1 /* Common capabilities, needed by capability.o and root_plug.o
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.
10 #include <linux/capability.h>
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>
17 #include <linux/mman.h>
18 #include <linux/pagemap.h>
19 #include <linux/swap.h>
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>
25 #include <linux/mount.h>
26 #include <linux/sched.h>
28 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
30 * Because of the reduced scope of CAP_SETPCAP when filesystem
31 * capabilities are in effect, it is safe to allow this capability to
32 * be available in the default configuration.
34 # define CAP_INIT_BSET CAP_FULL_SET
35 #else /* ie. ndef CONFIG_SECURITY_FILE_CAPABILITIES */
36 # define CAP_INIT_BSET CAP_INIT_EFF_SET
37 #endif /* def CONFIG_SECURITY_FILE_CAPABILITIES */
39 kernel_cap_t cap_bset = CAP_INIT_BSET; /* systemwide capability bound */
40 EXPORT_SYMBOL(cap_bset);
42 /* Global security state */
44 unsigned securebits = SECUREBITS_DEFAULT; /* systemwide security settings */
45 EXPORT_SYMBOL(securebits);
47 int cap_netlink_send(struct sock *sk, struct sk_buff *skb)
49 NETLINK_CB(skb).eff_cap = current->cap_effective;
53 int cap_netlink_recv(struct sk_buff *skb, int cap)
55 if (!cap_raised(NETLINK_CB(skb).eff_cap, cap))
60 EXPORT_SYMBOL(cap_netlink_recv);
63 * NOTE WELL: cap_capable() cannot be used like the kernel's capable()
64 * function. That is, it has the reverse semantics: cap_capable()
65 * returns 0 when a task has a capability, but the kernel's capable()
66 * returns 1 for this case.
68 int cap_capable (struct task_struct *tsk, int cap)
70 /* Derived from include/linux/sched.h:capable. */
71 if (cap_raised(tsk->cap_effective, cap))
76 int cap_settime(struct timespec *ts, struct timezone *tz)
78 if (!capable(CAP_SYS_TIME))
83 int cap_ptrace (struct task_struct *parent, struct task_struct *child)
85 /* Derived from arch/i386/kernel/ptrace.c:sys_ptrace. */
86 if (!cap_issubset(child->cap_permitted, parent->cap_permitted) &&
87 !__capable(parent, CAP_SYS_PTRACE))
92 int cap_capget (struct task_struct *target, kernel_cap_t *effective,
93 kernel_cap_t *inheritable, kernel_cap_t *permitted)
95 /* Derived from kernel/capability.c:sys_capget. */
96 *effective = target->cap_effective;
97 *inheritable = target->cap_inheritable;
98 *permitted = target->cap_permitted;
102 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
104 static inline int cap_block_setpcap(struct task_struct *target)
107 * No support for remote process capability manipulation with
108 * filesystem capability support.
110 return (target != current);
113 static inline int cap_inh_is_capped(void)
116 * Return 1 if changes to the inheritable set are limited
117 * to the old permitted set. That is, if the current task
118 * does *not* possess the CAP_SETPCAP capability.
