Re: [PATCH] settrans -gl only kills the active translator

[Roland McGrath]

I don't understand what your option is supposed to do.  
With -a and but no -p, regardless of the other options, settrans
will not affect the passive translator.  The file_set_translator
RPC affects either passive or active or both, according to the flags
arguments.  

Your code does nothing to either translator setting, but just sends an
fsys_goaway to the active translator.  This leaves it up to the translator
to die or not as it chooses.  The file_set_translator RPC that will be made
by "settrans -a FILE" to the parent filesystem will do the same fsys_goaway
call to the translator.  So you haven't really added anything new.

The intent of that TODO item is more like the opposite of the direction you
seem to be thinking.  If you follow the code path on the server side, you
will see that it will always call fsys_goaway on the active translator if
there is one, and if that server returns some error code, the parent
filesystem will just propagate the error and refuse to reset the active
translator.  So, a truly reticent filesystem server can't be "unmounted"
from a node by any means in the filesystem protocols.  You have to just
kill the process.

So, what "without using goaway" means in this TODO item is having a way in
the file_set_translator protocol to detach the old translator without
regard for its feelings on the subject.  Add a flag bit to the ACTIVE_FLAGS
argument that tells the server to skip the fsys_goaway call and just reset
its port.  Then settrans would get a new option to use this flag bit.