Skip to contentSkip to conversation

btrfs-progs: extent tree v2 support, global roots

PATCH In review Josef Bacik 4 participants 29 messages Mar 7, 2022 – Jun 14, 2022 Btrfs file system

lore
Share & tools
lore
#0 lore raw

v5 cover btrfs-progs: extent tree v2 support, global roots

v4->v3:
- Rebase onto devel, depends on "btrfs-progs: cleanup btrfs_item* accessors".
- Dropped the various patches that have already been merged into -progs.

v3->v4:
- Rebase onto devel, depends on the v3 prep patches that were sent on December 1st which has the rest of the "don't access ->*_root" patches.
- I think I screwed up the versioning of this, but I lost the other submission, so call this v3.

v1->v2:
- These depend on the v3 of the prep patches (it's marked as v2 because I'm stupid, but the second v2 posting I sent.)
- I've moved the global root rb tree patches into this series to differentiate them from the actual fixes in the prep series.

--- Original email ---
Hello,

These patches are the first chunk of the extent tree v2 format changes. This includes the separate block group root which will hold all of the block group items. This also includes the global root support, which is the work to allow us to have multiple extent, csum, and free space trees in the same file system.

The goal of these two changes are straightforward. For the block group root, on very large file systems the block group items are very widely separated, which means it takes a very long time to mount the file system on large, slow disks. Putting the block group items in their own root will allow us to densely populate the tree and dramatically increase mount times in these cases.

The global roots change is motivated by lock contention on the root nodes of these global roots. I've had to make many changes to how we run delayed refs to speed up things like the transaction commit because of all the delayed refs going into one tree and contending on the root node of the extent tree. In the same token you can have heavy lock contention on the csum roots when writing to many files. Allowing for multiple roots will let us spread the lock contention load around.

I have disabled a few key features, namely balance and qgroups. There will be more to come as I make more and more invasive changes, and then they will slowly be re-enabled as the work is added. These are disabled to avoid a bunch of work that would be thrown away by future changes.

These patches have passed xfstests without panicing, but clearly failing a lot of tests because of the disabled features. I've also run it through fsperf to validate that there are no major performance regressions.

WARNING: there are many more format changes planned, this is just the first batch. If you want to test then please feel free, but know that the format is still in flux. Thanks,

Josef

Josef Bacik (19):
btrfs-progs: add support for loading the block group root
btrfs-progs: add print support for the block group tree
btrfs-progs: mkfs: use the btrfs_block_group_root helper
btrfs-progs: check-lowmem: use the btrfs_block_group_root helper
btrfs-progs: handle no bg item in extent tree for free space tree
btrfs-progs: mkfs: add support for the block group tree
btrfs-progs: check: add block group tree support
btrfs-progs: qgroup-verify: scan extents based on block groups
btrfs-progs: check: make free space tree validation extent tree v2
aware
btrfs-progs: check: add helper to reinit the root based on a key
btrfs-progs: check: handle the block group tree properly
btrfs-progs: set the number of global roots in the super block
btrfs-progs: handle the per-block group global root id
btrfs-progs: add a btrfs_delete_and_free_root helper
btrfs-progs: make btrfs_clear_free_space_tree extent tree v2 aware
btrfs-progs: make btrfs_create_tree take a key for the root key
btrfs-progs: mkfs: set chunk_item_objectid properly for extent tree v2
btrfs-progs: mkfs: create the global root's
btrfs-progs: check: don't do the root item check for extent tree v2

 check/main.c                    | 233 +++++++++++++++++--------------
 check/mode-lowmem.c             |  12 +-
 check/qgroup-verify.c           |  32 +++--
 cmds/inspect-dump-tree.c        |  30 +++-
 common/repair.c                 |   3 +
 kernel-shared/ctree.h           |   9 +-
 kernel-shared/disk-io.c         | 235 ++++++++++++++++++++++++--------
 kernel-shared/disk-io.h         |  15 +-
 kernel-shared/extent-tree.c     |  32 ++++-
 kernel-shared/free-space-tree.c |  72 +++++-----
 kernel-shared/print-tree.c      |  23 +++-
 kernel-shared/transaction.c     |   2 +
 mkfs/common.c                   |  94 ++++++++++---
 mkfs/common.h                   |  12 ++
 mkfs/main.c                     |  93 ++++++++++++-
 15 files changed, 658 insertions(+), 239 deletions(-)
Signature · 1 line
2.26.3
#1 lore raw

PATCH v5 1/19 btrfs-progs: add support for loading the block group root

This adds the ability to load the block group root, as well as make sure the various backup super block and super block updates are made appropriately.

---
 kernel-shared/ctree.h       |   1 +
 kernel-shared/disk-io.c     | 161 +++++++++++++++++++++++++++---------
 kernel-shared/disk-io.h     |  10 ++-
 kernel-shared/extent-tree.c |   8 +-
 kernel-shared/transaction.c |   2 +
 5 files changed, 138 insertions(+), 44 deletions(-)
kernel-shared/ctree.h +1 -0
index addfafc7..b12dbff1 100644
@@ -1201,6 +1201,7 @@ struct btrfs_fs_info {
 	struct btrfs_root *dev_root;
 	struct btrfs_root *quota_root;
 	struct btrfs_root *uuid_root;
+	struct btrfs_root *block_group_root;
 
 	struct rb_root global_roots_tree;
 	struct rb_root fs_root_tree;
kernel-shared/disk-io.c +122 -39

266 lines not on this page · raw message

kernel-shared/disk-io.h +9 -1
index d55ced1e..1e9044f8 100644
@@ -223,9 +223,17 @@ struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
 				     u64 objectid);
 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr);
 struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_inf, u64 bytenr);
-struct btrfs_root *btrfs_block_group_root(struct btrfs_fs_info *fs_info);
 struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
 				     struct btrfs_key *key);
 int btrfs_global_root_insert(struct btrfs_fs_info *fs_info,
 			     struct btrfs_root *root);
+
+static inline struct btrfs_root *btrfs_block_group_root(
+						struct btrfs_fs_info *fs_info)
+{
+	if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
+		return fs_info->block_group_root;
+	return btrfs_extent_root(fs_info, 0);
+}
+
 #endif
kernel-shared/extent-tree.c +4 -4
index e36745ca..b2b99d4f 100644
@@ -1540,7 +1540,7 @@ static int update_block_group_item(struct btrfs_trans_handle *trans,
 {
 	int ret;
 	struct btrfs_fs_info *fs_info = trans->fs_info;
-	struct btrfs_root *root = btrfs_extent_root(fs_info, 0);
+	struct btrfs_root *root = btrfs_block_group_root(fs_info);
 	unsigned long bi;
 	struct btrfs_block_group_item bgi;
 	struct extent_buffer *leaf;
@@ -2731,7 +2731,7 @@ int btrfs_read_block_groups(struct btrfs_fs_info *fs_info)
 	int ret;
 	struct btrfs_key key;
 
