From: "Pedro Falcato" <pedro.falcato@gmail.com>
To: "Marvin Häuser" <mhaeuser@posteo.de>
Cc: edk2-devel-groups-io <devel@edk2.groups.io>,
Jeff Brasen <jbrasen@nvidia.com>
Subject: Re: [edk2-devel] [PATCH v3 1/2] Ext4Pkg: Improve Ext4IsBindingSupported() behavior
Date: Mon, 13 Sep 2021 17:35:51 +0100 [thread overview]
Message-ID: <CAKbZUD1q5kkA=7U6c45jFZT1neuua21B1ZVn6DMwfjPHbPzNPQ@mail.gmail.com> (raw)
In-Reply-To: <23908cee-adf2-bee3-3bf9-d65f245070b8@posteo.de>
On BY_DRIVER, Jeff reached the conclusion that we can't open the
BLOCK_IO protocol with BY_DRIVER since it's already owned by
DiskIoDxe. The finding makes sense and is consistent
with FatPkg's behaviour.
As for the GET_PROTOCOL issue, feel free to send a patch ;)
On Sun, Sep 12, 2021 at 11:40 AM Marvin Häuser <mhaeuser@posteo.de> wrote:
>
> On 11/09/2021 00:11, Jeff Brasen via groups.io wrote:
> > A couple of improvements to improve performance.
> > Add check to return ACCESS_DENIED if already connected
> > Add check to verify superblock magic during supported to reduce start calls
> >
> > Signed-off-by: Jeff Brasen <jbrasen@nvidia.com>
> > ---
> > Features/Ext4Pkg/Ext4Dxe/Ext4Dxe.h | 14 +++++++
> > Features/Ext4Pkg/Ext4Dxe/Ext4Dxe.c | 54 +++++++++++++++++++++------
> > Features/Ext4Pkg/Ext4Dxe/Superblock.c | 35 +++++++++++++++++
> > 3 files changed, 92 insertions(+), 11 deletions(-)
> >
> > diff --git a/Features/Ext4Pkg/Ext4Dxe/Ext4Dxe.h b/Features/Ext4Pkg/Ext4Dxe/Ext4Dxe.h
> > index 64eab455db..a9b932ed52 100644
> > --- a/Features/Ext4Pkg/Ext4Dxe/Ext4Dxe.h
> > +++ b/Features/Ext4Pkg/Ext4Dxe/Ext4Dxe.h
> > @@ -1117,4 +1117,18 @@ Ext4GetVolumeName (
> > OUT UINTN *VolNameLen
> > );
> >
> > +/**
> > + Checks the superblock's magic value.
> > +
> > + @param[in] DiskIo Pointer to the DiskIo.
> > + @param[in] BlockIo Pointer to the BlockIo.
> > +
> > + @returns Whether the partition has a valid EXT4 superblock magic value.
> > +**/
> > +BOOLEAN
> > +Ext4SuperblockCheckMagic (
> > + IN EFI_DISK_IO_PROTOCOL *DiskIo,
> > + IN EFI_BLOCK_IO_PROTOCOL *BlockIo
> > + );
> > +
> > #endif
> > diff --git a/Features/Ext4Pkg/Ext4Dxe/Ext4Dxe.c b/Features/Ext4Pkg/Ext4Dxe/Ext4Dxe.c
> > index ea2e048d77..d9fbe9ea78 100644
> > --- a/Features/Ext4Pkg/Ext4Dxe/Ext4Dxe.c
> > +++ b/Features/Ext4Pkg/Ext4Dxe/Ext4Dxe.c
> > @@ -631,7 +631,6 @@ Ext4Unload (
> > @retval EFI_ACCESS_DENIED The device specified by ControllerHandle and
> > RemainingDevicePath is already being managed by a different
> > driver or an application that requires exclusive access.
> > - Currently not implemented.
> > @retval EFI_UNSUPPORTED The device specified by ControllerHandle and
> > RemainingDevicePath is not supported by the driver specified by This.
