Merge remote-tracking branch 'bonzini/migration-writev' into staging

# By Paolo Bonzini
# Via Paolo Bonzini
* bonzini/migration-writev:
  qemu-file: do not use stdio for qemu_fdopen
  iov: handle partial writes from sendmsg and recvmsg
  iov: reorganize iov_send_recv, part 3
  iov: reorganize iov_send_recv, part 2
  iov: reorganize iov_send_recv, part 1
  qemu-file: drop socket_put_buffer

Message-id: 1366192012-14872-1-git-send-email-pbonzini@redhat.com
Signed-off-by: Anthony Liguori <aliguori@us.ibm.com>

2 files changed, 149 insertions, 69 deletions

diff --git a/savevm.c b/savevm.c
index 53515cb794..31dcce975e 100644
--- a/savevm.c
+++ b/savevm.c
@@ -219,18 +219,6 @@ static int socket_get_buffer(void *opaque, uint8_t *buf, int64_t pos, int size)
     return len;
 }
 
-static int socket_put_buffer(void *opaque, const uint8_t *buf, int64_t pos, int size)
-{
-    QEMUFileSocket *s = opaque;
-    ssize_t len;
-
-    len = qemu_send_full(s->fd, buf, size, 0);
-    if (len < size) {
-        len = -socket_error();
-    }
-    return len;
-}
-
 static int socket_close(void *opaque)
 {
     QEMUFileSocket *s = opaque;
@@ -368,9 +356,94 @@ static const QEMUFileOps stdio_file_write_ops = {
     .close =      stdio_fclose
 };
 
+static ssize_t unix_writev_buffer(void *opaque, struct iovec *iov, int iovcnt,
+                                  int64_t pos)
+{
+    QEMUFileSocket *s = opaque;
+    ssize_t len, offset;
+    ssize_t size = iov_size(iov, iovcnt);
+    ssize_t total = 0;
+
+    assert(iovcnt > 0);
+    offset = 0;
+    while (size > 0) {
+        /* Find the next start position; skip all full-sized vector elements  */
+        while (offset >= iov[0].iov_len) {
+            offset -= iov[0].iov_len;
+            iov++, iovcnt--;
+        }
+
+        /* skip `offset' bytes from the (now) first element, undo it on exit */
+        assert(iovcnt > 0);
+        iov[0].iov_base += offset;
+        iov[0].iov_len -= offset;
+
+        do {
+            len = writev(s->fd, iov, iovcnt);
+        } while (len == -1 && errno == EINTR);
+        if (len == -1) {
+            return -errno;
+        }
+
+        /* Undo the changes above */
+        iov[0].iov_base -= offset;
+        iov[0].iov_len += offset;
+
+        /* Prepare for the next iteration */
+        offset += len;
+        total += len;
+        size -= len;
+    }
+
+    return total;
+}
+
+static int unix_get_buffer(void *opaque, uint8_t *buf, int64_t pos, int size)
+{
+    QEMUFileSocket *s = opaque;
+    ssize_t len;
+
+    for (;;) {
+        len = read(s->fd, buf, size);
+        if (len != -1) {
+            break;
+        }
+        if (errno == EAGAIN) {
+            yield_until_fd_readable(s->fd);
+        } else if (errno != EINTR) {
+            break;
+        }
+    }
+
+    if (len == -1) {
+        len = -errno;
+    }
+    return len;
+}
+
+static int unix_close(void *opaque)
+{
+    QEMUFileSocket *s = opaque;
+    close(s->fd);
+    g_free(s);
+    return 0;
+}
+
+static const QEMUFileOps unix_read_ops = {
+    .get_fd =     socket_get_fd,
+    .get_buffer = unix_get_buffer,
+    .close =      unix_close
+};
+
+static const QEMUFileOps unix_write_ops = {
+    .get_fd =     socket_get_fd,
+    .writev_buffer = unix_writev_buffer,
+    .close =      unix_close
+};
+
 QEMUFile *qemu_fdopen(int fd, const char *mode)
 {
-    QEMUFileStdio *s;
+    QEMUFileSocket *s;
 
     if (mode == NULL ||
 	(mode[0] != 'r' && mode[0] != 'w') ||
@@ -379,21 +452,15 @@ QEMUFile *qemu_fdopen(int fd, const char *mode)
         return NULL;
     }
 
