Skip to contentSkip to conversation

Prevent reading uninitialized memory with socket filters

PATCH Superseded Dan Rosenberg 6 participants 14 messages Nov 9, 2010 – Nov 10, 2010

Subsystems (4)

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 Prevent reading uninitialized memory with socket filters

The "mem" array used as scratch space for socket filters is not initialized, allowing unprivileged users to leak kernel stack bytes.

---
 net/core/filter.c               |    2 ++
 1 file changed, 2 insertions(+), 0 deletions(-)
net/core/filter.c +2 -0
index 7beaec3..2749ba0 100644
@@ -121,6 +121,8 @@ unsigned int sk_run_filter(struct sk_buff *skb, struct sock_filter *filter, int
 	int k;
 	int pc;
 
+	memset(mem, 0, sizeof(mem));
+
 	/*
 	 * Process array of filter instructions.
 	 */
#1 lore raw
Quoted 2 lines · Dan Rosenberg

The "mem" array used as scratch space for socket filters is not initialized, allowing unprivileged users to leak kernel stack bytes.

Hi Dan.

Using

	type var[count] = {};
instead of
	type var[count];
	...
	memset(var, 0, sizeof(var));

at least for gcc 4.4 and 4.5 generally results in smaller code.

$ size net/core/filter.o*
   text	   data	    bss	    dec	    hex	filename
   6751	     56	   1736	   8543	   215f	net/core/filter.o.memset
   6749	     56	   1736	   8541	   215d	net/core/filter.o.init
#3 lore raw
Quoted 9 lines · David Miller

From: Dan Rosenberg <[email protected]>
Date: Tue, 09 Nov 2010 17:28:44 -0500

Quoted 4 lines

The "mem" array used as scratch space for socket filters is not initialized, allowing unprivileged users to leak kernel stack bytes.

Prove it.

And once done, add the checks in sk_chk_filter() ?

Allow a load of mem[X] only if a prior store of mem[X] is proven.

#4 lore raw
Quoted 7 lines · Eric Dumazet
Quoted 6 lines · David Miller

From: Dan Rosenberg <[email protected]>
Date: Tue, 09 Nov 2010 17:28:44 -0500

Quoted 3 lines

The "mem" array used as scratch space for socket filters is not initialized, allowing unprivileged users to leak kernel stack bytes.

Quoted 9 lines
Quoted 3 lines
Quoted 1 line

Prove it.

And once done, add the checks in sk_chk_filter() ?

Allow a load of mem[X] only if a prior store of mem[X] is proven.

This seems complex, and might fail on some valid filters.

What about the following patch then ?

[PATCH] 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.

---
 net/core/filter.c |   10 ++++++++--
 1 file changed, 8 insertions(+), 2 deletions(-)
net/core/filter.c +8 -2
index 7beaec3..4d84dc2 100644
@@ -117,10 +117,12 @@ unsigned int sk_run_filter(struct sk_buff *skb, struct sock_filter *filter, int
 	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.
 	 */
@@ -264,10 +266,12 @@ load_b:
 			X = fentry->k;
 			continue;
 		case BPF_S_LD_MEM:
-			A = mem[fentry->k];
+			A = (memvalid & (1UL << fentry->k)) ?
+				mem[fentry->k] : 0;
 			continue;
 		case BPF_S_LDX_MEM:
-			X = mem[fentry->k];
+			X = (memvalid & (1UL << fentry->k)) ?
+				mem[fentry->k] : 0;
 			continue;
 		case BPF_S_MISC_TAX:
 			X = A;
@@ -280,9 +284,11 @@ load_b:
 		case BPF_S_RET_A:
 			return A;
 		case BPF_S_ST:
+			memvalid |= 1UL << fentry->k;
 			mem[fentry->k] = A;
 			continue;
 		case BPF_S_STX:
+			memvalid |= 1UL << fentry->k;
 			mem[fentry->k] = X;
 			continue;
 		default:
#12 lore raw

For reasons I don't understand, this code seems to cause some people's machines to lock up:

BUG: soft lockup - CPU#0 stuck for 10s! [dz:11844]

I haven't been able to reproduce this myself. Very strange. I'll send a stack trace if I can reproduce.

-Dan

Quoted 6 lines · Dan Rosenberg
Quoted 2 lines

Prove it.

