Use fork and exec for the command

[pvdrz]

Describe the changes done on this pull request

This PR uses fork and Command::exec to run the command instead of using Command::spawn. This is done because we need to set the command's process group as the foreground process group for the terminal and we cannot execute the command until it is set.

We cannot use Command::pre_exec to wait for the foreground process group to be set before executing the command because the monitor process will get stuck as Command::spawn will block until all the Command::pre_exec closures are executed.

This PR also adds the new concept of "exit reason" for the event loop. The main difference between an exit and a break is that setting the loop to break will stop the loop after the current callback is done, meanwhile setting the loop to exit will stop the loop after all the callbacks for the events that have already been polled are done. Meaning that "exit" is the "nice" way to stop the event loop.

This is another chunk of #363 and it is blocked by #467.

Pull Request Checklist

• I have read and accepted the code of conduct for this project.
• I have tested, formatted and ran clippy over my changes.
• I have commented and documented my changes.
• This pull request will help fix issue sudo-rs ignores SIGTSTP signal #325 where a proper discussion about a solution has taken place.

[github-actions]

Number of dependencies and binary size impact report

Metric	main	PR #468	Delta
Direct dependencies	6	6	-
Total dependencies	13	13	-
Binary size	927.5 KiB	930.1 KiB	+0.3%
Text size	536.2 KiB	540.2 KiB	+0.7%

Dependencies diff

 └─ sudo [v0.1.0-alpha.1]
    ├─ env_logger [v0.9.3]
    |  └─ log [v0.4.17]
    |     └─ cfg-if [v1.0.0]
    ├─ glob [v0.3.1]
    ├─ libc [v0.2.144]
    ├─ log [v0.4.17]
    ├─ signal-hook [v0.3.15]
    |  ├─ libc [v0.2.144]
    |  ├─ signal-hook-registry [v1.4.1]
    |  |  └─ libc [v0.2.144]
    |  └─ cc [v1.0.79]
    └─ signal-hook-registry [v1.4.1]

[codecov]

Codecov Report

Patch coverage: 73.17% and project coverage change: +0.15 🎉

Comparison is base (95093c4) 82.77% compared to head (1e8b055) 82.93%.

Additional details and impacted files

@@            Coverage Diff             @@
##             main     #468      +/-   ##
==========================================
+ Coverage   82.77%   82.93%   +0.15%     
==========================================
  Files          52       52              
  Lines        6852     6961     +109     
==========================================
+ Hits         5672     5773     +101     
- Misses       1180     1188       +8     

Impacted Files	Coverage Δ
sudo/lib/exec/parent.rs	70.10% <51.61%> (-5.45%)	⬇️
sudo/lib/exec/monitor.rs	69.90% <76.13%> (+3.23%)	⬆️
sudo/lib/exec/event.rs	88.29% <80.48%> (-5.74%)	⬇️
sudo/lib/exec/backchannel.rs	96.36% <100.00%> (+0.03%)	⬆️

... and 5 files with indirect coverage changes

☔ View full report in Codecov by Sentry.
📢 Do you have feedback about the report comment? Let us know in this issue.

[japaric]

LGTM

[japaric]

this pattern comes up often when fork-ing. perhaps the signature of fork could be changed to take a child closure then this if ret == 0 can be moved into the body of the fork function. not something that needs to be addressed in this PR; just an idle though

[japaric]

idle thought: this binary serialization over a pipe comes up in other places. perhaps a bit of abstraction could help with the buffer sizes and ser/de steps, e.g. Pipe<T: BinSerDe>, impl BinSerDef for i32 { const NUM_BYTES: usize = 4; /* .. */, Pipe::write(val: &T), Pipe::read() -> io::Result<T>, etc.