]>
git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blob - security/commoncap.c
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_inh_is_capped(void)
102 * Return 1 if changes to the inheritable set are limited
103 * to the old permitted set. That is, if the current task
104 * does *not* possess the CAP_SETPCAP capability.
106 return (cap_capable(current
, CAP_SETPCAP
, SECURITY_CAP_AUDIT
) != 0);
109 static inline int cap_limit_ptraced_target(void) { return 1; }
111 #else /* ie., ndef CONFIG_SECURITY_FILE_CAPABILITIES */
113 static inline int cap_inh_is_capped(void) { return 1; }
114 static inline int cap_limit_ptraced_target(void)
116 return !capable(CAP_SETPCAP
);
119 #endif /* def CONFIG_SECURITY_FILE_CAPABILITIES */
121 int cap_capset_check (kernel_cap_t
*effective
,
122 kernel_cap_t
*inheritable
, kernel_cap_t
*permitted
)
124 if (cap_inh_is_capped()
125 && !cap_issubset(*inheritable
,
126 cap_combine(current
->cap_inheritable
,
127 current
->cap_permitted
))) {
128 /* incapable of using this inheritable set */
131 if (!cap_issubset(*inheritable
,
132 cap_combine(current
->cap_inheritable
,
133 current
->cap_bset
))) {
134 /* no new pI capabilities outside bounding set */
138 /* verify restrictions on target's new Permitted set */
139 if (!cap_issubset (*permitted
,
140 cap_combine (current
->cap_permitted
,
141 current
->cap_permitted
))) {
145 /* verify the _new_Effective_ is a subset of the _new_Permitted_ */
146 if (!cap_issubset (*effective
, *permitted
)) {
153 void cap_capset_set (kernel_cap_t
*effective
,
154 kernel_cap_t
*inheritable
, kernel_cap_t
*permitted
)
156 current
->cap_effective
= *effective
;
157 current
->cap_inheritable
= *inheritable
;
158 current
->cap_permitted
= *permitted
;
161 static inline void bprm_clear_caps(struct linux_binprm
*bprm
)
163 cap_clear(bprm
->cap_post_exec_permitted
);
164 bprm
->cap_effective
= false;
167 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
169 int cap_inode_need_killpriv(struct dentry
*dentry
)
171 struct inode
*inode
= dentry
->d_inode
;
174 if (!inode
->i_op
|| !inode
->i_op
->getxattr
)
177 error
= inode
->i_op
->getxattr(dentry
, XATTR_NAME_CAPS
, NULL
, 0);
183 int cap_inode_killpriv(struct dentry
*dentry
)
185 struct inode
*inode
= dentry
->d_inode
;
187 if (!inode
->i_op
|| !inode
->i_op
->removexattr
)
190 return inode
->i_op
->removexattr(dentry
, XATTR_NAME_CAPS
);
193 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data
*caps
,
194 struct linux_binprm
*bprm
)
199 if (caps
->magic_etc
& VFS_CAP_FLAGS_EFFECTIVE
)
200 bprm
->cap_effective
= true;
202 bprm
->cap_effective
= false;
204 CAP_FOR_EACH_U32(i
) {
205 __u32 permitted
= caps
->permitted
.cap
[i
];
206 __u32 inheritable
= caps
->inheritable
.cap
[i
];
209 * pP' = (X & fP) | (pI & fI)
211 bprm
->cap_post_exec_permitted
.cap
[i
] =
212 (current
->cap_bset
.cap
[i
] & permitted
) |
213 (current
->cap_inheritable
.cap
[i
] & inheritable
);
215 if (permitted
& ~bprm
->cap_post_exec_permitted
.cap
[i
]) {
217 * insufficient to execute correctly
224 * For legacy apps, with no internal support for recognizing they
225 * do not have enough capabilities, we return an error if they are
226 * missing some "forced" (aka file-permitted) capabilities.
