2 * Copyright (c) 2008-2011 Robert N. M. Watson
3 * Copyright (c) 2010-2011 Jonathan Anderson
4 * Copyright (c) 2012 FreeBSD Foundation
7 * This software was developed at the University of Cambridge Computer
8 * Laboratory with support from a grant from Google, Inc.
10 * Portions of this software were developed by Pawel Jakub Dawidek under
11 * sponsorship from the FreeBSD Foundation.
13 * Redistribution and use in source and binary forms, with or without
14 * modification, are permitted provided that the following conditions
16 * 1. Redistributions of source code must retain the above copyright
17 * notice, this list of conditions and the following disclaimer.
18 * 2. Redistributions in binary form must reproduce the above copyright
19 * notice, this list of conditions and the following disclaimer in the
20 * documentation and/or other materials provided with the distribution.
22 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * FreeBSD kernel capability facility.
38 * Two kernel features are implemented here: capability mode, a sandboxed mode
39 * of execution for processes, and capabilities, a refinement on file
40 * descriptors that allows fine-grained control over operations on the file
41 * descriptor. Collectively, these allow processes to run in the style of a
42 * historic "capability system" in which they can use only resources
43 * explicitly delegated to them. This model is enforced by restricting access
44 * to global namespaces in capability mode.
46 * Capabilities wrap other file descriptor types, binding them to a constant
47 * rights mask set when the capability is created. New capabilities may be
48 * derived from existing capabilities, but only if they have the same or a
49 * strict subset of the rights on the original capability.
51 * System calls permitted in capability mode are defined in capabilities.conf;
52 * calls must be carefully audited for safety to ensure that they don't allow
53 * escape from a sandbox. Some calls permit only a subset of operations in
54 * capability mode -- for example, shm_open(2) is limited to creating
55 * anonymous, rather than named, POSIX shared memory objects.
58 #include <sys/cdefs.h>
59 __FBSDID("$FreeBSD$");
61 #include "opt_capsicum.h"
62 #include "opt_ktrace.h"
64 #include <sys/param.h>
65 #include <sys/capability.h>
67 #include <sys/filedesc.h>
68 #include <sys/kernel.h>
69 #include <sys/limits.h>
71 #include <sys/mutex.h>
73 #include <sys/syscallsubr.h>
74 #include <sys/sysproto.h>
75 #include <sys/sysctl.h>
76 #include <sys/systm.h>
77 #include <sys/ucred.h>
79 #include <sys/ktrace.h>
81 #include <security/audit/audit.h>
86 #ifdef CAPABILITY_MODE
88 FEATURE(security_capability_mode, "Capsicum Capability Mode");
91 * System call to enter capability mode for the process.
94 sys_cap_enter(struct thread *td, struct cap_enter_args *uap)
96 struct ucred *newcred, *oldcred;
99 if (IN_CAPABILITY_MODE(td))
105 oldcred = p->p_ucred;
106 crcopy(newcred, oldcred);
107 newcred->cr_flags |= CRED_FLAG_CAPMODE;
108 p->p_ucred = newcred;
115 * System call to query whether the process is in capability mode.
118 sys_cap_getmode(struct thread *td, struct cap_getmode_args *uap)
122 i = IN_CAPABILITY_MODE(td) ? 1 : 0;
123 return (copyout(&i, uap->modep, sizeof(i)));
126 #else /* !CAPABILITY_MODE */
129 sys_cap_enter(struct thread *td, struct cap_enter_args *uap)
136 sys_cap_getmode(struct thread *td, struct cap_getmode_args *uap)
142 #endif /* CAPABILITY_MODE */
146 FEATURE(security_capabilities, "Capsicum Capabilities");
148 MALLOC_DECLARE(M_FILECAPS);
151 _cap_check(cap_rights_t have, cap_rights_t need, enum ktr_cap_fail_type type)
155 if ((need & ~have) != 0) {
157 if (KTRPOINT(curthread, KTR_CAPFAIL))
158 ktrcapfail(type, need, have);
160 return (ENOTCAPABLE);
166 * Test whether a capability grants the requested rights.
169 cap_check(cap_rights_t have, cap_rights_t need)
172 return (_cap_check(have, need, CAPFAIL_NOTCAPABLE));
176 * Convert capability rights into VM access flags.
179 cap_rights_to_vmprot(cap_rights_t have)
183 maxprot = VM_PROT_NONE;
184 if (have & CAP_MMAP_R)
185 maxprot |= VM_PROT_READ;
186 if (have & CAP_MMAP_W)
187 maxprot |= VM_PROT_WRITE;
188 if (have & CAP_MMAP_X)
189 maxprot |= VM_PROT_EXECUTE;
195 * Extract rights from a capability for monitoring purposes -- not for use in
196 * any other way, as we want to keep all capability permission evaluation in
200 cap_rights(struct filedesc *fdp, int fd)
203 return (fdp->fd_ofiles[fd].fde_rights);
207 * System call to limit rights of the given capability.
