Skip to contentSkip to conversation

inet_diag insufficient validation?

PATCH In review Dan Rosenberg 4 participants 7 messages Jun 1, 2011 – Jun 17, 2011

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

It seems to me that the auditing performed by inet_diag_bc_audit() is insufficient to prevent pathological INET_DIAG bytecode from doing bad things.

Firstly, it's possible to cause an infinite loop in inet_diag_bc_audit() with a INET_DIAG_BC_JMP opcode with a "yes" value of 0. The valid_cc() function, also called from here, seems suspicious as well.

Once the bytecode is actually run in inet_diag_bc_run(), it looks like more infinite loops are possible, if appropriate "yes" or "no" values are set to zero and weren't validated by the audit.

Finally, I can't seem to find any validation that the reported length of the netlink message header doesn't exceed the skb length, as checked in some other netlink receive functions, which could result in reading beyond the bounds of the socket data. I could just be missing something here though.

Regards,
Dan

#1 lore raw
Quoted 3 lines · Dan Rosenberg

Once the bytecode is actually run in inet_diag_bc_run(), it looks like more infinite loops are possible, if appropriate "yes" or "no" values are set to zero and weren't validated by the audit.

On second glance, there don't appear to be any infinite loops at runtime (but the audit loop seems real). Thanks to Nelson Elhage for setting me straight.

-Dan

#2 lore raw
In reply to Dan Rosenberg
Quoted 6 lines · Dan Rosenberg

Finally, I can't seem to find any validation that the reported length of the netlink message header doesn't exceed the skb length, as checked in some other netlink receive functions, which could result in reading beyond the bounds of the socket data. I could just be missing something here though.

And for the second time, I was missing something - this validation happens in netlink_rcv_skb().

That leaves the infinite loop in bytecode auditing, which I've confirmed via reproducer.

-Dan

#3 lore raw
In reply to Dan Rosenberg

Cc'ed acme.

Quoted 26 lines · Dan Rosenberg

It seems to me that the auditing performed by inet_diag_bc_audit() is insufficient to prevent pathological INET_DIAG bytecode from doing bad things.

Firstly, it's possible to cause an infinite loop in inet_diag_bc_audit() with a INET_DIAG_BC_JMP opcode with a "yes" value of 0.  The valid_cc() function, also called from here, seems suspicious as well.

Once the bytecode is actually run in inet_diag_bc_run(), it looks like more infinite loops are possible, if appropriate "yes" or "no" values are set to zero and weren't validated by the audit.

Finally, I can't seem to find any validation that the reported length of the netlink message header doesn't exceed the skb length, as checked in some other netlink receive functions, which could result in reading beyond the bounds of the socket data.  I could just be missing something here though.

Regards,
Dan

_______________________________________________
Security mailing list
[email protected]
http://linux.kernel.org/mailman/listinfo/security

#4 lore raw
Quoted 11 lines · Eugene Teo

Cc'ed acme.

Quoted 8 lines · Dan Rosenberg

It seems to me that the auditing performed by inet_diag_bc_audit() is insufficient to prevent pathological INET_DIAG bytecode from doing bad things.

Firstly, it's possible to cause an infinite loop in inet_diag_bc_audit() with a INET_DIAG_BC_JMP opcode with a "yes" value of 0. The valid_cc() function, also called from here, seems suspicious as well.

Any chance of getting this fixed? I have a reproducer available if necessary.

-Dan

#5 lore raw

PATCH 1/1 inet_diag: fix inet_diag_bc_audit()

In reply to Eugene Teo
Quoted 24 lines · Eugene Teo

Cc'ed acme.

Quoted 21 lines · Dan Rosenberg

It seems to me that the auditing performed by inet_diag_bc_audit() is insufficient to prevent pathological INET_DIAG bytecode from doing bad things.

