Linux RDMA and InfiniBand development
 help / color / mirror / Atom feed
* [PATCH rdma-rc] RDMA/siw: Honor the sg_offset in siw_map_mr_sg()
@ 2026-11-22 19:11 Andrew Stellman
  2026-11-22 19:16 ` sashiko-bot
  0 siblings, 1 reply; 2+ messages in thread
From: Andrew Stellman @ 2026-11-22 19:11 UTC (permalink / raw)
  To: Bernard Metzler
  Cc: Jason Gunthorpe, Leon Romanovsky, linux-rdma, linux-kernel

siw_map_mr_sg() builds the first PBL entry from sg_dma_address() and
sg_dma_len() of the first scatterlist entry and ignores *sg_off.
ib_sg_to_pages() includes *sg_off in mr->iova, which siw stores as
mem->va, so siw_pbl_get_buffer() resolves an address in the MR to a
location *sg_off bytes before it. rdma_rw_ctx_init() passes its
sg_offset to ib_map_mr_sg(), and isert can pass a nonzero offset.

A test module registered a one page buffer on siw with ib_map_mr_sg()
and sg_offset 0x200, which gave mr->iova at page offset 0x200 and
mr->length 3584. A user space client then wrote 16 bytes with an RDMA
WRITE at mr->iova. The bytes landed at page offset 0x000 of the buffer
instead of 0x200. With sg_offset 0 they landed at page offset 0x000,
as expected.

Start the first PBL entry *sg_off bytes into the first scatterlist
entry, and reject an offset past the end of that entry, as
ib_sg_to_pages() does. With this change the bytes land at page offset
0x200.

The issue was found by Claude Opus 5.5 running Quality Playbook, an
LLM-driven code review tool:
https://github.com/andrewstellman/quality-playbook
The fix and this changelog were drafted with Claude Opus 5.5 and
reviewed by me.

Fixes: 303ae1cdfdf7 ("rdma/siw: application interface")
Signed-off-by: Andrew Stellman <astellman@stellman-greene.com>
---
Tested in QEMU on v7.4-rc3 with KASAN, with an in-kernel test module
and a librdmacm client over siw. A reproducer is available on request.

 drivers/infiniband/sw/siw/siw_verbs.c | 11 +++++++++--
 1 file changed, 9 insertions(+), 2 deletions(-)

diff --git a/drivers/infiniband/sw/siw/siw_verbs.c b/drivers/infiniband/sw/siw/siw_verbs.c
index 80ef4cb95fc..c5f67c8e7ce 100644
--- a/drivers/infiniband/sw/siw/siw_verbs.c
+++ b/drivers/infiniband/sw/siw/siw_verbs.c
@@ -1505,8 +1505,15 @@ int siw_map_mr_sg(struct ib_mr *base_mr, struct scatterlist *sl, int num_sle,
 			return -EINVAL;
 		}
 		if (i == 0) {
-			pble->addr = sg_dma_address(slp);
-			pble->size = sg_dma_len(slp);
+			unsigned int off = sg_off ? *sg_off : 0;
+
+			/* The MR starts *sg_off bytes into the first entry */
+			if (off > sg_dma_len(slp)) {
+				siw_dbg_mem(mem, "sg_off %u beyond first SGE\n", off);
+				return -EINVAL;
+			}
+			pble->addr = sg_dma_address(slp) + off;
+			pble->size = sg_dma_len(slp) - off;
 			pble->pbl_off = 0;
 			pbl_size = pble->size;
 			pbl->num_buf = 1;
-- 
2.43.0


^ permalink raw reply related	[flat|nested] 2+ messages in thread

end of thread, other threads:[~2026-11-22 19:16 UTC | newest]

Thread overview: 2+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2026-11-22 19:11 [PATCH rdma-rc] RDMA/siw: Honor the sg_offset in siw_map_mr_sg() Andrew Stellman
2026-11-22 19:16 ` sashiko-bot

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox