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/sysproto.h>
74 #include <sys/sysctl.h>
75 #include <sys/systm.h>
76 #include <sys/ucred.h>
78 #include <sys/ktrace.h>
80 #include <security/audit/audit.h>
85 #ifdef CAPABILITY_MODE
87 FEATURE(security_capability_mode, "Capsicum Capability Mode");
90 * System call to enter capability mode for the process.
93 sys_cap_enter(struct thread *td, struct cap_enter_args *uap)
95 struct ucred *newcred, *oldcred;
98 if (IN_CAPABILITY_MODE(td))
104 oldcred = p->p_ucred;
105 crcopy(newcred, oldcred);
106 newcred->cr_flags |= CRED_FLAG_CAPMODE;
107 p->p_ucred = newcred;
114 * System call to query whether the process is in capability mode.
117 sys_cap_getmode(struct thread *td, struct cap_getmode_args *uap)
121 i = IN_CAPABILITY_MODE(td) ? 1 : 0;
122 return (copyout(&i, uap->modep, sizeof(i)));
125 #else /* !CAPABILITY_MODE */
128 sys_cap_enter(struct thread *td, struct cap_enter_args *uap)
135 sys_cap_getmode(struct thread *td, struct cap_getmode_args *uap)
141 #endif /* CAPABILITY_MODE */
145 FEATURE(security_capabilities, "Capsicum Capabilities");
148 _cap_check(cap_rights_t have, cap_rights_t need, enum ktr_cap_fail_type type)
152 if ((need & ~have) != 0) {
154 if (KTRPOINT(curthread, KTR_CAPFAIL))
155 ktrcapfail(type, need, have);
157 return (ENOTCAPABLE);
163 * Test whether a capability grants the requested rights.
166 cap_check(cap_rights_t have, cap_rights_t need)
169 return (_cap_check(have, need, CAPFAIL_NOTCAPABLE));
173 * Convert capability rights into VM access flags.
176 cap_rights_to_vmprot(cap_rights_t have)
180 maxprot = VM_PROT_NONE;
181 if (have & CAP_MMAP_R)
182 maxprot |= VM_PROT_READ;
183 if (have & CAP_MMAP_W)
184 maxprot |= VM_PROT_WRITE;
185 if (have & CAP_MMAP_X)
186 maxprot |= VM_PROT_EXECUTE;
192 * Extract rights from a capability for monitoring purposes -- not for use in
193 * any other way, as we want to keep all capability permission evaluation in
197 cap_rights(struct filedesc *fdp, int fd)
200 return (fdp->fd_ofiles[fd].fde_rights);
204 * System call to limit rights of the given capability.
207 sys_cap_rights_limit(struct thread *td, struct cap_rights_limit_args *uap)
209 struct filedesc *fdp;
214 rights = uap->rights;
217 AUDIT_ARG_RIGHTS(rights);
219 if ((rights & ~CAP_ALL) != 0)
222 fdp = td->td_proc->p_fd;
224 if (fget_locked(fdp, fd) == NULL) {
225 FILEDESC_XUNLOCK(fdp);
228 error = _cap_check(cap_rights(fdp, fd), rights, CAPFAIL_INCREASE);
230 fdp->fd_ofiles[fd].fde_rights = rights;
231 if ((rights & CAP_IOCTL) == 0) {
232 free(fdp->fd_ofiles[fd].fde_ioctls, M_TEMP);
233 fdp->fd_ofiles[fd].fde_ioctls = NULL;
234 fdp->fd_ofiles[fd].fde_nioctls = 0;
236 if ((rights & CAP_FCNTL) == 0)
237 fdp->fd_ofiles[fd].fde_fcntls = 0;
239 FILEDESC_XUNLOCK(fdp);
244 * System call to query the rights mask associated with a capability.
