Re: [PATCH] luks2: Add support for LUKS2 in (proc)/luks_script

[Glenn Washburn]

On Wed, 8 Feb 2023 19:39:19 +0100
Daniel Kiper <dkiper@net-space.pl> wrote:

> On Sun, Feb 05, 2023 at 01:05:31AM -0600, Glenn Washburn wrote:
> > The sector size in bytes is added to each line and it is allowed to be 5
> > decimal digits long, which covers the most common cases of 512 and 4096
> > byte sectors with space for an additional digit as future-proofing. The
> > size allocation is updated to reflect this additional field, allow up to
> > 5 characters and 1 space added.
> >
> > Signed-off-by: Glenn Washburn <development@efficientek.com>
> > ---
> >  grub-core/disk/cryptodisk.c | 28 ++++++++++++++++++++--------
> >  1 file changed, 20 insertions(+), 8 deletions(-)
> >
> > diff --git a/grub-core/disk/cryptodisk.c b/grub-core/disk/cryptodisk.c
> > index 34b67a705f..f2c8c8d3fe 100644
> > --- a/grub-core/disk/cryptodisk.c
> > +++ b/grub-core/disk/cryptodisk.c
> > @@ -1478,12 +1478,22 @@ luks_script_get (grub_size_t *sz)
> >    *sz = 0;
> >
> >    for (i = cryptodisk_list; i != NULL; i = i->next)
> > -    if (grub_strcmp (i->modname, "luks") == 0)
> > +    if (grub_strcmp (i->modname, "luks") == 0 ||
> > +   grub_strcmp (i->modname, "luks2") == 0)
> >        {
> > -   size += sizeof ("luks_mount ");
> > +   size += grub_strlen (i->modname);
> > +   size += sizeof ("_mount");
> >     size += grub_strlen (i->uuid);
> >     size += grub_strlen (i->cipher->cipher->name);
> > -   size += 54;
> > +   /*
> > +    * Add space in the line for (in order) spaces, cipher mode, cipher IV
> > +    * mode, sector offset, sector size and the trailing newline. This is
> > +    * an upper bound on the size of this data. There are 16 extra bytes
> > +    * in an earlier version of this code that are unaccounted for. It is
> > +    * left in the calculations in case it is needed. At worst, its short-
> > +    * lived wasted space.
> > +    */
> > +   size += 5 + 5 + 8 + 20 + 5 + 1 + 16;
> >     if (i->essiv_hash)
> >       size += grub_strlen (i->essiv_hash->name);
> >     size += i->keysize * 2;
> > @@ -1496,16 +1506,18 @@ luks_script_get (grub_size_t *sz)
> >    ptr = ret;
> >
> >    for (i = cryptodisk_list; i != NULL; i = i->next)
> > -    if (grub_strcmp (i->modname, "luks") == 0)
> > +    if (grub_strcmp (i->modname, "luks") == 0 ||
> > +   grub_strcmp (i->modname, "luks2") == 0)
> >        {
> >     unsigned j;
> >     const char *iptr;
> > -   ptr = grub_stpcpy (ptr, "luks_mount ");
> > +   ptr = grub_stpcpy (ptr, i->modname);
> > +   ptr = grub_stpcpy (ptr, "_mount ");
> >     ptr = grub_stpcpy (ptr, i->uuid);
> >     *ptr++ = ' ';
> > -   grub_snprintf (ptr, 21, "%" PRIuGRUB_UINT64_T " ", i->offset_sectors);
> > -   while (*ptr)
> > -     ptr++;
> > +   ptr += grub_snprintf (ptr, 21, "%" PRIuGRUB_UINT64_T " ",
> > +                         i->offset_sectors);
> 
> You, OK not you :-), blindly assume offset_sectors is 64-bit long. Today
> it is but in the future it may not. May I ask you to define 
> PRIuGRUB_DISK_ADDR,
> etc. properly and fix the problem here and in the other places if needed?

Ok, I've added this in a recent patch.

> 
> > +   ptr += grub_snprintf (ptr, 7, "%d ", 1 << i->log_sector_size);
> 
> Ugh... Simple "grep -Ir log_sector_size include" revealed we use
> different types in different places for log_sector_size. The most
> worrying for me are include/grub/disk.h and include/grub/cryptodisk.h.
> OK, probably it is not big problem here but I would prefer if we
> are consistent all over the place. I think it would be good to do
> "typedef grub_uint32_t grub_disk_secs_t" and use grub_disk_secs_t
> consistently everywhere. Could you do that? Of course it can be
> a separate patch set...

I agree consistency is a good thing. In this case, I don't see it
realistically posing any problems. Keeping in mind that what we are
talking about here is the log() base 2 of the sector size, a terabyte
sector size will have a log of 40. This fits in 6 bits. To have a log
sector size larger than 8 bits you're talking about an unbelievably
large number. And this is the sector size we're talking about here. So
we could have this be represented in an 8-bit char and not expect it to
be an issue for at least several centuries.

I'm don't quite see the point of "typedef grub_uint32_t
grub_disk_secs_t" with regards to the log sector size, unless you're
wanting it to be used in conjunction with (1 << log_sector_size). Are
you concerned with that shift overflowing the "%d"?

Glenn