libglusterfsclient: implement glusterfs_sendfile.

Signed-off-by: Anand V. Avati <avati@dev.gluster.com>

2 files changed, 171 insertions, 0 deletions

diff --git a/libglusterfsclient/src/libglusterfsclient.c b/libglusterfsclient/src/libglusterfsclient.c
index 8dc63a6eb..bfde21c61 100755
--- a/libglusterfsclient/src/libglusterfsclient.c
+++ b/libglusterfsclient/src/libglusterfsclient.c
@@ -53,6 +53,7 @@
 
 #define LIBGF_XL_NAME "libglusterfsclient"
 #define LIBGLUSTERFS_INODE_TABLE_LRU_LIMIT 1000 //14057
+#define LIBGF_SENDFILE_BLOCK_SIZE 4096
 #define LIBGF_READDIR_BLOCK     4096
 
 static inline xlator_t *
@@ -6683,6 +6684,150 @@ out:
         return off;
 }
 
+
+struct libgf_client_sendfile_data {
+        int                reads_sent;
+        int                reads_completed;
+        int                out_fd;
+        int32_t            op_ret;
+        int32_t            op_errno;
+        pthread_mutex_t    lock;
+        pthread_cond_t     cond;
+};
+
+int
+libgf_client_sendfile_read_cbk (int op_ret, int op_errno,
+                                glusterfs_iobuf_t *buf, void *cbk_data)
+{
+        struct libgf_client_sendfile_data *sendfile_data = cbk_data;
+        int                                bytes = 0;
+
+        if (op_ret > 0) {
+                bytes = writev (sendfile_data->out_fd, buf->vector, buf->count);
+                if (bytes != op_ret) {
+                        op_ret = -1;
+                        op_errno = errno;
+                }
+
+                glusterfs_free (buf);
+        }
+
+        pthread_mutex_lock (&sendfile_data->lock);
+        {
+                if (sendfile_data->op_ret != -1) {
+                        if (op_ret == -1) {
+                                sendfile_data->op_ret = -1;
+                                sendfile_data->op_errno = op_errno;
+                        } else {
+                                sendfile_data->op_ret += op_ret;
+                        }
+                }
+
+                sendfile_data->reads_completed++;
+
+                if (sendfile_data->reads_completed
+                    == sendfile_data->reads_sent) {
+                        pthread_cond_broadcast (&sendfile_data->cond);
+                } 
+        }
+        pthread_mutex_unlock (&sendfile_data->lock);
+
+        return 0;
+}
+
+
+ssize_t
+glusterfs_sendfile (int out_fd, glusterfs_file_t in_fd, off_t *offset,
+                    size_t count)
+{
+        ssize_t                           ret = -1;
+        struct libgf_client_sendfile_data cbk_data = {0, };
+        off_t                             off = -1;
+        size_t                            size = 0;
+        int                               flags = 0;
+        int                               non_block = 0;
+
+        
+        pthread_mutex_init (&cbk_data.lock, NULL);
+        pthread_cond_init (&cbk_data.cond, NULL);
+        cbk_data.out_fd = out_fd;
+        
+        if (offset) {
+                off = *offset;
+        }
+
+        flags = fcntl (out_fd, F_GETFL);
+
+        if (flags != -1) {
+                non_block = flags & O_NONBLOCK;
+
+                if (non_block) {
+                        ret = fcntl (out_fd, F_SETFL, flags & ~O_NONBLOCK);
+                }
+        }
+
+        while (count != 0) {
+                /* 
+                 * FIXME: what's the optimal size for reads and writes?
+                 */
+                size = (count > LIBGF_SENDFILE_BLOCK_SIZE) ? 
+                        LIBGF_SENDFILE_BLOCK_SIZE : count;
+
+                /* 
+                 * we don't wait for reply to previous read, we just send all
+                 * reads in a single go.
+                 */
+                ret = glusterfs_read_async (in_fd, size, off,
+                                            libgf_client_sendfile_read_cbk,
+                                            &cbk_data);
+                if (ret == -1) {
+                        break;
+                }
+
+                pthread_mutex_lock (&cbk_data.lock);
+                {
+                        cbk_data.reads_sent++;
+                }
+                pthread_mutex_unlock (&cbk_data.lock);
+
+                if (offset) {
+                        off += size;
+                }
+
+                count -= size;
+        }
+
+        pthread_mutex_lock (&cbk_data.lock);
+        {
+                /* 
+                 * if we've not received replies to all the reads we've sent,
+                 * wait for them
+                 */
+                if (cbk_data.reads_sent > cbk_data.reads_completed) {
+                        pthread_cond_wait (&cbk_data.cond,
+                                           &cbk_data.lock);
+                }
+        }
+        pthread_mutex_unlock (&cbk_data.lock);
+
+        if (offset != NULL) {
+                *offset = off;
+        }
+ 
+        /* if we were able to stack_wind all the reads */
+
+        if (ret == 0) {
+                ret = cbk_data.op_ret;
+                errno = cbk_data.op_errno;
+        }
+
+        if (non_block) {
+                fcntl (out_fd, F_SETFL, flags);
+        }
+
+        return ret;
+}
+
 static struct xlator_fops libgf_client_fops = {
 };
 
diff --git a/libglusterfsclient/src/libglusterfsclient.h b/libglusterfsclient/src/libglusterfsclient.h
index 9e03a8c59..e8b19f4bc 100755
--- a/libglusterfsclient/src/libglusterfsclient.h
+++ b/libglusterfsclient/src/libglusterfsclient.h
@@ -1234,6 +1234,32 @@ glusterfs_seekdir (glusterfs_dir_t dirfd, off_t offset);
  */
 off_t
 glusterfs_telldir (glusterfs_dir_t dirfd);
+
+
+/* Write count bytes from in_fd to out_fd, starting at *offset.
+ * glusterfs_sendfile aims at eliminating memory copy at the end of
+ * each read from in_fd, copying the file directly to out_fd from the buffer 
+ * provided by glusterfs.
+ *
+ * @out_fd: file descriptor opened for writing
+ *
+ * @in_fd: glusterfs file handle to the file to be read from.
+ *
+ * @offset: If offset is not NULL, then it points to a variable holding the file
+ *          offset  from  which  glusterfs_sendfile()  will  start reading data
+ *          from in_fd.  When glusterfs_sendfile() returns, this variable will 
+ *          be set to the offset of the byte following the last byte that was 
+ *          read.  If offset is  not  NULL, then glusterfs_sendfile()  does  not
+ *          modify the current file offset of in_fd; otherwise the current file
+ *          offset is adjusted to reflect the number of bytes read from in_fd.
+ *
+ * @count:  number of bytes to copy between the file descriptors.
+ */
+
+ssize_t
+glusterfs_sendfile (int out_fd, glusterfs_file_t in_fd, off_t *offset,
+                    size_t count);
+
 /* FIXME: review the need for these apis */
 /* added for log related initialization in booster fork implementation */
 void