Firstly, it's possible to cause an infinite loop in inet_diag_bc_audit() with a INET_DIAG_BC_JMP opcode with a "yes" value of 0. The valid_cc() function, also called from here, seems suspicious as well.

Once the bytecode is actually run in inet_diag_bc_run(), it looks like more infinite loops are possible, if appropriate "yes" or "no" values are set to zero and weren't validated by the audit.

Finally, I can't seem to find any validation that the reported length of the netlink message header doesn't exceed the skb length, as checked in some other netlink receive functions, which could result in reading beyond the bounds of the socket data. I could just be missing something here though.

Regards,
Dan

Thanks guys, here is the patch I cooked to address this problem, sorry for the long delay again.

[PATCH] inet_diag: fix inet_diag_bc_audit()

A malicious user or buggy application can inject code and trigger an infinite loop in inet_diag_bc_audit()

Also make sure each instruction is aligned on 4 bytes boundary, to avoid unaligned accesses.

CC Stephen Hemminger <[email protected]>
CC Arnaldo Carvalho de Melo <[email protected]>
CC Eugene Teo <[email protected]>
CC Alexey Kuznetsov <[email protected]>
---
 net/ipv4/inet_diag.c |   14 ++++++--------
 1 files changed, 6 insertions(+), 8 deletions(-)
net/ipv4/inet_diag.c +6 -8
index 6ffe94c..3267d38 100644
@@ -437,7 +437,7 @@ static int valid_cc(const void *bc, int len, int cc)
 			return 0;
 		if (cc == len)
 			return 1;
-		if (op->yes < 4)
+		if (op->yes < 4 || op->yes & 3)
 			return 0;
 		len -= op->yes;
 		bc  += op->yes;
@@ -447,11 +447,11 @@ static int valid_cc(const void *bc, int len, int cc)
 
 static int inet_diag_bc_audit(const void *bytecode, int bytecode_len)
 {
-	const unsigned char *bc = bytecode;
+	const void *bc = bytecode;
 	int  len = bytecode_len;
 
 	while (len > 0) {
-		struct inet_diag_bc_op *op = (struct inet_diag_bc_op *)bc;
+		const struct inet_diag_bc_op *op = bc;
 
 //printk("BC: %d %d %d {%d} / %d\n", op->code, op->yes, op->no, op[1].no, len);
 		switch (op->code) {
@@ -462,22 +462,20 @@ static int inet_diag_bc_audit(const void *bytecode, int bytecode_len)
 		case INET_DIAG_BC_S_LE:
 		case INET_DIAG_BC_D_GE:
 		case INET_DIAG_BC_D_LE:
-			if (op->yes < 4 || op->yes > len + 4)
-				return -EINVAL;
 		case INET_DIAG_BC_JMP:
-			if (op->no < 4 || op->no > len + 4)
+			if (op->no < 4 || op->no > len + 4 || op->no & 3)
 				return -EINVAL;
 			if (op->no < len &&
 			    !valid_cc(bytecode, bytecode_len, len - op->no))
 				return -EINVAL;
 			break;
 		case INET_DIAG_BC_NOP:
-			if (op->yes < 4 || op->yes > len + 4)
-				return -EINVAL;
 			break;
 		default:
 			return -EINVAL;
 		}
+		if (op->yes < 4 || op->yes > len + 4 || op->yes & 3)
+			return -EINVAL;
 		bc  += op->yes;
 		len -= op->yes;
 	}
maintainer
#6 lore raw

Re: [PATCH] inet_diag: fix inet_diag_bc_audit()

From: Eric Dumazet <[email protected]>
Date: Fri, 17 Jun 2011 22:19:50 +0200

Quoted 10 lines

[PATCH] inet_diag: fix inet_diag_bc_audit()

A malicious user or buggy application can inject code and trigger an infinite loop in inet_diag_bc_audit()

Also make sure each instruction is aligned on 4 bytes boundary, to avoid unaligned accesses.

Applied and queued up for -stable, thanks!