Attention is currently required from: arehbein, laforge, pespin.

daniel has posted comments on this change. ( https://gerrit.osmocom.org/c/libosmo-netif/+/34338?usp=email )

Change subject: stream_srv: Fix connection error handling ......................................................................

Patch Set 2:

(2 comments)

File src/stream_srv.c:

https://gerrit.osmocom.org/c/libosmo-netif/+/34338/comment/3cb40480_27675525 PS2, Line 489: if (OSMO_UNLIKELY(res <= 0)) {

Why do we not call `osmo_stream_srv_clear_tx_queue()` here, if the connection is dead anyways?

Because that's what osmo_stream_srv_destroy() does already (either through msgb_queue_free() or osmo_iofd_free())

https://gerrit.osmocom.org/c/libosmo-netif/+/34338/comment/ff657801_7ece556d PS2, Line 496: if (osmo_iofd_txqueue_len(iofd) == 0)

Can it not happen that messages can't be sent out, so that this condition never holds true? In that […]

I just noticed that osmo_stream_srv_clear_tx_queue() is not aware of osmo_io. I'll prepare a patch.

We only end up here if there's actual data to be read, so it's not really an endless loop.

The only way where a message is never sent out is if the fd never is writable. Even if sending then fails that message has been removed from the tx queue so it will eventually be empty.