210 sys_cap_rights_limit(struct thread *td, struct cap_rights_limit_args *uap)
212 struct filedesc *fdp;
217 rights = uap->rights;
220 AUDIT_ARG_RIGHTS(rights);
222 if ((rights & ~CAP_ALL) != 0)
225 fdp = td->td_proc->p_fd;
227 if (fget_locked(fdp, fd) == NULL) {
228 FILEDESC_XUNLOCK(fdp);
231 error = _cap_check(cap_rights(fdp, fd), rights, CAPFAIL_INCREASE);
233 fdp->fd_ofiles[fd].fde_rights = rights;
234 if ((rights & CAP_IOCTL) == 0) {
235 free(fdp->fd_ofiles[fd].fde_ioctls, M_FILECAPS);
236 fdp->fd_ofiles[fd].fde_ioctls = NULL;
237 fdp->fd_ofiles[fd].fde_nioctls = 0;
239 if ((rights & CAP_FCNTL) == 0)
240 fdp->fd_ofiles[fd].fde_fcntls = 0;
242 FILEDESC_XUNLOCK(fdp);
247 * System call to query the rights mask associated with a capability.
250 sys_cap_rights_get(struct thread *td, struct cap_rights_get_args *uap)
252 struct filedesc *fdp;
260 fdp = td->td_proc->p_fd;
262 if (fget_locked(fdp, fd) == NULL) {
263 FILEDESC_SUNLOCK(fdp);
266 rights = cap_rights(fdp, fd);
267 FILEDESC_SUNLOCK(fdp);
268 return (copyout(&rights, uap->rightsp, sizeof(*uap->rightsp)));
272 * Test whether a capability grants the given ioctl command.
273 * If descriptor doesn't have CAP_IOCTL, then ioctls list is empty and
274 * ENOTCAPABLE will be returned.
277 cap_ioctl_check(struct filedesc *fdp, int fd, u_long cmd)
283 FILEDESC_LOCK_ASSERT(fdp);
284 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
285 ("%s: invalid fd=%d", __func__, fd));
287 ncmds = fdp->fd_ofiles[fd].fde_nioctls;
291 cmds = fdp->fd_ofiles[fd].fde_ioctls;
292 for (i = 0; i < ncmds; i++) {
297 return (ENOTCAPABLE);
301 * Check if the current ioctls list can be replaced by the new one.
304 cap_ioctl_limit_check(struct filedesc *fdp, int fd, const u_long *cmds,
312 oncmds = fdp->fd_ofiles[fd].fde_nioctls;
315 if (oncmds < (ssize_t)ncmds)
316 return (ENOTCAPABLE);
318 ocmds = fdp->fd_ofiles[fd].fde_ioctls;
319 for (i = 0; i < ncmds; i++) {
320 for (j = 0; j < oncmds; j++) {
321 if (cmds[i] == ocmds[j])
325 return (ENOTCAPABLE);
332 kern_cap_ioctls_limit(struct thread *td, int fd, u_long *cmds, size_t ncmds)
334 struct filedesc *fdp;
340 fdp = td->td_proc->p_fd;
343 if (fget_locked(fdp, fd) == NULL) {
348 error = cap_ioctl_limit_check(fdp, fd, cmds, ncmds);
352 ocmds = fdp->fd_ofiles[fd].fde_ioctls;
353 fdp->fd_ofiles[fd].fde_ioctls = cmds;
354 fdp->fd_ofiles[fd].fde_nioctls = ncmds;
359 FILEDESC_XUNLOCK(fdp);
360 free(cmds, M_FILECAPS);
365 sys_cap_ioctls_limit(struct thread *td, struct cap_ioctls_limit_args *uap)
373 if (ncmds > 256) /* XXX: Is 256 sane? */
379 cmds = malloc(sizeof(cmds[0]) * ncmds, M_FILECAPS, M_WAITOK);
380 error = copyin(uap->cmds, cmds, sizeof(cmds[0]) * ncmds);
382 free(cmds, M_FILECAPS);
387 return (kern_cap_ioctls_limit(td, uap->fd, cmds, ncmds));
391 sys_cap_ioctls_get(struct thread *td, struct cap_ioctls_get_args *uap)
393 struct filedesc *fdp;
394 struct filedescent *fdep;
401 maxcmds = uap->maxcmds;
405 fdp = td->td_proc->p_fd;
408 if (fget_locked(fdp, fd) == NULL) {
414 * If all ioctls are allowed (fde_nioctls == -1 && fde_ioctls == NULL)
415 * the only sane thing we can do is to not populate the given array and
416 * return CAP_IOCTLS_ALL.
