2 * SPDX-License-Identifier: BSD-2-Clause
4 * Copyright (c) 2008-2011 Robert N. M. Watson
5 * Copyright (c) 2010-2011 Jonathan Anderson
6 * Copyright (c) 2012 FreeBSD Foundation
9 * This software was developed at the University of Cambridge Computer
10 * Laboratory with support from a grant from Google, Inc.
12 * Portions of this software were developed by Pawel Jakub Dawidek under
13 * sponsorship from the FreeBSD Foundation.
15 * Redistribution and use in source and binary forms, with or without
16 * modification, are permitted provided that the following conditions
18 * 1. Redistributions of source code must retain the above copyright
19 * notice, this list of conditions and the following disclaimer.
20 * 2. Redistributions in binary form must reproduce the above copyright
21 * notice, this list of conditions and the following disclaimer in the
22 * documentation and/or other materials provided with the distribution.
24 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
38 * FreeBSD kernel capability facility.
40 * Two kernel features are implemented here: capability mode, a sandboxed mode
41 * of execution for processes, and capabilities, a refinement on file
42 * descriptors that allows fine-grained control over operations on the file
43 * descriptor. Collectively, these allow processes to run in the style of a
44 * historic "capability system" in which they can use only resources
45 * explicitly delegated to them. This model is enforced by restricting access
46 * to global namespaces in capability mode.
48 * Capabilities wrap other file descriptor types, binding them to a constant
49 * rights mask set when the capability is created. New capabilities may be
50 * derived from existing capabilities, but only if they have the same or a
51 * strict subset of the rights on the original capability.
53 * System calls permitted in capability mode are defined in capabilities.conf;
54 * calls must be carefully audited for safety to ensure that they don't allow
55 * escape from a sandbox. Some calls permit only a subset of operations in
56 * capability mode -- for example, shm_open(2) is limited to creating
57 * anonymous, rather than named, POSIX shared memory objects.
60 #include <sys/cdefs.h>
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 bool __read_frequently trap_enotcap;
87 SYSCTL_BOOL(_kern, OID_AUTO, trap_enotcap, CTLFLAG_RWTUN, &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 cap_check_failed_notcapable(const cap_rights_t *havep, const cap_rights_t *needp)
185 if (KTRPOINT(curthread, KTR_CAPFAIL))
186 ktrcapfail(CAPFAIL_NOTCAPABLE, needp, havep);
188 return (ENOTCAPABLE);
192 * Convert capability rights into VM access flags.
195 cap_rights_to_vmprot(const cap_rights_t *havep)
199 maxprot = VM_PROT_NONE;
200 if (cap_rights_is_set(havep, CAP_MMAP_R))
201 maxprot |= VM_PROT_READ;
202 if (cap_rights_is_set(havep, CAP_MMAP_W))
203 maxprot |= VM_PROT_WRITE;
204 if (cap_rights_is_set(havep, CAP_MMAP_X))
205 maxprot |= VM_PROT_EXECUTE;
211 * Extract rights from a capability for monitoring purposes -- not for use in
212 * any other way, as we want to keep all capability permission evaluation in
217 cap_rights_fde(const struct filedescent *fdep)
220 return (cap_rights_fde_inline(fdep));
224 cap_rights(struct filedesc *fdp, int fd)
227 return (cap_rights_fde(&fdp->fd_ofiles[fd]));
231 kern_cap_rights_limit(struct thread *td, int fd, cap_rights_t *rights)
233 struct filedesc *fdp;
234 struct filedescent *fdep;
238 fdp = td->td_proc->p_fd;
240 fdep = fdeget_locked(fdp, fd);
242 FILEDESC_XUNLOCK(fdp);
246 error = _cap_check(cap_rights(fdp, fd), rights, CAPFAIL_INCREASE);
248 seqc_write_begin(&fdep->fde_seqc);
249 fdep->fde_rights = *rights;
250 if (!cap_rights_is_set(rights, CAP_IOCTL)) {
251 ioctls = fdep->fde_ioctls;
252 fdep->fde_ioctls = NULL;
253 fdep->fde_nioctls = 0;
255 if (!cap_rights_is_set(rights, CAP_FCNTL))
256 fdep->fde_fcntls = 0;
257 seqc_write_end(&fdep->fde_seqc);
259 FILEDESC_XUNLOCK(fdp);
260 free(ioctls, M_FILECAPS);
265 * System call to limit rights of the given capability.
