xref: /freebsd/sys/kern/sys_capability.c (revision 09d325677d53a12c79a43664ff29871e92247629) 
1 /*-
2  * Copyright (c) 2008-2011 Robert N. M. Watson
3  * Copyright (c) 2010-2011 Jonathan Anderson
4  * Copyright (c) 2012 FreeBSD Foundation
5  * All rights reserved.
6  *
7  * This software was developed at the University of Cambridge Computer
8  * Laboratory with support from a grant from Google, Inc.
9  *
10  * Portions of this software were developed by Pawel Jakub Dawidek under
11  * sponsorship from the FreeBSD Foundation.
12  *
13  * Redistribution and use in source and binary forms, with or without
14  * modification, are permitted provided that the following conditions
15  * are met:
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.
21  *
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
32  * SUCH DAMAGE.
33  */
34 
35 /*
36  * FreeBSD kernel capability facility.
37  *
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.
45  *
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.
50  *
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.
56  */
57 
58 #include <sys/cdefs.h>
59 __FBSDID("$FreeBSD$");
60 
61 #include "opt_capsicum.h"
62 #include "opt_ktrace.h"
63 
64 #include <sys/param.h>
65 #include <sys/capability.h>
66 #include <sys/file.h>
67 #include <sys/filedesc.h>
68 #include <sys/kernel.h>
69 #include <sys/limits.h>
70 #include <sys/lock.h>
71 #include <sys/mutex.h>
72 #include <sys/proc.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>
78 #include <sys/uio.h>
79 #include <sys/ktrace.h>
80 
81 #include <security/audit/audit.h>
82 
83 #include <vm/uma.h>
84 #include <vm/vm.h>
85 
86 #ifdef CAPABILITY_MODE
87 
88 FEATURE(security_capability_mode, "Capsicum Capability Mode");
89 
90 /*
91  * System call to enter capability mode for the process.
92  */
93 int
94 sys_cap_enter(struct thread *td, struct cap_enter_args *uap)
95 {
96 	struct ucred *newcred, *oldcred;
97 	struct proc *p;
98 
99 	if (IN_CAPABILITY_MODE(td))
100 		return (0);
101 
102 	newcred = crget();
103 	p = td->td_proc;
104 	PROC_LOCK(p);
105 	oldcred = p->p_ucred;
106 	crcopy(newcred, oldcred);
107 	newcred->cr_flags |= CRED_FLAG_CAPMODE;
108 	p->p_ucred = newcred;
109 	PROC_UNLOCK(p);
110 	crfree(oldcred);
111 	return (0);
112 }
113 
114 /*
115  * System call to query whether the process is in capability mode.
116  */
117 int
118 sys_cap_getmode(struct thread *td, struct cap_getmode_args *uap)
119 {
120 	u_int i;
121 
122 	i = IN_CAPABILITY_MODE(td) ? 1 : 0;
123 	return (copyout(&i, uap->modep, sizeof(i)));
124 }
125 
126 #else /* !CAPABILITY_MODE */
127 
128 int
129 sys_cap_enter(struct thread *td, struct cap_enter_args *uap)
130 {
131 
132 	return (ENOSYS);
133 }
134 
135 int
136 sys_cap_getmode(struct thread *td, struct cap_getmode_args *uap)
137 {
138 
139 	return (ENOSYS);
140 }
141 
142 #endif /* CAPABILITY_MODE */
143 
144 #ifdef CAPABILITIES
145 
146 FEATURE(security_capabilities, "Capsicum Capabilities");
147 
148 MALLOC_DECLARE(M_FILECAPS);
149 
150 static inline int
151 _cap_check(const cap_rights_t *havep, const cap_rights_t *needp,
152     enum ktr_cap_fail_type type)
153 {
154 	int i;
155 
156 	for (i = 0; i < nitems(havep->cr_rights); i++) {
157 		if (!cap_rights_contains(havep, needp)) {
158 #ifdef KTRACE
159 			if (KTRPOINT(curthread, KTR_CAPFAIL))
160 				ktrcapfail(type, needp, havep);
161 #endif
162 			return (ENOTCAPABLE);
163 		}
164 	}
165 	return (0);
166 }
167 
168 /*
169  * Test whether a capability grants the requested rights.
170  */
171 int
172 cap_check(const cap_rights_t *havep, const cap_rights_t *needp)
173 {
174 
175 	return (_cap_check(havep, needp, CAPFAIL_NOTCAPABLE));
176 }
177 
178 /*
179  * Convert capability rights into VM access flags.
180  */
181 u_char
182 cap_rights_to_vmprot(cap_rights_t *havep)
183 {
184 	u_char maxprot;
185 
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;
193 
194 	return (maxprot);
195 }
196 
197 /*
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
200  * this one file.
