[Qemu-devel] [PATCH] lsi53c895a: check message length value

[Qemu-devel] [PATCH] lsi53c895a: check message length value

[P J P]

From: Prasad J Pandit <pjp@fedoraproject.org>

While writing a message in 'lsi_do_msgin', message length value
in msg_len could be invalid, add check to avoid OOB access issue.

Reported-by: Ameya More <ameya.more@oracle.com>
Signed-off-by: Prasad J Pandit <pjp@fedoraproject.org>
---
 hw/scsi/lsi53c895a.c | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/hw/scsi/lsi53c895a.c b/hw/scsi/lsi53c895a.c
index d1e6534311..a266c5a113 100644
--- a/hw/scsi/lsi53c895a.c
+++ b/hw/scsi/lsi53c895a.c
@@ -205,7 +205,7 @@ typedef struct {
     /* Action to take at the end of a MSG IN phase.
        0 = COMMAND, 1 = disconnect, 2 = DATA OUT, 3 = DATA IN.  */
     int msg_action;
-    int msg_len;
+    uint8_t msg_len;
     uint8_t msg[LSI_MAX_MSGIN_LEN];
     /* 0 if SCRIPTS are running or stopped.
      * 1 if a Wait Reselect instruction has been issued.
@@ -861,12 +861,15 @@ static void lsi_do_status(LSIState *s)
 
 static void lsi_do_msgin(LSIState *s)
 {
-    int len;
+    uint8_t len;
     trace_lsi_do_msgin(s->dbc, s->msg_len);
     s->sfbr = s->msg[0];
     len = s->msg_len;
     if (len > s->dbc)
         len = s->dbc;
+    if (len > LSI_MAX_MSGIN_LEN) {
+        len = LSI_MAX_MSGIN_LEN;
+    }
     pci_dma_write(PCI_DEVICE(s), s->dnad, s->msg, len);
     /* Linux drivers rely on the last byte being in the SIDL.  */
     s->sidl = s->msg[len - 1];
@@ -2114,7 +2117,7 @@ static const VMStateDescription vmstate_lsi_scsi = {
         VMSTATE_INT32(carry, LSIState),
         VMSTATE_INT32(status, LSIState),
         VMSTATE_INT32(msg_action, LSIState),
-        VMSTATE_INT32(msg_len, LSIState),
+        VMSTATE_UINT8(msg_len, LSIState),
         VMSTATE_BUFFER(msg, LSIState),
         VMSTATE_INT32(waiting, LSIState),
 
-- 
2.17.2

Re: [Qemu-devel] [PATCH] lsi53c895a: check message length value

[Ameya More]

Hi Prasad,

Thanks for following up on this. While Mark and I reported this issue to 
you, it was actually discovered by Dejvau Security and they should 
receive credit for reporting this issue.
http://www.dejavusecurity.com

Thanks,
Ameya


On 10/25/2018 03:09 PM, P J P wrote:
> From: Prasad J Pandit <pjp@fedoraproject.org>
>
> While writing a message in 'lsi_do_msgin', message length value
> in msg_len could be invalid, add check to avoid OOB access issue.
>
> Reported-by: Ameya More <ameya.more@oracle.com>
> Signed-off-by: Prasad J Pandit <pjp@fedoraproject.org>
> ---
>   hw/scsi/lsi53c895a.c | 9 ++++++---
>   1 file changed, 6 insertions(+), 3 deletions(-)
>
> diff --git a/hw/scsi/lsi53c895a.c b/hw/scsi/lsi53c895a.c
> index d1e6534311..a266c5a113 100644
> --- a/hw/scsi/lsi53c895a.c
> +++ b/hw/scsi/lsi53c895a.c
> @@ -205,7 +205,7 @@ typedef struct {
>       /* Action to take at the end of a MSG IN phase.
>          0 = COMMAND, 1 = disconnect, 2 = DATA OUT, 3 = DATA IN.  */
>       int msg_action;
> -    int msg_len;
> +    uint8_t msg_len;
>       uint8_t msg[LSI_MAX_MSGIN_LEN];
>       /* 0 if SCRIPTS are running or stopped.
>        * 1 if a Wait Reselect instruction has been issued.
> @@ -861,12 +861,15 @@ static void lsi_do_status(LSIState *s)
>   
>   static void lsi_do_msgin(LSIState *s)
>   {
> -    int len;
> +    uint8_t len;
>       trace_lsi_do_msgin(s->dbc, s->msg_len);
>       s->sfbr = s->msg[0];
>       len = s->msg_len;
>       if (len > s->dbc)
>           len = s->dbc;
> +    if (len > LSI_MAX_MSGIN_LEN) {
> +        len = LSI_MAX_MSGIN_LEN;
> +    }
>       pci_dma_write(PCI_DEVICE(s), s->dnad, s->msg, len);
>       /* Linux drivers rely on the last byte being in the SIDL.  */
>       s->sidl = s->msg[len - 1];
> @@ -2114,7 +2117,7 @@ static const VMStateDescription vmstate_lsi_scsi = {
>           VMSTATE_INT32(carry, LSIState),
>           VMSTATE_INT32(status, LSIState),
>           VMSTATE_INT32(msg_action, LSIState),
> -        VMSTATE_INT32(msg_len, LSIState),
> +        VMSTATE_UINT8(msg_len, LSIState),
>           VMSTATE_BUFFER(msg, LSIState),
>           VMSTATE_INT32(waiting, LSIState),
>   