228 return bprm
->cap_effective
? ret
: 0;
231 int get_vfs_caps_from_disk(const struct dentry
*dentry
, struct cpu_vfs_cap_data
*cpu_caps
)
233 struct inode
*inode
= dentry
->d_inode
;
237 struct vfs_cap_data caps
;
239 memset(cpu_caps
, 0, sizeof(struct cpu_vfs_cap_data
));
241 if (!inode
|| !inode
->i_op
|| !inode
->i_op
->getxattr
)
244 size
= inode
->i_op
->getxattr((struct dentry
*)dentry
, XATTR_NAME_CAPS
, &caps
,
246 if (size
== -ENODATA
|| size
== -EOPNOTSUPP
) {
247 /* no data, that's ok */
253 if (size
< sizeof(magic_etc
))
256 cpu_caps
->magic_etc
= magic_etc
= le32_to_cpu(caps
.magic_etc
);
258 switch ((magic_etc
& VFS_CAP_REVISION_MASK
)) {
259 case VFS_CAP_REVISION_1
:
260 if (size
!= XATTR_CAPS_SZ_1
)
262 tocopy
= VFS_CAP_U32_1
;
264 case VFS_CAP_REVISION_2
:
265 if (size
!= XATTR_CAPS_SZ_2
)
267 tocopy
= VFS_CAP_U32_2
;
273 CAP_FOR_EACH_U32(i
) {
276 cpu_caps
->permitted
.cap
[i
] = le32_to_cpu(caps
.data
[i
].permitted
);
277 cpu_caps
->inheritable
.cap
[i
] = le32_to_cpu(caps
.data
[i
].inheritable
);
282 /* Locate any VFS capabilities: */
283 static int get_file_caps(struct linux_binprm
*bprm
)
285 struct dentry
*dentry
;
287 struct cpu_vfs_cap_data vcaps
;
289 bprm_clear_caps(bprm
);
291 if (!file_caps_enabled
)
294 if (bprm
->file
->f_vfsmnt
->mnt_flags
& MNT_NOSUID
)
297 dentry
= dget(bprm
->file
->f_dentry
);
299 rc
= get_vfs_caps_from_disk(dentry
, &vcaps
);
302 printk(KERN_NOTICE
"%s: get_vfs_caps_from_disk returned %d for %s\n",
303 __func__
, rc
, bprm
->filename
);
304 else if (rc
== -ENODATA
)
309 rc
= bprm_caps_from_vfs_caps(&vcaps
, bprm
);
314 bprm_clear_caps(bprm
);
320 int cap_inode_need_killpriv(struct dentry
*dentry
)
325 int cap_inode_killpriv(struct dentry
*dentry
)
330 static inline int get_file_caps(struct linux_binprm
*bprm
)
332 bprm_clear_caps(bprm
);
337 int cap_bprm_set_security (struct linux_binprm
*bprm
)
341 ret
= get_file_caps(bprm
);
343 if (!issecure(SECURE_NOROOT
)) {
345 * To support inheritance of root-permissions and suid-root
346 * executables under compatibility mode, we override the
347 * capability sets for the file.
349 * If only the real uid is 0, we do not set the effective
352 if (bprm
->e_uid
== 0 || current_uid() == 0) {
353 /* pP' = (cap_bset & ~0) | (pI & ~0) */
354 bprm
->cap_post_exec_permitted
= cap_combine(
355 current
->cap_bset
, current
->cap_inheritable
357 bprm
->cap_effective
= (bprm
->e_uid
== 0);
365 void cap_bprm_apply_creds (struct linux_binprm
*bprm
, int unsafe
)
367 kernel_cap_t pP
= current
->cap_permitted
;
368 kernel_cap_t pE
= current
->cap_effective
;
372 current_uid_gid(&uid
, &gid
);
374 if (bprm
->e_uid
!= uid
|| bprm
->e_gid
!= gid
||
375 !cap_issubset(bprm
->cap_post_exec_permitted
,
376 current
->cap_permitted
)) {
377 set_dumpable(current
->mm
, suid_dumpable
);
378 current
->pdeath_signal
= 0;
380 if (unsafe
& ~LSM_UNSAFE_PTRACE_CAP
) {
381 if (!capable(CAP_SETUID
)) {
385 if (cap_limit_ptraced_target()) {
386 bprm
->cap_post_exec_permitted
= cap_intersect(
387 bprm
->cap_post_exec_permitted
,
388 current
->cap_permitted
);
393 current
->suid
= current
->euid
= current
->fsuid
= bprm
->e_uid
;
394 current
->sgid
= current
->egid
= current
->fsgid
= bprm
->e_gid
;
396 /* For init, we want to retain the capabilities set
397 * in the init_task struct. Thus we skip the usual
398 * capability rules */
399 if (!is_global_init(current
)) {
400 current
->cap_permitted
= bprm
->cap_post_exec_permitted
;
401 if (bprm
->cap_effective
)
402 current
->cap_effective
= bprm
->cap_post_exec_permitted
;
404 cap_clear(current
->cap_effective
);
408 * Audit candidate if current->cap_effective is set
410 * We do not bother to audit if 3 things are true:
411 * 1) cap_effective has all caps
413 * 3) root is supposed to have all caps (SECURE_NOROOT)
414 * Since this is just a normal root execing a process.
416 * Number 1 above might fail if you don't have a full bset, but I think
417 * that is interesting information to audit.
