Re: [edk2-devel] [PATCH 12/13] OvmfPkg/QemuKernelLoaderFsDxe: add support for new Linux initrd device path

["Ard Biesheuvel" <ard.biesheuvel@linaro.org>]

On Tue, 3 Mar 2020 at 11:10, Laszlo Ersek <lersek@redhat.com> wrote:
>
> On 03/02/20 08:29, Ard Biesheuvel wrote:
> > Linux v5.7 will introduce a new method to load the initial ramdisk
> > (initrd) from the loader, using the LoadFile2 protocol installed on a
> > special vendor GUIDed media device path.
> >
> > Add support for this to our QEMU command line kernel/initrd loader.
> >
> > Ref: https://bugzilla.tianocore.org/show_bug.cgi?id=2566
> > Signed-off-by: Ard Biesheuvel <ard.biesheuvel@linaro.org>
> > ---
> >  OvmfPkg/QemuKernelLoaderFsDxe/QemuKernelLoaderFsDxe.c   | 79 ++++++++++++++++++++
> >  OvmfPkg/QemuKernelLoaderFsDxe/QemuKernelLoaderFsDxe.inf |  2 +
> >  2 files changed, 81 insertions(+)
> >
> > diff --git a/OvmfPkg/QemuKernelLoaderFsDxe/QemuKernelLoaderFsDxe.c b/OvmfPkg/QemuKernelLoaderFsDxe/QemuKernelLoaderFsDxe.c
> > index 77d8fedb738a..415946edf22a 100644
> > --- a/OvmfPkg/QemuKernelLoaderFsDxe/QemuKernelLoaderFsDxe.c
> > +++ b/OvmfPkg/QemuKernelLoaderFsDxe/QemuKernelLoaderFsDxe.c
> > @@ -13,15 +13,18 @@
> >  #include <Guid/FileInfo.h>
> >  #include <Guid/FileSystemInfo.h>
> >  #include <Guid/FileSystemVolumeLabelInfo.h>
> > +#include <Guid/LinuxEfiInitrdMedia.h>
> >  #include <Guid/QemuKernelLoaderFsMedia.h>
> >  #include <Library/BaseLib.h>
> >  #include <Library/BaseMemoryLib.h>
> >  #include <Library/DebugLib.h>
> > +#include <Library/DevicePathLib.h>
> >  #include <Library/MemoryAllocationLib.h>
> >  #include <Library/QemuFwCfgLib.h>
> >  #include <Library/UefiBootServicesTableLib.h>
> >  #include <Library/UefiRuntimeServicesTableLib.h>
> >  #include <Protocol/DevicePath.h>
> > +#include <Protocol/LoadFile2.h>
> >  #include <Protocol/SimpleFileSystem.h>
> >
> >  //
> > @@ -84,6 +87,19 @@ STATIC CONST SINGLE_VENMEDIA_NODE_DEVPATH mFileSystemDevicePath = {
> >    }
> >  };
> >
> > +STATIC CONST SINGLE_VENMEDIA_NODE_DEVPATH mInitrdDevicePath = {
> > +  {
> > +    {
> > +      MEDIA_DEVICE_PATH, MEDIA_VENDOR_DP,
> > +      { sizeof (VENDOR_DEVICE_PATH) }
> > +    },
> > +    LINUX_EFI_INITRD_MEDIA_GUID
> > +  }, {
> > +    END_DEVICE_PATH_TYPE, END_ENTIRE_DEVICE_PATH_SUBTYPE,
> > +    { sizeof (EFI_DEVICE_PATH_PROTOCOL) }
> > +  }
> > +};
> > +
> >  //
> >  // The "file in the EFI stub filesystem" abstraction.
> >  //
> > @@ -839,6 +855,48 @@ STATIC CONST EFI_SIMPLE_FILE_SYSTEM_PROTOCOL mFileSystem = {
> >    StubFileSystemOpenVolume
> >  };
> >
> > +STATIC
> > +EFI_STATUS
> > +EFIAPI
> > +InitrdLoadFile2 (
> > +  IN EFI_LOAD_FILE2_PROTOCOL          *This,
> > +  IN EFI_DEVICE_PATH_PROTOCOL         *FilePath,
> > +  IN BOOLEAN                          BootPolicy,
> > +  IN OUT UINTN                        *BufferSize,
> > +  IN VOID                             *Buffer OPTIONAL
> > +  )
> > +{
> > +  CONST KERNEL_BLOB   *InitrdBlob = &mKernelBlob[KernelBlobTypeInitrd];
> > +
> > +  ASSERT (InitrdBlob->Size > 0);
> > +
> > +  if (BootPolicy) {
> > +    return EFI_UNSUPPORTED;
> > +  }
> > +
> > +  if (BufferSize == NULL || !IsDevicePathValid (FilePath, 0)) {
> > +    return EFI_INVALID_PARAMETER;
> > +  }
> > +
> > +  if (FilePath->Type != END_DEVICE_PATH_TYPE ||
> > +      FilePath->SubType != END_ENTIRE_DEVICE_PATH_SUBTYPE) {
> > +    return EFI_NOT_FOUND;
> > +  }
> > +
> > +  if (Buffer == NULL || *BufferSize < InitrdBlob->Size) {
> > +    *BufferSize = InitrdBlob->Size;
> > +    return EFI_BUFFER_TOO_SMALL;
> > +  }
> > +
> > +  CopyMem (Buffer, InitrdBlob->Data, InitrdBlob->Size);
> > +
> > +  *BufferSize = InitrdBlob->Size;
> > +  return EFI_SUCCESS;
> > +}
> > +
> > +STATIC CONST EFI_LOAD_FILE2_PROTOCOL     mInitrdLoadFile2 = {
> > +  InitrdLoadFile2,
> > +};
> >
> >  //
> >  // Utility functions.
> > @@ -945,6 +1003,7 @@ QemuKernelLoaderFsDxeEntrypoint (
> >    KERNEL_BLOB               *KernelBlob;
> >    EFI_STATUS                Status;
> >    EFI_HANDLE                FileSystemHandle;
> > +  EFI_HANDLE                InitrdLoadFile2Handle;
> >
> >    if (!QemuFwCfgIsAvailable ()) {
> >      return EFI_NOT_FOUND;
