On Thu 2014-07-17 09:09:07, Alex Elder wrote:
> Two log record flags--LOG_CONT and LOG_NEWLINE--are mutually
> exclusive. That is, one or the other is always set, but they are
> never both set at the same time in a log record flags field. What
> follows is a great deal of explanation that aims to prove this
> assertion.
>
> Having that knowledge allows us to simplify a bit of logic, and with
> a little more work (in follow-on patches) it allows us to do without
> some flag values, considerably simplifying things.
>
>
> Log record flags are held in the "cont" continuation buffer, as well
> as in "syslog_prev" and "console_prev". Those are discussed later.
>
> Other than that, log record flags are only set in log_store():
> msg->flags = flags & 0x1f;
> There are 5 places log_store() is called: twice from cont_flush();
> and three times from vprintk_emit().
>
> Only two single-flag values are ever passed to cont_flush():
> LOG_CONT; and LOG_NEWLINE. That passed-in value is provided to
> log_store() either as-is, or modified to include LOG_NOCONS. There
> are thus four possible flag combinations supplied to log_store() by
> cont_flush(): LOG_CONT; LOG_NEWLINE; LOG_CONT|LOG_NOCONS; or
> LOG_NEWLINE|LOG_NOCONS.
>
> The first call vprintk_emit() makes to log_store() passes a flags
> value of LOG_PREFIX|LOG_NEWLINE. The second and third calls pass a
> locally-computed "lflags" value, possibly with LOG_CONT added. The
> only possible flag combinations held in "lflags" are: 0;
> LOG_NEWLINE; LOG_PREFIX; or LOG_NEWLINE|LOG_PREFIX. If LOG_NEWLINE
> is not set, LOG_CONT flag is added in the call to log_store(). So
> there are four possible flag combinations supplied by cont_flush():
> LOG_CONT; LOG_NEWLINE; LOG_PREFIX|LOG_CONT; or LOG_PREFIX|LOG_NEWLINE.
>
> Therefore log_store() is never provided (so never records) a flag
> value that contains both LOG_CONT and LOG_NEWLINE, and one of those
> flags is always present.
>
>
> Meanwhile, the "cont" flags field is only ever assigned value 0, or
> a value passed to cont_flush(). The value of cont.flags is only
> ever *used* if cont.flushed is true, and that only gets set after
> the cont.flags field is assigned to the value passed to
> cont_flush(). As mentioned above, cont_flush() is never provided
> more than one flag value, and it's always either LOG_NEWLINE
> or LOG_CONT. Therefore, for all intents and purposes, cont.flags
> only ever holds LOG_NEWLINE or LOG_CONT.
>
> The only values assigned to "syslog_prev" and "console_prev" are the
> initial value LOG_NEWLINE or the flags value from a log record. So
> none of these ever hold LOG_CONT and LOG_NEWLINE at the same time.
>
> This proves that at LOG_CONT and LOG_NEWLINE are in fact mutually
> exclusive flags.
>
> This patch makes the following changes:
> - Changes two spots in msg_print_txt() so they no longer bother
> checking for LOG_NEWLINE once its known that LOG_CONT is set.
> - Changes two |= assignments in vprintk_emit() to be simple
> assignments instead, because the result is known to be the
> same and it makes it obvious no other flags are involved.
> - Assigns LOG_CONT to "lflags" in vprintk_emit() if it does not
> contain LOG_NEWLINE (rather than just adding it at the time
> of the call to log_store()).
> - Adds a short explanatory comment.
> - Explicitly pass NULL rather than "dict" (and 0 as its length)
> to log_store() in one spot in vprintk_emit(), because we know
> the value of dict at that point is always NULL.
>
> Signed-off-by: Alex Elder <[email protected]>
> ---
> kernel/printk/printk.c | 23 +++++++++++++----------
> 1 file changed, 13 insertions(+), 10 deletions(-)
>
> diff --git a/kernel/printk/printk.c b/kernel/printk/printk.c
> index ff1c7ad..1ee1862 100644
> --- a/kernel/printk/printk.c
> +++ b/kernel/printk/printk.c
> @@ -1006,11 +1006,9 @@ static size_t msg_print_text(const struct printk_log
> *msg, enum log_flags prev,
> prefix = false;
>
> if (msg->flags & LOG_CONT) {
> - if ((prev & LOG_CONT) && !(prev & LOG_NEWLINE))
> + if (prev & LOG_CONT)
> prefix = false;
> -
> - if (!(msg->flags & LOG_NEWLINE))
> - newline = false;
> + newline = false;
> }
>
> do {
> @@ -1639,10 +1637,16 @@ asmlinkage int vprintk_emit(int facility, int level,
> text_len += vscnprintf(text + text_len,
> sizeof(textbuf) - text_len, fmt, args);
>
> - /* mark and strip a trailing newline */
> + /*
> + * If there's a trailing newline, flag it and strip it off.
> + * Otherwise we assume this is a partial log message, to be
> + * continued with the next call.
> + */
> if (text_len && text[text_len-1] == '\n') {
> text_len--;
> - lflags |= LOG_NEWLINE;
> + lflags = LOG_NEWLINE;
> + } else {
> + lflags = LOG_CONT;
> }
>
> /* strip kernel syslog prefix and extract log level or control flags */
> @@ -1672,7 +1676,7 @@ asmlinkage int vprintk_emit(int facility, int level,
> level = default_message_loglevel;
>
> if (dict)
> - lflags |= LOG_PREFIX|LOG_NEWLINE;
> + lflags = LOG_PREFIX|LOG_NEWLINE;
>
> if (!(lflags & LOG_NEWLINE)) {
> /*
All the above changes make sense and I support them after all.
> @@ -1686,9 +1690,8 @@ asmlinkage int vprintk_emit(int facility, int level,
> if (cont_add(facility, level, text, text_len))
> printed_len += text_len;
> else
> - printed_len += log_store(facility, level,
> - lflags | LOG_CONT, 0,
> - dict, dictlen, text, text_len);
> + printed_len += log_store(facility, level, lflags, 0,
> + NULL, 0, text, text_len);
> } else {
> bool stored = false;
>
I do not see real benefit in this change. I think that it would cause
more confusion than help. I would suggest to remove it. Then I will
agree with this patch :-)
Best Regards,
Petr
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/
