Re: [RFC][PATCH] Make cryptoapi non-optional?

["David S. Miller" <davem@redhat.com>]

On Sat, 9 Aug 2003 02:44:59 -0500
Matt Mackall <mpm@selenic.com> wrote:

> The attached (lightly tested) patch gets rid of the SHA in the
> /dev/random code and replaces it with cryptoapi, leaving us with just
> one SHA implementation.
 ...
>  __u32 secure_tcp_syn_cookie(__u32 saddr, __u32 daddr, __u16 sport,
>  		__u16 dport, __u32 sseq, __u32 count, __u32 data)
 ...
> +	tfm = crypto_alloc_tfm("sha1", 0);

secure_tcp_syn_cookie() is called from software interrupt
context, therefore it may not sleep.

So you cannot call crypto_alloc_tfm() here, which can sleep.

Not to mention that calling crypto_alloc_tfm() for every TCP
connection creation is absurdly expensive.

Same thing in check_tcp_syn_cookie().

Also, same problem exists with extract_entropy() which also must be
callable from software interrupt context, and thus the
crypto_alloc_tfm() alloc calls you added there are illegal too.

This patch needs tons of work.