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();
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
->cred
->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
->cred
->cap_permitted
,
71 current
->cred
->cap_permitted
))
73 if (capable(CAP_SYS_PTRACE
))
78 int cap_ptrace_traceme(struct task_struct
*parent
)
80 if (cap_issubset(current
->cred
->cap_permitted
,
81 parent
->cred
->cap_permitted
))
83 if (has_capability(parent
, CAP_SYS_PTRACE
))
88 int cap_capget (struct task_struct
*target
, kernel_cap_t
*effective
,
89 kernel_cap_t
*inheritable
, kernel_cap_t
*permitted
)
91 struct cred
*cred
= target
->cred
;
93 /* Derived from kernel/capability.c:sys_capget. */
94 *effective
= cred
->cap_effective
;
95 *inheritable
= cred
->cap_inheritable
;
96 *permitted
= cred
->cap_permitted
;
100 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
102 static inline int cap_inh_is_capped(void)
105 * Return 1 if changes to the inheritable set are limited
106 * to the old permitted set. That is, if the current task
107 * does *not* possess the CAP_SETPCAP capability.
109 return (cap_capable(current
, CAP_SETPCAP
, SECURITY_CAP_AUDIT
) != 0);
112 static inline int cap_limit_ptraced_target(void) { return 1; }
114 #else /* ie., ndef CONFIG_SECURITY_FILE_CAPABILITIES */
116 static inline int cap_inh_is_capped(void) { return 1; }
117 static inline int cap_limit_ptraced_target(void)
119 return !capable(CAP_SETPCAP
);
122 #endif /* def CONFIG_SECURITY_FILE_CAPABILITIES */
124 int cap_capset_check(const kernel_cap_t
*effective
,
125 const kernel_cap_t
*inheritable
,
126 const kernel_cap_t
*permitted
)
128 const struct cred
*cred
= current
->cred
;
130 if (cap_inh_is_capped()
131 && !cap_issubset(*inheritable
,
132 cap_combine(cred
->cap_inheritable
,
133 cred
->cap_permitted
))) {
134 /* incapable of using this inheritable set */
137 if (!cap_issubset(*inheritable
,
138 cap_combine(cred
->cap_inheritable
,
140 /* no new pI capabilities outside bounding set */
144 /* verify restrictions on target's new Permitted set */
145 if (!cap_issubset (*permitted
,
146 cap_combine (cred
->cap_permitted
,
147 cred
->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(const kernel_cap_t
*effective
,
160 const kernel_cap_t
*inheritable
,
161 const kernel_cap_t
*permitted
)
163 struct cred
*cred
= current
->cred
;
165 cred
->cap_effective
= *effective
;
166 cred
->cap_inheritable
= *inheritable
;
167 cred
->cap_permitted
= *permitted
;
170 static inline void bprm_clear_caps(struct linux_binprm
*bprm
)
172 cap_clear(bprm
->cap_post_exec_permitted
);
173 bprm
->cap_effective
= false;
176 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
178 int cap_inode_need_killpriv(struct dentry
*dentry
)
180 struct inode
*inode
= dentry
->d_inode
;
183 if (!inode
->i_op
|| !inode
->i_op
->getxattr
)
186 error
= inode
->i_op
->getxattr(dentry
, XATTR_NAME_CAPS
, NULL
, 0);
192 int cap_inode_killpriv(struct dentry
*dentry
)
194 struct inode
*inode
= dentry
->d_inode
;
196 if (!inode
->i_op
|| !inode
->i_op
->removexattr
)
199 return inode
->i_op
->removexattr(dentry
, XATTR_NAME_CAPS
);
202 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data
*caps
,
203 struct linux_binprm
*bprm
)
208 if (caps
->magic_etc
& VFS_CAP_FLAGS_EFFECTIVE
)
209 bprm
->cap_effective
= true;
211 bprm
->cap_effective
= false;
213 CAP_FOR_EACH_U32(i
) {
214 __u32 permitted
= caps
->permitted
.cap
[i
];
215 __u32 inheritable
= caps
->inheritable
.cap
[i
];
218 * pP' = (X & fP) | (pI & fI)
220 bprm
->cap_post_exec_permitted
.cap
[i
] =
221 (current
->cred
->cap_bset
.cap
[i
] & permitted
) |
222 (current
->cred
->cap_inheritable
.cap
[i
] & inheritable
);
224 if (permitted
& ~bprm
->cap_post_exec_permitted
.cap
[i
]) {
226 * insufficient to execute correctly
233 * For legacy apps, with no internal support for recognizing they
234 * do not have enough capabilities, we return an error if they are
235 * missing some "forced" (aka file-permitted) capabilities.
