Skip to contentSkip to conversation

hw/usb/hcd-dwc2: fix divide-by-zero in dwc2_handle_packet()

PATCH Superseded Mauro Matteo Cascella 3 participants 9 messages Oct 12, 2020 – Oct 15, 2020

v1 of 2 Superseded Latest is v2 →

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 hw/usb/hcd-dwc2: fix divide-by-zero in dwc2_handle_packet()

Check the value of mps before it is used as divisor. Since HCCHAR_MPS is guest controllable, this prevents a malicious/buggy guest from crashing the QEMU process on the host.

---
 hw/usb/hcd-dwc2.c | 6 ++++++
 1 file changed, 6 insertions(+)
hw/usb/hcd-dwc2.c +6 -0
index 97688d21bf..91476fd781 100644
@@ -324,6 +324,12 @@ babble:
             }
         }
 
+        if (mps == 0) {
+            qemu_log_mask(LOG_GUEST_ERROR,
+                    "%s: Bad HCCHAR_MPS set to zero\n", __func__);
+            return;
+        }
+
         tpcnt = actual / mps;
         if (actual % mps) {
             tpcnt++;
Signature · 1 line
2.26.2
Reviewed-by on v1 1/1
#1 lore raw
Quoted 30 lines · Mauro Matteo Cascella

Check the value of mps before it is used as divisor. Since HCCHAR_MPS is guest controllable, this prevents a malicious/buggy guest from crashing the QEMU process on the host.

---
 hw/usb/hcd-dwc2.c | 6 ++++++
 1 file changed, 6 insertions(+)
hw/usb/hcd-dwc2.c +6 -1
index 97688d21bf..91476fd781 100644
@@ -324,6 +324,12 @@ babble:
             }
         }
 
+        if (mps == 0) {
+            qemu_log_mask(LOG_GUEST_ERROR,
+                    "%s: Bad HCCHAR_MPS set to zero\n", __func__);
+            return;
+        }
+
         tpcnt = actual / mps;
         if (actual % mps) {
             tpcnt++;
--
Quoted 2 lines

2.26.2

Hi Mauro,

I think it would be better to move this check earlier in the function, just after 'mps' is read from the register. Otherwise it can get assigned to 'tlen' and 'p->mps', and who knows what mischief an invalid value there might cause.

After that, you can add my

Thanks,
Paul

#2 lore raw
Quoted 10 lines

hw/usb/hcd-dwc2.c babble: · line 331

Quoted 5 lines
hw/usb/hcd-dwc2.c +5 -0
+        if (mps == 0) {
+            qemu_log_mask(LOG_GUEST_ERROR,
+                    "%s: Bad HCCHAR_MPS set to zero\n", __func__);
+            return;
+        }

I think it would be better to move this check earlier in the function, just after 'mps' is read from the register. Otherwise it can get assigned to 'tlen' and 'p->mps', and who knows what mischief an invalid value there might cause.

Makes sense. While being at it maybe handle len > DWC2_MAX_XFER_SIZE the same way, the assert looks like it can be triggered by the guest.

Also: What would be the effect of simply returning here? Would dwc2 emulation simply stop processing queues? Should we maybe raise an error IRQ? What will real dwc2 hardware do in this case?

take care,
Gerd

#3 lore raw

Hi Gerd,

Quoted 7 lines · Gerd Hoffmann
Quoted 6 lines
Quoted 6 lines
hw/usb/hcd-dwc2.c +5 -0
+        if (mps == 0) {
+            qemu_log_mask(LOG_GUEST_ERROR,
+                    "%s: Bad HCCHAR_MPS set to zero\n", __func__);
+            return;
+        }
Quoted 7 lines
Quoted 4 lines

I think it would be better to move this check earlier in the function, just after 'mps' is read from the register. Otherwise it can get assigned to 'tlen' and 'p->mps', and who knows what mischief an invalid value there might cause.

Makes sense. While being at it maybe handle len > DWC2_MAX_XFER_SIZE the same way, the assert looks like it can be triggered by the guest.

I sent you a patch to fix up several assert()s, including that one, about a month ago. Did you miss it?
https://lore.kernel.org/qemu-devel/[email protected]

Quoted 3 lines

Also: What would be the effect of simply returning here? Would dwc2 emulation simply stop processing queues? Should we maybe raise an error IRQ?

Not entirely sure, I imagine the emulation will just stop working. I can test it tomorrow. Also, can you give me a hint what an error IRQ is? Is that a Qemu thing, or do you mean we should emulate what the real core does in this case?

Quoted 1 line

What will real dwc2 hardware do in this case?

No idea. I don't think it's mentioned in the manual.