-	root = btrfs_extent_root(fs_info, 0);
+	root = btrfs_block_group_root(fs_info);
 	key.objectid = 0;
 	key.offset = 0;
 	key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
@@ -2812,7 +2812,7 @@ static int insert_block_group_item(struct btrfs_trans_handle *trans,
 	key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
 	key.offset = block_group->length;
 
-	root = btrfs_extent_root(fs_info, 0);
+	root = btrfs_block_group_root(fs_info);
 	return btrfs_insert_item(trans, root, &key, &bgi, sizeof(bgi));
 }
 
@@ -2929,7 +2929,7 @@ static int remove_block_group_item(struct btrfs_trans_handle *trans,
 {
 	struct btrfs_fs_info *fs_info = trans->fs_info;
 	struct btrfs_key key;
-	struct btrfs_root *root = btrfs_extent_root(fs_info, 0);
+	struct btrfs_root *root = btrfs_block_group_root(fs_info);
 	int ret = 0;
 
 	key.objectid = block_group->start;
kernel-shared/transaction.c +2 -0
index 5b991651..02012266 100644
@@ -185,6 +185,8 @@ int btrfs_commit_transaction(struct btrfs_trans_handle *trans,
 		goto commit_tree;
 	if (root == root->fs_info->chunk_root)
 		goto commit_tree;
+	if (root == root->fs_info->block_group_root)
+		goto commit_tree;
 
 	free_extent_buffer(root->commit_root);
 	root->commit_root = NULL;
Signature · 1 line
2.26.3
#2 lore raw

PATCH v5 2/19 btrfs-progs: add print support for the block group tree

In reply to Josef Bacik

Add the appropriate support to the print tree and dump tree code to spit out the block group tree.

---
 cmds/inspect-dump-tree.c   | 30 +++++++++++++++++++++++++++++-
 kernel-shared/print-tree.c | 23 +++++++++++++++++++----
 2 files changed, 48 insertions(+), 5 deletions(-)
cmds/inspect-dump-tree.c +29 -1
index 6332b46d..daa7f925 100644
@@ -83,8 +83,14 @@ out:
 
 static void print_old_roots(struct btrfs_super_block *super)
 {
+	const char *extent_tree_str = "extent root";
 	struct btrfs_root_backup *backup;
 	int i;
+	bool extent_tree_v2 = (btrfs_super_incompat_flags(super) &
+		BTRFS_FEATURE_INCOMPAT_EXTENT_TREE_V2);
+
+	if (extent_tree_v2)
+		extent_tree_str = "block group root";
 
 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
 		backup = super->super_roots + i;
@@ -93,7 +99,7 @@ static void print_old_roots(struct btrfs_super_block *super)
 		       (unsigned long long)btrfs_backup_tree_root_gen(backup),
 		       (unsigned long long)btrfs_backup_tree_root(backup));
 
-		printf("\t\textent root gen %llu block %llu\n",
+		printf("\t\t%s gen %llu block %llu\n", extent_tree_str,
 		       (unsigned long long)btrfs_backup_extent_root_gen(backup),
 		       (unsigned long long)btrfs_backup_extent_root(backup));
 
@@ -510,6 +516,11 @@ static int cmd_inspect_dump_tree(const struct cmd_struct *cmd,
 				       info->log_root_tree->node->start,
 					btrfs_header_level(
 						info->log_root_tree->node));
+			if (info->block_group_root)
+				printf("block group tree: %llu level %d\n",
+				       info->block_group_root->node->start,
+					btrfs_header_level(
+						info->block_group_root->node));
 		} else {
 			if (info->tree_root->node) {
 				printf("root tree\n");
@@ -528,6 +539,12 @@ static int cmd_inspect_dump_tree(const struct cmd_struct *cmd,
 				btrfs_print_tree(info->log_root_tree->node,
 					BTRFS_PRINT_TREE_FOLLOW | print_mode);
 			}
+
+			if (info->block_group_root) {
+				printf("block group tree\n");
+				btrfs_print_tree(info->block_group_root->node,
+					BTRFS_PRINT_TREE_FOLLOW | print_mode);
+			}
 		}
 	}
 	tree_root_scan = info->tree_root;
@@ -573,6 +590,17 @@ again:
 		goto close_root;
 	}
 
+	if (tree_id && tree_id == BTRFS_BLOCK_GROUP_TREE_OBJECTID) {
+		if (!info->block_group_root) {
+			error("cannot print block group tree, invalid pointer");
+			goto close_root;
+		}
+		printf("block group tree\n");
+		btrfs_print_tree(info->block_group_root->node,
+					BTRFS_PRINT_TREE_FOLLOW | print_mode);
+		goto close_root;
+	}
+
 	key.offset = 0;
 	key.objectid = 0;
 	key.type = BTRFS_ROOT_ITEM_KEY;
kernel-shared/print-tree.c +19 -4
index 717be5d5..978d92bc 100644
@@ -1872,8 +1872,14 @@ static int empty_backup(struct btrfs_root_backup *backup)
 	return 0;
 }
 
-static void print_root_backup(struct btrfs_root_backup *backup)
+static void print_root_backup(struct btrfs_root_backup *backup,
+			      bool extent_tree_v2)
 {
+	const char *extent_tree_str = "backup_extent_root";
+
+	if (extent_tree_v2)
+		extent_tree_str = "backup_block_group_root";
+
 	printf("\t\tbackup_tree_root:\t%llu\tgen: %llu\tlevel: %d\n",
 			btrfs_backup_tree_root(backup),
 			btrfs_backup_tree_root_gen(backup),
@@ -1882,7 +1888,8 @@ static void print_root_backup(struct btrfs_root_backup *backup)
 			btrfs_backup_chunk_root(backup),
 			btrfs_backup_chunk_root_gen(backup),
 			btrfs_backup_chunk_root_level(backup));
-	printf("\t\tbackup_extent_root:\t%llu\tgen: %llu\tlevel: %d\n",
+	printf("\t\t%s:\t%llu\tgen: %llu\tlevel: %d\n",
+			extent_tree_str,
 			btrfs_backup_extent_root(backup),
 			btrfs_backup_extent_root_gen(backup),
 			btrfs_backup_extent_root_level(backup));
@@ -1894,7 +1901,7 @@ static void print_root_backup(struct btrfs_root_backup *backup)
 			btrfs_backup_dev_root(backup),
 			btrfs_backup_dev_root_gen(backup),
 			btrfs_backup_dev_root_level(backup));
-	printf("\t\tbackup_csum_root:\t%llu\tgen: %llu\tlevel: %d\n",
+	printf("\t\tcsum_root:\t%llu\tgen: %llu\tlevel: %d\n",
 			btrfs_backup_csum_root(backup),
 			btrfs_backup_csum_root_gen(backup),
 			btrfs_backup_csum_root_level(backup));
@@ -1912,12 +1919,14 @@ static void print_backup_roots(struct btrfs_super_block *sb)
 {
 	struct btrfs_root_backup *backup;
 	int i;
+	bool extent_tree_v2 = (btrfs_super_incompat_flags(sb) &
+		BTRFS_FEATURE_INCOMPAT_EXTENT_TREE_V2);
 
