]>
git.proxmox.com Git - mirror_ubuntu-jammy-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/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_may_access(struct task_struct
*child
, unsigned int mode
)
68 /* Derived from arch/i386/kernel/ptrace.c:sys_ptrace. */
69 if (cap_issubset(child
->cap_permitted
, current
->cap_permitted
))
71 if (capable(CAP_SYS_PTRACE
))
76 int cap_ptrace_traceme(struct task_struct
*parent
)
78 /* Derived from arch/i386/kernel/ptrace.c:sys_ptrace. */
79 if (cap_issubset(current
->cap_permitted
, parent
->cap_permitted
))
81 if (has_capability(parent
, CAP_SYS_PTRACE
))
86 int cap_capget (struct task_struct
*target
, kernel_cap_t
*effective
,
87 kernel_cap_t
*inheritable
, kernel_cap_t
*permitted
)
89 /* Derived from kernel/capability.c:sys_capget. */
90 *effective
= target
->cap_effective
;
91 *inheritable
= target
->cap_inheritable
;
92 *permitted
= target
->cap_permitted
;
96 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
98 static inline int cap_block_setpcap(struct task_struct
*target
)
101 * No support for remote process capability manipulation with
102 * filesystem capability support.
104 return (target
!= current
);
107 static inline int cap_inh_is_capped(void)
110 * Return 1 if changes to the inheritable set are limited
111 * to the old permitted set. That is, if the current task
112 * does *not* possess the CAP_SETPCAP capability.
114 return (cap_capable(current
, CAP_SETPCAP
) != 0);
117 static inline int cap_limit_ptraced_target(void) { return 1; }
119 #else /* ie., ndef CONFIG_SECURITY_FILE_CAPABILITIES */
121 static inline int cap_block_setpcap(struct task_struct
*t
) { return 0; }
122 static inline int cap_inh_is_capped(void) { return 1; }
123 static inline int cap_limit_ptraced_target(void)
125 return !capable(CAP_SETPCAP
);
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 */
143 if (!cap_issubset(*inheritable
,
144 cap_combine(target
->cap_inheritable
,
145 current
->cap_bset
))) {
146 /* no new pI capabilities outside bounding set */
150 /* verify restrictions on target's new Permitted set */
151 if (!cap_issubset (*permitted
,
152 cap_combine (target
->cap_permitted
,
153 current
->cap_permitted
))) {
157 /* verify the _new_Effective_ is a subset of the _new_Permitted_ */
158 if (!cap_issubset (*effective
, *permitted
)) {
165 void cap_capset_set (struct task_struct
*target
, kernel_cap_t
*effective
,
166 kernel_cap_t
*inheritable
, kernel_cap_t
*permitted
)
168 target
->cap_effective
= *effective
;
169 target
->cap_inheritable
= *inheritable
;
170 target
->cap_permitted
= *permitted
;
173 static inline void bprm_clear_caps(struct linux_binprm
*bprm
)
175 cap_clear(bprm
->cap_post_exec_permitted
);
176 bprm
->cap_effective
= false;
179 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
181 int cap_inode_need_killpriv(struct dentry
*dentry
)
183 struct inode
*inode
= dentry
->d_inode
;
186 if (!inode
->i_op
|| !inode
->i_op
->getxattr
)
189 error
= inode
->i_op
->getxattr(dentry
, XATTR_NAME_CAPS
, NULL
, 0);
195 int cap_inode_killpriv(struct dentry
*dentry
)
197 struct inode
*inode
= dentry
->d_inode
;
199 if (!inode
->i_op
|| !inode
->i_op
->removexattr
)
202 return inode
->i_op
->removexattr(dentry
, XATTR_NAME_CAPS
);
205 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data
*caps
,
206 struct linux_binprm
*bprm
)
211 if (caps
->magic_etc
& VFS_CAP_FLAGS_EFFECTIVE
)
212 bprm
->cap_effective
= true;
214 bprm
->cap_effective
= false;
216 CAP_FOR_EACH_U32(i
) {
217 __u32 permitted
= caps
->permitted
.cap
[i
];
218 __u32 inheritable
= caps
->inheritable
.cap
[i
];
221 * pP' = (X & fP) | (pI & fI)
223 bprm
->cap_post_exec_permitted
.cap
[i
] =
224 (current
->cap_bset
.cap
[i
] & permitted
) |
225 (current
->cap_inheritable
.cap
[i
] & inheritable
);
227 if (permitted
& ~bprm
->cap_post_exec_permitted
.cap
[i
]) {
229 * insufficient to execute correctly
236 * For legacy apps, with no internal support for recognizing they
237 * do not have enough capabilities, we return an error if they are
238 * missing some "forced" (aka file-permitted) capabilities.
