Re: [PATCH] bcache: set io_disable to true when stop bcache device

[Coly Li <colyli@suse.de>]

> 2023年3月8日 17:20，mingzhe <mingzhe.zou@easystack.cn> 写道：
> 
> Stop is an operation that cannot be aborted. If there are still
> IO requests being processed, we can never stop the device.
> So, all new IO requests should fail when we set io_disable to true.
> However, sysfs has been unlinked at this time, user cannot modify
> io_disable via sysfs.
> 
> Signed-off-by: mingzhe <mingzhe.zou@easystack.cn>

NACK. I tried this years ago, it might introduce potential risk for meta data inconsistency.


Coly Li


> ---
> drivers/md/bcache/request.c | 16 ++++++++++++++++
> drivers/md/bcache/super.c   |  9 +++++++++
> 2 files changed, 25 insertions(+)
> 
> diff --git a/drivers/md/bcache/request.c b/drivers/md/bcache/request.c
> index 67a2e29e0b40..9b85aad20022 100644
> --- a/drivers/md/bcache/request.c
> +++ b/drivers/md/bcache/request.c
> @@ -758,6 +758,15 @@ static void cached_dev_bio_complete(struct closure *cl)
> search_free(cl);
> }
> 
> +static void cached_dev_bio_fail(struct closure *cl)
> +{
> + struct search *s = container_of(cl, struct search, cl);
> + struct cached_dev *dc = container_of(s->d, struct cached_dev, disk);
> +
> + s->iop.status = BLK_STS_IOERR;
> + cached_dev_bio_complete(cl);
> +}
> +
> /* Process reads */
> 
> static void cached_dev_read_error_done(struct closure *cl)
> @@ -971,6 +980,9 @@ static void cached_dev_write(struct cached_dev *dc, struct search *s)
> struct bkey start = KEY(dc->disk.id, bio->bi_iter.bi_sector, 0);
> struct bkey end = KEY(dc->disk.id, bio_end_sector(bio), 0);
> 
> + if (unlikely((dc->io_disable)))
> + goto fail_bio;
> +
> bch_keybuf_check_overlapping(&s->iop.c->moving_gc_keys, &start, &end);
> 
> down_read_non_owner(&dc->writeback_lock);
> @@ -1046,6 +1058,10 @@ static void cached_dev_write(struct cached_dev *dc, struct search *s)
> insert_data:
> closure_call(&s->iop.cl, bch_data_insert, NULL, cl);
> continue_at(cl, cached_dev_write_complete, NULL);
> + return;
> +
> +fail_bio:
> + continue_at(cl, cached_dev_bio_fail, NULL);
> }
> 
> static void cached_dev_nodata(struct closure *cl)
> diff --git a/drivers/md/bcache/super.c b/drivers/md/bcache/super.c
> index ba3909bb6bea..a2a82942f85b 100644
> --- a/drivers/md/bcache/super.c
> +++ b/drivers/md/bcache/super.c
> @@ -1389,6 +1389,15 @@ static void cached_dev_flush(struct closure *cl)
> bch_cache_accounting_destroy(&dc->accounting);
> kobject_del(&d->kobj);
> 
> + /*
> + * Stop is an operation that cannot be aborted. If there are still
> + * IO requests being processed, we can never stop the device.
> + * So, all new IO requests should fail when we set io_disable to true.
> + * However, sysfs has been unlinked at this time, user cannot modify
> + * io_disable via sysfs.
> + */
> + dc->io_disable = true;
> +
> continue_at(cl, cached_dev_free, system_wq);
> }
> 
> -- 
> 2.17.1.windows.2
>