diff mbox

Prevent reading uninitialized memory with socketfilters

Message ID 20101110.103923.59670339.davem@davemloft.net
State Accepted, archived
Delegated to: David Miller
Headers show

Commit Message

David Miller Nov. 10, 2010, 6:39 p.m. UTC 
From: Tetsuo Handa <penguin-kernel@I-love.SAKURA.ne.jp>
Date: Wed, 10 Nov 2010 23:25:08 +0900

> Just I thought...
> 
>> unsigned int sk_run_filter(struct sk_buff *skb, struct sock_filter *filter, int flen)
>> {
>> 	struct sock_filter *fentry;     /* We walk down these */
> Can't this be "const struct sock_filter *"?
>> (...snipped...)
>> 	for (pc = 0; pc < flen; pc++) {
>> 		fentry = &filter[pc];
> Can't we do
> 		u32 f_k = fentry->k;
> and replace 27 repetition of fentry->k with f_k?

Ok, here is what I'm adding to net-2.6 (and will submit to -stable),
if anyone can spot any silly errors let me know now.

Thanks!

--------------------
filter: make sure filters dont read uninitialized memory

There is a possibility malicious users can get limited information about
uninitialized stack mem array. Even if sk_run_filter() result is bound
to packet length (0 .. 65535), we could imagine this can be used by
hostile user.

Initializing mem[] array, like Dan Rosenberg suggested in his patch is
expensive since most filters dont even use this array.

Its hard to make the filter validation in sk_chk_filter(), because of
the jumps. This might be done later.

In this patch, I use a bitmap (a single long var) so that only filters
using mem[] loads/stores pay the price of added security checks.

For other filters, additional cost is a single instruction.

[ Since we access fentry->k a lot now, cache it in a local variable
  and mark filter entry pointer as const. -DaveM ]

Reported-by: Dan Rosenberg <drosenberg@vsecurity.com>
Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
---
 net/core/filter.c |   64 +++++++++++++++++++++++++++++------------------------
 1 files changed, 35 insertions(+), 29 deletions(-)

Comments

Ben Hutchings Nov. 10, 2010, 8:57 p.m. UTC | #1 
On Wed, 2010-11-10 at 10:39 -0800, David Miller wrote:
[...]
> In this patch, I use a bitmap (a single long var) so that only filters
> using mem[] loads/stores pay the price of added security checks.
> 
> For other filters, additional cost is a single instruction.
> 
> [ Since we access fentry->k a lot now, cache it in a local variable
>   and mark filter entry pointer as const. -DaveM ]
[...]

I don't see the justification for combining these changes.  One patch,
one fix, right?

Ben.
David Miller Nov. 10, 2010, 8:59 p.m. UTC | #2 
From: Ben Hutchings <bhutchings@solarflare.com>
Date: Wed, 10 Nov 2010 20:57:44 +0000

> On Wed, 2010-11-10 at 10:39 -0800, David Miller wrote:
> [...]
>> In this patch, I use a bitmap (a single long var) so that only filters
>> using mem[] loads/stores pay the price of added security checks.
>> 
>> For other filters, additional cost is a single instruction.
>> 
>> [ Since we access fentry->k a lot now, cache it in a local variable
>>   and mark filter entry pointer as const. -DaveM ]
> [...]
> 
> I don't see the justification for combining these changes.  One patch,
> one fix, right?

I'm minimizing the performance impact of the new bitmap checks.
--
To unsubscribe from this list: send the line "unsubscribe netdev" in
the body of a message to majordomo@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Ben Hutchings Nov. 10, 2010, 9:25 p.m. UTC | #3 
On Wed, 2010-11-10 at 12:59 -0800, David Miller wrote:
> From: Ben Hutchings <bhutchings@solarflare.com>
> Date: Wed, 10 Nov 2010 20:57:44 +0000
> 
> > On Wed, 2010-11-10 at 10:39 -0800, David Miller wrote:
> > [...]
> >> In this patch, I use a bitmap (a single long var) so that only filters
> >> using mem[] loads/stores pay the price of added security checks.
> >> 
> >> For other filters, additional cost is a single instruction.
> >> 
> >> [ Since we access fentry->k a lot now, cache it in a local variable
> >>   and mark filter entry pointer as const. -DaveM ]
> > [...]
> > 
> > I don't see the justification for combining these changes.  One patch,
> > one fix, right?
> 
> I'm minimizing the performance impact of the new bitmap checks.

