2022-11-23 05:32:29

by butt3rflyh4ck

[permalink] [raw]
Subject: [RESEND PATCH] ALSA: rawmidi: fix infoleak in snd_rawmidi_ioctl_status_compat64

The compat_status is declared off of the stack, so it needs to
be zeroed out before copied back to userspace to prevent any
unintentional data leakage.

Fixes: d9e5582c4bb2 ("ALSA: Avoid using timespec for struct snd_rawmidi_status")
Signed-off-by: Xiaolong Huang <[email protected]>

---

Reason for resend:
1. add Fixes line.
---
sound/core/rawmidi_compat.c | 1 +
1 file changed, 1 insertion(+)

diff --git a/sound/core/rawmidi_compat.c b/sound/core/rawmidi_compat.c
index 68a93443583c..6afa68165b17 100644
--- a/sound/core/rawmidi_compat.c
+++ b/sound/core/rawmidi_compat.c
@@ -80,6 +80,7 @@ static int snd_rawmidi_ioctl_status_compat64(struct snd_rawmidi_file *rfile,
if (err < 0)
return err;

+ memset(&compat_status, 0, sizeof(compat_status));
compat_status = (struct compat_snd_rawmidi_status64) {
.stream = status.stream,
.tstamp_sec = status.tstamp_sec,

base-commit: eb7081409f94a9a8608593d0fb63a1aa3d6f95d8
--
2.25.1


2022-11-23 07:30:02

by Takashi Iwai

[permalink] [raw]
Subject: Re: [RESEND PATCH] ALSA: rawmidi: fix infoleak in snd_rawmidi_ioctl_status_compat64

On Wed, 23 Nov 2022 06:09:11 +0100,
Xiaolong Huang wrote:
>
> The compat_status is declared off of the stack, so it needs to
> be zeroed out before copied back to userspace to prevent any
> unintentional data leakage.
>
> Fixes: d9e5582c4bb2 ("ALSA: Avoid using timespec for struct snd_rawmidi_status")
> Signed-off-by: Xiaolong Huang <[email protected]>
>
> ---
>
> Reason for resend:
> 1. add Fixes line.
> ---
> sound/core/rawmidi_compat.c | 1 +
> 1 file changed, 1 insertion(+)
>
> diff --git a/sound/core/rawmidi_compat.c b/sound/core/rawmidi_compat.c
> index 68a93443583c..6afa68165b17 100644
> --- a/sound/core/rawmidi_compat.c
> +++ b/sound/core/rawmidi_compat.c
> @@ -80,6 +80,7 @@ static int snd_rawmidi_ioctl_status_compat64(struct snd_rawmidi_file *rfile,
> if (err < 0)
> return err;
>
> + memset(&compat_status, 0, sizeof(compat_status));
> compat_status = (struct compat_snd_rawmidi_status64) {
> .stream = status.stream,
> .tstamp_sec = status.tstamp_sec,

Here at the line just after your addition, compat_status is fully
initialized by substitution, so I believe the memset is superfluous.

Or have you verified that it really leaks the uninitialized memory?


thanks,

Takashi

2022-11-23 07:45:00

by butt3rflyh4ck

[permalink] [raw]
Subject: Re: [RESEND PATCH] ALSA: rawmidi: fix infoleak in snd_rawmidi_ioctl_status_compat64

Hi, the compat_status structure is struct compat_snd_rawmidi_status64 {
s32 stream;
u8 rsvd[4]; /* alignment */
s64 tstamp_sec;
s64 tstamp_nsec;
u32 avail;
u32 xruns;
unsigned char reserved[16];
} __attribute__((packed));
The rsvd[4] and reserved[16] are not initialized.


Regards,
butt3rflyh4ck.

