crypto: algif_hash - Fix race between MORE and non-MORE sends

The 'MSG_MORE' state of the previous sendmsg() is fetched without the
socket lock held, so two sendmsg calls can race.  This can be seen with a
large sendfile() as that now does a series of sendmsg() calls, and if a
write() comes in on the same socket at an inopportune time, it can flip the
state.

Fix this by moving the fetch of ctx->more inside the socket lock.

Fixes: c662b043cdca ("crypto: af_alg/hash: Support MSG_SPLICE_PAGES")
Reported-by: syzbot+689ec3afb1ef07b766b2@syzkaller.appspotmail.com
Link: https://lore.kernel.org/r/000000000000554b8205ffdea64e@google.com/
Signed-off-by: David Howells <dhowells@redhat.com>
Tested-by: syzbot+689ec3afb1ef07b766b2@syzkaller.appspotmail.com
cc: Herbert Xu <herbert@gondor.apana.org.au>
cc: Paolo Abeni <pabeni@redhat.com>
cc: "David S. Miller" <davem@davemloft.net>
cc: Eric Dumazet <edumazet@google.com>
cc: Jakub Kicinski <kuba@kernel.org>
cc: linux-crypto@vger.kernel.org
cc: netdev@vger.kernel.org
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>

diff --git a/crypto/algif_hash.c b/crypto/algif_hash.c
index 0ab43e1..82c44d4 100644 (file)
--- a/crypto/algif_hash.c
+++ b/crypto/algif_hash.c
@@ -68,13 +68,15 @@ static int hash_sendmsg(struct socket *sock, struct msghdr *msg,
        struct hash_ctx *ctx = ask->private;
        ssize_t copied = 0;
        size_t len, max_pages, npages;
-       bool continuing = ctx->more, need_init = false;
+       bool continuing, need_init = false;
        int err;
 
        max_pages = min_t(size_t, ALG_MAX_PAGES,
                          DIV_ROUND_UP(sk->sk_sndbuf, PAGE_SIZE));
 
        lock_sock(sk);
+       continuing = ctx->more;
+
        if (!continuing) {
                /* Discard a previous request that wasn't marked MSG_MORE. */
                hash_free_result(sk, ctx);