The structure "rte_flow_item_geneve_opt" is not a protocol header of
geneve tunnel option from rfc8926. The field "data" is a pointer
which points to the actual variable-length option data. So the
structure is not packed.
There is 4 bytes hole before the pointer in a 64-bit system. The
option header is just 4 bytes. When using offsetof() to get the
fixed part's size of option header, the wrong value 8 was got. When
constructing the encap header, a wrong size and offset was used due
to this hole.
With this commit, the fixed part's size is calculated explicitly
based on all fields.
Fixes: 55c074f3ba1d ("app/testpmd: support GENEVE option item")
Cc: viachesl...@nvidia.com
Cc: sta...@dpdk.org
Signed-off-by: Bing Zhao <bi...@nvidia.com>
Reviewed-by: Viacheslav Ovsiienko <viachesl...@nvidia.com>
---
app/test-pmd/cmdline_flow.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/app/test-pmd/cmdline_flow.c b/app/test-pmd/cmdline_flow.c
index c0644d678c..60424e26aa 100644
--- a/app/test-pmd/cmdline_flow.c
+++ b/app/test-pmd/cmdline_flow.c
@@ -9426,7 +9426,8 @@ cmd_set_raw_parsed(const struct buffer *in)
case RTE_FLOW_ITEM_TYPE_GENEVE_OPT:
opt = (const struct rte_flow_item_geneve_opt *)
item->spec;
- size = offsetof(struct rte_flow_item_geneve_opt, data);
+ size = offsetof(struct rte_flow_item_geneve_opt,
+ option_len) + sizeof(uint8_t);
if (opt->option_len && opt->data) {
*total_size += opt->option_len *
sizeof(uint32_t);
--
2.27.0
