Skip to contentSkip to conversation

btrfs: zoned: fix alloc_offset calculation for partly conventional block groups

PATCH In review Johannes Thumshirn 3 participants 3 messages Jun 6, 2025 – Jun 12, 2025 Btrfs file system

lore
Share & tools
lore
#0 lore raw

PATCH v3 1/1 btrfs: zoned: fix alloc_offset calculation for partly conventional block groups

From: Johannes Thumshirn <[email protected]>

When one of two zones composing a DUP block group is a conventional zone, we have the zone_info[i]->alloc_offset = WP_CONVENTIONAL. That will, of course, not match the write pointer of the other zone, and fails that block group.

This commit solves that issue by properly recovering the emulated write pointer from the last allocated extent. The offset for the SINGLE, DUP, and RAID1 are straight-forward: it is same as the end of last allocated extent. The RAID0 and RAID10 are a bit tricky that we need to do the math of striping.

This is the kernel equivalent of Naohiro's user-space commit:
1e85aa96e107 ("btrfs-progs: zoned: fix alloc_offset calculation for partly conventional block groups")

---
Changes to v2:
- Fix kbuild error on 32bit due to modulos

Changes to v1:
- Fix kbuild error on 32bit due to divisions

---
 fs/btrfs/zoned.c | 86 ++++++++++++++++++++++++++++++++++++++++--------
 1 file changed, 72 insertions(+), 14 deletions(-)
fs/btrfs/zoned.c +72 -14
index 4e122d6c19c0..79b72f6673e7 100644
@@ -1404,7 +1404,8 @@ static int btrfs_load_block_group_single(struct btrfs_block_group *bg,
 static int btrfs_load_block_group_dup(struct btrfs_block_group *bg,
 				      struct btrfs_chunk_map *map,
 				      struct zone_info *zone_info,
-				      unsigned long *active)
+				      unsigned long *active,
+				      u64 last_alloc)
 {
 	struct btrfs_fs_info *fs_info = bg->fs_info;
 
@@ -1427,6 +1428,13 @@ static int btrfs_load_block_group_dup(struct btrfs_block_group *bg,
 			  zone_info[1].physical);
 		return -EIO;
 	}
+
+	if (zone_info[0].alloc_offset == WP_CONVENTIONAL)
+		zone_info[0].alloc_offset = last_alloc;
+
+	if (zone_info[1].alloc_offset == WP_CONVENTIONAL)
+		zone_info[1].alloc_offset = last_alloc;
+
 	if (zone_info[0].alloc_offset != zone_info[1].alloc_offset) {
 		btrfs_err(bg->fs_info,
 			  "zoned: write pointer offset mismatch of zones in DUP profile");
@@ -1447,7 +1455,8 @@ static int btrfs_load_block_group_dup(struct btrfs_block_group *bg,
 static int btrfs_load_block_group_raid1(struct btrfs_block_group *bg,
 					struct btrfs_chunk_map *map,
 					struct zone_info *zone_info,
-					unsigned long *active)
+					unsigned long *active,
+					u64 last_alloc)
 {
 	struct btrfs_fs_info *fs_info = bg->fs_info;
 	int i;
@@ -1462,10 +1471,12 @@ static int btrfs_load_block_group_raid1(struct btrfs_block_group *bg,
 	bg->zone_capacity = min_not_zero(zone_info[0].capacity, zone_info[1].capacity);
 
 	for (i = 0; i < map->num_stripes; i++) {
-		if (zone_info[i].alloc_offset == WP_MISSING_DEV ||
-		    zone_info[i].alloc_offset == WP_CONVENTIONAL)
+		if (zone_info[i].alloc_offset == WP_MISSING_DEV)
 			continue;
 
+		if (zone_info[i].alloc_offset == WP_CONVENTIONAL)
+			zone_info[i].alloc_offset = last_alloc;
+
 		if ((zone_info[0].alloc_offset != zone_info[i].alloc_offset) &&
 		    !btrfs_test_opt(fs_info, DEGRADED)) {
 			btrfs_err(fs_info,
@@ -1495,7 +1506,8 @@ static int btrfs_load_block_group_raid1(struct btrfs_block_group *bg,
 static int btrfs_load_block_group_raid0(struct btrfs_block_group *bg,
 					struct btrfs_chunk_map *map,
 					struct zone_info *zone_info,
-					unsigned long *active)
+					unsigned long *active,
+					u64 last_alloc)
 {
 	struct btrfs_fs_info *fs_info = bg->fs_info;
 
@@ -1506,10 +1518,29 @@ static int btrfs_load_block_group_raid0(struct btrfs_block_group *bg,
 	}
 
 	for (int i = 0; i < map->num_stripes; i++) {
-		if (zone_info[i].alloc_offset == WP_MISSING_DEV ||
-		    zone_info[i].alloc_offset == WP_CONVENTIONAL)
+		if (zone_info[i].alloc_offset == WP_MISSING_DEV)
 			continue;
 
