Re: [PATCH 1/4] lib/sprintf: add scnprintf_append() helper function

[Petr Mladek <pmladek@suse.com>]

On Tue 2025-11-11 13:31:00, Junrui Luo wrote:
> On Mon, Nov 10, 2025 at 03:13:56PM +0100, Petr Mladek wrote:
> > On Fri 2025-11-07 17:51:23, David Laight wrote:
> > > On Fri, 7 Nov 2025 13:52:27 +0100
> > > Petr Mladek <pmladek@suse.com> wrote:
> > > 
> > > > On Fri 2025-11-07 11:35:35, Andy Shevchenko wrote:
> > > > > On Fri, Nov 07, 2025 at 09:12:46AM +0000, David Laight wrote:  
> > > > > > On Thu, 6 Nov 2025 21:38:33 -0800
> > > > > > Andrew Morton <akpm@linux-foundation.org> wrote:  
> > > > > > > On Fri,  7 Nov 2025 13:16:13 +0800 Junrui Luo <moonafterrain@outlook.com> wrote:  
> > > > > 
> > > > > ...
> > > > >   
> > > > > > That is true for all the snprintf() functions.
> > > > > >   
> > > > > > > I wonder if we should instead implement a kasprintf() version of this
> > > > > > > which reallocs each time and then switch all the callers over to that.  
> > > > > > 
> > > > > > That adds the cost of a malloc, and I, like kasprintf() probably ends up
> > > > > > doing all the work of snprintf twice.
> > > > > > 
> > > > > > I'd be tempted to avoid the strlen() by passing in the offset.
> > > > > > So (say):
> > > > > > #define scnprintf_at(buf, len, off, ...) \
> > > > > > 	scnprintf((buf) + off, (len) - off, __VA_ARGS__)  
> > > > 
> > > > It does not handle correctly the situation when len < off.
> > > > Othersise, it looks good.
> > > 
> > > That shouldn't happen unless the calling code is really buggy.
> > > There is also a WARN_ON_ONCE() at the top of snprintf().
> > 
> > Fair enough.
> > 
> > BTW: I have found there exists a userspace library which implements
> > this idea, the funtion is called vsnoprintf(), see
> > https://arpa2.gitlab.io/arpa2common/group__snoprintf.html
> > 
> > I know that it is cryptic. But I like the name. The letters "no"
> > match the ordering of the parameters "size, offset".
> > 
> > In our case, it would be scnoprintf() ...
> > 
> 
> Thanks for the feedback. Based on the discussion above, I plan to prepare a v2 patch.
> int scnprintf_append(char *buf, size_t size, const char *fmt, ...)
> {
> 	va_list args;
> 	size_t len;
> 
> 	len = strnlen(buf, size);
> 	if (len == size)
> 		return len;
> 	va_start(args, fmt);
> 	len += vscnprintf(buf + len, size - len, fmt, args);
> 	va_end(args);
> 	return len;
> }
> EXPORT_SYMBOL(scnprintf_append);

I am fine with this. Just please add a comment that that it can be
inefficient when using massively because it does strlen().
I see similar comment in "CAVEATS" section in man 3 strcat.

> I agree that using a macro like David suggested, with an explicit offset, is a reasonable and efficient approach.
> The `scnprintf_append()` function, however, does not require such a variable; though if used improperly, it could introduce an extra `strlen()` overhead.
> 
> However, if the consensus is to prefer the macro approach, I can rework the series to use `scnoprintf()`, as suggested by Petr instead.
> 
> That said, I believe `scnprintf_append()` also has its merits:
> it simplifies one-off string constructions and provides built-in bound checking for safety.
> Some existing code that appends strings in the kernel lacks proper bound checks, and this function could serve as a graceful replacement.
> The benefits were also demonstrated in other patches.

I think that both functions might have their users. You might consider
adding the other variant when doing the conversions.

Best Regards,
Petr