120 return (cap_capable(current, CAP_SETPCAP) != 0);
123 #else /* ie., ndef CONFIG_SECURITY_FILE_CAPABILITIES */
125 static inline int cap_block_setpcap(struct task_struct *t) { return 0; }
126 static inline int cap_inh_is_capped(void) { return 1; }
128 #endif /* def CONFIG_SECURITY_FILE_CAPABILITIES */
130 int cap_capset_check (struct task_struct *target, kernel_cap_t *effective,
131 kernel_cap_t *inheritable, kernel_cap_t *permitted)
133 if (cap_block_setpcap(target)) {
136 if (cap_inh_is_capped()
137 && !cap_issubset(*inheritable,
138 cap_combine(target->cap_inheritable,
139 current->cap_permitted))) {
140 /* incapable of using this inheritable set */
144 /* verify restrictions on target's new Permitted set */
145 if (!cap_issubset (*permitted,
146 cap_combine (target->cap_permitted,
147 current->cap_permitted))) {
151 /* verify the _new_Effective_ is a subset of the _new_Permitted_ */
152 if (!cap_issubset (*effective, *permitted)) {
159 void cap_capset_set (struct task_struct *target, kernel_cap_t *effective,
160 kernel_cap_t *inheritable, kernel_cap_t *permitted)
162 target->cap_effective = *effective;
163 target->cap_inheritable = *inheritable;
164 target->cap_permitted = *permitted;
167 static inline void bprm_clear_caps(struct linux_binprm *bprm)
169 cap_clear(bprm->cap_inheritable);
170 cap_clear(bprm->cap_permitted);
171 bprm->cap_effective = false;
174 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
176 int cap_inode_need_killpriv(struct dentry *dentry)
178 struct inode *inode = dentry->d_inode;
181 if (!inode->i_op || !inode->i_op->getxattr)
184 error = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, NULL, 0);
190 int cap_inode_killpriv(struct dentry *dentry)
192 struct inode *inode = dentry->d_inode;
194 if (!inode->i_op || !inode->i_op->removexattr)
197 return inode->i_op->removexattr(dentry, XATTR_NAME_CAPS);
200 static inline int cap_from_disk(struct vfs_cap_data *caps,
201 struct linux_binprm *bprm, unsigned size)
206 if (size < sizeof(magic_etc))
209 magic_etc = le32_to_cpu(caps->magic_etc);
211 switch ((magic_etc & VFS_CAP_REVISION_MASK)) {
212 case VFS_CAP_REVISION_1:
213 if (size != XATTR_CAPS_SZ_1)
215 tocopy = VFS_CAP_U32_1;
217 case VFS_CAP_REVISION_2:
218 if (size != XATTR_CAPS_SZ_2)
220 tocopy = VFS_CAP_U32_2;
226 if (magic_etc & VFS_CAP_FLAGS_EFFECTIVE) {
227 bprm->cap_effective = true;
229 bprm->cap_effective = false;
232 for (i = 0; i < tocopy; ++i) {
233 bprm->cap_permitted.cap[i] =
234 le32_to_cpu(caps->data[i].permitted);
235 bprm->cap_inheritable.cap[i] =
236 le32_to_cpu(caps->data[i].inheritable);
238 while (i < VFS_CAP_U32) {
239 bprm->cap_permitted.cap[i] = 0;
240 bprm->cap_inheritable.cap[i] = 0;
247 /* Locate any VFS capabilities: */
248 static int get_file_caps(struct linux_binprm *bprm)
250 struct dentry *dentry;
252 struct vfs_cap_data vcaps;
255 if (bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID) {
256 bprm_clear_caps(bprm);
260 dentry = dget(bprm->file->f_dentry);
261 inode = dentry->d_inode;
262 if (!inode->i_op || !inode->i_op->getxattr)
265 rc = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, &vcaps,
267 if (rc == -ENODATA || rc == -EOPNOTSUPP) {
268 /* no data, that's ok */
275 rc = cap_from_disk(&vcaps, bprm, rc);
277 printk(KERN_NOTICE "%s: cap_from_disk returned %d for %s\n",
278 __FUNCTION__, rc, bprm->filename);
283 bprm_clear_caps(bprm);
289 int cap_inode_need_killpriv(struct dentry *dentry)
294 int cap_inode_killpriv(struct dentry *dentry)
299 static inline int get_file_caps(struct linux_binprm *bprm)
301 bprm_clear_caps(bprm);
306 int cap_bprm_set_security (struct linux_binprm *bprm)
310 ret = get_file_caps(bprm);
312 printk(KERN_NOTICE "%s: get_file_caps returned %d for %s\n",
313 __FUNCTION__, ret, bprm->filename);
315 /* To support inheritance of root-permissions and suid-root
316 * executables under compatibility mode, we raise all three
317 * capability sets for the file.
319 * If only the real uid is 0, we only raise the inheritable
320 * and permitted sets of the executable file.
