mountinfo.Mounted: optimize by adding fast paths

[kolyshkin]

Add two fast paths to avoid (slow) mountinfo parsing:

1. In case path does not exist, it is definitely NOT a mount point.

2. In case path's st.Dev differs from that of its parent, it is
   definitely a mount point.

In all the other cases (including bind mounts, case of path="/", other
errors from unix.Lstat) the code falls back to the old slow method
of parsing mountinfo.

[kolyshkin]

So, the method is not working for bind mounts, so I changed the commit to try comparing devices first, then falling back to mountinfo parsing. This way, if there is a proper (not bind) mount, we can skip parsing mountinfo.

[thaJeztah]

Left some thoughts/comments

[kolyshkin]

I have reworked it considerably, PTAL @tianon @thaJeztah @cpuguy83

[thaJeztah]

Looks like there's still a duplicate declaration for mounted() on darwin;

file	darwin	freebsd	linux	windows
mountinfo_unix	✅	✅	✅	-
mountinfo_unsupported	✅	-	-	-
mountinfo_windows	-	-	-	✅
conflict?	⚠️	✅	✅	✅

[thaJeztah]

Cross is failing;

for os in linux freebsd darwin windows; do \
	for arch in amd64 arm arm64 ppc64le s390x; do \
		echo "$os/$arch" | grep -qE '^((freebsd|darwin|windows)/(ppc64le|s390x)|windows/arm64|freebsd/s390x)$' && continue; \
		echo "# building for $os/$arch"; \
		GOOS=$os GOARCH=$arch go build ./...; \
	done; \
done
# building for linux/amd64
##[error]mount/mount.go:9:2: cannot find package "github.com/moby/sys/mountinfo" in any of:
	/opt/hostedtoolcache/go/1.13.12/x64/src/github.com/moby/sys/mountinfo (from $GOROOT)
	/home/runner/go/src/github.com/moby/sys/mountinfo (from $GOPATH)
##[error]mount/flags_linux.go:4:2: cannot find package "golang.org/x/sys/unix" in any of:
	/opt/hostedtoolcache/go/1.13.12/x64/src/golang.org/x/sys/unix (from $GOROOT)
	/home/runner/go/src/golang.org/x/sys/unix (from $GOPATH)
##[error]Makefile:33: recipe for target 'cross' failed
make: *** [cross] Error 1
##[error]Process completed with exit code 2.

Does it need the same trickery as we use for test?

sys/Makefile

Lines 10 to 12 in fb9f8cf

	for p in $(PACKAGES); do \
	(cd $$p && go test -v .); \
	done

Starting to wonder if it would be cleaner to have separate Makefiles in each module, and then call those targets from the main Makefile (also beginning to feel the pain of the submodules 😓 - do you think it causes more issues to have them in the same repository than to have to maintain two repositories?)

[kolyshkin]

Will fix

Starting to wonder if it would be cleaner to have separate Makefiles in each module

It will result in more code (same/similar for in the top makefile, plus go test in every sub-makefile).

[kolyshkin]

OK we need replace in go.mod for such tests, and once replace is there, we can ditch the for p in $(PACKAGES) loop in the top Makefile.

[kolyshkin]

Separated out cross build to #23

[thaJeztah]

LGTM, thanks!

[thaJeztah]

@cpuguy83 PTAL

[thaJeztah]

oh, and @tianon 🤗

[tianon]

LGTM 😄

[tianon]

I had some concern that a "broken" FUSE mount might hit this case, but it returns ENOTCONN. 👍

(Just including a note in case anyone else reading this has the same question/concern. 😅)

[kolyshkin]

@cpuguy83 PTAL

[cpuguy83]

LGTM