2022-01-07 07:34:29

by Dan Carpenter

Subject: [PATCH] ACPI: pfr_telemetry: Fix info leak in pfrt_log_ioctl()

The "data_info" struct is copied to the user. It has a 4 byte struct
hole after the last struct member so we need to memset that to avoid
copying uninitialized stack data to the user.

Fixes: b0013e037a8b ("ACPI: Introduce Platform Firmware Runtime Telemetry driver")
Signed-off-by: Dan Carpenter <[email protected]>
---
When you're adding a new driver to the kernel then please use the new
driver's prefix instead of just the subsystem prefix.

Bad: ACPI: Introduce Platform Firmware Runtime Telemetry driver
Good: ACPI / pfr_telemetry: Introduce Platform Firmware Runtime Telemetry driver

Otherwise it's just up to me to guess what prefix you wanted.

drivers/acpi/pfr_telemetry.c | 1 +
1 file changed, 1 insertion(+)

diff --git a/drivers/acpi/pfr_telemetry.c b/drivers/acpi/pfr_telemetry.c
index da50dd80192c..9abf350bd7a5 100644
--- a/drivers/acpi/pfr_telemetry.c
+++ b/drivers/acpi/pfr_telemetry.c
@@ -83,6 +83,7 @@ static int get_pfrt_log_data_info(struct pfrt_log_data_info *data_info,
union acpi_object *out_obj, in_obj, in_buf;
int ret = -EBUSY;

+ memset(data_info, 0, sizeof(*data_info));
memset(&in_obj, 0, sizeof(in_obj));
memset(&in_buf, 0, sizeof(in_buf));
in_obj.type = ACPI_TYPE_PACKAGE;
--
2.20.1

2022-01-07 13:46:59

by Chen Yu

Subject: Re: [PATCH] ACPI: pfr_telemetry: Fix info leak in pfrt_log_ioctl()

On Fri, Jan 07, 2022 at 10:34:07AM +0300, Dan Carpenter wrote:
> The "data_info" struct is copied to the user. It has a 4 byte struct
> hole after the last struct member so we need to memset that to avoid
> copying uninitialized stack data to the user.
>
> Fixes: b0013e037a8b ("ACPI: Introduce Platform Firmware Runtime Telemetry driver")
> Signed-off-by: Dan Carpenter <[email protected]>
> ---
> When you're adding a new driver to the kernel then please use the new
> driver's prefix instead of just the subsystem prefix.
>
> Bad: ACPI: Introduce Platform Firmware Runtime Telemetry driver
> Good: ACPI / pfr_telemetry: Introduce Platform Firmware Runtime Telemetry driver
>
Thanks for pointing this out.
> Otherwise it's just up to me to guess what prefix you wanted.
>
> drivers/acpi/pfr_telemetry.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/drivers/acpi/pfr_telemetry.c b/drivers/acpi/pfr_telemetry.c
> index da50dd80192c..9abf350bd7a5 100644
> --- a/drivers/acpi/pfr_telemetry.c
> +++ b/drivers/acpi/pfr_telemetry.c
> @@ -83,6 +83,7 @@ static int get_pfrt_log_data_info(struct pfrt_log_data_info *data_info,
> union acpi_object *out_obj, in_obj, in_buf;
> int ret = -EBUSY;
>
> + memset(data_info, 0, sizeof(*data_info));
Just one minor question, how about moving above before:
data_info->status = out_obj->package.elements[LOG_STATUS_IDX].integer.value;
after the sanity check of the _DSM result?

thanks,
Chenyu
> memset(&in_obj, 0, sizeof(in_obj));
> memset(&in_buf, 0, sizeof(in_buf));
> in_obj.type = ACPI_TYPE_PACKAGE;

2022-01-10 06:17:42

by Dan Carpenter

Subject: Re: [PATCH] ACPI: pfr_telemetry: Fix info leak in pfrt_log_ioctl()

