Re: scsi: scsi_debug: make timeout faults by set delay to maximum value

[John Garry <john.g.garry@oracle.com>]

On 06/11/2025 12:03, JiangJianJun wrote:
>> I got this patch 3x times...
> Sorry, command has an error, i thought it had failed, so I sent it again.
> 
>> What do you mean by "checked during cancellation"?
> &
>> I don't know what is meant by "cancel the command properly".
> Maybe I shouldn't use "cancel" to describe it; in the code, it's called "abort".
> The function `sdebug_timeout_cmd` is designed to simulate a command timeout, by
> discarding it. I noticed that you added a check to see if the command was
> executed before "abort"; otherwise, it returns a failure.

When we get discard the command, we get a timeout, and the scsi error 
handling kicks in eventually. The first thing that the scsi error 
handler tries to do is abort the command. All that the scsi_debug abort 
handler can do is ensure that we no longer have a reference to the 
scsi_command, which may mean cancelling any pending completion (if 
possible). Is your problem that sometimes the abort handler may fail, 
and we have to escalate?

> Therefore, I change discarding to long-long-delay, so that "abort" will succeed.
> If we needed abort-failure, we can inject ERR_ABORT_CMD_FAILED.
> 
> The fault-injection can be seen this link:
> https://urldefense.com/v3/__https://lore.kernel.org/r/20231010092051.608007-5-haowenchao2@huawei.com__;!!ACWV5N9M2RV99hQ!L_o5Xt_9lhXLe0R8AWsJWnt33qXlNijbVEWyiq9kjgA2lBsuV7E2s7ficvGM37RlSxxvXVZ1YEkAQxZ-WkFvKGTURrU$
> 
>