Skip to contentSkip to conversation

RDMA/ucma: Allow path records to exactly fit the output buffer

PATCH Applied Serhat Kumral 2 participants 2 messages Aug 6 – Aug 13 InfiniBand

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 RDMA/ucma: Allow path records to exactly fit the output buffer

ucma_query_path() emits a path record only when the remaining output buffer is strictly larger than struct ib_path_rec_data. A buffer sized exactly for the response header and N complete records therefore gets only N - 1 records, while resp->num_paths still advertises N. A caller sizing its buffer for a single record gets a header claiming one path and no path data at all.

ucma_query_ib_service() in the same file computes the record count with a plain division and so accepts an exact fit; make ucma_query_path() behave the same way.

Current librdmacm is unaffected because it always sizes the response for six records while the kernel currently reports at most two paths. Other users of the UAPI that provide an exactly sized buffer can observe the truncated response.

Fixes ac53b264b2f3 ("RDMA/ucma: Support querying when IB paths are not reversible")

---
Reproduced with soft-RoCE (rxe) under qemu, on two 7.2.0-rc3 kernels that differ only in this patch; the kernel config, the test program and the VM were identical across both runs. The test program drives /dev/infiniband/rdma_cm directly so that hdr.out can be set to exactly

	sizeof(struct rdma_ucm_query_path_resp) +
	sizeof(struct ib_path_rec_data)		= 8 + 72 = 80

and poisons the response buffer first, so a record slot the kernel never wrote stays recognisable afterwards.

Before:

requesting room for 1 record(s): hdr.out = 80 resp->num_paths (advertised by kernel) = 1
record slot 0: UNTOUCHED (still poison)
records that fit in the buffer and should have been copied: 1
records actually copied: 0

After:

requesting room for 1 record(s): hdr.out = 80 resp->num_paths (advertised by kernel) = 1
record slot 0: written by kernel flags=0x0000002b
records that fit in the buffer and should have been copied: 1
records actually copied: 1

flags 0x2b is IB_PATH_GMP | IB_PATH_PRIMARY | IB_PATH_BIDIRECTIONAL, i.e. exactly what ucma_query_path() writes into the record.

 drivers/infiniband/core/ucma.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/infiniband/core/ucma.c +1 -1
index 878561fa1cb5..cfd202325458 100644
@@ -951,7 +951,7 @@ static ssize_t ucma_query_path(struct ucma_context *ctx,
 
 	resp->num_paths = ctx->cm_id->route.num_pri_alt_paths;
 	for (i = 0, out_len -= sizeof(*resp);
-	     i < resp->num_paths && out_len > sizeof(struct ib_path_rec_data);
+	     i < resp->num_paths && out_len >= sizeof(struct ib_path_rec_data);
 	     i++, out_len -= sizeof(struct ib_path_rec_data)) {
 		struct sa_path_rec *rec = &ctx->cm_id->route.path_rec[i];
 
Signature · 1 line
2.55.0
#1 lore raw
Quoted 8 lines · Serhat Kumral

ucma_query_path() emits a path record only when the remaining output buffer is strictly larger than struct ib_path_rec_data. A buffer sized exactly for the response header and N complete records therefore gets only N - 1 records, while resp->num_paths still advertises N. A caller sizing its buffer for a single record gets a header claiming one path and no path data at all.

[...]

Applied, thanks!

[1/1] RDMA/ucma: Allow path records to exactly fit the output buffer

      https://git.kernel.org/rdma/rdma/c/8827f0c17f50f1

Best regards,

Signature · 1 line
Leon Romanovsky <[email protected]>