237 return bprm
->cap_effective
? ret
: 0;
240 int get_vfs_caps_from_disk(const struct dentry
*dentry
, struct cpu_vfs_cap_data
*cpu_caps
)
242 struct inode
*inode
= dentry
->d_inode
;
246 struct vfs_cap_data caps
;
248 memset(cpu_caps
, 0, sizeof(struct cpu_vfs_cap_data
));
250 if (!inode
|| !inode
->i_op
|| !inode
->i_op
->getxattr
)
253 size
= inode
->i_op
->getxattr((struct dentry
*)dentry
, XATTR_NAME_CAPS
, &caps
,
255 if (size
== -ENODATA
|| size
== -EOPNOTSUPP
) {
256 /* no data, that's ok */
262 if (size
< sizeof(magic_etc
))
265 cpu_caps
->magic_etc
= magic_etc
= le32_to_cpu(caps
.magic_etc
);
267 switch ((magic_etc
& VFS_CAP_REVISION_MASK
)) {
268 case VFS_CAP_REVISION_1
:
269 if (size
!= XATTR_CAPS_SZ_1
)
271 tocopy
= VFS_CAP_U32_1
;
273 case VFS_CAP_REVISION_2
:
274 if (size
!= XATTR_CAPS_SZ_2
)
276 tocopy
= VFS_CAP_U32_2
;
282 CAP_FOR_EACH_U32(i
) {
285 cpu_caps
->permitted
.cap
[i
] = le32_to_cpu(caps
.data
[i
].permitted
);
286 cpu_caps
->inheritable
.cap
[i
] = le32_to_cpu(caps
.data
[i
].inheritable
);
291 /* Locate any VFS capabilities: */
292 static int get_file_caps(struct linux_binprm
*bprm
)
294 struct dentry
*dentry
;
296 struct cpu_vfs_cap_data vcaps
;
298 bprm_clear_caps(bprm
);
300 if (!file_caps_enabled
)
303 if (bprm
->file
->f_vfsmnt
->mnt_flags
& MNT_NOSUID
)
306 dentry
= dget(bprm
->file
->f_dentry
);
308 rc
= get_vfs_caps_from_disk(dentry
, &vcaps
);
311 printk(KERN_NOTICE
"%s: get_vfs_caps_from_disk returned %d for %s\n",
312 __func__
, rc
, bprm
->filename
);
313 else if (rc
== -ENODATA
)
318 rc
= bprm_caps_from_vfs_caps(&vcaps
, bprm
);
323 bprm_clear_caps(bprm
);
329 int cap_inode_need_killpriv(struct dentry
*dentry
)
334 int cap_inode_killpriv(struct dentry
*dentry
)
339 static inline int get_file_caps(struct linux_binprm
*bprm
)
341 bprm_clear_caps(bprm
);
346 int cap_bprm_set_security (struct linux_binprm
*bprm
)
350 ret
= get_file_caps(bprm
);
352 if (!issecure(SECURE_NOROOT
)) {
354 * To support inheritance of root-permissions and suid-root
355 * executables under compatibility mode, we override the
356 * capability sets for the file.