This seems like an entirely separate optimisation, since fentry->k was
*already* being used all over the place.  (And a smart compiler should
optimise that anyway... though I realise gcc is often not that smart.)

Ben.
diff mbox

Patch

diff --git a/net/core/filter.c b/net/core/filter.c
index 7beaec3..23e9b2a 100644
--- a/net/core/filter.c
+++ b/net/core/filter.c
@@ -112,39 +112,41 @@  EXPORT_SYMBOL(sk_filter);
  */
 unsigned int sk_run_filter(struct sk_buff *skb, struct sock_filter *filter, int flen)
 {
-	struct sock_filter *fentry;	/* We walk down these */
 	void *ptr;
 	u32 A = 0;			/* Accumulator */
 	u32 X = 0;			/* Index Register */
 	u32 mem[BPF_MEMWORDS];		/* Scratch Memory Store */
+	unsigned long memvalid = 0;
 	u32 tmp;
 	int k;
 	int pc;
 
+	BUILD_BUG_ON(BPF_MEMWORDS > BITS_PER_LONG);
 	/*
 	 * Process array of filter instructions.
 	 */
 	for (pc = 0; pc < flen; pc++) {
-		fentry = &filter[pc];
+		const struct sock_filter *fentry = &filter[pc];
+		u32 f_k = fentry->k;
 