 	for (i = 0; i < BTRFS_NUM_BACKUP_ROOTS; i++) {
 		backup = sb->super_roots + i;
 		if (!empty_backup(backup)) {
 			printf("\tbackup %d:\n", i);
-			print_root_backup(backup);
+			print_root_backup(backup, extent_tree_v2);
 		}
 	}
 }
@@ -2034,6 +2043,12 @@ void btrfs_print_superblock(struct btrfs_super_block *sb, int full)
 	       (unsigned long long)btrfs_super_cache_generation(sb));
 	printf("uuid_tree_generation\t%llu\n",
 	       (unsigned long long)btrfs_super_uuid_tree_generation(sb));
+	printf("block_group_root\t%llu\n",
+	       (unsigned long long)btrfs_super_block_group_root(sb));
+	printf("block_group_root_generation\t%llu\n",
+	       (unsigned long long)btrfs_super_block_group_root_generation(sb));
+	printf("block_group_root_level\t%llu\n",
+	       (unsigned long long)btrfs_super_block_group_root_level(sb));
 
 	uuid_unparse(sb->dev_item.uuid, buf);
 	printf("dev_item.uuid\t\t%s\n", buf);
Signature · 1 line
2.26.3
#3 lore raw

PATCH v5 3/19 btrfs-progs: mkfs: use the btrfs_block_group_root helper

In reply to Josef Bacik

Instead of accessing the extent root directory for modifying block groups, use the helper which will do the correct thing based on the flags of the file system.

---
 check/main.c | 4 ++--
 mkfs/main.c  | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)
check/main.c +2 -2
index 39cb1ce5..6ddfd18a 100644
@@ -9426,6 +9426,7 @@ static int reinit_global_roots(struct btrfs_trans_handle *trans, u64 objectid)
 
 static int reinit_extent_tree(struct btrfs_trans_handle *trans, bool pin)
 {
+	struct btrfs_root *bg_root = btrfs_block_group_root(trans->fs_info);
 	u64 start = 0;
 	int ret;
 
@@ -9499,7 +9500,6 @@ again:
 	while (1) {
 		struct btrfs_block_group_item bgi;
 		struct btrfs_block_group *cache;
-		struct btrfs_root *extent_root = btrfs_extent_root(gfs_info, 0);
 		struct btrfs_key key;
 
 		cache = btrfs_lookup_first_block_group(gfs_info, start);
@@ -9513,7 +9513,7 @@ again:
 		key.objectid = cache->start;
 		key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
 		key.offset = cache->length;
-		ret = btrfs_insert_item(trans, extent_root, &key, &bgi,
+		ret = btrfs_insert_item(trans, bg_root, &key, &bgi,
 					sizeof(bgi));
 		if (ret) {
 			fprintf(stderr, "Error adding block group\n");
mkfs/main.c +1 -1
index 3dd06979..20dc0436 100644
@@ -596,7 +596,7 @@ static int cleanup_temp_chunks(struct btrfs_fs_info *fs_info,
 {
 	struct btrfs_trans_handle *trans = NULL;
 	struct btrfs_block_group_item *bgi;
-	struct btrfs_root *root = btrfs_extent_root(fs_info, 0);
+	struct btrfs_root *root = btrfs_block_group_root(fs_info);
 	struct btrfs_key key;
 	struct btrfs_key found_key;
 	struct btrfs_path path;
Signature · 1 line
2.26.3
#4 lore raw

PATCH v5 4/19 btrfs-progs: check-lowmem: use the btrfs_block_group_root helper

In reply to Josef Bacik

When we're messing with block group items use the btrfs_block_group_root() helper to get the correct root to search, and this will do the right thing based on the file system flags.

---
 check/mode-lowmem.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
check/mode-lowmem.c +4 -4
index 99d04945..8535e684 100644
@@ -266,7 +266,7 @@ static int modify_block_group_cache(struct btrfs_block_group *block_group, int c
  */
 static int modify_block_groups_cache(u64 flags, int cache)
 {
-	struct btrfs_root *root = btrfs_extent_root(gfs_info, 0);
+	struct btrfs_root *root = btrfs_block_group_root(gfs_info);
 	struct btrfs_key key;
 	struct btrfs_path path;
 	struct btrfs_block_group *bg_cache;
@@ -331,7 +331,7 @@ static int clear_block_groups_full(u64 flags)
 static int create_chunk_and_block_group(u64 flags, u64 *start, u64 *nbytes)
 {
 	struct btrfs_trans_handle *trans;
-	struct btrfs_root *root = btrfs_extent_root(gfs_info, 0);
+	struct btrfs_root *root = btrfs_block_group_root(gfs_info);
 	int ret;
 
 	if ((flags & BTRFS_BLOCK_GROUP_TYPE_MASK) == 0)
@@ -419,7 +419,7 @@ static int is_chunk_almost_full(u64 start)
 {
 	struct btrfs_path path;
 	struct btrfs_key key;
-	struct btrfs_root *root = btrfs_extent_root(gfs_info, 0);
+	struct btrfs_root *root = btrfs_block_group_root(gfs_info);
 	struct btrfs_block_group_item *bi;
 	struct btrfs_block_group_item bg_item;
 	struct extent_buffer *eb;
@@ -4601,7 +4601,7 @@ next:
 static int find_block_group_item(struct btrfs_path *path, u64 bytenr, u64 len,
 				 u64 type)
 {
-	struct btrfs_root *root = btrfs_extent_root(gfs_info, 0);
+	struct btrfs_root *root = btrfs_block_group_root(gfs_info);
 	struct btrfs_block_group_item bgi;
 	struct btrfs_key key;
 	int ret;
Signature · 1 line
2.26.3
#5 lore raw

PATCH v5 5/19 btrfs-progs: handle no bg item in extent tree for free space tree

In reply to Josef Bacik

We have an ASSERT(ret == 0) when populating the free space tree as we should at least find the block group item with extent tree v1. However with v2 we no longer have the block group item in the extent tree, so fix the population logic to handle an empty block group (which occurs during mkfs) and only assert if ret != 0 and we don't have extent tree v2 turned on.

