[OpenWrt-Devel] Squashfs breakage lottery with UBI WAS: [PATCH RFC 2/2] amp821xx: use newly added pad-squashfs for Meraki MR24

[Jonas Gorski]

On Sun, 1 Sep 2019 at 13:52, Russell Senior <russell at personaltelco.net> wrote:
>
> >>>>> "Jonas" == Jonas Gorski <jonas.gorski at gmail.com> writes:
>
> >> It contains a patch at the end titled: "[PATCH] base-files: pad
> >> root.squashfs to 64KiB in ubi volumes" This is another approach that
> >> just deals with the UBI+squashfs issue but works with
> >> "-nopad". Soooooo.... do we all agree there?
>
> Jonas> a) 64k is excessive, we only need 4k (actually 1k would be
> Jonas> enough, since we don't enable CONFIG_SQUASHFS_4K_DEVBLK_SIZE).
>
> Jonas> The referenced issue with 64k page size happens when
> Jonas> loop-mounting a squashfs, since loop defaults to PAGE_SIZE as its
> Jonas> block size. But we never do that in OpenWrt, and we don't support
> Jonas> any targets with that huge PAGE_SIZE - biggest is ARC with 8k.
>
> Jonas> b) it misses the squashfs's in generic sysupgrade images itself -
> Jonas> we need to pad their length as well, to avoid breaking devices
> Jonas> with a sysupgrade image hitting the corner case being flashed
> Jonas> from an unfixed firmware with the old nand.sh.
>
> Jonas> Also IMHO "1c0290c5cc6258c48b8ba46b4f9c85a21de4f875" should be
> Jonas> reverted, for the previously mentioned issues.
>
> Afaict, only devices with LEB sizes of non-integer kilobytes (like the
> MR24 with its 15.5k LEBs) need any intervention at all. Because
> squashfs's are read in 1k blocks, there is a 1 in 62 chance of creating
> a rootfs that is an inopportune size on 15.5k LEBs.  I have a PogoPlug
> v3 with LEBs of 126k, and a MikroTik RouterBOARD 493G with LEBs of
> 124k. Neither of those is affected.
>
> I still kind of like my solution where we explicitly ask for padding for
> devices that need it.

The PogoPlug would also be affected if we enabled
SQUASHFS_4K_DEVBLK_SIZE - the 493G wouldn't be. In that case we would
need to pad to 4k.

But in the current situation, If we just pad to 1k, neither one would
see an increase in LEBs used. At worst will the padding make the
squashfs image fit exactly, but it will not cause it to overspill into
a new LEB.

So padding to 1k is harmless for devices with LEBs of a multiple of
1k, and fixes devices with LEBs that have an odd .5k size.


Jonas

_______________________________________________
openwrt-devel mailing list
openwrt-devel at lists.openwrt.org
https://lists.openwrt.org/mailman/listinfo/openwrt-devel