security: filesystem capabilities: fix fragile setuid fixup code
[linux-2.6-block.git] / security / commoncap.c
CommitLineData
e338d263 1/* Common capabilities, needed by capability.o and root_plug.o
1da177e4
LT
2 *
3 * This program is free software; you can redistribute it and/or modify
4 * it under the terms of the GNU General Public License as published by
5 * the Free Software Foundation; either version 2 of the License, or
6 * (at your option) any later version.
7 *
8 */
9
c59ede7b 10#include <linux/capability.h>
1da177e4
LT
11#include <linux/module.h>
12#include <linux/init.h>
13#include <linux/kernel.h>
14#include <linux/security.h>
15#include <linux/file.h>
16#include <linux/mm.h>
17#include <linux/mman.h>
18#include <linux/pagemap.h>
19#include <linux/swap.h>
1da177e4
LT
20#include <linux/skbuff.h>
21#include <linux/netlink.h>
22#include <linux/ptrace.h>
23#include <linux/xattr.h>
24#include <linux/hugetlb.h>
b5376771 25#include <linux/mount.h>
b460cbc5 26#include <linux/sched.h>
3898b1b4
AM
27#include <linux/prctl.h>
28#include <linux/securebits.h>
72c2d582 29
1da177e4
LT
30int cap_netlink_send(struct sock *sk, struct sk_buff *skb)
31{
32 NETLINK_CB(skb).eff_cap = current->cap_effective;
33 return 0;
34}
35
c7bdb545 36int cap_netlink_recv(struct sk_buff *skb, int cap)
1da177e4 37{
c7bdb545 38 if (!cap_raised(NETLINK_CB(skb).eff_cap, cap))
1da177e4
LT
39 return -EPERM;
40 return 0;
41}
42
43EXPORT_SYMBOL(cap_netlink_recv);
44
a6dbb1ef
AM
45/*
46 * NOTE WELL: cap_capable() cannot be used like the kernel's capable()
47 * function. That is, it has the reverse semantics: cap_capable()
48 * returns 0 when a task has a capability, but the kernel's capable()
49 * returns 1 for this case.
50 */
1da177e4
LT
51int cap_capable (struct task_struct *tsk, int cap)
52{
53 /* Derived from include/linux/sched.h:capable. */
54 if (cap_raised(tsk->cap_effective, cap))
55 return 0;
56 return -EPERM;
57}
58
59int cap_settime(struct timespec *ts, struct timezone *tz)
60{
61 if (!capable(CAP_SYS_TIME))
62 return -EPERM;
63 return 0;
64}
65
66int cap_ptrace (struct task_struct *parent, struct task_struct *child)
67{
68 /* Derived from arch/i386/kernel/ptrace.c:sys_ptrace. */
d4eb82c7
CW
69 if (!cap_issubset(child->cap_permitted, parent->cap_permitted) &&
70 !__capable(parent, CAP_SYS_PTRACE))
1da177e4
LT
71 return -EPERM;
72 return 0;
73}
74
75int cap_capget (struct task_struct *target, kernel_cap_t *effective,
76 kernel_cap_t *inheritable, kernel_cap_t *permitted)
77{
78 /* Derived from kernel/capability.c:sys_capget. */
e338d263
AM
79 *effective = target->cap_effective;
80 *inheritable = target->cap_inheritable;
81 *permitted = target->cap_permitted;
1da177e4
LT
82 return 0;
83}
84
72c2d582
AM
85#ifdef CONFIG_SECURITY_FILE_CAPABILITIES
86
87static inline int cap_block_setpcap(struct task_struct *target)
88{
89 /*
90 * No support for remote process capability manipulation with
91 * filesystem capability support.
92 */
93 return (target != current);
94}
95
96static inline int cap_inh_is_capped(void)
97{
98 /*
a6dbb1ef
AM
99 * Return 1 if changes to the inheritable set are limited
100 * to the old permitted set. That is, if the current task
101 * does *not* possess the CAP_SETPCAP capability.
