Xen devel - Mar 2012 - libxc: Report a better error on state file EOF.

If a EOF is detected in state file 4 error are reported
 - 0-length
 - read_exact_timed failed
 - Error when reading batch size (0 = Success)
 - Error when reading batch (0 = Success)

With this patch just one error is reported
 - Error when reading batch size (32 = Broken pipe)

Signed-off-by: Frediano Ziglio <frediano.ziglio@citrix.com>

# HG changeset patch
# Parent 4bd752a4cdf323c41c50f8cd6286f566d67adeae
diff -r 4bd752a4cdf3 -r beb5ec839f95 tools/libxc/xc_domain_restore.c
--- a/tools/libxc/xc_domain_restore.c	Fri Mar 23 20:51:48 2012 +0000
+++ b/tools/libxc/xc_domain_restore.c	Tue Mar 27 15:13:10 2012 +0100
@@ -86,8 +86,8 @@ static ssize_t rdexact(xc_interface *xch
         if ( (len == -1) && ((errno == EINTR) || (errno == EAGAIN)) )
             continue;
         if ( len == 0 ) {
-            ERROR("0-length read");
-            errno = 0;
+            errno = EPIPE;
+            return -1;
         }
         if ( len <= 0 ) {
             ERROR("read_exact_timed failed (read rc: %d, errno: %d)",
len, errno);
@@ -1479,7 +1479,7 @@ int xc_domain_restore(xc_interface *xch,
             pagebuf.nr_physpages = pagebuf.nr_pages = 0;
             pagebuf.compbuf_pos = pagebuf.compbuf_size = 0;
             if ( pagebuf_get_one(xch, ctx, &pagebuf, io_fd, dom) < 0 ) {
-                PERROR("Error when reading batch");
+                /* pagebuf_get_one already returned a proper error */
                 goto out;
             }
         }


Regards
  Frediano Ziglio



_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xen.org
http://lists.xen.org/xen-devel

Frediano Ziglio writes ("[Xen-devel] libxc: Report a better error on state
file EOF."):
> If a EOF is detected in state file 4 error are reported
>  - 0-length
>  - read_exact_timed failed
>  - Error when reading batch size (0 = Success)
>  - Error when reading batch (0 = Success)
> 
> With this patch just one error is reported
>  - Error when reading batch size (32 = Broken pipe)
I approve of the idea of improving the error messages.  But I''m afraid
I must quibble with the reuse of EPIPE this way.  In general it is not
a good idea to synthesise errno values; people expect them to mean
specific things.  In this case EPIPE is an error that only the writer
on a pipe sees, so an admin who sees "Broken pipe" will expect that
the problem was that the writer detected that the reader went away.

I think a better approach would be to state explicitly (in
documentation comments) which situations cause an error report and
which don''t, and simply reduce duplication.

Thanks,
Ian.