419 if (!cap_isclear(current
->cap_effective
)) {
420 if (!cap_issubset(CAP_FULL_SET
, current
->cap_effective
) ||
421 (bprm
->e_uid
!= 0) || (current
->uid
!= 0) ||
422 issecure(SECURE_NOROOT
))
423 audit_log_bprm_fcaps(bprm
, &pP
, &pE
);
426 current
->securebits
&= ~issecure_mask(SECURE_KEEP_CAPS
);
429 int cap_bprm_secureexec (struct linux_binprm
*bprm
)
431 if (current_uid() != 0) {
432 if (bprm
->cap_effective
)
434 if (!cap_isclear(bprm
->cap_post_exec_permitted
))
438 return (current_euid() != current_uid() ||
439 current_egid() != current_gid());
442 int cap_inode_setxattr(struct dentry
*dentry
, const char *name
,
443 const void *value
, size_t size
, int flags
)
445 if (!strcmp(name
, XATTR_NAME_CAPS
)) {
446 if (!capable(CAP_SETFCAP
))
449 } else if (!strncmp(name
, XATTR_SECURITY_PREFIX
,
450 sizeof(XATTR_SECURITY_PREFIX
) - 1) &&
451 !capable(CAP_SYS_ADMIN
))
456 int cap_inode_removexattr(struct dentry
*dentry
, const char *name
)
458 if (!strcmp(name
, XATTR_NAME_CAPS
)) {
459 if (!capable(CAP_SETFCAP
))
462 } else if (!strncmp(name
, XATTR_SECURITY_PREFIX
,
463 sizeof(XATTR_SECURITY_PREFIX
) - 1) &&
464 !capable(CAP_SYS_ADMIN
))
469 /* moved from kernel/sys.c. */
471 * cap_emulate_setxuid() fixes the effective / permitted capabilities of
472 * a process after a call to setuid, setreuid, or setresuid.
474 * 1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
475 * {r,e,s}uid != 0, the permitted and effective capabilities are
478 * 2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
479 * capabilities of the process are cleared.
481 * 3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
482 * capabilities are set to the permitted capabilities.
484 * fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
489 * cevans - New behaviour, Oct '99
490 * A process may, via prctl(), elect to keep its capabilities when it
491 * calls setuid() and switches away from uid==0. Both permitted and
492 * effective sets will be retained.
493 * Without this change, it was impossible for a daemon to drop only some
494 * of its privilege. The call to setuid(!=0) would drop all privileges!
495 * Keeping uid 0 is not an option because uid 0 owns too many vital
497 * Thanks to Olaf Kirch and Peter Benie for spotting this.
499 static inline void cap_emulate_setxuid (int old_ruid
, int old_euid
,
502 uid_t euid
= current_euid();
504 if ((old_ruid
== 0 || old_euid
== 0 || old_suid
== 0) &&
505 (current_uid() != 0 && euid
!= 0 && current_suid() != 0) &&
506 !issecure(SECURE_KEEP_CAPS
)) {
507 cap_clear (current
->cap_permitted
);
508 cap_clear (current
->cap_effective
);
510 if (old_euid
== 0 && euid
!= 0) {
511 cap_clear (current
->cap_effective
);
513 if (old_euid
!= 0 && euid
== 0) {
514 current
->cap_effective
= current
->cap_permitted
;
518 int cap_task_post_setuid (uid_t old_ruid
, uid_t old_euid
, uid_t old_suid
,
525 /* Copied from kernel/sys.c:setreuid/setuid/setresuid. */
526 if (!issecure (SECURE_NO_SETUID_FIXUP
)) {
527 cap_emulate_setxuid (old_ruid
, old_euid
, old_suid
);
532 uid_t old_fsuid
= old_ruid
;
534 /* Copied from kernel/sys.c:setfsuid. */
537 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
538 * if not, we might be a bit too harsh here.
541 if (!issecure (SECURE_NO_SETUID_FIXUP
)) {
542 if (old_fsuid
== 0 && current_fsuid() != 0) {
543 current
->cap_effective
=
545 current
->cap_effective
);
547 if (old_fsuid
!= 0 && current_fsuid() == 0) {
548 current
->cap_effective
=
550 current
->cap_effective
,
551 current
->cap_permitted
);
563 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
565 * Rationale: code calling task_setscheduler, task_setioprio, and
566 * task_setnice, assumes that
567 * . if capable(cap_sys_nice), then those actions should be allowed
568 * . if not capable(cap_sys_nice), but acting on your own processes,
569 * then those actions should be allowed
570 * This is insufficient now since you can call code without suid, but
571 * yet with increased caps.
572 * So we check for increased caps on the target process.
