On Mon, Dec 12, 2022 at 11:20:48AM +0800, mingzhe.zou@xxxxxxxxxxxx wrote:
> From: mingzhe <mingzhe.zou@xxxxxxxxxxxx>
>
> Currently, the dirty_data of cached_dev and flash_dev depend on the stripe.
>
> Since the flash device supports resize, it may cause a bug (resize the flash
> from 1T to 2T, and nr_stripes from 1 to 2).
>
> The patch add dirty_data in struct bcache_device, we can get the value of
> dirty_data quickly and fixes the bug of resize flash device.
>
> Signed-off-by: mingzhe <mingzhe.zou@xxxxxxxxxxxx>
> ---
> drivers/md/bcache/bcache.h | 1 +
> drivers/md/bcache/writeback.c | 2 ++
> drivers/md/bcache/writeback.h | 7 +------
> 3 files changed, 4 insertions(+), 6 deletions(-)
>
> diff --git a/drivers/md/bcache/bcache.h b/drivers/md/bcache/bcache.h
> index 621a2ae1767b..5da991505b45 100644
> --- a/drivers/md/bcache/bcache.h
> +++ b/drivers/md/bcache/bcache.h
> @@ -268,6 +268,7 @@ struct bcache_device {
> unsigned int stripe_size;
> atomic_t *stripe_sectors_dirty;
> unsigned long *full_dirty_stripes;
> + atomic_long_t dirty_sectors;
>
> struct bio_set bio_split;
>
> diff --git a/drivers/md/bcache/writeback.c b/drivers/md/bcache/writeback.c
> index f21295dea71b..7b5009e8b4ff 100644
> --- a/drivers/md/bcache/writeback.c
> +++ b/drivers/md/bcache/writeback.c
> @@ -769,6 +769,8 @@ void bcache_dev_sectors_dirty_add(struct cache_set *c, unsigned int inode,
> if (stripe < 0)
> return;
>
> + atomic_long_add(nr_sectors, &d->dirty_sectors);
> +
> if (UUID_FLASH_ONLY(&c->uuids[inode]))
> atomic_long_add(nr_sectors, &c->flash_dev_dirty_sectors);
>
Indeed, if the bcache device is a flash only volume, all the rested code since
here to the end of bcache_dev_sectors_dirty_add() can be ignored. Setting
stripe_sectors_dirty and full_dirty_stripes is for writeback related stuffs,
for flsh only device there is no writeback.
And then you may add atomic_long_add(nr_sectors, &d->dirty_sectors) after
atomic_long_add(nr_sectors, &c->flash_dev_dirty_sectors). After this, for
flash only device, its stripe_sectors_dirty and full_dirty_stripes can be
always 0, and its dirty sectors are counted in flash_dev_dirty_sectors.
Yes current code still allocate memory for stripes counter when it is flash
only device. I will fix this firstly.
> diff --git a/drivers/md/bcache/writeback.h b/drivers/md/bcache/writeback.h
> index 7e5a2fe03429..12765c0dfd5c 100644
> --- a/drivers/md/bcache/writeback.h
> +++ b/drivers/md/bcache/writeback.h
> @@ -56,12 +56,7 @@ struct bch_dirty_init_state {
>
> static inline uint64_t bcache_dev_sectors_dirty(struct bcache_device *d)
> {
> - uint64_t i, ret = 0;
> -
> - for (i = 0; i < d->nr_stripes; i++)
> - ret += atomic_read(d->stripe_sectors_dirty + i);
> -
> - return ret;
> + return atomic_long_read(&d->dirty_sectors);
> }
>
This looks really nice.
> static inline int offset_to_stripe(struct bcache_device *d,
> --
> 2.17.1
>
--
Coly Li
