Skip to contentSkip to conversation

usbip: vudc: Fix use after free bug in vudc_remove due to race condition

PATCH In review Zheng Wang 4 participants 8 messages Mar 17, 2023 – May 5

Subsystems (2)

v3 of 4 In review Latest is v4 →

lore
Share & tools
lore

Zheng Wang

author
#0 lore raw

PATCH v3 1/1 usbip: vudc: Fix use after free bug in vudc_remove due to race condition

In vudc_probe, it calls init_vudc_hw, which bound &udc->timer with v_timer.

When it calls usbip_sockfd_store, it will call v_start_timer to start the timer work.

When we call vudc_remove to remove the driver, theremay be a sequence as follows:

Fix it by shutdown the timer work before cleanup in vudc_remove.

Note that removing a driver is a root-only operation, and should never happen. But the attacker can directly unplug the usb to trigger the remove function.

CPU0 CPU1

                     |v_timer
vudc_remove          |
kfree(udc);          |
//free shost         |
                     |udc->gadget
                     |//use

The udc might be removed before v_timer finished, and UAF happens.

This bug was found by Codeql static analysis and might by false positive.

Fixes b6a0ca111867 ("usbip: vudc: Add UDC specific ops")

---
v3:
- fix the issue by adding del_timer_sync in v_stop_timer and invoke it in vudc_remove

v2:
- add more details about how the bug was found suggested by Shuah

---
 drivers/usb/usbip/vudc_dev.c      | 1 +
 drivers/usb/usbip/vudc_transfer.c | 2 ++
 2 files changed, 3 insertions(+)
drivers/usb/usbip/vudc_dev.c +1 -0
index 2bc428f2e261..dcbfed30806d 100644
@@ -633,6 +633,7 @@ int vudc_remove(struct platform_device *pdev)
 {
 	struct vudc *udc = platform_get_drvdata(pdev);
 
+	v_stop_timer(udc);
 	usb_del_gadget_udc(&udc->gadget);
 	cleanup_vudc_hw(udc);
 	kfree(udc);
drivers/usb/usbip/vudc_transfer.c +2 -0
index 7e801fee33bf..562ea7b6ea2e 100644
@@ -492,5 +492,7 @@ void v_stop_timer(struct vudc *udc)
 
 	/* timer itself will take care of stopping */
 	dev_dbg(&udc->pdev->dev, "timer stop");
+	
+	del_timer_sync(&t->timer);
 	t->state = VUDC_TR_STOPPED;
 }
Signature · 1 line
2.25.1
maintainer
#1 lore raw
Quoted 23 lines · Zheng Wang

In vudc_probe, it calls init_vudc_hw, which bound &udc->timer with v_timer.

When it calls usbip_sockfd_store, it will call v_start_timer to start the timer work.

When we call vudc_remove to remove the driver, theremay be a sequence as follows:

Fix it by shutdown the timer work before cleanup in vudc_remove.

Note that removing a driver is a root-only operation, and should never happen. But the attacker can directly unplug the usb to trigger the remove function.

CPU0 CPU1

|v_timer

vudc_remove          |
kfree(udc);          |
//free shost         |
                      |udc->gadget
                      |//use
Quoted 3 lines

The udc might be removed before v_timer finished, and UAF happens.

This bug was found by Codeql static analysis and might by false positive.

This statement that this could be a false positive makes me hesitate taking this patch.

What kind of testing have you done with this fix? Were you able to test the scenario of unplugging usb?

thanks,
-- Shuah

#2 lore raw
Quoted 27 lines · Shuah Khan <[email protected]> 于2023年3月18日周六 06:53写道:
Quoted 25 lines · Zheng Wang

In vudc_probe, it calls init_vudc_hw, which bound &udc->timer with v_timer.

When it calls usbip_sockfd_store, it will call v_start_timer to start the timer work.

When we call vudc_remove to remove the driver, theremay be a sequence as follows:

Fix it by shutdown the timer work before cleanup in vudc_remove.

