diff mbox series

[bpf-next] xsk: fix possible crash in socket_release when out-of-memory

Message ID 1601112373-10595-1-git-send-email-magnus.karlsson@gmail.com
State Accepted
Delegated to: BPF Maintainers
Headers show
Series [bpf-next] xsk: fix possible crash in socket_release when out-of-memory | expand

Commit Message

Magnus Karlsson Sept. 26, 2020, 9:26 a.m. UTC 
From: Magnus Karlsson <magnus.karlsson@intel.com>

Fix possible crash in socket_release when an out-of-memory error has
occurred in the bind call. If a socket using the XDP_SHARED_UMEM flag
encountered an error in xp_create_and_assign_umem, the bind code
jumped to the exit routine but erroneously forgot to set the err value
before jumping. This meant that the exit routine thought the setup
went well and set the state of the socket to XSK_BOUND. The xsk socket
release code will then, at application exit, think that this is a
properly setup socket, when it is not, leading to a crash when all
fields in the socket have in fact not been initialized properly. Fix
this by setting the err variable in xsk_bind so that the socket is not
set to XSK_BOUND which leads to the clean-up in xsk_release not being
triggered.

Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
Reported-by: syzbot+ddc7b4944bc61da19b81@syzkaller.appspotmail.com
Fixes: 1c1efc2af158 ("xsk: Create and free buffer pool independently from umem")
---
I have not been able to reproduce this issue using the syzkaller
config and reproducer, so I cannot guarantee it fixes it. But this bug
is real and it is triggered by an out-of-memory in
xp_create_and_assign_umem, just like syzcaller injects, and would lead
to the same crash in dev_hold in xsk_release.
---
 net/xdp/xsk.c | 1 +
 1 file changed, 1 insertion(+)

Comments

Daniel Borkmann Sept. 28, 2020, 7:25 p.m. UTC | #1 
On 9/26/20 11:26 AM, Magnus Karlsson wrote:
> From: Magnus Karlsson <magnus.karlsson@intel.com>
> 
> Fix possible crash in socket_release when an out-of-memory error has
> occurred in the bind call. If a socket using the XDP_SHARED_UMEM flag
> encountered an error in xp_create_and_assign_umem, the bind code
> jumped to the exit routine but erroneously forgot to set the err value
> before jumping. This meant that the exit routine thought the setup
> went well and set the state of the socket to XSK_BOUND. The xsk socket
> release code will then, at application exit, think that this is a
> properly setup socket, when it is not, leading to a crash when all
> fields in the socket have in fact not been initialized properly. Fix
> this by setting the err variable in xsk_bind so that the socket is not
> set to XSK_BOUND which leads to the clean-up in xsk_release not being
> triggered.
> 
> Signed-off-by: Magnus Karlsson <magnus.karlsson@intel.com>
> Reported-by: syzbot+ddc7b4944bc61da19b81@syzkaller.appspotmail.com
> Fixes: 1c1efc2af158 ("xsk: Create and free buffer pool independently from umem")

Looks good either way, applied, thanks!

> I have not been able to reproduce this issue using the syzkaller
> config and reproducer, so I cannot guarantee it fixes it. But this bug
> is real and it is triggered by an out-of-memory in
> xp_create_and_assign_umem, just like syzcaller injects, and would lead
> to the same crash in dev_hold in xsk_release.

You can just asked syzbot (which I just did on the original report) via:

#syz test: git://git.kernel.org/pub/scm/linux/kernel/git/bpf/bpf-next.git master

Thanks,
Daniel
diff mbox series

Patch

diff --git a/net/xdp/xsk.c b/net/xdp/xsk.c
index 3895697..ba4dfb1 100644
--- a/net/xdp/xsk.c
+++ b/net/xdp/xsk.c
@@ -703,6 +703,7 @@  static int xsk_bind(struct socket *sock, struct sockaddr *addr, int addr_len)
 			xs->pool = xp_create_and_assign_umem(xs,
 							     umem_xs->umem);
 			if (!xs->pool) {
+				err = -ENOMEM;
 				sockfd_put(sock);
 				goto out_unlock;
 			}