I hope this was a joke. In either case, my reply is a joke:

http://lists.grok.org.uk/pipermail/full-disclosure/2010-November/077321.html

#5 lore raw

Re: [PATCH] Prevent reading uninitialized memory with socketfilters

In reply to Eric Dumazet

Just I thought...

Quoted 3 lines

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 *"?

Quoted 3 lines

(...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?

maintainer
#13 lore raw
In reply to Dan Rosenberg

From: Dan Rosenberg <[email protected]>
Date: Wed, 10 Nov 2010 06:12:47 -0500

Quoted 5 lines
Quoted 3 lines

Prove it.

I hope this was a joke.

It absolutely is not.

You are very much not the first person ever to try and add an expensive memset() here.

So the onus is really on you to prove this assertion and show the exact code path by which the user can actually see any uninitialized kernel stack memory (he can't, he can peek at certain values in a certain extremely contrived range, making the leak useless), rather than point us at some web external site archive of a list posting which we cannot easily quote and reply to here.

I think you cannot do it, really. Except in the AF_PACKET case, the sockets can only see "0" or a negative error code, not the actual sk_run_filter() return value.

In the one exception, AF_PACKET, the range of values the user can see are in the range of MTU of the device being accessed, which realistically is 1500 bytes. This means the user cannot see any kernel stack value outside of the range 0 to 1500, which isn't worth using this expensive memset to guard against at all.

I don't even think it's worth adding all of the extra cpu cycles incurred by Eric Dumazet's scheme of using a bitmap test on every single memory buffer access.

maintainer
#6 lore raw

Re: [PATCH] Prevent reading uninitialized memory with socketfilters

In reply to Tetsuo Handa

From: Tetsuo Handa <[email protected]>
Date: Wed, 10 Nov 2010 23:25:08 +0900

Quoted 5 lines

Just I thought...

Quoted 3 lines

unsigned int sk_run_filter(struct sk_buff *skb, struct sock_filter *filter, int flen) {
struct sock_filter *fentry; /* We walk down these */

Quoted 7 lines

Can't this be "const struct sock_filter *"?

Quoted 3 lines

(...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?

Yes, this feedback seems reasonable, I'll make these changes when I apply Eric's patch, thanks!

maintainer
#7 lore raw

Re: [PATCH] Prevent reading uninitialized memory with socketfilters

In reply to Tetsuo Handa

From: Tetsuo Handa <[email protected]>
Date: Wed, 10 Nov 2010 23:25:08 +0900

Quoted 5 lines

Just I thought...

Quoted 3 lines

unsigned int sk_run_filter(struct sk_buff *skb, struct sock_filter *filter, int flen) {
struct sock_filter *fentry; /* We walk down these */

Quoted 7 lines

Can't this be "const struct sock_filter *"?

Quoted 3 lines

(...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 ]

---
 net/core/filter.c |   64 +++++++++++++++++++++++++++++------------------------
 1 files changed, 35 insertions(+), 29 deletions(-)
net/core/filter.c +35 -29
index 7beaec3..23e9b2a 100644
@@ -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);
Signature · 1 line
1.7.3.2
#8 lore raw

Re: [PATCH] Prevent reading uninitialized memory with socketfilters

On Wed, 2010-11-10 at 10:39 -0800, David Miller wrote:
[...]

Quoted 7 lines

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.

Signature · 3 lines
Ben Hutchings, Senior Software Engineer, Solarflare Communications
Not speaking for my employer; that's the marketing department's job.
They asked us to note that Solarflare product names are trademarked.
maintainer
#9 lore raw

Re: [PATCH] Prevent reading uninitialized memory with socketfilters

From: Ben Hutchings <[email protected]>
Date: Wed, 10 Nov 2010 20:57:44 +0000

Quoted 5 lines · David Miller

[...]

Quoted 3 lines

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.

Quoted 8 lines
Quoted 4 lines

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.

#10 lore raw

Re: [PATCH] Prevent reading uninitialized memory with socketfilters

Quoted 10 lines · David Miller

From: Ben Hutchings <[email protected]>
Date: Wed, 10 Nov 2010 20:57:44 +0000

Quoted 7 lines · David Miller

[...]

Quoted 5 lines

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.

Quoted 8 lines
Quoted 6 lines
Quoted 2 lines

[ 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.

Signature · 3 lines
Ben Hutchings, Senior Software Engineer, Solarflare Communications
Not speaking for my employer; that's the marketing department's job.
They asked us to note that Solarflare product names are trademarked.