This is a necessary follow-up on commit 00b144bc9d093 ("obj/ct_timeout:
Avoid array overrun in timeout_parse_attr_data()") which fixed array out
of bounds access but missed the logic behind it:
The nested attribute type values are incremented by one when being
transferred between kernel and userspace, the zero type value is
reserved for "unspecified".
Kernel uses CTA_TIMEOUT_* symbols for that, libnftnl simply mangles the
type values in nftnl_obj_ct_timeout_build().
Return path was broken as it overstepped its nlattr array but apart from
that worked: Type values were decremented by one in
timeout_parse_attr_data().
This patch moves the type value mangling into
parse_timeout_attr_policy_cb() (which still overstepped nlattr array).
Consequently, when copying values from nlattr array into ct timeout
object in timeout_parse_attr_data(), loop is adjusted to start at index
0 and the type value decrement is dropped there.
Fixes: 0adceeab1597a ("src: add ct timeout support")
Signed-off-by: Phil Sutter <[email protected]>
---
src/obj/ct_timeout.c | 8 ++++----
1 file changed, 4 insertions(+), 4 deletions(-)
diff --git a/src/obj/ct_timeout.c b/src/obj/ct_timeout.c
index a09e25ae5d44f..2662cac69438d 100644
--- a/src/obj/ct_timeout.c
+++ b/src/obj/ct_timeout.c
@@ -108,10 +108,10 @@ parse_timeout_attr_policy_cb(const struct nlattr *attr,
void *data)
if (mnl_attr_type_valid(attr, data_cb->nlattr_max) < 0)
return MNL_CB_OK;
- if (type <= data_cb->nlattr_max) {
+ if (type > 0 && type <= data_cb->nlattr_max) {
if (mnl_attr_validate(attr, MNL_TYPE_U32) < 0)
abi_breakage();
- tb[type] = attr;
+ tb[type - 1] = attr;
}
return MNL_CB_OK;
}
@@ -134,9 +134,9 @@ timeout_parse_attr_data(struct nftnl_obj *e,
if (mnl_attr_parse_nested(nest, parse_timeout_attr_policy_cb, &cnt) < 0)
return -1;
- for (i = 1; i < array_size(tb); i++) {
+ for (i = 0; i < array_size(tb); i++) {
if (tb[i]) {
- nftnl_timeout_policy_attr_set_u32(e, i-1,
+ nftnl_timeout_policy_attr_set_u32(e, i,
ntohl(mnl_attr_get_u32(tb[i])));
}
}
--
2.23.0