323 if (!issecure (SECURE_NOROOT)) {
324 if (bprm->e_uid == 0 || current->uid == 0) {
325 cap_set_full (bprm->cap_inheritable);
326 cap_set_full (bprm->cap_permitted);
328 if (bprm->e_uid == 0)
329 bprm->cap_effective = true;
335 void cap_bprm_apply_creds (struct linux_binprm *bprm, int unsafe)
337 /* Derived from fs/exec.c:compute_creds. */
338 kernel_cap_t new_permitted, working;
340 new_permitted = cap_intersect (bprm->cap_permitted, cap_bset);
341 working = cap_intersect (bprm->cap_inheritable,
342 current->cap_inheritable);
343 new_permitted = cap_combine (new_permitted, working);
345 if (bprm->e_uid != current->uid || bprm->e_gid != current->gid ||
346 !cap_issubset (new_permitted, current->cap_permitted)) {
347 set_dumpable(current->mm, suid_dumpable);
348 current->pdeath_signal = 0;
350 if (unsafe & ~LSM_UNSAFE_PTRACE_CAP) {
351 if (!capable(CAP_SETUID)) {
352 bprm->e_uid = current->uid;
353 bprm->e_gid = current->gid;
355 if (!capable (CAP_SETPCAP)) {
356 new_permitted = cap_intersect (new_permitted,
357 current->cap_permitted);
362 current->suid = current->euid = current->fsuid = bprm->e_uid;
363 current->sgid = current->egid = current->fsgid = bprm->e_gid;
365 /* For init, we want to retain the capabilities set
366 * in the init_task struct. Thus we skip the usual
367 * capability rules */
368 if (!is_global_init(current)) {
369 current->cap_permitted = new_permitted;
370 if (bprm->cap_effective)
371 current->cap_effective = new_permitted;
373 cap_clear(current->cap_effective);
376 /* AUD: Audit candidate if current->cap_effective is set */
378 current->keep_capabilities = 0;
381 int cap_bprm_secureexec (struct linux_binprm *bprm)
383 if (current->uid != 0) {
384 if (bprm->cap_effective)
386 if (!cap_isclear(bprm->cap_permitted))
388 if (!cap_isclear(bprm->cap_inheritable))
392 return (current->euid != current->uid ||
393 current->egid != current->gid);
396 int cap_inode_setxattr(struct dentry *dentry, char *name, void *value,
397 size_t size, int flags)
399 if (!strcmp(name, XATTR_NAME_CAPS)) {
400 if (!capable(CAP_SETFCAP))
403 } else if (!strncmp(name, XATTR_SECURITY_PREFIX,
404 sizeof(XATTR_SECURITY_PREFIX) - 1) &&
405 !capable(CAP_SYS_ADMIN))
410 int cap_inode_removexattr(struct dentry *dentry, char *name)
412 if (!strcmp(name, XATTR_NAME_CAPS)) {
413 if (!capable(CAP_SETFCAP))
416 } else if (!strncmp(name, XATTR_SECURITY_PREFIX,
417 sizeof(XATTR_SECURITY_PREFIX) - 1) &&
418 !capable(CAP_SYS_ADMIN))
423 /* moved from kernel/sys.c. */
425 * cap_emulate_setxuid() fixes the effective / permitted capabilities of
426 * a process after a call to setuid, setreuid, or setresuid.
428 * 1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
429 * {r,e,s}uid != 0, the permitted and effective capabilities are
432 * 2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
433 * capabilities of the process are cleared.
435 * 3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
436 * capabilities are set to the permitted capabilities.
438 * fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
443 * cevans - New behaviour, Oct '99
444 * A process may, via prctl(), elect to keep its capabilities when it
445 * calls setuid() and switches away from uid==0. Both permitted and
446 * effective sets will be retained.
447 * Without this change, it was impossible for a daemon to drop only some
448 * of its privilege. The call to setuid(!=0) would drop all privileges!
449 * Keeping uid 0 is not an option because uid 0 owns too many vital
451 * Thanks to Olaf Kirch and Peter Benie for spotting this.
