2024-01-12 05:34:16

by Deren Wu

Subject: Re: [BUG] Unloading mt7921e module cause use-after-free

On Wed, 2024-01-10 at 20:15 +0500, Mikhail Gavrilov wrote:
>
> Greetings,
> For bug reproduction just type:
> # rmmod mt7921e
>
> Backtrace:
> BUG: KASAN: use-after-free in
> tasklet_action_common.isra.0+0x6a4/0x7a0
> Read of size 8 at addr ffff888146806748 by task ksoftirqd/5/48
> CPU: 5 PID: 48 Comm: ksoftirqd/5 Tainted: G W L -------
> --- 6.8.0-0.rc0.20240109git9f8413c4a66f.1.fc40.x86_64+debug #1
> Hardware name: Micro-Star International Co., Ltd. MS-7D73/MPG B650I
> EDGE WIFI (MS-7D73), BIOS 1.81 01/05/2024
> Call Trace:
> <TASK>
> dump_stack_lvl+0x76/0xd0
> print_report+0xcf/0x670
> ? tasklet_action_common.isra.0+0x6a4/0x7a0
> kasan_report+0xa6/0xe0
> ? tasklet_action_common.isra.0+0x6a4/0x7a0
> tasklet_action_common.isra.0+0x6a4/0x7a0
> __do_softirq+0x215/0x8b9
> ? __pfx___do_softirq+0x10/0x10
> ? run_ksoftirqd+0x73/0x80
> ? __pfx_run_ksoftirqd+0x10/0x10
> run_ksoftirqd+0x4b/0x80
> smpboot_thread_fn+0x56d/0x900
> ? __kthread_parkme+0xbd/0x1f0
> ? __pfx_smpboot_thread_fn+0x10/0x10
> kthread+0x2f2/0x3d0
> ? _raw_spin_unlock_irq+0x28/0x60
> ? __pfx_kthread+0x10/0x10
> ret_from_fork+0x31/0x70
> ? __pfx_kthread+0x10/0x10
> ret_from_fork_asm+0x1b/0x30
> </TASK>
>

Hi Mikhail,

I can reproduce a similar problem when CONFI_DEBUG_SHIRQ is enabled.

Pleaese try this patch in mainline kernel (6.7+)
---
--- a/drivers/net/wireless/mediatek/mt76/mt7921/pci.c
+++ b/drivers/net/wireless/mediatek/mt76/mt7921/pci.c
@@ -387,6 +387,7 @@ static void mt7921_pci_remove(struct pci_dev *pdev)
struct mt792x_dev *dev = container_of(mdev, struct mt792x_dev,
mt76);

mt7921e_unregister_device(dev);
+ set_bit(MT76_REMOVED, &mdev->phy.state);
devm_free_irq(&pdev->dev, pdev->irq, dev);
mt76_free_device(&dev->mt76);
pci_free_irq_vectors(pdev);
diff --git a/drivers/net/wireless/mediatek/mt76/mt792x_dma.c
b/drivers/net/wireless/mediatek/mt76/mt792x_dma.c
index 488326ce5ed4..3893dbe866fe 100644
--- a/drivers/net/wireless/mediatek/mt76/mt792x_dma.c
+++ b/drivers/net/wireless/mediatek/mt76/mt792x_dma.c
@@ -12,6 +12,8 @@ irqreturn_t mt792x_irq_handler(int irq, void
*dev_instance)
{
struct mt792x_dev *dev = dev_instance;

+ if (test_bit(MT76_REMOVED, &dev->mt76.phy.state))
+ return IRQ_NONE;
mt76_wr(dev, dev->irq_map->host_irq_enable, 0);

if (!test_bit(MT76_STATE_INITIALIZED, &dev->mphy.state))

2024-01-12 07:31:29

by Mikhail Gavrilov

Subject: Re: [BUG] Unloading mt7921e module cause use-after-free

On Fri, Jan 12, 2024 at 10:34 AM Deren Wu (武德仁) <[email protected]> wrote:
>
> Hi Mikhail,
>
> I can reproduce a similar problem when CONFI_DEBUG_SHIRQ is enabled.
>
> Pleaese try this patch in mainline kernel (6.7+)
> ---
> --- a/drivers/net/wireless/mediatek/mt76/mt7921/pci.c
> +++ b/drivers/net/wireless/mediatek/mt76/mt7921/pci.c
> @@ -387,6 +387,7 @@ static void mt7921_pci_remove(struct pci_dev *pdev)
> struct mt792x_dev *dev = container_of(mdev, struct mt792x_dev,
> mt76);
>
> mt7921e_unregister_device(dev);
> + set_bit(MT76_REMOVED, &mdev->phy.state);
> devm_free_irq(&pdev->dev, pdev->irq, dev);
> mt76_free_device(&dev->mt76);
> pci_free_irq_vectors(pdev);
> diff --git a/drivers/net/wireless/mediatek/mt76/mt792x_dma.c
> b/drivers/net/wireless/mediatek/mt76/mt792x_dma.c
> index 488326ce5ed4..3893dbe866fe 100644
> --- a/drivers/net/wireless/mediatek/mt76/mt792x_dma.c
> +++ b/drivers/net/wireless/mediatek/mt76/mt792x_dma.c
> @@ -12,6 +12,8 @@ irqreturn_t mt792x_irq_handler(int irq, void
> *dev_instance)
> {
> struct mt792x_dev *dev = dev_instance;
>
> + if (test_bit(MT76_REMOVED, &dev->mt76.phy.state))
> + return IRQ_NONE;
> mt76_wr(dev, dev->irq_map->host_irq_enable, 0);
>
> if (!test_bit(MT76_STATE_INITIALIZED, &dev->mphy.state))
>
>

Thanks, this patch looks good to me.
Demonstration: https://youtu.be/nKnA2ftVoXw

Tested-by: Mikhail Gavrilov <[email protected]>

--
Best Regards,
Mike Gavrilov.