240 return bprm
->cap_effective
? ret
: 0;
243 int get_vfs_caps_from_disk(const struct dentry
*dentry
, struct cpu_vfs_cap_data
*cpu_caps
)
245 struct inode
*inode
= dentry
->d_inode
;
249 struct vfs_cap_data caps
;
251 memset(cpu_caps
, 0, sizeof(struct cpu_vfs_cap_data
));
253 if (!inode
|| !inode
->i_op
|| !inode
->i_op
->getxattr
)
256 size
= inode
->i_op
->getxattr((struct dentry
*)dentry
, XATTR_NAME_CAPS
, &caps
,
258 if (size
== -ENODATA
|| size
== -EOPNOTSUPP
) {
259 /* no data, that's ok */
265 if (size
< sizeof(magic_etc
))
268 cpu_caps
->magic_etc
= magic_etc
= le32_to_cpu(caps
.magic_etc
);
270 switch ((magic_etc
& VFS_CAP_REVISION_MASK
)) {
271 case VFS_CAP_REVISION_1
:
272 if (size
!= XATTR_CAPS_SZ_1
)
274 tocopy
= VFS_CAP_U32_1
;
276 case VFS_CAP_REVISION_2
:
277 if (size
!= XATTR_CAPS_SZ_2
)
279 tocopy
= VFS_CAP_U32_2
;
285 CAP_FOR_EACH_U32(i
) {
288 cpu_caps
->permitted
.cap
[i
] = le32_to_cpu(caps
.data
[i
].permitted
);
289 cpu_caps
->inheritable
.cap
[i
] = le32_to_cpu(caps
.data
[i
].inheritable
);
294 /* Locate any VFS capabilities: */
295 static int get_file_caps(struct linux_binprm
*bprm
)
297 struct dentry
*dentry
;
299 struct cpu_vfs_cap_data vcaps
;
301 bprm_clear_caps(bprm
);
303 if (!file_caps_enabled
)
306 if (bprm
->file
->f_vfsmnt
->mnt_flags
& MNT_NOSUID
)
309 dentry
= dget(bprm
->file
->f_dentry
);
311 rc
= get_vfs_caps_from_disk(dentry
, &vcaps
);
314 printk(KERN_NOTICE
"%s: get_vfs_caps_from_disk returned %d for %s\n",
315 __func__
, rc
, bprm
->filename
);
316 else if (rc
== -ENODATA
)
321 rc
= bprm_caps_from_vfs_caps(&vcaps
, bprm
);
326 bprm_clear_caps(bprm
);
332 int cap_inode_need_killpriv(struct dentry
*dentry
)
337 int cap_inode_killpriv(struct dentry
*dentry
)
342 static inline int get_file_caps(struct linux_binprm
*bprm
)
344 bprm_clear_caps(bprm
);
349 int cap_bprm_set_security (struct linux_binprm
*bprm
)
353 ret
= get_file_caps(bprm
);
355 if (!issecure(SECURE_NOROOT
)) {
357 * To support inheritance of root-permissions and suid-root
358 * executables under compatibility mode, we override the
359 * capability sets for the file.
361 * If only the real uid is 0, we do not set the effective
364 if (bprm
->e_uid
== 0 || current
->uid
== 0) {
365 /* pP' = (cap_bset & ~0) | (pI & ~0) */
366 bprm
->cap_post_exec_permitted
= cap_combine(
367 current
->cap_bset
, current
->cap_inheritable
369 bprm
->cap_effective
= (bprm
->e_uid
== 0);
377 void cap_bprm_apply_creds (struct linux_binprm
*bprm
, int unsafe
)
379 if (bprm
->e_uid
!= current
->uid
|| bprm
->e_gid
!= current
->gid
||
380 !cap_issubset(bprm
->cap_post_exec_permitted
,
381 current
->cap_permitted
)) {
382 set_dumpable(current
->mm
, suid_dumpable
);
383 current
->pdeath_signal
= 0;
385 if (unsafe
& ~LSM_UNSAFE_PTRACE_CAP
) {
386 if (!capable(CAP_SETUID
)) {
387 bprm
->e_uid
= current
->uid
;
388 bprm
->e_gid
= current
->gid
;
390 if (cap_limit_ptraced_target()) {
391 bprm
->cap_post_exec_permitted
= cap_intersect(
392 bprm
->cap_post_exec_permitted
,
393 current
->cap_permitted
);
398 current
->suid
= current
->euid
= current
->fsuid
= bprm
->e_uid
;
399 current
->sgid
= current
->egid
= current
->fsgid
= bprm
->e_gid
;
401 /* For init, we want to retain the capabilities set
402 * in the init_task struct. Thus we skip the usual
403 * capability rules */
404 if (!is_global_init(current
)) {
405 current
->cap_permitted
= bprm
->cap_post_exec_permitted
;
406 if (bprm
->cap_effective
)
407 current
->cap_effective
= bprm
->cap_post_exec_permitted
;
409 cap_clear(current
->cap_effective
);
412 /* AUD: Audit candidate if current->cap_effective is set */
414 current
->securebits
&= ~issecure_mask(SECURE_KEEP_CAPS
);