453 static inline void cap_emulate_setxuid (int old_ruid, int old_euid,
456 if ((old_ruid == 0 || old_euid == 0 || old_suid == 0) &&
457 (current->uid != 0 && current->euid != 0 && current->suid != 0) &&
458 !current->keep_capabilities) {
459 cap_clear (current->cap_permitted);
460 cap_clear (current->cap_effective);
462 if (old_euid == 0 && current->euid != 0) {
463 cap_clear (current->cap_effective);
465 if (old_euid != 0 && current->euid == 0) {
466 current->cap_effective = current->cap_permitted;
470 int cap_task_post_setuid (uid_t old_ruid, uid_t old_euid, uid_t old_suid,
477 /* Copied from kernel/sys.c:setreuid/setuid/setresuid. */
478 if (!issecure (SECURE_NO_SETUID_FIXUP)) {
479 cap_emulate_setxuid (old_ruid, old_euid, old_suid);
484 uid_t old_fsuid = old_ruid;
486 /* Copied from kernel/sys.c:setfsuid. */
489 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
490 * if not, we might be a bit too harsh here.
493 if (!issecure (SECURE_NO_SETUID_FIXUP)) {
494 if (old_fsuid == 0 && current->fsuid != 0) {
495 current->cap_effective =
497 current->cap_effective);
499 if (old_fsuid != 0 && current->fsuid == 0) {
500 current->cap_effective =
502 current->cap_effective,
503 current->cap_permitted);
515 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
517 * Rationale: code calling task_setscheduler, task_setioprio, and
518 * task_setnice, assumes that
519 * . if capable(cap_sys_nice), then those actions should be allowed
520 * . if not capable(cap_sys_nice), but acting on your own processes,
521 * then those actions should be allowed
522 * This is insufficient now since you can call code without suid, but
523 * yet with increased caps.
524 * So we check for increased caps on the target process.
526 static inline int cap_safe_nice(struct task_struct *p)
528 if (!cap_issubset(p->cap_permitted, current->cap_permitted) &&
529 !__capable(current, CAP_SYS_NICE))
534 int cap_task_setscheduler (struct task_struct *p, int policy,
535 struct sched_param *lp)
537 return cap_safe_nice(p);
540 int cap_task_setioprio (struct task_struct *p, int ioprio)
542 return cap_safe_nice(p);
545 int cap_task_setnice (struct task_struct *p, int nice)
547 return cap_safe_nice(p);
550 int cap_task_kill(struct task_struct *p, struct siginfo *info,
553 if (info != SEND_SIG_NOINFO && (is_si_special(info) || SI_FROMKERNEL(info)))
557 * Running a setuid root program raises your capabilities.
558 * Killing your own setuid root processes was previously
560 * We must preserve legacy signal behavior in this case.
562 if (p->euid == 0 && p->uid == current->uid)
565 /* sigcont is permitted within same session */
566 if (sig == SIGCONT && (task_session_nr(current) == task_session_nr(p)))
571 * Signal sent as a particular user.
572 * Capabilities are ignored. May be wrong, but it's the
573 * only thing we can do at the moment.
574 * Used only by usb drivers?
577 if (cap_issubset(p->cap_permitted, current->cap_permitted))
579 if (capable(CAP_KILL))
585 int cap_task_setscheduler (struct task_struct *p, int policy,
586 struct sched_param *lp)
590 int cap_task_setioprio (struct task_struct *p, int ioprio)
594 int cap_task_setnice (struct task_struct *p, int nice)
598 int cap_task_kill(struct task_struct *p, struct siginfo *info,
605 void cap_task_reparent_to_init (struct task_struct *p)
607 cap_set_init_eff(p->cap_effective);
608 cap_clear(p->cap_inheritable);
609 cap_set_full(p->cap_permitted);
610 p->keep_capabilities = 0;
614 int cap_syslog (int type)
616 if ((type != 3 && type != 10) && !capable(CAP_SYS_ADMIN))
621 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
623 int cap_sys_admin = 0;
625 if (cap_capable(current, CAP_SYS_ADMIN) == 0)
627 return __vm_enough_memory(mm, pages, cap_sys_admin);