corruption on mbox file (linux debian)

Depending of the client, the bug could appear every day. I know this is a 
old bug, but I would appreciate to know more on this subject

The observation :
1. mbox is corrupted, first line (or more) is overwritten where first bytes 
are
    027 003 001 000 150 142 072 157 235 351 341 361 302 024 322 264

2/ there is always such messages in /var/log/mail.info
        /usr/sbin/mlock[XXXX]: (64) fd not regular file
        imapd[29107]: Killed (lost mailbox lock) user=XXX
    (*)seems strange that fd is always 64
3/ when there is only one line like this one, no corruption
        imapd[29107]: Killed (lost mailbox lock) user=XXX

My current conclusions are:

1/ mlock die (/usr/sbin/mlock) it return ? and log a message to syslog

2/ this code seems not ok in src/osdep/unix

 int unix_lock (char *file,int flags,int mode,DOTLOCK *lock,int op)
{
    int fd;
    blocknotify_t bn = (blocknotify_t) mail_parameters 
(NIL,GET_BLOCKNOTIFY,NIL);
    (*bn) (BLOCK_FILELOCK,NIL);
                                /* try locking the easy way */
    if (dotlock_lock (file,lock,-1)) {
                                /* got dotlock file, easy open */
        if ((fd = open (file,flags,mode)) >= 0) flock (fd,op);
        else dotlock_unlock (lock); /* open failed, free the dotlock */
    }
                                /* no dot lock file, open file now */
    else if ((fd = open (file,flags,mode)) >= 0) {
                                /* try paranoid way to make a dot lock file 
*/
        if (dotlock_lock (file,lock,fd)) {
              close (fd);               /* get fresh fd in case of timing 
race */
              if ((fd = open (file,flags,mode)) >= 0) flock (fd,op);
                                /* open failed, free the dotlock */
              else dotlock_unlock (lock);
        }
        else flock (fd,op);         /* paranoid way failed, just flock() it 
*/
    }
    (*bn) (BLOCK_NONE,NIL);
    return fd;
}

dotlock_lock (file,lock,fd) fails because fd not a regular file. It could
potentially fails for other reasons but never in my server. So my proposal
would be :
    else {
        close (fd);
        fd = -1;
    }
in place of
        else flock (fd,op);         /* paranoid way failed, just flock() it 
*/

Any comment would be appreciated.

Michel 

_______________________________________________
Imap-uw mailing list
[email protected]
http://mailman2.u.washington.edu/mailman/listinfo/imap-uw