-    s = g_malloc0(sizeof(QEMUFileStdio));
-    s->stdio_file = fdopen(fd, mode);
-    if (!s->stdio_file)
-        goto fail;
+    s = g_malloc0(sizeof(QEMUFileSocket));
+    s->fd = fd;
 
     if(mode[0] == 'r') {
-        s->file = qemu_fopen_ops(s, &stdio_file_read_ops);
+        s->file = qemu_fopen_ops(s, &unix_read_ops);
     } else {
-        s->file = qemu_fopen_ops(s, &stdio_file_write_ops);
+        s->file = qemu_fopen_ops(s, &unix_write_ops);
     }
     return s->file;
-
-fail:
-    g_free(s);
-    return NULL;
 }
 
 static const QEMUFileOps socket_read_ops = {
@@ -404,7 +471,6 @@ static const QEMUFileOps socket_read_ops = {
 
 static const QEMUFileOps socket_write_ops = {
     .get_fd =     socket_get_fd,
-    .put_buffer = socket_put_buffer,
     .writev_buffer = socket_writev_buffer,
     .close =      socket_close
 };
diff --git a/util/iov.c b/util/iov.c
index 9dae318197..d32226d644 100644
--- a/util/iov.c
+++ b/util/iov.c
@@ -144,57 +144,71 @@ ssize_t iov_send_recv(int sockfd, struct iovec *iov, unsigned iov_cnt,
                       size_t offset, size_t bytes,
                       bool do_send)
 {
+    ssize_t total = 0;
     ssize_t ret;
-    unsigned si, ei;            /* start and end indexes */
-    if (bytes == 0) {
-        /* Catch the do-nothing case early, as otherwise we will pass an
-         * empty iovec to sendmsg/recvmsg(), and not all implementations
-         * accept this.
-         */
-        return 0;
-    }
+    size_t orig_len, tail;
+    unsigned niov;
+
+    while (bytes > 0) {
+        /* Find the start position, skipping `offset' bytes:
+         * first, skip all full-sized vector elements, */
+        for (niov = 0; niov < iov_cnt && offset >= iov[niov].iov_len; ++niov) {
+            offset -= iov[niov].iov_len;
+        }
 
-    /* Find the start position, skipping `offset' bytes:
-     * first, skip all full-sized vector elements, */
-    for (si = 0; si < iov_cnt && offset >= iov[si].iov_len; ++si) {
-        offset -= iov[si].iov_len;
-    }
-    if (offset) {
-        assert(si < iov_cnt);
-        /* second, skip `offset' bytes from the (now) first element,
-         * undo it on exit */
-        iov[si].iov_base += offset;
-        iov[si].iov_len -= offset;
-    }
-    /* Find the end position skipping `bytes' bytes: */
-    /* first, skip all full-sized elements */
-    for (ei = si; ei < iov_cnt && iov[ei].iov_len <= bytes; ++ei) {
-        bytes -= iov[ei].iov_len;
-    }
-    if (bytes) {
-        /* second, fixup the last element, and remember
-         * the length we've cut from the end of it in `bytes' */
-        size_t tail;
-        assert(ei < iov_cnt);
-        assert(iov[ei].iov_len > bytes);
-        tail = iov[ei].iov_len - bytes;
-        iov[ei].iov_len = bytes;
-        bytes = tail;  /* bytes is now equal to the tail size */
-        ++ei;
-    }
+        /* niov == iov_cnt would only be valid if bytes == 0, which
+         * we already ruled out in the loop condition.  */
+        assert(niov < iov_cnt);
+        iov += niov;
+        iov_cnt -= niov;
+
+        if (offset) {
+            /* second, skip `offset' bytes from the (now) first element,
+             * undo it on exit */
+            iov[0].iov_base += offset;
+            iov[0].iov_len -= offset;
+        }
+        /* Find the end position skipping `bytes' bytes: */
+        /* first, skip all full-sized elements */
+        tail = bytes;
+        for (niov = 0; niov < iov_cnt && iov[niov].iov_len <= tail; ++niov) {
+            tail -= iov[niov].iov_len;
+        }
+        if (tail) {
+            /* second, fixup the last element, and remember the original
+             * length */
+            assert(niov < iov_cnt);
+            assert(iov[niov].iov_len > tail);
+            orig_len = iov[niov].iov_len;
+            iov[niov++].iov_len = tail;
+        }
 
-    ret = do_send_recv(sockfd, iov + si, ei - si, do_send);
+        ret = do_send_recv(sockfd, iov, niov, do_send);
 
-    /* Undo the changes above */
-    if (offset) {
-        iov[si].iov_base -= offset;
-        iov[si].iov_len += offset;
-    }
-    if (bytes) {
-        iov[ei-1].iov_len += bytes;
+        /* Undo the changes above before checking for errors */
+        if (tail) {
+            iov[niov-1].iov_len = orig_len;
+        }
+        if (offset) {
+            iov[0].iov_base -= offset;
+            iov[0].iov_len += offset;
+        }
+
+        if (ret < 0) {
+            assert(errno != EINTR);
+            if (errno == EAGAIN && total > 0) {
+                return total;
+            }
+            return -1;
+        }
+
+        /* Prepare for the next iteration */
+        offset += ret;
+        total += ret;
+        bytes -= ret;
     }
 
-    return ret;
+    return total;
 }