[PATCH] IB/hfi1: Fix potential use-after-free in PIO and SDMA map teardown

[PATCH] IB/hfi1: Fix potential use-after-free in PIO and SDMA map teardown

[lirongqing]

From: Li RongQing <lirongqing@baidu.com>

The current teardown logic for dd->pio_map and dd->sdma_map frees the
structures while they might still be accessed by RCU readers. Although
the pointer is nulled under a spinlock, the memory is reclaimed before
waiting for the grace period to end.

This patch fixes the sequence by:
1. Extracting the pointer under the lock.
2. Clearing the RCU-protected pointer.
3. Waiting for readers to finish with synchronize_rcu().
4. Finally freeing the memory.

Fixes: 7724105686e7 ("IB/hfi1: add driver files")
Signed-off-by: Li RongQing <lirongqing@baidu.com>
---
 drivers/infiniband/hw/hfi1/pio.c  | 5 ++++-
 drivers/infiniband/hw/hfi1/sdma.c | 4 +++-
 2 files changed, 7 insertions(+), 2 deletions(-)

diff --git a/drivers/infiniband/hw/hfi1/pio.c b/drivers/infiniband/hw/hfi1/pio.c
index 764286d..acf40f8 100644
--- a/drivers/infiniband/hw/hfi1/pio.c
+++ b/drivers/infiniband/hw/hfi1/pio.c
@@ -1946,13 +1946,16 @@ int pio_map_init(struct hfi1_devdata *dd, u8 port, u8 num_vls, u8 *vl_scontexts)
 
 void free_pio_map(struct hfi1_devdata *dd)
 {
+	struct pio_vl_map *map;
+
 	/* Free PIO map if allocated */
 	if (rcu_access_pointer(dd->pio_map)) {
 		spin_lock_irq(&dd->pio_map_lock);
-		pio_map_free(rcu_access_pointer(dd->pio_map));
+		map = rcu_access_pointer(dd->pio_map);
 		RCU_INIT_POINTER(dd->pio_map, NULL);
 		spin_unlock_irq(&dd->pio_map_lock);
 		synchronize_rcu();
+		pio_map_free(map);
 	}
 	kfree(dd->kernel_send_context);
 	dd->kernel_send_context = NULL;
diff --git a/drivers/infiniband/hw/hfi1/sdma.c b/drivers/infiniband/hw/hfi1/sdma.c
index 5cfa4f8..7c01616 100644
--- a/drivers/infiniband/hw/hfi1/sdma.c
+++ b/drivers/infiniband/hw/hfi1/sdma.c
@@ -1255,6 +1255,7 @@ void sdma_clean(struct hfi1_devdata *dd, size_t num_engines)
 {
 	size_t i;
 	struct sdma_engine *sde;
+	struct sdma_vl_map *map;
 
 	if (dd->sdma_pad_dma) {
 		dma_free_coherent(&dd->pcidev->dev, SDMA_PAD,
@@ -1291,10 +1292,11 @@ void sdma_clean(struct hfi1_devdata *dd, size_t num_engines)
 	}
 	if (rcu_access_pointer(dd->sdma_map)) {
 		spin_lock_irq(&dd->sde_map_lock);
-		sdma_map_free(rcu_access_pointer(dd->sdma_map));
+		map = rcu_access_pointer(dd->sdma_map);
 		RCU_INIT_POINTER(dd->sdma_map, NULL);
 		spin_unlock_irq(&dd->sde_map_lock);
 		synchronize_rcu();
+		sdma_map_free(map);
 	}
 	kfree(dd->per_sdma);
 	dd->per_sdma = NULL;
-- 
2.9.4

Re: [PATCH] IB/hfi1: Fix potential use-after-free in PIO and SDMA map teardown

[Jason Gunthorpe]

On Fri, Feb 06, 2026 at 12:08:36AM -0500, lirongqing wrote:
> From: Li RongQing <lirongqing@baidu.com>
> 
> The current teardown logic for dd->pio_map and dd->sdma_map frees the
> structures while they might still be accessed by RCU readers. Although
> the pointer is nulled under a spinlock, the memory is reclaimed before
> waiting for the grace period to end.
> 
> This patch fixes the sequence by:
> 1. Extracting the pointer under the lock.
> 2. Clearing the RCU-protected pointer.
> 3. Waiting for readers to finish with synchronize_rcu().
> 4. Finally freeing the memory.
> 
> Fixes: 7724105686e7 ("IB/hfi1: add driver files")
> Signed-off-by: Li RongQing <lirongqing@baidu.com>
> ---
>  drivers/infiniband/hw/hfi1/pio.c  | 5 ++++-
>  drivers/infiniband/hw/hfi1/sdma.c | 4 +++-
>  2 files changed, 7 insertions(+), 2 deletions(-)

This does seem to be a legitimate mis-use of RCU, applied to rc

Jason