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/audit.h>
12 #include <linux/module.h>
13 #include <linux/init.h>
14 #include <linux/kernel.h>
15 #include <linux/security.h>
16 #include <linux/file.h>
18 #include <linux/mman.h>
19 #include <linux/pagemap.h>
20 #include <linux/swap.h>
21 #include <linux/skbuff.h>
22 #include <linux/netlink.h>
23 #include <linux/ptrace.h>
24 #include <linux/xattr.h>
25 #include <linux/hugetlb.h>
26 #include <linux/mount.h>
27 #include <linux/sched.h>
28 #include <linux/prctl.h>
29 #include <linux/securebits.h>
31 int cap_netlink_send(struct sock
*sk
, struct sk_buff
*skb
)
33 NETLINK_CB(skb
).eff_cap
= current
->cap_effective
;
37 int cap_netlink_recv(struct sk_buff
*skb
, int cap
)
39 if (!cap_raised(NETLINK_CB(skb
).eff_cap
, cap
))
44 EXPORT_SYMBOL(cap_netlink_recv
);
47 * NOTE WELL: cap_capable() cannot be used like the kernel's capable()
48 * function. That is, it has the reverse semantics: cap_capable()
49 * returns 0 when a task has a capability, but the kernel's capable()
50 * returns 1 for this case.
52 int cap_capable(struct task_struct
*tsk
, int cap
, int audit
)
54 /* Derived from include/linux/sched.h:capable. */
55 if (cap_raised(tsk
->cap_effective
, cap
))
60 int cap_settime(struct timespec
*ts
, struct timezone
*tz
)
62 if (!capable(CAP_SYS_TIME
))
67 int cap_ptrace_may_access(struct task_struct
*child
, unsigned int mode
)
69 /* Derived from arch/i386/kernel/ptrace.c:sys_ptrace. */
70 if (cap_issubset(child
->cap_permitted
, current
->cap_permitted
))
72 if (capable(CAP_SYS_PTRACE
))
77 int cap_ptrace_traceme(struct task_struct
*parent
)
79 /* Derived from arch/i386/kernel/ptrace.c:sys_ptrace. */
80 if (cap_issubset(current
->cap_permitted
, parent
->cap_permitted
))
82 if (has_capability(parent
, CAP_SYS_PTRACE
))
87 int cap_capget (struct task_struct
*target
, kernel_cap_t
*effective
,
88 kernel_cap_t
*inheritable
, kernel_cap_t
*permitted
)
90 /* Derived from kernel/capability.c:sys_capget. */
91 *effective
= target
->cap_effective
;
92 *inheritable
= target
->cap_inheritable
;
93 *permitted
= target
->cap_permitted
;
97 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
99 static inline int cap_block_setpcap(struct task_struct
*target
)
102 * No support for remote process capability manipulation with
103 * filesystem capability support.
105 return (target
!= current
);
108 static inline int cap_inh_is_capped(void)
111 * Return 1 if changes to the inheritable set are limited
112 * to the old permitted set. That is, if the current task
113 * does *not* possess the CAP_SETPCAP capability.
