Re: [PATCH 1/1] mvebu: fix end-of-array check

["Pali Rohár" <pali@kernel.org>]

On Tuesday 06 December 2022 07:10:15 Stefan Roese wrote:
> Hi Pali,
> 
> On 12/5/22 19:18, Pali Rohár wrote:
> > On Monday 05 December 2022 12:42:44 Stefan Roese wrote:
> > > Hi!
> > > 
> > > On 12/4/22 11:39, Pali Rohár wrote:
> > > > Hello!
> > > > 
> > > > I would suggest to change description of the patch from
> > > > 
> > > >     "mvebu: fix end-of-array check"
> > > > 
> > > > to something like:
> > > > 
> > > >     "arm: mvebu: Espressobin: fix end-of-array check in env"
> > > > 
> > > > as it affects only Espressobin boards (not other mvebu).
> > > 
> > > Yes, please update the commit subject here.
> > > 
> > > > Stefan, please look below as this issue/fix is important.
> > > 
> > > Yes.
> > > 
> > > > On Wednesday 30 November 2022 13:33:40 Derek LaHousse wrote:
> > > > > Properly seek the end of default_environment variables.
> > > > > 
> > > > > The current algorithm overwrites from the second variable.  This
> > > > > replacement finds the end of the array of strings.
> > > > > 
> > > > > Stomped variables include "board", "soc", "loadaddr".  These can be
> > > > > seen on a "env default -a" after patch, but they are not seen with a
> > > > > version before the patch.
> > > > 
> > > > This is a real issue which I introduced in the past. So it some fix for
> > > > this issue should be pulled into the v2023.01 release.
> > > 
> > > Understood.
> > > 
> > > > > Signed-off-by: Derek LaHousse <derek@seaofdirac.org>
> > > > > ---
> > > > >    board/Marvell/mvebu_armada-37xx/board.c | 7 +++++--
> > > > >    1 file changed, 5 insertions(+), 2 deletions(-)
> > > > > 
> > > > > diff --git a/board/Marvell/mvebu_armada-37xx/board.c
> > > > > b/board/Marvell/mvebu_armada-37xx/board.c
> > > > > index c6ecc323bb..ac29ac5b95 100644
> > > > > --- a/board/Marvell/mvebu_armada-37xx/board.c
> > > > > +++ b/board/Marvell/mvebu_armada-37xx/board.c
> > > > > @@ -100,8 +100,11 @@ int board_late_init(void)
> > > > >    		return 0;
> > > > >    	/* Find free buffer in default_environment[] for new variables
> > > > > */
> > > > > -	while (*ptr != '\0' && *(ptr+1) != '\0') ptr++;
> > > > > -	ptr += 2;
> > > > > +	if (*ptr != '\0') { // Defending against empty default env
> > > > > +		while ((i = strlen(ptr)) != 0) {
> > > > > +			ptr += i + 1;
> > > > > +		}
> > > > > +	}
> > > > 
> > > > If I'm looking at the outer-if condition correctly then it is not
> > > > needed. strlen("") returns zero and so inner-while loop stops
> > > > immediately.
> > > > 
> > > > My proposed fix for this issue was just changing correct while loop
> > > > check to ensure that ptr is set after the _last_ variable.
> > > > 
> > > > -	while (*ptr != '\0' && *(ptr+1) != '\0') ptr++;
> > > > -	ptr += 2;
> > > > +	while (*ptr != '\0' || *(ptr+1) != '\0') ptr++;
> > > > +	ptr++;
> > > > 
> > > > Both changes should be equivalent, but I'm not sure which one is more
> > > > readable. The original issue was introduced really by non-readable
> > > > code...
> > > > 
> > > > Stefan, do you have a preference which one fix is better / more
> > > > readable?
> > > 
> > > I would prefer to get Pali's corrected version included. Could you
> > > please prepare a v2 patch with this update and also with the added
> > > or changed patch subject.
> > 
> > Originally this issue was reported half year ago on the armbian forum:
> > https://forum.armbian.com/topic/19564-making-espressobin-v7-work-in-2022/?do=findComment&comment=138136
> > 
> > U-Boot "changed" variable name scriptaddr to criptaddr (without leading
> > s) and broke booting.
> > 
> > Details are later in comments:
> > https://forum.armbian.com/topic/19564-making-espressobin-v7-work-in-2022/?do=findComment&comment=154062
> > https://forum.armbian.com/topic/19564-making-espressobin-v7-work-in-2022/?do=findComment&comment=154235
> > 
> > If you prefer my variant, I can prepare a v2 patch.
> 
> Yes, please do. That's easiest for me. I'll push this quickly then.
> 
> Thanks,
> Stefan

Hello Stefan! During discussion on the forum, Derek pointed that my
"fix" does not work for another edge case when *ptr is '\0' before
entering into while-loop. That is a valid point and code needs to be
adjusted.

With Derek we do not have an agreement how to write this peace of the
code in readable way, which should move ptr pointer to the end of env
list - find two nul bytes which indicates end of the env list and then
set ptr to the second nul byte, so at this position can be put another
new variable. Plus handle special case nul byte is at the beginning.

My idea was to use single while loop to find this location to have code
minimal in its size. Derek thinks that code is better readable if
iteration is done via strlen() calls in while-loop, like it is in this
version.

Stefan, do you have an idea how to write this code in a way that is fast
and also readable and maintainable?