On Wed, Nov 10, 2021 at 10:56:30PM +0900, Masami Hiramatsu wrote: > On Tue, 9 Nov 2021 11:08:44 -0800 > Beau Belgrave <beaub@xxxxxxxxxxxxxxxxxxx> wrote: > I would like to keep verifying in writer side then we can ensure the > data on ring buffer (of perf and of ftrace) is sane. If you add the unsafe > flag, you have to change all the code which access the ring buffer, not only > the filter but also eprobes, histograms, perf-tools, and other user-space > tracing tools which reads the tracing buffer directly. > > > It sounded like Steven wanted to think about this a bit, so I'll wait a > > bit before poking again for consensus :) > > > > Do you have any strong feelings about where it goes? > > I recommend you to start verifying the writer side, it should make the > change as small as possible. Unsafe flag idea may involve many other > tools. And it is not fundamentary required for user-events. > > Thank you, > > -- > Masami Hiramatsu <mhiramat@xxxxxxxxxx> Ok, I will start there. Are static string buffers required as well for the null check? Or is this only for dyn strings that require the check? Also, I am assuming that __rel_loc offset is based after the __rel_loc payload, IE: Offset 0 of __rel_loc is immediately after the 4 byte __rel_loc description? Thanks, -Beau
|