2018-10-06 16:09:19

by Wenwen Wang

[permalink] [raw]
Subject: [PATCH] s390/qeth: fix a missing-check bug

In qeth_snmp_command(), the length of the user request is firstly copied
from the user-space buffer 'udata' to the kernel variable 'req_len' and
checked to see whether it is too large. If the check fails, an error code
EINVAL is returned. Otherwise, the execution continues and the whole buffer
is copied again from 'udata' and saved to the kernel buffer 'ureq'.
However, after the second copy, no re-check is enforced on the newly-copied
request length. Given that the buffer 'udata' is in the user space, a
malicious user can race to change the request length between the two
copies. In this way, the attacker can supply malicious data to the kernel
and cause undefined behavior.

This patch adds a re-check on the request length after the second copy from
the buffer 'udata'. If the newly-copied value is different from the value
obtained in the first copy, i.e., 'req_len', an error code EINVAL will be
returned after the buffer 'ureq' is freed.

Signed-off-by: Wenwen Wang <[email protected]>
---
drivers/s390/net/qeth_core_main.c | 4 ++++
1 file changed, 4 insertions(+)

diff --git a/drivers/s390/net/qeth_core_main.c b/drivers/s390/net/qeth_core_main.c
index de82824..6199743 100644
--- a/drivers/s390/net/qeth_core_main.c
+++ b/drivers/s390/net/qeth_core_main.c
@@ -4613,6 +4613,10 @@ static int qeth_snmp_command(struct qeth_card *card, char __user *udata)
QETH_CARD_TEXT(card, 2, "snmpnome");
return PTR_ERR(ureq);
}
+ if (ureq->hdr.req_len != req_len) {
+ kfree(ureq);
+ return -EINVAL;
+ }
qinfo.udata_len = ureq->hdr.data_len;
qinfo.udata = kzalloc(qinfo.udata_len, GFP_KERNEL);
if (!qinfo.udata) {
--
2.7.4



2018-10-08 05:02:38

by Heiko Carstens

[permalink] [raw]
Subject: Re: [PATCH] s390/qeth: fix a missing-check bug

On Sat, Oct 06, 2018 at 11:08:23AM -0500, Wenwen Wang wrote:
> In qeth_snmp_command(), the length of the user request is firstly copied
> from the user-space buffer 'udata' to the kernel variable 'req_len' and
> checked to see whether it is too large. If the check fails, an error code
> EINVAL is returned. Otherwise, the execution continues and the whole buffer
> is copied again from 'udata' and saved to the kernel buffer 'ureq'.
> However, after the second copy, no re-check is enforced on the newly-copied
> request length. Given that the buffer 'udata' is in the user space, a
> malicious user can race to change the request length between the two
> copies. In this way, the attacker can supply malicious data to the kernel
> and cause undefined behavior.
>
> This patch adds a re-check on the request length after the second copy from
> the buffer 'udata'. If the newly-copied value is different from the value
> obtained in the first copy, i.e., 'req_len', an error code EINVAL will be
> returned after the buffer 'ureq' is freed.
>
> Signed-off-by: Wenwen Wang <[email protected]>
> ---
> drivers/s390/net/qeth_core_main.c | 4 ++++
> 1 file changed, 4 insertions(+)
>
> diff --git a/drivers/s390/net/qeth_core_main.c b/drivers/s390/net/qeth_core_main.c
> index de82824..6199743 100644
> --- a/drivers/s390/net/qeth_core_main.c
> +++ b/drivers/s390/net/qeth_core_main.c
> @@ -4613,6 +4613,10 @@ static int qeth_snmp_command(struct qeth_card *card, char __user *udata)
> QETH_CARD_TEXT(card, 2, "snmpnome");
> return PTR_ERR(ureq);
> }
> + if (ureq->hdr.req_len != req_len) {
> + kfree(ureq);
> + return -EINVAL;
> + }

ureq->hdr.req_len is not used anywhere in the code, so could you please
explain what the undefined behavior is?

You could argue that adding a second sanity check may help to avoid
potential future bugs, but currently the code looks sane to me.