247 sys_cap_rights_get(struct thread *td, struct cap_rights_get_args *uap)
249 struct filedesc *fdp;
257 fdp = td->td_proc->p_fd;
259 if (fget_locked(fdp, fd) == NULL) {
260 FILEDESC_SUNLOCK(fdp);
263 rights = cap_rights(fdp, fd);
264 FILEDESC_SUNLOCK(fdp);
265 return (copyout(&rights, uap->rightsp, sizeof(*uap->rightsp)));
269 * Test whether a capability grants the given ioctl command.
270 * If descriptor doesn't have CAP_IOCTL, then ioctls list is empty and
271 * ENOTCAPABLE will be returned.
274 cap_ioctl_check(struct filedesc *fdp, int fd, u_long cmd)
280 FILEDESC_LOCK_ASSERT(fdp);
281 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
282 ("%s: invalid fd=%d", __func__, fd));
284 ncmds = fdp->fd_ofiles[fd].fde_nioctls;
288 cmds = fdp->fd_ofiles[fd].fde_ioctls;
289 for (i = 0; i < ncmds; i++) {
294 return (ENOTCAPABLE);
298 * Check if the current ioctls list can be replaced by the new one.
301 cap_ioctl_limit_check(struct filedesc *fdp, int fd, const u_long *cmds,
309 oncmds = fdp->fd_ofiles[fd].fde_nioctls;
312 if (oncmds < (ssize_t)ncmds)
313 return (ENOTCAPABLE);
315 ocmds = fdp->fd_ofiles[fd].fde_ioctls;
316 for (i = 0; i < ncmds; i++) {
317 for (j = 0; j < oncmds; j++) {
318 if (cmds[i] == ocmds[j])
322 return (ENOTCAPABLE);
329 sys_cap_ioctls_limit(struct thread *td, struct cap_ioctls_limit_args *uap)
331 struct filedesc *fdp;
332 u_long *cmds, *ocmds;
341 if (ncmds > 256) /* XXX: Is 256 sane? */
347 cmds = malloc(sizeof(cmds[0]) * ncmds, M_TEMP, M_WAITOK);
348 error = copyin(uap->cmds, cmds, sizeof(cmds[0]) * ncmds);
355 fdp = td->td_proc->p_fd;
358 if (fget_locked(fdp, fd) == NULL) {
363 error = cap_ioctl_limit_check(fdp, fd, cmds, ncmds);
367 ocmds = fdp->fd_ofiles[fd].fde_ioctls;
368 fdp->fd_ofiles[fd].fde_ioctls = cmds;
369 fdp->fd_ofiles[fd].fde_nioctls = ncmds;
374 FILEDESC_XUNLOCK(fdp);
380 sys_cap_ioctls_get(struct thread *td, struct cap_ioctls_get_args *uap)
382 struct filedesc *fdp;
383 struct filedescent *fdep;
390 maxcmds = uap->maxcmds;
394 fdp = td->td_proc->p_fd;
397 if (fget_locked(fdp, fd) == NULL) {
403 * If all ioctls are allowed (fde_nioctls == -1 && fde_ioctls == NULL)
404 * the only sane thing we can do is to not populate the given array and
405 * return CAP_IOCTLS_ALL.
408 fdep = &fdp->fd_ofiles[fd];
409 if (cmds != NULL && fdep->fde_ioctls != NULL) {
410 error = copyout(fdep->fde_ioctls, cmds,
411 sizeof(cmds[0]) * MIN(fdep->fde_nioctls, maxcmds));
415 if (fdep->fde_nioctls == -1)
416 td->td_retval[0] = CAP_IOCTLS_ALL;
418 td->td_retval[0] = fdep->fde_nioctls;
422 FILEDESC_SUNLOCK(fdp);
427 * Test whether a capability grants the given fcntl command.
