>
> Hi,
>
> On Fri, 2017-06-02 at 11:31 -0400, Frediano Ziglio wrote:
> > >
> > > Inform client that transfer will not work, instead of only
> > > dropping the agent messages.
> > >
> > > The file transfer messages can be disabled using qemu cli option:
> > > "disable-agent-file-xfer"
> > >
> > > Resolves:
> > > https://bugzilla.redhat.com/show_bug.cgi?id=1373725
> > >
> > > Signed-off-by: Pavel Grunt <pgrunt at redhat.com>
> > > ---
> > > server/reds.c | 34 ++++++++++++++++++++++++++++++++++
> > > 1 file changed, 34 insertions(+)
> > >
> > > diff --git a/server/reds.c b/server/reds.c
> > > index bafd04bab..fd163250d 100644
> > > --- a/server/reds.c
> > > +++ b/server/reds.c
> > > @@ -1111,6 +1111,39 @@ static void
> > > reds_on_main_agent_monitors_config(RedsState *reds,
> > > reds_client_monitors_config_cleanup(reds);
> > > }
> > >
> > > +static void agent_data_free(uint8_t *data, G_GNUC_UNUSED void
> > > *opaque)
> > > +{
> > > + free(data);
> > > +}
> > > +
> > > +/*
> > > + * Inform client about discarded file transfer
> > > + * https://bugzilla.redhat.com/show_bug.cgi?id=1373725
> > > + */
> > > +static void agent_xfer_error(MainChannelClient *mcc, const
> > > VDAgentMessage
> > > *message)
> > > +{
> > > + VDAgentMessage *xfer_msg;
> > > + VDAgentFileXferStatusMessage *xfer_status;
> > > + size_t len;
> > > +
> > > + if (message->type < VD_AGENT_FILE_XFER_START || message->type
> > > >
> > > VD_AGENT_FILE_XFER_DATA) {
> > > + return;
> > > + }
> > > +
> > > + len = sizeof(VDAgentMessage) +
> > > sizeof(VDAgentFileXferStatusMessage);
> > > + xfer_msg = spice_malloc0(len);
> > > + xfer_msg->protocol = VD_AGENT_PROTOCOL;
> > > + xfer_msg->type = VD_AGENT_FILE_XFER_STATUS;
> > > + xfer_msg->size = sizeof(VDAgentFileXferStatusMessage);
> > > + xfer_status = (VDAgentFileXferStatusMessage *) xfer_msg-
> > > >data;
> > > + xfer_status->id = ((VDAgentFileXferStartMessage *) message-
> > > >data)->id;
> >
> > VDAgentFileXferStartMessage should be const but the real problem is
> > this:
> >
> > reds.c: In function 'agent_xfer_error':
> > reds.c:1139:5: error: dereferencing type-punned pointer will break
> > strict-aliasing rules [-Werror=strict-aliasing]
> > xfer_status->id = ((VDAgentFileXferStartMessage *) message-
> > >data)->id;
> >
> > ^~~~~~~~~~~
> >
> Do you have a clue why I don't get the warning?
>
> my CFLAGS are:
> -Wall -Wextra -g -Wstrict-aliasing
>
> gcc-7.1.1-1.fc26.x86_64
> clang-4.0.0-3.fc26.x86_64
>
> Thanks,
> Pavel
>
Mine is gcc-6.3.1-1.fc25.x86_64 (Fedora 25).
No idea, as far as I know there should be no alias but maybe the tracking
of 6.3 is not enough to detect it.
Note that using uint32_t instead of VDAgentFileXferStartMessage have the
same issue.
xfer_status->id = ((const VDAgentFileXferStartMessage *) (message+1))->id;
works for me. A bit less readable but not that much.
Frediano
> > no idea... but if I write (message+1) instead of message->data
> > (which is from ABI the same)
> > the warning disappears :-( Not counting that data is uint8_t, a type
> > that is supposed to
> > be ok to be aliases... but maybe I didn't understand something about
> > aliasing.
> > Also spice_malloc0 is marked as __attribute__((malloc)) which should
> > tell compiler
> > that xfer_msg (and so xfer_status) cannot alias...
> >
> > > + xfer_status->result = VD_AGENT_FILE_XFER_STATUS_ERROR;
> > > +
> > > + spice_debug("canceling file transfer %u", xfer_status->id);
> > > +
> > > + main_channel_client_push_agent_data(mcc, (uint8_t *)
> > > xfer_msg, len,
> > > agent_data_free, NULL);
> > > +}
> > > +
> > > void reds_on_main_agent_data(RedsState *reds, MainChannelClient
> > > *mcc, const
> > > void *message,
> > > size_t size)
> > > {
> > > @@ -1124,6 +1157,7 @@ void reds_on_main_agent_data(RedsState
> > > *reds,
> > > MainChannelClient *mcc, const void
> > > case AGENT_MSG_FILTER_OK:
> > > break;
> > > case AGENT_MSG_FILTER_DISCARD:
> > > + agent_xfer_error(mcc, message);
> > > return;
> > > case AGENT_MSG_FILTER_MONITORS_CONFIG:
> > > reds_on_main_agent_monitors_config(reds, mcc, message,
> > > size);
> >
> > Frediano
>