RE: [patch] direct-io: bug fix in dio handling write error

[Chen, Kenneth W]

Badari Pulavarty wrote on Tuesday, March 21, 2006 8:57 AM
> I hate to do this you - but your patch breaks error handling on
> synchronous DIO requests.
> 
> Since you are using "dio->io_error" instead of "dio->result" to
> represent an error - you need to make sure to check that (also ?)
> instead of dio->result in direct_io_worker() before calling 
> dio_complete().
> 
> Isn't it ? Am I missing something ?


That's the other part of the maze.  AFAICS, in the synchronous path,
dio_bio_complete already implicitly checks -EIO error:

static int dio_bio_complete(struct dio *dio, struct bio *bio)
{ ...
    return uptodate ? 0 : -EIO;
}

And such error code bubbles up to direct_io_worker's sync path:

direct_io_worker {
    ...
    if (dio->is_async) {
    ...
    } else {
        ret2 = dio_await_completion(dio);
        if (ret == 0)
            ret = ret2;

I've also explicitly ran test case for synchronous write and found no
regression there.  I admit my test coverage may not be very comprehensive.
But I've done the best I can.

It's entirely possible there are more corner cases.  But let's get some
coverage here with -mm and then add fixes as we go.

- Ken

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/