current)) {
134                 target = find_task_by_vpid(pid);
135                 if (!target) {
136                         ret = -ESRCH;
137                         goto out;
138                 }
139         } else
140                 target = current;
141
142         ret = security_capget(target, pEp, pIp, pPp);
143
144 out:
145         read_unlock(&tasklist_lock);
146         spin_unlock(&task_capability_lock);
147
148         return ret;
149 }
150
151 /*
152  * cap_set_pg - set capabilities for all processes in a given process
153  * group.  We call this holding task_capability_lock and tasklist_lock.
154  */
155 static inline int cap_set_pg(int pgrp_nr, kernel_cap_t *effective,
156                              kernel_cap_t *inheritable,
157                              kernel_cap_t *permitted)
158 {
159         struct task_struct *g, *target;
160         int ret = -EPERM;
161         int found = 0;
162         struct pid *pgrp;
163
164         spin_lock(&task_capability_lock);
165         read_lock(&tasklist_lock);
166
167         pgrp = find_vpid(pgrp_nr);
168         do_each_pid_task(pgrp, PIDTYPE_PGID, g) {
169                 target = g;
170                 while_each_thread(g, target) {
171                         if (!security_capset_check(target, effective,
172                                                    inheritable, permitted)) {
173                                 security_capset_set(target, effective,
174                                                     inheritable, permitted);
175                                 ret = 0;
176                         }
177                         found = 1;
178                 }
179         } while_each_pid_task(pgrp, PIDTYPE_PGID, g);
180
181         read_unlock(&tasklist_lock);
182         spin_unlock(&task_capability_lock);
183
184         if (!found)
185                 ret = 0;
186         return ret;
187 }
188
189 /*
190  * cap_set_all - set capabilities for all processes other than init
191  * and self.  We call this holding task_capability_lock and tasklist_lock.
192  */
193 static inline int cap_set_all(kernel_cap_t *effective,
194                               kernel_cap_t *inheritable,
195                               kernel_cap_t *permitted)
196 {
197         struct task_struct *g, *target;
198         int ret = -EPERM;
199         int found = 0;
200
201         spin_lock(&task_capability_lock);
202         read_lock(&tasklist_lock);
203
204         do_each_thread(g, target) {
205                 if (target == current
206                     || 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         read_unlock(&tasklist_lock);
217         spin_unlock(&task_capability_lock);
218
219         if (!found)
220                 ret = 0;
221
222         return ret;
223 }
224
225 /*
226  * Given the target pid does not refer to the current process we
227  * need more elaborate support... (This support is not present when
228  * filesystem capabilities are configured.)
229  */
230 static inline int do_sys_capset_other_tasks(pid_t pid, kernel_cap_t *effective,
231                                             kernel_cap_t *inheritable,
232                                             kernel_cap_t *permitted)
233 {
234         struct task_struct *target;
235         int ret;
236
237         if (!capable(CAP_SETPCAP))
238                 return -EPERM;
239
240         if (pid == -1)            /* all procs other than current and init */
241                 return cap_set_all(effective, inheritable, permitted);
242
243         else if (pid < 0)                    /* all procs in process group */
244                 return cap_set_pg(-pid, effective, inheritable, permitted);
245
246         /* target != current */
247         spin_lock(&task_capability_lock);
248         read_lock(&tasklist_lock);
249
250         target = find_task_by_vpid(pid);
251         if (!target)
252                 ret = -ESRCH;
253         else {
254                 ret = security_capset_check(target, effective, inheritable,
255                                             permitted);
256
257                 /* having verified that the proposed changes are legal,
258                    we now put them into effect. */
259                 if (!ret)
260                         security_capset_set(target, effective, inheritable,
261                                             permitted);
262         }
263
264         read_unlock(&tasklist_lock);
265         spin_unlock(&task_capability_lock);
266
267         return ret;
268 }
269
270 #else /* ie., def CONFIG_SECURITY_FILE_CAPABILITIES */
271
272 /*
273  * If we have configured with filesystem capability support, then the
274  * only thing that can change the capabilities of the current process
275  * is the current process. As such, we can't be in this code at the
276  * same time as we are in the process of setting capabilities in this
277  * process. The net result is that we can limit our use of locks to
278  * when we are reading the caps of another process.
279  */
280 static inline int cap_get_target_pid(pid_t pid, kernel_cap_t *pEp,
281                                      kernel_cap_t *pIp, kernel_cap_t *pPp)
282 {
283         int ret;
284
285         if (pid && (pid != task_pid_vnr(current))) {
286                 struct task_struct *target;
287
288                 spin_lock(&task_capability_lock);
289                 read_lock(&tasklist_lock);
290
291                 target = find_task_by_vpid(pid);
292                 if (!target)
293                         ret = -ESRCH;
294                 else
295                         ret = security_capget(target, pEp, pIp, pPp);
296
297                 read_unlock(&tasklist_lock);
298                 spin_unlock(&task_capability_lock);
299         } else
300                 ret = security_capget(current, pEp, pIp, pPp);
301
302         return ret;
303 }
304
305 /*
306  * With filesystem capability support configured, the kernel does not
307  * permit the changing of capabilities in one process by another
308  * process. (CAP_SETPCAP has much less broad semantics when configured
309  * this way.)
310  */
311 static inline int do_sys_capset_other_tasks(pid_t pid,
312                                             kernel_cap_t *effective,
313                                             kernel_cap_t *inheritable,
314                                             kernel_cap_t *permitted)
315 {
316         return -EPERM;
317 }
318
319 #endif /* ie., ndef CONFIG_SECURITY_FILE_CAPABILITIES */
320
321 /*
322  * Atomically modify the effective capabilities returning the original
323  * value. No permission check is performed here - it is assumed that the
324  * caller is permitted to set the desired effective capabilities.
325  */
326 kernel_cap_t cap_set_effective(const kernel_cap_t pE_new)
327 {
328         kernel_cap_t pE_old;
329
330         spin_lock(&task_capability_lock);
331
332         pE_old = current->cap_effective;
333         current->cap_effective = pE_new;
334
335         spin_unlock(&task_capability_lock);
336
337         return pE_old;
338 }
339
340 EXPORT_SYMBOL(cap_set_effective);
341
342 /**
343  * sys_capget - get the capabilities of a given process.
344  * @header: pointer to struct that contains capability version and
345  *      target pid data
346  * @dataptr: pointer to struct that contains the effective, permitted,
347  *      and inheritable capabilities that are returned
348  *
349  * Returns 0 on success and < 0 on error.
350  */
351 asmlinkage long sys_capget(cap_user_header_t header, cap_user_data_t dataptr)
352 {
353         int ret = 0;
354         pid_t pid;
355         unsigned tocopy;
356         kernel_cap_t pE, pI, pP;
357
358         ret = cap_validate_magic(header, &tocopy);
359         if (ret != 0)
360                 return ret;
361
362         if (get_user(pid, &header->pid))
363                 return -EFAULT;
364
365         if (pid < 0)
366                 return -EINVAL;
367
368         ret = cap_get_target_pid(pid, &pE, &pI, &pP);
369
370         if (!ret) {
371                 struct __user_cap_data_struct kdata[_KERNEL_CAPABILITY_U32S];
372                 unsigned i;
373
374                 for (i = 0; i < tocopy; i++) {
375                         kdata[i].effective = pE.cap[i];
376                         kdata[i].permitted = pP.cap[i];
377                         kdata[i].inheritable = pI.cap[i];
378                 }
379
380                 /*
381                  * Note, in the case, tocopy < _KERNEL_CAPABILITY_U32S,
382                  * we silently drop the upper capabilities here. This
383                  * has the effect of making older libcap
384                  * implementations implicitly drop upper capability
385                  * bits when they perform a: capget/modify/capset
386                  * sequence.
