Re: [PATCH] printk: avoid prb_first_valid_seq() where possible

[Petr Mladek <pmladek@suse.com>]

On Wed 2021-02-10 19:32:10, John Ogness wrote:
> On 2021-02-09, Petr Mladek <pmladek@suse.com> wrote:
> >> @@ -1629,9 +1631,13 @@ int do_syslog(int type, char __user *buf, int len, int source)
> >>  	/* Number of chars in the log buffer */
> >>  	case SYSLOG_ACTION_SIZE_UNREAD:
> >>  		logbuf_lock_irq();
> >> -		if (syslog_seq < prb_first_valid_seq(prb)) {
> >> -			/* messages are gone, move to first one */
> >> -			syslog_seq = prb_first_valid_seq(prb);
> >> +		if (prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
> >> +			if (info.seq != syslog_seq) {
> >> +				/* messages are gone, move to first one */
> >> +				syslog_seq = info.seq;
> >> +				syslog_partial = 0;
> >> +			}
> >> +		} else {
> >>  			syslog_partial = 0;
> >
> > I am scratching my head when prb_read_valid_info(prb,
> > syslog_seq, &info, NULL)) might fail.
> 
> It can fail because the descriptor has been invalidated/recycled by
> writers and perhaps there is no valid record that has yet come after it.

I see. From some reasons I though that there should always be at
least one message in the commited state. But it is enough when
it is in reusable state. I should have double checked it.

> I recommend changing your suggestion to:
> 
> > 		if (!prb_read_valid_info(prb, syslog_seq, &info, NULL)) {
> >			/*
> >			 * No unread messages. No need to check/reset
> >			 * syslog_partial. When a reader does read a new
> >			 * message it will notice and appropriately update
> >			 * syslog_seq and reset syslog_partial.
> >			 */

The following comment might be enough after all.

			/* No unread messages. */

My main concern was that we cleared syslog_partial and continued.
I thought that we might miss a bug this way. But it seems to
be perfectly fine. I just have to update my mental picture.

Otherwise. the fact that syslog_partial will be fixed by the next
successful call is more or less obvious if we change the code as you
propose.

Please, send an updated patch.

Best Regards,
Petr


> > 			logbuf_unlock_irq();
> > 			return 0;
> > 		}
> > 		if (info.seq != syslog_seq) {
> > 			/* messages are gone, move to first one */
> > 			syslog_seq = info.seq;
> > 			syslog_partial = 0;
> > 		}
> 
> John Ogness