417 int cap_bprm_secureexec (struct linux_binprm
*bprm
)
419 if (current
->uid
!= 0) {
420 if (bprm
->cap_effective
)
422 if (!cap_isclear(bprm
->cap_post_exec_permitted
))
426 return (current
->euid
!= current
->uid
||
427 current
->egid
!= current
->gid
);
430 int cap_inode_setxattr(struct dentry
*dentry
, const char *name
,
431 const void *value
, size_t size
, int flags
)
433 if (!strcmp(name
, XATTR_NAME_CAPS
)) {
434 if (!capable(CAP_SETFCAP
))
437 } else if (!strncmp(name
, XATTR_SECURITY_PREFIX
,
438 sizeof(XATTR_SECURITY_PREFIX
) - 1) &&
439 !capable(CAP_SYS_ADMIN
))
444 int cap_inode_removexattr(struct dentry
*dentry
, const char *name
)
446 if (!strcmp(name
, XATTR_NAME_CAPS
)) {
447 if (!capable(CAP_SETFCAP
))
450 } else if (!strncmp(name
, XATTR_SECURITY_PREFIX
,
451 sizeof(XATTR_SECURITY_PREFIX
) - 1) &&
452 !capable(CAP_SYS_ADMIN
))
457 /* moved from kernel/sys.c. */
459 * cap_emulate_setxuid() fixes the effective / permitted capabilities of
460 * a process after a call to setuid, setreuid, or setresuid.
462 * 1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
463 * {r,e,s}uid != 0, the permitted and effective capabilities are
466 * 2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
467 * capabilities of the process are cleared.
469 * 3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
470 * capabilities are set to the permitted capabilities.
472 * fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
477 * cevans - New behaviour, Oct '99
478 * A process may, via prctl(), elect to keep its capabilities when it
479 * calls setuid() and switches away from uid==0. Both permitted and
480 * effective sets will be retained.
481 * Without this change, it was impossible for a daemon to drop only some
482 * of its privilege. The call to setuid(!=0) would drop all privileges!
483 * Keeping uid 0 is not an option because uid 0 owns too many vital
485 * Thanks to Olaf Kirch and Peter Benie for spotting this.
487 static inline void cap_emulate_setxuid (int old_ruid
, int old_euid
,
490 if ((old_ruid
== 0 || old_euid
== 0 || old_suid
== 0) &&
491 (current
->uid
!= 0 && current
->euid
!= 0 && current
->suid
!= 0) &&
492 !issecure(SECURE_KEEP_CAPS
)) {
493 cap_clear (current
->cap_permitted
);
494 cap_clear (current
->cap_effective
);
496 if (old_euid
== 0 && current
->euid
!= 0) {
497 cap_clear (current
->cap_effective
);
499 if (old_euid
!= 0 && current
->euid
== 0) {
500 current
->cap_effective
= current
->cap_permitted
;
504 int cap_task_post_setuid (uid_t old_ruid
, uid_t old_euid
, uid_t old_suid
,
511 /* Copied from kernel/sys.c:setreuid/setuid/setresuid. */
512 if (!issecure (SECURE_NO_SETUID_FIXUP
)) {
513 cap_emulate_setxuid (old_ruid
, old_euid
, old_suid
);
518 uid_t old_fsuid
= old_ruid
;
520 /* Copied from kernel/sys.c:setfsuid. */
523 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
524 * if not, we might be a bit too harsh here.
527 if (!issecure (SECURE_NO_SETUID_FIXUP
)) {
528 if (old_fsuid
== 0 && current
->fsuid
!= 0) {
529 current
->cap_effective
=
531 current
->cap_effective
);
533 if (old_fsuid
!= 0 && current
->fsuid
== 0) {
534 current
->cap_effective
=
536 current
->cap_effective
,
537 current
->cap_permitted
);
549 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
551 * Rationale: code calling task_setscheduler, task_setioprio, and
552 * task_setnice, assumes that
553 * . if capable(cap_sys_nice), then those actions should be allowed
554 * . if not capable(cap_sys_nice), but acting on your own processes,
555 * then those actions should be allowed
556 * This is insufficient now since you can call code without suid, but
557 * yet with increased caps.
558 * So we check for increased caps on the target process.
