Re: [arch-projects] [netctl] Still some weird behaviour with bridges

On Mon, May 06, 2013 at 07:49:48PM +0200, Jouke Witteveen wrote:

On Mon, May 6, 2013 at 7:36 PM, Dave Reisner wrote:

...

On Mon, May 06, 2013 at 07:23:58PM +0200, Jouke Witteveen wrote:

...

On Mon, May 6, 2013 at 10:22 AM, Thomas Bächler wrote:

...

I have this profile:

Interface=vmnet Connection=bridge BindsToInterfaces=() IP=static SkipNoCarrier=yes Address=('1.2.3.4/5')

While it seems to work fine, systemd prints these messages:

Starting network profile 'vmnet'... Cannot find device "on" Usage: ip addr {add|change|replace} IFADDR dev STRING [ LIFETIME ] [ CONFFLAG-LIST ] ip addr del IFADDR dev STRING ip addr {show|save|flush} [ dev STRING ] [ scope SCOPE-ID ] [ to PREFIX ] [ FLAG-LIST ] [ label PATTERN ] [up] ip addr {showdump|restore} IFADDR := PREFIX | ADDR peer PREFIX [ broadcast ADDR ] [ anycast ADDR ] [ label STRING ] [ scope SCOPE-ID ] SCOPE-ID := [ host | link | global | NUMBER ] FLAG-LIST := [ FLAG-LIST ] FLAG FLAG := [ permanent | dynamic | secondary | primary | tentative | deprecated | dadfailed | temporary | CONFFLAG-LIST ] CONFFLAG-LIST := [ CONFFLAG-LIST ] CONFFLAG CONFFLAG := [ home | nodad ] LIFETIME := [ valid_lft LFT ] [ preferred_lft LFT ] LFT := forever | SECONDS interface does not exist! Started network profile 'vmnet'

This is most likely related to netctl not handling the empty BindsToInterfaces properly.

You are right. This (arrays in Bash) is pretty tricky. I believe it is fixed now. If so, the fix will land in netctl 1.1.

Thanks, - Jouke

The problem is that you're declaring it initially as a string, and then expecting it to be an array. You simply cannot interchange the types without encountering unwanted behavior like this.

I'd like to keep the possibility of having declarations like this in a profile:

Address='1.2.3.4/5'

Although Address is treated like an array internally, this works. As far as I can tell it is relatively safe to treat non-empty strings as 1-element arrays.

I think this is the wrong thing to do. You need to enforce types if you're consuming raw bash as configuration. You're telling people that k='foo' is valid, but k='foo bar' is not and suddenly they must use k=(foo bar). This is bizzare.

...

Note that you're still doing this in src/netctl.in:112. I'll also point out that your -v check allows people to create profiles that do not bind to any interfaces since it checks for the definition and not the existence of a value -- are you sure you want this?

Regarding src/netctl.in:112, the variable is set to a non-empty string (in src/lib/globals:load_profile, it is actually verified that $Interface is non-empty). By the above safety assumption, this should be acceptable. The current behavior is to bind to the Interface by default. By explicitly setting BindsToInterfaces, this behavior can be altered. Indeed, it is possible to do

BindsToInterfaces=''

in a profile, which will most likely screw things up (it will instruct systemd to bind to a device without a name). However, I consider this a usage error. The current configuration allows setting BindsToInterfaces to a possibly empty array as well as specifying it correctly by a non-empty string. Those are the uses that I would like to support. If a better way exists, I'd be happy to hear about it :-).

And based on what you said above, you don't believe this is a double standard? If someone were to adhere to the fact that BindsToInterfaces is an array and not a string, then BindsToInterfaces=() is a valid way of declaring this as empty and netctl will respect that. d