Re: [PATCH v4] dma: at_hdmac: Fix use-after-free by proper tasklet cleanup
From: Hongling Zeng
Date: Wed Jul 08 2026 - 22:42:32 EST
在 2026年07月09日 05:32, Frank Li 写道:
On Wed, Jul 08, 2026 at 10:59:59AM +0800, Hongling Zeng wrote:Thank you for the suggestion. I've updated to v6 following your advice:
Current cleanup paths have a use-after-free vulnerability:I forget the reason why need goto here. Can you call disable_irq() or
- vchan_init() creates tasklets that access at_dma_chan memory
- free_irq() only waits for IRQ handler, NOT tasklets
- atdma is devm-managed and freed after probe/remove
- Running tasklets accessing freed memory → Use-After-Free!
The fix requires careful ordering:
- free_irq() FIRST to synchronize with running IRQ handlers and prevent
them from scheduling new tasklets
- Then kill tasklets to wait for already-scheduled ones to complete
- Only then free other resources
Fixes: ac803b56860f ("dmaengine: at_hdmac: Convert driver to use virt-dma")
Reported-by: sashiko-bot@xxxxxxxxxx
Closes: https://lore.kernel.org/all/20260604073945.54B311F00898@xxxxxxxxxxxxxxx/
Signed-off-by: Hongling Zeng <zenghongling@xxxxxxxxxx>
---
Change in v4:
- Fix error path fallthrough causing double-free_irq()
- Use channel iteration index (chan_id not initialized before registration)
- Remove unnecessary defensive checks
---
drivers/dma/at_hdmac.c | 31 ++++++++++++++++++++++---------
1 file changed, 22 insertions(+), 9 deletions(-)
diff --git a/drivers/dma/at_hdmac.c b/drivers/dma/at_hdmac.c
index e5b30a57c477..044a0fb38b7a 100644
--- a/drivers/dma/at_hdmac.c
+++ b/drivers/dma/at_hdmac.c
@@ -1940,6 +1940,20 @@ static void at_dma_off(struct at_dma *atdma)
cpu_relax();
}
+static void at_dma_cleanup_channels(struct at_dma *atdma)
+{
+ struct dma_chan *chan, *_chan;
+ int i = 0;
+
+ list_for_each_entry_safe(chan, _chan, &atdma->dma_device.channels,
+ device_node) {
+ /* Disable interrupts */
+ atc_disable_chan_irq(atdma, i++);
+ tasklet_kill(&to_at_dma_chan(chan)->vc.task);
+ list_del(&chan->device_node);
+ }
+}
+
static int __init at_dma_probe(struct platform_device *pdev)
{
struct at_dma *atdma;
@@ -2105,12 +2119,17 @@ static int __init at_dma_probe(struct platform_device *pdev)
err_of_dma_controller_register:
dma_async_device_unregister(&atdma->dma_device);
err_dma_async_device_register:
+ free_irq(platform_get_irq(pdev, 0), atdma);
+ at_dma_cleanup_channels(atdma);
dma_pool_destroy(atdma->memset_pool);
+ dma_pool_destroy(atdma->lli_pool);
+ goto err_clk;
disable hardware irq and call synchronize_irq() at free_irq() place. then
goto can fallback to below clean up code
Frank
- Replaced free_irq() with disable_irq() to allow fallthrough without goto
- free_irq() is now called only once at err_desc_pool_create label
The new approach ensures IRQ handler cannot schedule tasklets before tasklet_kill(), then free_irq() handles the actual release.
Please help to review again.
Thanks!
err_memset_pool_create:
dma_pool_destroy(atdma->lli_pool);
err_desc_pool_create:
free_irq(platform_get_irq(pdev, 0), atdma);
err_irq:
+err_clk:
clk_disable_unprepare(atdma->clk);
return err;
}
@@ -2118,23 +2137,17 @@ static int __init at_dma_probe(struct platform_device *pdev)
static void at_dma_remove(struct platform_device *pdev)
{
struct at_dma *atdma = platform_get_drvdata(pdev);
- struct dma_chan *chan, *_chan;
at_dma_off(atdma);
if (pdev->dev.of_node)
of_dma_controller_free(pdev->dev.of_node);
dma_async_device_unregister(&atdma->dma_device);
- dma_pool_destroy(atdma->memset_pool);
- dma_pool_destroy(atdma->lli_pool);
free_irq(platform_get_irq(pdev, 0), atdma);
- list_for_each_entry_safe(chan, _chan, &atdma->dma_device.channels,
- device_node) {
- /* Disable interrupts */
- atc_disable_chan_irq(atdma, chan->chan_id);
- list_del(&chan->device_node);
- }
+ at_dma_cleanup_channels(atdma);
+ dma_pool_destroy(atdma->memset_pool);
+ dma_pool_destroy(atdma->lli_pool);
clk_disable_unprepare(atdma->clk);
}
--
2.25.1