The server will only overwrite the mac if it is zero.
Signed-off-by: Dominik Heidler dominik@heidler.eu ---
The alfred client sets the mac to zero by default so this shouldn't break existing behaviour.
unix_sock.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/unix_sock.c b/unix_sock.c index 3c7e583..12a10e6 100644 --- a/unix_sock.c +++ b/unix_sock.c @@ -119,7 +119,9 @@ static int unix_sock_add_data(struct globals *globals,
data = push->data; data_len = ntohs(data->header.length); - memcpy(data->source, &interface->hwaddr, sizeof(interface->hwaddr)); + static const char zero[ETH_ALEN] = { 0 }; + if (!memcmp(zero, data->source, sizeof(data->source))) + memcpy(data->source, &interface->hwaddr, sizeof(interface->hwaddr));
if ((int)(data_len + sizeof(*data)) > len) goto err;
On Wednesday 17 February 2016 00:51:10 Dominik Heidler wrote:
The server will only overwrite the mac if it is zero.
Signed-off-by: Dominik Heidler dominik@heidler.eu
I think this breaks the forwarding/synchronization rules of alfred.
@Simon, what do you say?
kind regards, Sven
The alfred client sets the mac to zero by default so this shouldn't break existing behaviour.
unix_sock.c | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-)
diff --git a/unix_sock.c b/unix_sock.c index 3c7e583..12a10e6 100644 --- a/unix_sock.c +++ b/unix_sock.c @@ -119,7 +119,9 @@ static int unix_sock_add_data(struct globals *globals,
data = push->data; data_len = ntohs(data->header.length);
- memcpy(data->source, &interface->hwaddr, sizeof(interface->hwaddr));
static const char zero[ETH_ALEN] = { 0 };
if (!memcmp(zero, data->source, sizeof(data->source)))
memcpy(data->source, &interface->hwaddr, sizeof(interface->hwaddr));if ((int)(data_len + sizeof(*data)) > len) goto err;
On Wednesday 17 February 2016 00:51:10 Dominik Heidler wrote:
The alfred client sets the mac to zero by default so this shouldn't break existing behaviour.
This also may not to be true. Some users of alfred may not set the source in alfred_data and thus they may end up with random bytes when reusing a buffer, allocating a buffer without memset or using an uninitialized buffer on stack.
It is bad practice to send uninitialized bytes around but I just wanted to point out that potential problem.
Your patch also doesn't explain why this feature may be required.
Btw. you can grep for "data_source" to find the code related to the sync/forwarded rules of alfred.
Kind regards, Sven
Am 17.02.2016 um 09:39 schrieb Sven Eckelmann:
Your patch also doesn't explain why this feature may be required.
I'm running an alfred master server with this patch, so that I can run a script that will inject data of legacy clients (that don't have an alfred slave, but need to be crawled).
Regards, Dominik
On Wednesday 17 February 2016 11:36:01 Dominik Heidler wrote:
Am 17.02.2016 um 09:39 schrieb Sven Eckelmann:
Your patch also doesn't explain why this feature may be required.
I'm running an alfred master server with this patch, so that I can run a script that will inject data of legacy clients (that don't have an alfred slave, but need to be crawled).
Sorry for the late reply. In generally I think this change is acceptable, however I'd like to request the following changes and to resend your patch:
* Please write a proper commit message in the patch, explaining your scenario and why you need it * Write a comment above the change you made to explain that clients should set the mac to 00:00:00:00:00:00 by default, unless they want alfred to use its mac for them.
Thank you! Simon
b.a.t.m.a.n@lists.open-mesh.org
-
Dominik Heidler -
Simon Wunderlich -
Sven Eckelmann