Re: [PATCH bpf-next 4/4] bpf: inet_diag: Dump bpf_sk_storages in inet_diag_dump()

On Feb 25, 2020, at 9:08 AM, Martin Lau [off-list ref] wrote:

On Mon, Feb 24, 2020 at 09:47:33PM -0800, Song Liu wrote:

quoted

On Fri, Feb 21, 2020 at 10:49 AM Martin KaFai Lau [off-list ref] wrote:

quoted

This patch will dump out the bpf_sk_storages of a sk
if the request has the INET_DIAG_REQ_SK_BPF_STORAGES nlattr.

An array of SK_DIAG_BPF_STORAGE_REQ_MAP_FD can be specified in
INET_DIAG_REQ_SK_BPF_STORAGES to select which bpf_sk_storage to dump.
If no map_fd is specified, all bpf_sk_storages of a sk will be dumped.

[...]

quoted

Signed-off-by: Martin KaFai Lau <redacted>

[...]

quoted

quoted

@@ -1022,8 +1069,11 @@ static int __inet_diag_dump(struct sk_buff *skb, struct netlink_callback *cb,
                           const struct inet_diag_req_v2 *r)

{
       const struct inet_diag_handler *handler;
+       u32 prev_min_dump_alloc;
       int err = 0;

+again:
+       prev_min_dump_alloc = cb->min_dump_alloc;
       handler = inet_diag_lock_handler(r->sdiag_protocol);
       if (!IS_ERR(handler))
               handler->dump(skb, cb, r);

@@ -1031,6 +1081,12 @@ static int __inet_diag_dump(struct sk_buff *skb, struct netlink_callback *cb,
               err = PTR_ERR(handler);
       inet_diag_unlock_handler(handler);

+       if (!skb->len && cb->min_dump_alloc > prev_min_dump_alloc) {

Why do we check for !skb->len here?

skb contains the info of sk(s) to be dumped to the userspace.
It may contain no sk info (i.e. !skb->len),  1 sk info, 2 sk info...etc.
It only retries if there is no sk info and the cb->min_dump_alloc becomes
larger (together, it means the current skb is not large enough to fit one
sk info).

I see. Thanks for the explanation. 

Acked-by: Song Liu <redacted>