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/capsicum.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 = crcopysafe(p, newcred);
106 newcred->cr_flags |= CRED_FLAG_CAPMODE;
107 proc_set_cred(p, 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(const cap_rights_t *havep, const cap_rights_t *needp,
151 enum ktr_cap_fail_type type)
155 for (i = 0; i < nitems(havep->cr_rights); i++) {
156 if (!cap_rights_contains(havep, needp)) {
158 if (KTRPOINT(curthread, KTR_CAPFAIL))
159 ktrcapfail(type, needp, havep);
161 return (ENOTCAPABLE);
168 * Test whether a capability grants the requested rights.
171 cap_check(const cap_rights_t *havep, const cap_rights_t *needp)
174 return (_cap_check(havep, needp, CAPFAIL_NOTCAPABLE));
178 * Convert capability rights into VM access flags.
181 cap_rights_to_vmprot(cap_rights_t *havep)
185 maxprot = VM_PROT_NONE;
186 if (cap_rights_is_set(havep, CAP_MMAP_R))
187 maxprot |= VM_PROT_READ;
188 if (cap_rights_is_set(havep, CAP_MMAP_W))
189 maxprot |= VM_PROT_WRITE;
190 if (cap_rights_is_set(havep, CAP_MMAP_X))
191 maxprot |= VM_PROT_EXECUTE;
197 * Extract rights from a capability for monitoring purposes -- not for use in
198 * any other way, as we want to keep all capability permission evaluation in
203 cap_rights_fde(struct filedescent *fde)
206 return (&fde->fde_rights);
210 cap_rights(struct filedesc *fdp, int fd)
213 return (cap_rights_fde(&fdp->fd_ofiles[fd]));
217 kern_cap_rights_limit(struct thread *td, int fd, cap_rights_t *rights)
219 struct filedesc *fdp;
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 (!cap_rights_is_set(rights, CAP_IOCTL)) {
232 free(fdp->fd_ofiles[fd].fde_ioctls, M_FILECAPS);
233 fdp->fd_ofiles[fd].fde_ioctls = NULL;
234 fdp->fd_ofiles[fd].fde_nioctls = 0;
236 if (!cap_rights_is_set(rights, CAP_FCNTL))
237 fdp->fd_ofiles[fd].fde_fcntls = 0;
239 FILEDESC_XUNLOCK(fdp);
244 * System call to limit rights of the given capability.
247 sys_cap_rights_limit(struct thread *td, struct cap_rights_limit_args *uap)
252 cap_rights_init(&rights);
254 error = copyin(uap->rightsp, &rights, sizeof(rights.cr_rights[0]));
257 version = CAPVER(&rights);
258 if (version != CAP_RIGHTS_VERSION_00)
261 error = copyin(uap->rightsp, &rights,
262 sizeof(rights.cr_rights[0]) * CAPARSIZE(&rights));
265 /* Check for race. */
266 if (CAPVER(&rights) != version)
269 if (!cap_rights_is_valid(&rights))
272 if (version != CAP_RIGHTS_VERSION) {
273 rights.cr_rights[0] &= ~(0x3ULL << 62);
274 rights.cr_rights[0] |= ((uint64_t)CAP_RIGHTS_VERSION << 62);
277 if (KTRPOINT(td, KTR_STRUCT))
278 ktrcaprights(&rights);
281 AUDIT_ARG_FD(uap->fd);
282 AUDIT_ARG_RIGHTS(&rights);
283 return (kern_cap_rights_limit(td, uap->fd, &rights));
287 * System call to query the rights mask associated with a capability.
290 sys___cap_rights_get(struct thread *td, struct __cap_rights_get_args *uap)
292 struct filedesc *fdp;
296 if (uap->version != CAP_RIGHTS_VERSION_00)
303 fdp = td->td_proc->p_fd;
305 if (fget_locked(fdp, fd) == NULL) {
306 FILEDESC_SUNLOCK(fdp);
309 rights = *cap_rights(fdp, fd);
310 FILEDESC_SUNLOCK(fdp);
311 n = uap->version + 2;
312 if (uap->version != CAPVER(&rights)) {
314 * For older versions we need to check if the descriptor
315 * doesn't contain rights not understood by the caller.
316 * If it does, we have to return an error.
318 for (i = n; i < CAPARSIZE(&rights); i++) {
319 if ((rights.cr_rights[i] & ~(0x7FULL << 57)) != 0)
323 error = copyout(&rights, uap->rightsp, sizeof(rights.cr_rights[0]) * n);
325 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
326 ktrcaprights(&rights);
332 * Test whether a capability grants the given ioctl command.
333 * If descriptor doesn't have CAP_IOCTL, then ioctls list is empty and
334 * ENOTCAPABLE will be returned.
337 cap_ioctl_check(struct filedesc *fdp, int fd, u_long cmd)
343 FILEDESC_LOCK_ASSERT(fdp);
344 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
345 ("%s: invalid fd=%d", __func__, fd));
347 ncmds = fdp->fd_ofiles[fd].fde_nioctls;
351 cmds = fdp->fd_ofiles[fd].fde_ioctls;
352 for (i = 0; i < ncmds; i++) {
357 return (ENOTCAPABLE);
361 * Check if the current ioctls list can be replaced by the new one.
