[lxc-users] LXCFS - dash read builtin behavior oddity?

[Serge Hallyn]

Interesting!  I can reproduce it trivially.

This is most likely a result of the change to using direct io.  This causes a
stat to return size 0.  Would you mind filing an issue at
github.com/lxc/lxcfs/issues ?  I expect that a quick look at the cat and dash
sources will show us whether dash needs a patch or whether lxcfs should be
providing anohter hint about the fact that its /proc files should be read to
get the full size.

Quoting Robert Pendell (shinji at elite-systems.org):
> Ok.  I'm left scratching a little.  When running under 0.9 (latest
> code compiled from git) dash behaves in a weird way inside the
> container.  This behavior is non reproducable on 0.7.  When doing the
> same experiment using bash instead it does not reproduce so this is
> application specific from what I can tell. Running strace on lxcfs did
> give some interesting insight into this.
> 
> While running this: read b c < /proc/uptime
> from dash inside the container 0.7 gives appropriate data where 0.9
> returns /0 for unknown reasons.
> Using cat from dash works as expected.  Running that same command
> above from within bash also gives the expected results.
> 
> This discovery was made while using byobu since I had it reporting
> uptime and after updating the binary it started reporting "s" for
> uptime.
> 
> Robert Pendell
> shinji at elite-systems.org
> A perfect world is one of chaos.
> Keybase: http://keybase.io/shinji257
> _______________________________________________
> lxc-users mailing list
> lxc-users at lists.linuxcontainers.org
> http://lists.linuxcontainers.org/listinfo/lxc-users