 		switch (fentry->code) {
 		case BPF_S_ALU_ADD_X:
 			A += X;
 			continue;
 		case BPF_S_ALU_ADD_K:
-			A += fentry->k;
+			A += f_k;
 			continue;
 		case BPF_S_ALU_SUB_X:
 			A -= X;
 			continue;
 		case BPF_S_ALU_SUB_K:
-			A -= fentry->k;
+			A -= f_k;
 			continue;
 		case BPF_S_ALU_MUL_X:
 			A *= X;
 			continue;
 		case BPF_S_ALU_MUL_K:
-			A *= fentry->k;
+			A *= f_k;
 			continue;
 		case BPF_S_ALU_DIV_X:
 			if (X == 0)
@@ -152,49 +154,49 @@  unsigned int sk_run_filter(struct sk_buff *skb, struct sock_filter *filter, int
 			A /= X;
 			continue;
 		case BPF_S_ALU_DIV_K:
-			A /= fentry->k;
+			A /= f_k;
 			continue;
 		case BPF_S_ALU_AND_X:
 			A &= X;
 			continue;
 		case BPF_S_ALU_AND_K:
-			A &= fentry->k;
+			A &= f_k;
 			continue;
 		case BPF_S_ALU_OR_X:
 			A |= X;
 			continue;
 		case BPF_S_ALU_OR_K:
-			A |= fentry->k;
+			A |= f_k;
 			continue;
 		case BPF_S_ALU_LSH_X:
 			A <<= X;
 			continue;
 		case BPF_S_ALU_LSH_K:
-			A <<= fentry->k;
+			A <<= f_k;
 			continue;
 		case BPF_S_ALU_RSH_X:
 			A >>= X;
 			continue;
 		case BPF_S_ALU_RSH_K:
-			A >>= fentry->k;
+			A >>= f_k;
 			continue;
 		case BPF_S_ALU_NEG:
 			A = -A;
 			continue;
 		case BPF_S_JMP_JA:
-			pc += fentry->k;
+			pc += f_k;
 			continue;
 		case BPF_S_JMP_JGT_K:
-			pc += (A > fentry->k) ? fentry->jt : fentry->jf;
+			pc += (A > f_k) ? fentry->jt : fentry->jf;
 			continue;
 		case BPF_S_JMP_JGE_K:
-			pc += (A >= fentry->k) ? fentry->jt : fentry->jf;
+			pc += (A >= f_k) ? fentry->jt : fentry->jf;
 			continue;
 		case BPF_S_JMP_JEQ_K:
-			pc += (A == fentry->k) ? fentry->jt : fentry->jf;
+			pc += (A == f_k) ? fentry->jt : fentry->jf;
 			continue;
 		case BPF_S_JMP_JSET_K:
-			pc += (A & fentry->k) ? fentry->jt : fentry->jf;
+			pc += (A & f_k) ? fentry->jt : fentry->jf;
 			continue;
 		case BPF_S_JMP_JGT_X:
 			pc += (A > X) ? fentry->jt : fentry->jf;
@@ -209,7 +211,7 @@  unsigned int sk_run_filter(struct sk_buff *skb, struct sock_filter *filter, int
 			pc += (A & X) ? fentry->jt : fentry->jf;
 			continue;
 		case BPF_S_LD_W_ABS:
-			k = fentry->k;
+			k = f_k;
 load_w:
 			ptr = load_pointer(skb, k, 4, &tmp);
 			if (ptr != NULL) {
@@ -218,7 +220,7 @@  load_w:
 			}
 			break;
 		case BPF_S_LD_H_ABS:
-			k = fentry->k;
+			k = f_k;
 load_h:
 			ptr = load_pointer(skb, k, 2, &tmp);
 			if (ptr != NULL) {
@@ -227,7 +229,7 @@  load_h:
 			}
 			break;
 		case BPF_S_LD_B_ABS:
-			k = fentry->k;
+			k = f_k;
 load_b:
 			ptr = load_pointer(skb, k, 1, &tmp);
 			if (ptr != NULL) {
@@ -242,32 +244,34 @@  load_b:
 			X = skb->len;
 			continue;
 		case BPF_S_LD_W_IND:
-			k = X + fentry->k;
+			k = X + f_k;
 			goto load_w;
 		case BPF_S_LD_H_IND:
-			k = X + fentry->k;
+			k = X + f_k;
 			goto load_h;
 		case BPF_S_LD_B_IND:
-			k = X + fentry->k;
+			k = X + f_k;
 			goto load_b;
 		case BPF_S_LDX_B_MSH:
-			ptr = load_pointer(skb, fentry->k, 1, &tmp);
+			ptr = load_pointer(skb, f_k, 1, &tmp);
 			if (ptr != NULL) {
 				X = (*(u8 *)ptr & 0xf) << 2;
 				continue;
 			}
 			return 0;
 		case BPF_S_LD_IMM:
-			A = fentry->k;
+			A = f_k;
 			continue;
 		case BPF_S_LDX_IMM:
-			X = fentry->k;
+			X = f_k;
 			continue;
 		case BPF_S_LD_MEM:
-			A = mem[fentry->k];
+			A = (memvalid & (1UL << f_k)) ?
+				mem[f_k] : 0;
 			continue;
 		case BPF_S_LDX_MEM:
-			X = mem[fentry->k];
+			X = (memvalid & (1UL << f_k)) ?
+				mem[f_k] : 0;
 			continue;
 		case BPF_S_MISC_TAX:
 			X = A;
@@ -276,14 +280,16 @@  load_b:
 			A = X;
 			continue;
 		case BPF_S_RET_K:
-			return fentry->k;
+			return f_k;
 		case BPF_S_RET_A:
 			return A;
 		case BPF_S_ST:
-			mem[fentry->k] = A;
+			memvalid |= 1UL << f_k;
+			mem[f_k] = A;
 			continue;
 		case BPF_S_STX:
-			mem[fentry->k] = X;
+			memvalid |= 1UL << f_k;
+			mem[f_k] = X;
 			continue;
 		default:
 			WARN_ON(1);