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>
27 #include <linux/prctl.h>
28 #include <linux/securebits.h>
30 int cap_netlink_send(struct sock *sk, struct sk_buff *skb)
32 NETLINK_CB(skb).eff_cap = current->cap_effective;
36 int cap_netlink_recv(struct sk_buff *skb, int cap)
38 if (!cap_raised(NETLINK_CB(skb).eff_cap, cap))
43 EXPORT_SYMBOL(cap_netlink_recv);
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.
51 int cap_capable (struct task_struct *tsk, int cap)
53 /* Derived from include/linux/sched.h:capable. */
54 if (cap_raised(tsk->cap_effective, cap))
59 int cap_settime(struct timespec *ts, struct timezone *tz)
61 if (!capable(CAP_SYS_TIME))
66 int cap_ptrace (struct task_struct *parent, struct task_struct *child,
69 /* Derived from arch/i386/kernel/ptrace.c:sys_ptrace. */
70 if (!cap_issubset(child->cap_permitted, parent->cap_permitted) &&
71 !__capable(parent, CAP_SYS_PTRACE))
76 int cap_capget (struct task_struct *target, kernel_cap_t *effective,
77 kernel_cap_t *inheritable, kernel_cap_t *permitted)
79 /* Derived from kernel/capability.c:sys_capget. */
80 *effective = target->cap_effective;
81 *inheritable = target->cap_inheritable;
82 *permitted = target->cap_permitted;
86 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
88 static inline int cap_block_setpcap(struct task_struct *target)
91 * No support for remote process capability manipulation with
92 * filesystem capability support.
94 return (target != current);
97 static inline int cap_inh_is_capped(void)
100 * Return 1 if changes to the inheritable set are limited
101 * to the old permitted set. That is, if the current task
102 * does *not* possess the CAP_SETPCAP capability.
104 return (cap_capable(current, CAP_SETPCAP) != 0);
107 static inline int cap_limit_ptraced_target(void) { return 1; }
109 #else /* ie., ndef CONFIG_SECURITY_FILE_CAPABILITIES */
111 static inline int cap_block_setpcap(struct task_struct *t) { return 0; }
112 static inline int cap_inh_is_capped(void) { return 1; }
113 static inline int cap_limit_ptraced_target(void)
115 return !capable(CAP_SETPCAP);
118 #endif /* def CONFIG_SECURITY_FILE_CAPABILITIES */
120 int cap_capset_check (struct task_struct *target, kernel_cap_t *effective,
121 kernel_cap_t *inheritable, kernel_cap_t *permitted)
123 if (cap_block_setpcap(target)) {
126 if (cap_inh_is_capped()
127 && !cap_issubset(*inheritable,
128 cap_combine(target->cap_inheritable,
129 current->cap_permitted))) {
130 /* incapable of using this inheritable set */
133 if (!cap_issubset(*inheritable,
134 cap_combine(target->cap_inheritable,
135 current->cap_bset))) {
136 /* no new pI capabilities outside bounding set */
140 /* verify restrictions on target's new Permitted set */
141 if (!cap_issubset (*permitted,
142 cap_combine (target->cap_permitted,
143 current->cap_permitted))) {
147 /* verify the _new_Effective_ is a subset of the _new_Permitted_ */
148 if (!cap_issubset (*effective, *permitted)) {
155 void cap_capset_set (struct task_struct *target, kernel_cap_t *effective,
156 kernel_cap_t *inheritable, kernel_cap_t *permitted)
158 target->cap_effective = *effective;
159 target->cap_inheritable = *inheritable;
160 target->cap_permitted = *permitted;
163 static inline void bprm_clear_caps(struct linux_binprm *bprm)
165 cap_clear(bprm->cap_post_exec_permitted);
166 bprm->cap_effective = false;
169 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
171 int cap_inode_need_killpriv(struct dentry *dentry)
173 struct inode *inode = dentry->d_inode;
176 if (!inode->i_op || !inode->i_op->getxattr)
179 error = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, NULL, 0);
185 int cap_inode_killpriv(struct dentry *dentry)
187 struct inode *inode = dentry->d_inode;
189 if (!inode->i_op || !inode->i_op->removexattr)