364 cap_ioctl_limit_check(struct filedesc *fdp, int fd, const u_long *cmds,
372 oncmds = fdp->fd_ofiles[fd].fde_nioctls;
375 if (oncmds < (ssize_t)ncmds)
376 return (ENOTCAPABLE);
378 ocmds = fdp->fd_ofiles[fd].fde_ioctls;
379 for (i = 0; i < ncmds; i++) {
380 for (j = 0; j < oncmds; j++) {
381 if (cmds[i] == ocmds[j])
385 return (ENOTCAPABLE);
392 kern_cap_ioctls_limit(struct thread *td, int fd, u_long *cmds, size_t ncmds)
394 struct filedesc *fdp;
400 fdp = td->td_proc->p_fd;
403 if (fget_locked(fdp, fd) == NULL) {
408 error = cap_ioctl_limit_check(fdp, fd, cmds, ncmds);
412 ocmds = fdp->fd_ofiles[fd].fde_ioctls;
413 fdp->fd_ofiles[fd].fde_ioctls = cmds;
414 fdp->fd_ofiles[fd].fde_nioctls = ncmds;
419 FILEDESC_XUNLOCK(fdp);
420 free(cmds, M_FILECAPS);
425 sys_cap_ioctls_limit(struct thread *td, struct cap_ioctls_limit_args *uap)
433 if (ncmds > 256) /* XXX: Is 256 sane? */
439 cmds = malloc(sizeof(cmds[0]) * ncmds, M_FILECAPS, M_WAITOK);
440 error = copyin(uap->cmds, cmds, sizeof(cmds[0]) * ncmds);
442 free(cmds, M_FILECAPS);
447 return (kern_cap_ioctls_limit(td, uap->fd, cmds, ncmds));
451 sys_cap_ioctls_get(struct thread *td, struct cap_ioctls_get_args *uap)
453 struct filedesc *fdp;
454 struct filedescent *fdep;
461 maxcmds = uap->maxcmds;
465 fdp = td->td_proc->p_fd;
468 if (fget_locked(fdp, fd) == NULL) {
474 * If all ioctls are allowed (fde_nioctls == -1 && fde_ioctls == NULL)
475 * the only sane thing we can do is to not populate the given array and
476 * return CAP_IOCTLS_ALL.
479 fdep = &fdp->fd_ofiles[fd];
480 if (cmds != NULL && fdep->fde_ioctls != NULL) {
481 error = copyout(fdep->fde_ioctls, cmds,
482 sizeof(cmds[0]) * MIN(fdep->fde_nioctls, maxcmds));
486 if (fdep->fde_nioctls == -1)
487 td->td_retval[0] = CAP_IOCTLS_ALL;
489 td->td_retval[0] = fdep->fde_nioctls;
493 FILEDESC_SUNLOCK(fdp);
498 * Test whether a capability grants the given fcntl command.
501 cap_fcntl_check_fde(struct filedescent *fde, int cmd)
505 fcntlcap = (1 << cmd);
506 KASSERT((CAP_FCNTL_ALL & fcntlcap) != 0,
507 ("Unsupported fcntl=%d.", cmd));
509 if ((fde->fde_fcntls & fcntlcap) != 0)
512 return (ENOTCAPABLE);
516 cap_fcntl_check(struct filedesc *fdp, int fd, int cmd)
519 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
520 ("%s: invalid fd=%d", __func__, fd));
522 return (cap_fcntl_check_fde(&fdp->fd_ofiles[fd], cmd));
526 sys_cap_fcntls_limit(struct thread *td, struct cap_fcntls_limit_args *uap)
528 struct filedesc *fdp;
529 uint32_t fcntlrights;
533 fcntlrights = uap->fcntlrights;
536 AUDIT_ARG_FCNTL_RIGHTS(fcntlrights);
538 if ((fcntlrights & ~CAP_FCNTL_ALL) != 0)
541 fdp = td->td_proc->p_fd;
544 if (fget_locked(fdp, fd) == NULL) {
545 FILEDESC_XUNLOCK(fdp);
549 if ((fcntlrights & ~fdp->fd_ofiles[fd].fde_fcntls) != 0) {
550 FILEDESC_XUNLOCK(fdp);
551 return (ENOTCAPABLE);
554 fdp->fd_ofiles[fd].fde_fcntls = fcntlrights;
555 FILEDESC_XUNLOCK(fdp);
561 sys_cap_fcntls_get(struct thread *td, struct cap_fcntls_get_args *uap)
563 struct filedesc *fdp;
571 fdp = td->td_proc->p_fd;
573 if (fget_locked(fdp, fd) == NULL) {
574 FILEDESC_SUNLOCK(fdp);
577 rights = fdp->fd_ofiles[fd].fde_fcntls;
578 FILEDESC_SUNLOCK(fdp);
580 return (copyout(&rights, uap->fcntlrightsp, sizeof(rights)));
583 #else /* !CAPABILITIES */
586 * Stub Capability functions for when options CAPABILITIES isn't compiled
591 sys_cap_rights_limit(struct thread *td, struct cap_rights_limit_args *uap)
598 sys___cap_rights_get(struct thread *td, struct __cap_rights_get_args *uap)
605 sys_cap_ioctls_limit(struct thread *td, struct cap_ioctls_limit_args *uap)
612 sys_cap_ioctls_get(struct thread *td, struct cap_ioctls_get_args *uap)
619 sys_cap_fcntls_limit(struct thread *td, struct cap_fcntls_limit_args *uap)
626 sys_cap_fcntls_get(struct thread *td, struct cap_fcntls_get_args *uap)
632 #endif /* CAPABILITIES */