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(const kernel_cap_t *effective,
122 const kernel_cap_t *inheritable,
123 const kernel_cap_t *permitted)
125 if (cap_inh_is_capped()
126 && !cap_issubset(*inheritable,
127 cap_combine(current->cap_inheritable,
128 current->cap_permitted))) {
129 /* incapable of using this inheritable set */
132 if (!cap_issubset(*inheritable,
133 cap_combine(current->cap_inheritable,
134 current->cap_bset))) {
135 /* no new pI capabilities outside bounding set */
139 /* verify restrictions on target's new Permitted set */
140 if (!cap_issubset (*permitted,
141 cap_combine (current->cap_permitted,
142 current->cap_permitted))) {
146 /* verify the _new_Effective_ is a subset of the _new_Permitted_ */
147 if (!cap_issubset (*effective, *permitted)) {
154 void cap_capset_set(const kernel_cap_t *effective,
155 const kernel_cap_t *inheritable,
156 const kernel_cap_t *permitted)
158 current->cap_effective = *effective;
159 current->cap_inheritable = *inheritable;
160 current->cap_permitted = *permitted;
163 static inline void bprm_clear_caps(struct linux_binprm *bprm)
165 cap_clear(bprm->cap_post_exec_permitted);
166 bprm->cap_effective = false;
169 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
171 int cap_inode_need_killpriv(struct dentry *dentry)
173 struct inode *inode = dentry->d_inode;
176 if (!inode->i_op || !inode->i_op->getxattr)
179 error = inode->i_op->getxattr(dentry, XATTR_NAME_CAPS, NULL, 0);
185 int cap_inode_killpriv(struct dentry *dentry)
187 struct inode *inode = dentry->d_inode;
189 if (!inode->i_op || !inode->i_op->removexattr)
192 return inode->i_op->removexattr(dentry, XATTR_NAME_CAPS);
195 static inline int bprm_caps_from_vfs_caps(struct cpu_vfs_cap_data *caps,
196 struct linux_binprm *bprm)
201 if (caps->magic_etc & VFS_CAP_FLAGS_EFFECTIVE)
202 bprm->cap_effective = true;
204 bprm->cap_effective = false;
206 CAP_FOR_EACH_U32(i) {
207 __u32 permitted = caps->permitted.cap[i];
208 __u32 inheritable = caps->inheritable.cap[i];
211 * pP' = (X & fP) | (pI & fI)
213 bprm->cap_post_exec_permitted.cap[i] =
214 (current->cap_bset.cap[i] & permitted) |
215 (current->cap_inheritable.cap[i] & inheritable);
217 if (permitted & ~bprm->cap_post_exec_permitted.cap[i]) {
219 * insufficient to execute correctly
226 * For legacy apps, with no internal support for recognizing they
227 * do not have enough capabilities, we return an error if they are
228 * missing some "forced" (aka file-permitted) capabilities.
230 return bprm->cap_effective ? ret : 0;
233 int get_vfs_caps_from_disk(const struct dentry *dentry, struct cpu_vfs_cap_data *cpu_caps)
235 struct inode *inode = dentry->d_inode;
239 struct vfs_cap_data caps;
241 memset(cpu_caps, 0, sizeof(struct cpu_vfs_cap_data));
243 if (!inode || !inode->i_op || !inode->i_op->getxattr)
246 size = inode->i_op->getxattr((struct dentry *)dentry, XATTR_NAME_CAPS, &caps,
248 if (size == -ENODATA || size == -EOPNOTSUPP) {
249 /* no data, that's ok */
255 if (size < sizeof(magic_etc))
258 cpu_caps->magic_etc = magic_etc = le32_to_cpu(caps.magic_etc);
260 switch ((magic_etc & VFS_CAP_REVISION_MASK)) {
261 case VFS_CAP_REVISION_1:
262 if (size != XATTR_CAPS_SZ_1)
264 tocopy = VFS_CAP_U32_1;
266 case VFS_CAP_REVISION_2:
267 if (size != XATTR_CAPS_SZ_2)
269 tocopy = VFS_CAP_U32_2;
275 CAP_FOR_EACH_U32(i) {
278 cpu_caps->permitted.cap[i] = le32_to_cpu(caps.data[i].permitted);
279 cpu_caps->inheritable.cap[i] = le32_to_cpu(caps.data[i].inheritable);
284 /* Locate any VFS capabilities: */
285 static int get_file_caps(struct linux_binprm *bprm)
287 struct dentry *dentry;
289 struct cpu_vfs_cap_data vcaps;
291 bprm_clear_caps(bprm);
293 if (!file_caps_enabled)
296 if (bprm->file->f_vfsmnt->mnt_flags & MNT_NOSUID)
299 dentry = dget(bprm->file->f_dentry);
301 rc = get_vfs_caps_from_disk(dentry, &vcaps);
304 printk(KERN_NOTICE "%s: get_vfs_caps_from_disk returned %d for %s\n",
305 __func__, rc, bprm->filename);
306 else if (rc == -ENODATA)
311 rc = bprm_caps_from_vfs_caps(&vcaps, bprm);
316 bprm_clear_caps(bprm);
322 int cap_inode_need_killpriv(struct dentry *dentry)
327 int cap_inode_killpriv(struct dentry *dentry)
332 static inline int get_file_caps(struct linux_binprm *bprm)
334 bprm_clear_caps(bprm);
339 int cap_bprm_set_security (struct linux_binprm *bprm)
343 ret = get_file_caps(bprm);
345 if (!issecure(SECURE_NOROOT)) {
347 * To support inheritance of root-permissions and suid-root
348 * executables under compatibility mode, we override the
349 * capability sets for the file.
351 * If only the real uid is 0, we do not set the effective
354 if (bprm->e_uid == 0 || current_uid() == 0) {
355 /* pP' = (cap_bset & ~0) | (pI & ~0) */
356 bprm->cap_post_exec_permitted = cap_combine(
357 current->cap_bset, current->cap_inheritable
359 bprm->cap_effective = (bprm->e_uid == 0);
367 void cap_bprm_apply_creds (struct linux_binprm *bprm, int unsafe)
369 kernel_cap_t pP = current->cap_permitted;
370 kernel_cap_t pE = current->cap_effective;
374 current_uid_gid(&uid, &gid);
376 if (bprm->e_uid != uid || bprm->e_gid != gid ||
377 !cap_issubset(bprm->cap_post_exec_permitted,
378 current->cap_permitted)) {
379 set_dumpable(current->mm, suid_dumpable);
380 current->pdeath_signal = 0;
382 if (unsafe & ~LSM_UNSAFE_PTRACE_CAP) {
383 if (!capable(CAP_SETUID)) {
387 if (cap_limit_ptraced_target()) {
388 bprm->cap_post_exec_permitted = cap_intersect(
389 bprm->cap_post_exec_permitted,
390 current->cap_permitted);
395 current->suid = current->euid = current->fsuid = bprm->e_uid;
396 current->sgid = current->egid = current->fsgid = bprm->e_gid;
398 /* For init, we want to retain the capabilities set
399 * in the init_task struct. Thus we skip the usual
400 * capability rules */
401 if (!is_global_init(current)) {
402 current->cap_permitted = bprm->cap_post_exec_permitted;
403 if (bprm->cap_effective)
404 current->cap_effective = bprm->cap_post_exec_permitted;
406 cap_clear(current->cap_effective);
410 * Audit candidate if current->cap_effective is set
412 * We do not bother to audit if 3 things are true:
413 * 1) cap_effective has all caps
415 * 3) root is supposed to have all caps (SECURE_NOROOT)
416 * Since this is just a normal root execing a process.
418 * Number 1 above might fail if you don't have a full bset, but I think
419 * that is interesting information to audit.
