Skip to content

Commit 0a8e411

Browse files
committed
nsenter: clone /proc/self/exe to avoid exposing host binary to container
There are quite a few circumstances where /proc/self/exe pointing to a pretty important container binary is a _bad_ thing, so to avoid this we have to make a copy (preferably doing self-clean-up and not being writeable). We require memfd_create(2) -- though there is an O_TMPFILE fallback -- but we can always extend this to use a scratch MNT_DETACH overlayfs or tmpfs. The main downside to this approach is no page-cache sharing for the runc binary (which overlayfs would give us) but this is far less complicated. This is only done during nsenter so that it happens transparently to the Go code, and any libcontainer users benefit from it. This also makes ExtraFiles and --preserve-fds handling trivial (because we don't need to worry about it). Fixes: CVE-2019-5736 Co-developed-by: Christian Brauner <[email protected]> Signed-off-by: Aleksa Sarai <[email protected]>
1 parent dd023c4 commit 0a8e411

File tree

2 files changed

+279
-0
lines changed

2 files changed

+279
-0
lines changed

libcontainer/nsenter/cloned_binary.c

+268
Original file line numberDiff line numberDiff line change
@@ -0,0 +1,268 @@
1+
/*
2+
* Copyright (C) 2019 Aleksa Sarai <[email protected]>
3+
* Copyright (C) 2019 SUSE LLC
4+
*
5+
* Licensed under the Apache License, Version 2.0 (the "License");
6+
* you may not use this file except in compliance with the License.
7+
* You may obtain a copy of the License at
8+
*
9+
* http://www.apache.org/licenses/LICENSE-2.0
10+
*
11+
* Unless required by applicable law or agreed to in writing, software
12+
* distributed under the License is distributed on an "AS IS" BASIS,
13+
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
14+
* See the License for the specific language governing permissions and
15+
* limitations under the License.
16+
*/
17+
18+
#define _GNU_SOURCE
19+
#include <unistd.h>
20+
#include <stdio.h>
21+
#include <stdlib.h>
22+
#include <stdbool.h>
23+
#include <string.h>
24+
#include <limits.h>
25+
#include <fcntl.h>
26+
#include <errno.h>
27+
28+
#include <sys/types.h>
29+
#include <sys/stat.h>
30+
#include <sys/vfs.h>
31+
#include <sys/mman.h>
32+
#include <sys/sendfile.h>
33+
#include <sys/syscall.h>
34+
35+
/* Use our own wrapper for memfd_create. */
36+
#if !defined(SYS_memfd_create) && defined(__NR_memfd_create)
37+
# define SYS_memfd_create __NR_memfd_create
38+
#endif
39+
#ifdef SYS_memfd_create
40+
# define HAVE_MEMFD_CREATE
41+
/* memfd_create(2) flags -- copied from <linux/memfd.h>. */
42+
# ifndef MFD_CLOEXEC
43+
# define MFD_CLOEXEC 0x0001U
44+
# define MFD_ALLOW_SEALING 0x0002U
45+
# endif
46+
int memfd_create(const char *name, unsigned int flags)
47+
{
48+
return syscall(SYS_memfd_create, name, flags);
49+
}
50+
#endif
51+
52+
/* This comes directly from <linux/fcntl.h>. */
53+
#ifndef F_LINUX_SPECIFIC_BASE
54+
# define F_LINUX_SPECIFIC_BASE 1024
55+
#endif
56+
#ifndef F_ADD_SEALS
57+
# define F_ADD_SEALS (F_LINUX_SPECIFIC_BASE + 9)
58+
# define F_GET_SEALS (F_LINUX_SPECIFIC_BASE + 10)
59+
#endif
60+
#ifndef F_SEAL_SEAL
61+
# define F_SEAL_SEAL 0x0001 /* prevent further seals from being set */
62+
# define F_SEAL_SHRINK 0x0002 /* prevent file from shrinking */
63+
# define F_SEAL_GROW 0x0004 /* prevent file from growing */
64+
# define F_SEAL_WRITE 0x0008 /* prevent writes */
65+
#endif
66+
67+
#define RUNC_SENDFILE_MAX 0x7FFFF000 /* sendfile(2) is limited to 2GB. */
68+
#ifdef HAVE_MEMFD_CREATE
69+
# define RUNC_MEMFD_COMMENT "runc_cloned:/proc/self/exe"
70+
# define RUNC_MEMFD_SEALS \
71+
(F_SEAL_SEAL | F_SEAL_SHRINK | F_SEAL_GROW | F_SEAL_WRITE)
72+
#endif
73+
74+
static void *must_realloc(void *ptr, size_t size)
75+
{
76+
void *old = ptr;
77+
do {
78+
ptr = realloc(old, size);
79+
} while(!ptr);
80+
return ptr;
81+
}
82+
83+
/*
84+
* Verify whether we are currently in a self-cloned program (namely, is
85+
* /proc/self/exe a memfd). F_GET_SEALS will only succeed for memfds (or rather
86+
* for shmem files), and we want to be sure it's actually sealed.
87+
*/
88+
static int is_self_cloned(void)
89+
{
90+
int fd, ret, is_cloned = 0;
91+
92+
fd = open("/proc/self/exe", O_RDONLY|O_CLOEXEC);
93+
if (fd < 0)
94+
return -ENOTRECOVERABLE;
95+
96+
#ifdef HAVE_MEMFD_CREATE
97+
ret = fcntl(fd, F_GET_SEALS);
98+
is_cloned = (ret == RUNC_MEMFD_SEALS);
99+
#else
100+
struct stat statbuf = {0};
101+
ret = fstat(fd, &statbuf);
102+
if (ret >= 0)
103+
is_cloned = (statbuf.st_nlink == 0);
104+
#endif
105+
close(fd);
106+
return is_cloned;
107+
}
108+
109+
/*
110+
* Basic wrapper around mmap(2) that gives you the file length so you can
111+
* safely treat it as an ordinary buffer. Only gives you read access.
112+
*/
113+
static char *read_file(char *path, size_t *length)
114+
{
115+
int fd;
116+
char buf[4096], *copy = NULL;
117+
118+
if (!length)
119+
return NULL;
120+
121+
fd = open(path, O_RDONLY | O_CLOEXEC);
122+
if (fd < 0)
123+
return NULL;
124+
125+
*length = 0;
126+
for (;;) {
127+
int n;
128+
129+
n = read(fd, buf, sizeof(buf));
130+
if (n < 0)
131+
goto error;
132+
if (!n)
133+
break;
134+
135+
copy = must_realloc(copy, (*length + n) * sizeof(*copy));
136+
memcpy(copy + *length, buf, n);
137+
*length += n;
138+
}
139+
close(fd);
140+
return copy;
141+
142+
error:
143+
close(fd);
144+
free(copy);
145+
return NULL;
146+
}
147+
148+
/*
149+
* A poor-man's version of "xargs -0". Basically parses a given block of
150+
* NUL-delimited data, within the given length and adds a pointer to each entry
151+
* to the array of pointers.
152+
*/
153+
static int parse_xargs(char *data, int data_length, char ***output)
154+
{
155+
int num = 0;
156+
char *cur = data;
157+
158+
if (!data || *output != NULL)
159+
return -1;
160+
161+
while (cur < data + data_length) {
162+
num++;
163+
*output = must_realloc(*output, (num + 1) * sizeof(**output));
164+
(*output)[num - 1] = cur;
165+
cur += strlen(cur) + 1;
166+
}
167+
(*output)[num] = NULL;
168+
return num;
169+
}
170+
171+
/*
172+
* "Parse" out argv and envp from /proc/self/cmdline and /proc/self/environ.
173+
* This is necessary because we are running in a context where we don't have a
174+
* main() that we can just get the arguments from.
175+
