Skip to contentSkip to conversation

RDMA/irdma: Fix out-of-bounds write in irdma_copy_user_pgaddrs

PATCH Applied Jacob Moroni 2 participants 3 messages May 12 – May 26

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 RDMA/irdma: Fix out-of-bounds write in irdma_copy_user_pgaddrs

The irdma_copy_user_pgaddrs function loops through all of the umem DMA blocks to populate the PBLEs and will stop when either the last DMA block is reached or palloc->total_cnt is reached. The issue is that the logic for checking palloc->total_cnt would only work for non-zero values.

When irdma_setup_pbles is called with lvl==0, it
calls irdma_copy_user_pgaddrs with palloc->total_cnt==0, which means the only way to break out of the loop is to reach the last umem DMA block, which means it could end up going beyond the fixed size of 4 iwmr->pgaddrmem array that is used in the lvl==0 case.

In the case of QP/CQ/SRQ rings, the value of lvl is determined by a separate input (for example, req.cq_pages in the case of a CQ). So, we must perform explicit checking to ensure we don't overflow the pgaddrmem array if the user provides a umem that consists of more blocks than their provided req.cq_pages.

Fixes b48c24c2d710 ("RDMA/irdma: Implement device supported verb APIs")
---
 drivers/infiniband/hw/irdma/verbs.c | 14 ++++++++++----
 1 file changed, 10 insertions(+), 4 deletions(-)
drivers/infiniband/hw/irdma/verbs.c +10 -4
index 17086048d2d7..b30e81d2b933 100644
@@ -2781,10 +2781,11 @@ static inline u64 *irdma_next_pbl_addr(u64 *pbl, struct irdma_pble_info **pinfo,
  * irdma_copy_user_pgaddrs - copy user page address to pble's os locally
  * @iwmr: iwmr for IB's user page addresses
  * @pbl: ple pointer to save 1 level or 0 level pble
+ * @pbl_len: Max number of PBL entries to populate
  * @level: indicated level 0, 1 or 2
  */
 static void irdma_copy_user_pgaddrs(struct irdma_mr *iwmr, u64 *pbl,
-				    enum irdma_pble_level level)
+				    u32 pbl_len, enum irdma_pble_level level)
 {
 	struct ib_umem *region = iwmr->region;
 	struct irdma_pbl *iwpbl = &iwmr->iwpbl;
@@ -2792,7 +2793,9 @@ static void irdma_copy_user_pgaddrs(struct irdma_mr *iwmr, u64 *pbl,
 	struct irdma_pble_info *pinfo;
 	struct ib_block_iter biter;
 	u32 idx = 0;
-	u32 pbl_cnt = 0;
+
+	if (!pbl_len)
+		return;
 
 	pinfo = (level == PBLE_LEVEL_1) ? NULL : palloc->level2.leaf;
 
@@ -2801,7 +2804,7 @@ static void irdma_copy_user_pgaddrs(struct irdma_mr *iwmr, u64 *pbl,
 
 	rdma_umem_for_each_dma_block(region, &biter, iwmr->page_size) {
 		*pbl = rdma_block_iter_dma_address(&biter);
-		if (++pbl_cnt == palloc->total_cnt)
+		if (!--pbl_len)
 			break;
 		pbl = irdma_next_pbl_addr(pbl, &pinfo, &idx);
 	}
@@ -2877,6 +2880,7 @@ static int irdma_setup_pbles(struct irdma_pci_f *rf, struct irdma_mr *iwmr,
 	u64 *pbl;
 	int status;
 	enum irdma_pble_level level = PBLE_LEVEL_1;
+	u32 pbl_len;
 
 	if (lvl) {
 		status = irdma_get_pble(rf->pble_rsrc, palloc, iwmr->page_cnt,
@@ -2884,16 +2888,18 @@ static int irdma_setup_pbles(struct irdma_pci_f *rf, struct irdma_mr *iwmr,
 		if (status)
 			return status;
 
+		pbl_len = palloc->total_cnt;
 		iwpbl->pbl_allocated = true;
 		level = palloc->level;
 		pinfo = (level == PBLE_LEVEL_1) ? &palloc->level1 :
 						  palloc->level2.leaf;
 		pbl = pinfo->addr;
 	} else {
+		pbl_len = IRDMA_MAX_SAVED_PHY_PGADDR;
 		pbl = iwmr->pgaddrmem;
 	}
 
-	irdma_copy_user_pgaddrs(iwmr, pbl, level);
+	irdma_copy_user_pgaddrs(iwmr, pbl, pbl_len, level);
 
 	if (lvl)
 		iwmr->pgaddrmem[0] = *pbl;
Signature · 1 line
2.54.0.563.g4f69b47b94-goog
maintainer Applied
#1 lore raw
Quoted 18 lines · Jacob Moroni

The irdma_copy_user_pgaddrs function loops through all of the umem DMA blocks to populate the PBLEs and will stop when either the last DMA block is reached or palloc->total_cnt is reached. The issue is that the logic for checking palloc->total_cnt would only work for non-zero values.

When irdma_setup_pbles is called with lvl==0, it
calls irdma_copy_user_pgaddrs with palloc->total_cnt==0, which means the only way to break out of the loop is to reach the last umem DMA block, which means it could end up going beyond the fixed size of 4 iwmr->pgaddrmem array that is used in the lvl==0 case.

In the case of QP/CQ/SRQ rings, the value of lvl is determined by a separate input (for example, req.cq_pages in the case of a CQ). So, we must perform explicit checking to ensure we don't overflow the pgaddrmem array if the user provides a umem that consists of more blocks than their provided req.cq_pages.

Quoted 5 lines
Fixes b48c24c2d710 ("RDMA/irdma: Implement device supported verb APIs")
---
 drivers/infiniband/hw/irdma/verbs.c | 14 ++++++++++----
 1 file changed, 10 insertions(+), 4 deletions(-)

Applied to for-next

Can you also address the unrelated bug sashiko found:

https://sashiko.dev/#/patchset/20260512183852.614045-1-jmoroni%40google.com

Thanks,
Jason

#2 lore raw

Thanks!

Quoted 1 line

Can you also address the unrelated bug sashiko found:

Sure, I will take a look.

- Jake