Re: [Qemu-devel] [PATCH] lsi53c895a: check message length value

[P J P]

+-- On Thu, 25 Oct 2018, Ameya More wrote --+
| While Mark and I reported this issue to you, it was actually discovered by 
| Dejvau Security and they should receive credit for reporting this issue. 
| http://www.dejavusecurity.com

I see; Would it be possible to share email-id of the original reporter to 
include in the commit log message?

Thank you.
--
Prasad J Pandit / Red Hat Product Security Team
47AF CE69 3A90 54AA 9045 1053 DD13 3D32 FE5B 041F

Re: [Qemu-devel] [PATCH] lsi53c895a: check message length value

[Mark Kanda]

On 10/26/2018 4:25 AM, P J P wrote:
> +-- On Thu, 25 Oct 2018, Ameya More wrote --+
> | While Mark and I reported this issue to you, it was actually discovered by
> | Dejvau Security and they should receive credit for reporting this issue.
> | http://www.dejavusecurity.com
> 
> I see; Would it be possible to share email-id of the original reporter to
> include in the commit log message?

Deja vu requested that we include the following text in the commit message:

	Discovered by Deja vu Security. Reported by Oracle.

Would that be acceptable?

Thanks,

-Mark

Re: [Qemu-devel] [PATCH] lsi53c895a: check message length value

[P J P]

+-- On Fri, 26 Oct 2018, Mark Kanda wrote --+
| Deja vu requested that we include the following text in the commit message:
| 
| 	Discovered by Deja vu Security. Reported by Oracle.
| 
| Would that be acceptable?

Generally an email-id is used/preferred in the commit log message. We could 
use above for acknowledgement and avoid Reported-by in the commit log message 
if that suits Deja vu team.

Please let me know your/their preference.

Thank you.
--
Prasad J Pandit / Red Hat Product Security Team
47AF CE69 3A90 54AA 9045 1053 DD13 3D32 FE5B 041F

Re: [Qemu-devel] [PATCH] lsi53c895a: check message length value

[Mark Kanda]

On 10/26/2018 1:37 PM, P J P wrote:
> +-- On Fri, 26 Oct 2018, Mark Kanda wrote --+
> | Deja vu requested that we include the following text in the commit message:
> |
> | 	Discovered by Deja vu Security. Reported by Oracle.
> |
> | Would that be acceptable?
> 
> Generally an email-id is used/preferred in the commit log message. We could
> use above for acknowledgement and avoid Reported-by in the commit log message
> if that suits Deja vu team.
> 
> Please let me know your/their preference.
> 

Yes, please use that acknowledgement text in lieu of a 'Reported-by' line.

Thanks,

-Mark

Re: [Qemu-devel] [PATCH] lsi53c895a: check message length value

[P J P]

+-- On Fri, 26 Oct 2018, Mark Kanda wrote --+
| Yes, please use that acknowledgement text in lieu of a 'Reported-by' line.

Okay, thank you.
--
Prasad J Pandit / Red Hat Product Security Team
47AF CE69 3A90 54AA 9045 1053 DD13 3D32 FE5B 041F

Re: [Qemu-devel] [PATCH] lsi53c895a: check message length value

[Paolo Bonzini]

On 25/10/2018 22:09, P J P wrote:
> From: Prasad J Pandit <pjp@fedoraproject.org>
> 
> While writing a message in 'lsi_do_msgin', message length value
> in msg_len could be invalid, add check to avoid OOB access issue.
> 
> Reported-by: Ameya More <ameya.more@oracle.com>
> Signed-off-by: Prasad J Pandit <pjp@fedoraproject.org>
> ---
>  hw/scsi/lsi53c895a.c | 9 ++++++---
>  1 file changed, 6 insertions(+), 3 deletions(-)
> 
> diff --git a/hw/scsi/lsi53c895a.c b/hw/scsi/lsi53c895a.c
> index d1e6534311..a266c5a113 100644
> --- a/hw/scsi/lsi53c895a.c
> +++ b/hw/scsi/lsi53c895a.c
> @@ -205,7 +205,7 @@ typedef struct {
>      /* Action to take at the end of a MSG IN phase.
>         0 = COMMAND, 1 = disconnect, 2 = DATA OUT, 3 = DATA IN.  */
>      int msg_action;
> -    int msg_len;
> +    uint8_t msg_len;

Not wrong per se, but it's also not clear why it's needed.  I understand
that you want to switch from signed to unsigned, but it is not mentioned
in the commit message.

