bpf: sockmap update/simplify memory accounting scheme

Instead of tracking wmem_queued and sk_mem_charge by incrementing
in the verdict SK_REDIRECT paths and decrementing in the tx work
path use skb_set_owner_w and sock_writeable helpers. This solves
a few issues with the current code. First, in SK_REDIRECT inc on
sk_wmem_queued and sk_mem_charge were being done without the peers
sock lock being held. Under stress this can result in accounting
errors when tx work and/or multiple verdict decisions are working
on the peer psock.

Additionally, this cleans up the code because we can rely on the
default destructor to decrement memory accounting on kfree_skb. Also
this will trigger sk_write_space when space becomes available on
kfree_skb() which wasn't happening before and prevent __sk_free
from being called until all in-flight packets are completed.

Fixes: 174a79ff9515 ("bpf: sockmap with sk redirect support")
Signed-off-by: John Fastabend <john.fastabend@gmail.com>
Acked-by: Daniel Borkmann <daniel@iogearbox.net>
Acked-by: Alexei Starovoitov <ast@kernel.org>
Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/kernel/bpf/sockmap.c b/kernel/bpf/sockmap.c
index db0d99d..f6ffde9 100644 (file)
--- a/kernel/bpf/sockmap.c
+++ b/kernel/bpf/sockmap.c
@@ -111,7 +111,7 @@ static int smap_verdict_func(struct smap_psock *psock, struct sk_buff *skb)
 
 static void smap_do_verdict(struct smap_psock *psock, struct sk_buff *skb)
 {
-       struct sock *sock;
+       struct sock *sk;
        int rc;
 
        /* Because we use per cpu values to feed input from sock redirect
@@ -123,16 +123,16 @@ static void smap_do_verdict(struct smap_psock *psock, struct sk_buff *skb)
        rc = smap_verdict_func(psock, skb);
        switch (rc) {
        case SK_REDIRECT:
-               sock = do_sk_redirect_map();
+               sk = do_sk_redirect_map();
                preempt_enable();
-               if (likely(sock)) {
-                       struct smap_psock *peer = smap_psock_sk(sock);
+               if (likely(sk)) {
+                       struct smap_psock *peer = smap_psock_sk(sk);
 
                        if (likely(peer &&
                                   test_bit(SMAP_TX_RUNNING, &peer->state) &&
-                                  sk_stream_memory_free(peer->sock))) {
-                               peer->sock->sk_wmem_queued += skb->truesize;
-                               sk_mem_charge(peer->sock, skb->truesize);
+                                  !sock_flag(sk, SOCK_DEAD) &&
+                                  sock_writeable(sk))) {
+                               skb_set_owner_w(skb, sk);
                                skb_queue_tail(&peer->rxqueue, skb);
                                schedule_work(&peer->tx_work);
                                break;
@@ -282,16 +282,12 @@ start:
                                /* Hard errors break pipe and stop xmit */
                                smap_report_sk_error(psock, n ? -n : EPIPE);
                                clear_bit(SMAP_TX_RUNNING, &psock->state);
-                               sk_mem_uncharge(psock->sock, skb->truesize);
-                               psock->sock->sk_wmem_queued -= skb->truesize;
                                kfree_skb(skb);
                                goto out;
                        }
                        rem -= n;
                        off += n;
                } while (rem);
-               sk_mem_uncharge(psock->sock, skb->truesize);
-               psock->sock->sk_wmem_queued -= skb->truesize;
                kfree_skb(skb);
        }
 out: