On Tue, Jun 26, 2018 at 11:41:02PM +0800, zhiyuanme@xxxxxxxx wrote:
>
> Hi,
> There is a doubt about the use of bcache recently.
> In this scenario,current bio is going to bypass,but overlaped with
> writeback_keys.then bch_keybuf_check_overlapping will delete the
> overlaping bkey from writeback_keys rbtree when private (the member of
> rbnode) is NULL.
>
> Next,s->iop.bypass will be true,current bio is written directly to the
> HDD.Just at this time,the system is down for some reason,however the dirty
> data in the cache(SSD) haved falled to HDD.When the system is up
> again,journal will replay.the old dirty will be writeback resulting in
> overlapping the last data.
>
> Is there consistency in data in this case? Or any other way for bcache to
> deal with the problem?
We don't complete the write until we also delete the dirty data from the cache.
Notice that cached_dev_write _always_ calls bch_data_insert - if bypass is true,
bch_data_insert discards data from the cache instead of doing a write.
> Looking forward to your reply. Thank you.
>
> bcache version:linux-4.16.rc7
> code as follows:
> static void cached_dev_write(struct cached_dev *dc, struct search *s)
> { //if s->iop.bypass is true and overlapping with writeback_keys. ...
> if (bch_keybuf_check_overlapping(&dc->writeback_keys, &start, &end)) //if return false {
> s->iop.bypass = false;
> s->iop.writeback = true;
> } ... if (s->iop.bypass) { s->iop.bio = s->orig_bio; bio_get(s->iop.bio); if ((bio_op(bio) != REQ_OP_DISCARD) ||blk_queue_discard(bdev_get_queue(dc->bdev))) closure_bio_submit(bio, cl); } //system down ... closure_call(&s->iop.cl, bch_data_insert, NULL, cl); ...}
>
>
> zhiyuanme@xxxxxxxx
--
To unsubscribe from this list: send the line "unsubscribe linux-bcache" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
