Re: [PATCH v4 6/7] printk: insert newline for truncated records

["Petr Mládek" <pmladek@suse.cz>]

On Fri 2014-07-18 16:28:04, Alex Elder wrote:
> If a log record has LOG_PREFIX set, its predecessor record should be
> terminated if it was marked LOG_CONT.
> 
> In devkmsg_read(), this condition was being ignored, which would
> lead to such records showing up combined when reading /dev/kmsg.
> Fix this oversight.
> 
> We should similarly insert a newline in msg_print_text() in this
> case, to avoid formatted records getting merged.
> 
> Suggested-by: Petr Mládek <pmladek@suse.cz>
> Signed-off-by: Alex Elder <elder@linaro.org>
> ---
>  kernel/printk/printk.c | 18 +++++++++++++++---
>  1 file changed, 15 insertions(+), 3 deletions(-)
> 
> diff --git a/kernel/printk/printk.c b/kernel/printk/printk.c
> index e9f0632..a5ad81c 100644
> --- a/kernel/printk/printk.c
> +++ b/kernel/printk/printk.c
> @@ -575,6 +575,7 @@ static ssize_t devkmsg_read(struct file *file, char __user *buf,
>  	char cont;
>  	size_t len;
>  	ssize_t ret;
> +	bool insert_newline;
>  
>  	if (!user)
>  		return -EBADF;
> @@ -626,7 +627,10 @@ static ssize_t devkmsg_read(struct file *file, char __user *buf,
>  	else
>  		cont = '-';
>  
> -	len = sprintf(user->buf, "%u,%llu,%llu,%c;",
> +	/* Insert a newline if the previous line was not terminated properly */
> +	insert_newline = (user->prev & LOG_CONT) && (msg->flags & LOG_PREFIX);
> +	len = sprintf(user->buf, "%s%u,%llu,%llu,%c;",
> +		      insert_newline ? "\n" : "",
>  		      (msg->facility << 3) | msg->level,
>  		      user->seq, ts_usec, cont);
>  	user->prev = msg->flags;
> @@ -999,10 +1003,12 @@ static size_t msg_print_text(const struct printk_log *msg, enum log_flags prev,
>  {
>  	const char *text = log_text(msg);
>  	size_t text_size = msg->text_len;
> +	size_t len = 0;
> +	bool insert_newline;
>  	bool prefix = true;
>  	bool newline = true;
> -	size_t len = 0;
>  
> +	insert_newline = (prev & LOG_CONT) && (msg->flags & LOG_PREFIX);
>  	if ((prev & LOG_CONT) && !(msg->flags & LOG_PREFIX))
>  		prefix = false;
>  
> @@ -1023,9 +1029,13 @@ static size_t msg_print_text(const struct printk_log *msg, enum log_flags prev,
>  
>  		if (buf) {
>  			if (print_prefix(msg, syslog, NULL) +
> -			    text_len + 1 >= size - len)
> +			    text_len + 2 >= size - len)

It counts the '\n' even when it is not used.
I think that it is even wrong that it calculates prefix when it is not used.


>  				break;
>  
> +			if (insert_newline) {
> +				insert_newline = false;
> +				buf[len++] = '\n';
> +			}
>  			if (prefix)
>  				len += print_prefix(msg, syslog, buf + len);
>  			memcpy(buf + len, text, text_len);
> @@ -1034,6 +1044,8 @@ static size_t msg_print_text(const struct printk_log *msg, enum log_flags prev,
>  				buf[len++] = '\n';
>  		} else {
>  			/* SYSLOG_ACTION_* buffer size only calculation */
> +			if (insert_newline)
> +				len++;

You forgot "insert_newline = false" here.

>  			if (prefix)
>  				len += print_prefix(msg, syslog, NULL);
>  			len += text_len;

It is just matter of personal style but I would suggest to do this
before the do-while cycle:

	/* Force newline if the previous text was not properly finished */
	if ((prev & LOG_CONT) && (msg->flags & LOG_PREFIX) && (len < size)) {
		if (buf)
			buf[len++] = '\n';
		else
			len++;
	}

IMHO, it is more clear. The do-while cycle already is complex enough.

BTW: This is relared to the first patch. I would either patch all
three locations in one patch or better split it into three patches.


Best Regards,
Petr