af_unix: improve STREAM behavior with fragmented memory

unix_stream_sendmsg() currently uses order-2 allocations,
and we had numerous reports this can fail.

The __GFP_REPEAT flag present in sock_alloc_send_pskb() is
not helping.

This patch extends the work done in commit eb6a24816b247c
("af_unix: reduce high order page allocations) for
datagram sockets.

This opens the possibility of zero copy IO (splice() and
friends)

The trick is to not use skb_pull() anymore in recvmsg() path,
and instead add a @consumed field in UNIXCB() to track amount
of already read payload in the skb.

There is a performance regression for large sends
because of extra page allocations that will be addressed
in a follow-up patch, allowing sock_alloc_send_pskb()
to attempt high order page allocations.

Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: David Rientjes <rientjes@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

diff --git a/include/net/af_unix.h b/include/net/af_unix.h
index 05442df..a175ba4 100644 (file)
--- a/include/net/af_unix.h
+++ b/include/net/af_unix.h
@@ -35,6 +35,7 @@ struct unix_skb_parms {
 #ifdef CONFIG_SECURITY_NETWORK
        u32                     secid;          /* Security ID          */
 #endif
+       u32                     consumed;
 };
 
 #define UNIXCB(skb)    (*(struct unix_skb_parms *)&((skb)->cb))

diff --git a/net/unix/af_unix.c b/net/unix/af_unix.c
index c4ce243..99dc760 100644 (file)
--- a/net/unix/af_unix.c
+++ b/net/unix/af_unix.c
@@ -1596,6 +1596,10 @@ out:
        return err;
 }
 
+/* We use paged skbs for stream sockets, and limit occupancy to 32768
+ * bytes, and a minimun of a full page.
+ */
+#define UNIX_SKB_FRAGS_SZ (PAGE_SIZE << get_order(32768))
 
 static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
                               struct msghdr *msg, size_t len)
@@ -1609,6 +1613,7 @@ static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
        struct scm_cookie tmp_scm;
        bool fds_sent = false;
        int max_level;
+       int data_len;
 
        if (NULL == siocb->scm)
                siocb->scm = &tmp_scm;
@@ -1635,40 +1640,21 @@ static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
                goto pipe_err;
 
        while (sent < len) {
-               /*
-                *      Optimisation for the fact that under 0.01% of X
-                *      messages typically need breaking up.
-                */
-
-               size = len-sent;
+               size = len - sent;
 
                /* Keep two messages in the pipe so it schedules better */
-               if (size > ((sk->sk_sndbuf >> 1) - 64))
-                       size = (sk->sk_sndbuf >> 1) - 64;
+               size = min_t(int, size, (sk->sk_sndbuf >> 1) - 64);
 
-               if (size > SKB_MAX_ALLOC)
-                       size = SKB_MAX_ALLOC;
-
-               /*
-                *      Grab a buffer
-                */
+               /* allow fallback to order-0 allocations */
+               size = min_t(int, size, SKB_MAX_HEAD(0) + UNIX_SKB_FRAGS_SZ);
 
-               skb = sock_alloc_send_skb(sk, size, msg->msg_flags&MSG_DONTWAIT,
-                                         &err);
+               data_len = max_t(int, 0, size - SKB_MAX_HEAD(0));
 
-               if (skb == NULL)
+               skb = sock_alloc_send_pskb(sk, size - data_len, data_len,
+                                          msg->msg_flags & MSG_DONTWAIT, &err);
+               if (!skb)
                        goto out_err;
 
-               /*
-                *      If you pass two values to the sock_alloc_send_skb
-                *      it tries to grab the large buffer with GFP_NOFS
-                *      (which can fail easily), and if it fails grab the
-                *      fallback size buffer which is under a page and will
-                *      succeed. [Alan]
-                */
-               size = min_t(int, size, skb_tailroom(skb));
-
-
                /* Only send the fds in the first buffer */
                err = unix_scm_to_skb(siocb->scm, skb, !fds_sent);
                if (err < 0) {
@@ -1678,7 +1664,10 @@ static int unix_stream_sendmsg(struct kiocb *kiocb, struct socket *sock,
                max_level = err + 1;
                fds_sent = true;
 
-               err = memcpy_fromiovec(skb_put(skb, size), msg->msg_iov, size);
+               skb_put(skb, size - data_len);
+               skb->data_len = data_len;
+               skb->len = size;
+               err = skb_copy_datagram_from_iovec(skb, 0, msg->msg_iov, 0, size);
                if (err) {
                        kfree_skb(skb);
                        goto out_err;
@@ -1890,6 +1879,11 @@ static long unix_stream_data_wait(struct sock *sk, long timeo,
        return timeo;
 }
 
+static unsigned int unix_skb_len(const struct sk_buff *skb)
+{
+       return skb->len - UNIXCB(skb).consumed;
+}
+
 static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
                               struct msghdr *msg, size_t size,
                               int flags)
@@ -1977,8 +1971,8 @@ again:
                }
 
                skip = sk_peek_offset(sk, flags);
-               while (skip >= skb->len) {
-                       skip -= skb->len;
+               while (skip >= unix_skb_len(skb)) {
+                       skip -= unix_skb_len(skb);
                        last = skb;
                        skb = skb_peek_next(skb, &sk->sk_receive_queue);
                        if (!skb)
@@ -2005,8 +1999,9 @@ again:
                        sunaddr = NULL;
                }
 
-               chunk = min_t(unsigned int, skb->len - skip, size);
-               if (memcpy_toiovec(msg->msg_iov, skb->data + skip, chunk)) {
+               chunk = min_t(unsigned int, unix_skb_len(skb) - skip, size);
+               if (skb_copy_datagram_iovec(skb, UNIXCB(skb).consumed + skip,
+                                           msg->msg_iov, chunk)) {
                        if (copied == 0)
                                copied = -EFAULT;
                        break;
@@ -2016,14 +2011,14 @@ again:
 
                /* Mark read part of skb as used */
                if (!(flags & MSG_PEEK)) {
-                       skb_pull(skb, chunk);
+                       UNIXCB(skb).consumed += chunk;
 
                        sk_peek_offset_bwd(sk, chunk);
 
                        if (UNIXCB(skb).fp)
                                unix_detach_fds(siocb->scm, skb);
 
-                       if (skb->len)
+                       if (unix_skb_len(skb))
                                break;
 
                        skb_unlink(skb, &sk->sk_receive_queue);
@@ -2107,7 +2102,7 @@ long unix_inq_len(struct sock *sk)
        if (sk->sk_type == SOCK_STREAM ||
            sk->sk_type == SOCK_SEQPACKET) {
                skb_queue_walk(&sk->sk_receive_queue, skb)
-                       amount += skb->len;
+                       amount += unix_skb_len(skb);
        } else {
                skb = skb_peek(&sk->sk_receive_queue);
                if (skb)