72c2d582 102 */
a6dbb1ef 103 return (cap_capable(current, CAP_SETPCAP) != 0);
72c2d582
AM
104}
105
106#else /* ie., ndef CONFIG_SECURITY_FILE_CAPABILITIES */
107
108static inline int cap_block_setpcap(struct task_struct *t) { return 0; }
109static inline int cap_inh_is_capped(void) { return 1; }
110
111#endif /* def CONFIG_SECURITY_FILE_CAPABILITIES */
112
1da177e4
LT
113int cap_capset_check (struct task_struct *target, kernel_cap_t *effective,
114 kernel_cap_t *inheritable, kernel_cap_t *permitted)
115{
72c2d582
AM
116 if (cap_block_setpcap(target)) {
117 return -EPERM;
118 }
119 if (cap_inh_is_capped()
120 && !cap_issubset(*inheritable,
121 cap_combine(target->cap_inheritable,
122 current->cap_permitted))) {
123 /* incapable of using this inheritable set */
1da177e4
LT
124 return -EPERM;
125 }
3b7391de
SH
126 if (!cap_issubset(*inheritable,
127 cap_combine(target->cap_inheritable,
128 current->cap_bset))) {
129 /* no new pI capabilities outside bounding set */
130 return -EPERM;
131 }
1da177e4
LT
132
133 /* verify restrictions on target's new Permitted set */
134 if (!cap_issubset (*permitted,
135 cap_combine (target->cap_permitted,
136 current->cap_permitted))) {
137 return -EPERM;
138 }
139
140 /* verify the _new_Effective_ is a subset of the _new_Permitted_ */
141 if (!cap_issubset (*effective, *permitted)) {
142 return -EPERM;
143 }
144
145 return 0;
146}
147
148void cap_capset_set (struct task_struct *target, kernel_cap_t *effective,
149 kernel_cap_t *inheritable, kernel_cap_t *permitted)
150{
151 target->cap_effective = *effective;
152 target->cap_inheritable = *inheritable;
153 target->cap_permitted = *permitted;
154}
155
b5376771
SH
156static inline void bprm_clear_caps(struct linux_binprm *bprm)
157{
158 cap_clear(bprm->cap_inheritable);
159 cap_clear(bprm->cap_permitted);
160 bprm->cap_effective = false;
161}
162
163#ifdef CONFIG_SECURITY_FILE_CAPABILITIES
164
165int cap_inode_need_killpriv(struct dentry *dentry)
166{
167 struct inode *inode = dentry->d_inode;
168 int error;
169
170 if (!inode->i_op || !inode->i_op->getxattr)
171 return 0;
172
173 error = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, NULL, 0);
174 if (error <= 0)
175 return 0;
176 return 1;
177}
178
179int cap_inode_killpriv(struct dentry *dentry)
180{
181 struct inode *inode = dentry->d_inode;
182
183 if (!inode->i_op || !inode->i_op->removexattr)
184 return 0;
185
186 return inode->i_op->removexattr(dentry, XATTR_NAME_CAPS);
187}
188
e338d263
AM
189static inline int cap_from_disk(struct vfs_cap_data *caps,
190 struct linux_binprm *bprm, unsigned size)
b5376771
SH
191{
192 __u32 magic_etc;
e338d263 193 unsigned tocopy, i;
b5376771 194
e338d263 195 if (size < sizeof(magic_etc))
b5376771
SH
196 return -EINVAL;
197
e338d263 198 magic_etc = le32_to_cpu(caps->magic_etc);
b5376771
SH
199
200 switch ((magic_etc & VFS_CAP_REVISION_MASK)) {
e338d263
AM
201 case VFS_CAP_REVISION_1:
202 if (size != XATTR_CAPS_SZ_1)
203 return -EINVAL;
204 tocopy = VFS_CAP_U32_1;
205 break;
206 case VFS_CAP_REVISION_2:
207 if (size != XATTR_CAPS_SZ_2)
208 return -EINVAL;
209 tocopy = VFS_CAP_U32_2;
210 break;
b5376771
SH
211 default:
212 return -EINVAL;
213 }
e338d263
AM
214
215 if (magic_etc & VFS_CAP_FLAGS_EFFECTIVE) {
216 bprm->cap_effective = true;
217 } else {
218 bprm->cap_effective = false;
219 }
220
221 for (i = 0; i < tocopy; ++i) {
222 bprm->cap_permitted.cap[i] =
223 le32_to_cpu(caps->data[i].permitted);
224 bprm->cap_inheritable.cap[i] =
225 le32_to_cpu(caps->data[i].inheritable);
226 }
227 while (i < VFS_CAP_U32) {
228 bprm->cap_permitted.cap[i] = 0;
229 bprm->cap_inheritable.cap[i] = 0;
230 i++;
231 }
232
233 return 0;
b5376771
SH
234}
235
236/* Locate any VFS capabilities: */
237static int get_file_caps(struct linux_binprm *bprm)
238{
239 struct dentry *dentry;
240 int rc = 0;
e338d263 241 struct vfs_cap_data vcaps;
b5376771
SH
242 struct inode *inode;
243
244 if (bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID) {
245 bprm_clear_caps(bprm);
246 return 0;
247 }
248
249 dentry = dget(bprm->file->f_dentry);
250 inode = dentry->d_inode;
251 if (!inode->i_op || !inode->i_op->getxattr)
252 goto out;
253
e338d263
AM
254 rc = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, &vcaps,
255 XATTR_CAPS_SZ);
b5376771
SH
256 if (rc == -ENODATA || rc == -EOPNOTSUPP) {
257 /* no data, that's ok */
258 rc = 0;
259 goto out;
260 }
261 if (rc < 0)
262 goto out;
263
e338d263 264 rc = cap_from_disk(&vcaps, bprm, rc);
b5376771
SH
265 if (rc)
266 printk(KERN_NOTICE "%s: cap_from_disk returned %d for %s\n",
dd6f953a 267 __func__, rc, bprm->filename);
b5376771
SH
268
269out:
270 dput(dentry);
271 if (rc)
272 bprm_clear_caps(bprm);
273
274 return rc;
275}
276
277#else
278int cap_inode_need_killpriv(struct dentry *dentry)
279{
280 return 0;
281}
282
283int cap_inode_killpriv(struct dentry *dentry)
284{
285 return 0;
286}
287
288static inline int get_file_caps(struct linux_binprm *bprm)
289{
290 bprm_clear_caps(bprm);
291 return 0;
292}
293#endif
294
1da177e4
LT
295int cap_bprm_set_security (struct linux_binprm *bprm)
296{
b5376771 297 int ret;
1da177e4 298
b5376771
SH
299 ret = get_file_caps(bprm);
300 if (ret)
301 printk(KERN_NOTICE "%s: get_file_caps returned %d for %s\n",
dd6f953a 302 __func__, ret, bprm->filename);
1da177e4
LT
303
304 /* To support inheritance of root-permissions and suid-root
305 * executables under compatibility mode, we raise all three
306 * capability sets for the file.
307 *
308 * If only the real uid is 0, we only raise the inheritable
309 * and permitted sets of the executable file.