*/
176+
static int fetchve(char ***argv, char ***envp)
177+
{
178+
char *cmdline = NULL, *environ = NULL;
179+
size_t cmdline_size, environ_size;
180+
181+
cmdline = read_file("/proc/self/cmdline", &cmdline_size);
182+
if (!cmdline)
183+
goto error;
184+
environ = read_file("/proc/self/environ", &environ_size);
185+
if (!environ)
186+
goto error;
187+
188+
if (parse_xargs(cmdline, cmdline_size, argv) <= 0)
189+
goto error;
190+
if (parse_xargs(environ, environ_size, envp) <= 0)
191+
goto error;
192+
193+
return 0;
194+
195+
error:
196+
free(environ);
197+
free(cmdline);
198+
return -EINVAL;
199+
}
200+
201+
static int clone_binary(void)
202+
{
203+
int binfd, memfd;
204+
ssize_t sent = 0;
205+
206+
#ifdef HAVE_MEMFD_CREATE
207+
memfd = memfd_create(RUNC_MEMFD_COMMENT, MFD_CLOEXEC | MFD_ALLOW_SEALING);
208+
#else
209+
memfd = open("/tmp", O_TMPFILE | O_EXCL | O_RDWR | O_CLOEXEC, 0711);
210+
#endif
211+
if (memfd < 0)
212+
return -ENOTRECOVERABLE;
213+
214+
binfd = open("/proc/self/exe", O_RDONLY | O_CLOEXEC);
215+
if (binfd < 0)
216+
goto error;
217+
218+
sent = sendfile(memfd, binfd, NULL, RUNC_SENDFILE_MAX);
219+
close(binfd);
220+
if (sent < 0)
221+
goto error;
222+
223+
#ifdef HAVE_MEMFD_CREATE
224+
int err = fcntl(memfd, F_ADD_SEALS, RUNC_MEMFD_SEALS);
225+
if (err < 0)
226+
goto error;
227+
#else
228+
/* Need to re-open "memfd" as read-only to avoid execve(2) giving -EXTBUSY. */
229+
int newfd;
230+
char *fdpath = NULL;
231+
232+
if (asprintf(&fdpath, "/proc/self/fd/%d", memfd) < 0)
233+
goto error;
234+
newfd = open(fdpath, O_RDONLY | O_CLOEXEC);
235+
free(fdpath);
236+
if (newfd < 0)
237+
goto error;
238+
239+
close(memfd);
240+
memfd = newfd;
241+
#endif
242+
return memfd;
243+
244+
error:
245+
close(memfd);
246+
return -EIO;
247+
}
248+
249+
int ensure_cloned_binary(void)
250+
{
251+
int execfd;
252+
char **argv = NULL, **envp = NULL;
253+
254+
/* Check that we're not self-cloned, and if we are then bail. */
255+
int cloned = is_self_cloned();
256+
if (cloned > 0 || cloned == -ENOTRECOVERABLE)
257+
return cloned;
258+
259+
if (fetchve(&argv, &envp) < 0)
260+
return -EINVAL;
261+
262+
execfd = clone_binary();
263+
if (execfd < 0)
264+
return -EIO;
265+
266+
fexecve(execfd, argv, envp);
267+
return -ENOEXEC;
268+
}

libcontainer/nsenter/nsexec.c

+11
Original file line numberDiff line numberDiff line change
@@ -534,6 +534,9 @@ void join_namespaces(char *nslist)
534534
free(namespaces);
535535
}
536536

537+
/* Defined in cloned_binary.c. */
538+
extern int ensure_cloned_binary(void);
539+
537540
void nsexec(void)
538541
{
539542
int pipenum;
@@ -549,6 +552,14 @@ void nsexec(void)
549552
if (pipenum == -1)
550553
return;
551554

555+
/*
556+
* We need to re-exec if we are not in a cloned binary. This is necessary
557+
* to ensure that containers won't be able to access the host binary
558+
* through /proc/self/exe. See CVE-2019-5736.
559+
*/
560+
if (ensure_cloned_binary() < 0)
561+
bail("could not ensure we are a cloned binary");
562+
552563
/* Parse all of the netlink configuration. */
553564
nl_parse(pipenum, &config);
554565

0 commit comments

Comments
 (0)