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)
56 /* Derived from include/linux/sched.h:capable. */
58 cap_raised = cap_raised(__task_cred(tsk)->cap_effective, cap);
60 return cap_raised ? 0 : -EPERM;
63 int cap_settime(struct timespec *ts, struct timezone *tz)
65 if (!capable(CAP_SYS_TIME))
70 int cap_ptrace_may_access(struct task_struct *child, unsigned int mode)
75 if (!cap_issubset(child->cred->cap_permitted,
76 current->cred->cap_permitted) &&
77 !capable(CAP_SYS_PTRACE))
83 int cap_ptrace_traceme(struct task_struct *parent)
88 if (!cap_issubset(current->cred->cap_permitted,
89 parent->cred->cap_permitted) &&
90 !has_capability(parent, CAP_SYS_PTRACE))
96 int cap_capget (struct task_struct *target, kernel_cap_t *effective,
97 kernel_cap_t *inheritable, kernel_cap_t *permitted)
99 const struct cred *cred;
101 /* Derived from kernel/capability.c:sys_capget. */
103 cred = __task_cred(target);
104 *effective = cred->cap_effective;
105 *inheritable = cred->cap_inheritable;
106 *permitted = cred->cap_permitted;
111 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
113 static inline int cap_inh_is_capped(void)
116 * Return 1 if changes to the inheritable set are limited
117 * to the old permitted set. That is, if the current task
118 * does *not* possess the CAP_SETPCAP capability.
120 return (cap_capable(current, CAP_SETPCAP, SECURITY_CAP_AUDIT) != 0);
123 static inline int cap_limit_ptraced_target(void) { return 1; }
125 #else /* ie., ndef CONFIG_SECURITY_FILE_CAPABILITIES */
127 static inline int cap_inh_is_capped(void) { return 1; }
128 static inline int cap_limit_ptraced_target(void)
130 return !capable(CAP_SETPCAP);
133 #endif /* def CONFIG_SECURITY_FILE_CAPABILITIES */
135 int cap_capset_check(const kernel_cap_t *effective,
136 const kernel_cap_t *inheritable,
137 const kernel_cap_t *permitted)
139 const struct cred *cred = current->cred;
141 if (cap_inh_is_capped()
142 && !cap_issubset(*inheritable,
143 cap_combine(cred->cap_inheritable,
144 cred->cap_permitted))) {
145 /* incapable of using this inheritable set */
148 if (!cap_issubset(*inheritable,
149 cap_combine(cred->cap_inheritable,
151 /* no new pI capabilities outside bounding set */
155 /* verify restrictions on target's new Permitted set */
156 if (!cap_issubset (*permitted,
157 cap_combine (cred->cap_permitted,
158 cred->cap_permitted))) {
162 /* verify the _new_Effective_ is a subset of the _new_Permitted_ */
163 if (!cap_issubset (*effective, *permitted)) {
170 void cap_capset_set(const kernel_cap_t *effective,
171 const kernel_cap_t *inheritable,
172 const kernel_cap_t *permitted)
174 struct cred *cred = current->cred;
176 cred->cap_effective = *effective;
177 cred->cap_inheritable = *inheritable;
178 cred->cap_permitted = *permitted;
181 static inline void bprm_clear_caps(struct linux_binprm *bprm)
183 cap_clear(bprm->cap_post_exec_permitted);
184 bprm->cap_effective = false;
187 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
189 int cap_inode_need_killpriv(struct dentry *dentry)
191 struct inode *inode = dentry->d_inode;
194 if (!inode->i_op || !inode->i_op->getxattr)
197 error = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, NULL, 0);
203 int cap_inode_killpriv(struct dentry *dentry)
205 struct inode *inode = dentry->d_inode;
207 if (!inode->i_op || !inode->i_op->removexattr)
210 return inode->i_op->removexattr(dentry, XATTR_NAME_CAPS);
213 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
214 struct linux_binprm *bprm)
219 if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
220 bprm->cap_effective = true;
222 bprm->cap_effective = false;
224 CAP_FOR_EACH_U32(i) {
225 __u32 permitted = caps->permitted.cap[i];
226 __u32 inheritable = caps->inheritable.cap[i];
229 * pP' = (X & fP) | (pI & fI)
231 bprm->cap_post_exec_permitted.cap[i] =
232 (current->cred->cap_bset.cap[i] & permitted) |
233 (current->cred->cap_inheritable.cap[i] & inheritable);
235 if (permitted & ~bprm->cap_post_exec_permitted.cap[i]) {
237 * insufficient to execute correctly
244 * For legacy apps, with no internal support for recognizing they
245 * do not have enough capabilities, we return an error if they are
246 * missing some "forced" (aka file-permitted) capabilities.
