Re: Still a pretty bad time on 5.4.6 with fuse_request_end.

[Miklos Szeredi]

On Wed, Feb 12, 2020 at 8:58 AM Michael Stapelberg
<michael+lkml@xxxxxxxxxxxxx> wrote:

> (gdb) p *req->args
> $5 = {
>   nodeid = 18446683600620026424,
>   opcode = 2167928246,
>   in_numargs = 65535,
>   out_numargs = 65535,
>   force = false,
>   noreply = false,
>   nocreds = false,
>   in_pages = false,
>   out_pages = false,
>   out_argvar = true,
>   page_zeroing = true,
>   page_replace = false,
>   in_args = {{
>       size = 978828800,
>       value = 0x2fafce0
>     }, {
>       size = 978992728,
>       value = 0xffffffff8138efaa <fuse_alloc_forget+26>
>     }, {
>       size = 50002688,
>       value = 0xffffffff8138635f <fuse_lookup_name+255>
>     }},
>   out_args = {{
>       size = 570,
>       value = 0xffffc90002fafb10
>     }, {
>       size = 6876,
>       value = 0x3000000001adc
>     }},
>   end = 0x1000100000001
> }

Okay, that looks like rubbish, the request was possibly freed and overwritten.

> Independently, as a separate test, I have also modified the source like this:
>
> bool async;
> bool async_early = req->args->end;
>
> if (test_and_set_bit(FR_FINISHED, &req->flags))
> goto put_request;
>
> async = req->args->end;
>
> âand printed the value of async and async_early. async is true,
> async_early is false.

Can you save and print out the value of req->opcode before the
test_and_set_bit()?

Thanks,
Miklos