-Paul

Quoted 3 lines

take care,
Gerd

#4 lore raw
Quoted 3 lines · Paul Zimmerman

I sent you a patch to fix up several assert()s, including that one, about a month ago. Did you miss it?
https://lore.kernel.org/qemu-devel/[email protected]

I'll check. There is a bunch of pending stuff in my qemu-patches mailbox I'll try to process this week ...

Quoted 8 lines
Quoted 3 lines

Also: What would be the effect of simply returning here? Would dwc2 emulation simply stop processing queues? Should we maybe raise an error IRQ?

Not entirely sure, I imagine the emulation will just stop working. I can test it tomorrow. Also, can you give me a hint what an error IRQ is? Is that a Qemu thing, or do you mean we should emulate what the real core does in this case?

Same real hardware does. ehci for example has the USBSTS_ERRINT bit in the IRQ status register to signal errors.

take care,
Gerd

#5 lore raw
Quoted 18 lines · Gerd Hoffmann
Quoted 3 lines · Paul Zimmerman

I sent you a patch to fix up several assert()s, including that one, about a month ago. Did you miss it?
https://lore.kernel.org/qemu-devel/[email protected]

I'll check. There is a bunch of pending stuff in my qemu-patches mailbox I'll try to process this week ...

Quoted 8 lines
Quoted 3 lines

Also: What would be the effect of simply returning here? Would dwc2 emulation simply stop processing queues? Should we maybe raise an error IRQ?

Not entirely sure, I imagine the emulation will just stop working. I can test it tomorrow. Also, can you give me a hint what an error IRQ is? Is that a Qemu thing, or do you mean we should emulate what the real core does in this case?

Quoted 6 lines

Same real hardware does. ehci for example has the USBSTS_ERRINT bit in the IRQ status register to signal errors.

take care,
Gerd

I'll send a new version of the patch with the check moved earlier in the function, as suggested by Paul. If raising an error turns out to be the right thing to do, I think other checks may need to be updated in addition to this one. Hence we can possibly address that in a later patch. Thanks a lot for your comments.

Signature · 2 lines
Mauro Matteo Cascella, Red Hat Product Security
6F78 E20B 5935 928C F0A8  1A9D 4E55 23B8 BB34 10B0
#6 lore raw
Quoted 7 lines · Mauro Matteo Cascella
Quoted 5 lines · Gerd Hoffmann
Quoted 5 lines
Quoted 4 lines

Also: What would be the effect of simply returning here? Would dwc2 emulation simply stop processing queues? Should we maybe raise an error IRQ?

Quoted 7 lines
Quoted 7 lines
Quoted 4 lines

Not entirely sure, I imagine the emulation will just stop working. I can test it tomorrow. Also, can you give me a hint what an error IRQ is? Is that a Qemu thing, or do you mean we should emulate what the real core does in this case?

Same real hardware does. ehci for example has the USBSTS_ERRINT bit in the IRQ status register to signal errors.

So I tested this patch (modified as I suggested) in two ways. First, I hacked up the dwc_otg driver in the Raspbian kernel to always set the 'mps' field to zero. Then I booted the 2019-09-26-raspbian-buster Raspbian image with 4 usb devices, keyboard, tablet, storage, and network.

In this test, the first USB device failed to enumerate, giving the kernel error message "usb 1-1: device descriptor read/64, error -110". The kernel retried the enumeration several times, finally giving the message "usb usb1-port1: unable to enumerate usb device". After that, there was no more USB activity shown in the kernel log. Otherwise the kernel seemed to keep running normally and the raspi emulation continued to run.

In the second test, I changed the hack to only zero the 'mps' field for bulk transfers. In that case, the non-bulk devices (keyboard, tablet) continued to operate normally. The bulk devices (storage, network) enumerated fine, but after that they failed to work, with various error messages in the kernel log. Again the kernel seemed to keep running normally and the raspi emulation continued to run.

So I think the patch works fine, and I don't think we need to do anything fancier.

- Paul

Quoted 15 lines
Quoted 4 lines

take care,
Gerd

I'll send a new version of the patch with the check moved earlier in the function, as suggested by Paul. If raising an error turns out to be the right thing to do, I think other checks may need to be updated in addition to this one. Hence we can possibly address that in a later patch. Thanks a lot for your comments.

--
Mauro Matteo Cascella, Red Hat Product Security 6F78 E20B 5935 928C F0A8 1A9D 4E55 23B8 BB34 10B0

#7 lore raw

Hi,

Quoted 2 lines

So I think the patch works fine, and I don't think we need to do anything fancier.

Cool, thanks for checking.

take care,
Gerd