---
 kernel-shared/free-space-tree.c | 11 ++++++++---
 1 file changed, 8 insertions(+), 3 deletions(-)
kernel-shared/free-space-tree.c +8 -3
index 0fdf5004..896bd3a2 100644
@@ -1057,6 +1057,9 @@ int populate_free_space_tree(struct btrfs_trans_handle *trans,
 	if (ret)
 		goto out;
 
+	start = block_group->start;
+	end = block_group->start + block_group->length;
+
 	/*
 	 * Iterate through all of the extent and metadata items in this block
 	 * group, adding the free space between them and the free space at the
@@ -1071,10 +1074,11 @@ int populate_free_space_tree(struct btrfs_trans_handle *trans,
 	ret = btrfs_search_slot_for_read(extent_root, &key, path, 1, 0);
 	if (ret < 0)
 		goto out;
-	ASSERT(ret == 0);
+	if (ret > 0) {
+		ASSERT(btrfs_fs_incompat(trans->fs_info, EXTENT_TREE_V2));
+		goto done;
+	}
 
-	start = block_group->start;
-	end = block_group->start + block_group->length;
 	while (1) {
 		btrfs_item_key_to_cpu(path->nodes[0], &key, path->slots[0]);
 
@@ -1106,6 +1110,7 @@ int populate_free_space_tree(struct btrfs_trans_handle *trans,
 		if (ret)
 			break;
 	}
+done:
 	if (start < end) {
 		ret = __add_to_free_space_tree(trans, block_group, path2,
 				start, end - start);
Signature · 1 line
2.26.3
#6 lore raw

PATCH v5 6/19 btrfs-progs: mkfs: add support for the block group tree

In reply to Josef Bacik

Add the extent tree v2 table with the block group tree as a root, and then create the empty root and use the proper root for cleanup up the temporary block groups.

---
 mkfs/common.c | 93 ++++++++++++++++++++++++++++++++++++++++-----------
 mkfs/common.h | 12 +++++++
 mkfs/main.c   |  5 +++
 3 files changed, 91 insertions(+), 19 deletions(-)
mkfs/common.c +74 -19
index 11d92c8b..aa65543b 100644
@@ -39,6 +39,7 @@ static u64 reference_root_table[] = {
 	[MKFS_FS_TREE]		=	BTRFS_FS_TREE_OBJECTID,
 	[MKFS_CSUM_TREE]	=	BTRFS_CSUM_TREE_OBJECTID,
 	[MKFS_FREE_SPACE_TREE]	=	BTRFS_FREE_SPACE_TREE_OBJECTID,
+	[MKFS_BLOCK_GROUP_TREE]	=	BTRFS_BLOCK_GROUP_TREE_OBJECTID,
 };
 
 static int btrfs_write_empty_tree(int fd, struct btrfs_mkfs_config *cfg,
@@ -97,7 +98,8 @@ static int btrfs_create_tree_root(int fd, struct btrfs_mkfs_config *cfg,
 
 	for (i = 0; i < blocks_nr; i++) {
 		blk = blocks[i];
-		if (blk == MKFS_ROOT_TREE || blk == MKFS_CHUNK_TREE)
+		if (blk == MKFS_ROOT_TREE || blk == MKFS_CHUNK_TREE ||
+		    blk == MKFS_BLOCK_GROUP_TREE)
 			continue;
 
 		btrfs_set_root_bytenr(&root_item, cfg->blocks[blk]);
@@ -187,6 +189,50 @@ static int create_free_space_tree(int fd, struct btrfs_mkfs_config *cfg,
 	return 0;
 }
 
+static void write_block_group_item(struct extent_buffer *buf, u32 nr,
+				   u64 objectid, u64 offset, u64 used,
+				   u32 itemoff)
+{
+	struct btrfs_block_group_item *bg_item;
+	struct btrfs_disk_key disk_key;
+
+	btrfs_set_disk_key_objectid(&disk_key, objectid);
+	btrfs_set_disk_key_offset(&disk_key, offset);
+	btrfs_set_disk_key_type(&disk_key, BTRFS_BLOCK_GROUP_ITEM_KEY);
+	btrfs_set_item_key(buf, &disk_key, nr);
+	btrfs_set_item_offset(buf, nr, itemoff);
+	btrfs_set_item_size(buf, nr, sizeof(*bg_item));
+
+	bg_item = btrfs_item_ptr(buf, nr, struct btrfs_block_group_item);
+	btrfs_set_block_group_used(buf, bg_item, used);
+	btrfs_set_block_group_flags(buf, bg_item, BTRFS_BLOCK_GROUP_SYSTEM);
+	btrfs_set_block_group_chunk_objectid(buf, bg_item,
+					     BTRFS_FIRST_CHUNK_TREE_OBJECTID);
+}
+
+static int create_block_group_tree(int fd, struct btrfs_mkfs_config *cfg,
+				   struct extent_buffer *buf,
+				   u64 bg_offset, u64 bg_size, u64 bg_used)
+{
+	int ret;
+
+	memset(buf->data + sizeof(struct btrfs_header), 0,
+		cfg->nodesize - sizeof(struct btrfs_header));
+	write_block_group_item(buf, 0, bg_offset, bg_size, bg_used,
+			       cfg->leaf_data_size -
+			       sizeof(struct btrfs_block_group_item));
+	btrfs_set_header_bytenr(buf, cfg->blocks[MKFS_BLOCK_GROUP_TREE]);
+	btrfs_set_header_owner(buf, BTRFS_BLOCK_GROUP_TREE_OBJECTID);
+	btrfs_set_header_nritems(buf, 1);
+	csum_tree_block_size(buf, btrfs_csum_type_size(cfg->csum_type), 0,
+			     cfg->csum_type);
+	ret = pwrite(fd, buf->data, cfg->nodesize,
+		     cfg->blocks[MKFS_BLOCK_GROUP_TREE]);
+	if (ret != cfg->nodesize)
+		return ret < 0 ? -errno : -EIO;
+	return 0;
+}
+
 /*
  * @fs_uuid - if NULL, generates a UUID, returns back the new filesystem UUID
  *
@@ -239,11 +285,19 @@ int make_btrfs(int fd, struct btrfs_mkfs_config *cfg)
 	bool add_block_group = true;
 	bool free_space_tree = !!(cfg->runtime_features &
 				  BTRFS_RUNTIME_FEATURE_FREE_SPACE_TREE);
+	bool extent_tree_v2 = !!(cfg->features &
+				 BTRFS_FEATURE_INCOMPAT_EXTENT_TREE_V2);
 
 	/* Don't include the free space tree in the blocks to process. */
 	if (!free_space_tree)
 		blocks_nr--;
 
+	if (extent_tree_v2) {
+		blocks = extent_tree_v2_blocks;
+		blocks_nr = ARRAY_SIZE(extent_tree_v2_blocks);
+		add_block_group = false;
+	}
+
 	if ((cfg->features & BTRFS_FEATURE_INCOMPAT_ZONED)) {
 		system_group_offset = cfg->zone_size * BTRFS_NR_SB_LOG_ZONES;
 		system_group_size = cfg->zone_size;
@@ -300,6 +354,12 @@ int make_btrfs(int fd, struct btrfs_mkfs_config *cfg)
 		btrfs_set_super_compat_ro_flags(&super, ro_flags);
 		btrfs_set_super_cache_generation(&super, 0);
 	}
+	if (extent_tree_v2) {
+		btrfs_set_super_block_group_root(&super,
+						 cfg->blocks[MKFS_BLOCK_GROUP_TREE]);
+		btrfs_set_super_block_group_root_generation(&super, 1);
+		btrfs_set_super_block_group_root_level(&super, 0);
+	}
 	if (cfg->label)
 		__strncpy_null(super.label, cfg->label, BTRFS_LABEL_SIZE - 1);
 
@@ -331,25 +391,12 @@ int make_btrfs(int fd, struct btrfs_mkfs_config *cfg)
 