560 static int cap_safe_nice(struct task_struct
*p
)
562 if (!cap_issubset(p
->cap_permitted
, current
->cap_permitted
) &&
563 !capable(CAP_SYS_NICE
))
568 int cap_task_setscheduler (struct task_struct
*p
, int policy
,
569 struct sched_param
*lp
)
571 return cap_safe_nice(p
);
574 int cap_task_setioprio (struct task_struct
*p
, int ioprio
)
576 return cap_safe_nice(p
);
579 int cap_task_setnice (struct task_struct
*p
, int nice
)
581 return cap_safe_nice(p
);
585 * called from kernel/sys.c for prctl(PR_CABSET_DROP)
586 * done without task_capability_lock() because it introduces
587 * no new races - i.e. only another task doing capget() on
588 * this task could get inconsistent info. There can be no
589 * racing writer bc a task can only change its own caps.
591 static long cap_prctl_drop(unsigned long cap
)
593 if (!capable(CAP_SETPCAP
))
597 cap_lower(current
->cap_bset
, cap
);
602 int cap_task_setscheduler (struct task_struct
*p
, int policy
,
603 struct sched_param
*lp
)
607 int cap_task_setioprio (struct task_struct
*p
, int ioprio
)
611 int cap_task_setnice (struct task_struct
*p
, int nice
)
617 int cap_task_prctl(int option
, unsigned long arg2
, unsigned long arg3
,
618 unsigned long arg4
, unsigned long arg5
, long *rc_p
)
623 case PR_CAPBSET_READ
:
624 if (!cap_valid(arg2
))
627 error
= !!cap_raised(current
->cap_bset
, arg2
);
629 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
630 case PR_CAPBSET_DROP
:
631 error
= cap_prctl_drop(arg2
);
635 * The next four prctl's remain to assist with transitioning a
636 * system from legacy UID=0 based privilege (when filesystem
637 * capabilities are not in use) to a system using filesystem
638 * capabilities only - as the POSIX.1e draft intended.
642 * PR_SET_SECUREBITS =
643 * issecure_mask(SECURE_KEEP_CAPS_LOCKED)
644 * | issecure_mask(SECURE_NOROOT)
645 * | issecure_mask(SECURE_NOROOT_LOCKED)
646 * | issecure_mask(SECURE_NO_SETUID_FIXUP)
647 * | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
649 * will ensure that the current process and all of its
650 * children will be locked into a pure
651 * capability-based-privilege environment.
653 case PR_SET_SECUREBITS
:
654 if ((((current
->securebits
& SECURE_ALL_LOCKS
) >> 1)
655 & (current
->securebits
^ arg2
)) /*[1]*/
656 || ((current
->securebits
& SECURE_ALL_LOCKS
658 || (arg2
& ~(SECURE_ALL_LOCKS
| SECURE_ALL_BITS
)) /*[3]*/
659 || (cap_capable(current
, CAP_SETPCAP
) != 0)) { /*[4]*/
661 * [1] no changing of bits that are locked
662 * [2] no unlocking of locks
663 * [3] no setting of unsupported bits
664 * [4] doing anything requires privilege (go read about
665 * the "sendmail capabilities bug")
667 error
= -EPERM
; /* cannot change a locked bit */
669 current
->securebits
= arg2
;
672 case PR_GET_SECUREBITS
:
673 error
= current
->securebits
;
676 #endif /* def CONFIG_SECURITY_FILE_CAPABILITIES */
678 case PR_GET_KEEPCAPS
:
679 if (issecure(SECURE_KEEP_CAPS
))
682 case PR_SET_KEEPCAPS
:
683 if (arg2
> 1) /* Note, we rely on arg2 being unsigned here */
685 else if (issecure(SECURE_KEEP_CAPS_LOCKED
))
688 current
->securebits
|= issecure_mask(SECURE_KEEP_CAPS
);
690 current
->securebits
&=
691 ~issecure_mask(SECURE_KEEP_CAPS
);
695 /* No functionality available - continue with default */
699 /* Functionality provided */
704 void cap_task_reparent_to_init (struct task_struct
*p
)
706 cap_set_init_eff(p
->cap_effective
);
707 cap_clear(p
->cap_inheritable
);
708 cap_set_full(p
->cap_permitted
);
709 p
->securebits
= SECUREBITS_DEFAULT
;
713 int cap_syslog (int type
)
715 if ((type
!= 3 && type
!= 10) && !capable(CAP_SYS_ADMIN
))
720 int cap_vm_enough_memory(struct mm_struct
*mm
, long pages
)
722 int cap_sys_admin
= 0;
724 if (cap_capable(current
, CAP_SYS_ADMIN
) == 0)
726 return __vm_enough_memory(mm
, pages
, cap_sys_admin
);