On 11/22/23 6:28 AM, Yonghong Song wrote:
On 11/21/23 7:50 PM, Edward Adam Davis wrote:Confirm that skb->len is not 0 to ensure that skb length is valid. Fixes: 114039b34201 ("bpf: Move skb->len == 0 checks into __bpf_redirect") Reported-by: syzbot+e2c932aec5c8a6e1d31c@xxxxxxxxxxxxxxxxxxxxxxxxx Signed-off-by: Edward Adam Davis <eadavis@xxxxxx>Stan, Could you take a look at this patch?
I think this only papers over the bug.. also BPF selftests seem to break with this change. Looking again at the syzkaller trace : [...] Please remove unsupported %\0 in format string WARNING: CPU: 0 PID: 5068 at lib/vsprintf.c:2675 format_decode+0xa03/0xba0 lib/vsprintf.c:2675 [...] We need to fix bpf_bprintf_prepare() instead to reject invalid fmts such as %0 and similar.
net/bpf/test_run.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/net/bpf/test_run.c b/net/bpf/test_run.c index c9fdcc5cdce1..78258a822a5c 100644 --- a/net/bpf/test_run.c +++ b/net/bpf/test_run.c @@ -845,6 +845,9 @@ static int convert___skb_to_skb(struct sk_buff *skb, struct __sk_buff *__skb) { struct qdisc_skb_cb *cb = (struct qdisc_skb_cb *)skb->cb; + if (!skb->len) + return -EINVAL; + if (!__skb) return 0;
