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>
87 SYSCTL_INT(_kern, OID_AUTO, trap_enotcap, CTLFLAG_RW, &trap_enotcap, 0,
88 "Deliver SIGTRAP on ENOTCAPABLE");
90 #ifdef CAPABILITY_MODE
92 #define IOCTLS_MAX_COUNT 256 /* XXX: Is 256 sane? */
94 FEATURE(security_capability_mode, "Capsicum Capability Mode");
97 * System call to enter capability mode for the process.
100 sys_cap_enter(struct thread *td, struct cap_enter_args *uap)
102 struct ucred *newcred, *oldcred;
105 if (IN_CAPABILITY_MODE(td))
111 oldcred = crcopysafe(p, newcred);
112 newcred->cr_flags |= CRED_FLAG_CAPMODE;
113 proc_set_cred(p, newcred);
120 * System call to query whether the process is in capability mode.
123 sys_cap_getmode(struct thread *td, struct cap_getmode_args *uap)
127 i = IN_CAPABILITY_MODE(td) ? 1 : 0;
128 return (copyout(&i, uap->modep, sizeof(i)));
131 #else /* !CAPABILITY_MODE */
134 sys_cap_enter(struct thread *td, struct cap_enter_args *uap)
141 sys_cap_getmode(struct thread *td, struct cap_getmode_args *uap)
147 #endif /* CAPABILITY_MODE */
151 FEATURE(security_capabilities, "Capsicum Capabilities");
153 MALLOC_DECLARE(M_FILECAPS);
156 _cap_check(const cap_rights_t *havep, const cap_rights_t *needp,
157 enum ktr_cap_fail_type type)
160 if (!cap_rights_contains(havep, needp)) {
162 if (KTRPOINT(curthread, KTR_CAPFAIL))
163 ktrcapfail(type, needp, havep);
165 return (ENOTCAPABLE);
171 * Test whether a capability grants the requested rights.
174 cap_check(const cap_rights_t *havep, const cap_rights_t *needp)
177 return (_cap_check(havep, needp, CAPFAIL_NOTCAPABLE));
181 * Convert capability rights into VM access flags.
184 cap_rights_to_vmprot(cap_rights_t *havep)
188 maxprot = VM_PROT_NONE;
189 if (cap_rights_is_set(havep, CAP_MMAP_R))
190 maxprot |= VM_PROT_READ;
191 if (cap_rights_is_set(havep, CAP_MMAP_W))
192 maxprot |= VM_PROT_WRITE;
193 if (cap_rights_is_set(havep, CAP_MMAP_X))
194 maxprot |= VM_PROT_EXECUTE;
200 * Extract rights from a capability for monitoring purposes -- not for use in
201 * any other way, as we want to keep all capability permission evaluation in
206 cap_rights_fde(struct filedescent *fde)
209 return (&fde->fde_rights);
213 cap_rights(struct filedesc *fdp, int fd)
216 return (cap_rights_fde(&fdp->fd_ofiles[fd]));
220 kern_cap_rights_limit(struct thread *td, int fd, cap_rights_t *rights)
222 struct filedesc *fdp;
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 (!cap_rights_is_set(rights, CAP_IOCTL)) {
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 (!cap_rights_is_set(rights, CAP_FCNTL))
240 fdp->fd_ofiles[fd].fde_fcntls = 0;
242 FILEDESC_XUNLOCK(fdp);
247 * System call to limit rights of the given capability.
250 sys_cap_rights_limit(struct thread *td, struct cap_rights_limit_args *uap)
255 cap_rights_init(&rights);
257 error = copyin(uap->rightsp, &rights, sizeof(rights.cr_rights[0]));
260 version = CAPVER(&rights);
261 if (version != CAP_RIGHTS_VERSION_00)
264 error = copyin(uap->rightsp, &rights,
265 sizeof(rights.cr_rights[0]) * CAPARSIZE(&rights));
268 /* Check for race. */
269 if (CAPVER(&rights) != version)
272 if (!cap_rights_is_valid(&rights))
275 if (version != CAP_RIGHTS_VERSION) {
276 rights.cr_rights[0] &= ~(0x3ULL << 62);
277 rights.cr_rights[0] |= ((uint64_t)CAP_RIGHTS_VERSION << 62);
280 if (KTRPOINT(td, KTR_STRUCT))
281 ktrcaprights(&rights);
284 AUDIT_ARG_FD(uap->fd);
285 AUDIT_ARG_RIGHTS(&rights);
286 return (kern_cap_rights_limit(td, uap->fd, &rights));
290 * System call to query the rights mask associated with a capability.
