site_archiver@lists.apple.com Delivered-To: darwin-kernel@lists.apple.com Domainkey-signature: a=rsa-sha1; q=dns; c=nofws; s=beta; d=gmail.com; h=received:message-id:date:from:to:subject:cc:in-reply-to:mime-version:content-type:content-transfer-encoding:content-disposition:references; b=nH+j8S0cxQjiELAYP46LqSKHUGpg+6ycNUMLsEH5VL3+TOXQjIsj5SQr0Xpnpy7ma6UEwP/RK3Oy2Jn9HJ0FCqhYDBoVde13ebxVQrJMSBvtDHhzfRJBmTAF6i4OyX7m2Y1ZZEzRBZt+l5jhwfq1bQvoO7I3Ok4Rv2Y6EEm+3Ko= I tried this approach, and it doesn't work consistently. The problem is that not every packet sent to my iff_input_func has the ethernet header stripped (okay not stripped, but the mbuf_data pointer points beyond it). As a result, now I am seeing the ethernet header repeated back to back for those packets that happened to be delivered with the ethernet header intact to my iff_input_func. I need a smarter way to figure out when the kernel is going to strip the ethernet header when delivering it to my iff_input_func. Will the following heuristic work consistently? if (mbuf_data(*packet) != mbuf_pktdat(*packet)) { // the ethernet drivers must have stripped the L2 header. Prepend it } Alternatively, I could compare mbuf_data(*packet) to *frame_ptr passed into the iff_input_func. Thanks - Bhavesh Yuck! But thanks for the code snippet, Josh! Especially because in the last few days, I've had to discover the hard way what I should really pass to mbuf_pkthdr_setheader() as the second parameter :) Thanks! I'll give this a shot too. Incidentally, my approach also "works" in the sense that now my cloned mbufs in my NKE do contain the ethernet header, but as I was afraid, mutating the input mbuf breaks the stack. I'll try your approach instead. Thanks - Bhavesh On 9/18/06, Josh Graessley <jgraessley@apple.com> wrote:

The input function is allowed to make modifications.

You will want to leave the packet in the same format you found it before you return from the function if you're going to let the packet continue up the stack. So something like this would work:

errno_t myFilterInput(ifnet_t interface, protocol_family_t protocol, mbuf_t *packet, char **frame_ptr) { struct ether_header *en_header = (struct ether_header*)*frame_ptr;

/* Prepend the ethernet header */ errno_t result = mbuf_prepend(packet, sizeof(*en_header), MBUF_WAITOK); if (result != 0) { *packet = NULL; return result; } *((struct ether_header*)mbuf_data(*packet)) = *en_header;

/* Do any work you want here, including calling mbuf_dup */ result = doSomeWork(packet);

/* Fixup the packet to allow it to continue up the stack */ if (result == 0 && *packet) { /* Fixup the original so the packet starts after the ethernet header */ /* This same code should be used before injecting an ethernet packet up the stack (ifnet_input). You may also need to set the receive interface. mbuf_pkthdr_setheader(*packet, mbuf_data(*packet)); mbuf_adj(*packet, sizeof(*en_header)); }

return result; }

enjoy, -josh

On Sep 18, 2006, at 4:17 PM, Bhavesh Davda wrote:

...

Thanks for your response, Josh.

I was thinking of an alternate workaround:

I'm not sure what the kpi_mbuf comment about "data" being something the iff_input_func can modify means, but if it indeed can modify it, I was going to "fix up" the m_data pointer to point to m_pktdat, before cloning the packet via mbuf_dup.

The downside of this is that if the "data" parameter is really supposed to be non-mutable, then I've just broken the assumption for the rest of the kernel :(

What do you think?

Thanks

- Bhavesh

On 9/18/06, Josh Graessley <jgraessley@apple.com> wrote:

...

The packet header field is a big disaster.

When an ethernet driver inputs a packet it usually:

- has the whole packet copied in to the mbuf - Sets the mbuf's packet header to point to where the ethernet frame starts in the mbuf - uses mbuf_adj or something similar to move the start of the mbuf pointer to the end of the ethernet header

If you're doing to use mbuf_dup on one of these packets your best bet is probably to: - mbuf_prepend for the length of the ethernet header - copy the ethernet header in to the preceeding space you just created using mbuf_prepend - duplicate the packet - set the packet header's header field to the mbuf_data value - mbuf_adj for the size of the ethernet header

If there was one thing I could undo it would be the wacky way in which this header stuff is handled. It is especially frustrating when you're working with interface filters. On the outbound path, you get the entire packet in the mbuf chain. On the inbound path, the ethernet header is separate, in the packet header's header field. It usually points to some place in the allocated space for the first mbuf, but there are no guarantees this will be the case.

-josh

On Sep 18, 2006, at 1:46 PM, Bhavesh Davda wrote:

...

For certain packets that my iff_input_func is receiving, when I looked at the entire mbuf structure received in parameter 4, I noticed that the m_hdr.mh_data field offsets 14 bytes beyond the start of the ethernet frame.

This is causing my mbuf_dup() cloned mbuf to not contain the 14 bytes of ethernet header in my NKE.

Looking in the original mbuf received in my iff_input_func, the 14 bytes of ethernet header are in the mbuf body at mbuf->M_dat.MH.MH_dat.MH_databuf[0..13].

What am I doing wrong? How do I clone the mbuf to include the DLIL header?

Thanks!

- Bhavesh

-- Bhavesh P. Davda _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel@lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/jgraessley% 40apple.com

This email sent to jgraessley@apple.com

-- Bhavesh P. Davda

-- Bhavesh P. Davda -- Bhavesh P. Davda _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel@lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.a... On 9/18/06, Bhavesh Davda <bhavesh.davda@gmail.com> wrote: This email sent to site_archiver@lists.apple.com