Using this as a dependency is a big no no. Either reimplement or find another solution.

DRY is a liability here.

Yes @stevvooe. I will find other way to handle this. 😄

Was looking around for a suggestion, but I think @dmcgowan covered the options in #2497 (comment).

Found a way to create a Fmountat which we could use here dmcgowan@d8f6d09

@stevvooe yes. the option provided by @dmcgowan works! I have changed the code. Please take a look.

Make a copy of m before altering

should be ${root}/io.containerd.v1.overlayfs/snapshots

Is this officially supported across different versions of Go?

# https://github.com/golang/go/blob/release-branch.go1.8/src/syscall/exec_linux.go#L46
func runtime_BeforeFork()
func runtime_AfterFork()

# https://github.com/golang/go/blob/release-branch.go1.9/src/syscall/exec_linux.go#L52
func runtime_BeforeFork()
func runtime_AfterFork()
func runtime_AfterForkInChild()

1.9 or higher supports the three features. But, I think go1.10 can work well than go1.9 since the runtime.LockOSThread has been improved.

more input from https://golang.org/doc/go1.10#runtime

There should still be a page size check in here, we need to offload that to a linux only section though

This copy should be done as part of the next function, and only when necessary

Is there no other way to do this without a raw syscall? That should only be necessary in the child process

I think you are still missing a waitpid or else we will have zombie babies

I actually don't think you need pipes where, just waitpid(wait4) the exit status and use that. You can always wait on a process, running or not, that you forked off.

I make it complicated and it can cause the zombie process. I have updated the code. Please take a look.

i would almost recommend using unix.Wait4 here to follow the rest of the theme of low level syscalls in this function

make senses. I will update it.

thanks for all your help

should we add loop to retry if the Wait4 return EINTR? it is possible?

ya, good idea

We should still set some buffer room here, maybe start size at 512

the buffer will be used in pagesize check. have updated it.

Looks like there should be another check here that after compacting the options do fit a page, i.e. something like

if optionsSize(options) >= pagesize-512 {
	return errors.Errorf("can't mount: lower layers list too long")
}

Yet better, move this check to after parseMountOptions() as it's simpler and more precise in there:

	flags, data := parseMountOptions(options)
	if len(data) >= pagesize {
		return errors.Errorf("can't mount: mount options string too long")
	}
```
or smth like that

make senses. will update it.

@kolyshkin I add another check after compact and show the error if the mount option is too long.
if we got empty chdir or the data is too long, return the mount options is too long. It is readable to user instead of failed to mount: no such file or directory.