358 * If only the real uid is 0, we do not set the effective
361 if (bprm
->e_uid
== 0 || current_uid() == 0) {
362 /* pP' = (cap_bset & ~0) | (pI & ~0) */
363 bprm
->cap_post_exec_permitted
= cap_combine(
364 current
->cred
->cap_bset
,
365 current
->cred
->cap_inheritable
);
366 bprm
->cap_effective
= (bprm
->e_uid
== 0);
374 void cap_bprm_apply_creds (struct linux_binprm
*bprm
, int unsafe
)
376 struct cred
*cred
= current
->cred
;
378 if (bprm
->e_uid
!= cred
->uid
|| bprm
->e_gid
!= cred
->gid
||
379 !cap_issubset(bprm
->cap_post_exec_permitted
,
380 cred
->cap_permitted
)) {
381 set_dumpable(current
->mm
, suid_dumpable
);
382 current
->pdeath_signal
= 0;
384 if (unsafe
& ~LSM_UNSAFE_PTRACE_CAP
) {
385 if (!capable(CAP_SETUID
)) {
386 bprm
->e_uid
= cred
->uid
;
387 bprm
->e_gid
= cred
->gid
;
389 if (cap_limit_ptraced_target()) {
390 bprm
->cap_post_exec_permitted
= cap_intersect(
391 bprm
->cap_post_exec_permitted
,
392 cred
->cap_permitted
);
397 cred
->suid
= cred
->euid
= cred
->fsuid
= bprm
->e_uid
;
398 cred
->sgid
= cred
->egid
= cred
->fsgid
= bprm
->e_gid
;
400 /* For init, we want to retain the capabilities set
401 * in the init_task struct. Thus we skip the usual
402 * capability rules */
403 if (!is_global_init(current
)) {
404 cred
->cap_permitted
= bprm
->cap_post_exec_permitted
;
405 if (bprm
->cap_effective
)
406 cred
->cap_effective
= bprm
->cap_post_exec_permitted
;
408 cap_clear(cred
->cap_effective
);
412 * Audit candidate if current->cap_effective is set
414 * We do not bother to audit if 3 things are true:
415 * 1) cap_effective has all caps
417 * 3) root is supposed to have all caps (SECURE_NOROOT)
418 * Since this is just a normal root execing a process.
420 * Number 1 above might fail if you don't have a full bset, but I think
421 * that is interesting information to audit.
423 if (!cap_isclear(cred
->cap_effective
)) {
424 if (!cap_issubset(CAP_FULL_SET
, cred
->cap_effective
) ||
425 (bprm
->e_uid
!= 0) || (cred
->uid
!= 0) ||
426 issecure(SECURE_NOROOT
))
427 audit_log_bprm_fcaps(bprm
, &cred
->cap_permitted
,
428 &cred
->cap_effective
);
431 cred
->securebits
&= ~issecure_mask(SECURE_KEEP_CAPS
);
434 int cap_bprm_secureexec (struct linux_binprm
*bprm
)
436 const struct cred
*cred
= current
->cred
;
438 if (cred
->uid
!= 0) {
439 if (bprm
->cap_effective
)
441 if (!cap_isclear(bprm
->cap_post_exec_permitted
))
445 return (cred
->euid
!= cred
->uid
||
446 cred
->egid
!= cred
->gid
);
449 int cap_inode_setxattr(struct dentry
*dentry
, const char *name
,
450 const void *value
, size_t size
, int flags
)
452 if (!strcmp(name
, XATTR_NAME_CAPS
)) {
453 if (!capable(CAP_SETFCAP
))
456 } else if (!strncmp(name
, XATTR_SECURITY_PREFIX
,
457 sizeof(XATTR_SECURITY_PREFIX
) - 1) &&
458 !capable(CAP_SYS_ADMIN
))
463 int cap_inode_removexattr(struct dentry
*dentry
, const char *name
)
465 if (!strcmp(name
, XATTR_NAME_CAPS
)) {
466 if (!capable(CAP_SETFCAP
))
469 } else if (!strncmp(name
, XATTR_SECURITY_PREFIX
,
470 sizeof(XATTR_SECURITY_PREFIX
) - 1) &&
471 !capable(CAP_SYS_ADMIN
))
476 /* moved from kernel/sys.c. */
478 * cap_emulate_setxuid() fixes the effective / permitted capabilities of
479 * a process after a call to setuid, setreuid, or setresuid.
481 * 1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
482 * {r,e,s}uid != 0, the permitted and effective capabilities are
485 * 2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
486 * capabilities of the process are cleared.
488 * 3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
489 * capabilities are set to the permitted capabilities.
491 * fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
496 * cevans - New behaviour, Oct '99
497 * A process may, via prctl(), elect to keep its capabilities when it
498 * calls setuid() and switches away from uid==0. Both permitted and
499 * effective sets will be retained.
500 * Without this change, it was impossible for a daemon to drop only some
501 * of its privilege. The call to setuid(!=0) would drop all privileges!
502 * Keeping uid 0 is not an option because uid 0 owns too many vital
504 * Thanks to Olaf Kirch and Peter Benie for spotting this.
506 static inline void cap_emulate_setxuid (int old_ruid
, int old_euid
,
509 struct cred
*cred
= current
->cred
;
511 if ((old_ruid
== 0 || old_euid
== 0 || old_suid
== 0) &&
512 (cred
->uid
!= 0 && cred
->euid
!= 0 && cred
->suid
!= 0) &&
513 !issecure(SECURE_KEEP_CAPS
)) {
514 cap_clear (cred
->cap_permitted
);
515 cap_clear (cred
->cap_effective
);
517 if (old_euid
== 0 && cred
->euid
!= 0) {
518 cap_clear (cred
->cap_effective
);
520 if (old_euid
!= 0 && cred
->euid
== 0) {
521 cred
->cap_effective
= cred
->cap_permitted
;
525 int cap_task_post_setuid (uid_t old_ruid
, uid_t old_euid
, uid_t old_suid
,
528 struct cred
*cred
= current
->cred
;
534 /* Copied from kernel/sys.c:setreuid/setuid/setresuid. */
535 if (!issecure (SECURE_NO_SETUID_FIXUP
)) {
536 cap_emulate_setxuid (old_ruid
, old_euid
, old_suid
);
541 uid_t old_fsuid
= old_ruid
;
543 /* Copied from kernel/sys.c:setfsuid. */
546 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
547 * if not, we might be a bit too harsh here.
550 if (!issecure (SECURE_NO_SETUID_FIXUP
)) {
551 if (old_fsuid
== 0 && cred
->fsuid
!= 0) {
552 cred
->cap_effective
=
554 cred
->cap_effective
);
556 if (old_fsuid
!= 0 && cred
->fsuid
== 0) {
557 cred
->cap_effective
=
560 cred
->cap_permitted
);
572 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
574 * Rationale: code calling task_setscheduler, task_setioprio, and
575 * task_setnice, assumes that
576 * . if capable(cap_sys_nice), then those actions should be allowed
577 * . if not capable(cap_sys_nice), but acting on your own processes,
578 * then those actions should be allowed
579 * This is insufficient now since you can call code without suid, but
580 * yet with increased caps.
581 * So we check for increased caps on the target process.
583 static int cap_safe_nice(struct task_struct
*p
)
585 if (!cap_issubset(p
->cred
->cap_permitted
,
586 current
->cred
->cap_permitted
) &&
587 !capable(CAP_SYS_NICE
))
592 int cap_task_setscheduler (struct task_struct
*p
, int policy
,
593 struct sched_param
*lp
)
595 return cap_safe_nice(p
);
598 int cap_task_setioprio (struct task_struct
*p
, int ioprio
)
600 return cap_safe_nice(p
);
603 int cap_task_setnice (struct task_struct
*p
, int nice
)
605 return cap_safe_nice(p
);
609 * called from kernel/sys.c for prctl(PR_CABSET_DROP)
610 * done without task_capability_lock() because it introduces
611 * no new races - i.e. only another task doing capget() on
612 * this task could get inconsistent info. There can be no
613 * racing writer bc a task can only change its own caps.