Note that removing a driver is a root-only operation, and should never happen. But the attacker can directly unplug the usb to trigger the remove function.

CPU0 CPU1

|v_timer

vudc_remove          |
kfree(udc);          |
//free shost         |
                      |udc->gadget
                      |//use

The udc might be removed before v_timer finished, and UAF happens.

Quoted 8 lines
Quoted 1 line

This bug was found by Codeql static analysis and might by false positive.

This statement that this could be a false positive makes me hesitate taking this patch.

What kind of testing have you done with this fix? Were you able to test the scenario of unplugging usb?

Sorry I did't make a full test for I did't have the device. The attacking scenario if based on other cases.

Best regads,
Zheng

Quoted 2 lines

thanks,
-- Shuah

#3 lore raw

Friendly ping about the issue.
Sorry that I couldn't make test about the driver.

Thanks,
Zheng

Quoted 46 lines · Zheng Hacker <[email protected]> 于2023年3月18日周六 15:39写道:
Quoted 27 lines · Shuah Khan <[email protected]> 于2023年3月18日周六 06:53写道:
Quoted 25 lines · Zheng Wang

In vudc_probe, it calls init_vudc_hw, which bound &udc->timer with v_timer.

When it calls usbip_sockfd_store, it will call v_start_timer to start the timer work.

When we call vudc_remove to remove the driver, theremay be a sequence as follows:

Fix it by shutdown the timer work before cleanup in vudc_remove.

Note that removing a driver is a root-only operation, and should never happen. But the attacker can directly unplug the usb to trigger the remove function.

CPU0 CPU1

|v_timer

vudc_remove          |
kfree(udc);          |
//free shost         |
                      |udc->gadget
                      |//use

The udc might be removed before v_timer finished, and UAF happens.

Quoted 8 lines
Quoted 1 line

This bug was found by Codeql static analysis and might by false positive.

This statement that this could be a false positive makes me hesitate taking this patch.

What kind of testing have you done with this fix? Were you able to test the scenario of unplugging usb?

Sorry I did't make a full test for I did't have the device. The attacking scenario if based on other cases.

Best regads,
Zheng

Quoted 2 lines

thanks,
-- Shuah

maintainer
#4 lore raw
Quoted 48 lines · Zheng Hacker

Friendly ping about the issue.
Sorry that I couldn't make test about the driver.

Thanks,
Zheng

Quoted 41 lines · Zheng Hacker <[email protected]> 于2023年3月18日周六 15:39写道:
Quoted 27 lines · Shuah Khan <[email protected]> 于2023年3月18日周六 06:53写道:
Quoted 25 lines · Zheng Wang

In vudc_probe, it calls init_vudc_hw, which bound &udc->timer with v_timer.

When it calls usbip_sockfd_store, it will call v_start_timer to start the timer work.

When we call vudc_remove to remove the driver, theremay be a sequence as follows:

Fix it by shutdown the timer work before cleanup in vudc_remove.

Note that removing a driver is a root-only operation, and should never happen. But the attacker can directly unplug the usb to trigger the remove function.

CPU0 CPU1

|v_timer

vudc_remove          |
kfree(udc);          |
//free shost         |
                       |udc->gadget
                       |//use

The udc might be removed before v_timer finished, and UAF happens.

Quoted 8 lines
Quoted 1 line

This bug was found by Codeql static analysis and might by false positive.

This statement that this could be a false positive makes me hesitate taking this patch.

What kind of testing have you done with this fix? Were you able to test the scenario of unplugging usb?

Sorry I did't make a full test for I did't have the device. The attacking scenario if based on other cases.

Sorry. I really need for you test this and provide information on how it was tested.

thanks,
-- Shuah

#5 lore raw

Hi Shuah,

I got it. I'll try it.

