On 12/21/20 10:48 PM, Naohiro Aota wrote:
We cannot use log-structured superblock writing in conventional zones since
there is no write pointer to determine the last written superblock
position. So, we write a superblock at a static location in a conventional
zone.
The written position is at the beginning of a zone, which is different from
an SB position of regular btrfs. This difference causes a "chicken-and-egg
problem" when supporting zoned emulation on a regular device. To know if
btrfs is (emulated) zoned btrfs, we need to load an SB and check the
feature flag. However, to load an SB, we need to know that it is zoned
btrfs to load it from a different position.
This patch moves the SB location on conventional zones so that the first SB
location will be the same as regular btrfs.
Signed-off-by: Naohiro Aota <naohiro.aota@xxxxxxx>
---
fs/btrfs/zoned.c | 3 ++-
1 file changed, 2 insertions(+), 1 deletion(-)
diff --git a/fs/btrfs/zoned.c b/fs/btrfs/zoned.c
index 90b8d1d5369f..e5619c8bcebb 100644
--- a/fs/btrfs/zoned.c
+++ b/fs/btrfs/zoned.c
@@ -465,7 +465,8 @@ static int sb_log_location(struct block_device *bdev, struct blk_zone *zones,
int ret;
if (zones[0].type == BLK_ZONE_TYPE_CONVENTIONAL) {
- *bytenr_ret = zones[0].start << SECTOR_SHIFT;
+ *bytenr_ret = (zones[0].start << SECTOR_SHIFT) +
+ btrfs_sb_offset(0);
return 0;
}
I'm confused, we call btrfs_sb_log_location_bdev(), which does
if (!bdev_is_zoned(bdev)) {
*bytenr_ret = btrfs_sb_offset(mirror);
return 0;
}
so how does the emulation work, if we short circuit this if the block device
isn't zoned? And then why does it matter where in the conventional zone that we
put the super block? Can't we just emulate a conventional zone that starts at
offset 0, and then the btrfs_sb_offset() will be the same as zones[0].start +
btrfs_sb_offset(0)? Thanks,
Josef
