On 1/3/24 4:04 AM, Maciej Fijalkowski wrote:
On Tue, Jan 02, 2024 at 02:58:00PM -0800, Martin KaFai Lau wrote:
On 12/21/23 5:26 AM, Maciej Fijalkowski wrote:
This comes from __xdp_return() call with xdp_buff argument passed as
NULL which is supposed to be consumed by xsk_buff_free() call.
To address this properly, in ZC case, a node that represents the frag
being removed has to be pulled out of xskb_list. Introduce
appriopriate xsk helpers to do such node operation and use them
accordingly within bpf_xdp_adjust_tail().
[ ... ]
+static inline struct xdp_buff *xsk_buff_get_tail(struct xdp_buff *first)
+{
+ struct xdp_buff_xsk *xskb = container_of(first, struct xdp_buff_xsk, xdp);
+ struct xdp_buff_xsk *frag;
+
+ frag = list_last_entry(&xskb->pool->xskb_list, struct xdp_buff_xsk,
+ xskb_list_node);
+ return &frag->xdp;
+}
+
[ ... ]
+static void __shrink_data(struct xdp_buff *xdp, struct xdp_mem_info *mem_info,
+ skb_frag_t *frag, int shrink)
+{
+ if (mem_info->type == MEM_TYPE_XSK_BUFF_POOL) {
+ struct xdp_buff *tail = xsk_buff_get_tail(xdp);
+
+ if (tail)
+ tail->data_end -= shrink;
+ }
+ skb_frag_size_sub(frag, shrink);
+}
+
+static bool shrink_data(struct xdp_buff *xdp, skb_frag_t *frag, int shrink)
+{
+ struct xdp_mem_info *mem_info = &xdp->rxq->mem;
+
+ if (skb_frag_size(frag) == shrink) {
+ struct page *page = skb_frag_page(frag);
+ struct xdp_buff *zc_frag = NULL;
+
+ if (mem_info->type == MEM_TYPE_XSK_BUFF_POOL) {
+ zc_frag = xsk_buff_get_tail(xdp);
+
+ if (zc_frag) {
Based on the xsk_buff_get_tail(), would zc_frag ever be NULL?
Hey Martin thanks for taking a look, I had to do this in order to satisfy
!CONFIG_XDP_SOCKETS builds :/
There is compilation/checker warning if it does not check for NULL?
hmm... but it still should not reach here in the runtime and call
xsk_buff_get_tail() in the !CONFIG_XDP_SOCKETS build. Can the NULL test on the
get_tail() return value be removed? The above "mem_info->type ==
MEM_TYPE_XSK_BUFF_POOL" should have avoided the get_tail() call for the
!CONFIG_XDP_SOCKETS build. Otherwise, it could be passing NULL to the
__xdp_return() and hit the same bug again. The NULL check here is pretty hard to
reason logically.
+ xdp_buff_clear_frags_flag(zc_frag);
+ xsk_buff_del_tail(zc_frag);
+ }
+ }
+
+ __xdp_return(page_address(page), mem_info, false, zc_frag);
and iiuc, this patch is fixing a bug when zc_frag is NULL and
MEM_TYPE_XSK_BUFF_POOL.
Generally I don't see the need for xdp_return_buff() (which calls in the
end __xdp_return() being discussed) to handle MEM_TYPE_XSK_BUFF_POOL, this
could be refactored later and then probably this fix would look different,
but this is out of the scope now.
+ return true;
+ }
+ __shrink_data(xdp, mem_info, frag, shrink);
+ return false;
+}
+
