Syzbot was able to trigger use of uninitialized memory in
af_alg_free_resources.
Bug is caused by missing initialization of rsgl->sgl.need_unpin before
adding to rsgl_list. Then in case of extract_iter_to_sg() failure, rsgl
is left with uninitialized need_unpin which is read during clean up
BUG: KMSAN: uninit-value in af_alg_free_sg crypto/af_alg.c:545 [inline]
BUG: KMSAN: uninit-value in af_alg_free_areq_sgls crypto/af_alg.c:778 [inline]
BUG: KMSAN: uninit-value in af_alg_free_resources+0x3d1/0xf60 crypto/af_alg.c:1117
af_alg_free_sg crypto/af_alg.c:545 [inline]
af_alg_free_areq_sgls crypto/af_alg.c:778 [inline]
af_alg_free_resources+0x3d1/0xf60 crypto/af_alg.c:1117
_skcipher_recvmsg crypto/algif_skcipher.c:144 [inline]
...
Uninit was created at:
slab_post_alloc_hook+0x12f/0xb70 mm/slab.h:767
slab_alloc_node mm/slub.c:3470 [inline]
__kmem_cache_alloc_node+0x536/0x8d0 mm/slub.c:3509
__do_kmalloc_node mm/slab_common.c:984 [inline]
__kmalloc+0x121/0x3c0 mm/slab_common.c:998
kmalloc include/linux/slab.h:586 [inline]
sock_kmalloc+0x128/0x1c0 net/core/sock.c:2683
af_alg_alloc_areq+0x41/0x2a0 crypto/af_alg.c:1188
_skcipher_recvmsg crypto/algif_skcipher.c:71 [inline]
Fixes: c1abe6f570af ("crypto: af_alg: Use extract_iter_to_sg() to create scatterlists")
Reported-and-tested-by: [email protected]
Closes: https://syzkaller.appspot.com/bug?extid=cba21d50095623218389
Signed-off-by: Pavel Skripkin <[email protected]>
---
crypto/af_alg.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/crypto/af_alg.c b/crypto/af_alg.c
index 06b15b9f661c..4e4f2d837481 100644
--- a/crypto/af_alg.c
+++ b/crypto/af_alg.c
@@ -1241,6 +1241,8 @@ int af_alg_get_rsgl(struct sock *sk, struct msghdr *msg, int flags,
return -ENOMEM;
}
+ rsgl->sgl.need_unpin = 0;
+
rsgl->sgl.sgt.sgl = rsgl->sgl.sgl;
rsgl->sgl.sgt.nents = 0;
rsgl->sgl.sgt.orig_nents = 0;
--
2.40.1
Pavel Skripkin <[email protected]> wrote:
> Syzbot was able to trigger use of uninitialized memory in
> af_alg_free_resources.
>
> Bug is caused by missing initialization of rsgl->sgl.need_unpin before
> adding to rsgl_list. Then in case of extract_iter_to_sg() failure, rsgl
> is left with uninitialized need_unpin which is read during clean up
Looks feasible :-).
> + rsgl->sgl.need_unpin = 0;
> +
The blank line isn't really necessary and it's a bool, so can you use 'false'
rather than '0'?
Alternatively, it might be better to move:
rsgl->sgl.need_unpin =
iov_iter_extract_will_pin(&msg->msg_iter);
up instead.
David