268 sys_cap_rights_limit(struct thread *td, struct cap_rights_limit_args *uap)
273 cap_rights_init_zero(&rights);
275 error = copyin(uap->rightsp, &rights, sizeof(rights.cr_rights[0]));
278 version = CAPVER(&rights);
279 if (version != CAP_RIGHTS_VERSION_00)
282 error = copyin(uap->rightsp, &rights,
283 sizeof(rights.cr_rights[0]) * CAPARSIZE(&rights));
286 /* Check for race. */
287 if (CAPVER(&rights) != version)
290 if (!cap_rights_is_valid(&rights))
293 if (version != CAP_RIGHTS_VERSION) {
294 rights.cr_rights[0] &= ~(0x3ULL << 62);
295 rights.cr_rights[0] |= ((uint64_t)CAP_RIGHTS_VERSION << 62);
298 if (KTRPOINT(td, KTR_STRUCT))
299 ktrcaprights(&rights);
302 AUDIT_ARG_FD(uap->fd);
303 AUDIT_ARG_RIGHTS(&rights);
304 return (kern_cap_rights_limit(td, uap->fd, &rights));
308 * System call to query the rights mask associated with a capability.
311 sys___cap_rights_get(struct thread *td, struct __cap_rights_get_args *uap)
313 struct filedesc *fdp;
317 if (uap->version != CAP_RIGHTS_VERSION_00)
324 fdp = td->td_proc->p_fd;
326 if (fget_locked(fdp, fd) == NULL) {
327 FILEDESC_SUNLOCK(fdp);
330 rights = *cap_rights(fdp, fd);
331 FILEDESC_SUNLOCK(fdp);
332 n = uap->version + 2;
333 if (uap->version != CAPVER(&rights)) {
335 * For older versions we need to check if the descriptor
336 * doesn't contain rights not understood by the caller.
337 * If it does, we have to return an error.
339 for (i = n; i < CAPARSIZE(&rights); i++) {
340 if ((rights.cr_rights[i] & ~(0x7FULL << 57)) != 0)
344 error = copyout(&rights, uap->rightsp, sizeof(rights.cr_rights[0]) * n);
346 if (error == 0 && KTRPOINT(td, KTR_STRUCT))
347 ktrcaprights(&rights);
353 * Test whether a capability grants the given ioctl command.
354 * If descriptor doesn't have CAP_IOCTL, then ioctls list is empty and
355 * ENOTCAPABLE will be returned.
358 cap_ioctl_check(struct filedesc *fdp, int fd, u_long cmd)
360 struct filedescent *fdep;
365 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
366 ("%s: invalid fd=%d", __func__, fd));
368 fdep = fdeget_locked(fdp, fd);
369 KASSERT(fdep != NULL,
370 ("%s: invalid fd=%d", __func__, fd));
372 ncmds = fdep->fde_nioctls;
376 cmds = fdep->fde_ioctls;
377 for (i = 0; i < ncmds; i++) {
382 return (ENOTCAPABLE);
386 * Check if the current ioctls list can be replaced by the new one.
389 cap_ioctl_limit_check(struct filedescent *fdep, const u_long *cmds,
397 oncmds = fdep->fde_nioctls;
400 if (oncmds < (ssize_t)ncmds)
401 return (ENOTCAPABLE);
403 ocmds = fdep->fde_ioctls;
404 for (i = 0; i < ncmds; i++) {
405 for (j = 0; j < oncmds; j++) {
406 if (cmds[i] == ocmds[j])
410 return (ENOTCAPABLE);
417 kern_cap_ioctls_limit(struct thread *td, int fd, u_long *cmds, size_t ncmds)
419 struct filedesc *fdp;
420 struct filedescent *fdep;
426 if (ncmds > IOCTLS_MAX_COUNT) {
431 fdp = td->td_proc->p_fd;
434 fdep = fdeget_locked(fdp, fd);
440 error = cap_ioctl_limit_check(fdep, cmds, ncmds);
444 ocmds = fdep->fde_ioctls;
445 seqc_write_begin(&fdep->fde_seqc);
446 fdep->fde_ioctls = cmds;
447 fdep->fde_nioctls = ncmds;
448 seqc_write_end(&fdep->fde_seqc);
453 FILEDESC_XUNLOCK(fdp);
455 free(cmds, M_FILECAPS);
460 sys_cap_ioctls_limit(struct thread *td, struct cap_ioctls_limit_args *uap)
468 if (ncmds > IOCTLS_MAX_COUNT)
474 cmds = malloc(sizeof(cmds[0]) * ncmds, M_FILECAPS, M_WAITOK);
475 error = copyin(uap->cmds, cmds, sizeof(cmds[0]) * ncmds);
477 free(cmds, M_FILECAPS);
482 return (kern_cap_ioctls_limit(td, uap->fd, cmds, ncmds));
486 sys_cap_ioctls_get(struct thread *td, struct cap_ioctls_get_args *uap)
488 struct filedesc *fdp;
489 struct filedescent *fdep;
490 u_long *cmdsp, *dstcmds;
491 size_t maxcmds, ncmds;
497 maxcmds = uap->maxcmds;
501 fdp = td->td_proc->p_fd;
504 if (dstcmds != NULL) {
505 cmdsp = malloc(sizeof(cmdsp[0]) * IOCTLS_MAX_COUNT, M_FILECAPS,
510 fdep = fdeget_locked(fdp, fd);
513 FILEDESC_SUNLOCK(fdp);
516 count = fdep->fde_nioctls;
517 if (count != -1 && cmdsp != NULL) {
518 ncmds = MIN(count, maxcmds);
519 memcpy(cmdsp, fdep->fde_ioctls, sizeof(cmdsp[0]) * ncmds);
521 FILEDESC_SUNLOCK(fdp);
524 * If all ioctls are allowed (fde_nioctls == -1 && fde_ioctls == NULL)
525 * the only sane thing we can do is to not populate the given array and
526 * return CAP_IOCTLS_ALL.