574 static int cap_safe_nice(struct task_struct
*p
)
576 if (!cap_issubset(p
->cap_permitted
, current
->cap_permitted
) &&
577 !capable(CAP_SYS_NICE
))
582 int cap_task_setscheduler (struct task_struct
*p
, int policy
,
583 struct sched_param
*lp
)
585 return cap_safe_nice(p
);
588 int cap_task_setioprio (struct task_struct
*p
, int ioprio
)
590 return cap_safe_nice(p
);
593 int cap_task_setnice (struct task_struct
*p
, int nice
)
595 return cap_safe_nice(p
);
599 * called from kernel/sys.c for prctl(PR_CABSET_DROP)
600 * done without task_capability_lock() because it introduces
601 * no new races - i.e. only another task doing capget() on
602 * this task could get inconsistent info. There can be no
603 * racing writer bc a task can only change its own caps.
605 static long cap_prctl_drop(unsigned long cap
)
607 if (!capable(CAP_SETPCAP
))
611 cap_lower(current
->cap_bset
, cap
);
616 int cap_task_setscheduler (struct task_struct
*p
, int policy
,
617 struct sched_param
*lp
)
621 int cap_task_setioprio (struct task_struct
*p
, int ioprio
)
625 int cap_task_setnice (struct task_struct
*p
, int nice
)
631 int cap_task_prctl(int option
, unsigned long arg2
, unsigned long arg3
,
632 unsigned long arg4
, unsigned long arg5
, long *rc_p
)
637 case PR_CAPBSET_READ
:
638 if (!cap_valid(arg2
))
641 error
= !!cap_raised(current
->cap_bset
, arg2
);
643 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
644 case PR_CAPBSET_DROP
:
645 error
= cap_prctl_drop(arg2
);
649 * The next four prctl's remain to assist with transitioning a
650 * system from legacy UID=0 based privilege (when filesystem
651 * capabilities are not in use) to a system using filesystem
652 * capabilities only - as the POSIX.1e draft intended.
656 * PR_SET_SECUREBITS =
657 * issecure_mask(SECURE_KEEP_CAPS_LOCKED)
658 * | issecure_mask(SECURE_NOROOT)
659 * | issecure_mask(SECURE_NOROOT_LOCKED)
660 * | issecure_mask(SECURE_NO_SETUID_FIXUP)
661 * | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
663 * will ensure that the current process and all of its
664 * children will be locked into a pure
665 * capability-based-privilege environment.
667 case PR_SET_SECUREBITS
:
668 if ((((current
->securebits
& SECURE_ALL_LOCKS
) >> 1)
669 & (current
->securebits
^ arg2
)) /*[1]*/
670 || ((current
->securebits
& SECURE_ALL_LOCKS
672 || (arg2
& ~(SECURE_ALL_LOCKS
| SECURE_ALL_BITS
)) /*[3]*/
673 || (cap_capable(current
, CAP_SETPCAP
, SECURITY_CAP_AUDIT
) != 0)) { /*[4]*/
675 * [1] no changing of bits that are locked
676 * [2] no unlocking of locks
677 * [3] no setting of unsupported bits
678 * [4] doing anything requires privilege (go read about
679 * the "sendmail capabilities bug")
681 error
= -EPERM
; /* cannot change a locked bit */
683 current
->securebits
= arg2
;
686 case PR_GET_SECUREBITS
:
687 error
= current
->securebits
;
690 #endif /* def CONFIG_SECURITY_FILE_CAPABILITIES */
692 case PR_GET_KEEPCAPS
:
693 if (issecure(SECURE_KEEP_CAPS
))
696 case PR_SET_KEEPCAPS
:
697 if (arg2
> 1) /* Note, we rely on arg2 being unsigned here */
699 else if (issecure(SECURE_KEEP_CAPS_LOCKED
))
702 current
->securebits
|= issecure_mask(SECURE_KEEP_CAPS
);
704 current
->securebits
&=
705 ~issecure_mask(SECURE_KEEP_CAPS
);
709 /* No functionality available - continue with default */
713 /* Functionality provided */
718 void cap_task_reparent_to_init (struct task_struct
*p
)
720 cap_set_init_eff(p
->cap_effective
);
721 cap_clear(p
->cap_inheritable
);
722 cap_set_full(p
->cap_permitted
);
723 p
->securebits
= SECUREBITS_DEFAULT
;
727 int cap_syslog (int type
)
729 if ((type
!= 3 && type
!= 10) && !capable(CAP_SYS_ADMIN
))
734 int cap_vm_enough_memory(struct mm_struct
*mm
, long pages
)
736 int cap_sys_admin
= 0;
738 if (cap_capable(current
, CAP_SYS_ADMIN
, SECURITY_CAP_NOAUDIT
) == 0)
740 return __vm_enough_memory(mm
, pages
, cap_sys_admin
);