io_uring: fix races with file table unregister
authorPavel Begunkov <asml.silence@gmail.com>
Mon, 13 Jun 2022 05:32:44 +0000 (06:32 +0100)
committerGreg Kroah-Hartman <gregkh@linuxfoundation.org>
Wed, 22 Jun 2022 12:22:00 +0000 (14:22 +0200)
[ Upstream commit b0380bf6dad4601d92025841e2b7a135d566c6e3 ]

Fixed file table quiesce might unlock ->uring_lock, potentially letting
new requests to be submitted, don't allow those requests to use the
table as they will race with unregistration.

Reported-and-tested-by: van fantasy <g1042620637@gmail.com>
Fixes: 05f3fb3c53975 ("io_uring: avoid ring quiesce for fixed file set unregister and update")
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: Sasha Levin <sashal@kernel.org>
fs/io_uring.c

index b8e6398..5f111a6 100644 (file)
@@ -7933,11 +7933,19 @@ static void __io_sqe_files_unregister(struct io_ring_ctx *ctx)
 
 static int io_sqe_files_unregister(struct io_ring_ctx *ctx)
 {
+       unsigned nr = ctx->nr_user_files;
        int ret;
 
        if (!ctx->file_data)
                return -ENXIO;
+
+       /*
+        * Quiesce may unlock ->uring_lock, and while it's not held
+        * prevent new requests using the table.
+        */
+       ctx->nr_user_files = 0;
        ret = io_rsrc_ref_quiesce(ctx->file_data, ctx);
+       ctx->nr_user_files = nr;
        if (!ret)
                __io_sqe_files_unregister(ctx);
        return ret;