1/*
2 * Seccomp filter example for x86 (32-bit and 64-bit) with BPF macros
3 *
4 * Copyright (c) 2012 The Chromium OS Authors <chromium-os-dev@chromium.org>
5 * Author: Will Drewry <wad@chromium.org>
6 *
7 * The code may be used by anyone for any purpose,
8 * and can serve as a starting point for developing
9 * applications using prctl(PR_SET_SECCOMP, 2, ...).
10 */
11#if defined(__i386__) || defined(__x86_64__)
12#define SUPPORTED_ARCH 1
13#endif
14
15#if defined(SUPPORTED_ARCH)
16#define __USE_GNU 1
17#define _GNU_SOURCE 1
18
19#include <linux/types.h>
20#include <linux/filter.h>
21#include <linux/seccomp.h>
22#include <linux/unistd.h>
23#include <signal.h>
24#include <stdio.h>
25#include <stddef.h>
26#include <string.h>
27#include <sys/prctl.h>
28#include <unistd.h>
29
30#define syscall_arg(_n) (offsetof(struct seccomp_data, args[_n]))
31#define syscall_nr (offsetof(struct seccomp_data, nr))
32
33#if defined(__i386__)
34#define REG_RESULT	REG_EAX
35#define REG_SYSCALL	REG_EAX
36#define REG_ARG0	REG_EBX
37#define REG_ARG1	REG_ECX
38#define REG_ARG2	REG_EDX
39#define REG_ARG3	REG_ESI
40#define REG_ARG4	REG_EDI
41#define REG_ARG5	REG_EBP
42#elif defined(__x86_64__)
43#define REG_RESULT	REG_RAX
44#define REG_SYSCALL	REG_RAX
45#define REG_ARG0	REG_RDI
46#define REG_ARG1	REG_RSI
47#define REG_ARG2	REG_RDX
48#define REG_ARG3	REG_R10
49#define REG_ARG4	REG_R8
50#define REG_ARG5	REG_R9
51#endif
52
53#ifndef PR_SET_NO_NEW_PRIVS
54#define PR_SET_NO_NEW_PRIVS 38
55#endif
56
57#ifndef SYS_SECCOMP
58#define SYS_SECCOMP 1
59#endif
60
61static void emulator(int nr, siginfo_t *info, void *void_context)
62{
63	ucontext_t *ctx = (ucontext_t *)(void_context);
64	int syscall;
65	char *buf;
66	ssize_t bytes;
67	size_t len;
68	if (info->si_code != SYS_SECCOMP)
69		return;
70	if (!ctx)
71		return;
72	syscall = ctx->uc_mcontext.gregs[REG_SYSCALL];
73	buf = (char *) ctx->uc_mcontext.gregs[REG_ARG1];
74	len = (size_t) ctx->uc_mcontext.gregs[REG_ARG2];
75
76	if (syscall != __NR_write)
77		return;
78	if (ctx->uc_mcontext.gregs[REG_ARG0] != STDERR_FILENO)
79		return;
80	/* Redirect stderr messages to stdout. Doesn't handle EINTR, etc */
81	ctx->uc_mcontext.gregs[REG_RESULT] = -1;
82	if (write(STDOUT_FILENO, "[ERR] ", 6) > 0) {
83		bytes = write(STDOUT_FILENO, buf, len);
84		ctx->uc_mcontext.gregs[REG_RESULT] = bytes;
85	}
86	return;
87}
88
89static int install_emulator(void)
90{
91	struct sigaction act;
92	sigset_t mask;
93	memset(&act, 0, sizeof(act));
94	sigemptyset(&mask);
95	sigaddset(&mask, SIGSYS);
96
97	act.sa_sigaction = &emulator;
98	act.sa_flags = SA_SIGINFO;
99	if (sigaction(SIGSYS, &act, NULL) < 0) {
100		perror("sigaction");
101		return -1;
102	}
103	if (sigprocmask(SIG_UNBLOCK, &mask, NULL)) {
104		perror("sigprocmask");
105		return -1;
106	}
107	return 0;
108}
109
110static int install_filter(void)
111{
112	struct sock_filter filter[] = {
113		/* Grab the system call number */
114		BPF_STMT(BPF_LD+BPF_W+BPF_ABS, syscall_nr),
115		/* Jump table for the allowed syscalls */
116		BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, __NR_rt_sigreturn, 0, 1),
117		BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ALLOW),
118#ifdef __NR_sigreturn
119		BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, __NR_sigreturn, 0, 1),
120		BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ALLOW),
121#endif
122		BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, __NR_exit_group, 0, 1),
123		BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ALLOW),
124		BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, __NR_exit, 0, 1),
125		BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ALLOW),
126		BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, __NR_read, 1, 0),
127		BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, __NR_write, 3, 2),
128
129		/* Check that read is only using stdin. */
130		BPF_STMT(BPF_LD+BPF_W+BPF_ABS, syscall_arg(0)),
131		BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, STDIN_FILENO, 4, 0),
132		BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_KILL),
133
134		/* Check that write is only using stdout */
135		BPF_STMT(BPF_LD+BPF_W+BPF_ABS, syscall_arg(0)),
136		BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, STDOUT_FILENO, 1, 0),
137		/* Trap attempts to write to stderr */
138		BPF_JUMP(BPF_JMP+BPF_JEQ+BPF_K, STDERR_FILENO, 1, 2),
139
140		BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_ALLOW),
141		BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_TRAP),
142		BPF_STMT(BPF_RET+BPF_K, SECCOMP_RET_KILL),
143	};
144	struct sock_fprog prog = {
145		.len = (unsigned short)(sizeof(filter)/sizeof(filter[0])),
146		.filter = filter,
147	};
148
149	if (prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0)) {
150		perror("prctl(NO_NEW_PRIVS)");
151		return 1;
152	}
153
154
155	if (prctl(PR_SET_SECCOMP, SECCOMP_MODE_FILTER, &prog)) {
156		perror("prctl");
157		return 1;
158	}
159	return 0;
160}
161
162#define payload(_c) (_c), sizeof((_c))
163int main(int argc, char **argv)
164{
165	char buf[4096];
166	ssize_t bytes = 0;
167	if (install_emulator())
168		return 1;
169	if (install_filter())
170		return 1;
171	syscall(__NR_write, STDOUT_FILENO,
172		payload("OHAI! WHAT IS YOUR NAME? "));
173	bytes = syscall(__NR_read, STDIN_FILENO, buf, sizeof(buf));
174	syscall(__NR_write, STDOUT_FILENO, payload("HELLO, "));
175	syscall(__NR_write, STDOUT_FILENO, buf, bytes);
176	syscall(__NR_write, STDERR_FILENO,
177		payload("Error message going to STDERR\n"));
178	return 0;
179}
180#else	/* SUPPORTED_ARCH */
181/*
182 * This sample is x86-only.  Since kernel samples are compiled with the
183 * host toolchain, a non-x86 host will result in using only the main()
184 * below.
185 */
186int main(void)
187{
188	return 1;
189}
190#endif	/* SUPPORTED_ARCH */
191