Dmitrii Kuvaiskii
Dmitrii Kuvaiskii
I analyzed this issue, and it's a very subtle problem with pipes. Here is the relevant snippet in Gramine-SGX log: ``` (libos_parser.c:1658:buf_write_all) [P4:T34:sed] trace: ---- write(1, 0x154054d0, 0x20) ... (libos_parser.c:1658:buf_write_all)...
I don't see any specific problems with this approach. The description by @jonathan-sha seems solid, I appreciate the details! > we don't need to "sleep" for this to work, we...
> > How do you calculate the latency of each of these calls? > > I think you don't need to? You can do rdtsc -> gettimeofday() -> wait some...
Thanks @jonathan-sha! This looks very good. Feel free to submit a PR. (Though depending on the complexity of the PR, it may take a long time to review it.)
Jenkins, test this please
Jenkins, test this please
Jenkins, test this please
I confirm that `fchmod()` and `fchown()` are not correctly emulated in Gramine on `O_PATH`-opened file descriptors. Technically, `O_PATH`-opened FDs do not have an underlying file to manipulate, so syscalls like...
Please show us your manifest.template file
Hm, can you set `loader.log_level = "all"` and show the complete Gramine log?