-block-fd unblocks on EOF and command get executed

[a-liashenko]

Is it intentional that --block-fd treats EOF as "green flag" to execute process?
A normal workflow:

1. Start bwrap --block-fd <fd> <cmd>
2. Do some extra work, launch extra services, etc
3. Write 1 byte into fd

If step 2 failed I want to stop bwrap WITHOUT executing command. But closing pipe or SIGTERM/SIGKILL bwrap will allow child process to execute command. Unfortunately, only way is killing bwrap child process by pid directly.

What do you think the best way to solve it? For compatibility reasons I can prepare PR with extra info about EOF. But I think child process should die if no data arrived to fd.
From man: Block the sandbox on reading from FD until some data is available.

Strace log

user@pc $ strace -f -e trace=read,execve bwrap --block-fd 3 --bind / / env 3</dev/null
execve("/usr/bin/bwrap", ["bwrap", "--block-fd", "3", "--bind", "/", "/", "env"], 0x7ffeaa1a3878 /* 60 vars */) = 0
strace: Process 399130 attached
[pid 399130] read(5, "\1\0\0\0\0\0\0\0", 8) = 8
[pid 399130] read(6, "327 1221 0:36 /@ /oldroot rw,noa"..., 4079) = 4074
[pid 399130] read(6, "1219 ", 5)        = 5
[pid 399130] read(6, "327 259:1 / /oldroot/boot rw,noa"..., 4080) = 4080
[pid 399130] read(6, ",nosuid,nodev master:155 - tmpfs"..., 8160) = 949
[pid 399130] read(6, "", 7211)          = 0
========== HERE IS EOF HAPPENS =====
[pid 399130] read(3, "", 1)             = 0
========== HERE CHILD WILL EXECUTE COMMAND ========
[pid 399130] execve("/usr/bin/env", ["env"], 0x7ffff0974608 /* 60 vars */) = 0

Code

Looks like result of read(...) is never checked against 0 (EOF).

bubblewrap/bubblewrap.c

Lines 48 to 54 in 8e51677

	#define TEMP_FAILURE_RETRY(expression) \
	(__extension__ \
	({ long int __result; \
	do __result = (long int) (expression); \
	while (__result == -1L && errno == EINTR); \
	__result; }))
	#endif

bubblewrap/bubblewrap.c

Lines 3261 to 3266 in 8e51677

	if (opt_block_fd != -1)
	{
	char b[1];
	(void) TEMP_FAILURE_RETRY (read (opt_block_fd, b, 1));
	close (opt_block_fd);
	}

[smcv]

Is it intentional that --block-fd treats EOF as "green flag" to execute process?

I believe so. In other tools I've used a name more like --block-until-readable for the equivalent feature, which makes it more obvious how it works / what it does.

Unfortunately, only way is killing bwrap child process by pid directly.

If you want to terminate bwrap then that would be the way to do it.

I think child process should die if no data arrived to fd

That would be a backward-compatibility break for anything that is relying on the current behaviour.

[a-liashenko]

Thank you for response, I was a bit confused.

If you want to terminate bwrap then that would be the way to do it.

Any pitfalls to call SIGTERM -> SIGKILL for child_pid from --json-status-fd in my scenario?