[PATCH] io_uring/fdinfo: validate opcode before checking if it's an 128b one

[PATCH] io_uring/fdinfo: validate opcode before checking if it's an 128b one

[Jens Axboe]

The mixed SQE support assumes that userspace always passes valid data,
that is not the case. Validate the opcode properly before indexing
the io_issue_defs[] array, and pass it through the nospec indexing
as well as it's a user valid indexing a kernel array.

Fixes: 1cba30bf9fdd ("io_uring: add support for IORING_SETUP_SQE_MIXED")
Reported-by: syzbot+b883b008a0b1067d5833@syzkaller.appspotmail.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>

---

diff --git a/io_uring/fdinfo.c b/io_uring/fdinfo.c
index 248006424cab..ac6e7edc7027 100644
--- a/io_uring/fdinfo.c
+++ b/io_uring/fdinfo.c
@@ -5,6 +5,7 @@
 #include <linux/file.h>
 #include <linux/proc_fs.h>
 #include <linux/seq_file.h>
+#include <linux/nospec.h>
 #include <linux/io_uring.h>
 
 #include <uapi/linux/io_uring.h>
@@ -107,6 +108,9 @@ static void __io_uring_show_fdinfo(struct io_ring_ctx *ctx, struct seq_file *m)
 
 		sqe = &ctx->sq_sqes[sq_idx << sq_shift];
 		opcode = READ_ONCE(sqe->opcode);
+		if (opcode >= IORING_OP_LAST)
+			continue;
+		opcode = array_index_nospec(opcode, IORING_OP_LAST);
 		if (sq_shift) {
 			sqe128 = true;
 		} else if (io_issue_defs[opcode].is_128) {

-- 
Jens Axboe