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");
147 MALLOC_DECLARE(M_FILECAPS);
150 _cap_check(cap_rights_t have, cap_rights_t need, enum ktr_cap_fail_type type)
154 if ((need & ~have) != 0) {
156 if (KTRPOINT(curthread, KTR_CAPFAIL))
157 ktrcapfail(type, need, have);
159 return (ENOTCAPABLE);
165 * Test whether a capability grants the requested rights.
168 cap_check(cap_rights_t have, cap_rights_t need)
171 return (_cap_check(have, need, CAPFAIL_NOTCAPABLE));
175 * Convert capability rights into VM access flags.
178 cap_rights_to_vmprot(cap_rights_t have)
182 maxprot = VM_PROT_NONE;
183 if (have & CAP_MMAP_R)
184 maxprot |= VM_PROT_READ;
185 if (have & CAP_MMAP_W)
186 maxprot |= VM_PROT_WRITE;
187 if (have & CAP_MMAP_X)
188 maxprot |= VM_PROT_EXECUTE;
194 * Extract rights from a capability for monitoring purposes -- not for use in
195 * any other way, as we want to keep all capability permission evaluation in
199 cap_rights(struct filedesc *fdp, int fd)
202 return (fdp->fd_ofiles[fd].fde_rights);
206 * System call to limit rights of the given capability.
209 sys_cap_rights_limit(struct thread *td, struct cap_rights_limit_args *uap)
211 struct filedesc *fdp;
216 rights = uap->rights;
219 AUDIT_ARG_RIGHTS(rights);
221 if ((rights & ~CAP_ALL) != 0)
224 fdp = td->td_proc->p_fd;
226 if (fget_locked(fdp, fd) == NULL) {
227 FILEDESC_XUNLOCK(fdp);
230 error = _cap_check(cap_rights(fdp, fd), rights, CAPFAIL_INCREASE);
232 fdp->fd_ofiles[fd].fde_rights = rights;
233 if ((rights & CAP_IOCTL) == 0) {
234 free(fdp->fd_ofiles[fd].fde_ioctls, M_FILECAPS);
235 fdp->fd_ofiles[fd].fde_ioctls = NULL;
236 fdp->fd_ofiles[fd].fde_nioctls = 0;
238 if ((rights & CAP_FCNTL) == 0)
239 fdp->fd_ofiles[fd].fde_fcntls = 0;
241 FILEDESC_XUNLOCK(fdp);
246 * System call to query the rights mask associated with a capability.
249 sys_cap_rights_get(struct thread *td, struct cap_rights_get_args *uap)
251 struct filedesc *fdp;
259 fdp = td->td_proc->p_fd;
261 if (fget_locked(fdp, fd) == NULL) {
262 FILEDESC_SUNLOCK(fdp);
265 rights = cap_rights(fdp, fd);
266 FILEDESC_SUNLOCK(fdp);
267 return (copyout(&rights, uap->rightsp, sizeof(*uap->rightsp)));
271 * Test whether a capability grants the given ioctl command.
272 * If descriptor doesn't have CAP_IOCTL, then ioctls list is empty and
273 * ENOTCAPABLE will be returned.
276 cap_ioctl_check(struct filedesc *fdp, int fd, u_long cmd)
282 FILEDESC_LOCK_ASSERT(fdp);
283 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
284 ("%s: invalid fd=%d", __func__, fd));
286 ncmds = fdp->fd_ofiles[fd].fde_nioctls;
290 cmds = fdp->fd_ofiles[fd].fde_ioctls;
291 for (i = 0; i < ncmds; i++) {
296 return (ENOTCAPABLE);
300 * Check if the current ioctls list can be replaced by the new one.
303 cap_ioctl_limit_check(struct filedesc *fdp, int fd, const u_long *cmds,
311 oncmds = fdp->fd_ofiles[fd].fde_nioctls;
314 if (oncmds < (ssize_t)ncmds)
315 return (ENOTCAPABLE);
317 ocmds = fdp->fd_ofiles[fd].fde_ioctls;
318 for (i = 0; i < ncmds; i++) {
319 for (j = 0; j < oncmds; j++) {
320 if (cmds[i] == ocmds[j])
324 return (ENOTCAPABLE);
331 sys_cap_ioctls_limit(struct thread *td, struct cap_ioctls_limit_args *uap)
333 struct filedesc *fdp;
334 u_long *cmds, *ocmds;
343 if (ncmds > 256) /* XXX: Is 256 sane? */
349 cmds = malloc(sizeof(cmds[0]) * ncmds, M_FILECAPS, M_WAITOK);
350 error = copyin(uap->cmds, cmds, sizeof(cmds[0]) * ncmds);
352 free(cmds, M_FILECAPS);
357 fdp = td->td_proc->p_fd;
360 if (fget_locked(fdp, fd) == NULL) {
365 error = cap_ioctl_limit_check(fdp, fd, cmds, ncmds);
369 ocmds = fdp->fd_ofiles[fd].fde_ioctls;
370 fdp->fd_ofiles[fd].fde_ioctls = cmds;
371 fdp->fd_ofiles[fd].fde_nioctls = ncmds;
376 FILEDESC_XUNLOCK(fdp);
377 free(cmds, M_FILECAPS);
382 sys_cap_ioctls_get(struct thread *td, struct cap_ioctls_get_args *uap)
384 struct filedesc *fdp;
385 struct filedescent *fdep;
392 maxcmds = uap->maxcmds;
396 fdp = td->td_proc->p_fd;
399 if (fget_locked(fdp, fd) == NULL) {
405 * If all ioctls are allowed (fde_nioctls == -1 && fde_ioctls == NULL)
406 * the only sane thing we can do is to not populate the given array and
407 * return CAP_IOCTLS_ALL.
