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

[Neal H Walfield]

> I have no objection to adding a flag for file_set_translator to avoid
> sending the goaway.  But calling the flag "disconnect" is wrong: a
> disconnect happens whether you set that flag or not.

I fail to see how this is true: if the translator, in response to
fsys_goaway, returns an error, diskfs_S_file_set_translator deallocates
a user reference to the control port, which it obtained in
fshelp_fetch_control, and returns the error.

Most translators interpret fsys_goaway as meaning: shutdown (i.e. call
exit).  Here, the disconnection is an obvious result of the exit.

As far as I can tell, libdiskfs currently has no facility to disconnect
an active translator from the file system (i.e. deallocate the user
reference to the control port to the active translator) without also
asking the filesystem to shutdown.

> I suggest FSYS_GOAWAY_NORPC or FSYS_GOAWAY_NOGOAWAY or something like
> that, that accurate describes what the option requests.

Again, I fail to see how this is more accurate than what I have proposed
(in fact, I think it is less accurate).  fsys_goaway means please call
exit.  If anything, FSYS_GOAWAY_DISCONNECT should be renamed to
FS_TRANS_DISCONNECT and be or'ed into the active flags rather than the
goaway flags.

> As for the shell command settrans, I think that -f should timeout, and
> if nothing happens then repeat the call with the new flag attached.
> Perhaps there should be a new flag with the current meaning of -f
> (that is, set the FORCE bit, but don't bypass the goaway process).

I do not agree: the new bit (disconnect) means disconnect the active
translator from the filesystem while fsys_goaway means shutdown (i.e.
call exit).  Adding the FORCE bit means shutdown even if there are
users.  This, in my mind, is orthogonal to disconnecting the active
translator from the filesystem and, furthermore, the latter does not
imply the former.

pgpjsE9djmPnD.pgp
Description: PGP signature