I came up with a new way of avoiding the error handling lockup bug in

rsync. It isn't pretty, but it does work and should completely avoid
that class of lockup.

2 files changed, 68 insertions, 10 deletions

diff --git a/io.c b/io.c
index 894e05a8..649bd758 100644
--- a/io.c
+++ b/io.c
@@ -48,6 +48,8 @@ static void check_timeout(void)
 {
 	extern int am_server, am_daemon;
 	time_t t;
+
+	err_list_push();
 	
 	if (!io_timeout) return;
 
@@ -339,6 +341,8 @@ static void writefd_unbuffered(int fd,char *buf,int len)
 	int fd_count, count;
 	struct timeval tv;
 
+	err_list_push();
+
 	no_flush++;
 
 	while (total < len) {
@@ -462,6 +466,9 @@ static void mplex_write(int fd, enum logcode code, char *buf, int len)
 void io_flush(void)
 {
 	int fd = multiplex_out_fd;
+
+	err_list_push();
+
 	if (!io_buffer_count || no_flush) return;
 
 	if (io_multiplexing_out) {
@@ -487,6 +494,7 @@ void io_end_buffering(int fd)
    a socket to be flushed. Do an explicit shutdown to try to prevent this */
 void io_shutdown(void)
 {
+	err_list_push();
 	if (multiplex_out_fd != -1) close(multiplex_out_fd);
 	if (io_error_fd != -1) close(io_error_fd);
 	multiplex_out_fd = -1;
@@ -498,6 +506,8 @@ static void writefd(int fd,char *buf,int len)
 {
 	stats.total_written += len;
 
+	err_list_push();
+
 	if (!io_buffer || fd != multiplex_out_fd) {
 		writefd_unbuffered(fd, buf, len);
 		return;
@@ -635,14 +645,6 @@ int io_multiplex_write(enum logcode code, char *buf, int len)
 	return 1;
 }
 
-/* write a message to the special error fd */
-int io_error_write(int f, enum logcode code, char *buf, int len)
-{
-	if (f == -1) return 0;
-	mplex_write(f, code, buf, len);
-	return 1;
-}
-
 /* stop output multiplexing */
 void io_multiplexing_close(void)
 {
diff --git a/log.c b/log.c
index aef3ac8a..a826b185 100644
--- a/log.c
+++ b/log.c
@@ -67,6 +67,59 @@ static char const *rerr_name(int code)
         return NULL;
 }
 
+struct err_list {
+	struct err_list *next;
+	char *buf;
+	int len;
+	int written; /* how many bytes we have written so far */
+};
+
+static struct err_list *err_list_head;
+static struct err_list *err_list_tail;
+
+/* add an error message to the pending error list */
+static void err_list_add(int code, char *buf, int len)
+{
+	struct err_list *el;
+	el = (struct err_list *)malloc(sizeof(*el));
+	if (!el) exit_cleanup(RERR_MALLOC);
+	el->next = NULL;
+	el->buf = malloc(len+4);
+	if (!el->buf) exit_cleanup(RERR_MALLOC);
+	memcpy(el->buf+4, buf, len);
+	SIVAL(el->buf, 0, ((code+MPLEX_BASE)<<24) | len);
+	el->len = len+4;
+	el->written = 0;
+	if (err_list_tail) {
+		err_list_tail->next = el;
+	} else {
+		err_list_head = el;
+	}
+	err_list_tail = el;
+}
+
+
+/* try to push errors off the error list onto the wire */
+void err_list_push(void)
+{
+	if (log_error_fd == -1) return;
+
+	while (err_list_head) {
+		struct err_list *el = err_list_head;
+		int n = write(log_error_fd, el->buf+el->written, el->len - el->written);
+		if (n == -1) break;
+		if (n > 0) {
+			el->written += n;
+		}
+		if (el->written == el->len) {
+			free(el->buf);
+			err_list_head = el->next;
+			if (!err_list_head) err_list_tail = NULL;
+			free(el);
+		}
+	}
+}
+
 
 static void logit(int priority, char *buf)
 {
@@ -144,6 +197,7 @@ void log_close()
 void set_error_fd(int fd)
 {
 	log_error_fd = fd;
+	set_nonblocking(log_error_fd);
 }
 
 /* this is the underlying (unformatted) rsync debugging function. Call
@@ -167,8 +221,10 @@ void rwrite(enum logcode code, char *buf, int len)
 		return;
 	}
 
-	/* first try to pass it off the our sibling */
-	if (am_server && io_error_write(log_error_fd, code, buf, len)) {
+	/* first try to pass it off to our sibling */
+	if (am_server && log_error_fd != -1) {
+		err_list_add(code, buf, len);
+		err_list_push();
 		return;
 	}