 		/* Add the block group item for our temporary chunk. */
 		if (cfg->blocks[blk] > system_group_offset && add_block_group) {
-			struct btrfs_block_group_item *bg_item;
-
+			itemoff -= sizeof(struct btrfs_block_group_item);
+			write_block_group_item(buf, nritems,
+					       system_group_offset,
+					       system_group_size, total_used,
+					       itemoff);
 			add_block_group = false;
-
-			itemoff -= sizeof(*bg_item);
-			btrfs_set_disk_key_objectid(&disk_key, system_group_offset);
-			btrfs_set_disk_key_offset(&disk_key, system_group_size);
-			btrfs_set_disk_key_type(&disk_key, BTRFS_BLOCK_GROUP_ITEM_KEY);
-			btrfs_set_item_key(buf, &disk_key, nritems);
-			btrfs_set_item_offset(buf, nritems, itemoff);
-			btrfs_set_item_size(buf, nritems, sizeof(*bg_item));
-
-			bg_item = btrfs_item_ptr(buf, nritems,
-						 struct btrfs_block_group_item);
-			btrfs_set_block_group_used(buf, bg_item, total_used);
-			btrfs_set_block_group_flags(buf, bg_item,
-						    BTRFS_BLOCK_GROUP_SYSTEM);
-			btrfs_set_block_group_chunk_objectid(buf, bg_item,
-					BTRFS_FIRST_CHUNK_TREE_OBJECTID);
 			nritems++;
 		}
 
@@ -565,6 +612,14 @@ int make_btrfs(int fd, struct btrfs_mkfs_config *cfg)
 			goto out;
 	}
 
+	if (extent_tree_v2) {
+		ret = create_block_group_tree(fd, cfg, buf,
+					      system_group_offset,
+					      system_group_size, total_used);
+		if (ret)
+			goto out;
+	}
+
 	/* and write out the super block */
 	memset(buf->data, 0, BTRFS_SUPER_INFO_SIZE);
 	memcpy(buf->data, &super, sizeof(super));
mkfs/common.h +12 -0
index 428cd366..3533e114 100644
@@ -51,6 +51,7 @@ enum btrfs_mkfs_block {
 	MKFS_FS_TREE,
 	MKFS_CSUM_TREE,
 	MKFS_FREE_SPACE_TREE,
+	MKFS_BLOCK_GROUP_TREE,
 	MKFS_BLOCK_COUNT
 };
 
@@ -69,6 +70,17 @@ static const enum btrfs_mkfs_block extent_tree_v1_blocks[] = {
 	MKFS_FREE_SPACE_TREE,
 };
 
+static const enum btrfs_mkfs_block extent_tree_v2_blocks[] = {
+	MKFS_ROOT_TREE,
+	MKFS_EXTENT_TREE,
+	MKFS_CHUNK_TREE,
+	MKFS_DEV_TREE,
+	MKFS_FS_TREE,
+	MKFS_CSUM_TREE,
+	MKFS_FREE_SPACE_TREE,
+	MKFS_BLOCK_GROUP_TREE,
+};
+
 struct btrfs_mkfs_config {
 	/* Label of the new filesystem */
 	const char *label;
mkfs/main.c +5 -0
index 20dc0436..7f79ba1a 100644
@@ -299,6 +299,11 @@ static int recow_roots(struct btrfs_trans_handle *trans,
 	ret = __recow_root(trans, info->dev_root);
 	if (ret)
 		return ret;
+        if (btrfs_fs_incompat(info, EXTENT_TREE_V2)) {
+		ret = __recow_root(trans, info->block_group_root);
+		if (ret)
+			return ret;
+        }
 	ret = recow_global_roots(trans);
 	if (ret)
 		return ret;
Signature · 1 line
2.26.3
#7 lore raw

PATCH v5 7/19 btrfs-progs: check: add block group tree support

In reply to Josef Bacik

This makes the appropriate changes to enable the block group tree checking for both lowmem and normal check modes. This is relatively straightforward, simply need to use the helper to get the right root for dealing with block groups.

---
 check/main.c        | 21 ++++++++++++++++++++-
 check/mode-lowmem.c |  4 ++--
 2 files changed, 22 insertions(+), 3 deletions(-)
check/main.c +20 -1
index 6ddfd18a..a0f4ee91 100644
@@ -6268,10 +6268,17 @@ static int check_type_with_root(u64 rootid, u8 key_type)
 		break;
 	case BTRFS_EXTENT_ITEM_KEY:
 	case BTRFS_METADATA_ITEM_KEY:
-	case BTRFS_BLOCK_GROUP_ITEM_KEY:
 		if (rootid != BTRFS_EXTENT_TREE_OBJECTID)
 			goto err;
 		break;
+	case BTRFS_BLOCK_GROUP_ITEM_KEY:
+		if (btrfs_fs_incompat(gfs_info, EXTENT_TREE_V2)) {
+			if (rootid != BTRFS_BLOCK_GROUP_TREE_OBJECTID)
+				goto err;
+		} else if (rootid != BTRFS_EXTENT_TREE_OBJECTID) {
+			goto err;
+		}
+		break;
 	case BTRFS_ROOT_ITEM_KEY:
 		if (rootid != BTRFS_ROOT_TREE_OBJECTID)
 			goto err;
@@ -9492,6 +9499,18 @@ again:
 		return ret;
 	}
 
+	/*
+	 * If we are extent tree v2 then we can reint the block group root as
+	 * well.
+	 */
+	if (btrfs_fs_incompat(gfs_info, EXTENT_TREE_V2)) {
+		ret = btrfs_fsck_reinit_root(trans, gfs_info->block_group_root);
+		if (ret) {
+			fprintf(stderr, "block group initialization failed\n");
+			return ret;
+		}
+	}
+
 	/*
 	 * Now we have all the in-memory block groups setup so we can make
 	 * allocations properly, and the metadata we care about is safe since we
check/mode-lowmem.c +2 -2
index 8535e684..68c1adfd 100644
@@ -5540,7 +5540,7 @@ int check_chunks_and_extents_lowmem(void)
 	key.offset = 0;
 	key.type = BTRFS_ROOT_ITEM_KEY;
 
-	ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
+	ret = btrfs_search_slot(NULL, gfs_info->tree_root, &key, &path, 0, 0);
 	if (ret) {
 		error("cannot find extent tree in tree_root");
 		goto out;
@@ -5575,7 +5575,7 @@ int check_chunks_and_extents_lowmem(void)
 		if (ret)
 			goto out;
 next:
-		ret = btrfs_next_item(root, &path);
+		ret = btrfs_next_item(gfs_info->tree_root, &path);
 		if (ret)
 			goto out;
 	}
Signature · 1 line
2.26.3
#8 lore raw

PATCH v5 8/19 btrfs-progs: qgroup-verify: scan extents based on block groups

In reply to Josef Bacik

When we switch to per-block group extent roots we'll need to scan each individual extent root. To make this easier in the future go ahead and use the range of the block groups to scan the extents.

---
 check/qgroup-verify.c | 32 ++++++++++++++++++++++++--------
 1 file changed, 24 insertions(+), 8 deletions(-)
check/qgroup-verify.c +24 -8
index 2c05f875..6e012c1f 100644
@@ -1400,6 +1400,7 @@ static bool is_bad_qgroup(struct qgroup_count *count)
  */
 int qgroup_verify_all(struct btrfs_fs_info *info)
 {
+	struct rb_node *n;
 	int ret;
 	bool found_err = false;
 	bool skip_err = false;
@@ -1430,10 +1431,17 @@ int qgroup_verify_all(struct btrfs_fs_info *info)
 	/*
 	 * Put all extent refs into our rbtree
 	 */
-	ret = scan_extents(info, 0, ~0ULL);
-	if (ret) {
-		fprintf(stderr, "ERROR: while scanning extent tree: %d\n", ret);
-		goto out;
+	for (n = rb_first(&info->block_group_cache_tree); n; n = rb_next(n)) {
+		struct btrfs_block_group *bg;
+
+		bg = rb_entry(n, struct btrfs_block_group, cache_node);
+		ret = scan_extents(info, bg->start,
+				   bg->start + bg->length - 1);
+		if (ret) {
+			fprintf(stderr, "ERROR: while scanning extent tree: %d\n",
+				ret);
+			goto out;
+		}
 	}
 
 	ret = map_implied_refs(info);
@@ -1507,6 +1515,7 @@ static void print_subvol_info(u64 subvolid, u64 bytenr, u64 num_bytes,
 
 int print_extent_state(struct btrfs_fs_info *info, u64 subvol)
 {
+	struct rb_node *n;
 	int ret;
 
 	tree_blocks = ulist_alloc(0);
@@ -1519,10 +1528,17 @@ int print_extent_state(struct btrfs_fs_info *info, u64 subvol)
 	/*
 	 * Put all extent refs into our rbtree
 	 */
-	ret = scan_extents(info, 0, ~0ULL);
-	if (ret) {
-		fprintf(stderr, "ERROR: while scanning extent tree: %d\n", ret);
-		goto out;
+	for (n = rb_first(&info->block_group_cache_tree); n; n = rb_next(n)) {
+		struct btrfs_block_group *bg;
+
+		bg = rb_entry(n, struct btrfs_block_group, cache_node);
+		ret = scan_extents(info, bg->start,
+				   bg->start + bg->length - 1);
+		if (ret) {
+			fprintf(stderr, "ERROR: while scanning extent tree: %d\n",
+				ret);
+			goto out;
+		}
 	}
 
 	ret = map_implied_refs(info);
Signature · 1 line
2.26.3
#9 lore raw

PATCH v5 9/19 btrfs-progs: check: make free space tree validation extent tree v2 aware

In reply to Josef Bacik

The free space tree needs to be validated against all referenced blocks in the file system, so use the btrfs_mark_used_blocks() helper to check the free space tree and free space cache against. This will do the right thing for both extent tree v1 and extent tree v2.

---
 check/main.c | 90 ++++++++++++++++++----------------------------------
 1 file changed, 31 insertions(+), 59 deletions(-)
check/main.c +31 -59
index a0f4ee91..9d090fdc 100644
@@ -5637,72 +5637,38 @@ static int check_cache_range(struct btrfs_root *root,
 }
 
 static int verify_space_cache(struct btrfs_root *root,
-			      struct btrfs_block_group *cache)
+			      struct btrfs_block_group *cache,
+			      struct extent_io_tree *used)
 {
-	struct btrfs_path path;
-	struct extent_buffer *leaf;
-	struct btrfs_key key;
-	u64 last;
+	u64 start, end, last_end, bg_end;
 	int ret = 0;
 
-	root = btrfs_extent_root(root->fs_info, cache->start);
+	start = cache->start;
+	bg_end = cache->start + cache->length;
+	last_end = start;
 
-	last = max_t(u64, cache->start, BTRFS_SUPER_INFO_OFFSET);
-
-	btrfs_init_path(&path);
-	key.objectid = last;
-	key.offset = 0;
-	key.type = BTRFS_EXTENT_ITEM_KEY;
-	ret = btrfs_search_slot(NULL, root, &key, &path, 0, 0);
-	if (ret < 0)
-		goto out;
-	ret = 0;
-	while (1) {
-		if (path.slots[0] >= btrfs_header_nritems(path.nodes[0])) {
-			ret = btrfs_next_leaf(root, &path);
-			if (ret < 0)
-				goto out;
-			if (ret > 0) {
-				ret = 0;
-				break;
-			}
-		}
-		leaf = path.nodes[0];
-		btrfs_item_key_to_cpu(leaf, &key, path.slots[0]);
-		if (key.objectid >= cache->start + cache->length)
+	while (start < bg_end) {
+		ret = find_first_extent_bit(used, cache->start, &start, &end,
+					    EXTENT_DIRTY);
+		if (ret || start >= bg_end) {
+			ret = 0;
 			break;
-		if (key.type != BTRFS_EXTENT_ITEM_KEY &&
-		    key.type != BTRFS_METADATA_ITEM_KEY) {
-			path.slots[0]++;
-			continue;
 		}
-
-		if (last == key.objectid) {
-			if (key.type == BTRFS_EXTENT_ITEM_KEY)
-				last = key.objectid + key.offset;
-			else
-				last = key.objectid + gfs_info->nodesize;
-			path.slots[0]++;
-			continue;
+		if (last_end < start) {
+			ret = check_cache_range(root, cache, last_end,
+						start - last_end);
+			if (ret)
+				return ret;
 		}
-
-		ret = check_cache_range(root, cache, last,
-					key.objectid - last);
-		if (ret)
-			break;
-		if (key.type == BTRFS_EXTENT_ITEM_KEY)
-			last = key.objectid + key.offset;
-		else
-			last = key.objectid + gfs_info->nodesize;
-		path.slots[0]++;
+		end = min(end, bg_end - 1);
+		clear_extent_dirty(used, start, end);
+		start = end + 1;
+		last_end = start;
 	}
 
-	if (last < cache->start + cache->length)
-		ret = check_cache_range(root, cache, last,
-					cache->start + cache->length - last);
-
-out:
-	btrfs_release_path(&path);
+	if (last_end < bg_end)
+		ret = check_cache_range(root, cache, last_end,
+					bg_end - last_end);
 
 	if (!ret &&
 	    !RB_EMPTY_ROOT(&cache->free_space_ctl->free_space_offset)) {
@@ -5716,11 +5682,17 @@ out:
 
 static int check_space_cache(struct btrfs_root *root)
 {
+	struct extent_io_tree used;
 	struct btrfs_block_group *cache;
 	u64 start = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
 	int ret;
 	int error = 0;
 
+	extent_io_tree_init(&used);
+	ret = btrfs_mark_used_blocks(gfs_info, &used);
+	if (ret)
+		return ret;
+
 	while (1) {
 		ctx.item_count++;
 		cache = btrfs_lookup_first_block_group(gfs_info, start);
@@ -5765,14 +5737,14 @@ static int check_space_cache(struct btrfs_root *root)
 				continue;
 		}
 
-		ret = verify_space_cache(root, cache);
+		ret = verify_space_cache(root, cache, &used);
 		if (ret) {
 			fprintf(stderr, "cache appears valid but isn't %llu\n",
 				cache->start);
 			error++;
 		}
 	}
-
+	extent_io_tree_cleanup(&used);
 	return error ? -EINVAL : 0;
 }
 
Signature · 1 line
2.26.3
#10 lore raw

PATCH v5 10/19 btrfs-progs: check: add helper to reinit the root based on a key

In reply to Josef Bacik

In the case of per-bg roots we may be missing the root items. To re-initialize them we want to add the root item as well as allocate the empty block. To achieve this extract out the reinit root logic to a helper that just takes the root key and then does the appropriate work to allocate an empty root and update the root item. Fix the normal reinit root helper to use this new helper.

