linux-kernel - Re: cBPF socket filters failing

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <CAFSh4UwAmR+sdfbdyxHRDnDr8r+TXxo2bvWtY3gmLAJekWc3Sw@mail.gmail.com>
Date:   Wed, 6 Jan 2021 10:07:08 +0000
From:   Tom Cook <tom.k.cook@...il.com>
To:     linux-kernel@...r.kernel.org
Subject: Re: cBPF socket filters failing - inexplicably?

Just to note I have also reproduced this on a 5.10.0 kernel.

On Tue, Jan 5, 2021 at 1:42 PM Tom Cook <tom.k.cook@...il.com> wrote:
>
> In the course of tracking down a defect in some existing software,
> I've found the failure demonstrated by the short program below.
> Essentially, a cBPF program that just rejects every frame (ie always
> returns zero) and is attached to a socket using setsockopt(SOL_SOCKET,
> SO_ATTACH_FILTER, ...) still occasionally lets frames through to
> userspace.
>
> The code is based on the first example in
> Documentation/networking/filter.txt, except that I've changed the
> content of the filter program and added a timeout on the socket.
>
> To reproduce the problem:
>
> # gcc test.c -o test
> # sudo ./test
> ... and in another console start a large network operation.
>
> In my case, I copied a ~300MB core file I had lying around to another
> host on the LAN.  The test code should print the string "Failed to
> read from socket" 100 times.  In practice, it produces about 10%
> "Received packet with ethertype..." messages.
>
> I've observed the same result on Ubuntu amd64 glibc system running a
> 5.9.0 kernel and also on Alpine arm64v8 muslc system running a 4.9.1
> kernel.  I've written test code in both C and Python.  I'm fairly sure
> this is not something I'm doing wrong - but very keen to have things
> thrown at me if it is.
>
> Regards,
> Tom Cook
>
>
> #include <stdio.h>
> #include <sys/socket.h>
> #include <sys/types.h>
> #include <arpa/inet.h>
> #include <linux/if_ether.h>
> #include <linux/filter.h>
> #include <stdint.h>
> #include <unistd.h>
>
> struct sock_filter code[] = {
>     { 0x06,    0,    0,    0x00 }  /* BPF_RET | BPF_K   0   0   0 */
> };
>
> struct sock_fprog bpf = {
>     .len = 1,
>     .filter = code,
> };
>
> void test() {
>     uint8_t buf[2048];
>
>     int sock = socket(PF_PACKET, SOCK_RAW, htons(ETH_P_ALL));
>     if (sock < 0) {
>         printf("Failed to open socket\n");
>         return;
>     }
>     int ret = setsockopt(sock, SOL_SOCKET, SO_ATTACH_FILTER, &bpf, sizeof(bpf));
>     if (ret < 0) {
>         printf("Failed to set socket filter\n");
>         return;
>     }
>     struct timeval tv = {
>         .tv_sec = 1
>     };
>
>     ret = setsockopt(sock, SOL_SOCKET, SO_RCVTIMEO, &tv, sizeof(tv));
>     if (ret < 0) {
>         printf("Failed to set socket timeout\n");
>         return;
>     }
>
>     ssize_t count = recv(sock, buf, 2048, 0);
>     if (count <= 0) {
>         printf("Failed to read from socket\n");
>         return;
>     }
>
>     close(sock);
>
>     uint16_t *ethertype = (short*)(buf + 12);
>     uint8_t *proto = (unsigned char *)(buf + 23);
>     uint16_t *dport = (uint16_t *)(buf + 14 + 20);
>
>     printf("Received packet with ethertype 0x%04hu, protocol 0x%02hhu
> and dport 0x%04hu\n", *ethertype, *proto, *dport);
> }
>
> int main() {
>     for (size_t ii = 0; ii < 100; ++ii) {
>         test();
>     }
> }