> 2022年9月7日 19:29，mingzhe.zou@xxxxxxxxxxxx 写道：
>
> From: mingzhe <mingzhe.zou@xxxxxxxxxxxx>
>
> Currently, size is specified and not checked when creating a flash device.
> This will cause a problem, IO maybe hang when creating a flash device with
> the actual size of the device.
>
> ```
> if (attr == &sysfs_flash_vol_create) {
> int r;
> uint64_t v;
>
> strtoi_h_or_return(buf, v);
>
> r = bch_flash_dev_create(c, v);
> if (r)
> return r;
> }
> ```
>
> Because the flash device needs some space for superblock, journal and btree.
> If the size of data reaches the available size, the new IO cannot allocate
> space and will hang. At this time, the gc thread will be started frequently.
>
> Even more unreasonable, we can create flash devices larger than actual size.
>
> ```
> [root@zou ~]# echo 2G > /sys/block/vdb/bcache/set/flash_vol_create
> [root@zou ~]# lsblk /dev/vdb
> NAME MAJ:MIN RM SIZE RO TYPE MOUNTPOINT
> vdb 252:16 0 1G 0 disk
> └─bcache0 251:0 0 2G 0 disk
> ```
>
> This patch will limit the size of flash device, reserving at least 5% of
> available size for the btree.
>
> ```
> [root@zou ~]# echo 2G > /sys/block/vdb/bcache/set/flash_vol_create
> [root@zou ~]# lsblk /dev/vdb
> NAME MAJ:MIN RM SIZE RO TYPE MOUNTPOINT
> vdb 252:16 0 1G 0 disk
> └─bcache0 251:0 0 950M 0 disk
> ```
>
> Signed-off-by: mingzhe <mingzhe.zou@xxxxxxxxxxxx>
> ---
> drivers/md/bcache/super.c | 13 ++++++++++++-
> 1 file changed, 12 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/md/bcache/super.c b/drivers/md/bcache/super.c
> index ba3909bb6bea..f41e09e0e8ee 100644
> --- a/drivers/md/bcache/super.c
> +++ b/drivers/md/bcache/super.c
> @@ -1579,6 +1579,17 @@ static int flash_devs_run(struct cache_set *c)
> return ret;
> }
>
Hi Mingzhe,
> +static inline sector_t flash_dev_max_sectors(struct cache_set *c)
> +{
> + size_t avail_nbuckets;
> + struct cache *ca = c->cache;
> + size_t first_bucket = ca->sb.first_bucket;
> + size_t njournal_buckets = ca->sb.njournal_buckets;
> +
> + avail_nbuckets = c->nbuckets - first_bucket - njournal_buckets;
> + return bucket_to_sector(c, avail_nbuckets / 100 * 95);
> +}
Overall I like this idea. This is really something I didn’t realize to fix, nice catch!
BTW, I feel 95% is still quite high rate, how about using 90%? And you may define the rate as a macro in bcache.h.
Thanks for this patch.
Coly Li
> +
> int bch_flash_dev_create(struct cache_set *c, uint64_t size)
> {
> struct uuid_entry *u;
> @@ -1600,7 +1611,7 @@ int bch_flash_dev_create(struct cache_set *c, uint64_t size)
> u->first_reg = u->last_reg = cpu_to_le32((u32)ktime_get_real_seconds());
>
> SET_UUID_FLASH_ONLY(u, 1);
> - u->sectors = size >> 9;
> + u->sectors = min(flash_dev_max_sectors(c), size >> 9);
>
> bch_uuid_write(c);
>
> --
> 2.17.1
>