201  */
202 cap_rights_t *
203 cap_rights(struct filedesc *fdp, int fd)
204 {
205 
206 	return (&fdp->fd_ofiles[fd].fde_rights);
207 }
208 
209 /*
210  * System call to limit rights of the given capability.
211  */
212 int
213 sys_cap_rights_limit(struct thread *td, struct cap_rights_limit_args *uap)
214 {
215 	struct filedesc *fdp;
216 	cap_rights_t rights;
217 	int error, fd, version;
218 
219 	cap_rights_init(&rights);
220 
221 	error = copyin(uap->rightsp, &rights, sizeof(rights.cr_rights[0]));
222 	if (error != 0)
223 		return (error);
224 	version = CAPVER(&rights);
225 	if (version != CAP_RIGHTS_VERSION_00)
226 		return (EINVAL);
227 
228 	error = copyin(uap->rightsp, &rights,
229 	    sizeof(rights.cr_rights[0]) * CAPARSIZE(&rights));
230 	if (error != 0)
231 		return (error);
232 	/* Check for race. */
233 	if (CAPVER(&rights) != version)
234 		return (EINVAL);
235 
236 	if (!cap_rights_is_valid(&rights))
237 		return (EINVAL);
238 
239 	if (version != CAP_RIGHTS_VERSION) {
240 		rights.cr_rights[0] &= ~(0x3ULL << 62);
241 		rights.cr_rights[0] |= ((uint64_t)CAP_RIGHTS_VERSION << 62);
242 	}
243 #ifdef KTRACE
244 	if (KTRPOINT(td, KTR_STRUCT))
245 		ktrcaprights(&rights);
246 #endif
247 
248 	fd = uap->fd;
249 
250 	AUDIT_ARG_FD(fd);
251 	AUDIT_ARG_RIGHTS(&rights);
252 
253 	fdp = td->td_proc->p_fd;
254 	FILEDESC_XLOCK(fdp);
255 	if (fget_locked(fdp, fd) == NULL) {
256 		FILEDESC_XUNLOCK(fdp);
257 		return (EBADF);
258 	}
259 	error = _cap_check(cap_rights(fdp, fd), &rights, CAPFAIL_INCREASE);
260 	if (error == 0) {
261 		fdp->fd_ofiles[fd].fde_rights = rights;
262 		if (!cap_rights_is_set(&rights, CAP_IOCTL)) {
263 			free(fdp->fd_ofiles[fd].fde_ioctls, M_FILECAPS);
264 			fdp->fd_ofiles[fd].fde_ioctls = NULL;
265 			fdp->fd_ofiles[fd].fde_nioctls = 0;
266 		}
267 		if (!cap_rights_is_set(&rights, CAP_FCNTL))
268 			fdp->fd_ofiles[fd].fde_fcntls = 0;
269 	}
270 	FILEDESC_XUNLOCK(fdp);
271 	return (error);
272 }
273 
274 /*
275  * System call to query the rights mask associated with a capability.
276  */
277 int
278 sys___cap_rights_get(struct thread *td, struct __cap_rights_get_args *uap)
279 {
280 	struct filedesc *fdp;
281 	cap_rights_t rights;
282 	int error, fd, i, n;
283 
284 	if (uap->version != CAP_RIGHTS_VERSION_00)
285 		return (EINVAL);
286 
287 	fd = uap->fd;
288 
289 	AUDIT_ARG_FD(fd);
290 
291 	fdp = td->td_proc->p_fd;
292 	FILEDESC_SLOCK(fdp);
293 	if (fget_locked(fdp, fd) == NULL) {
294 		FILEDESC_SUNLOCK(fdp);
295 		return (EBADF);
296 	}
297 	rights = *cap_rights(fdp, fd);
298 	FILEDESC_SUNLOCK(fdp);
299 	n = uap->version + 2;
300 	if (uap->version != CAPVER(&rights)) {
301 		/*
302 		 * For older versions we need to check if the descriptor
303 		 * doesn't contain rights not understood by the caller.
304 		 * If it does, we have to return an error.
305 		 */
306 		for (i = n; i < CAPARSIZE(&rights); i++) {
307 			if ((rights.cr_rights[i] & ~(0x7FULL << 57)) != 0)
308 				return (EINVAL);
309 		}
310 	}
311 	error = copyout(&rights, uap->rightsp, sizeof(rights.cr_rights[0]) * n);
312 #ifdef KTRACE
313 	if (error == 0 && KTRPOINT(td, KTR_STRUCT))
314 		ktrcaprights(&rights);
315 #endif
316 	return (error);
317 }
318 
319 /*
320  * Test whether a capability grants the given ioctl command.
321  * If descriptor doesn't have CAP_IOCTL, then ioctls list is empty and
322  * ENOTCAPABLE will be returned.
