tipc_crypto_rcv() invokes tipc_aead_get(), which returns a reference of
the tipc_aead object to "aead" with increased refcnt.
When tipc_crypto_rcv() returns, the original local reference of "aead"
becomes invalid, so the refcount should be decreased to keep refcount
balanced.
The issue happens in one error path of tipc_crypto_rcv(). When TIPC
message decryption status is EINPROGRESS or EBUSY, the function forgets
to decrease the refcnt increased by tipc_aead_get() and causes a refcnt
leak.
Fix this issue by calling tipc_aead_put() on the error path when TIPC
message decryption status is EINPROGRESS or EBUSY.
Signed-off-by: Xiyu Yang <[email protected]>
Signed-off-by: Xin Tan <[email protected]>
---
net/tipc/crypto.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/net/tipc/crypto.c b/net/tipc/crypto.c
index c8c47fc72653..8c47ded2edb6 100644
--- a/net/tipc/crypto.c
+++ b/net/tipc/crypto.c
@@ -1712,6 +1712,7 @@ int tipc_crypto_rcv(struct net *net, struct tipc_crypto *rx,
case -EBUSY:
this_cpu_inc(stats->stat[STAT_ASYNC]);
*skb = NULL;
+ tipc_aead_put(aead);
return rc;
default:
this_cpu_inc(stats->stat[STAT_NOK]);
--
2.7.4
From: Xiyu Yang <[email protected]>
Date: Wed, 15 Apr 2020 16:39:56 +0800
> tipc_crypto_rcv() invokes tipc_aead_get(), which returns a reference of
> the tipc_aead object to "aead" with increased refcnt.
>
> When tipc_crypto_rcv() returns, the original local reference of "aead"
> becomes invalid, so the refcount should be decreased to keep refcount
> balanced.
>
> The issue happens in one error path of tipc_crypto_rcv(). When TIPC
> message decryption status is EINPROGRESS or EBUSY, the function forgets
> to decrease the refcnt increased by tipc_aead_get() and causes a refcnt
> leak.
>
> Fix this issue by calling tipc_aead_put() on the error path when TIPC
> message decryption status is EINPROGRESS or EBUSY.
>
> Signed-off-by: Xiyu Yang <[email protected]>
> Signed-off-by: Xin Tan <[email protected]>
Applied and queued up for -stable.
This code is harder to audit than it needs to be due to the special
casing of things like -ENOKEY etc. It should rather explicitly handle
the NULL test on aead in this top-level piece of code, which would
make the validation of aead reference counting much more explicit and
clear.