248 return bprm->cap_effective ? ret : 0;
251 int get_vfs_caps_from_disk(const struct dentry *dentry, struct cpu_vfs_cap_data *cpu_caps)
253 struct inode *inode = dentry->d_inode;
257 struct vfs_cap_data caps;
259 memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
261 if (!inode || !inode->i_op || !inode->i_op->getxattr)
264 size = inode->i_op->getxattr((struct dentry *)dentry, XATTR_NAME_CAPS, &caps,
266 if (size == -ENODATA || size == -EOPNOTSUPP) {
267 /* no data, that's ok */
273 if (size < sizeof(magic_etc))
276 cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps.magic_etc);
278 switch ((magic_etc & VFS_CAP_REVISION_MASK)) {
279 case VFS_CAP_REVISION_1:
280 if (size != XATTR_CAPS_SZ_1)
282 tocopy = VFS_CAP_U32_1;
284 case VFS_CAP_REVISION_2:
285 if (size != XATTR_CAPS_SZ_2)
287 tocopy = VFS_CAP_U32_2;
293 CAP_FOR_EACH_U32(i) {
296 cpu_caps->permitted.cap[i] = le32_to_cpu(caps.data[i].permitted);
297 cpu_caps->inheritable.cap[i] = le32_to_cpu(caps.data[i].inheritable);
302 /* Locate any VFS capabilities: */
303 static int get_file_caps(struct linux_binprm *bprm)
305 struct dentry *dentry;
307 struct cpu_vfs_cap_data vcaps;
309 bprm_clear_caps(bprm);
311 if (!file_caps_enabled)
314 if (bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)
317 dentry = dget(bprm->file->f_dentry);
319 rc = get_vfs_caps_from_disk(dentry, &vcaps);
322 printk(KERN_NOTICE "%s: get_vfs_caps_from_disk returned %d for %s\n",
323 __func__, rc, bprm->filename);
324 else if (rc == -ENODATA)
329 rc = bprm_caps_from_vfs_caps(&vcaps, bprm);
334 bprm_clear_caps(bprm);
340 int cap_inode_need_killpriv(struct dentry *dentry)
345 int cap_inode_killpriv(struct dentry *dentry)
350 static inline int get_file_caps(struct linux_binprm *bprm)
352 bprm_clear_caps(bprm);
357 int cap_bprm_set_security (struct linux_binprm *bprm)
361 ret = get_file_caps(bprm);
363 if (!issecure(SECURE_NOROOT)) {
365 * To support inheritance of root-permissions and suid-root
366 * executables under compatibility mode, we override the
367 * capability sets for the file.
369 * If only the real uid is 0, we do not set the effective
372 if (bprm->e_uid == 0 || current_uid() == 0) {
373 /* pP' = (cap_bset & ~0) | (pI & ~0) */
374 bprm->cap_post_exec_permitted = cap_combine(
375 current->cred->cap_bset,
376 current->cred->cap_inheritable);
377 bprm->cap_effective = (bprm->e_uid == 0);
385 void cap_bprm_apply_creds (struct linux_binprm *bprm, int unsafe)
387 struct cred *cred = current->cred;
389 if (bprm->e_uid != cred->uid || bprm->e_gid != cred->gid ||
390 !cap_issubset(bprm->cap_post_exec_permitted,
391 cred->cap_permitted)) {
392 set_dumpable(current->mm, suid_dumpable);
393 current->pdeath_signal = 0;
395 if (unsafe & ~LSM_UNSAFE_PTRACE_CAP) {
396 if (!capable(CAP_SETUID)) {
397 bprm->e_uid = cred->uid;
398 bprm->e_gid = cred->gid;
400 if (cap_limit_ptraced_target()) {
401 bprm->cap_post_exec_permitted = cap_intersect(
402 bprm->cap_post_exec_permitted,
403 cred->cap_permitted);
408 cred->suid = cred->euid = cred->fsuid = bprm->e_uid;
409 cred->sgid = cred->egid = cred->fsgid = bprm->e_gid;
411 /* For init, we want to retain the capabilities set
412 * in the init_task struct. Thus we skip the usual
413 * capability rules */
414 if (!is_global_init(current)) {
415 cred->cap_permitted = bprm->cap_post_exec_permitted;
416 if (bprm->cap_effective)
417 cred->cap_effective = bprm->cap_post_exec_permitted;
419 cap_clear(cred->cap_effective);
423 * Audit candidate if current->cap_effective is set
425 * We do not bother to audit if 3 things are true:
426 * 1) cap_effective has all caps
428 * 3) root is supposed to have all caps (SECURE_NOROOT)
429 * Since this is just a normal root execing a process.
