Re: [PATCH 1/2] plainmount: Support decryption of devices encrypted in plain mode.

[Milan Broz]

On 01/02/2022 03:45, Glenn Washburn wrote:
> On Mon, 31 Jan 2022 17:40:24 +0000
> Maxim Fomin <maxim@fomin.one> wrote:
>
> > ------- Original Message -------
> >
> > On Monday, January 31st, 2022 at 14:15, Milan Broz <gmazyland@gmail.com> wrote:
> >
> > > On 30/01/2022 20:40, Maxim Fomin wrote:
> > >
> > > > This patch introduces support for plain encryption mode (plain dm-crypt) via
> > > >
> > > > new module and command named 'plainmount'. The command allows to open devices
> > > >
> > > > encrypted in plain mode (without LUKS) with following syntax:
> > > >
> > > > +
> > >
> > > ...
> > >
> > > > +#define GRUB_PLAINMOUNT_UUID "00000000000000000000000000000000"
> > > >
> > > > +#define GRUB_PLAINMOUNT_CIPHER "aes-cbc-essiv:sha256"
> > > >
> > > > +#define GRUB_PLAINMOUNT_DIGEST "ripemd160"
> > > >
> > > > +#define GRUB_PLAINMOUNT_KEY_SIZE 256
> > > >
> > > > +#define GRUB_PLAINMOUNT_SECTOR_SIZE 512
> > >
> > > Sooner or later we will have to change this default in cryptsetup
> > >
> > > (as ripemd and CBC mode are no longer the best options) and you
> > >
> > > you will create data corruption here (as there is no way in plain
> > >
> > > mode to check that the mode is set correctly).
> > >
> > > Not sure if it is possible, but in normal system it should be required
> > >
> > > that these parameters are set in /etc/crypttab, grub should perhaps
> > >
> > > require explicit setting them in config too?
> > >
> > > Milan
> >
> > These default values can be fixed or removed altogether. Some additional points.
> >
> > 1. Just to clarify - mounting device in plain mode with wrong parameters (it is 
> > irrelevant
> > whether default values or explicitly set arguments are wrong) does not case 
> > data corruption
> > per se, data corruption requires writing to such device. And writing to 
> > unformatted device
> > is complicated by the fact that no command operating on files will work with 
> > such device.
> > User can corrupt data only if he uses some grub command which has functionality 
> > to write
> > arbitrary byte data to arbitrary device (like 'dd' command) at arbitrary device 
> > offset.
>
> I agree. I don't think data corruption is a big issue because GRUB
> doesn't (generally) write to disks.

Yes, what I meant is in some situations like when the encryption use 4k sectors 
while
default here is 512Bytes. Then you can have the first 512B decrypted correctly
(the rest 7 512B sectors is garbage, as it fits in 4k larger sector - depends 
on IV setting).
This is enough to detect FS signature but obviously not correct data.
(In most real cases you get garbage in the whole sector.)

It is perhaps ok to ignore it, though.

...
 
> I'm fine with default arguments, _if_ the defaults are common (eg. if
> cryptsetup has defaults use those). As a user of plainmount, I think
> I'd want to always be explicit incase the defaults changed. Of course,
> not having defaults is okay too.

FYI: The defaults from patch are default for plain device in cryptsetup since
version 1.1.0 (2010).
But we have to change it soon, perhaps in 2.5 release.

Milan