310 */
311
312 if (!issecure (SECURE_NOROOT)) {
313 if (bprm->e_uid == 0 || current->uid == 0) {
314 cap_set_full (bprm->cap_inheritable);
315 cap_set_full (bprm->cap_permitted);
316 }
317 if (bprm->e_uid == 0)
b5376771 318 bprm->cap_effective = true;
1da177e4 319 }
b5376771
SH
320
321 return ret;
1da177e4
LT
322}
323
324void cap_bprm_apply_creds (struct linux_binprm *bprm, int unsafe)
325{
326 /* Derived from fs/exec.c:compute_creds. */
327 kernel_cap_t new_permitted, working;
328
3b7391de
SH
329 new_permitted = cap_intersect(bprm->cap_permitted,
330 current->cap_bset);
331 working = cap_intersect(bprm->cap_inheritable,
1da177e4 332 current->cap_inheritable);
3b7391de 333 new_permitted = cap_combine(new_permitted, working);
1da177e4
LT
334
335 if (bprm->e_uid != current->uid || bprm->e_gid != current->gid ||
336 !cap_issubset (new_permitted, current->cap_permitted)) {
6c5d5238 337 set_dumpable(current->mm, suid_dumpable);
b5376771 338 current->pdeath_signal = 0;
1da177e4
LT
339
340 if (unsafe & ~LSM_UNSAFE_PTRACE_CAP) {
341 if (!capable(CAP_SETUID)) {
342 bprm->e_uid = current->uid;
343 bprm->e_gid = current->gid;
344 }
345 if (!capable (CAP_SETPCAP)) {
346 new_permitted = cap_intersect (new_permitted,
347 current->cap_permitted);
348 }
349 }
350 }
351
352 current->suid = current->euid = current->fsuid = bprm->e_uid;
353 current->sgid = current->egid = current->fsgid = bprm->e_gid;
354
355 /* For init, we want to retain the capabilities set
356 * in the init_task struct. Thus we skip the usual
357 * capability rules */
b460cbc5 358 if (!is_global_init(current)) {
1da177e4 359 current->cap_permitted = new_permitted;
e338d263
AM
360 if (bprm->cap_effective)
361 current->cap_effective = new_permitted;
362 else
363 cap_clear(current->cap_effective);
1da177e4
LT
364 }
365
366 /* AUD: Audit candidate if current->cap_effective is set */
367
3898b1b4 368 current->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
1da177e4
LT
369}
370
371int cap_bprm_secureexec (struct linux_binprm *bprm)
372{
b5376771
SH
373 if (current->uid != 0) {
374 if (bprm->cap_effective)
375 return 1;
376 if (!cap_isclear(bprm->cap_permitted))
377 return 1;
378 if (!cap_isclear(bprm->cap_inheritable))
379 return 1;
380 }
381
1da177e4
LT
382 return (current->euid != current->uid ||
383 current->egid != current->gid);
384}
385
8f0cfa52
DH
386int cap_inode_setxattr(struct dentry *dentry, const char *name,
387 const void *value, size_t size, int flags)
1da177e4 388{
b5376771
SH
389 if (!strcmp(name, XATTR_NAME_CAPS)) {
390 if (!capable(CAP_SETFCAP))
391 return -EPERM;
392 return 0;
393 } else if (!strncmp(name, XATTR_SECURITY_PREFIX,
1da177e4
LT
394 sizeof(XATTR_SECURITY_PREFIX) - 1) &&
395 !capable(CAP_SYS_ADMIN))
396 return -EPERM;
397 return 0;
398}
399
8f0cfa52 400int cap_inode_removexattr(struct dentry *dentry, const char *name)
1da177e4 401{
b5376771
SH
402 if (!strcmp(name, XATTR_NAME_CAPS)) {
403 if (!capable(CAP_SETFCAP))
404 return -EPERM;
405 return 0;
406 } else if (!strncmp(name, XATTR_SECURITY_PREFIX,
1da177e4
LT
407 sizeof(XATTR_SECURITY_PREFIX) - 1) &&
408 !capable(CAP_SYS_ADMIN))
409 return -EPERM;
410 return 0;
411}
412
413/* moved from kernel/sys.c. */
414/*
415 * cap_emulate_setxuid() fixes the effective / permitted capabilities of
416 * a process after a call to setuid, setreuid, or setresuid.
417 *
418 * 1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
419 * {r,e,s}uid != 0, the permitted and effective capabilities are
420 * cleared.
421 *
422 * 2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
423 * capabilities of the process are cleared.
424 *
425 * 3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
426 * capabilities are set to the permitted capabilities.
427 *
428 * fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
429 * never happen.
430 *
431 * -astor
432 *
433 * cevans - New behaviour, Oct '99
434 * A process may, via prctl(), elect to keep its capabilities when it
435 * calls setuid() and switches away from uid==0. Both permitted and
436 * effective sets will be retained.
437 * Without this change, it was impossible for a daemon to drop only some
438 * of its privilege. The call to setuid(!=0) would drop all privileges!
439 * Keeping uid 0 is not an option because uid 0 owns too many vital
440 * files..
441 * Thanks to Olaf Kirch and Peter Benie for spotting this.
