On Fri, Feb 21, 2020 at 10:47 AM Martin KaFai Lau <kafai@xxxxxx> wrote: > > In a latter patch, there is a need to update "cb->min_dump_alloc" > in inet_sk_diag_fill() as it learns the diffierent bpf_sk_storages > stored in a sk while dumping all sk(s) (e.g. tcp_hashinfo). > > The inet_sk_diag_fill() currently does not take the "cb" as an argument. > One of the reason is inet_sk_diag_fill() is used by both dump_one() > and dump() (which belong to the "struct inet_diag_handler". The dump_one() > interface does not pass the "cb" along. > > This patch is to make dump_one() pass a "cb". The "cb" is created in > inet_diag_cmd_exact(). The "nlh" and "in_skb" are stored in "cb" as > the dump() interface does. The total number of args in > inet_sk_diag_fill() is also cut from 10 to 7 and > that helps many callers to pass fewer args. > > In particular, > "struct user_namespace *user_ns", "u32 pid", and "u32 seq" > can be replaced by accessing "cb->nlh" and "cb->skb". > > A similar argument reduction is also made to > inet_twsk_diag_fill() and inet_req_diag_fill(). > > inet_csk_diag_dump() and inet_csk_diag_fill() are also removed. > They are mostly equivalent to inet_sk_diag_fill(). Their repeated > usages are very limited. Thus, inet_sk_diag_fill() is directly used > in those occasions. > > Signed-off-by: Martin KaFai Lau <kafai@xxxxxx> Acked-by: Song Liu <songliubraving@xxxxxx>