+		if (zone_info[i].alloc_offset == WP_CONVENTIONAL) {
+			u64 stripe_nr, full_stripe_nr;
+			u64 stripe_offset;
+			int stripe_index;
+
+			stripe_nr = div64_u64(last_alloc, map->stripe_size);
+			stripe_offset = stripe_nr * map->stripe_size;
+			full_stripe_nr = div_u64(stripe_nr, map->num_stripes);
+			div_u64_rem(stripe_nr, map->num_stripes, &stripe_index);
+
+			zone_info[i].alloc_offset =
+				full_stripe_nr * map->stripe_size;
+
+			if (stripe_index > i)
+				zone_info[i].alloc_offset += map->stripe_size;
+			else if (stripe_index == i)
+				zone_info[i].alloc_offset +=
+					(last_alloc - stripe_offset);
+		}
+
 		if (test_bit(0, active) != test_bit(i, active)) {
 			if (!btrfs_zone_activate(bg))
 				return -EIO;
@@ -1527,7 +1558,8 @@ static int btrfs_load_block_group_raid0(struct btrfs_block_group *bg,
 static int btrfs_load_block_group_raid10(struct btrfs_block_group *bg,
 					 struct btrfs_chunk_map *map,
 					 struct zone_info *zone_info,
-					 unsigned long *active)
+					 unsigned long *active,
+					 u64 last_alloc)
 {
 	struct btrfs_fs_info *fs_info = bg->fs_info;
 
@@ -1538,8 +1570,7 @@ static int btrfs_load_block_group_raid10(struct btrfs_block_group *bg,
 	}
 
 	for (int i = 0; i < map->num_stripes; i++) {
-		if (zone_info[i].alloc_offset == WP_MISSING_DEV ||
-		    zone_info[i].alloc_offset == WP_CONVENTIONAL)
+		if (zone_info[i].alloc_offset == WP_MISSING_DEV)
 			continue;
 
 		if (test_bit(0, active) != test_bit(i, active)) {
@@ -1550,6 +1581,29 @@ static int btrfs_load_block_group_raid10(struct btrfs_block_group *bg,
 				set_bit(BLOCK_GROUP_FLAG_ZONE_IS_ACTIVE, &bg->runtime_flags);
 		}
 
+		if (zone_info[i].alloc_offset == WP_CONVENTIONAL) {
+			u64 stripe_nr, full_stripe_nr;
+			u64 stripe_offset;
+			int stripe_index;
+
+			stripe_nr = div64_u64(last_alloc, map->stripe_size);
+			stripe_offset = stripe_nr * map->stripe_size;
+			full_stripe_nr = div_u64(stripe_nr,
+					 map->num_stripes / map->sub_stripes);
+			div_u64_rem(stripe_nr,
+				    (map->num_stripes / map->sub_stripes),
+				    &stripe_index);
+
+			zone_info[i].alloc_offset =
+				full_stripe_nr * map->stripe_size;
+
+			if (stripe_index > (i / map->sub_stripes))
+				zone_info[i].alloc_offset += map->stripe_size;
+			else if (stripe_index == (i / map->sub_stripes))
+				zone_info[i].alloc_offset +=
+					(last_alloc - stripe_offset);
+		}
+
 		if ((i % map->sub_stripes) == 0) {
 			bg->zone_capacity += zone_info[i].capacity;
 			bg->alloc_offset += zone_info[i].alloc_offset;
@@ -1638,18 +1692,22 @@ int btrfs_load_block_group_zone_info(struct btrfs_block_group *cache, bool new)
 		ret = btrfs_load_block_group_single(cache, &zone_info[0], active);
 		break;
 	case BTRFS_BLOCK_GROUP_DUP:
-		ret = btrfs_load_block_group_dup(cache, map, zone_info, active);
+		ret = btrfs_load_block_group_dup(cache, map, zone_info, active,
+						 last_alloc);
 		break;
 	case BTRFS_BLOCK_GROUP_RAID1:
 	case BTRFS_BLOCK_GROUP_RAID1C3:
 	case BTRFS_BLOCK_GROUP_RAID1C4:
-		ret = btrfs_load_block_group_raid1(cache, map, zone_info, active);
+		ret = btrfs_load_block_group_raid1(cache, map, zone_info,
+						   active, last_alloc);
 		break;
 	case BTRFS_BLOCK_GROUP_RAID0:
-		ret = btrfs_load_block_group_raid0(cache, map, zone_info, active);
+		ret = btrfs_load_block_group_raid0(cache, map, zone_info,
+						   active, last_alloc);
 		break;
 	case BTRFS_BLOCK_GROUP_RAID10:
-		ret = btrfs_load_block_group_raid10(cache, map, zone_info, active);
+		ret = btrfs_load_block_group_raid10(cache, map, zone_info,
+						    active, last_alloc);
 		break;
 	case BTRFS_BLOCK_GROUP_RAID5:
 	case BTRFS_BLOCK_GROUP_RAID6:
Signature · 1 line
2.49.0
Reviewed-by on v3 1/1
#2 lore raw
In reply to Johannes Thumshirn
Quoted 13 lines · Johannes Thumshirn

From: Johannes Thumshirn <[email protected]>

When one of two zones composing a DUP block group is a conventional zone, we have the zone_info[i]->alloc_offset = WP_CONVENTIONAL. That will, of course, not match the write pointer of the other zone, and fails that block group.

This commit solves that issue by properly recovering the emulated write pointer from the last allocated extent. The offset for the SINGLE, DUP, and RAID1 are straight-forward: it is same as the end of last allocated extent. The RAID0 and RAID10 are a bit tricky that we need to do the math of striping.

Quoted 4 lines

This is the kernel equivalent of Naohiro's user-space commit:
1e85aa96e107 ("btrfs-progs: zoned: fix alloc_offset calculation for partly conventional block groups")

Thanks,