431 * Number 1 above might fail if you don't have a full bset, but I think
432 * that is interesting information to audit.
434 if (!cap_isclear(cred->cap_effective)) {
435 if (!cap_issubset(CAP_FULL_SET, cred->cap_effective) ||
436 (bprm->e_uid != 0) || (cred->uid != 0) ||
437 issecure(SECURE_NOROOT))
438 audit_log_bprm_fcaps(bprm, &cred->cap_permitted,
439 &cred->cap_effective);
442 cred->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
445 int cap_bprm_secureexec (struct linux_binprm *bprm)
447 const struct cred *cred = current_cred();
449 if (cred->uid != 0) {
450 if (bprm->cap_effective)
452 if (!cap_isclear(bprm->cap_post_exec_permitted))
456 return (cred->euid != cred->uid ||
457 cred->egid != cred->gid);
460 int cap_inode_setxattr(struct dentry *dentry, const char *name,
461 const void *value, size_t size, int flags)
463 if (!strcmp(name, XATTR_NAME_CAPS)) {
464 if (!capable(CAP_SETFCAP))
467 } else if (!strncmp(name, XATTR_SECURITY_PREFIX,
468 sizeof(XATTR_SECURITY_PREFIX) - 1) &&
469 !capable(CAP_SYS_ADMIN))
474 int cap_inode_removexattr(struct dentry *dentry, const char *name)
476 if (!strcmp(name, XATTR_NAME_CAPS)) {
477 if (!capable(CAP_SETFCAP))
480 } else if (!strncmp(name, XATTR_SECURITY_PREFIX,
481 sizeof(XATTR_SECURITY_PREFIX) - 1) &&
482 !capable(CAP_SYS_ADMIN))
487 /* moved from kernel/sys.c. */
489 * cap_emulate_setxuid() fixes the effective / permitted capabilities of
490 * a process after a call to setuid, setreuid, or setresuid.
492 * 1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
493 * {r,e,s}uid != 0, the permitted and effective capabilities are
496 * 2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
497 * capabilities of the process are cleared.
499 * 3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
500 * capabilities are set to the permitted capabilities.
502 * fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
507 * cevans - New behaviour, Oct '99
508 * A process may, via prctl(), elect to keep its capabilities when it
509 * calls setuid() and switches away from uid==0. Both permitted and
510 * effective sets will be retained.
511 * Without this change, it was impossible for a daemon to drop only some
512 * of its privilege. The call to setuid(!=0) would drop all privileges!
513 * Keeping uid 0 is not an option because uid 0 owns too many vital
515 * Thanks to Olaf Kirch and Peter Benie for spotting this.
517 static inline void cap_emulate_setxuid (int old_ruid, int old_euid,
520 struct cred *cred = current->cred;
522 if ((old_ruid == 0 || old_euid == 0 || old_suid == 0) &&
523 (cred->uid != 0 && cred->euid != 0 && cred->suid != 0) &&
524 !issecure(SECURE_KEEP_CAPS)) {
525 cap_clear(cred->cap_permitted);
526 cap_clear(cred->cap_effective);
528 if (old_euid == 0 && cred->euid != 0) {
529 cap_clear(cred->cap_effective);
531 if (old_euid != 0 && cred->euid == 0) {
532 cred->cap_effective = cred->cap_permitted;
536 int cap_task_post_setuid (uid_t old_ruid, uid_t old_euid, uid_t old_suid,
539 struct cred *cred = current->cred;
545 /* Copied from kernel/sys.c:setreuid/setuid/setresuid. */
546 if (!issecure (SECURE_NO_SETUID_FIXUP)) {
547 cap_emulate_setxuid (old_ruid, old_euid, old_suid);
552 uid_t old_fsuid = old_ruid;
554 /* Copied from kernel/sys.c:setfsuid. */
557 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
558 * if not, we might be a bit too harsh here.
561 if (!issecure (SECURE_NO_SETUID_FIXUP)) {
562 if (old_fsuid == 0 && cred->fsuid != 0) {
563 cred->cap_effective =
565 cred->cap_effective);
567 if (old_fsuid != 0 && cred->fsuid == 0) {
568 cred->cap_effective =
571 cred->cap_permitted);
583 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
585 * Rationale: code calling task_setscheduler, task_setioprio, and
586 * task_setnice, assumes that
587 * . if capable(cap_sys_nice), then those actions should be allowed
588 * . if not capable(cap_sys_nice), but acting on your own processes,
589 * then those actions should be allowed
590 * This is insufficient now since you can call code without suid, but
591 * yet with increased caps.
