[Qemu-devel] [PATCH 3/3] net: EAGAIN handling for net/socket.c TCP

[Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>]

Replace spinning send_all() with a proper non-blocking send.  When the
socket write buffer limit is reached, we should stop trying to send and
wait for the socket to become writable again.

Non-blocking TCP sockets can return in two different ways when the write
buffer limit is reached:

1. ret = -1 and errno = EAGAIN/EWOULDBLOCK.  No data has been written.

2. ret < total_size.  Short write, only part of the message was
   transmitted.

Handle both cases and keep track of how many bytes have been written in
s->send_index.  (This includes the 'length' header before the actual
payload buffer.)

Signed-off-by: Stefan Hajnoczi <stefanha@linux.vnet.ibm.com>
---
 net/socket.c |   50 +++++++++++++++++++++++++++++++++++++++++++++-----
 1 file changed, 45 insertions(+), 5 deletions(-)

diff --git a/net/socket.c b/net/socket.c
index e5e4e8d..be44105 100644
--- a/net/socket.c
+++ b/net/socket.c
@@ -40,6 +40,7 @@ typedef struct NetSocketState {
     int state; /* 0 = getting length, 1 = getting data */
     unsigned int index;
     unsigned int packet_len;
+    unsigned int send_index;      /* number of bytes sent (only SOCK_STREAM) */
     uint8_t buf[4096];
     struct sockaddr_in dgram_dst; /* contains inet host and port destination iff connectionless (SOCK_DGRAM) */
     IOHandler *send_fn;           /* differs between SOCK_STREAM/SOCK_DGRAM */
@@ -88,15 +89,54 @@ static void net_socket_writable(void *opaque)
     qemu_flush_queued_packets(&s->nc);
 }
 
-/* XXX: we consider we can send the whole packet without blocking */
 static ssize_t net_socket_receive(NetClientState *nc, const uint8_t *buf, size_t size)
 {
     NetSocketState *s = DO_UPCAST(NetSocketState, nc, nc);
-    uint32_t len;
-    len = htonl(size);
+    struct iovec iov[2];
+    int iovcnt = 0;
+    ssize_t ret;
+    ssize_t total_size = 0;
+    uint32_t len = htonl(size);
+    unsigned int buf_index;
+
+    /* Length header */
+    if (s->send_index < sizeof(len)) {
+        iov[iovcnt].iov_base = (uint8_t *)&len + s->send_index;
+        iov[iovcnt].iov_len  = sizeof(len) - s->send_index;
+        total_size += iov[iovcnt].iov_len;
+        iovcnt++;
+
+        buf_index = 0;
+    } else {
+        buf_index = s->send_index - sizeof(len);
+    }
+
+    assert(buf_index < size);
+
+    /* Payload buffer */
+    iov[iovcnt].iov_base = (uint8_t *)buf + buf_index;
+    iov[iovcnt].iov_len  = size - buf_index;
+    total_size += iov[iovcnt].iov_len;
+    iovcnt++;
+
+    do {
+        ret = writev(s->fd, iov, iovcnt);
+    } while (ret == -1 && ret == EINTR);
 
-    send_all(s->fd, (const uint8_t *)&len, sizeof(len));
-    return send_all(s->fd, buf, size);
+    if (ret == -1 && errno == EAGAIN) {
+        ret = 0; /* handled further down */
+    }
+    if (ret == -1) {
+        s->send_index = 0;
+        return -errno;
+    }
+    if (ret < total_size) {
+        s->send_index += ret;
+        net_socket_write_poll(s, true);
+        return 0;
+    }
+    s->send_index = 0;
+    return size;
 }
 
 static ssize_t net_socket_receive_dgram(NetClientState *nc, const uint8_t *buf, size_t size)
-- 
1.7.10.4