421 if (!cap_isclear(current->cap_effective)) {
422 if (!cap_issubset(CAP_FULL_SET, current->cap_effective) ||
423 (bprm->e_uid != 0) || (current->uid != 0) ||
424 issecure(SECURE_NOROOT))
425 audit_log_bprm_fcaps(bprm, &pP, &pE);
428 current->securebits &= ~issecure_mask(SECURE_KEEP_CAPS);
431 int cap_bprm_secureexec (struct linux_binprm *bprm)
433 if (current_uid() != 0) {
434 if (bprm->cap_effective)
436 if (!cap_isclear(bprm->cap_post_exec_permitted))
440 return (current_euid() != current_uid() ||
441 current_egid() != current_gid());
444 int cap_inode_setxattr(struct dentry *dentry, const char *name,
445 const void *value, size_t size, int flags)
447 if (!strcmp(name, XATTR_NAME_CAPS)) {
448 if (!capable(CAP_SETFCAP))
451 } else if (!strncmp(name, XATTR_SECURITY_PREFIX,
452 sizeof(XATTR_SECURITY_PREFIX) - 1) &&
453 !capable(CAP_SYS_ADMIN))
458 int cap_inode_removexattr(struct dentry *dentry, const char *name)
460 if (!strcmp(name, XATTR_NAME_CAPS)) {
461 if (!capable(CAP_SETFCAP))
464 } else if (!strncmp(name, XATTR_SECURITY_PREFIX,
465 sizeof(XATTR_SECURITY_PREFIX) - 1) &&
466 !capable(CAP_SYS_ADMIN))
471 /* moved from kernel/sys.c. */
473 * cap_emulate_setxuid() fixes the effective / permitted capabilities of
474 * a process after a call to setuid, setreuid, or setresuid.
476 * 1) When set*uiding _from_ one of {r,e,s}uid == 0 _to_ all of
477 * {r,e,s}uid != 0, the permitted and effective capabilities are
480 * 2) When set*uiding _from_ euid == 0 _to_ euid != 0, the effective
481 * capabilities of the process are cleared.
483 * 3) When set*uiding _from_ euid != 0 _to_ euid == 0, the effective
484 * capabilities are set to the permitted capabilities.
486 * fsuid is handled elsewhere. fsuid == 0 and {r,e,s}uid!= 0 should
491 * cevans - New behaviour, Oct '99
492 * A process may, via prctl(), elect to keep its capabilities when it
493 * calls setuid() and switches away from uid==0. Both permitted and
494 * effective sets will be retained.
495 * Without this change, it was impossible for a daemon to drop only some
496 * of its privilege. The call to setuid(!=0) would drop all privileges!
497 * Keeping uid 0 is not an option because uid 0 owns too many vital
499 * Thanks to Olaf Kirch and Peter Benie for spotting this.
501 static inline void cap_emulate_setxuid (int old_ruid, int old_euid,
504 uid_t euid = current_euid();
506 if ((old_ruid == 0 || old_euid == 0 || old_suid == 0) &&
507 (current_uid() != 0 && euid != 0 && current_suid() != 0) &&
508 !issecure(SECURE_KEEP_CAPS)) {
509 cap_clear (current->cap_permitted);
510 cap_clear (current->cap_effective);
512 if (old_euid == 0 && euid != 0) {
513 cap_clear (current->cap_effective);
515 if (old_euid != 0 && euid == 0) {
516 current->cap_effective = current->cap_permitted;
520 int cap_task_post_setuid (uid_t old_ruid, uid_t old_euid, uid_t old_suid,
527 /* Copied from kernel/sys.c:setreuid/setuid/setresuid. */
528 if (!issecure (SECURE_NO_SETUID_FIXUP)) {
529 cap_emulate_setxuid (old_ruid, old_euid, old_suid);
534 uid_t old_fsuid = old_ruid;
536 /* Copied from kernel/sys.c:setfsuid. */
539 * FIXME - is fsuser used for all CAP_FS_MASK capabilities?
540 * if not, we might be a bit too harsh here.
543 if (!issecure (SECURE_NO_SETUID_FIXUP)) {
544 if (old_fsuid == 0 && current_fsuid() != 0) {
545 current->cap_effective =
547 current->cap_effective);
549 if (old_fsuid != 0 && current_fsuid() == 0) {
550 current->cap_effective =
552 current->cap_effective,
553 current->cap_permitted);
565 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
567 * Rationale: code calling task_setscheduler, task_setioprio, and
568 * task_setnice, assumes that
569 * . if capable(cap_sys_nice), then those actions should be allowed
570 * . if not capable(cap_sys_nice), but acting on your own processes,
571 * then those actions should be allowed
572 * This is insufficient now since you can call code without suid, but
573 * yet with increased caps.
574 * So we check for increased caps on the target process.