On Fri, Jan 07, 2022 at 09:46:17PM +0800, Chen Yu wrote:
> On Fri, Jan 07, 2022 at 10:34:07AM +0300, Dan Carpenter wrote:
> > The "data_info" struct is copied to the user. It has a 4 byte struct
> > hole after the last struct member so we need to memset that to avoid
> > copying uninitialized stack data to the user.
> >
> > Fixes: b0013e037a8b ("ACPI: Introduce Platform Firmware Runtime Telemetry driver")
> > Signed-off-by: Dan Carpenter <[email protected]>
> > ---
> > When you're adding a new driver to the kernel then please use the new
> > driver's prefix instead of just the subsystem prefix.
> >
> > Bad: ACPI: Introduce Platform Firmware Runtime Telemetry driver
> > Good: ACPI / pfr_telemetry: Introduce Platform Firmware Runtime Telemetry driver
> >
> Thanks for pointing this out.
> > Otherwise it's just up to me to guess what prefix you wanted.
> >
> > drivers/acpi/pfr_telemetry.c | 1 +
> > 1 file changed, 1 insertion(+)
> >
> > diff --git a/drivers/acpi/pfr_telemetry.c b/drivers/acpi/pfr_telemetry.c
> > index da50dd80192c..9abf350bd7a5 100644
> > --- a/drivers/acpi/pfr_telemetry.c
> > +++ b/drivers/acpi/pfr_telemetry.c
> > @@ -83,6 +83,7 @@ static int get_pfrt_log_data_info(struct pfrt_log_data_info *data_info,
> > union acpi_object *out_obj, in_obj, in_buf;
> > int ret = -EBUSY;
> >
> > + memset(data_info, 0, sizeof(*data_info));
> Just one minor question, how about moving above before:
> data_info->status = out_obj->package.elements[LOG_STATUS_IDX].integer.value;
> after the sanity check of the _DSM result?

I guess I wanted to keep all the memsets together. I feel like if the
data is invalid, then it's going to be a slow path and it's not worth
optimizing that case. If the data is invalid then a little slow down is
the least of our concerns.

regards,
dan carpenter

2022-01-10 15:39:46

by Rafael J. Wysocki

Subject: Re: [PATCH] ACPI: pfr_telemetry: Fix info leak in pfrt_log_ioctl()

On Mon, Jan 10, 2022 at 7:17 AM Dan Carpenter <[email protected]> wrote:
>
> On Fri, Jan 07, 2022 at 09:46:17PM +0800, Chen Yu wrote:
> > On Fri, Jan 07, 2022 at 10:34:07AM +0300, Dan Carpenter wrote:
> > > The "data_info" struct is copied to the user. It has a 4 byte struct
> > > hole after the last struct member so we need to memset that to avoid
> > > copying uninitialized stack data to the user.
> > >
> > > Fixes: b0013e037a8b ("ACPI: Introduce Platform Firmware Runtime Telemetry driver")
> > > Signed-off-by: Dan Carpenter <[email protected]>
> > > ---
> > > When you're adding a new driver to the kernel then please use the new
> > > driver's prefix instead of just the subsystem prefix.
> > >
> > > Bad: ACPI: Introduce Platform Firmware Runtime Telemetry driver
> > > Good: ACPI / pfr_telemetry: Introduce Platform Firmware Runtime Telemetry driver
> > >
> > Thanks for pointing this out.
> > > Otherwise it's just up to me to guess what prefix you wanted.
> > >
> > > drivers/acpi/pfr_telemetry.c | 1 +
> > > 1 file changed, 1 insertion(+)
> > >
> > > diff --git a/drivers/acpi/pfr_telemetry.c b/drivers/acpi/pfr_telemetry.c
> > > index da50dd80192c..9abf350bd7a5 100644
> > > --- a/drivers/acpi/pfr_telemetry.c
> > > +++ b/drivers/acpi/pfr_telemetry.c
> > > @@ -83,6 +83,7 @@ static int get_pfrt_log_data_info(struct pfrt_log_data_info *data_info,
> > > union acpi_object *out_obj, in_obj, in_buf;
> > > int ret = -EBUSY;
> > >
> > > + memset(data_info, 0, sizeof(*data_info));
> > Just one minor question, how about moving above before:
> > data_info->status = out_obj->package.elements[LOG_STATUS_IDX].integer.value;
> > after the sanity check of the _DSM result?
>
> I guess I wanted to keep all the memsets together. I feel like if the
> data is invalid, then it's going to be a slow path and it's not worth
> optimizing that case. If the data is invalid then a little slow down is
> the least of our concerns.

Patch applied, thanks!

Yu, this series needs to spend a few days more in linux-next, because
of the fixes against it sent lately.

2022-01-11 00:56:40

by Chen Yu

Subject: Re: [PATCH] ACPI: pfr_telemetry: Fix info leak in pfrt_log_ioctl()

