From: Carlos Bilbao <carlos.bilbao@kernel.org>

If ecdb_ahdr->ahslength is zero, two bugs follow:

  kmalloc(be16_to_cpu(ecdb_ahdr->ahslength) + 15, ...)

allocates 15 bytes, but the immediately following memcpy writes
ISCSI_CDB_SIZE (16) bytes into it, a one-byte heap overflow. Also:

  memcpy(cdb + ISCSI_CDB_SIZE, ecdb_ahdr->ecdb,
           be16_to_cpu(ecdb_ahdr->ahslength) - 1);

(u16)0 - 1 promotes to (int)-1 which converts to SIZE_MAX as size_t,
causing a massive out-of-bounds write.

Reject ahslength == 0 with ISCSI_REASON_PROTOCOL_ERROR before the kmalloc.
Also reject ahslength values that exceed the actual AHS buffer advertised.

Fixes: 8f1f7d297bce ("scsi: target: iscsi: Add support for extended CDB AHS")
Signed-off-by: Carlos Bilbao (Lambda) <carlos.bilbao@kernel.org>
Reviewed-by: Dmitry Bogdanov <d.bogdanov@yadro.com>

Changes in v3:
- Move ahslength and cdb_length declarations to function scope following
  the reverse christmas tree ordering of existing locals.

Changes in v2:
- Add bounds check: ahslength must not exceed (hdr->hlength * 4) - 3.
- Replace opaque ahslength + 15 with explicit cdb_length variable.
---
 drivers/target/iscsi/iscsi_target.c | 22 ++++++++++++++++++----
 1 file changed, 18 insertions(+), 4 deletions(-)

diff --git a/drivers/target/iscsi/iscsi_target.c b/drivers/target/iscsi/iscsi_target.c
index e80449f6ce15..cb832fd523af 100644
--- a/drivers/target/iscsi/iscsi_target.c
+++ b/drivers/target/iscsi/iscsi_target.c
@@ -995,6 +995,7 @@ int iscsit_setup_scsi_cmd(struct iscsit_conn *conn, struct iscsit_cmd *cmd,
 	int data_direction, payload_length;
 	struct iscsi_ecdb_ahdr *ecdb_ahdr;
 	struct iscsi_scsi_req *hdr;
+	u16 ahslength, cdb_length;
 	int iscsi_task_attr;
 	unsigned char *cdb;
 	int sam_task_attr;
@@ -1108,14 +1109,27 @@ int iscsit_setup_scsi_cmd(struct iscsit_conn *conn, struct iscsit_cmd *cmd,
 				ISCSI_REASON_CMD_NOT_SUPPORTED, buf);
 		}
 
-		cdb = kmalloc(be16_to_cpu(ecdb_ahdr->ahslength) + 15,
-			      GFP_KERNEL);
+		ahslength = be16_to_cpu(ecdb_ahdr->ahslength);
+		if (!ahslength) {
+			pr_err("Extended CDB AHS with zero length, protocol error.\n");
+			return iscsit_add_reject_cmd(cmd,
+				ISCSI_REASON_PROTOCOL_ERROR, buf);
+		}
+		if (ahslength > (hdr->hlength * 4) - 3) {
+			pr_err("Extended CDB AHS length %u exceeds available PDU buffer.\n",
+			       ahslength);
+			return iscsit_add_reject_cmd(cmd,
+				ISCSI_REASON_PROTOCOL_ERROR, buf);
+		}
+
+		cdb_length = ahslength - 1 + ISCSI_CDB_SIZE;
+
+		cdb = kmalloc(cdb_length, GFP_KERNEL);
 		if (cdb == NULL)
 			return iscsit_add_reject_cmd(cmd,
 				ISCSI_REASON_BOOKMARK_NO_RESOURCES, buf);
 		memcpy(cdb, hdr->cdb, ISCSI_CDB_SIZE);
-		memcpy(cdb + ISCSI_CDB_SIZE, ecdb_ahdr->ecdb,
-		       be16_to_cpu(ecdb_ahdr->ahslength) - 1);
+		memcpy(cdb + ISCSI_CDB_SIZE, ecdb_ahdr->ecdb, cdb_length - ISCSI_CDB_SIZE);
 	}
 
 	data_direction = (hdr->flags & ISCSI_FLAG_CMD_WRITE) ? DMA_TO_DEVICE :
-- 
2.43.0

Carlos,

> If ecdb_ahdr->ahslength is zero, two bugs follow:
>
>   kmalloc(be16_to_cpu(ecdb_ahdr->ahslength) + 15, ...)
>
> allocates 15 bytes, but the immediately following memcpy writes
> ISCSI_CDB_SIZE (16) bytes into it, a one-byte heap overflow. Also:

Applied to 7.1/scsi-staging, thanks!

-- 
Martin K. Petersen