323  */
324 int
325 cap_ioctl_check(struct filedesc *fdp, int fd, u_long cmd)
326 {
327 	u_long *cmds;
328 	ssize_t ncmds;
329 	long i;
330 
331 	FILEDESC_LOCK_ASSERT(fdp);
332 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
333 	    ("%s: invalid fd=%d", __func__, fd));
334 
335 	ncmds = fdp->fd_ofiles[fd].fde_nioctls;
336 	if (ncmds == -1)
337 		return (0);
338 
339 	cmds = fdp->fd_ofiles[fd].fde_ioctls;
340 	for (i = 0; i < ncmds; i++) {
341 		if (cmds[i] == cmd)
342 			return (0);
343 	}
344 
345 	return (ENOTCAPABLE);
346 }
347 
348 /*
349  * Check if the current ioctls list can be replaced by the new one.
350  */
351 static int
352 cap_ioctl_limit_check(struct filedesc *fdp, int fd, const u_long *cmds,
353     size_t ncmds)
354 {
355 	u_long *ocmds;
356 	ssize_t oncmds;
357 	u_long i;
358 	long j;
359 
360 	oncmds = fdp->fd_ofiles[fd].fde_nioctls;
361 	if (oncmds == -1)
362 		return (0);
363 	if (oncmds < (ssize_t)ncmds)
364 		return (ENOTCAPABLE);
365 
366 	ocmds = fdp->fd_ofiles[fd].fde_ioctls;
367 	for (i = 0; i < ncmds; i++) {
368 		for (j = 0; j < oncmds; j++) {
369 			if (cmds[i] == ocmds[j])
370 				break;
371 		}
372 		if (j == oncmds)
373 			return (ENOTCAPABLE);
374 	}
375 
376 	return (0);
377 }
378 
379 int
380 kern_cap_ioctls_limit(struct thread *td, int fd, u_long *cmds, size_t ncmds)
381 {
382 	struct filedesc *fdp;
383 	u_long *ocmds;
384 	int error;
385 
386 	AUDIT_ARG_FD(fd);
387 
388 	fdp = td->td_proc->p_fd;
389 	FILEDESC_XLOCK(fdp);
390 
391 	if (fget_locked(fdp, fd) == NULL) {
392 		error = EBADF;
393 		goto out;
394 	}
395 
396 	error = cap_ioctl_limit_check(fdp, fd, cmds, ncmds);
397 	if (error != 0)
398 		goto out;
399 
400 	ocmds = fdp->fd_ofiles[fd].fde_ioctls;
401 	fdp->fd_ofiles[fd].fde_ioctls = cmds;
402 	fdp->fd_ofiles[fd].fde_nioctls = ncmds;
403 
404 	cmds = ocmds;
405 	error = 0;
406 out:
407 	FILEDESC_XUNLOCK(fdp);
408 	free(cmds, M_FILECAPS);
409 	return (error);
410 }
411 
412 int
413 sys_cap_ioctls_limit(struct thread *td, struct cap_ioctls_limit_args *uap)
414 {
415 	u_long *cmds;
416 	size_t ncmds;
417 	int error;
418 
419 	ncmds = uap->ncmds;
420 
421 	if (ncmds > 256)	/* XXX: Is 256 sane? */
422 		return (EINVAL);
423 
424 	if (ncmds == 0) {
425 		cmds = NULL;
426 	} else {
427 		cmds = malloc(sizeof(cmds[0]) * ncmds, M_FILECAPS, M_WAITOK);
428 		error = copyin(uap->cmds, cmds, sizeof(cmds[0]) * ncmds);
429 		if (error != 0) {
430 			free(cmds, M_FILECAPS);
431 			return (error);
432 		}
433 	}
434 
435 	return (kern_cap_ioctls_limit(td, uap->fd, cmds, ncmds));
436 }
437 
438 int
439 sys_cap_ioctls_get(struct thread *td, struct cap_ioctls_get_args *uap)
440 {
441 	struct filedesc *fdp;
442 	struct filedescent *fdep;
443 	u_long *cmds;
444 	size_t maxcmds;
445 	int error, fd;
446 
447 	fd = uap->fd;
448 	cmds = uap->cmds;
449 	maxcmds = uap->maxcmds;
450 
451 	AUDIT_ARG_FD(fd);
452 
453 	fdp = td->td_proc->p_fd;
454 	FILEDESC_SLOCK(fdp);
455 
456 	if (fget_locked(fdp, fd) == NULL) {
457 		error = EBADF;
458 		goto out;
459 	}
460 
461 	/*
462 	 * If all ioctls are allowed (fde_nioctls == -1 && fde_ioctls == NULL)
463 	 * the only sane thing we can do is to not populate the given array and
464 	 * return CAP_IOCTLS_ALL.
