On Tue, 27 May 2025 10:23:37 -0400 Xin Long <lucien.xin@xxxxxxxxx> wrote: > On Mon, May 26, 2025 at 3:38 PM Benjamin Poirier > <benjamin.poirier@xxxxxxxxx> wrote: > > > > On 2025-05-26 14:25 -0400, Xin Long wrote: > > > On Mon, May 26, 2025 at 1:47 AM Christoph Hellwig <hch@xxxxxx> wrote: > > > > > > > > sctp_do_peeloff is only used inside of net/sctp/socket.c, > > > > so mark it static. ... > > I don't see a problem with marking sctp_do_peeloff() static again. > > > > > While there’s no known in-tree usage beyond SCTP itself, we can’t be > > > sure whether this function has been used by out-of-tree kernel modules. > > > > The mainline kernel does not need to cater to out-of-tree users. > Thank you for chiming in. > > I didn't know it was exported for the in-tree kernel dlm, and this > patch should be applied to net-next. The most likely module use would be bpf or io_uring. But they'd probably end up using the sockopt interface (the same as applications). Mind you 'peeloff' is all a strange idea that seems (to me) solving a problem that has nothing at all to do with sctp (as a protocol). The entire 'many-to-one' seems to be there to avoid the overhead of a lot of sockets when the data data is low. I'm sure epoll() solves the actual problem. David
