Potential deadlock in psp cache

• Key: MODPYTHON-69
• Type: Bug
• Status: Closed
• Resolution: Fixed
• Priority: Major
• Assignee: Unassigned
• Reporter: Jim Gallacher
• Component/s: publisher
• Affects Version/s: 3.2.7
• Fix Version/s: 3.2.7

Description

This issue was discussed on the python-dev mailing list but not followed up on. Fixing that now.

In psp.py

def dbm_cache_store(srv, dbmfile, filename, mtime, val):

    dbm_type = dbm_cache_type(dbmfile)
    ### potential deadlock here! ###
    _apache._global_lock(srv, "pspcache")
    try:
        dbm = dbm_type.open(dbmfile, 'c')
        dbm[filename] = "%d %s" % (mtime, code2str(val))
    finally:
        try: dbm.close()
        except: pass
        _apache._global_unlock(srv, "pspcache")

"pspcache" will hash to one of 31 mutexes. Therefore there is a 1 in 31 chance for a hash collision if a session is used in the same request, which would result in a deadlock. (This has been confirmed by testing.)

Most obvious solution is to use the global lock 0, which will serialize all accesses to either pspcache.dbm. Global lock 0 is also used by DbmSession, but since the lock is not held for the duration of the request there should not be any additional deadlock issues.

The fix is to replace the _apache._global_lock(srv, "pspcache") with
_apache._global_lock(srv, None, 0)

The corresponding lock handling in dbm_cache_get() will also need the same fix.

I will commit the this fix shortly.