430 cap_fcntl_check(struct filedesc *fdp, int fd, int cmd)
434 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
435 ("%s: invalid fd=%d", __func__, fd));
437 fcntlcap = (1 << cmd);
438 KASSERT((CAP_FCNTL_ALL & fcntlcap) != 0,
439 ("Unsupported fcntl=%d.", cmd));
441 if ((fdp->fd_ofiles[fd].fde_fcntls & fcntlcap) != 0)
444 return (ENOTCAPABLE);
448 sys_cap_fcntls_limit(struct thread *td, struct cap_fcntls_limit_args *uap)
450 struct filedesc *fdp;
451 uint32_t fcntlrights;
455 fcntlrights = uap->fcntlrights;
458 AUDIT_ARG_FCNTL_RIGHTS(fcntlrights);
460 if ((fcntlrights & ~CAP_FCNTL_ALL) != 0)
463 fdp = td->td_proc->p_fd;
466 if (fget_locked(fdp, fd) == NULL) {
467 FILEDESC_XUNLOCK(fdp);
471 if ((fcntlrights & ~fdp->fd_ofiles[fd].fde_fcntls) != 0) {
472 FILEDESC_XUNLOCK(fdp);
473 return (ENOTCAPABLE);
476 fdp->fd_ofiles[fd].fde_fcntls = fcntlrights;
477 FILEDESC_XUNLOCK(fdp);
483 sys_cap_fcntls_get(struct thread *td, struct cap_fcntls_get_args *uap)
485 struct filedesc *fdp;
493 fdp = td->td_proc->p_fd;
495 if (fget_locked(fdp, fd) == NULL) {
496 FILEDESC_SUNLOCK(fdp);
499 rights = fdp->fd_ofiles[fd].fde_fcntls;
500 FILEDESC_SUNLOCK(fdp);
502 return (copyout(&rights, uap->fcntlrightsp, sizeof(rights)));
506 * For backward compatibility.
509 sys_cap_new(struct thread *td, struct cap_new_args *uap)
511 struct filedesc *fdp;
517 rights = uap->rights;
520 AUDIT_ARG_RIGHTS(rights);
522 if ((rights & ~CAP_ALL) != 0)
525 fdp = td->td_proc->p_fd;
527 if (fget_locked(fdp, fd) == NULL) {
528 FILEDESC_SUNLOCK(fdp);
531 error = _cap_check(cap_rights(fdp, fd), rights, CAPFAIL_INCREASE);
532 FILEDESC_SUNLOCK(fdp);
536 error = do_dup(td, 0, fd, 0, &newfd);
542 * We don't really care about the race between checking capability
543 * rights for the source descriptor and now. If capability rights
544 * were ok at that earlier point, the process had this descriptor
545 * with those rights, so we don't increase them in security sense,
546 * the process might have done the cap_new(2) a bit earlier to get
549 fdp->fd_ofiles[newfd].fde_rights = rights;
550 if ((rights & CAP_IOCTL) == 0) {
551 free(fdp->fd_ofiles[newfd].fde_ioctls, M_TEMP);
552 fdp->fd_ofiles[newfd].fde_ioctls = NULL;
553 fdp->fd_ofiles[newfd].fde_nioctls = 0;
555 if ((rights & CAP_FCNTL) == 0)
556 fdp->fd_ofiles[newfd].fde_fcntls = 0;
557 FILEDESC_XUNLOCK(fdp);
559 td->td_retval[0] = newfd;
564 #else /* !CAPABILITIES */
567 * Stub Capability functions for when options CAPABILITIES isn't compiled
572 sys_cap_rights_limit(struct thread *td, struct cap_rights_limit_args *uap)
579 sys_cap_rights_get(struct thread *td, struct cap_rights_get_args *uap)
586 sys_cap_ioctls_limit(struct thread *td, struct cap_ioctls_limit_args *uap)
593 sys_cap_ioctls_get(struct thread *td, struct cap_ioctls_get_args *uap)
600 sys_cap_fcntls_limit(struct thread *td, struct cap_fcntls_limit_args *uap)
607 sys_cap_fcntls_get(struct thread *td, struct cap_fcntls_get_args *uap)
614 sys_cap_new(struct thread *td, struct cap_new_args *uap)
620 #endif /* CAPABILITIES */