530 error = copyout(cmdsp, dstcmds,
531 sizeof(cmdsp[0]) * ncmds);
535 td->td_retval[0] = count;
537 td->td_retval[0] = CAP_IOCTLS_ALL;
542 free(cmdsp, M_FILECAPS);
547 * Test whether a capability grants the given fcntl command.
550 cap_fcntl_check_fde(struct filedescent *fdep, int cmd)
554 fcntlcap = (1 << cmd);
555 KASSERT((CAP_FCNTL_ALL & fcntlcap) != 0,
556 ("Unsupported fcntl=%d.", cmd));
558 if ((fdep->fde_fcntls & fcntlcap) != 0)
561 return (ENOTCAPABLE);
565 cap_fcntl_check(struct filedesc *fdp, int fd, int cmd)
568 KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
569 ("%s: invalid fd=%d", __func__, fd));
571 return (cap_fcntl_check_fde(&fdp->fd_ofiles[fd], cmd));
575 sys_cap_fcntls_limit(struct thread *td, struct cap_fcntls_limit_args *uap)
577 struct filedesc *fdp;
578 struct filedescent *fdep;
579 uint32_t fcntlrights;
583 fcntlrights = uap->fcntlrights;
586 AUDIT_ARG_FCNTL_RIGHTS(fcntlrights);
588 if ((fcntlrights & ~CAP_FCNTL_ALL) != 0)
591 fdp = td->td_proc->p_fd;
594 fdep = fdeget_locked(fdp, fd);
596 FILEDESC_XUNLOCK(fdp);
600 if ((fcntlrights & ~fdep->fde_fcntls) != 0) {
601 FILEDESC_XUNLOCK(fdp);
602 return (ENOTCAPABLE);
605 seqc_write_begin(&fdep->fde_seqc);
606 fdep->fde_fcntls = fcntlrights;
607 seqc_write_end(&fdep->fde_seqc);
608 FILEDESC_XUNLOCK(fdp);
614 sys_cap_fcntls_get(struct thread *td, struct cap_fcntls_get_args *uap)
616 struct filedesc *fdp;
617 struct filedescent *fdep;
625 fdp = td->td_proc->p_fd;
627 fdep = fdeget_locked(fdp, fd);
629 FILEDESC_SUNLOCK(fdp);
632 rights = fdep->fde_fcntls;
633 FILEDESC_SUNLOCK(fdp);
635 return (copyout(&rights, uap->fcntlrightsp, sizeof(rights)));
638 #else /* !CAPABILITIES */
641 * Stub Capability functions for when options CAPABILITIES isn't compiled
646 sys_cap_rights_limit(struct thread *td, struct cap_rights_limit_args *uap)
653 sys___cap_rights_get(struct thread *td, struct __cap_rights_get_args *uap)
660 sys_cap_ioctls_limit(struct thread *td, struct cap_ioctls_limit_args *uap)
667 sys_cap_ioctls_get(struct thread *td, struct cap_ioctls_get_args *uap)
674 sys_cap_fcntls_limit(struct thread *td, struct cap_fcntls_limit_args *uap)
681 sys_cap_fcntls_get(struct thread *td, struct cap_fcntls_get_args *uap)
687 #endif /* CAPABILITIES */