115 return (cap_capable(current
, CAP_SETPCAP
, SECURITY_CAP_AUDIT
) != 0);
118 static inline int cap_limit_ptraced_target(void) { return 1; }
120 #else /* ie., ndef CONFIG_SECURITY_FILE_CAPABILITIES */
122 static inline int cap_block_setpcap(struct task_struct
*t
) { return 0; }
123 static inline int cap_inh_is_capped(void) { return 1; }
124 static inline int cap_limit_ptraced_target(void)
126 return !capable(CAP_SETPCAP
);
129 #endif /* def CONFIG_SECURITY_FILE_CAPABILITIES */
131 int cap_capset_check (struct task_struct
*target
, kernel_cap_t
*effective
,
132 kernel_cap_t
*inheritable
, kernel_cap_t
*permitted
)
134 if (cap_block_setpcap(target
)) {
137 if (cap_inh_is_capped()
138 && !cap_issubset(*inheritable
,
139 cap_combine(target
->cap_inheritable
,
140 current
->cap_permitted
))) {
141 /* incapable of using this inheritable set */
144 if (!cap_issubset(*inheritable
,
145 cap_combine(target
->cap_inheritable
,
146 current
->cap_bset
))) {
147 /* no new pI capabilities outside bounding set */
151 /* verify restrictions on target's new Permitted set */
152 if (!cap_issubset (*permitted
,
153 cap_combine (target
->cap_permitted
,
154 current
->cap_permitted
))) {
158 /* verify the _new_Effective_ is a subset of the _new_Permitted_ */
159 if (!cap_issubset (*effective
, *permitted
)) {
166 void cap_capset_set (struct task_struct
*target
, kernel_cap_t
*effective
,
167 kernel_cap_t
*inheritable
, kernel_cap_t
*permitted
)
169 target
->cap_effective
= *effective
;
170 target
->cap_inheritable
= *inheritable
;
171 target
->cap_permitted
= *permitted
;
174 static inline void bprm_clear_caps(struct linux_binprm
*bprm
)
176 cap_clear(bprm
->cap_post_exec_permitted
);
177 bprm
->cap_effective
= false;
180 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
182 int cap_inode_need_killpriv(struct dentry
*dentry
)
184 struct inode
*inode
= dentry
->d_inode
;
187 if (!inode
->i_op
|| !inode
->i_op
->getxattr
)
190 error
= inode
->i_op
->getxattr(dentry
, XATTR_NAME_CAPS
, NULL
, 0);
196 int cap_inode_killpriv(struct dentry
*dentry
)
198 struct inode
*inode
= dentry
->d_inode
;
200 if (!inode
->i_op
|| !inode
->i_op
->removexattr
)
203 return inode
->i_op
->removexattr(dentry
, XATTR_NAME_CAPS
);
206 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data
*caps
,
207 struct linux_binprm
*bprm
)
212 if (caps
->magic_etc
& VFS_CAP_FLAGS_EFFECTIVE
)
213 bprm
->cap_effective
= true;
215 bprm
->cap_effective
= false;
217 CAP_FOR_EACH_U32(i
) {
218 __u32 permitted
= caps
->permitted
.cap
[i
];
219 __u32 inheritable
= caps
->inheritable
.cap
[i
];
222 * pP' = (X & fP) | (pI & fI)
224 bprm
->cap_post_exec_permitted
.cap
[i
] =
225 (current
->cap_bset
.cap
[i
] & permitted
) |
226 (current
->cap_inheritable
.cap
[i
] & inheritable
);
228 if (permitted
& ~bprm
->cap_post_exec_permitted
.cap
[i
]) {
230 * insufficient to execute correctly
237 * For legacy apps, with no internal support for recognizing they
238 * do not have enough capabilities, we return an error if they are
239 * missing some "forced" (aka file-permitted) capabilities.
241 return bprm
->cap_effective
? ret
: 0;
244 int get_vfs_caps_from_disk(const struct dentry
*dentry
, struct cpu_vfs_cap_data
*cpu_caps
)
246 struct inode
*inode
= dentry
->d_inode
;
250 struct vfs_cap_data caps
;
252 memset(cpu_caps
, 0, sizeof(struct cpu_vfs_cap_data
));
254 if (!inode
|| !inode
->i_op
|| !inode
->i_op
->getxattr
)
257 size
= inode
->i_op
->getxattr((struct dentry
*)dentry
, XATTR_NAME_CAPS
, &caps
,
259 if (size
== -ENODATA
|| size
== -EOPNOTSUPP
) {
260 /* no data, that's ok */
266 if (size
< sizeof(magic_etc
))
269 cpu_caps
->magic_etc
= magic_etc
= le32_to_cpu(caps
.magic_etc
);
271 switch ((magic_etc
& VFS_CAP_REVISION_MASK
)) {
272 case VFS_CAP_REVISION_1
:
273 if (size
!= XATTR_CAPS_SZ_1
)
275 tocopy
= VFS_CAP_U32_1
;
277 case VFS_CAP_REVISION_2
:
278 if (size
!= XATTR_CAPS_SZ_2
)
280 tocopy
= VFS_CAP_U32_2
;
286 CAP_FOR_EACH_U32(i
) {
289 cpu_caps
->permitted
.cap
[i
] = le32_to_cpu(caps
.data
[i
].permitted
);
290 cpu_caps
->inheritable
.cap
[i
] = le32_to_cpu(caps
.data
[i
].inheritable
);
295 /* Locate any VFS capabilities: */
296 static int get_file_caps(struct linux_binprm
*bprm
)
298 struct dentry
*dentry
;
300 struct cpu_vfs_cap_data vcaps
;
302 bprm_clear_caps(bprm
);
304 if (!file_caps_enabled
)
307 if (bprm
->file
->f_vfsmnt
->mnt_flags
& MNT_NOSUID
)
310 dentry
= dget(bprm
->file
->f_dentry
);
312 rc
= get_vfs_caps_from_disk(dentry
, &vcaps
);
315 printk(KERN_NOTICE
"%s: get_vfs_caps_from_disk returned %d for %s\n",
316 __func__
, rc
, bprm
->filename
);
317 else if (rc
== -ENODATA
)
322 rc
= bprm_caps_from_vfs_caps(&vcaps
, bprm
);
327 bprm_clear_caps(bprm
);
333 int cap_inode_need_killpriv(struct dentry
*dentry
)
338 int cap_inode_killpriv(struct dentry
*dentry
)
343 static inline int get_file_caps(struct linux_binprm
*bprm
)
345 bprm_clear_caps(bprm
);
350 int cap_bprm_set_security (struct linux_binprm
*bprm
)
354 ret
= get_file_caps(bprm
);
356 if (!issecure(SECURE_NOROOT
)) {
358 * To support inheritance of root-permissions and suid-root
359 * executables under compatibility mode, we override the
360 * capability sets for the file.