410 fdep = &fdp->fd_ofiles[fd];
411 if (cmds != NULL && fdep->fde_ioctls != NULL) {
412 error = copyout(fdep->fde_ioctls, cmds,
413 sizeof(cmds[0]) * MIN(fdep->fde_nioctls, maxcmds));
417 if (fdep->fde_nioctls == -1)
418 td->td_retval[0] = CAP_IOCTLS_ALL;
420 td->td_retval[0] = fdep->fde_nioctls;
424 FILEDESC_SUNLOCK(fdp);
429 * Test whether a capability grants the given fcntl command.
432 cap_fcntl_check(struct filedesc *fdp, int fd, int cmd)
436 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
437 ("%s: invalid fd=%d", __func__, fd));
439 fcntlcap = (1 << cmd);
440 KASSERT((CAP_FCNTL_ALL & fcntlcap) != 0,
441 ("Unsupported fcntl=%d.", cmd));
443 if ((fdp->fd_ofiles[fd].fde_fcntls & fcntlcap) != 0)
446 return (ENOTCAPABLE);
450 sys_cap_fcntls_limit(struct thread *td, struct cap_fcntls_limit_args *uap)
452 struct filedesc *fdp;
453 uint32_t fcntlrights;
457 fcntlrights = uap->fcntlrights;
460 AUDIT_ARG_FCNTL_RIGHTS(fcntlrights);
462 if ((fcntlrights & ~CAP_FCNTL_ALL) != 0)
465 fdp = td->td_proc->p_fd;
468 if (fget_locked(fdp, fd) == NULL) {
469 FILEDESC_XUNLOCK(fdp);
473 if ((fcntlrights & ~fdp->fd_ofiles[fd].fde_fcntls) != 0) {
474 FILEDESC_XUNLOCK(fdp);
475 return (ENOTCAPABLE);
478 fdp->fd_ofiles[fd].fde_fcntls = fcntlrights;
479 FILEDESC_XUNLOCK(fdp);
485 sys_cap_fcntls_get(struct thread *td, struct cap_fcntls_get_args *uap)
487 struct filedesc *fdp;
495 fdp = td->td_proc->p_fd;
497 if (fget_locked(fdp, fd) == NULL) {
498 FILEDESC_SUNLOCK(fdp);
501 rights = fdp->fd_ofiles[fd].fde_fcntls;
502 FILEDESC_SUNLOCK(fdp);
504 return (copyout(&rights, uap->fcntlrightsp, sizeof(rights)));
508 * For backward compatibility.
511 sys_cap_new(struct thread *td, struct cap_new_args *uap)
513 struct filedesc *fdp;
519 rights = uap->rights;
522 AUDIT_ARG_RIGHTS(rights);
524 if ((rights & ~CAP_ALL) != 0)
527 fdp = td->td_proc->p_fd;
529 if (fget_locked(fdp, fd) == NULL) {
530 FILEDESC_SUNLOCK(fdp);
533 error = _cap_check(cap_rights(fdp, fd), rights, CAPFAIL_INCREASE);
534 FILEDESC_SUNLOCK(fdp);
538 error = do_dup(td, 0, fd, 0, &newfd);
544 * We don't really care about the race between checking capability
545 * rights for the source descriptor and now. If capability rights
546 * were ok at that earlier point, the process had this descriptor
547 * with those rights, so we don't increase them in security sense,
548 * the process might have done the cap_new(2) a bit earlier to get
551 fdp->fd_ofiles[newfd].fde_rights = rights;
552 if ((rights & CAP_IOCTL) == 0) {
553 free(fdp->fd_ofiles[newfd].fde_ioctls, M_FILECAPS);
554 fdp->fd_ofiles[newfd].fde_ioctls = NULL;
555 fdp->fd_ofiles[newfd].fde_nioctls = 0;
557 if ((rights & CAP_FCNTL) == 0)
558 fdp->fd_ofiles[newfd].fde_fcntls = 0;
559 FILEDESC_XUNLOCK(fdp);
561 td->td_retval[0] = newfd;
566 #else /* !CAPABILITIES */
569 * Stub Capability functions for when options CAPABILITIES isn't compiled
574 sys_cap_rights_limit(struct thread *td, struct cap_rights_limit_args *uap)
581 sys_cap_rights_get(struct thread *td, struct cap_rights_get_args *uap)
588 sys_cap_ioctls_limit(struct thread *td, struct cap_ioctls_limit_args *uap)
595 sys_cap_ioctls_get(struct thread *td, struct cap_ioctls_get_args *uap)
602 sys_cap_fcntls_limit(struct thread *td, struct cap_fcntls_limit_args *uap)
609 sys_cap_fcntls_get(struct thread *td, struct cap_fcntls_get_args *uap)
616 sys_cap_new(struct thread *td, struct cap_new_args *uap)
622 #endif /* CAPABILITIES */