On Mon Nov 20 13:46:57 2023 +0000, Jinoh Kang wrote:
Are there enough differences between \Device\NamedPipe and \Device\NamedPipe\
Not sure about that. I don't see enough tests to judge. I could be wrong, though.
to warrant the separate fd_ops?
They can be merged. The only think stopping the merge is the fact that `get_fd_user(fd)` points to the device *itself*, not the file object that contains the `is_rootdir` flag.
Does it actually make sense to use FD_TYPE_DIR here?
Putting anything here doesn't make a difference (hence TODO). In fact, removing `get_fd_type` entirely makes sense, too. Both `ntdll:om` and `ntdll:pipe` work just fine with this patch:
diff --git a/server/named_pipe.c b/server/named_pipe.c index 3ef87c20145..3628c1d9200 100644 --- a/server/named_pipe.c +++ b/server/named_pipe.c @@ -316,7 +316,7 @@ static const struct fd_ops named_pipe_device_fd_ops = { default_fd_get_poll_events, /* get_poll_events */ default_poll_event, /* poll_event */ - named_pipe_device_file_get_fd_type, /* get_fd_type */ + NULL, /* get_fd_type */ no_fd_read, /* read */ no_fd_write, /* write */ no_fd_flush, /* flush */ @@ -332,7 +332,7 @@ static const struct fd_ops named_pipe_rootdir_fd_ops = { default_fd_get_poll_events, /* get_poll_events */ default_poll_event, /* poll_event */ - named_pipe_rootdir_get_fd_type, /* get_fd_type */ + NULL, /* get_fd_type */ no_fd_read, /* read */ no_fd_write, /* write */ no_fd_flush, /* flush */In fact, there is a precedence: https://gitlab.winehq.org/wine/wine/-/blob/7c45c7c5ebb59237d568a4e5b38626422... Shall I go ahead and set both of them to NULL? `\Device\NamedPipe` doesn't look like a "device" to me anyway.
Removed special logic for `is_rootdir = 1` case entirely. The `FSCTL_PIPE_WAIT` test is now left todo.