576 static int cap_safe_nice(struct task_struct *p)
578 if (!cap_issubset(p->cap_permitted, current->cap_permitted) &&
579 !capable(CAP_SYS_NICE))
584 int cap_task_setscheduler (struct task_struct *p, int policy,
585 struct sched_param *lp)
587 return cap_safe_nice(p);
590 int cap_task_setioprio (struct task_struct *p, int ioprio)
592 return cap_safe_nice(p);
595 int cap_task_setnice (struct task_struct *p, int nice)
597 return cap_safe_nice(p);
601 * called from kernel/sys.c for prctl(PR_CABSET_DROP)
602 * done without task_capability_lock() because it introduces
603 * no new races - i.e. only another task doing capget() on
604 * this task could get inconsistent info. There can be no
605 * racing writer bc a task can only change its own caps.
607 static long cap_prctl_drop(unsigned long cap)
609 if (!capable(CAP_SETPCAP))
613 cap_lower(current->cap_bset, cap);
618 int cap_task_setscheduler (struct task_struct *p, int policy,
619 struct sched_param *lp)
623 int cap_task_setioprio (struct task_struct *p, int ioprio)
627 int cap_task_setnice (struct task_struct *p, int nice)
633 int cap_task_prctl(int option, unsigned long arg2, unsigned long arg3,
634 unsigned long arg4, unsigned long arg5, long *rc_p)
639 case PR_CAPBSET_READ:
640 if (!cap_valid(arg2))
643 error = !!cap_raised(current->cap_bset, arg2);
645 #ifdef CONFIG_SECURITY_FILE_CAPABILITIES
646 case PR_CAPBSET_DROP:
647 error = cap_prctl_drop(arg2);
651 * The next four prctl's remain to assist with transitioning a
652 * system from legacy UID=0 based privilege (when filesystem
653 * capabilities are not in use) to a system using filesystem
654 * capabilities only - as the POSIX.1e draft intended.
658 * PR_SET_SECUREBITS =
659 * issecure_mask(SECURE_KEEP_CAPS_LOCKED)
660 * | issecure_mask(SECURE_NOROOT)
661 * | issecure_mask(SECURE_NOROOT_LOCKED)
662 * | issecure_mask(SECURE_NO_SETUID_FIXUP)
663 * | issecure_mask(SECURE_NO_SETUID_FIXUP_LOCKED)
665 * will ensure that the current process and all of its
666 * children will be locked into a pure
667 * capability-based-privilege environment.
669 case PR_SET_SECUREBITS:
670 if ((((current->securebits & SECURE_ALL_LOCKS) >> 1)
671 & (current->securebits ^ arg2)) /*[1]*/
672 || ((current->securebits & SECURE_ALL_LOCKS
674 || (arg2 & ~(SECURE_ALL_LOCKS | SECURE_ALL_BITS)) /*[3]*/
675 || (cap_capable(current, CAP_SETPCAP, SECURITY_CAP_AUDIT) != 0)) { /*[4]*/
677 * [1] no changing of bits that are locked
678 * [2] no unlocking of locks
679 * [3] no setting of unsupported bits
680 * [4] doing anything requires privilege (go read about
681 * the "sendmail capabilities bug")
683 error = -EPERM; /* cannot change a locked bit */
685 current->securebits = arg2;
688 case PR_GET_SECUREBITS:
689 error = current->securebits;
692 #endif /* def CONFIG_SECURITY_FILE_CAPABILITIES */
694 case PR_GET_KEEPCAPS:
695 if (issecure(SECURE_KEEP_CAPS))
698 case PR_SET_KEEPCAPS:
699 if (arg2 > 1) /* Note, we rely on arg2 being unsigned here */
701 else if (issecure(SECURE_KEEP_CAPS_LOCKED))
704 current->securebits |= issecure_mask(SECURE_KEEP_CAPS);
706 current->securebits &=
707 ~issecure_mask(SECURE_KEEP_CAPS);
711 /* No functionality available - continue with default */
715 /* Functionality provided */
720 void cap_task_reparent_to_init (struct task_struct *p)
722 cap_set_init_eff(p->cap_effective);
723 cap_clear(p->cap_inheritable);
724 cap_set_full(p->cap_permitted);
725 p->securebits = SECUREBITS_DEFAULT;
729 int cap_syslog (int type)
731 if ((type != 3 && type != 10) && !capable(CAP_SYS_ADMIN))
736 int cap_vm_enough_memory(struct mm_struct *mm, long pages)
738 int cap_sys_admin = 0;
740 if (cap_capable(current, CAP_SYS_ADMIN, SECURITY_CAP_NOAUDIT) == 0)
742 return __vm_enough_memory(mm, pages, cap_sys_admin);