---
 check/main.c | 88 ++++++++++++++++++++++++++++++++++------------------
 1 file changed, 58 insertions(+), 30 deletions(-)
check/main.c +58 -30
index 9d090fdc..5b700350 100644
@@ -9129,29 +9129,34 @@ static int do_check_chunks_and_extents(void)
 	return ret;
 }
 
-static int btrfs_fsck_reinit_root(struct btrfs_trans_handle *trans,
-				  struct btrfs_root *root)
+static struct extent_buffer *btrfs_fsck_clear_root(
+					struct btrfs_trans_handle *trans,
+					struct btrfs_key *key)
 {
+	struct btrfs_root_item ri = {};
+	struct btrfs_path *path;
 	struct extent_buffer *c;
-	struct extent_buffer *old = root->node;
-	int level;
+	struct btrfs_disk_key disk_key = {};
 	int ret;
-	struct btrfs_disk_key disk_key = {0,0,0};
 
-	level = 0;
+	path = btrfs_alloc_path();
+	if (!path)
+		return ERR_PTR(-ENOMEM);
 
-	c = btrfs_alloc_free_block(trans, root, gfs_info->nodesize,
-				   root->root_key.objectid,
-				   &disk_key, level, 0, 0);
-	if (IS_ERR(c))
-		return PTR_ERR(c);
+	c = btrfs_alloc_free_block(trans, gfs_info->tree_root,
+				   gfs_info->nodesize, key->objectid,
+				   &disk_key, 0, 0, 0);
+	if (IS_ERR(c)) {
+		btrfs_free_path(path);
+		return c;
+	}
 
 	memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
-	btrfs_set_header_level(c, level);
+	btrfs_set_header_level(c, 0);
 	btrfs_set_header_bytenr(c, c->start);
 	btrfs_set_header_generation(c, trans->transid);
 	btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
-	btrfs_set_header_owner(c, root->root_key.objectid);
+	btrfs_set_header_owner(c, key->objectid);
 
 	write_extent_buffer(c, gfs_info->fs_devices->metadata_uuid,
 			    btrfs_header_fsid(), BTRFS_FSID_SIZE);
@@ -9161,25 +9166,48 @@ static int btrfs_fsck_reinit_root(struct btrfs_trans_handle *trans,
 			    BTRFS_UUID_SIZE);
 
 	btrfs_mark_buffer_dirty(c);
+
 	/*
-	 * this case can happen in the following case:
-	 *
-	 * reinit reloc data root, this is because we skip pin
-	 * down reloc data tree before which means we can allocate
-	 * same block bytenr here.
+	 * The root item may not exist, try to insert an empty one so it exists,
+	 * otherwise simply update the existing one with the correct settings.
 	 */
-	if (old->start == c->start) {
-		btrfs_set_root_generation(&root->root_item,
-					  trans->transid);
-		root->root_item.level = btrfs_header_level(root->node);
-		ret = btrfs_update_root(trans, gfs_info->tree_root,
-					&root->root_key, &root->root_item);
-		if (ret) {
-			free_extent_buffer(c);
-			return ret;
-		}
-	}
-	free_extent_buffer(old);
+	ret = btrfs_insert_empty_item(trans, gfs_info->tree_root, path, key,
+				      sizeof(ri));
+	if (ret == -EEXIST) {
+		read_extent_buffer(path->nodes[0], &ri,
+				   btrfs_item_ptr_offset(path->nodes[0],
+							 path->slots[0]),
+				   sizeof(ri));
+	} else if (ret) {
+		btrfs_free_path(path);
+		free_extent_buffer(c);
+		return ERR_PTR(ret);
+	}
+	btrfs_set_root_bytenr(&ri, c->start);
+	btrfs_set_root_generation(&ri, trans->transid);
+	btrfs_set_root_refs(&ri, 1);
+	btrfs_set_root_used(&ri, c->len);
+	btrfs_set_root_generation_v2(&ri, trans->transid);
+
+	write_extent_buffer(path->nodes[0], &ri,
+			    btrfs_item_ptr_offset(path->nodes[0],
+						  path->slots[0]),
+			    sizeof(ri));
+	btrfs_mark_buffer_dirty(path->nodes[0]);
+	btrfs_free_path(path);
+	return c;
+}
+
+static int btrfs_fsck_reinit_root(struct btrfs_trans_handle *trans,
+				  struct btrfs_root *root)
+{
+	struct extent_buffer *c;
+
+	c = btrfs_fsck_clear_root(trans, &root->root_key);
+	if (IS_ERR(c))
+		return PTR_ERR(c);
+
+	free_extent_buffer(root->node);
 	root->node = c;
 	add_root_to_dirty_list(root);
 	return 0;
Signature · 1 line
2.26.3
#11 lore raw

PATCH v5 11/19 btrfs-progs: check: handle the block group tree properly

In reply to Josef Bacik

We need to make sure we process the block group root, and mark its blocks as used for the free space tree checking.

---
 check/main.c    | 27 +++++++++++++++++----------
 common/repair.c |  3 +++
 2 files changed, 20 insertions(+), 10 deletions(-)
check/main.c +17 -10
index 5b700350..45065989 100644
@@ -8947,6 +8947,18 @@ out:
 	return ret;
 }
 
