netdev - [PATCH v2 net] x86: bpf_jit: fix two bugs in eBPF JIT compiler

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [thread-next>] [day] [month] [year] [list]
Message-Id: <1412998223-3805-1-git-send-email-ast@plumgrid.com>
Date:	Fri, 10 Oct 2014 20:30:23 -0700
From:	Alexei Starovoitov <ast@...mgrid.com>
To:	"David S. Miller" <davem@...emloft.net>
Cc:	"Darrick J. Wong" <darrick.wong@...cle.com>,
	Eric Dumazet <edumazet@...gle.com>,
	Daniel Borkmann <dborkman@...hat.com>,
	"H. Peter Anvin" <hpa@...or.com>,
	Thomas Gleixner <tglx@...utronix.de>,
	Ingo Molnar <mingo@...nel.org>, netdev@...r.kernel.org,
	linux-kernel@...r.kernel.org
Subject: [PATCH v2 net] x86: bpf_jit: fix two bugs in eBPF JIT compiler

1.
JIT compiler using multi-pass approach to converge to final image size,
since x86 instructions are variable length. It starts with large
gaps between instructions (so some jumps may use imm32 instead of imm8)
and iterates until total program size is the same as in previous pass.
This algorithm works only if program size is strictly decreasing.
Programs that use LD_ABS insn need additional code in prologue, but it
was not emitted during 1st pass, so there was a chance that 2nd pass would
adjust imm32->imm8 jump offsets to the same number of bytes as increase in
prologue, which may cause algorithm to erroneously decide that size converged.
Fix it by always emitting largest prologue in the first pass which
is detected by oldproglen==0 check.
Also change error check condition 'proglen != oldproglen' to fail gracefully.

2.
while staring at the code realized that 64-byte buffer may not be enough
when 1st insn is large, so increase it to 128 to avoid buffer overflow
(theoretical maximum size of prologue+div is 109) and add runtime check.

Fixes: 622582786c9e ("net: filter: x86: internal BPF JIT")
Reported-by: Darrick J. Wong <darrick.wong@...cle.com>
Signed-off-by: Alexei Starovoitov <ast@...mgrid.com>
---
v1->v2: reduce chances of stack corruption in case of future bugs (suggested by Eric)

note in classic BPF programs 1st insn is always short move, but native eBPF
programs may trigger buffer overflow. I couldn't force the crash with overflow,
since there are no further calls while this part of stack is used.
Both are ugly bugs regardless.
When net-next opens I will add narrowed down testcase from 'nmap' to testsuite.

 arch/x86/net/bpf_jit_comp.c |   25 +++++++++++++++++++------
 1 file changed, 19 insertions(+), 6 deletions(-)

diff --git a/arch/x86/net/bpf_jit_comp.c b/arch/x86/net/bpf_jit_comp.c
index d56cd1f..3f62734 100644
--- a/arch/x86/net/bpf_jit_comp.c
+++ b/arch/x86/net/bpf_jit_comp.c
@@ -182,12 +182,17 @@ struct jit_context {
 	bool seen_ld_abs;
 };
 
+/* maximum number of bytes emitted while JITing one eBPF insn */
+#define BPF_MAX_INSN_SIZE	128
+#define BPF_INSN_SAFETY		64
+
 static int do_jit(struct bpf_prog *bpf_prog, int *addrs, u8 *image,
 		  int oldproglen, struct jit_context *ctx)
 {
 	struct bpf_insn *insn = bpf_prog->insnsi;
 	int insn_cnt = bpf_prog->len;
-	u8 temp[64];
+	bool seen_ld_abs = ctx->seen_ld_abs | (oldproglen == 0);
+	u8 temp[BPF_MAX_INSN_SIZE + BPF_INSN_SAFETY];
 	int i;
 	int proglen = 0;
 	u8 *prog = temp;
@@ -225,7 +230,7 @@ static int do_jit(struct bpf_prog *bpf_prog, int *addrs, u8 *image,
 	EMIT2(0x31, 0xc0); /* xor eax, eax */
 	EMIT3(0x4D, 0x31, 0xED); /* xor r13, r13 */
 
-	if (ctx->seen_ld_abs) {
+	if (seen_ld_abs) {
 		/* r9d : skb->len - skb->data_len (headlen)
 		 * r10 : skb->data
 		 */
@@ -685,7 +690,7 @@ xadd:			if (is_imm8(insn->off))
 		case BPF_JMP | BPF_CALL:
 			func = (u8 *) __bpf_call_base + imm32;
 			jmp_offset = func - (image + addrs[i]);
-			if (ctx->seen_ld_abs) {
+			if (seen_ld_abs) {
 				EMIT2(0x41, 0x52); /* push %r10 */
 				EMIT2(0x41, 0x51); /* push %r9 */
 				/* need to adjust jmp offset, since
@@ -699,7 +704,7 @@ xadd:			if (is_imm8(insn->off))
 				return -EINVAL;
 			}
 			EMIT1_off32(0xE8, jmp_offset);
-			if (ctx->seen_ld_abs) {
+			if (seen_ld_abs) {
 				EMIT2(0x41, 0x59); /* pop %r9 */
 				EMIT2(0x41, 0x5A); /* pop %r10 */
 			}
@@ -804,7 +809,8 @@ emit_jmp:
 			goto common_load;
 		case BPF_LD | BPF_ABS | BPF_W:
 			func = CHOOSE_LOAD_FUNC(imm32, sk_load_word);
-common_load:		ctx->seen_ld_abs = true;
+common_load:
+			ctx->seen_ld_abs = seen_ld_abs = true;
 			jmp_offset = func - (image + addrs[i]);
 			if (!func || !is_simm32(jmp_offset)) {
 				pr_err("unsupported bpf func %d addr %p image %p\n",
@@ -878,6 +884,11 @@ common_load:		ctx->seen_ld_abs = true;
 		}
 
 		ilen = prog - temp;
+		if (ilen > BPF_MAX_INSN_SIZE) {
+			pr_err("bpf_jit_compile fatal insn size error\n");
+			return -EFAULT;
+		}
+
 		if (image) {
 			if (unlikely(proglen + ilen > oldproglen)) {
 				pr_err("bpf_jit_compile fatal error\n");
@@ -934,9 +945,11 @@ void bpf_int_jit_compile(struct bpf_prog *prog)
 			goto out;
 		}
 		if (image) {
-			if (proglen != oldproglen)
+			if (proglen != oldproglen) {
 				pr_err("bpf_jit: proglen=%d != oldproglen=%d\n",
 				       proglen, oldproglen);
+				goto out;
+			}
 			break;
 		}
 		if (proglen == oldproglen) {
-- 
1.7.9.5

--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html