do_socketpair possible bug when fd table is full

[scaspin]

Similarly to #1171, the raw_descriptor_store lock is grabbed three times in succession for two inserts. If the second operation fails (because the table is full, for example) the first fd is removed. If a concurrent thread interleaves between these two lock acquisitions and closes the same slot, the second insert's failure mode will either try to close a non-existent fd or an unrelated fd.

litebox/litebox_shim_linux/src/syscalls/net.rs

Lines 1075 to 1083 in 9732adf

	let raw_fd1 = files.insert_raw_fd(typed1).map_err(|typed| {
	let _ = self.global.litebox.descriptor_table_mut().remove(&typed);
	Errno::EMFILE
	})?;
	let raw_fd2 = files.insert_raw_fd(typed2).map_err(|typed| {
	self.do_close(raw_fd1).unwrap();
	let _ = self.global.litebox.descriptor_table_mut().remove(&typed);
	Errno::EMFILE
	})?;