[Xen-tools] Re: xenstored crash

[Rusty Russell <rusty@xxxxxxxxxxxxxxx>]

On Tue, 2005-09-13 at 22:12 +0100, Christian Limpach wrote:
> Rusty,
> 
> could you take a look at the following crash:
> (gdb) run -N --pid-file=/var/run/xenstore.pid -T /var/log/xenstored-trace.log
> Starting program: /usr/sbin/xenstored -N --pid-file=/var/run/xenstore.pid -T 
> /var/log/xenstored-trace.log
> xenstored: xenstored_core.c:512: send_reply: Assertion `conn->state != 
> BLOCKED' failed.

And here's the one-line fix, plus test case.

Rusty.

# HG changeset patch
# User Rusty Russell <rusty@xxxxxxxxxxxxxxx>
# Node ID 7882eb8eec92a90929904a9daa7857a2ce44f294
# Parent  0d8c0db042580571be501be5b59cb5beb8417cef
Fix Christian's xenstored watch crash.

When a connection blocked waiting on a transaction, don't queue watch events.
Sure, they'd be ignored and re-transmitted, but it hits an assert that we don't
send data out blocked connections, and it's wasteful.

Signed-off-by: Rusty Russell <rusty@xxxxxxxxxxxxxxx> (authored)

diff -r 0d8c0db04258 -r 7882eb8eec92 tools/xenstore/xenstored_watch.c
--- a/tools/xenstore/xenstored_watch.c  Tue Sep 13 21:52:24 2005
+++ b/tools/xenstore/xenstored_watch.c  Wed Sep 14 05:07:51 2005
@@ -144,7 +144,7 @@
                        else
                                continue;
                        /* If connection not doing anything, queue this. */
-                       if (!i->out)
+                       if (i->state == OK)
                                queue_next_event(i);
                }
        }
diff -r 0d8c0db04258 -r 7882eb8eec92 tools/xenstore/xs_test.c
--- a/tools/xenstore/xs_test.c  Tue Sep 13 21:52:24 2005
+++ b/tools/xenstore/xs_test.c  Wed Sep 14 05:07:51 2005
@@ -398,12 +398,16 @@
 static void do_readack(unsigned int handle)
 {
        enum xsd_sockmsg_type type;
-       char *ret;
-
-       ret = read_reply(handles[handle]->fd, &type, NULL);
-       if (!ret)
-               failed(handle);
-       free(ret);
+       char *ret = NULL;
+
+       /* Watches can have fired before reply comes: daemon detects
+        * and re-transmits, so we can ignore this. */
+       do {
+               free(ret);
+               ret = read_reply(handles[handle]->fd, &type, NULL);
+               if (!ret)
+                       failed(handle);
+       } while (type == XS_WATCH_EVENT);
 }
 
 static void do_setid(unsigned int handle, char *id)
diff -r 0d8c0db04258 -r 7882eb8eec92 
tools/xenstore/testsuite/16block-watch-crash.test
--- /dev/null   Tue Sep 13 21:52:24 2005
+++ b/tools/xenstore/testsuite/16block-watch-crash.test Wed Sep 14 05:07:51 2005
@@ -0,0 +1,13 @@
+# Test case where blocked connection gets sent watch.
+
+mkdir /test
+watch /test token
+1 start /test
+# This will block on above
+noackwrite /test/entry create contents
+1 write /test/entry2 create contents
+1 commit
+readack
+expect /test/entry2:token
+waitwatch
+ackwatch token

-- 
A bad analogy is like a leaky screwdriver -- Richard Braakman


_______________________________________________
Xen-tools mailing list
Xen-tools@xxxxxxxxxxxxxxxxxxx
http://lists.xensource.com/xen-tools