The switch to 8-bit, and below from VMSTATE_INT32 to VMSTATE_UINT8, is
wrong because it changes the format of the live migration stream.

>      uint8_t msg[LSI_MAX_MSGIN_LEN];
>      /* 0 if SCRIPTS are running or stopped.
>       * 1 if a Wait Reselect instruction has been issued.
> @@ -861,12 +861,15 @@ static void lsi_do_status(LSIState *s)
>  
>  static void lsi_do_msgin(LSIState *s)
>  {
> -    int len;
> +    uint8_t len;
>      trace_lsi_do_msgin(s->dbc, s->msg_len);
>      s->sfbr = s->msg[0];
>      len = s->msg_len;
>      if (len > s->dbc)
>          len = s->dbc;
> +    if (len > LSI_MAX_MSGIN_LEN) {
> +        len = LSI_MAX_MSGIN_LEN;
> +    }

I'm not sure it's appropriate to check for out of bounds reads here,
because if s->msg_len is greater than LSI_MAX_MSGIN_LEN, then this test
doesn't exclude you've already had an out of bounds write before.
Indeed the msg_len is checked in lsi_add_msg_byte in order to avoid out
of bounds accesses in either lsi_add_msg_byte or lsi_do_msgin.  You
could assert here that the variable is in range, I guess.

However, the out of bounds s->msg_len can actually happen in one other
case: namely, if a malicious live migration stream includes a bogus
s->msg_len.  Such live migration stream should be rejected; the fix for
that is to add a lsi_post_load function, point to it in
vmstate_lsi_scsi, and check there for s->msg_len <= LSI_MAX_MSGIN_LEN.

Thanks,

Paolo

>      pci_dma_write(PCI_DEVICE(s), s->dnad, s->msg, len);
>      /* Linux drivers rely on the last byte being in the SIDL.  */
>      s->sidl = s->msg[len - 1];
> @@ -2114,7 +2117,7 @@ static const VMStateDescription vmstate_lsi_scsi = {
>          VMSTATE_INT32(carry, LSIState),
>          VMSTATE_INT32(status, LSIState),
>          VMSTATE_INT32(msg_action, LSIState),
> -        VMSTATE_INT32(msg_len, LSIState),
> +        VMSTATE_UINT8(msg_len, LSIState),
>          VMSTATE_BUFFER(msg, LSIState),
>          VMSTATE_INT32(waiting, LSIState),
>  
> 

Re: [Qemu-devel] [PATCH] lsi53c895a: check message length value

[P J P]

+-- On Fri, 26 Oct 2018, Paolo Bonzini wrote --+
| > -    int msg_len;
| > +    uint8_t msg_len;
| 
| Not wrong per se, but it's also not clear why it's needed.  I understand
| that you want to switch from signed to unsigned, but it is not mentioned
| in the commit message.

Changed to uint8_t because IIUC 'msg_len' is likely be < LSI_MAX_MSGIN_LEN=8. 
And uint32_t seems rather large for it.
 
| The switch to 8-bit, and below from VMSTATE_INT32 to VMSTATE_UINT8, is
| wrong because it changes the format of the live migration stream.

I see.

| I'm not sure it's appropriate to check for out of bounds reads here,
| because if s->msg_len is greater than LSI_MAX_MSGIN_LEN, then this test
| doesn't exclude you've already had an out of bounds write before.
| Indeed the msg_len is checked in lsi_add_msg_byte in order to avoid out
| of bounds accesses in either lsi_add_msg_byte or lsi_do_msgin. You
| could assert here that the variable is in range, I guess.

Okay.

| However, the out of bounds s->msg_len can actually happen in one other
| case: namely, if a malicious live migration stream includes a bogus
| s->msg_len.  Such live migration stream should be rejected; the fix for
| that is to add a lsi_post_load function, point to it in
| vmstate_lsi_scsi, and check there for s->msg_len <= LSI_MAX_MSGIN_LEN.

Okay, sending a revised patch v1 in a bit.

Thank you.
--
Prasad J Pandit / Red Hat Product Security Team
47AF CE69 3A90 54AA 9045 1053 DD13 3D32 FE5B 041F