419 fdep = &fdp->fd_ofiles[fd];
420 if (cmds != NULL && fdep->fde_ioctls != NULL) {
421 error = copyout(fdep->fde_ioctls, cmds,
422 sizeof(cmds[0]) * MIN(fdep->fde_nioctls, maxcmds));
426 if (fdep->fde_nioctls == -1)
427 td->td_retval[0] = CAP_IOCTLS_ALL;
429 td->td_retval[0] = fdep->fde_nioctls;
433 FILEDESC_SUNLOCK(fdp);
438 * Test whether a capability grants the given fcntl command.
441 cap_fcntl_check(struct filedesc *fdp, int fd, int cmd)
445 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
446 ("%s: invalid fd=%d", __func__, fd));
448 fcntlcap = (1 << cmd);
449 KASSERT((CAP_FCNTL_ALL & fcntlcap) != 0,
450 ("Unsupported fcntl=%d.", cmd));
452 if ((fdp->fd_ofiles[fd].fde_fcntls & fcntlcap) != 0)
455 return (ENOTCAPABLE);
459 sys_cap_fcntls_limit(struct thread *td, struct cap_fcntls_limit_args *uap)
461 struct filedesc *fdp;
462 uint32_t fcntlrights;
466 fcntlrights = uap->fcntlrights;
469 AUDIT_ARG_FCNTL_RIGHTS(fcntlrights);
471 if ((fcntlrights & ~CAP_FCNTL_ALL) != 0)
474 fdp = td->td_proc->p_fd;
477 if (fget_locked(fdp, fd) == NULL) {
478 FILEDESC_XUNLOCK(fdp);
482 if ((fcntlrights & ~fdp->fd_ofiles[fd].fde_fcntls) != 0) {
483 FILEDESC_XUNLOCK(fdp);
484 return (ENOTCAPABLE);
487 fdp->fd_ofiles[fd].fde_fcntls = fcntlrights;
488 FILEDESC_XUNLOCK(fdp);
494 sys_cap_fcntls_get(struct thread *td, struct cap_fcntls_get_args *uap)
496 struct filedesc *fdp;
504 fdp = td->td_proc->p_fd;
506 if (fget_locked(fdp, fd) == NULL) {
507 FILEDESC_SUNLOCK(fdp);
510 rights = fdp->fd_ofiles[fd].fde_fcntls;
511 FILEDESC_SUNLOCK(fdp);
513 return (copyout(&rights, uap->fcntlrightsp, sizeof(rights)));
517 * For backward compatibility.
520 sys_cap_new(struct thread *td, struct cap_new_args *uap)
522 struct filedesc *fdp;
528 rights = uap->rights;
531 AUDIT_ARG_RIGHTS(rights);
533 if ((rights & ~CAP_ALL) != 0)
536 fdp = td->td_proc->p_fd;
538 if (fget_locked(fdp, fd) == NULL) {
539 FILEDESC_SUNLOCK(fdp);
542 error = _cap_check(cap_rights(fdp, fd), rights, CAPFAIL_INCREASE);
543 FILEDESC_SUNLOCK(fdp);
547 error = do_dup(td, 0, fd, 0, &newfd);
553 * We don't really care about the race between checking capability
554 * rights for the source descriptor and now. If capability rights
555 * were ok at that earlier point, the process had this descriptor
556 * with those rights, so we don't increase them in security sense,
557 * the process might have done the cap_new(2) a bit earlier to get
560 fdp->fd_ofiles[newfd].fde_rights = rights;
561 if ((rights & CAP_IOCTL) == 0) {
562 free(fdp->fd_ofiles[newfd].fde_ioctls, M_FILECAPS);
563 fdp->fd_ofiles[newfd].fde_ioctls = NULL;
564 fdp->fd_ofiles[newfd].fde_nioctls = 0;
566 if ((rights & CAP_FCNTL) == 0)
567 fdp->fd_ofiles[newfd].fde_fcntls = 0;
568 FILEDESC_XUNLOCK(fdp);
570 td->td_retval[0] = newfd;
575 #else /* !CAPABILITIES */
578 * Stub Capability functions for when options CAPABILITIES isn't compiled
583 sys_cap_rights_limit(struct thread *td, struct cap_rights_limit_args *uap)
590 sys_cap_rights_get(struct thread *td, struct cap_rights_get_args *uap)
597 sys_cap_ioctls_limit(struct thread *td, struct cap_ioctls_limit_args *uap)
604 sys_cap_ioctls_get(struct thread *td, struct cap_ioctls_get_args *uap)
611 sys_cap_fcntls_limit(struct thread *td, struct cap_fcntls_limit_args *uap)
618 sys_cap_fcntls_get(struct thread *td, struct cap_fcntls_get_args *uap)
625 sys_cap_new(struct thread *td, struct cap_new_args *uap)
631 #endif /* CAPABILITIES */