linux-kernel - Re: UFS writing request failure handling

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [day] [month] [year] [list]

Message-ID: <6adc73db15b84425805a6adc09c2bfcf@SIWEX5A.sing.micron.com>
Date:   Tue, 13 Feb 2018 15:18:32 +0000
From:   "Bean Huo (beanhuo)" <beanhuo@...ron.com>
To:     Bart Van Assche <bart.vanassche@....com>
CC:     "linux-kernel@...r.kernel.org" <linux-kernel@...r.kernel.org>,
        "linux-scsi@...r.kernel.org" <linux-scsi@...r.kernel.org>
Subject: Re: UFS writing request failure handling

Hi Bart
Thanks for your answer.
I looked at SCSI core source codes these days, UFS also follows SCSI core error handling.
There is already re-issue behavior, and assign 5 retries for each UFS request if there is error.

>
>On 02/12/18 04:21, Bean Huo (beanhuo) wrote
>> I am looking at UFS error handling, but I didn't notice re-issues
>> requests with UTP error to the host controller. According UFS host
>> spec, "host software either completes the request that had the error
>> and requests still outstanding with error to higher level software, or
>> re-issues these requests to the host controller". For the latest Linux
>> UFS driver, it belongs to former one? If I am wrong, please correct
>> me.
>
>Hello Bean,
>
>I'm afraid that reissuing a failed write forever could cause an infinite loop that
>makes e.g. user space processes unkillable or that could cause an orderly
>system shutdown to hang. I'm not sure that we want such behavior.
>
>If you want to learn more about how the SCSI core handles failed writes,
>please have a look at scsi_decide_disposition(). A possible approach to
>modify the error handling behavior is to stack a dm driver on top of the SCSI
>core that implements the desired behavior.
>
>Bart.