592 * So we check for increased caps on the target process.
594 static int cap_safe_nice(struct task_struct *p)
599 is_subset = cap_issubset(__task_cred(p)->cap_permitted,
600 current_cred()->cap_permitted);
603 if (!is_subset && !capable(CAP_SYS_NICE))
608 int cap_task_setscheduler (struct task_struct *p, int policy,
609 struct sched_param *lp)
611 return cap_safe_nice(p);
614 int cap_task_setioprio (struct task_struct *p, int ioprio)
616 return cap_safe_nice(p);
619 int cap_task_setnice (struct task_struct *p, int nice)
621 return cap_safe_nice(p);
625 * called from kernel/sys.c for prctl(PR_CABSET_DROP)
626 * done without task_capability_lock() because it introduces
627 * no new races - i.e. only another task doing capget() on
628 * this task could get inconsistent info. There can be no
629 * racing writer bc a task can only change its own caps.
631 static long cap_prctl_drop(unsigned long cap)
633 if (!capable(CAP_SETPCAP))
637 cap_lower(current->cred->cap_bset, cap);
642 int cap_task_setscheduler (struct task_struct *p, int policy,
643 struct sched_param *lp)
647 int cap_task_setioprio (struct task_struct *p, int ioprio)
651 int cap_task_setnice (struct task_struct *p, int nice)
657 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
658 unsigned long arg4, unsigned long arg5, long *rc_p)
660 struct cred *cred = current_cred();
664 case PR_CAPBSET_READ:
665 if (!cap_valid(arg2))
668 error = !!cap_raised(cred->cap_bset, arg2);
670 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
671 case PR_CAPBSET_DROP:
672 error = cap_prctl_drop(arg2);
676 * The next four prctl's remain to assist with transitioning a
677 * system from legacy UID=0 based privilege (when filesystem
678 * capabilities are not in use) to a system using filesystem
679 * capabilities only - as the POSIX.1e draft intended.
683 * PR_SET_SECUREBITS =
684 * issecure_mask(SECURE_KEEP_CAPS_LOCKED)
685 * | issecure_mask(SECURE_NOROOT)
686 * | issecure_mask(SECURE_NOROOT_LOCKED)
687 * | issecure_mask(SECURE_NO_SETUID_FIXUP)
688 * | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
690 * will ensure that the current process and all of its
691 * children will be locked into a pure
692 * capability-based-privilege environment.
694 case PR_SET_SECUREBITS:
695 if ((((cred->securebits & SECURE_ALL_LOCKS) >> 1)
696 & (cred->securebits ^ arg2)) /*[1]*/
697 || ((cred->securebits & SECURE_ALL_LOCKS
699 || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS)) /*[3]*/
700 || (cap_capable(current, CAP_SETPCAP, SECURITY_CAP_AUDIT) != 0)) { /*[4]*/
702 * [1] no changing of bits that are locked
703 * [2] no unlocking of locks
704 * [3] no setting of unsupported bits
705 * [4] doing anything requires privilege (go read about
706 * the "sendmail capabilities bug")
708 error = -EPERM; /* cannot change a locked bit */
710 cred->securebits = arg2;
713 case PR_GET_SECUREBITS:
714 error = cred->securebits;
717 #endif /* def CONFIG_SECURITY_FILE_CAPABILITIES */
719 case PR_GET_KEEPCAPS:
720 if (issecure(SECURE_KEEP_CAPS))
723 case PR_SET_KEEPCAPS:
724 if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
726 else if (issecure(SECURE_KEEP_CAPS_LOCKED))
729 cred->securebits |= issecure_mask(SECURE_KEEP_CAPS);
731 cred->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
735 /* No functionality available - continue with default */
739 /* Functionality provided */
744 void cap_task_reparent_to_init (struct task_struct *p)
746 struct cred *cred = p->cred;
748 cap_set_init_eff(cred->cap_effective);
749 cap_clear(cred->cap_inheritable);
750 cap_set_full(cred->cap_permitted);
751 p->cred->securebits = SECUREBITS_DEFAULT;
754 int cap_syslog (int type)
756 if ((type != 3 && type != 10) && !capable(CAP_SYS_ADMIN))
761 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
763 int cap_sys_admin = 0;
765 if (cap_capable(current, CAP_SYS_ADMIN, SECURITY_CAP_NOAUDIT) == 0)
767 return __vm_enough_memory(mm, pages, cap_sys_admin);