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(const cap_rights_t *havep, const cap_rights_t *needp,
152 enum ktr_cap_fail_type type)
156 for (i = 0; i < nitems(havep->cr_rights); i++) {
157 if (!cap_rights_contains(havep, needp)) {
159 if (KTRPOINT(curthread, KTR_CAPFAIL))
160 ktrcapfail(type, needp, havep);
162 return (ENOTCAPABLE);
169 * Test whether a capability grants the requested rights.
172 cap_check(const cap_rights_t *havep, const cap_rights_t *needp)
175 return (_cap_check(havep, needp, CAPFAIL_NOTCAPABLE));
179 * Convert capability rights into VM access flags.
182 cap_rights_to_vmprot(cap_rights_t *havep)
186 maxprot = VM_PROT_NONE;
187 if (cap_rights_is_set(havep, CAP_MMAP_R))
188 maxprot |= VM_PROT_READ;
189 if (cap_rights_is_set(havep, CAP_MMAP_W))
190 maxprot |= VM_PROT_WRITE;
191 if (cap_rights_is_set(havep, CAP_MMAP_X))
192 maxprot |= VM_PROT_EXECUTE;
198 * Extract rights from a capability for monitoring purposes -- not for use in
199 * any other way, as we want to keep all capability permission evaluation in
204 cap_rights_fde(struct filedescent *fde)
207 return (&fde->fde_rights);
211 cap_rights(struct filedesc *fdp, int fd)
214 return (cap_rights_fde(&fdp->fd_ofiles[fd]));
218 * System call to limit rights of the given capability.
221 sys_cap_rights_limit(struct thread *td, struct cap_rights_limit_args *uap)
223 struct filedesc *fdp;
225 int error, fd, version;
227 cap_rights_init(&rights);
229 error = copyin(uap->rightsp, &rights, sizeof(rights.cr_rights[0]));
232 version = CAPVER(&rights);
233 if (version != CAP_RIGHTS_VERSION_00)
236 error = copyin(uap->rightsp, &rights,
237 sizeof(rights.cr_rights[0]) * CAPARSIZE(&rights));
240 /* Check for race. */
241 if (CAPVER(&rights) != version)
244 if (!cap_rights_is_valid(&rights))
247 if (version != CAP_RIGHTS_VERSION) {
248 rights.cr_rights[0] &= ~(0x3ULL << 62);
249 rights.cr_rights[0] |= ((uint64_t)CAP_RIGHTS_VERSION << 62);
252 if (KTRPOINT(td, KTR_STRUCT))
253 ktrcaprights(&rights);
259 AUDIT_ARG_RIGHTS(&rights);
261 fdp = td->td_proc->p_fd;
263 if (fget_locked(fdp, fd) == NULL) {
264 FILEDESC_XUNLOCK(fdp);
267 error = _cap_check(cap_rights(fdp, fd), &rights, CAPFAIL_INCREASE);
269 fdp->fd_ofiles[fd].fde_rights = rights;
270 if (!cap_rights_is_set(&rights, CAP_IOCTL)) {
271 free(fdp->fd_ofiles[fd].fde_ioctls, M_FILECAPS);
272 fdp->fd_ofiles[fd].fde_ioctls = NULL;
273 fdp->fd_ofiles[fd].fde_nioctls = 0;
275 if (!cap_rights_is_set(&rights, CAP_FCNTL))
276 fdp->fd_ofiles[fd].fde_fcntls = 0;
278 FILEDESC_XUNLOCK(fdp);
283 * System call to query the rights mask associated with a capability.
286 sys___cap_rights_get(struct thread *td, struct __cap_rights_get_args *uap)
288 struct filedesc *fdp;
292 if (uap->version != CAP_RIGHTS_VERSION_00)
299 fdp = td->td_proc->p_fd;
301 if (fget_locked(fdp, fd) == NULL) {
302 FILEDESC_SUNLOCK(fdp);
305 rights = *cap_rights(fdp, fd);
306 FILEDESC_SUNLOCK(fdp);
307 n = uap->version + 2;
308 if (uap->version != CAPVER(&rights)) {
310 * For older versions we need to check if the descriptor
311 * doesn't contain rights not understood by the caller.
312 * If it does, we have to return an error.
314 for (i = n; i < CAPARSIZE(&rights); i++) {
315 if ((rights.cr_rights[i] & ~(0x7FULL << 57)) != 0)
319 error = copyout(&rights, uap->rightsp, sizeof(rights.cr_rights[0]) * n);
321 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
322 ktrcaprights(&rights);
328 * Test whether a capability grants the given ioctl command.
329 * If descriptor doesn't have CAP_IOCTL, then ioctls list is empty and
330 * ENOTCAPABLE will be returned.
333 cap_ioctl_check(struct filedesc *fdp, int fd, u_long cmd)
339 FILEDESC_LOCK_ASSERT(fdp);
340 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
341 ("%s: invalid fd=%d", __func__, fd));
343 ncmds = fdp->fd_ofiles[fd].fde_nioctls;
347 cmds = fdp->fd_ofiles[fd].fde_ioctls;
348 for (i = 0; i < ncmds; i++) {
353 return (ENOTCAPABLE);
357 * Check if the current ioctls list can be replaced by the new one.
