On Mon, Feb 23, 2009 at 05:35:58PM +0800, Herbert Xu wrote: > > 4. Operate. > > sendmsg(fd, msg, flags); > > Here struct msghdr provides for both the IOV array and space for > ancillary information. In fact we should take this one step further. Instead of having write/send do both input as well as output, it should just do the former. That is, we should call read/recv to get the output. This might seem a step backwards, but in fact it's the only sane way to do zero-copy. In order to get zero copy with this, you'd use splice instead of write (the setup can be done with the usual corking method), and the receive side would use vmsplice + splice. You might argue that this is too much overhead. However, it isn't because 1) For the cases where you have less than a page of data, just use send/recv as the overhead of the copy is dwarfed by the overhead of the rest of the crypto operation. 2) Where zero-copy does matter the setup overhead is immaterial. Of course to make vmsplice + splice work, we don't want to start the operation until the first read request from the user is issued. Cheers, -- Visit Openswan at http://www.openswan.org/ Email: Herbert Xu ~{PmV>HI~} <herbert@xxxxxxxxxxxxxxxxxxx> Home Page: http://gondor.apana.org.au/~herbert/ PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt -- To unsubscribe from this list: send the line "unsubscribe linux-crypto" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