192 return inode->i_op->removexattr(dentry, XATTR_NAME_CAPS);
195 static inline int cap_from_disk(struct vfs_cap_data *caps,
196 struct linux_binprm *bprm, unsigned size)
202 if (size < sizeof(magic_etc))
205 magic_etc = le32_to_cpu(caps->magic_etc);
207 switch ((magic_etc & VFS_CAP_REVISION_MASK)) {
208 case VFS_CAP_REVISION_1:
209 if (size != XATTR_CAPS_SZ_1)
211 tocopy = VFS_CAP_U32_1;
213 case VFS_CAP_REVISION_2:
214 if (size != XATTR_CAPS_SZ_2)
216 tocopy = VFS_CAP_U32_2;
222 if (magic_etc & VFS_CAP_FLAGS_EFFECTIVE) {
223 bprm->cap_effective = true;
225 bprm->cap_effective = false;
230 CAP_FOR_EACH_U32(i) {
235 * Legacy capability sets have no upper bits
237 bprm->cap_post_exec_permitted.cap[i] = 0;
241 * pP' = (X & fP) | (pI & fI)
243 value_cpu = le32_to_cpu(caps->data[i].permitted);
244 bprm->cap_post_exec_permitted.cap[i] =
245 (current->cap_bset.cap[i] & value_cpu) |
246 (current->cap_inheritable.cap[i] &
247 le32_to_cpu(caps->data[i].inheritable));
248 if (value_cpu & ~bprm->cap_post_exec_permitted.cap[i]) {
250 * insufficient to execute correctly
257 * For legacy apps, with no internal support for recognizing they
258 * do not have enough capabilities, we return an error if they are
259 * missing some "forced" (aka file-permitted) capabilities.
261 return bprm->cap_effective ? ret : 0;
264 /* Locate any VFS capabilities: */
265 static int get_file_caps(struct linux_binprm *bprm)
267 struct dentry *dentry;
269 struct vfs_cap_data vcaps;
272 if (bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID) {
273 bprm_clear_caps(bprm);
277 dentry = dget(bprm->file->f_dentry);
278 inode = dentry->d_inode;
279 if (!inode->i_op || !inode->i_op->getxattr)
282 rc = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, &vcaps,
284 if (rc == -ENODATA || rc == -EOPNOTSUPP) {
285 /* no data, that's ok */
292 rc = cap_from_disk(&vcaps, bprm, rc);
294 printk(KERN_NOTICE "%s: cap_from_disk returned %d for %s\n",
295 __func__, rc, bprm->filename);
300 bprm_clear_caps(bprm);
306 int cap_inode_need_killpriv(struct dentry *dentry)
311 int cap_inode_killpriv(struct dentry *dentry)
316 static inline int get_file_caps(struct linux_binprm *bprm)
318 bprm_clear_caps(bprm);
323 int cap_bprm_set_security (struct linux_binprm *bprm)
327 ret = get_file_caps(bprm);
329 if (!issecure(SECURE_NOROOT)) {
331 * To support inheritance of root-permissions and suid-root
332 * executables under compatibility mode, we override the
333 * capability sets for the file.
335 * If only the real uid is 0, we do not set the effective
338 if (bprm->e_uid == 0 || current->uid == 0) {
339 /* pP' = (cap_bset & ~0) | (pI & ~0) */
340 bprm->cap_post_exec_permitted = cap_combine(
341 current->cap_bset, current->cap_inheritable
343 bprm->cap_effective = (bprm->e_uid == 0);
351 void cap_bprm_apply_creds (struct linux_binprm *bprm, int unsafe)
353 if (bprm->e_uid != current->uid || bprm->e_gid != current->gid ||
354 !cap_issubset(bprm->cap_post_exec_permitted,
355 current->cap_permitted)) {
356 set_dumpable(current->mm, suid_dumpable);
357 current->pdeath_signal = 0;
359 if (unsafe & ~LSM_UNSAFE_PTRACE_CAP) {
360 if (!capable(CAP_SETUID)) {
361 bprm->e_uid = current->uid;
362 bprm->e_gid = current->gid;
364 if (cap_limit_ptraced_target()) {
365 bprm->cap_post_exec_permitted = cap_intersect(
366 bprm->cap_post_exec_permitted,
367 current->cap_permitted);
372 current->suid = current->euid = current->fsuid = bprm->e_uid;
373 current->sgid = current->egid = current->fsgid = bprm->e_gid;
375 /* For init, we want to retain the capabilities set
376 * in the init_task struct. Thus we skip the usual
377 * capability rules */
378 if (!is_global_init(current)) {
379 current->cap_permitted = bprm->cap_post_exec_permitted;
380 if (bprm->cap_effective)
381 current->cap_effective = bprm->cap_post_exec_permitted;