362 * If only the real uid is 0, we do not set the effective
365 if (bprm
->e_uid
== 0 || current
->uid
== 0) {
366 /* pP' = (cap_bset & ~0) | (pI & ~0) */
367 bprm
->cap_post_exec_permitted
= cap_combine(
368 current
->cap_bset
, current
->cap_inheritable
370 bprm
->cap_effective
= (bprm
->e_uid
== 0);
378 void cap_bprm_apply_creds (struct linux_binprm
*bprm
, int unsafe
)
380 kernel_cap_t pP
= current
->cap_permitted
;
381 kernel_cap_t pE
= current
->cap_effective
;
383 if (bprm
->e_uid
!= current
->uid
|| bprm
->e_gid
!= current
->gid
||
384 !cap_issubset(bprm
->cap_post_exec_permitted
,
385 current
->cap_permitted
)) {
386 set_dumpable(current
->mm
, suid_dumpable
);
387 current
->pdeath_signal
= 0;
389 if (unsafe
& ~LSM_UNSAFE_PTRACE_CAP
) {
390 if (!capable(CAP_SETUID
)) {
391 bprm
->e_uid
= current
->uid
;
392 bprm
->e_gid
= current
->gid
;
394 if (cap_limit_ptraced_target()) {
395 bprm
->cap_post_exec_permitted
= cap_intersect(
396 bprm
->cap_post_exec_permitted
,
397 current
->cap_permitted
);
402 current
->suid
= current
->euid
= current
->fsuid
= bprm
->e_uid
;
403 current
->sgid
= current
->egid
= current
->fsgid
= bprm
->e_gid
;
405 /* For init, we want to retain the capabilities set
406 * in the init_task struct. Thus we skip the usual
407 * capability rules */
408 if (!is_global_init(current
)) {
409 current
->cap_permitted
= bprm
->cap_post_exec_permitted
;
410 if (bprm
->cap_effective
)
411 current
->cap_effective
= bprm
->cap_post_exec_permitted
;
413 cap_clear(current
->cap_effective
);
417 * Audit candidate if current->cap_effective is set
419 * We do not bother to audit if 3 things are true:
420 * 1) cap_effective has all caps
422 * 3) root is supposed to have all caps (SECURE_NOROOT)
423 * Since this is just a normal root execing a process.
425 * Number 1 above might fail if you don't have a full bset, but I think
426 * that is interesting information to audit.