442 */
443static inline void cap_emulate_setxuid (int old_ruid, int old_euid,
444 int old_suid)
445{
446 if ((old_ruid == 0 || old_euid == 0 || old_suid == 0) &&
447 (current->uid != 0 && current->euid != 0 && current->suid != 0) &&
3898b1b4 448 !issecure(SECURE_KEEP_CAPS)) {
1da177e4
LT
449 cap_clear (current->cap_permitted);
450 cap_clear (current->cap_effective);
451 }
452 if (old_euid == 0 && current->euid != 0) {
453 cap_clear (current->cap_effective);
454 }
455 if (old_euid != 0 && current->euid == 0) {
456 current->cap_effective = current->cap_permitted;
457 }
458}
459
460int cap_task_post_setuid (uid_t old_ruid, uid_t old_euid, uid_t old_suid,
461 int flags)
462{
463 switch (flags) {
464 case LSM_SETID_RE:
465 case LSM_SETID_ID:
466 case LSM_SETID_RES:
467 /* Copied from kernel/sys.c:setreuid/setuid/setresuid. */
468 if (!issecure (SECURE_NO_SETUID_FIXUP)) {
469 cap_emulate_setxuid (old_ruid, old_euid, old_suid);
470 }
471 break;
472 case LSM_SETID_FS:
473 {
474 uid_t old_fsuid = old_ruid;
475
476 /* Copied from kernel/sys.c:setfsuid. */
477
478 /*
479 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
480 * if not, we might be a bit too harsh here.
481 */
482
483 if (!issecure (SECURE_NO_SETUID_FIXUP)) {
484 if (old_fsuid == 0 && current->fsuid != 0) {
e338d263
AM
485 current->cap_effective =
486 cap_drop_fs_set(
487 current->cap_effective);
1da177e4
LT
488 }
489 if (old_fsuid != 0 && current->fsuid == 0) {
e338d263
AM
490 current->cap_effective =
491 cap_raise_fs_set(
492 current->cap_effective,
493 current->cap_permitted);
1da177e4
LT
494 }
495 }
496 break;
497 }
498 default:
499 return -EINVAL;
500 }
501
502 return 0;
503}
504
b5376771
SH
505#ifdef CONFIG_SECURITY_FILE_CAPABILITIES
506/*
507 * Rationale: code calling task_setscheduler, task_setioprio, and
508 * task_setnice, assumes that
509 * . if capable(cap_sys_nice), then those actions should be allowed
510 * . if not capable(cap_sys_nice), but acting on your own processes,
511 * then those actions should be allowed
512 * This is insufficient now since you can call code without suid, but
513 * yet with increased caps.
514 * So we check for increased caps on the target process.
515 */
516static inline int cap_safe_nice(struct task_struct *p)
517{
518 if (!cap_issubset(p->cap_permitted, current->cap_permitted) &&
519 !__capable(current, CAP_SYS_NICE))
520 return -EPERM;
521 return 0;
522}
523
524int cap_task_setscheduler (struct task_struct *p, int policy,
525 struct sched_param *lp)
526{
527 return cap_safe_nice(p);
528}
529
530int cap_task_setioprio (struct task_struct *p, int ioprio)
531{
532 return cap_safe_nice(p);
533}
534
535int cap_task_setnice (struct task_struct *p, int nice)
536{
537 return cap_safe_nice(p);
538}
539
3b7391de
SH
540/*
541 * called from kernel/sys.c for prctl(PR_CABSET_DROP)
542 * done without task_capability_lock() because it introduces
543 * no new races - i.e. only another task doing capget() on
544 * this task could get inconsistent info. There can be no
545 * racing writer bc a task can only change its own caps.
