[PATCH] conhost: Fix status for READ_CONSOLE ioctl

[Keno Fischer]

Without this patch conhost would attempt to process non-existent input,
spewing stdout with various cursor moves and incorrectly returning
the ReadConsole call early. Fix that by only attempting input processing
if there are actually input records to be processed and properly
returning STATUS_PENDING if the input processing is not yet complete.

Signed-off-by: Keno Fischer 
---
 programs/conhost/conhost.c | 9 ++++++++-
 1 file changed, 8 insertions(+), 1 deletion(-)

diff --git a/programs/conhost/conhost.c b/programs/conhost/conhost.c
index 78f6e345170..64a77e52b43 100644
--- a/programs/conhost/conhost.c
+++ b/programs/conhost/conhost.c
@@ -1385,7 +1385,14 @@ static NTSTATUS read_console( struct console *console, unsigned int ioctl, size_
     if (edit_line_grow( console, 1 )) console->edit_line.buf[0] = 0;

console->pending_read = out_size;
-    return process_console_input( console );
+
+    /* If there are any pending input records, cook them now. */
+    if (console->record_count)
+    {
+        process_console_input( console );
+    }
+
+    return console->edit_line.status;
 }

/* add input events to a console input queue */

[Jacek Caban]

Hi Keno,

On 12/29/21 16:10, Keno Fischer wrote:
> diff --git a/programs/conhost/conhost.c b/programs/conhost/conhost.c
> index 78f6e345170..64a77e52b43 100644
> --- a/programs/conhost/conhost.c
> +++ b/programs/conhost/conhost.c
> @@ -1385,7 +1385,14 @@ static NTSTATUS read_console( struct console *console, unsigned int ioctl, size_
>       if (edit_line_grow( console, 1 )) console->edit_line.buf[0] = 0;
>   console->pending_read = out_size;
> - return process_console_input( console );
> -
> - /* If there are any pending input records, cook them now. */
> - if (console->record_count)
> - {
> -
>    process_console_input( console );
> - }
> -
> - return console->edit_line.status;

process_console_input() generally handles 0 record_count, what's exactly 
wrong there? Does the attached patch help?

Thanks,

Jacek

[Keno Fischer]

Hi Jacek,

On Wed, Dec 29, 2021 at 11:13 AM Jacek Caban  wrote:
> process_console_input() generally handles 0 record_count, what's exactly
> wrong there? Does the attached patch help?

Yes, that patch would have a similar effect, except that I believe it
would still return STATUS_SUCCESS rather than STATUS_PENDING.
Unfortunately, I can't reproduce the exact issue I saw that caused
this right now (I upgraded my local tree from rc1 to master yesterday,
so that's likely the difference), but basically what I was seeing is that
conhost spammed about 100 or so unnecessary cursor
moves from inside `update_read_output`. That caused lots of blinking
in the terminal. This patch fixed that for me.

Keno