[Gnu-arch-users] [BUG] serious bug in vu_write_retry?

[Jeremy Shaw]

Hello,

Some people at work were attempting to call tla from perl using popen3
and complained of corrupt output when calling 'tla cat-archive-log' on
large log files (> 200 lines).

When calling cat-archive-log, it would start printing out the log,
then all of a sudden it would output a null followed by what appears
to be some random data in memory. After printing a bunch of random
data, it would then print the archive-log correctly, starting from the
beginning.

They said that the following patch fixed the problem for them (and
passes all existing unit tests):

--- orig/vu/vu.c
+++ mod/vu/vu.c
@@ -1241,8 +1241,11 @@
        return amt;
       if (amt == count)
        return orig_count;
-      count -= amt;
-      buf += amt;
+      if(amt > 0 )
+        {
+          count -= amt;
+          buf += amt;
+        }
     }
 }

Can anyone else comment on the validity of this bug/patch? In the
unpatched code, it looks like when vu_write returns -1 and (*errn ==
EAGAIN || *errn == EINTR), count will be incremented by 1 and bug will
be decremented by 1.

Jeremy Shaw.