428 if (!cap_isclear(current
->cap_effective
)) {
429 if (!cap_issubset(CAP_FULL_SET
, current
->cap_effective
) ||
430 (bprm
->e_uid
!= 0) || (current
->uid
!= 0) ||
431 issecure(SECURE_NOROOT
))
432 audit_log_bprm_fcaps(bprm
, &pP
, &pE
);
435 current
->securebits
&= ~issecure_mask(SECURE_KEEP_CAPS
);
438 int cap_bprm_secureexec (struct linux_binprm
*bprm
)
440 if (current
->uid
!= 0) {
441 if (bprm
->cap_effective
)
443 if (!cap_isclear(bprm
->cap_post_exec_permitted
))
447 return (current
->euid
!= current
->uid
||
448 current
->egid
!= current
->gid
);
451 int cap_inode_setxattr(struct dentry
*dentry
, const char *name
,
452 const void *value
, size_t size
, int flags
)
454 if (!strcmp(name
, XATTR_NAME_CAPS
)) {
455 if (!capable(CAP_SETFCAP
))
458 } else if (!strncmp(name
, XATTR_SECURITY_PREFIX
,
459 sizeof(XATTR_SECURITY_PREFIX
) - 1) &&
460 !capable(CAP_SYS_ADMIN
))
465 int cap_inode_removexattr(struct dentry
*dentry
, const char *name
)
467 if (!strcmp(name
, XATTR_NAME_CAPS
)) {
468 if (!capable(CAP_SETFCAP
))
471 } else if (!strncmp(name
, XATTR_SECURITY_PREFIX
,
472 sizeof(XATTR_SECURITY_PREFIX
) - 1) &&
473 !capable(CAP_SYS_ADMIN
))
478 /* moved from kernel/sys.c. */
480 * cap_emulate_setxuid() fixes the effective / permitted capabilities of
481 * a process after a call to setuid, setreuid, or setresuid.
483 * 1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
484 * {r,e,s}uid != 0, the permitted and effective capabilities are
487 * 2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
488 * capabilities of the process are cleared.
490 * 3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
491 * capabilities are set to the permitted capabilities.
493 * fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
498 * cevans - New behaviour, Oct '99
499 * A process may, via prctl(), elect to keep its capabilities when it
500 * calls setuid() and switches away from uid==0. Both permitted and
501 * effective sets will be retained.
502 * Without this change, it was impossible for a daemon to drop only some
503 * of its privilege. The call to setuid(!=0) would drop all privileges!
504 * Keeping uid 0 is not an option because uid 0 owns too many vital
506 * Thanks to Olaf Kirch and Peter Benie for spotting this.
508 static inline void cap_emulate_setxuid (int old_ruid
, int old_euid
,
511 if ((old_ruid
== 0 || old_euid
== 0 || old_suid
== 0) &&
512 (current
->uid
!= 0 && current
->euid
!= 0 && current
->suid
!= 0) &&
513 !issecure(SECURE_KEEP_CAPS
)) {
514 cap_clear (current
->cap_permitted
);
515 cap_clear (current
->cap_effective
);
517 if (old_euid
== 0 && current
->euid
!= 0) {
518 cap_clear (current
->cap_effective
);
520 if (old_euid
!= 0 && current
->euid
== 0) {
521 current
->cap_effective
= current
->cap_permitted
;
525 int cap_task_post_setuid (uid_t old_ruid
, uid_t old_euid
, uid_t old_suid
,
532 /* Copied from kernel/sys.c:setreuid/setuid/setresuid. */
533 if (!issecure (SECURE_NO_SETUID_FIXUP
)) {
534 cap_emulate_setxuid (old_ruid
, old_euid
, old_suid
);
539 uid_t old_fsuid
= old_ruid
;
541 /* Copied from kernel/sys.c:setfsuid. */
544 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
545 * if not, we might be a bit too harsh here.
548 if (!issecure (SECURE_NO_SETUID_FIXUP
)) {
549 if (old_fsuid
== 0 && current
->fsuid
!= 0) {
550 current
->cap_effective
=
552 current
->cap_effective
);
554 if (old_fsuid
!= 0 && current
->fsuid
== 0) {
555 current
->cap_effective
=
557 current
->cap_effective
,
558 current
->cap_permitted
);
570 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
572 * Rationale: code calling task_setscheduler, task_setioprio, and
573 * task_setnice, assumes that
574 * . if capable(cap_sys_nice), then those actions should be allowed
575 * . if not capable(cap_sys_nice), but acting on your own processes,
576 * then those actions should be allowed
577 * This is insufficient now since you can call code without suid, but
578 * yet with increased caps.
579 * So we check for increased caps on the target process.
581 static int cap_safe_nice(struct task_struct
*p
)
583 if (!cap_issubset(p
->cap_permitted
, current
->cap_permitted
) &&
584 !capable(CAP_SYS_NICE
))
589 int cap_task_setscheduler (struct task_struct
*p
, int policy
,
590 struct sched_param
*lp
)
592 return cap_safe_nice(p
);
595 int cap_task_setioprio (struct task_struct
*p
, int ioprio
)
597 return cap_safe_nice(p
);
600 int cap_task_setnice (struct task_struct
*p
, int nice
)
602 return cap_safe_nice(p
);
606 * called from kernel/sys.c for prctl(PR_CABSET_DROP)
607 * done without task_capability_lock() because it introduces
608 * no new races - i.e. only another task doing capget() on
609 * this task could get inconsistent info. There can be no
610 * racing writer bc a task can only change its own caps.
