RDMA/mlx5: issue with error checking

RDMA/mlx5: issue with error checking

[Colin King (gmail)]

[-- Attachment #1.1.1: Type: text/plain, Size: 1266 bytes --]

Hi

Static analysis detected an issue with the following commit:

commit e73242aa14d2ec7f4a1a13688366bb36dc0fe5b7
Author: Edward Srouji <edwards@nvidia.com>
Date:   Wed Jul 9 09:42:11 2025 +0300

     RDMA/mlx5: Optimize DMABUF mkey page size


The issue is as follows:

int mlx5r_umr_dmabuf_update_pgsz(struct mlx5_ib_mr *mr, u32 xlt_flags,
                                  unsigned int page_shift)
{
         unsigned int old_page_shift = mr->page_shift;
         size_t zapped_blocks;
         size_t total_blocks;
         int err;

         zapped_blocks = _mlx5r_umr_zap_mkey(mr, xlt_flags, page_shift,
                                             mr->data_direct);
         if (zapped_blocks < 0)
                 return zapped_blocks;

The variable zapped_blocks is a size_t type and is being assigned a int 
return value from the call to _mlx5r_umr_zap_mkey. Since zapped_blocks 
is an unsigned type, the error check for zapped_blocks < 0 will never be 
true.  I suspect total_blocks should be a ssize_t type, but that 
probably also means total_blocks should be ssize_t too, but don't have 
the hardware to test this fix and I'm concerned that this change may 
break the code. Hence I'm reporting this issue.

Colin



[-- Attachment #1.1.2: OpenPGP public key --]
[-- Type: application/pgp-keys, Size: 4901 bytes --]

[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 840 bytes --]

Re: RDMA/mlx5: issue with error checking

[Leon Romanovsky]

On Thu, Jul 17, 2025 at 12:36:06PM +0100, Colin King (gmail) wrote:
> Hi
> 
> Static analysis detected an issue with the following commit:
> 
> commit e73242aa14d2ec7f4a1a13688366bb36dc0fe5b7
> Author: Edward Srouji <edwards@nvidia.com>
> Date:   Wed Jul 9 09:42:11 2025 +0300
> 
>     RDMA/mlx5: Optimize DMABUF mkey page size
> 
> 
> The issue is as follows:
> 
> int mlx5r_umr_dmabuf_update_pgsz(struct mlx5_ib_mr *mr, u32 xlt_flags,
>                                  unsigned int page_shift)
> {
>         unsigned int old_page_shift = mr->page_shift;
>         size_t zapped_blocks;
>         size_t total_blocks;
>         int err;
> 
>         zapped_blocks = _mlx5r_umr_zap_mkey(mr, xlt_flags, page_shift,
>                                             mr->data_direct);
>         if (zapped_blocks < 0)
>                 return zapped_blocks;
> 
> The variable zapped_blocks is a size_t type and is being assigned a int
> return value from the call to _mlx5r_umr_zap_mkey. Since zapped_blocks is an
> unsigned type, the error check for zapped_blocks < 0 will never be true.  I
> suspect total_blocks should be a ssize_t type, but that probably also means
> total_blocks should be ssize_t too, but don't have the hardware to test this
> fix and I'm concerned that this change may break the code. Hence I'm
> reporting this issue.

Thanks for the report, this will probably fix it.

diff --git a/drivers/infiniband/hw/mlx5/umr.c b/drivers/infiniband/hw/mlx5/umr.c
index 993d24f35ccbe..a27bee9a38138 100644
--- a/drivers/infiniband/hw/mlx5/umr.c
+++ b/drivers/infiniband/hw/mlx5/umr.c
@@ -996,6 +996,7 @@ _mlx5r_dmabuf_umr_update_pas(struct mlx5_ib_mr *mr, unsigned int flags,
 static int _mlx5r_umr_zap_mkey(struct mlx5_ib_mr *mr,
 			       unsigned int flags,
 			       unsigned int page_shift,
+			       size_t *nblocks,
 			       bool dd)
 {
 	unsigned int old_page_shift = mr->page_shift;
@@ -1004,7 +1005,6 @@ static int _mlx5r_umr_zap_mkey(struct mlx5_ib_mr *mr,
 	size_t page_shift_nblocks;
 	unsigned int max_log_size;
 	int access_mode;
-	size_t nblocks;
 	int err;
 
