Failed assertion in detail reader on incomplete packet read

[John Morrissey]

We've had a few FreeRADIUS instances run out of space in /var due to
unresponsive remote accounting servers.

When this happens, a short packet can be written to a detail file, causing
this assertion failure:

ASSERT FAILED detail.c[573]: data->state == STATE_QUEUED

when the VP-reading loop hits EOF and falls through to the alloc_packet
label without having read a trailing newline indicating end of packet:

while (fgets(buffer, sizeof(buffer), data->fp)) {
[...]
    /*
     *  We're reading VP's, and got a blank line.
     *  Queue the packet.
     */
    if ((data->state == STATE_READING) &&
        (buffer[0] == '\n')) {
        data->state = STATE_QUEUED;
        break;
    }
[...]
}
[...]
 alloc_packet:
    rad_assert(data->state == STATE_QUEUED);

The writes in rlm_detail's do_detail() are unchecked, so failed writes allow
a successful Accounting-Response to be returned to the NAS even though the
detail is incomplete and won't be processed by the detail reader.

It seems the best way around this is for do_detail() to keep track of the
number of bytes it's written, ftruncate() the detail file on a short write,
and return RLM_MODULE_FAIL. fprintf() would need to be replaced with
separate sprintf() and fwrite() in order to keep track of the number of
bytes written for the current packet.

Does this course of action make sense? I can work on this, but it may be a
few weeks before I have more time.

john
-- 
John Morrissey          _o            /\         ----  __o
jwm at horde.net        _-< \_          /  \       ----  <  \,
www.horde.net/    __(_)/_(_)________/    \_______(_) /_(_)__