612 static long cap_prctl_drop(unsigned long cap
)
614 if (!capable(CAP_SETPCAP
))
618 cap_lower(current
->cap_bset
, cap
);
623 int cap_task_setscheduler (struct task_struct
*p
, int policy
,
624 struct sched_param
*lp
)
628 int cap_task_setioprio (struct task_struct
*p
, int ioprio
)
632 int cap_task_setnice (struct task_struct
*p
, int nice
)
638 int cap_task_prctl(int option
, unsigned long arg2
, unsigned long arg3
,
639 unsigned long arg4
, unsigned long arg5
, long *rc_p
)
644 case PR_CAPBSET_READ
:
645 if (!cap_valid(arg2
))
648 error
= !!cap_raised(current
->cap_bset
, arg2
);
650 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
651 case PR_CAPBSET_DROP
:
652 error
= cap_prctl_drop(arg2
);
656 * The next four prctl's remain to assist with transitioning a
657 * system from legacy UID=0 based privilege (when filesystem
658 * capabilities are not in use) to a system using filesystem
659 * capabilities only - as the POSIX.1e draft intended.
663 * PR_SET_SECUREBITS =
664 * issecure_mask(SECURE_KEEP_CAPS_LOCKED)
665 * | issecure_mask(SECURE_NOROOT)
666 * | issecure_mask(SECURE_NOROOT_LOCKED)
667 * | issecure_mask(SECURE_NO_SETUID_FIXUP)
668 * | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
670 * will ensure that the current process and all of its
671 * children will be locked into a pure
672 * capability-based-privilege environment.
674 case PR_SET_SECUREBITS
:
675 if ((((current
->securebits
& SECURE_ALL_LOCKS
) >> 1)
676 & (current
->securebits
^ arg2
)) /*[1]*/
677 || ((current
->securebits
& SECURE_ALL_LOCKS
679 || (arg2
& ~(SECURE_ALL_LOCKS
| SECURE_ALL_BITS
)) /*[3]*/
680 || (cap_capable(current
, CAP_SETPCAP
, SECURITY_CAP_AUDIT
) != 0)) { /*[4]*/
682 * [1] no changing of bits that are locked
683 * [2] no unlocking of locks
684 * [3] no setting of unsupported bits
685 * [4] doing anything requires privilege (go read about
686 * the "sendmail capabilities bug")
688 error
= -EPERM
; /* cannot change a locked bit */
690 current
->securebits
= arg2
;
693 case PR_GET_SECUREBITS
:
694 error
= current
->securebits
;
697 #endif /* def CONFIG_SECURITY_FILE_CAPABILITIES */
699 case PR_GET_KEEPCAPS
:
700 if (issecure(SECURE_KEEP_CAPS
))
703 case PR_SET_KEEPCAPS
:
704 if (arg2
> 1) /* Note, we rely on arg2 being unsigned here */
706 else if (issecure(SECURE_KEEP_CAPS_LOCKED
))
709 current
->securebits
|= issecure_mask(SECURE_KEEP_CAPS
);
711 current
->securebits
&=
712 ~issecure_mask(SECURE_KEEP_CAPS
);
716 /* No functionality available - continue with default */
720 /* Functionality provided */
725 void cap_task_reparent_to_init (struct task_struct
*p
)
727 cap_set_init_eff(p
->cap_effective
);
728 cap_clear(p
->cap_inheritable
);
729 cap_set_full(p
->cap_permitted
);
730 p
->securebits
= SECUREBITS_DEFAULT
;
734 int cap_syslog (int type
)
736 if ((type
!= 3 && type
!= 10) && !capable(CAP_SYS_ADMIN
))
741 int cap_vm_enough_memory(struct mm_struct
*mm
, long pages
)
743 int cap_sys_admin
= 0;
745 if (cap_capable(current
, CAP_SYS_ADMIN
, SECURITY_CAP_NOAUDIT
) == 0)
747 return __vm_enough_memory(mm
, pages
, cap_sys_admin
);