fd: add close_range(..., CLOSE_RANGE_CLOEXEC)

[mjg]

Summary

For compatibility with Linux.

[mjg]

I would like to get this in for 13.1

[mjg]

• rephrase manpage

[mjg]

Linux compat will be updated separately. For some reason it has DUMMY(close_range);

[mjg]

• use the same value for the flag as linux to avoid future mess

[markj]

Seems ok. It would be nice to have a simple regression test in tests/sys/file/closefrom_test.c to ensure this code gets exercised.

lib/libc/sys/closefrom.2

63

sys/kern/kern_descrip.c

1410

This error path leaks the lock.

1493–1494

This comment can be deleted.

sys/sys/unistd.h

206

Hmm but then you have to update the pattern used in libsysdecode. Alternately just write 0x4 since that's a more common style for flag definitions.

[mjg]

sys/sys/unistd.h

206

The above is copy-pasted from rfork. I had manually typed this off but it failed to patter-match, I don't think touching this is worth any effort.

[mjg]

• manpage

[mjg]

How do you suggest to sensibly test this though? To my surprise I found procstat does not report CLOEXEC, so as is I would have to actually exec.

[markj]

In D34424#779939, @mjg wrote:

How do you suggest to sensibly test this though? To my surprise I found procstat does not report CLOEXEC, so as is I would have to actually exec.

You can use fcntl(F_GETFD) to test for the flag.

[mjg]

• add a rudimentary unit test

[markj]

Thanks for adding the test.

tests/sys/file/closefrom_test.c

333

I'd check for an error from close_range(), though I see that the pre-existing tests do not for some reason.

343

This test will pass if fcntl() fails. It'd be more rigourous to check for failure before testing FD_CLOEXEC.