383 cap_clear(current->cap_effective);
386 /* AUD: Audit candidate if current->cap_effective is set */
388 current->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
391 int cap_bprm_secureexec (struct linux_binprm *bprm)
393 if (current->uid != 0) {
394 if (bprm->cap_effective)
396 if (!cap_isclear(bprm->cap_post_exec_permitted))
400 return (current->euid != current->uid ||
401 current->egid != current->gid);
404 int cap_inode_setxattr(struct dentry *dentry, const char *name,
405 const void *value, size_t size, int flags)
407 if (!strcmp(name, XATTR_NAME_CAPS)) {
408 if (!capable(CAP_SETFCAP))
411 } else if (!strncmp(name, XATTR_SECURITY_PREFIX,
412 sizeof(XATTR_SECURITY_PREFIX) - 1) &&
413 !capable(CAP_SYS_ADMIN))
418 int cap_inode_removexattr(struct dentry *dentry, const char *name)
420 if (!strcmp(name, XATTR_NAME_CAPS)) {
421 if (!capable(CAP_SETFCAP))
424 } else if (!strncmp(name, XATTR_SECURITY_PREFIX,
425 sizeof(XATTR_SECURITY_PREFIX) - 1) &&
426 !capable(CAP_SYS_ADMIN))
431 /* moved from kernel/sys.c. */
433 * cap_emulate_setxuid() fixes the effective / permitted capabilities of
434 * a process after a call to setuid, setreuid, or setresuid.
436 * 1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
437 * {r,e,s}uid != 0, the permitted and effective capabilities are
440 * 2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
441 * capabilities of the process are cleared.
443 * 3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
444 * capabilities are set to the permitted capabilities.
446 * fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
451 * cevans - New behaviour, Oct '99
452 * A process may, via prctl(), elect to keep its capabilities when it
453 * calls setuid() and switches away from uid==0. Both permitted and
454 * effective sets will be retained.
455 * Without this change, it was impossible for a daemon to drop only some
456 * of its privilege. The call to setuid(!=0) would drop all privileges!
457 * Keeping uid 0 is not an option because uid 0 owns too many vital
459 * Thanks to Olaf Kirch and Peter Benie for spotting this.
461 static inline void cap_emulate_setxuid (int old_ruid, int old_euid,
464 if ((old_ruid == 0 || old_euid == 0 || old_suid == 0) &&
465 (current->uid != 0 && current->euid != 0 && current->suid != 0) &&
466 !issecure(SECURE_KEEP_CAPS)) {
467 cap_clear (current->cap_permitted);
468 cap_clear (current->cap_effective);
470 if (old_euid == 0 && current->euid != 0) {
471 cap_clear (current->cap_effective);
473 if (old_euid != 0 && current->euid == 0) {
474 current->cap_effective = current->cap_permitted;
478 int cap_task_post_setuid (uid_t old_ruid, uid_t old_euid, uid_t old_suid,
485 /* Copied from kernel/sys.c:setreuid/setuid/setresuid. */
486 if (!issecure (SECURE_NO_SETUID_FIXUP)) {
487 cap_emulate_setxuid (old_ruid, old_euid, old_suid);
492 uid_t old_fsuid = old_ruid;
494 /* Copied from kernel/sys.c:setfsuid. */
497 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
498 * if not, we might be a bit too harsh here.
501 if (!issecure (SECURE_NO_SETUID_FIXUP)) {
502 if (old_fsuid == 0 && current->fsuid != 0) {
503 current->cap_effective =
505 current->cap_effective);
507 if (old_fsuid != 0 && current->fsuid == 0) {
508 current->cap_effective =
510 current->cap_effective,
511 current->cap_permitted);
523 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
525 * Rationale: code calling task_setscheduler, task_setioprio, and
526 * task_setnice, assumes that
527 * . if capable(cap_sys_nice), then those actions should be allowed
528 * . if not capable(cap_sys_nice), but acting on your own processes,
529 * then those actions should be allowed
530 * This is insufficient now since you can call code without suid, but
531 * yet with increased caps.
