(seq. #2454) support non bool values in setsockopt

[zyuiop]

Add non boolean values in setsockopt.

Initially, I used a similar structure for getsockopt but I rebased on the latest PRs and removed that to rely on c_int only, hoping we don't implement string sockopt values anytime soon. However, I have for now kept this flexibility in setsockopt. We may want to rewrite that to use c_int instead, for consistency?

This PR has been split from #2454 but relies on its feature flag (not completely critical to be fair)

[github-actions]

Benchmark Results

Details

Benchmark	Current: 555421a	Previous: a025f19	Performance Ratio
startup_benchmark Build Time	112.42 s	116.82 s	0.96 ❗
startup_benchmark File Size	0.77 MB	0.77 MB	1.00 ❗
Startup Time - 1 core	1.00 s (±0.03 s)	1.00 s (±0.03 s)	1.00
Startup Time - 2 cores	0.98 s (±0.04 s)	1.04 s (±0.05 s)	0.95
Startup Time - 4 cores	0.99 s (±0.04 s)	1.02 s (±0.05 s)	0.97
multithreaded_benchmark Build Time	108.38 s	117.48 s	0.92 ❗
multithreaded_benchmark File Size	0.87 MB	0.84 MB	1.04 ❗
Multithreaded Pi Efficiency - 2 Threads	89.82 % (±15.72 %)	85.06 % (±19.28 %)	1.06
Multithreaded Pi Efficiency - 4 Threads	43.93 % (±7.87 %)	42.20 % (±9.32 %)	1.04
Multithreaded Pi Efficiency - 8 Threads	24.06 % (±3.67 %)	23.57 % (±5.00 %)	1.02
micro_benchmarks Build Time	94.30 s	92.09 s	1.02 ❗
micro_benchmarks File Size	0.88 MB	0.84 MB	1.04 ❗
Scheduling time - 1 thread	74.14 ticks (±3.87 ticks)	73.19 ticks (±3.39 ticks)	1.01
Scheduling time - 2 threads	42.82 ticks (±4.89 ticks)	44.10 ticks (±6.54 ticks)	0.97
Micro - Time for syscall (getpid)	3.15 ticks (±0.25 ticks)	3.09 ticks (±0.16 ticks)	1.02
Memcpy speed - (built_in) block size 4096	72858.42 MByte/s (±50495.37 MByte/s)	73877.55 MByte/s (±50900.64 MByte/s)	0.99
Memcpy speed - (built_in) block size 1048576	29312.30 MByte/s (±24119.93 MByte/s)	29621.41 MByte/s (±24412.82 MByte/s)	0.99
Memcpy speed - (built_in) block size 16777216	25958.09 MByte/s (±21340.57 MByte/s)	26615.80 MByte/s (±21973.01 MByte/s)	0.98
Memset speed - (built_in) block size 4096	73458.80 MByte/s (±50852.11 MByte/s)	74082.46 MByte/s (±51039.25 MByte/s)	0.99
Memset speed - (built_in) block size 1048576	30089.96 MByte/s (±24574.74 MByte/s)	30381.08 MByte/s (±24836.90 MByte/s)	0.99
Memset speed - (built_in) block size 16777216	26622.27 MByte/s (±21728.36 MByte/s)	27365.04 MByte/s (±22427.19 MByte/s)	0.97
Memcpy speed - (rust) block size 4096	65721.81 MByte/s (±46150.30 MByte/s)	67998.13 MByte/s (±47775.53 MByte/s)	0.97
Memcpy speed - (rust) block size 1048576	29327.49 MByte/s (±24153.61 MByte/s)	29415.03 MByte/s (±24216.28 MByte/s)	1.00
Memcpy speed - (rust) block size 16777216	26410.63 MByte/s (±21732.02 MByte/s)	26645.15 MByte/s (±21980.78 MByte/s)	0.99
Memset speed - (rust) block size 4096	66613.20 MByte/s (±46784.99 MByte/s)	68410.01 MByte/s (±48051.67 MByte/s)	0.97
Memset speed - (rust) block size 1048576	30099.22 MByte/s (±24596.29 MByte/s)	30182.39 MByte/s (±24662.20 MByte/s)	1.00
Memset speed - (rust) block size 16777216	27137.14 MByte/s (±22174.34 MByte/s)	27320.66 MByte/s (±22364.88 MByte/s)	0.99
alloc_benchmarks Build Time	89.91 s	91.56 s	0.98 ❗
alloc_benchmarks File Size	0.84 MB	0.85 MB	1.00 ❗
Allocations - Allocation success	100.00 %	100.00 %	1
Allocations - Deallocation success	100.00 %	100.00 %	1
Allocations - Pre-fail Allocations	100.00 %	100.00 %	1
Allocations - Average Allocation time	4062.33 Ticks (±66.12 Ticks)	4553.50 Ticks (±57.46 Ticks)	0.89 ❗
Allocations - Average Allocation time (no fail)	4062.33 Ticks (±66.12 Ticks)	4553.50 Ticks (±57.46 Ticks)	0.89 ❗
Allocations - Average Deallocation time	650.27 Ticks (±67.64 Ticks)	1035.15 Ticks (±129.57 Ticks)	0.63 ❗
mutex_benchmark Build Time	90.28 s	88.37 s	1.02 ❗
mutex_benchmark File Size	0.88 MB	0.84 MB	1.04 ❗
Mutex Stress Test Average Time per Iteration - 1 Threads	13.80 ns (±0.75 ns)	14.02 ns (±0.84 ns)	0.98
Mutex Stress Test Average Time per Iteration - 2 Threads	17.52 ns (±9.67 ns)	17.80 ns (±4.95 ns)	0.98

This comment was automatically generated by workflow using github-action-benchmark.

[mkroening]

Thanks, this is great!

However, I have for now kept this flexibility in setsockopt. We may want to rewrite that to use c_int instead, for consistency?

I think properly representing the pointer and the length is great.
Do you think it would make sense to represent it as a byte slice instead? We could leverage zerocopy in that case. Not sure how well it would tie in with strings, though.

Regarding the feature, same question as in the other PR: which software wants this with which values? How feasible would it be to support that "properly"? Returning fakes with finer granularity is fine of course.

[zyuiop]

The three values (NODELAY, SNDBUF, RCVBUF) are used by the socket2 rust library, and I think the particular application here was the Axum server.

Representing as a slice would indeed be cool, also because it reduces the amount of unsafety required. Not sure we could leverage zerocopy as in practice even booleans are encoded as i32 (and I assume zerocopy uses one byte to represent them?)

[mkroening]

The three values (NODELAY, SNDBUF, RCVBUF) are used by the socket2 rust library, and I think the particular application here was the Axum server.

Huh, that's odd. We have an axum server that is tested in CI that does not need this. But that does currently patch socket2 and tokio (https://github.com/hermit-os/hermit-rs/blob/main/Cargo.toml#L35-L37), since I would like to upstream it only after cleaning up the libc a bit.

Anyway, I think adding this support for non-bool values is great, I just would like to avoid the feature and stub values at a finer granularity (for each cmd explicitly, for example).