546 */
3898b1b4 547static long cap_prctl_drop(unsigned long cap)
3b7391de
SH
548{
549 if (!capable(CAP_SETPCAP))
550 return -EPERM;
551 if (!cap_valid(cap))
552 return -EINVAL;
553 cap_lower(current->cap_bset, cap);
554 return 0;
555}
3898b1b4 556
b5376771
SH
557#else
558int cap_task_setscheduler (struct task_struct *p, int policy,
559 struct sched_param *lp)
560{
561 return 0;
562}
563int cap_task_setioprio (struct task_struct *p, int ioprio)
564{
565 return 0;
566}
567int cap_task_setnice (struct task_struct *p, int nice)
568{
569 return 0;
570}
b5376771
SH
571#endif
572
3898b1b4
AM
573int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
574 unsigned long arg4, unsigned long arg5, long *rc_p)
575{
576 long error = 0;
577
578 switch (option) {
579 case PR_CAPBSET_READ:
580 if (!cap_valid(arg2))
581 error = -EINVAL;
582 else
583 error = !!cap_raised(current->cap_bset, arg2);
584 break;
585#ifdef CONFIG_SECURITY_FILE_CAPABILITIES
586 case PR_CAPBSET_DROP:
587 error = cap_prctl_drop(arg2);
588 break;
589
590 /*
591 * The next four prctl's remain to assist with transitioning a
592 * system from legacy UID=0 based privilege (when filesystem
593 * capabilities are not in use) to a system using filesystem
594 * capabilities only - as the POSIX.1e draft intended.
595 *
596 * Note:
597 *
598 * PR_SET_SECUREBITS =
599 * issecure_mask(SECURE_KEEP_CAPS_LOCKED)
600 * | issecure_mask(SECURE_NOROOT)
601 * | issecure_mask(SECURE_NOROOT_LOCKED)
602 * | issecure_mask(SECURE_NO_SETUID_FIXUP)
603 * | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
604 *
605 * will ensure that the current process and all of its
606 * children will be locked into a pure
607 * capability-based-privilege environment.
608 */
609 case PR_SET_SECUREBITS:
610 if ((((current->securebits & SECURE_ALL_LOCKS) >> 1)
611 & (current->securebits ^ arg2)) /*[1]*/
612 || ((current->securebits & SECURE_ALL_LOCKS
613 & ~arg2)) /*[2]*/
614 || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS)) /*[3]*/
615 || (cap_capable(current, CAP_SETPCAP) != 0)) { /*[4]*/
616 /*
617 * [1] no changing of bits that are locked
618 * [2] no unlocking of locks
619 * [3] no setting of unsupported bits
620 * [4] doing anything requires privilege (go read about
621 * the "sendmail capabilities bug")
622 */
623 error = -EPERM; /* cannot change a locked bit */
624 } else {
625 current->securebits = arg2;
626 }
627 break;
628 case PR_GET_SECUREBITS:
629 error = current->securebits;
630 break;
631
632#endif /* def CONFIG_SECURITY_FILE_CAPABILITIES */
633
634 case PR_GET_KEEPCAPS:
635 if (issecure(SECURE_KEEP_CAPS))
636 error = 1;
637 break;
638 case PR_SET_KEEPCAPS:
639 if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
640 error = -EINVAL;
641 else if (issecure(SECURE_KEEP_CAPS_LOCKED))
642 error = -EPERM;
643 else if (arg2)
644 current->securebits |= issecure_mask(SECURE_KEEP_CAPS);
645 else
646 current->securebits &=
647 ~issecure_mask(SECURE_KEEP_CAPS);
648 break;
649
650 default:
651 /* No functionality available - continue with default */
652 return 0;
653 }
654
655 /* Functionality provided */
656 *rc_p = error;
657 return 1;
658}
659
1da177e4
LT
660void cap_task_reparent_to_init (struct task_struct *p)
661{
e338d263
AM
662 cap_set_init_eff(p->cap_effective);
663 cap_clear(p->cap_inheritable);
664 cap_set_full(p->cap_permitted);
3898b1b4 665 p->securebits = SECUREBITS_DEFAULT;
1da177e4
LT
666 return;
667}
668
669int cap_syslog (int type)
670{
671 if ((type != 3 && type != 10) && !capable(CAP_SYS_ADMIN))
672 return -EPERM;
673 return 0;
674}
675
34b4e4aa 676int cap_vm_enough_memory(struct mm_struct *mm, long pages)
1da177e4
LT
677{
678 int cap_sys_admin = 0;
679
680 if (cap_capable(current, CAP_SYS_ADMIN) == 0)
681 cap_sys_admin = 1;
34b4e4aa 682 return __vm_enough_memory(mm, pages, cap_sys_admin);
1da177e4
LT
683}
684