 	access_mode = dd ? MLX5_MKC_ACCESS_MODE_KSM : MLX5_MKC_ACCESS_MODE_MTT;
@@ -1018,26 +1018,26 @@ static int _mlx5r_umr_zap_mkey(struct mlx5_ib_mr *mr,
 	 * Block size must be aligned to MLX5_UMR_FLEX_ALIGNMENT since it may
 	 * be used as offset into the XLT later on.
 	 */
-	nblocks = ib_umem_num_dma_blocks(mr->umem, 1UL << max_page_shift);
+	*nblocks = ib_umem_num_dma_blocks(mr->umem, 1UL << max_page_shift);
 	if (dd)
-		nblocks = ALIGN(nblocks, MLX5_UMR_KSM_NUM_ENTRIES_ALIGNMENT);
+		*nblocks = ALIGN(*nblocks, MLX5_UMR_KSM_NUM_ENTRIES_ALIGNMENT);
 	else
-		nblocks = ALIGN(nblocks, MLX5_UMR_MTT_NUM_ENTRIES_ALIGNMENT);
+		*nblocks = ALIGN(*nblocks, MLX5_UMR_MTT_NUM_ENTRIES_ALIGNMENT);
 	page_shift_nblocks = ib_umem_num_dma_blocks(mr->umem,
 						    1UL << page_shift);
 	/* If the number of blocks at max possible page shift is greater than
 	 * the number of blocks at the new page size, we should just go over the
 	 * whole mkey entries.
 	 */
-	if (nblocks >= page_shift_nblocks)
-		nblocks = 0;
+	if (*nblocks >= page_shift_nblocks)
+		*nblocks = 0;
 
 	/* Make the first nblocks entries non-present without changing
 	 * page size yet.
 	 */
-	if (nblocks)
+	if (*nblocks)
 		mr->page_shift = max_page_shift;
-	err = _mlx5r_dmabuf_umr_update_pas(mr, flags, 0, nblocks, dd);
+	err = _mlx5r_dmabuf_umr_update_pas(mr, flags, 0, *nblocks, dd);
 	if (err) {
 		mr->page_shift = old_page_shift;
 		return err;
@@ -1046,7 +1046,7 @@ static int _mlx5r_umr_zap_mkey(struct mlx5_ib_mr *mr,
 	/* Change page size to the max page size now that the MR is completely
 	 * non-present.
 	 */
-	if (nblocks) {
+	if (*nblocks) {
 		err = mlx5r_umr_update_mr_page_shift(mr, max_page_shift, dd);
 		if (err) {
 			mr->page_shift = old_page_shift;
@@ -1054,7 +1054,7 @@ static int _mlx5r_umr_zap_mkey(struct mlx5_ib_mr *mr,
 		}
 	}
 
-	return err ? err : nblocks;
+	return 0;
 }
 
 /**
@@ -1089,10 +1089,10 @@ int mlx5r_umr_dmabuf_update_pgsz(struct mlx5_ib_mr *mr, u32 xlt_flags,
 	size_t total_blocks;
 	int err;
 
-	zapped_blocks = _mlx5r_umr_zap_mkey(mr, xlt_flags, page_shift,
-					    mr->data_direct);
-	if (zapped_blocks < 0)
-		return zapped_blocks;
+	err = _mlx5r_umr_zap_mkey(mr, xlt_flags, page_shift, &zapped_blocks,
+				  mr->data_direct);
+	if (err)
+		return err;
 
 	/* _mlx5r_umr_zap_mkey already enables the mkey */
 	xlt_flags &= ~MLX5_IB_UPD_XLT_ENABLE;

> 
> Colin
> 
>