On Fri, Jun 23, 2017 at 2:36 PM, Andreas Dilger <adilger@dilger.ca> wrote:
> On Jun 23, 2017, at 6:26 AM, Theodore Ts'o <tytso@mit.edu> wrote:
>>
>> The problem is that if we continue, successive reads may all take
>> seconds or minutes to fail, thus tieing up the process for a long
>> time.
>
> Sorry, I don't understand where the seconds or minutes of delay come from?
> Is that because of long SCSI retries in the block layer, or in the disk
> itself, or something caused specifically because of this code?

For a networked block device we may be retrying for a while before
giving up, although this also applies to the initial failed read.

>
>> By returning EIO right away, we can "fast fail".
>
> But it seems like you don't necessarily need to fail at all?  Something like the
> following would return an error if the entry is not found, but still search the
> rest of the leaf blocks (if any) before giving up:
>

Giving up early or checking future blocks both work, critical thing
here is not returning NULL after seeing a read error.
Previously to this the behavior was to continue to check future blocks
after a read error, and it seemed OK.

Khazhy