> > **/
> > @@ -643,32 +642,65 @@ Ext4IsBindingSupported (
> > IN EFI_DEVICE_PATH *RemainingDevicePath OPTIONAL
> > )
> > {
> > - // Note to self: EFI_OPEN_PROTOCOL_TEST_PROTOCOL lets us not close the
> > - // protocol and ignore the output argument entirely
> > + EFI_STATUS Status;
> > + EFI_DISK_IO_PROTOCOL *DiskIo;
> > + EFI_BLOCK_IO_PROTOCOL *BlockIo;
> >
> > - EFI_STATUS Status;
> > + DiskIo = NULL;
> > + BlockIo = NULL;
> >
> > + //
> > + // Open the IO Abstraction(s) needed to perform the supported test
> > + //
> > Status = gBS->OpenProtocol (
> > ControllerHandle,
> > &gEfiDiskIoProtocolGuid,
> > - NULL,
> > - BindingProtocol->ImageHandle,
> > + (VOID **) &DiskIo,
> > + BindingProtocol->DriverBindingHandle,
> > ControllerHandle,
> > - EFI_OPEN_PROTOCOL_TEST_PROTOCOL
> > + EFI_OPEN_PROTOCOL_BY_DRIVER
> > );
> >
> > if (EFI_ERROR (Status)) {
> > return Status;
> > }
> > -
> > + //
> > + // Open the IO Abstraction(s) needed to perform the supported test
> > + //
> > Status = gBS->OpenProtocol (
> > ControllerHandle,
> > &gEfiBlockIoProtocolGuid,
> > - NULL,
> > - BindingProtocol->ImageHandle,
> > + (VOID **) &BlockIo,
> > + BindingProtocol->DriverBindingHandle,
> > ControllerHandle,
> > - EFI_OPEN_PROTOCOL_TEST_PROTOCOL
> > + EFI_OPEN_PROTOCOL_GET_PROTOCOL
> > );
> > +
> > + if (!EFI_ERROR (Status)) {
> > + if (!Ext4SuperblockCheckMagic (DiskIo, BlockIo)) {
> > + Status = EFI_UNSUPPORTED;
> > + }
> > + }
> > +
> > + //
> > + // Close the I/O Abstraction(s) used to perform the supported test
> > + //
> > + if (DiskIo != NULL) {
> > + gBS->CloseProtocol (
> > + ControllerHandle,
> > + &gEfiDiskIoProtocolGuid,
> > + BindingProtocol->DriverBindingHandle,
> > + ControllerHandle
> > + );
> > + }
> > + if (BlockIo != NULL) {
> > + gBS->CloseProtocol (
> > + ControllerHandle,
> > + &gEfiBlockIoProtocolGuid,
> > + BindingProtocol->DriverBindingHandle,
> > + ControllerHandle
> > + );
> > + }
>
> GET_PROTOCOL protocols are not to be closed, I guess this was missed
> when there was still the question whether to use BY_DRIVER for BlockIo.
>
> Best regards,
> Marvin
>
> > return Status;
> > }
> >
> > diff --git a/Features/Ext4Pkg/Ext4Dxe/Superblock.c b/Features/Ext4Pkg/Ext4Dxe/Superblock.c
> > index c321d8c3d8..0c965415c5 100644
> > --- a/Features/Ext4Pkg/Ext4Dxe/Superblock.c
> > +++ b/Features/Ext4Pkg/Ext4Dxe/Superblock.c
> > @@ -34,6 +34,41 @@ STATIC CONST UINT32 gSupportedIncompatFeat =
> > // this is desired, it's fairly trivial to look for EFI_VOLUME_CORRUPTED
> > // references and add some Ext4SignalCorruption function + function call.
> >
> > +/**
> > + Checks the superblock's magic value.
> > +
> > + @param[in] DiskIo Pointer to the DiskIo.
> > + @param[in] BlockIo Pointer to the BlockIo.
> > +
> > + @returns Whether the partition has a valid EXT4 superblock magic value.
> > +**/
> > +BOOLEAN
> > +Ext4SuperblockCheckMagic (
> > + IN EFI_DISK_IO_PROTOCOL *DiskIo,
> > + IN EFI_BLOCK_IO_PROTOCOL *BlockIo
> > + )
> > +{
> > + UINT16 Magic;
> > + EFI_STATUS Status;
> > +
> > + Status = DiskIo->ReadDisk (
> > + DiskIo,
> > + BlockIo->Media->MediaId,
> > + EXT4_SUPERBLOCK_OFFSET + OFFSET_OF (EXT4_SUPERBLOCK, s_magic),
> > + sizeof (Magic),
> > + &Magic
> > + );
> > + if (EFI_ERROR (Status)) {
> > + return FALSE;
> > + }
> > +
> > + if (Magic != EXT4_SIGNATURE) {
> > + return FALSE;
> > + }
> > +
> > + return TRUE;
> > +}
> > +
> > /**
> > Does brief validation of the ext4 superblock.
> >
>
--
Pedro Falcato
next prev parent reply other threads:[~2021-09-13 16:36 UTC|newest]
Thread overview: 7+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-09-10 22:11 [PATCH v3 0/2] ExtPkg Updates Jeff Brasen
2021-09-10 22:11 ` [PATCH v3 1/2] Ext4Pkg: Improve Ext4IsBindingSupported() behavior Jeff Brasen
2021-09-12 10:40 ` [edk2-devel] " Marvin Häuser
2021-09-13 16:35 ` Pedro Falcato [this message]
2021-09-10 22:11 ` [PATCH v3 2/2] Ext4Pkg: Support uncleanly unmounted filesystems Jeff Brasen
2021-09-11 0:59 ` [PATCH v3 0/2] ExtPkg Updates Pedro Falcato
2021-09-11 1:02 ` Pedro Falcato
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-list from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to='CAKbZUD1q5kkA=7U6c45jFZT1neuua21B1ZVn6DMwfjPHbPzNPQ@mail.gmail.com' \
--to=devel@edk2.groups.io \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox