[RFC 02/14] efi_loader: library function efi_dp_merge

[Mark Kettenis]

> Date: Tue, 14 May 2024 14:49:47 +0200
> From: Heinrich Schuchardt <heinrich.schuchardt at canonical.com>

Hi Heinrich,

> On 4/26/24 17:47, Ilias Apalodimas wrote:
> > Hi Heinrich
> > 
> > On Fri, 26 Apr 2024 at 17:53, Heinrich Schuchardt
> > <heinrich.schuchardt at canonical.com> wrote:
> >>
> >> On 26.04.24 16:30, Ilias Apalodimas wrote:
> >>> Hi Heinrich,
> >>>
> >>> On Fri, 26 Apr 2024 at 17:13, Heinrich Schuchardt
> >>> <heinrich.schuchardt at canonical.com> wrote:
> >>>>
> >>>> Provide a function to append a device_path to a list of device paths
> >>>> that is separated by final end nodes.
> >>>>
> >>>> Signed-off-by: Heinrich Schuchardt <heinrich.schuchardt at canonical.com>
> >>>> ---
> >>>>    include/efi_loader.h             |  3 +++
> >>>>    lib/efi_loader/efi_device_path.c | 31 +++++++++++++++++++++++++++++++
> >>>>    2 files changed, 34 insertions(+)
> >>>>
> >>>> diff --git a/include/efi_loader.h b/include/efi_loader.h
> >>>> index 9600941aa32..a7d7b8324f1 100644
> >>>> --- a/include/efi_loader.h
> >>>> +++ b/include/efi_loader.h
> >>>> @@ -944,6 +944,9 @@ struct efi_load_option {
> >>>>
> >>>>    struct efi_device_path *efi_dp_from_lo(struct efi_load_option *lo,
> >>>>                                          const efi_guid_t *guid);
> >>>> +struct efi_device_path *efi_dp_merge(const struct efi_device_path *dp1,
> >>>> +                                    efi_uintn_t *size,
> >>>> +                                    const struct efi_device_path *dp2);
> >>>>    struct efi_device_path *efi_dp_concat(const struct efi_device_path *dp1,
> >>>>                                         const struct efi_device_path *dp2,
> >>>>                                         bool split_end_node);
> >>>> diff --git a/lib/efi_loader/efi_device_path.c b/lib/efi_loader/efi_device_path.c
> >>>> index 46aa59b9e40..16cbe41d32f 100644
> >>>> --- a/lib/efi_loader/efi_device_path.c
> >>>> +++ b/lib/efi_loader/efi_device_path.c
> >>>> @@ -270,6 +270,37 @@ struct efi_device_path *efi_dp_dup(const struct efi_device_path *dp)
> >>>>           return ndp;
> >>>>    }
> >>>>
> >>>> +/**
> >>>> + * efi_dp_merge() - Concatenate two device paths separated by final end node
> >>>> + *
> >>>> + * @dp1:       first device path
> >>>> + * @size:      pointer to length of @dp1, total size on return
> >>>> + * @dp2:       second device path
> >>>> + *
> >>>> + * Return:     concatenated device path or NULL
> >>>> + */
> >>>> +struct efi_device_path *efi_dp_merge(const struct efi_device_path *dp1,
> >>>> +                                    efi_uintn_t *size,
> >>>> +                                    const struct efi_device_path *dp2)
> >>>> +{
> >>>> +       efi_uintn_t len, len1, len2;
> >>>> +       struct efi_device_path *dp;
> >>>> +       u8 *p;
> >>>> +
> >>>> +       len1 = *size;
> >>>> +       len2 = efi_dp_size(dp2) + sizeof(END);
> >>>> +       len = len1 + len2;
> >>>> +       dp = efi_alloc(len);
> >>>> +       if (!dp)
> >>>> +               return NULL;
> >>>> +       memcpy(dp, dp1, len1);
> >>>> +       p = (u8 *)dp + len1;
> >>>> +       memcpy(p, dp2, len2);
> >>>> +       *size = len;
> >>>
> >>> Can't we just use efi_dp_concat()?
> >>
> >> efi_dp_concat cannot be used to put a device-path end-node in between
> >> two device-paths that are concatenated. We need that separator in the
> >> load options.
> > 
> > I am not sure I am following you. It's been a few years so please bear
> > with me until I manage to re-read that code and page it back to
> > memory.
> > 
> > What I remember though is that the format of the DP looks like this:
> > 
> > Loaded image DP - end node - initrd GUID DP (without and end node) -
> > initrd - end of device path.
> > So to jump from the special initrd GUID to the actual DP that contains
> > the initrd you need to do an efi_dp_next(). Also, efi_dp_concat() will
> > inject an end node with a DEVICE_PATH_SUB_TYPE_END type if the third
> > argument is 'true'.
> > 
> > What am I missing?
> 
> Let us assume
> 
> dp1 = /vmlinux/endnode/initrd/endnode
> dp2 = /dtb/endnode
> 
> and we invoke dp_concat(dp1, dp2, true):
> 
> sz1 is calculated via the efi_dp_size() function which looks for the 
> *first* device-path end node.
> 
> sz1 = efi_dp_size(dp1) = sizeof(/vmlinux)
> 
> sz1 will not include initrd and its endnode.
> 
> So dp_concat(/vmlinux/endnode/initrd/endnode, /dtb/endnode, true)
> will return:
> 
> /vmlinux/endnode/dtb/endnode
> 
> but in our boot option we want
> 
> /vmlinux/endnode/initrd/endnode/dtb/endnode

That makes me realize a potential flaw with the design of this boot
option.  The concept of an initrd is Linux-specific, but the concept
of loading a DTB to pass along to an EFI image far less so.  So how
would this work if I want to only pass the DTB?  Would that be:

/vmlinux/endnode/endnode/dtb/endnode

?