Best regards,
Zheng

Quoted 57 lines · Shuah Khan <[email protected]> 于2023年4月14日周五 02:01写道:
Quoted 48 lines · Zheng Hacker

Friendly ping about the issue.
Sorry that I couldn't make test about the driver.

Thanks,
Zheng

Quoted 41 lines · Zheng Hacker <[email protected]> 于2023年3月18日周六 15:39写道:
Quoted 27 lines · Shuah Khan <[email protected]> 于2023年3月18日周六 06:53写道:
Quoted 25 lines · Zheng Wang

In vudc_probe, it calls init_vudc_hw, which bound &udc->timer with v_timer.

When it calls usbip_sockfd_store, it will call v_start_timer to start the timer work.

When we call vudc_remove to remove the driver, theremay be a sequence as follows:

Fix it by shutdown the timer work before cleanup in vudc_remove.

Note that removing a driver is a root-only operation, and should never happen. But the attacker can directly unplug the usb to trigger the remove function.

CPU0 CPU1

|v_timer

vudc_remove          |
kfree(udc);          |
//free shost         |
                       |udc->gadget
                       |//use

The udc might be removed before v_timer finished, and UAF happens.

Quoted 8 lines
Quoted 1 line

This bug was found by Codeql static analysis and might by false positive.

This statement that this could be a false positive makes me hesitate taking this patch.

What kind of testing have you done with this fix? Were you able to test the scenario of unplugging usb?

Sorry I did't make a full test for I did't have the device. The attacking scenario if based on other cases.

Sorry. I really need for you test this and provide information on how it was tested.

thanks,
-- Shuah

#6 lore raw

PATCH 1/1 usbip: vudc: Fix use after free bug in vudc_remove due to race condition

In reply to Zheng Wang

This patch follows up Zheng Wang's 2023 report of a use-after-free in vudc_remove(). The original thread stalled on Shuah Khan's request for runtime testing of the unplug/unbind path. This patch supplies that testing and keeps Zheng's original fix shape.

In vudc_probe(), v_init_timer() binds udc->tr_timer.timer to v_timer(). usbip_sockfd_store() starts the timer via v_start_timer()/v_kick_timer(). vudc_remove() can then free the containing struct vudc while the timer is still pending or executing.

KASAN confirms the race on an unpatched x86_64 QEMU guest with CONFIG_KASAN=y, CONFIG_USBIP_VUDC=y, CONFIG_USB_ZERO=y, and a tight loop that repeatedly writes a socket fd to usbip_sockfd, closes the socket pair, and unbinds/rebinds usbip-vudc.0:

  BUG: KASAN: slab-use-after-free in __run_timer_base.part.0+0x8ba/0x8e0
  Write of size 8 at addr ffff888001b80740 by task trigger_and_unb/239
  Allocated by task 239:
    vudc_probe+0x4d/0xaa0
  Freed by task 239:
    kfree+0x18f/0x520
    device_release_driver_internal+0x388/0x540
    unbind_store+0xd9/0x100

This lands in the timer core rather than v_timer() itself because the embedded timer_list is being walked after its containing struct vudc has already been freed. The underlying lifetime bug is the same one Zheng reported.

With v_stop_timer() called from vudc_remove() and the timer deleted synchronously, the same harness completed 5000 bind/unbind iterations with no KASAN report.

Fixes b6a0ca111867 ("usbip: vudc: Add UDC specific ops")
---
 drivers/usb/usbip/vudc_dev.c      | 1 +
 drivers/usb/usbip/vudc_transfer.c | 3 ++-
 2 files changed, 3 insertions(+), 1 deletion(-)
drivers/usb/usbip/vudc_dev.c +1 -0
index 90383107b660..c5f079c5a1ea 100644
@@ -632,6 +632,7 @@ void vudc_remove(struct platform_device *pdev)
 {
 	struct vudc *udc = platform_get_drvdata(pdev);
 