On Wed, Nov 23, 2022 at 2:55 PM Takashi Iwai <[email protected]> wrote:
>
> On Wed, 23 Nov 2022 06:09:11 +0100,
> Xiaolong Huang wrote:
> >
> > The compat_status is declared off of the stack, so it needs to
> > be zeroed out before copied back to userspace to prevent any
> > unintentional data leakage.
> >
> > Fixes: d9e5582c4bb2 ("ALSA: Avoid using timespec for struct snd_rawmidi_status")
> > Signed-off-by: Xiaolong Huang <[email protected]>
> >
> > ---
> >
> > Reason for resend:
> > 1. add Fixes line.
> > ---
> > sound/core/rawmidi_compat.c | 1 +
> > 1 file changed, 1 insertion(+)
> >
> > diff --git a/sound/core/rawmidi_compat.c b/sound/core/rawmidi_compat.c
> > index 68a93443583c..6afa68165b17 100644
> > --- a/sound/core/rawmidi_compat.c
> > +++ b/sound/core/rawmidi_compat.c
> > @@ -80,6 +80,7 @@ static int snd_rawmidi_ioctl_status_compat64(struct snd_rawmidi_file *rfile,
> > if (err < 0)
> > return err;
> >
> > + memset(&compat_status, 0, sizeof(compat_status));
> > compat_status = (struct compat_snd_rawmidi_status64) {
> > .stream = status.stream,
> > .tstamp_sec = status.tstamp_sec,
>
> Here at the line just after your addition, compat_status is fully
> initialized by substitution, so I believe the memset is superfluous.
>
> Or have you verified that it really leaks the uninitialized memory?
>
>
> thanks,
>
> Takashi



--
Active Defense Lab of Venustech

2022-11-23 07:47:59

by Takashi Iwai

[permalink] [raw]
Subject: Re: [RESEND PATCH] ALSA: rawmidi: fix infoleak in snd_rawmidi_ioctl_status_compat64

On Wed, 23 Nov 2022 08:06:16 +0100,
butt3rflyh4ck wrote:
>
> Hi, the compat_status structure is struct compat_snd_rawmidi_status64 {
> s32 stream;
> u8 rsvd[4]; /* alignment */
> s64 tstamp_sec;
> s64 tstamp_nsec;
> u32 avail;
> u32 xruns;
> unsigned char reserved[16];
> } __attribute__((packed));
> The rsvd[4] and reserved[16] are not initialized.

Other members are initialized with zero.
https://gcc.gnu.org/onlinedocs/gcc/Designated-Inits.html


Takashi

>
>
> Regards,
> butt3rflyh4ck.
>
> On Wed, Nov 23, 2022 at 2:55 PM Takashi Iwai <[email protected]> wrote:
> >
> > On Wed, 23 Nov 2022 06:09:11 +0100,
> > Xiaolong Huang wrote:
> > >
> > > The compat_status is declared off of the stack, so it needs to
> > > be zeroed out before copied back to userspace to prevent any
> > > unintentional data leakage.
> > >
> > > Fixes: d9e5582c4bb2 ("ALSA: Avoid using timespec for struct snd_rawmidi_status")
> > > Signed-off-by: Xiaolong Huang <[email protected]>
> > >
> > > ---
> > >
> > > Reason for resend:
> > > 1. add Fixes line.
> > > ---
> > > sound/core/rawmidi_compat.c | 1 +
> > > 1 file changed, 1 insertion(+)
> > >
> > > diff --git a/sound/core/rawmidi_compat.c b/sound/core/rawmidi_compat.c
> > > index 68a93443583c..6afa68165b17 100644
> > > --- a/sound/core/rawmidi_compat.c
> > > +++ b/sound/core/rawmidi_compat.c
> > > @@ -80,6 +80,7 @@ static int snd_rawmidi_ioctl_status_compat64(struct snd_rawmidi_file *rfile,
> > > if (err < 0)
> > > return err;
> > >
> > > + memset(&compat_status, 0, sizeof(compat_status));
> > > compat_status = (struct compat_snd_rawmidi_status64) {
> > > .stream = status.stream,
> > > .tstamp_sec = status.tstamp_sec,
> >
> > Here at the line just after your addition, compat_status is fully
> > initialized by substitution, so I believe the memset is superfluous.
> >
> > Or have you verified that it really leaks the uninitialized memory?
> >
> >
> > thanks,
> >
> > Takashi
>
>
>
> --
> Active Defense Lab of Venustech
>