465 	 */
466 
467 	fdep = &fdp->fd_ofiles[fd];
468 	if (cmds != NULL && fdep->fde_ioctls != NULL) {
469 		error = copyout(fdep->fde_ioctls, cmds,
470 		    sizeof(cmds[0]) * MIN(fdep->fde_nioctls, maxcmds));
471 		if (error != 0)
472 			goto out;
473 	}
474 	if (fdep->fde_nioctls == -1)
475 		td->td_retval[0] = CAP_IOCTLS_ALL;
476 	else
477 		td->td_retval[0] = fdep->fde_nioctls;
478 
479 	error = 0;
480 out:
481 	FILEDESC_SUNLOCK(fdp);
482 	return (error);
483 }
484 
485 /*
486  * Test whether a capability grants the given fcntl command.
487  */
488 int
489 cap_fcntl_check(struct filedesc *fdp, int fd, int cmd)
490 {
491 	uint32_t fcntlcap;
492 
493 	KASSERT(fd >= 0 && fd < fdp->fd_nfiles,
494 	    ("%s: invalid fd=%d", __func__, fd));
495 
496 	fcntlcap = (1 << cmd);
497 	KASSERT((CAP_FCNTL_ALL & fcntlcap) != 0,
498 	    ("Unsupported fcntl=%d.", cmd));
499 
500 	if ((fdp->fd_ofiles[fd].fde_fcntls & fcntlcap) != 0)
501 		return (0);
502 
503 	return (ENOTCAPABLE);
504 }
505 
506 int
507 sys_cap_fcntls_limit(struct thread *td, struct cap_fcntls_limit_args *uap)
508 {
509 	struct filedesc *fdp;
510 	uint32_t fcntlrights;
511 	int fd;
512 
513 	fd = uap->fd;
514 	fcntlrights = uap->fcntlrights;
515 
516 	AUDIT_ARG_FD(fd);
517 	AUDIT_ARG_FCNTL_RIGHTS(fcntlrights);
518 
519 	if ((fcntlrights & ~CAP_FCNTL_ALL) != 0)
520 		return (EINVAL);
521 
522 	fdp = td->td_proc->p_fd;
523 	FILEDESC_XLOCK(fdp);
524 
525 	if (fget_locked(fdp, fd) == NULL) {
526 		FILEDESC_XUNLOCK(fdp);
527 		return (EBADF);
528 	}
529 
530 	if ((fcntlrights & ~fdp->fd_ofiles[fd].fde_fcntls) != 0) {
531 		FILEDESC_XUNLOCK(fdp);
532 		return (ENOTCAPABLE);
533 	}
534 
535 	fdp->fd_ofiles[fd].fde_fcntls = fcntlrights;
536 	FILEDESC_XUNLOCK(fdp);
537 
538 	return (0);
539 }
540 
541 int
542 sys_cap_fcntls_get(struct thread *td, struct cap_fcntls_get_args *uap)
543 {
544 	struct filedesc *fdp;
545 	uint32_t rights;
546 	int fd;
547 
548 	fd = uap->fd;
549 
550 	AUDIT_ARG_FD(fd);
551 
552 	fdp = td->td_proc->p_fd;
553 	FILEDESC_SLOCK(fdp);
554 	if (fget_locked(fdp, fd) == NULL) {
555 		FILEDESC_SUNLOCK(fdp);
556 		return (EBADF);
557 	}
558 	rights = fdp->fd_ofiles[fd].fde_fcntls;
559 	FILEDESC_SUNLOCK(fdp);
560 
561 	return (copyout(&rights, uap->fcntlrightsp, sizeof(rights)));
562 }
563 
564 #else /* !CAPABILITIES */
565 
566 /*
567  * Stub Capability functions for when options CAPABILITIES isn't compiled
568  * into the kernel.
569  */
570 
571 int
572 sys_cap_rights_limit(struct thread *td, struct cap_rights_limit_args *uap)
573 {
574 
575 	return (ENOSYS);
576 }
577 
578 int
579 sys___cap_rights_get(struct thread *td, struct __cap_rights_get_args *uap)
580 {
581 
582 	return (ENOSYS);
583 }
584 
585 int
586 sys_cap_ioctls_limit(struct thread *td, struct cap_ioctls_limit_args *uap)
587 {
588 
589 	return (ENOSYS);
590 }
591 
592 int
593 sys_cap_ioctls_get(struct thread *td, struct cap_ioctls_get_args *uap)
594 {
595 
596 	return (ENOSYS);
597 }
598 
599 int
600 sys_cap_fcntls_limit(struct thread *td, struct cap_fcntls_limit_args *uap)
601 {
602 
603 	return (ENOSYS);
604 }
605 
606 int
607 sys_cap_fcntls_get(struct thread *td, struct cap_fcntls_get_args *uap)
608 {
609 
610 	return (ENOSYS);
611 }
612 
613 #endif /* CAPABILITIES */
614