615 static long cap_prctl_drop(unsigned long cap
)
617 if (!capable(CAP_SETPCAP
))
621 cap_lower(current
->cred
->cap_bset
, cap
);
626 int cap_task_setscheduler (struct task_struct
*p
, int policy
,
627 struct sched_param
*lp
)
631 int cap_task_setioprio (struct task_struct
*p
, int ioprio
)
635 int cap_task_setnice (struct task_struct
*p
, int nice
)
641 int cap_task_prctl(int option
, unsigned long arg2
, unsigned long arg3
,
642 unsigned long arg4
, unsigned long arg5
, long *rc_p
)
644 struct cred
*cred
= current_cred();
648 case PR_CAPBSET_READ
:
649 if (!cap_valid(arg2
))
652 error
= !!cap_raised(cred
->cap_bset
, arg2
);
654 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
655 case PR_CAPBSET_DROP
:
656 error
= cap_prctl_drop(arg2
);
660 * The next four prctl's remain to assist with transitioning a
661 * system from legacy UID=0 based privilege (when filesystem
662 * capabilities are not in use) to a system using filesystem
663 * capabilities only - as the POSIX.1e draft intended.
667 * PR_SET_SECUREBITS =
668 * issecure_mask(SECURE_KEEP_CAPS_LOCKED)
669 * | issecure_mask(SECURE_NOROOT)
670 * | issecure_mask(SECURE_NOROOT_LOCKED)
671 * | issecure_mask(SECURE_NO_SETUID_FIXUP)
672 * | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
674 * will ensure that the current process and all of its
675 * children will be locked into a pure
676 * capability-based-privilege environment.
678 case PR_SET_SECUREBITS
:
679 if ((((cred
->securebits
& SECURE_ALL_LOCKS
) >> 1)
680 & (cred
->securebits
^ arg2
)) /*[1]*/
681 || ((cred
->securebits
& SECURE_ALL_LOCKS
683 || (arg2
& ~(SECURE_ALL_LOCKS
| SECURE_ALL_BITS
)) /*[3]*/
684 || (cap_capable(current
, CAP_SETPCAP
, SECURITY_CAP_AUDIT
) != 0)) { /*[4]*/
686 * [1] no changing of bits that are locked
687 * [2] no unlocking of locks
688 * [3] no setting of unsupported bits
689 * [4] doing anything requires privilege (go read about
690 * the "sendmail capabilities bug")
692 error
= -EPERM
; /* cannot change a locked bit */
694 cred
->securebits
= arg2
;
697 case PR_GET_SECUREBITS
:
698 error
= cred
->securebits
;
701 #endif /* def CONFIG_SECURITY_FILE_CAPABILITIES */
703 case PR_GET_KEEPCAPS
:
704 if (issecure(SECURE_KEEP_CAPS
))
707 case PR_SET_KEEPCAPS
:
708 if (arg2
> 1) /* Note, we rely on arg2 being unsigned here */
710 else if (issecure(SECURE_KEEP_CAPS_LOCKED
))
713 cred
->securebits
|= issecure_mask(SECURE_KEEP_CAPS
);
715 cred
->securebits
&= ~issecure_mask(SECURE_KEEP_CAPS
);
719 /* No functionality available - continue with default */
723 /* Functionality provided */
728 void cap_task_reparent_to_init (struct task_struct
*p
)
730 struct cred
*cred
= p
->cred
;
732 cap_set_init_eff(cred
->cap_effective
);
733 cap_clear(cred
->cap_inheritable
);
734 cap_set_full(cred
->cap_permitted
);
735 p
->cred
->securebits
= SECUREBITS_DEFAULT
;
738 int cap_syslog (int type
)
740 if ((type
!= 3 && type
!= 10) && !capable(CAP_SYS_ADMIN
))
745 int cap_vm_enough_memory(struct mm_struct
*mm
, long pages
)
747 int cap_sys_admin
= 0;
749 if (cap_capable(current
, CAP_SYS_ADMIN
, SECURITY_CAP_NOAUDIT
) == 0)
751 return __vm_enough_memory(mm
, pages
, cap_sys_admin
);