+	v_stop_timer(udc);
 	usb_del_gadget_udc(&udc->gadget);
 	cleanup_vudc_hw(udc);
 	kfree(udc);
drivers/usb/usbip/vudc_transfer.c +2 -1
index a4f02ea3e3ef..d4ce85c4c6a2 100644
@@ -490,7 +490,8 @@ void v_stop_timer(struct vudc *udc)
 {
 	struct transfer_timer *t = &udc->tr_timer;
 
-	/* timer itself will take care of stopping */
+	/* Delete the timer synchronously before teardown frees udc. */
 	dev_dbg(&udc->pdev->dev, "timer stop");
+	timer_delete_sync(&t->timer);
 	t->state = VUDC_TR_STOPPED;
 }
Signature · 1 line
2.53.0
maintainer Acked-by on v1 1/1
#7 lore raw
Quoted 59 lines · Michael Bommarito

This patch follows up Zheng Wang's 2023 report of a use-after-free in vudc_remove(). The original thread stalled on Shuah Khan's request for runtime testing of the unplug/unbind path. This patch supplies that testing and keeps Zheng's original fix shape.

In vudc_probe(), v_init_timer() binds udc->tr_timer.timer to v_timer(). usbip_sockfd_store() starts the timer via v_start_timer()/v_kick_timer(). vudc_remove() can then free the containing struct vudc while the timer is still pending or executing.

KASAN confirms the race on an unpatched x86_64 QEMU guest with CONFIG_KASAN=y, CONFIG_USBIP_VUDC=y, CONFIG_USB_ZERO=y, and a tight loop that repeatedly writes a socket fd to usbip_sockfd, closes the socket pair, and unbinds/rebinds usbip-vudc.0:

   BUG: KASAN: slab-use-after-free in __run_timer_base.part.0+0x8ba/0x8e0
   Write of size 8 at addr ffff888001b80740 by task trigger_and_unb/239
   Allocated by task 239:
     vudc_probe+0x4d/0xaa0
   Freed by task 239:
     kfree+0x18f/0x520
     device_release_driver_internal+0x388/0x540
     unbind_store+0xd9/0x100

This lands in the timer core rather than v_timer() itself because the embedded timer_list is being walked after its containing struct vudc has already been freed. The underlying lifetime bug is the same one Zheng reported.

With v_stop_timer() called from vudc_remove() and the timer deleted synchronously, the same harness completed 5000 bind/unbind iterations with no KASAN report.

Fixes b6a0ca111867 ("usbip: vudc: Add UDC specific ops")
---
  drivers/usb/usbip/vudc_dev.c      | 1 +
  drivers/usb/usbip/vudc_transfer.c | 3 ++-
  2 files changed, 3 insertions(+), 1 deletion(-)
drivers/usb/usbip/vudc_dev.c +1 -0
index 90383107b660..c5f079c5a1ea 100644
@@ -632,6 +632,7 @@ void vudc_remove(struct platform_device *pdev)
  {
  	struct vudc *udc = platform_get_drvdata(pdev);
 
+	v_stop_timer(udc);
  	usb_del_gadget_udc(&udc->gadget);
  	cleanup_vudc_hw(udc);
  	kfree(udc);
drivers/usb/usbip/vudc_transfer.c +0 -0
index a4f02ea3e3ef..d4ce85c4c6a2 100644

drivers/usb/usbip/vudc_transfer.c void v_stop_timer(struct vudc *udc) · line 497

Quoted 9 lines
drivers/usb/usbip/vudc_transfer.c +2 -1
@@ -490,7 +490,8 @@ void v_stop_timer(struct vudc *udc)
  {
  	struct transfer_timer *t = &udc->tr_timer;
 
-	/* timer itself will take care of stopping */
+	/* Delete the timer synchronously before teardown frees udc. */
  	dev_dbg(&udc->pdev->dev, "timer stop");
+	timer_delete_sync(&t->timer);
  	t->state = VUDC_TR_STOPPED;
  }

Looks good to me.

thanks,
-- Shuah