293 sys___cap_rights_get(struct thread *td, struct __cap_rights_get_args *uap)
295 struct filedesc *fdp;
299 if (uap->version != CAP_RIGHTS_VERSION_00)
306 fdp = td->td_proc->p_fd;
308 if (fget_locked(fdp, fd) == NULL) {
309 FILEDESC_SUNLOCK(fdp);
312 rights = *cap_rights(fdp, fd);
313 FILEDESC_SUNLOCK(fdp);
314 n = uap->version + 2;
315 if (uap->version != CAPVER(&rights)) {
317 * For older versions we need to check if the descriptor
318 * doesn't contain rights not understood by the caller.
319 * If it does, we have to return an error.
321 for (i = n; i < CAPARSIZE(&rights); i++) {
322 if ((rights.cr_rights[i] & ~(0x7FULL << 57)) != 0)
326 error = copyout(&rights, uap->rightsp, sizeof(rights.cr_rights[0]) * n);
328 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
329 ktrcaprights(&rights);
335 * Test whether a capability grants the given ioctl command.
336 * If descriptor doesn't have CAP_IOCTL, then ioctls list is empty and
337 * ENOTCAPABLE will be returned.
340 cap_ioctl_check(struct filedesc *fdp, int fd, u_long cmd)
346 FILEDESC_LOCK_ASSERT(fdp);
347 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
348 ("%s: invalid fd=%d", __func__, fd));
350 ncmds = fdp->fd_ofiles[fd].fde_nioctls;
354 cmds = fdp->fd_ofiles[fd].fde_ioctls;
355 for (i = 0; i < ncmds; i++) {
360 return (ENOTCAPABLE);
364 * Check if the current ioctls list can be replaced by the new one.
367 cap_ioctl_limit_check(struct filedesc *fdp, int fd, const u_long *cmds,
375 oncmds = fdp->fd_ofiles[fd].fde_nioctls;
378 if (oncmds < (ssize_t)ncmds)
379 return (ENOTCAPABLE);
381 ocmds = fdp->fd_ofiles[fd].fde_ioctls;
382 for (i = 0; i < ncmds; i++) {
383 for (j = 0; j < oncmds; j++) {
384 if (cmds[i] == ocmds[j])
388 return (ENOTCAPABLE);
395 kern_cap_ioctls_limit(struct thread *td, int fd, u_long *cmds, size_t ncmds)
397 struct filedesc *fdp;
403 if (ncmds > IOCTLS_MAX_COUNT) {
408 fdp = td->td_proc->p_fd;
411 if (fget_locked(fdp, fd) == NULL) {
416 error = cap_ioctl_limit_check(fdp, fd, cmds, ncmds);
420 ocmds = fdp->fd_ofiles[fd].fde_ioctls;
421 fdp->fd_ofiles[fd].fde_ioctls = cmds;
422 fdp->fd_ofiles[fd].fde_nioctls = ncmds;
427 FILEDESC_XUNLOCK(fdp);
429 free(cmds, M_FILECAPS);
434 sys_cap_ioctls_limit(struct thread *td, struct cap_ioctls_limit_args *uap)
442 if (ncmds > IOCTLS_MAX_COUNT)
448 cmds = malloc(sizeof(cmds[0]) * ncmds, M_FILECAPS, M_WAITOK);
449 error = copyin(uap->cmds, cmds, sizeof(cmds[0]) * ncmds);
451 free(cmds, M_FILECAPS);
456 return (kern_cap_ioctls_limit(td, uap->fd, cmds, ncmds));
460 sys_cap_ioctls_get(struct thread *td, struct cap_ioctls_get_args *uap)
462 struct filedesc *fdp;
463 struct filedescent *fdep;
464 u_long *cmdsp, *dstcmds;
465 size_t maxcmds, ncmds;
471 maxcmds = uap->maxcmds;
475 fdp = td->td_proc->p_fd;
478 if (dstcmds != NULL) {
479 cmdsp = malloc(sizeof(cmdsp[0]) * IOCTLS_MAX_COUNT, M_FILECAPS,
484 fdep = fdeget_locked(fdp, fd);
487 FILEDESC_SUNLOCK(fdp);
490 count = fdep->fde_nioctls;
491 if (count != -1 && cmdsp != NULL) {
492 ncmds = MIN(count, maxcmds);
493 memcpy(cmdsp, fdep->fde_ioctls, sizeof(cmdsp[0]) * ncmds);
495 FILEDESC_SUNLOCK(fdp);
498 * If all ioctls are allowed (fde_nioctls == -1 && fde_ioctls == NULL)
499 * the only sane thing we can do is to not populate the given array and
500 * return CAP_IOCTLS_ALL.
504 error = copyout(cmdsp, dstcmds,
505 sizeof(cmdsp[0]) * ncmds);
509 td->td_retval[0] = count;
511 td->td_retval[0] = CAP_IOCTLS_ALL;
516 free(cmdsp, M_FILECAPS);
521 * Test whether a capability grants the given fcntl command.
524 cap_fcntl_check_fde(struct filedescent *fde, int cmd)
528 fcntlcap = (1 << cmd);
529 KASSERT((CAP_FCNTL_ALL & fcntlcap) != 0,
530 ("Unsupported fcntl=%d.", cmd));
532 if ((fde->fde_fcntls & fcntlcap) != 0)
535 return (ENOTCAPABLE);
539 cap_fcntl_check(struct filedesc *fdp, int fd, int cmd)
542 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
543 ("%s: invalid fd=%d", __func__, fd));
545 return (cap_fcntl_check_fde(&fdp->fd_ofiles[fd], cmd));
549 sys_cap_fcntls_limit(struct thread *td, struct cap_fcntls_limit_args *uap)
551 struct filedesc *fdp;
552 uint32_t fcntlrights;
556 fcntlrights = uap->fcntlrights;
559 AUDIT_ARG_FCNTL_RIGHTS(fcntlrights);
561 if ((fcntlrights & ~CAP_FCNTL_ALL) != 0)
564 fdp = td->td_proc->p_fd;
567 if (fget_locked(fdp, fd) == NULL) {
568 FILEDESC_XUNLOCK(fdp);
572 if ((fcntlrights & ~fdp->fd_ofiles[fd].fde_fcntls) != 0) {
573 FILEDESC_XUNLOCK(fdp);
574 return (ENOTCAPABLE);
577 fdp->fd_ofiles[fd].fde_fcntls = fcntlrights;
578 FILEDESC_XUNLOCK(fdp);
584 sys_cap_fcntls_get(struct thread *td, struct cap_fcntls_get_args *uap)
586 struct filedesc *fdp;
594 fdp = td->td_proc->p_fd;
596 if (fget_locked(fdp, fd) == NULL) {
597 FILEDESC_SUNLOCK(fdp);
600 rights = fdp->fd_ofiles[fd].fde_fcntls;
601 FILEDESC_SUNLOCK(fdp);
603 return (copyout(&rights, uap->fcntlrightsp, sizeof(rights)));
606 #else /* !CAPABILITIES */
609 * Stub Capability functions for when options CAPABILITIES isn't compiled
614 sys_cap_rights_limit(struct thread *td, struct cap_rights_limit_args *uap)
621 sys___cap_rights_get(struct thread *td, struct __cap_rights_get_args *uap)
628 sys_cap_ioctls_limit(struct thread *td, struct cap_ioctls_limit_args *uap)
635 sys_cap_ioctls_get(struct thread *td, struct cap_ioctls_get_args *uap)
642 sys_cap_fcntls_limit(struct thread *td, struct cap_fcntls_limit_args *uap)
649 sys_cap_fcntls_get(struct thread *td, struct cap_fcntls_get_args *uap)
655 #endif /* CAPABILITIES */