Am Dienstag, 30. Dezember 2014, 04:33:41 schrieb Herbert Xu: Hi Herbert, > On Mon, Dec 29, 2014 at 04:05:40PM +0100, Stephan Mueller wrote: > > This would mean that the check must stay in recvmsg as only here we know > > that the caller wants data to be processed. > > On the send side you would do the check when MSG_MORE is unset. > On the receive side you should stop waiting only when ctx->more > is false and the send-side check succeeded. > > Perhaps rename ctx->more to ctx->done and then you can use it > to indicate to the receive side that we're ready and have valid > data for it. The receive side can then simply wait for ctx->done > to become true. I followed your advise and changed the sleep to wait for !ctx->more. Together with the patch suggested below that was just released, I removed aead_readable and aead_sufficient_data. Though, I did not rename ctx->more to ctx->done due to the following: - other AF_ALG implementations use ctx->more -- to aid code reviewers, I want to keep the logic as close together as possible - I do not want to negate the check for MSG_MORE -- at least for me, negating flags always twists my mind when reading code. The patch will come shortly after I tested all thoroughly. > > > > PS we should add a length check for missing/partial auth tags > > > to crypto_aead_decrypt. We can then remove such checks from > > > individual implementations. > > > > I agree in full here. Shall I create such a patch together with the AEAD > > AF_ALG interface, or can we merge the AEAD without that patch now and > > create a separate patch later? > > We should at least add a check in crypto_aead_decrypt first so as > to guarantee nothing slips through. > > Thanks, -- Ciao Stephan -- 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