532 * So we check for increased caps on the target process.
534 static inline int cap_safe_nice(struct task_struct *p)
536 if (!cap_issubset(p->cap_permitted, current->cap_permitted) &&
537 !__capable(current, CAP_SYS_NICE))
542 int cap_task_setscheduler (struct task_struct *p, int policy,
543 struct sched_param *lp)
545 return cap_safe_nice(p);
548 int cap_task_setioprio (struct task_struct *p, int ioprio)
550 return cap_safe_nice(p);
553 int cap_task_setnice (struct task_struct *p, int nice)
555 return cap_safe_nice(p);
559 * called from kernel/sys.c for prctl(PR_CABSET_DROP)
560 * done without task_capability_lock() because it introduces
561 * no new races - i.e. only another task doing capget() on
562 * this task could get inconsistent info. There can be no
563 * racing writer bc a task can only change its own caps.
565 static long cap_prctl_drop(unsigned long cap)
567 if (!capable(CAP_SETPCAP))
571 cap_lower(current->cap_bset, cap);
576 int cap_task_setscheduler (struct task_struct *p, int policy,
577 struct sched_param *lp)
581 int cap_task_setioprio (struct task_struct *p, int ioprio)
585 int cap_task_setnice (struct task_struct *p, int nice)
591 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
592 unsigned long arg4, unsigned long arg5, long *rc_p)
597 case PR_CAPBSET_READ:
598 if (!cap_valid(arg2))
601 error = !!cap_raised(current->cap_bset, arg2);
603 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
604 case PR_CAPBSET_DROP:
605 error = cap_prctl_drop(arg2);
609 * The next four prctl's remain to assist with transitioning a
610 * system from legacy UID=0 based privilege (when filesystem
611 * capabilities are not in use) to a system using filesystem
612 * capabilities only - as the POSIX.1e draft intended.
616 * PR_SET_SECUREBITS =
617 * issecure_mask(SECURE_KEEP_CAPS_LOCKED)
618 * | issecure_mask(SECURE_NOROOT)
619 * | issecure_mask(SECURE_NOROOT_LOCKED)
620 * | issecure_mask(SECURE_NO_SETUID_FIXUP)
621 * | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
623 * will ensure that the current process and all of its
624 * children will be locked into a pure
625 * capability-based-privilege environment.
627 case PR_SET_SECUREBITS:
628 if ((((current->securebits & SECURE_ALL_LOCKS) >> 1)
629 & (current->securebits ^ arg2)) /*[1]*/
630 || ((current->securebits & SECURE_ALL_LOCKS
632 || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS)) /*[3]*/
633 || (cap_capable(current, CAP_SETPCAP) != 0)) { /*[4]*/
635 * [1] no changing of bits that are locked
636 * [2] no unlocking of locks
637 * [3] no setting of unsupported bits
638 * [4] doing anything requires privilege (go read about
639 * the "sendmail capabilities bug")
641 error = -EPERM; /* cannot change a locked bit */
643 current->securebits = arg2;
646 case PR_GET_SECUREBITS:
647 error = current->securebits;
650 #endif /* def CONFIG_SECURITY_FILE_CAPABILITIES */
652 case PR_GET_KEEPCAPS:
653 if (issecure(SECURE_KEEP_CAPS))
656 case PR_SET_KEEPCAPS:
657 if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
659 else if (issecure(SECURE_KEEP_CAPS_LOCKED))
662 current->securebits |= issecure_mask(SECURE_KEEP_CAPS);
664 current->securebits &=
665 ~issecure_mask(SECURE_KEEP_CAPS);
669 /* No functionality available - continue with default */
673 /* Functionality provided */
678 void cap_task_reparent_to_init (struct task_struct *p)
680 cap_set_init_eff(p->cap_effective);
681 cap_clear(p->cap_inheritable);
682 cap_set_full(p->cap_permitted);
683 p->securebits = SECUREBITS_DEFAULT;
687 int cap_syslog (int type)
689 if ((type != 3 && type != 10) && !capable(CAP_SYS_ADMIN))
694 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
696 int cap_sys_admin = 0;
698 if (cap_capable(current, CAP_SYS_ADMIN) == 0)
700 return __vm_enough_memory(mm, pages, cap_sys_admin);