360 cap_ioctl_limit_check(struct filedesc *fdp, int fd, const u_long *cmds,
368 oncmds = fdp->fd_ofiles[fd].fde_nioctls;
371 if (oncmds < (ssize_t)ncmds)
372 return (ENOTCAPABLE);
374 ocmds = fdp->fd_ofiles[fd].fde_ioctls;
375 for (i = 0; i < ncmds; i++) {
376 for (j = 0; j < oncmds; j++) {
377 if (cmds[i] == ocmds[j])
381 return (ENOTCAPABLE);
388 kern_cap_ioctls_limit(struct thread *td, int fd, u_long *cmds, size_t ncmds)
390 struct filedesc *fdp;
396 fdp = td->td_proc->p_fd;
399 if (fget_locked(fdp, fd) == NULL) {
404 error = cap_ioctl_limit_check(fdp, fd, cmds, ncmds);
408 ocmds = fdp->fd_ofiles[fd].fde_ioctls;
409 fdp->fd_ofiles[fd].fde_ioctls = cmds;
410 fdp->fd_ofiles[fd].fde_nioctls = ncmds;
415 FILEDESC_XUNLOCK(fdp);
416 free(cmds, M_FILECAPS);
421 sys_cap_ioctls_limit(struct thread *td, struct cap_ioctls_limit_args *uap)
429 if (ncmds > 256) /* XXX: Is 256 sane? */
435 cmds = malloc(sizeof(cmds[0]) * ncmds, M_FILECAPS, M_WAITOK);
436 error = copyin(uap->cmds, cmds, sizeof(cmds[0]) * ncmds);
438 free(cmds, M_FILECAPS);
443 return (kern_cap_ioctls_limit(td, uap->fd, cmds, ncmds));
447 sys_cap_ioctls_get(struct thread *td, struct cap_ioctls_get_args *uap)
449 struct filedesc *fdp;
450 struct filedescent *fdep;
457 maxcmds = uap->maxcmds;
461 fdp = td->td_proc->p_fd;
464 if (fget_locked(fdp, fd) == NULL) {
470 * If all ioctls are allowed (fde_nioctls == -1 && fde_ioctls == NULL)
471 * the only sane thing we can do is to not populate the given array and
472 * return CAP_IOCTLS_ALL.
475 fdep = &fdp->fd_ofiles[fd];
476 if (cmds != NULL && fdep->fde_ioctls != NULL) {
477 error = copyout(fdep->fde_ioctls, cmds,
478 sizeof(cmds[0]) * MIN(fdep->fde_nioctls, maxcmds));
482 if (fdep->fde_nioctls == -1)
483 td->td_retval[0] = CAP_IOCTLS_ALL;
485 td->td_retval[0] = fdep->fde_nioctls;
489 FILEDESC_SUNLOCK(fdp);
494 * Test whether a capability grants the given fcntl command.
497 cap_fcntl_check_fde(struct filedescent *fde, int cmd)
501 fcntlcap = (1 << cmd);
502 KASSERT((CAP_FCNTL_ALL & fcntlcap) != 0,
503 ("Unsupported fcntl=%d.", cmd));
505 if ((fde->fde_fcntls & fcntlcap) != 0)
508 return (ENOTCAPABLE);
512 cap_fcntl_check(struct filedesc *fdp, int fd, int cmd)
515 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
516 ("%s: invalid fd=%d", __func__, fd));
518 return (cap_fcntl_check_fde(&fdp->fd_ofiles[fd], cmd));
522 sys_cap_fcntls_limit(struct thread *td, struct cap_fcntls_limit_args *uap)
524 struct filedesc *fdp;
525 uint32_t fcntlrights;
529 fcntlrights = uap->fcntlrights;
532 AUDIT_ARG_FCNTL_RIGHTS(fcntlrights);
534 if ((fcntlrights & ~CAP_FCNTL_ALL) != 0)
537 fdp = td->td_proc->p_fd;
540 if (fget_locked(fdp, fd) == NULL) {
541 FILEDESC_XUNLOCK(fdp);
545 if ((fcntlrights & ~fdp->fd_ofiles[fd].fde_fcntls) != 0) {
546 FILEDESC_XUNLOCK(fdp);
547 return (ENOTCAPABLE);
550 fdp->fd_ofiles[fd].fde_fcntls = fcntlrights;
551 FILEDESC_XUNLOCK(fdp);
557 sys_cap_fcntls_get(struct thread *td, struct cap_fcntls_get_args *uap)
559 struct filedesc *fdp;
567 fdp = td->td_proc->p_fd;
569 if (fget_locked(fdp, fd) == NULL) {
570 FILEDESC_SUNLOCK(fdp);
573 rights = fdp->fd_ofiles[fd].fde_fcntls;
574 FILEDESC_SUNLOCK(fdp);
576 return (copyout(&rights, uap->fcntlrightsp, sizeof(rights)));
579 #else /* !CAPABILITIES */
582 * Stub Capability functions for when options CAPABILITIES isn't compiled
587 sys_cap_rights_limit(struct thread *td, struct cap_rights_limit_args *uap)
594 sys___cap_rights_get(struct thread *td, struct __cap_rights_get_args *uap)
601 sys_cap_ioctls_limit(struct thread *td, struct cap_ioctls_limit_args *uap)
608 sys_cap_ioctls_get(struct thread *td, struct cap_ioctls_get_args *uap)
615 sys_cap_fcntls_limit(struct thread *td, struct cap_fcntls_limit_args *uap)
622 sys_cap_fcntls_get(struct thread *td, struct cap_fcntls_get_args *uap)
628 #endif /* CAPABILITIES */