On Mon, Jan 10, 2022 at 09:17:13AM +0300, Dan Carpenter wrote:
> On Fri, Jan 07, 2022 at 09:46:17PM +0800, Chen Yu wrote:
> > On Fri, Jan 07, 2022 at 10:34:07AM +0300, Dan Carpenter wrote:
> > > The "data_info" struct is copied to the user. It has a 4 byte struct
> > > hole after the last struct member so we need to memset that to avoid
> > > copying uninitialized stack data to the user.
> > >
> > > Fixes: b0013e037a8b ("ACPI: Introduce Platform Firmware Runtime Telemetry driver")
> > > Signed-off-by: Dan Carpenter <[email protected]>
> > > ---
> > > When you're adding a new driver to the kernel then please use the new
> > > driver's prefix instead of just the subsystem prefix.
> > >
> > > Bad: ACPI: Introduce Platform Firmware Runtime Telemetry driver
> > > Good: ACPI / pfr_telemetry: Introduce Platform Firmware Runtime Telemetry driver
> > >
> > Thanks for pointing this out.
> > > Otherwise it's just up to me to guess what prefix you wanted.
> > >
> > > drivers/acpi/pfr_telemetry.c | 1 +
> > > 1 file changed, 1 insertion(+)
> > >
> > > diff --git a/drivers/acpi/pfr_telemetry.c b/drivers/acpi/pfr_telemetry.c
> > > index da50dd80192c..9abf350bd7a5 100644
> > > --- a/drivers/acpi/pfr_telemetry.c
> > > +++ b/drivers/acpi/pfr_telemetry.c
> > > @@ -83,6 +83,7 @@ static int get_pfrt_log_data_info(struct pfrt_log_data_info *data_info,
> > > union acpi_object *out_obj, in_obj, in_buf;
> > > int ret = -EBUSY;
> > >
> > > + memset(data_info, 0, sizeof(*data_info));
> > Just one minor question, how about moving above before:
> > data_info->status = out_obj->package.elements[LOG_STATUS_IDX].integer.value;
> > after the sanity check of the _DSM result?
>
> I guess I wanted to keep all the memsets together. I feel like if the
> data is invalid, then it's going to be a slow path and it's not worth
> optimizing that case. If the data is invalid then a little slow down is
> the least of our concerns.
>
Ok, got it.

thanks,
Chenyu
>

2022-01-11 01:10:32

by Chen Yu

Subject: Re: [PATCH] ACPI: pfr_telemetry: Fix info leak in pfrt_log_ioctl()

On Mon, Jan 10, 2022 at 04:39:30PM +0100, Rafael J. Wysocki wrote:
> On Mon, Jan 10, 2022 at 7:17 AM Dan Carpenter <[email protected]> wrote:
> >
> > On Fri, Jan 07, 2022 at 09:46:17PM +0800, Chen Yu wrote:
> > > On Fri, Jan 07, 2022 at 10:34:07AM +0300, Dan Carpenter wrote:
> > > > The "data_info" struct is copied to the user. It has a 4 byte struct
> > > > hole after the last struct member so we need to memset that to avoid
> > > > copying uninitialized stack data to the user.
> > > >
> > > > Fixes: b0013e037a8b ("ACPI: Introduce Platform Firmware Runtime Telemetry driver")
> > > > Signed-off-by: Dan Carpenter <[email protected]>
> > > > ---
> > > > When you're adding a new driver to the kernel then please use the new
> > > > driver's prefix instead of just the subsystem prefix.
> > > >
> > > > Bad: ACPI: Introduce Platform Firmware Runtime Telemetry driver
> > > > Good: ACPI / pfr_telemetry: Introduce Platform Firmware Runtime Telemetry driver
> > > >
> > > Thanks for pointing this out.
> > > > Otherwise it's just up to me to guess what prefix you wanted.
> > > >
> > > > drivers/acpi/pfr_telemetry.c | 1 +
> > > > 1 file changed, 1 insertion(+)
> > > >
> > > > diff --git a/drivers/acpi/pfr_telemetry.c b/drivers/acpi/pfr_telemetry.c
> > > > index da50dd80192c..9abf350bd7a5 100644
> > > > --- a/drivers/acpi/pfr_telemetry.c
> > > > +++ b/drivers/acpi/pfr_telemetry.c
> > > > @@ -83,6 +83,7 @@ static int get_pfrt_log_data_info(struct pfrt_log_data_info *data_info,
> > > > union acpi_object *out_obj, in_obj, in_buf;
> > > > int ret = -EBUSY;
> > > >
> > > > + memset(data_info, 0, sizeof(*data_info));
> > > Just one minor question, how about moving above before:
> > > data_info->status = out_obj->package.elements[LOG_STATUS_IDX].integer.value;
> > > after the sanity check of the _DSM result?
> >
> > I guess I wanted to keep all the memsets together. I feel like if the
> > data is invalid, then it's going to be a slow path and it's not worth
> > optimizing that case. If the data is invalid then a little slow down is
> > the least of our concerns.
>
> Patch applied, thanks!
>
> Yu, this series needs to spend a few days more in linux-next, because
> of the fixes against it sent lately.
Ok, I see.

thanks,
Chenyu