On Sun, 2013-04-28 at 08:11 +0800, Herbert Xu wrote: > On Fri, Apr 26, 2013 at 09:44:17AM -0700, Tim Chen wrote: > > > > + old_tfm = crct10dif_tfm; > > + crc_t10dif_newalg = true; > > + /* make sure new alg flag is turned on before starting to switch tfm */ > > + mb(); > > + > > + new_tfm = crypto_alloc_shash("crct10dif", 0, 0); > > + if (IS_ERR(new_tfm)) > > + goto done; > > + > > + if (old_tfm) > > + if (crypto_tfm_alg_priority(&old_tfm->base) >= > > + crypto_tfm_alg_priority(&new_tfm->base)) { > > + crypto_free_shash(new_tfm); > > + goto done; > > + } > > + crct10dif_tfm = new_tfm; > > + /* make sure update to tfm pointer is completed */ > > + mb(); > > + crypto_free_shash(old_tfm); > > This is not safe at all. You'd need to use something like RCU. > > However, I think this is an overkill. Initialising it once should > be enough. If someone really wanted to change things at run-time, > they could always build this as a module and unload/reload it. > If I allocate the transform under the mod init instead, how can I make sure that the fast version is already registered if I have it compiled in? It is not clear to me how that's done looking at the libcrc32c code. Thanks. Tim -- 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