On 2017/9/26 下午3:15, 박병철/선임연구원/SW Platform(연)AOT팀(byungchul.park@xxxxxxx) wrote: >> -----Original Message----- >> From: Coly Li [mailto:i@xxxxxxx] >> Sent: Tuesday, September 26, 2017 4:09 PM >> To: 박병철/선임연구원/SW Platform(연)AOT팀(byungchul.park@xxxxxxx); >> Michael Lyle; Coly Li >> Cc: linux-bcache@xxxxxxxxxxxxxxx; linux-block@xxxxxxxxxxxxxxx; >> axboe@xxxxxxxxx; Eric Wheeler; Kent Overstreet; kernel-team@xxxxxxx >> Subject: Re: [PATCH 04/12] bcache: Don't reinvent the wheel but use existing >> llist API >> >> On 2017/9/26 下午2:39, 박병철/선임연구원/SW >> Platform(연)AOT팀(byungchul.park@xxxxxxx) wrote: >>>> -----Original Message----- >>>> From: Michael Lyle [mailto:mlyle@xxxxxxxx] >>>> Sent: Tuesday, September 26, 2017 1:38 PM >>>> To: Coly Li >>>> Cc: linux-bcache@xxxxxxxxxxxxxxx; linux-block@xxxxxxxxxxxxxxx; >>>> axboe@xxxxxxxxx; Eric Wheeler; Byungchul Park; Kent Overstreet >>>> Subject: Re: [PATCH 04/12] bcache: Don't reinvent the wheel but use existing >>>> llist API >>>> >>>> I believe this introduces a critical bug. >>>> >>>> cl->list is used to link together the llists for both things waiting, >>>> and for things that are being woken. >>>> >>>> If a closure that is woken decides to wait again, it will corrupt the >>>> llist that __closure_wake_up is using. >>>> >>>> The previous iteration structure gets the next element of the list >>>> before waking and is therefore safe. >>> >>> Do you mean we have to use llist_for_each_entry_safe() instead of non-safe >> version? >>> Is it ok if we use it instead? >> >> Yes, we should use llist_for_each_entry_safe(), there is a quite >> implicit race here. > > Hi coly, > > As you know, my first patch used the safe version, but you suggested to replace > It with non-safe one. :( No doubt, it's my fault. I didn't find the implicit potential race. > I will change it so it does the same as the original patch did. :) Yes, please. Because the original patch is merged into Linux upstream, post a fix to replace llist_for_each_entry() by llist_for_each_entry_safe(). We still have chance to catch up 4.14. Thank you ! -- Coly Li