387                  *
388                  * This behavior is considered fail-safe
389                  * behavior. Upgrading the application to a newer
390                  * version of libcap will enable access to the newer
391                  * capabilities.
392                  *
393                  * An alternative would be to return an error here
394                  * (-ERANGE), but that causes legacy applications to
395                  * unexpectidly fail; the capget/modify/capset aborts
396                  * before modification is attempted and the application
397                  * fails.
398                  */
399                 if (copy_to_user(dataptr, kdata, tocopy
400                                  * sizeof(struct __user_cap_data_struct))) {
401                         return -EFAULT;
402                 }
403         }
404
405         return ret;
406 }
407
408 /**
409  * sys_capset - set capabilities for a process or (*) a group of processes
410  * @header: pointer to struct that contains capability version and
411  *      target pid data
412  * @data: pointer to struct that contains the effective, permitted,
413  *      and inheritable capabilities
414  *
415  * Set capabilities for a given process, all processes, or all
416  * processes in a given process group.
417  *
418  * The restrictions on setting capabilities are specified as:
419  *
420  * [pid is for the 'target' task.  'current' is the calling task.]
421  *
422  * I: any raised capabilities must be a subset of the (old current) permitted
423  * P: any raised capabilities must be a subset of the (old current) permitted
424  * E: must be set to a subset of (new target) permitted
425  *
426  * Returns 0 on success and < 0 on error.
427  */
428 asmlinkage long sys_capset(cap_user_header_t header, const cap_user_data_t data)
429 {
430         struct __user_cap_data_struct kdata[_KERNEL_CAPABILITY_U32S];
431         unsigned i, tocopy;
432         kernel_cap_t inheritable, permitted, effective;
433         int ret;
434         pid_t pid;
435
436         ret = cap_validate_magic(header, &tocopy);
437         if (ret != 0)
438                 return ret;
439
440         if (get_user(pid, &header->pid))
441                 return -EFAULT;
442
443         if (copy_from_user(&kdata, data, tocopy
444                            * sizeof(struct __user_cap_data_struct))) {
445                 return -EFAULT;
446         }
447
448         for (i = 0; i < tocopy; i++) {
449                 effective.cap[i] = kdata[i].effective;
450                 permitted.cap[i] = kdata[i].permitted;
451                 inheritable.cap[i] = kdata[i].inheritable;
452         }
453         while (i < _KERNEL_CAPABILITY_U32S) {
454                 effective.cap[i] = 0;
455                 permitted.cap[i] = 0;
456                 inheritable.cap[i] = 0;
457                 i++;
458         }
459
460         if (pid && (pid != task_pid_vnr(current)))
461                 ret = do_sys_capset_other_tasks(pid, &effective, &inheritable,
462                                                 &permitted);
463         else {
464                 /*
465                  * This lock is required even when filesystem
466                  * capability support is configured - it protects the
467                  * sys_capget() call from returning incorrect data in
468                  * the case that the targeted process is not the
469                  * current one.
470                  */
471                 spin_lock(&task_capability_lock);
472
473                 ret = security_capset_check(current, &effective, &inheritable,
474                                             &permitted);
475                 /*
476                  * Having verified that the proposed changes are
477                  * legal, we now put them into effect.
478                  */
479                 if (!ret)
480                         security_capset_set(current, &effective, &inheritable,
481                                             &permitted);
482                 spin_unlock(&task_capability_lock);
483         }
484
485
486         return ret;
487 }
488
489 int __capable(struct task_struct *t, int cap)
490 {
491         if (security_capable(t, cap) == 0) {
492                 t->flags |= PF_SUPERPRIV;
493                 return 1;
494         }
495         return 0;
496 }
497
498 int capable(int cap)
499 {
500         return __capable(current, cap);
501 }
502 EXPORT_SYMBOL(capable);