io_uring: check file_slot early when accept use fix_file mode
PATCH 1/1 io_uring: check file_slot early when accept use fix_file mode
check file_slot early in io_accept_prep() to avoid wasted effort in failure cases.
--- fs/io_uring.c | 2 ++ 1 file changed, 2 insertions(+)
fs/io_uring.c +2 -0
@@ -4868,6 +4868,8 @@ static int io_accept_prep(struct io_kiocb *req, const struct io_uring_sqe *sqe) accept->nofile =rlimit (RLIMIT_NOFILE ); accept->file_slot =READ_ONCE (sqe->file_index); +if (accept->file_slot -1 >= req->ctx->nr_user_files) +return -EINVAL ;if (accept->file_slot && ((req->open.how.flags &O_CLOEXEC ) || (accept->flags &SOCK_CLOEXEC )))return -EINVAL ;
Signature · 1 line
2.24.4
Jens Axboe
maintainerQuoted 2 lines · Hao Xu
check
file_slotearly inio_accept_prep()to avoid wasted effort in failure cases.
It's generally better to just let the failure cases deal with it instead of having checks in multiple places. This is a failure path, so we don't care about making it fail early. Optimizations should be for the hot path, which is not a malformed sqe.
Signature · 1 line
Jens Axboe
Quoted 8 lines · Jens Axboe
Quoted 2 lines · Hao Xu
check
file_slotearly inio_accept_prep()to avoid wasted effort in failure cases.It's generally better to just let the failure cases deal with it instead of having checks in multiple places. This is a failure path, so we don't care about making it fail early. Optimizations should be for the hot path, which is not a malformed sqe.
I have a question here: if we do do_accept() and but fail in io_install_fixed_file(), do we lose the conn_fd return by do_accept() forever?
Quoted 1 line
Jens Axboe
maintainerQuoted 5 lines · Hao Xu
Quoted 4 lines · Jens Axboe
Quoted 2 lines · Hao Xu
check
file_slotearly inio_accept_prep()to avoid wasted effort in failure cases.
Quoted 7 lines
Quoted 4 lines
It's generally better to just let the failure cases deal with it instead of having checks in multiple places. This is a failure path, so we don't care about making it fail early. Optimizations should be for the hot path, which is not a malformed sqe.
I have a question here: if we do
do_accept()and but fail inio_install_fixed_file(), do we lose theconn_fdreturn bydo_accept()forever?
We do. The file is put and everything, so we're not leaking anything. But the actual connection is lost as the accept request failed.
Signature · 1 line
Jens Axboe
Quoted 10 lines · Jens Axboe
Quoted 9 lines · Hao Xu
Quoted 8 lines · Jens Axboe
Quoted 2 lines · Hao Xu
check
file_slotearly inio_accept_prep()to avoid wasted effort in failure cases.It's generally better to just let the failure cases deal with it instead of having checks in multiple places. This is a failure path, so we don't care about making it fail early. Optimizations should be for the hot path, which is not a malformed sqe.
Quoted 6 lines
Quoted 3 lines
I have a question here: if we do
do_accept()and but fail inio_install_fixed_file(), do we lose theconn_fdreturn bydo_accept()forever?We do. The file is put and everything, so we're not leaking anything. But the actual connection is lost as the accept request failed.
Does that cause any problem, since from client's perspective, connection is builded, and there is no way for the server to close it.
Quoted 1 line
Jens Axboe
maintainerQuoted 11 lines · Hao Xu
Quoted 10 lines · Jens Axboe
Quoted 9 lines · Hao Xu
Quoted 8 lines · Jens Axboe
Quoted 2 lines · Hao Xu
check
file_slotearly inio_accept_prep()to avoid wasted effort in failure cases.It's generally better to just let the failure cases deal with it instead of having checks in multiple places. This is a failure path, so we don't care about making it fail early. Optimizations should be for the hot path, which is not a malformed sqe.
Quoted 8 lines
Quoted 6 lines
Quoted 3 lines
I have a question here: if we do
do_accept()and but fail inio_install_fixed_file(), do we lose theconn_fdreturn bydo_accept()forever?We do. The file is put and everything, so we're not leaking anything. But the actual connection is lost as the accept request failed.
Does that cause any problem, since from client's perspective, connection is builded, and there is no way for the server to close it.
Maybe? But it's a program error, it's asking to accept a connection at a slot which is invalid. It kind of gets to keep the pieces in that particular case.
Signature · 1 line
Jens Axboe