[U-Boot] [PATCH v3 2/2] sunxi: binman: Add U-Boot binary size check

Mon May 7 20:46:58 UTC 2018

On Wed, May 02, 2018 at 03:24:50PM +0100, Måns Rullgård wrote:
> Maxime Ripard <maxime.ripard at bootlin.com> writes:
> 
> > 1;5201;0c
> > On Wed, May 02, 2018 at 10:34:49AM +0100, Måns Rullgård wrote:
> >> Siarhei Siamashka <siarhei.siamashka at gmail.com> writes:
> >> 
> >> > On Tue, 01 May 2018 18:25:06 +0100
> >> > Måns Rullgård <mans at mansr.com> wrote:
> >> >
> >> >> Maxime Ripard <maxime.ripard at free-electrons.com> writes:
> >> >> 
> >> >> > The U-Boot binary may trip over its actual allocated size in the storage.
> >> >> > In such a case, the environment will not be readable anymore (because
> >> >> > corrupted when the new image was flashed), and any attempt at using saveenv
> >> >> > to reconstruct the environment will result in a corrupted U-Boot binary.
> >> >> >
> >> >> > Reviewed-by: Andre Przywara <andre.przywara at arm.com>
> >> >> > Signed-off-by: Maxime Ripard <maxime.ripard at free-electrons.com>
> >> >> > ---
> >> >> >  arch/arm/dts/sunxi-u-boot.dtsi | 12 ++++++++++++
> >> >> >  1 file changed, 12 insertions(+)
> >> >> >
> >> >> > diff --git a/arch/arm/dts/sunxi-u-boot.dtsi b/arch/arm/dts/sunxi-u-boot.dtsi
> >> >> > index 5adfd9bca2ec..72e95afd780e 100644
> >> >> > --- a/arch/arm/dts/sunxi-u-boot.dtsi
> >> >> > +++ b/arch/arm/dts/sunxi-u-boot.dtsi
> >> >> > @@ -1,5 +1,14 @@
> >> >> >  #include <config.h>
> >> >> >
> >> >> > +/*
> >> >> > + * This is the maximum size the U-Boot binary can be, which is basically
> >> >> > + * the start of the environment, minus the start of the U-Boot binary in
> >> >> > + * the MMC. This makes the assumption that the MMC is using 512-bytes
> >> >> > + * blocks, but devices using something other than that remains to be
> >> >> > + * seen.
> >> >> > + */
> >> >> > +#define UBOOT_MMC_MAX_SIZE	(CONFIG_ENV_OFFSET - (CONFIG_SYS_MMCSD_RAW_MODE_U_BOOT_SECTOR * 512))
> >> >> > +
> >> >> >  / {
> >> >> >  	binman {
> >> >> >  		filename = "u-boot-sunxi-with-spl.bin";
> >> >> > @@ -8,6 +17,9 @@
> >> >> >  			filename = "spl/sunxi-spl.bin";
> >> >> >  		};
> >> >> >  		u-boot-img {
> >> >> > +#ifdef CONFIG_MMC
> >> >> > +			size = <UBOOT_MMC_MAX_SIZE>;
> >> >> > +#endif
> >> >> >  			pos = <CONFIG_SPL_PAD_TO>;
> >> >> >  		};
> >> >> >  	};
> >> >> > --   
> >> >> 
> >> >> This is broken in (at least) two ways:
> >> >> 
> >> >> 1. It is simply nonsensical if u-boot and env are on different devices
> >> >>    or not on mmc even if mmc support is enabled.
> >> >> 
> >> >> 2. It causes u-boot-sunxi-with-spl.bin to be padded to the env offset.
> >> >>    At best, this is useless.  If the env doesn't immediately follow
> >> >>    u-boot, it really breaks things.
> >> >> 
> >> >> Please fix or revert, I don't really care which.
> >> >
> >> > The padding is not useless. It reserves space for future size expansions
> >> > and makes it harder for the users to hurt themselves.
> >> >
> >> > For example, if the padded U-Boot size is 1024K while the actual size
> >> > is only ~800K, then adventurous users might be tempted to fit some of
> >> > their data into this gap. Yay, ~200K of storage space for free! Except
> >> > that the next U-Boot release may grow up to 900K without any warning
> >> > and if the users are not careful enough, then their data would be
> >> > corrupted during upgrade.
> >> 
> >> Do U-Boot users really need that level of hand-holding?
> >
> > Yes, and that's never a good argument to make, because...
> >
> >> > Could you please tell us what is your problem and why reverting this
> >> > patch would fix it?
> >> 
> >> See above.  Best case, it just wastes space in the created file.  If the
> >> configuration is anything other than U-Boot immediately followed by env
> >> on the same device, it _will_ break things horribly.  A few examples:
> >> 
> >> 1.  U-Boot and env are on different devices, e.g. U-Boot on mmc and env
> >>     in SPI NOR flash.  In this case, padding the file to the env offset
> >>     makes no sense.  Writing the image will corrupt anything stored
> >>     after U-Boot at a smaller (but still safe) offset.
> >
> > .. I could make pretty much the same one for all your cases, which
> > would be completely useless, and wouldn't fix anything.
> 
> Huh?  I'm saying we shouldn't "helpfully" do things that actually break
> perfectly valid setups.  U-Boot users are expected to know what they are
> doing, and shouldn't need that kind of help.  In my opinion.

My point is that this is a slippery slope. Anyway..

> > I guess this one could be solved using an ifdef guard with
> > ENV_IS_IN_MMC.
> 
> Not if env is on a different mmc device.

Ah, right...

> >> 2.  U-Boot at a non-zero offset, followed by env, but _not_ on mmc.  If
> >>     CONFIG_SYS_MMCSD_RAW_MODE_U_BOOT_SECTOR, probably at its default
> >>     value, is smaller than the offset of U-Boot in its actual device,
> >>     the padding will be too large.  Writing the image file will then
> >>     corrupt a stored env.
> >
> > This one would be covered too.
> >
> >> 3.  U-Boot at start of device, env at end as indicated by a negative
> >>     CONFIG_ENV_OFFSET.  With this configuration, binman tries to pad the
> >>     image to (nearly) 2^64 bytes and promptly fills up your storage
> >>     device.
> >
> > I'm not too sure about how to fix this one though. Any suggestion?
> 
> I just don't see the point in trying to pin down the very specific case
> of U-Boot and env being on the same device with only a (small) amount of
> padding between them.  There are a million other ways for users to screw
> up, so why should we be making a half-hearted effort to prevent this
> one?

Unfortunately, this isn't a very specific case this is the default we
shipped for years. And we have people relying on it in the wild.

This is why we added that check in the first place, because the size
of uboot increased so much that it was now overlapping with the
environment.

I guess now that we transitioned with the release of 2018.05, we can
remove that check entirely, since the default is the first partition,
wherever that is.

Maxime

-- 
Maxime Ripard, Bootlin (formerly Free Electrons)
Embedded Linux and Kernel engineering
https://bootlin.com