How many bytes must be added for your application? I have a driver that needs to insert 6 bytes before sending to the ethernet hardware, and after increasing net_device::hard_header_len by 6 I have not seen it rallocate packets due to insufficient headroom. I have only tested it with a simple FTP session; I don't know how often it would have to reallocate in a situation where packets received from other interfaces had to be transmitted. > -----Original Message----- > From: Venu Vadapalli [mailto:bvadapal@hotmail.com] > Sent: Wednesday, May 14, 2003 9:20 PM > To: linux-net@vger.kernel.org > Subject: RE: skbuff headroom guarantee in the driver? > > > The driver needs to have guaranteed space between the head > and data of > skbuff (prepend MAC header), so it could stuff some > information and send it > the MAC hardware. On the Rx side, the driver allocates the > skbuff, so it can > allocate the required headroom to receive similar information > from the MAC > hardware, without any problem. I'm afraid, the control buffer > of skb doesn't > help in this situation. > > -----Original Message----- > From: Jed Carlson [mailto:jcarlson@katalix.com] > Sent: Wednesday, May 14, 2003 3:44 PM > To: linux-net@vger.kernel.org > Subject: Re: skbuff headroom guarantee in the driver? > > > Have you considered whether you can use skb->cb[]? This can > be used as a > scratch area to pass information with the skb. But be aware that its > contents isn't guaranteed to be preserved as the skb passes > through the > kernel since it may be used by other kernel subsystems. Take > a look at how > it's used by the TCP code via the macro TCP_SKB_CB(). > > -Jed > > >Is there a way to guarantee headroom for the Tx packets to be used in > >the Mac driver. Increasing the hard_header_len doesn't seem to be the > >way to go because in the TCP processing skbuffs are allocated with > >MAX_HEADER_LEN and it doesn't seem appropriate to bank on this > >macro. If I want to pass some extra information (say, some descriptor > >realted) to the MAC by prepending the packet, what is the best way to > >do it? I guess, one could argue that it is no longer a true ethernet > >header and hence a new header should be introduced. However, > the extra > >information is not any mutated mac header but is some descriptor > >information that needs to be passed with packet for performance > >reasons. Needless to say, I'm hoping to avoid using > skb_realloc_header > >in the driver. > > > - > : send the line "unsubscribe > linux-net" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html > > _________________________________________________________________ > The new MSN 8: smart spam protection and 2 months FREE* > http://join.msn.com/?page=features/junkmail > > - > : send the line "unsubscribe > linux-net" in > the body of a message to majordomo@vger.kernel.org > More majordomo info at http://vger.kernel.org/majordomo-info.html > - : send the line "unsubscribe linux-net" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
