On Fri, Jun 23, 2023 at 9:24 PM Tom Talpey <tom@xxxxxxxxxx> wrote: > > On 6/23/2023 12:21 AM, Shyam Prasad N wrote: > > On Mon, Jun 12, 2023 at 8:59 PM Enzo Matsumiya <ematsumiya@xxxxxxx> wrote: > >> > >> On 06/12, Paulo Alcantara wrote: > >>> Shyam Prasad N <nspmangalore@xxxxxxxxx> writes: > >>> > >>>> I had to use kernel_getsockname to get socket source details, not > >>>> kernel_getpeername. > >>> > >>> Why can't you use @server->srcaddr directly? > >> > >> That's only filled if explicitly passed through srcaddr= mount option > >> (to bind it later in bind_socket()). > >> > >> Otherwise, it stays zeroed through @server's lifetime. > > > > Yes. As Enzo mentioned, srcaddr is only useful if the user gave that > > mount option. > > > > Also, here's an updated version of the patch. > > kernel_getsockname seems to be a blocking function, and we need to > > drop the spinlock before calling it. > > Why does this not do anything to report RDMA endpoint addresses/ports? > Many RDMA protocols employ IP addressing. > > If it's intended to not report such information, there should be some > string emitted to make it clear that this is TCP-specific. But let's > not be lazy here, the smbd_connection stores the rdma_cm_id which > holds similar information (the "rdma_addr"). > > Tom. Hi Tom, As always, thanks for reviewing. My understanding of RDMA is very limited. That's the only reason why my patch did not contain changes for RDMA. Let me dig around to understand what needs to be done here. -- Regards, Shyam