> > @@ -989,8 +1048,28 @@ QemuKernelLoaderFsDxeEntrypoint (
> >      goto FreeBlobs;
> >    }
> >
> > +  if (KernelBlob[KernelBlobTypeInitrd].Size > 0) {
> > +    InitrdLoadFile2Handle = NULL;
> > +    Status = gBS->InstallMultipleProtocolInterfaces (&InitrdLoadFile2Handle,
> > +                    &gEfiDevicePathProtocolGuid,  &mInitrdDevicePath,
> > +                    &gEfiLoadFile2ProtocolGuid,   &mInitrdLoadFile2,
> > +                    NULL);
> > +    if (EFI_ERROR (Status)) {
> > +      DEBUG ((DEBUG_ERROR, "%a: InstallMultipleProtocolInterfaces(): %r\n",
> > +        __FUNCTION__, Status));
> > +      goto UninstallFileSystemHandle;
> > +    }
> > +  }
> > +
> >    return EFI_SUCCESS;
> >
> > +UninstallFileSystemHandle:
> > +  Status = gBS->UninstallMultipleProtocolInterfaces (FileSystemHandle,
> > +                  &gEfiDevicePathProtocolGuid,       &mFileSystemDevicePath,
> > +                  &gEfiSimpleFileSystemProtocolGuid, &mFileSystem,
> > +                  NULL);
> > +  ASSERT_EFI_ERROR (Status);
> > +
> >  FreeBlobs:
> >    while (BlobType > 0) {
> >      CurrentBlob = &mKernelBlob[--BlobType];
> > diff --git a/OvmfPkg/QemuKernelLoaderFsDxe/QemuKernelLoaderFsDxe.inf b/OvmfPkg/QemuKernelLoaderFsDxe/QemuKernelLoaderFsDxe.inf
> > index f4b50c265027..737f9b87a7c7 100644
> > --- a/OvmfPkg/QemuKernelLoaderFsDxe/QemuKernelLoaderFsDxe.inf
> > +++ b/OvmfPkg/QemuKernelLoaderFsDxe/QemuKernelLoaderFsDxe.inf
> > @@ -28,6 +28,7 @@ [LibraryClasses]
> >    BaseLib
> >    BaseMemoryLib
> >    DebugLib
> > +  DevicePathLib
> >    MemoryAllocationLib
> >    UefiBootServicesTableLib
> >    QemuFwCfgLib
> > @@ -42,6 +43,7 @@ [Guids]
> >
> >  [Protocols]
> >    gEfiDevicePathProtocolGuid                ## PRODUCES
> > +  gEfiLoadFile2ProtocolGuid                 ## PRODUCES
> >    gEfiSimpleFileSystemProtocolGuid          ## PRODUCES
> >
> >  [Depex]
> >
>
> I think this patch conflicts with your other patch:
>
>   [edk2-devel] [PATCH v3 2/6] OvmfPkg: add 'initrd' shell command to
>                               expose Linux initrd via device path
>
> There should not be two different handles in the handle database that
> carry the same device path (I mean "same device path" by contents, not
> by reference).
>
> I believe that it's possible that the command-line specified -kernel /
> -initrd are attempted first, but the kernel's stub returns for some
> reason. Then the user can still go to the UEFI shell, and use the new
> dynamic shell command. In that case I think the device path will cease
> being unique (by contents).
>
> I think you can solve this as follows:
>
> - in both modules (this driver, and the dynamic shell command driver),
> install a NULL protocol interface with GUID = gEfiCallerIdGuid on the
> same handle that carries the device path and LoadFile2
>
> - in both modules, before you touch the handle that carries the device
> path, try to locate it first. If it exists but doesn't carry the subject
> module's FILE_GUID as a NULL protocol interface, we know the device path
> was created by the "other" module, and we can bail.
>
> Just an idea.
>

Thanks.

I did wonder about the rules regarding duplicates in the device path
space. If the DXE core doesn't catch that, this seems like a gross
oversight to me.

But it raises an interesting point: the idea is that any boot stage
could elect to provide this interface, not just UEFI or the shell
itself, but also shim, grub or whatever executes in between. That
means that, in general, it should probably be the responsibility of
the code that installs the protocol to ensure that it doesn't exist
already.

For this code, it means it could simply install it, since it
guarantees to execute first.

For the initrd implementation, it means we should check whether any
other implementation exists already, which could simply be done [in
that particular case] by locating the protocol and checking the
address against the statically allocated one in the driver.

I am aware that both are DXE drivers, but since the dynamic shell
command is not invoked by any driver that is invoked before EndOfDxe,
I think the above should be sufficient.