+static int load_super_root(struct list_head *head, struct btrfs_root *root)
+{
+	u8 level;
+
+	if (!root)
+		return 0;
+
+	level = btrfs_header_level(root->node);
+	return add_root_item_to_list(head, root->root_key.objectid,
+				     root->node->start, 0, level, 0, NULL);
+}
+
 static int check_chunks_and_extents(void)
 {
 	struct rb_root dev_cache;
@@ -8965,9 +8977,7 @@ static int check_chunks_and_extents(void)
 	int bits_nr;
 	struct list_head dropping_trees;
 	struct list_head normal_trees;
-	struct btrfs_root *root1;
 	struct btrfs_root *root;
-	u8 level;
 
 	root = gfs_info->fs_root;
 	dev_cache = RB_ROOT;
@@ -9000,16 +9010,13 @@ static int check_chunks_and_extents(void)
 	}
 
 again:
-	root1 = gfs_info->tree_root;
-	level = btrfs_header_level(root1->node);
-	ret = add_root_item_to_list(&normal_trees, root1->root_key.objectid,
-				    root1->node->start, 0, level, 0, NULL);
+	ret = load_super_root(&normal_trees, gfs_info->tree_root);
+	if (ret < 0)
+		goto out;
+	ret = load_super_root(&normal_trees, gfs_info->chunk_root);
 	if (ret < 0)
 		goto out;
-	root1 = gfs_info->chunk_root;
-	level = btrfs_header_level(root1->node);
-	ret = add_root_item_to_list(&normal_trees, root1->root_key.objectid,
-				    root1->node->start, 0, level, 0, NULL);
+	ret = load_super_root(&normal_trees, gfs_info->block_group_root);
 	if (ret < 0)
 		goto out;
 
common/repair.c +3 -0
index a73949b0..37a6943f 100644
@@ -149,6 +149,9 @@ int btrfs_mark_used_tree_blocks(struct btrfs_fs_info *fs_info,
 	ret = traverse_tree_blocks(tree, fs_info->chunk_root->node, 0);
 	if (!ret)
 		ret = traverse_tree_blocks(tree, fs_info->tree_root->node, 1);
+	if (!ret && fs_info->block_group_root)
+		ret = traverse_tree_blocks(tree,
+					   fs_info->block_group_root->node, 0);
 	return ret;
 }
 
Signature · 1 line
2.26.3
#12 lore raw

PATCH v5 12/19 btrfs-progs: set the number of global roots in the super block

In reply to Josef Bacik

In order to make sure the file system is consistent we need to record the number of global roots we should have in the super block. We could infer this from the number of global roots we find, however this could lead to interesting fuzzing problems, so add a source of truth to the super block in order to make it easier to verify the file system is consistent.

---
 kernel-shared/ctree.h   | 6 +++++-
 kernel-shared/disk-io.c | 4 ++++
 mkfs/common.c           | 1 +
 3 files changed, 10 insertions(+), 1 deletion(-)
kernel-shared/ctree.h +5 -1
index b12dbff1..90de7a65 100644
@@ -463,13 +463,15 @@ struct btrfs_super_block {
 
 	u8 metadata_uuid[BTRFS_FSID_SIZE];
 
+	__le64 nr_global_roots;
+
 	__le64 block_group_root;
 	__le64 block_group_root_generation;
 	u8 block_group_root_level;
 
 	/* future expansion */
 	u8 reserved8[7];
-	__le64 reserved[25];
+	__le64 reserved[24];
 	u8 sys_chunk_array[BTRFS_SYSTEM_CHUNK_ARRAY_SIZE];
 	struct btrfs_root_backup super_roots[BTRFS_NUM_BACKUP_ROOTS];
 	/* Padded to 4096 bytes */
@@ -2372,6 +2374,8 @@ BTRFS_SETGET_STACK_FUNCS(super_block_group_root_generation,
 			 block_group_root_generation, 64);
 BTRFS_SETGET_STACK_FUNCS(super_block_group_root_level,
 			 struct btrfs_super_block, block_group_root_level, 8);
+BTRFS_SETGET_STACK_FUNCS(super_nr_global_roots, struct btrfs_super_block,
+			 nr_global_roots, 64);
 
 static inline unsigned long btrfs_leaf_data(struct extent_buffer *l)
 {
kernel-shared/disk-io.c +4 -0
index 3d1157ad..fcef6e97 100644
@@ -1620,6 +1620,10 @@ static struct btrfs_fs_info *__open_ctree_fd(int fp, struct open_ctree_flags *oc
 	if (ret)
 		goto out_devices;
 
+	if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
+		fs_info->nr_global_roots =
+			btrfs_super_nr_global_roots(fs_info->super_copy);
+
 	/*
 	 * fs_info->zone_size (and zoned) are not known before reading the
 	 * chunk tree, so it's 0 at this point. But, fs_info->zoned == 0
mkfs/common.c +1 -0
index aa65543b..eac8c46c 100644
@@ -355,6 +355,7 @@ int make_btrfs(int fd, struct btrfs_mkfs_config *cfg)
 		btrfs_set_super_cache_generation(&super, 0);
 	}
 	if (extent_tree_v2) {
+		btrfs_set_super_nr_global_roots(&super, 1);
 		btrfs_set_super_block_group_root(&super,
 						 cfg->blocks[MKFS_BLOCK_GROUP_TREE]);
 		btrfs_set_super_block_group_root_generation(&super, 1);
Signature · 1 line
2.26.3
#13 lore raw

Re: [PATCH v5 12/19] btrfs-progs: set the number of global roots in the super block

Quoted 19 lines · Josef Bacik

In order to make sure the file system is consistent we need to record the number of global roots we should have in the super block. We could infer this from the number of global roots we find, however this could lead to interesting fuzzing problems, so add a source of truth to the super block in order to make it easier to verify the file system is consistent.

---
 kernel-shared/ctree.h   | 6 +++++-
 kernel-shared/disk-io.c | 4 ++++
 mkfs/common.c           | 1 +
 3 files changed, 10 insertions(+), 1 deletion(-)
kernel-shared/ctree.h +0 -0
index b12dbff1..90de7a65 100644

kernel-shared/ctree.h struct btrfs_super_block { · line 467

Quoted 5 lines
kernel-shared/ctree.h +2 -0
@@ -463,13 +463,15 @@ struct btrfs_super_block {
 
 	u8 metadata_uuid[BTRFS_FSID_SIZE];
 
+	__le64 nr_global_roots;
+

Shouldn't this be added after the last item?

Quoted 11 lines
kernel-shared/ctree.h +1 -1
 	__le64 block_group_root;
 	__le64 block_group_root_generation;
 	u8 block_group_root_level;
 
 	/* future expansion */
 	u8 reserved8[7];
-	__le64 reserved[25];
+	__le64 reserved[24];
 	u8 sys_chunk_array[BTRFS_SYSTEM_CHUNK_ARRAY_SIZE];
 	struct btrfs_root_backup super_roots[BTRFS_NUM_BACKUP_ROOTS];
 	/* Padded to 4096 bytes */
#14 lore raw

Re: [PATCH v5 12/19] btrfs-progs: set the number of global roots in the super block

Quoted 7 lines · David Sterba

kernel-shared/ctree.h struct btrfs_super_block { · line 467

Quoted 4 lines
kernel-shared/ctree.h +2 -0
 	u8 metadata_uuid[BTRFS_FSID_SIZE];
 
+	__le64 nr_global_roots;
+

Shouldn't this be added after the last item?

Quoted 8 lines

kernel-shared/ctree.h struct btrfs_super_block { · line 474

Quoted 8 lines
kernel-shared/ctree.h +1 -1
 	__le64 block_group_root;
 	__le64 block_group_root_generation;
 	u8 block_group_root_level;
 
 	/* future expansion */
 	u8 reserved8[7];
-	__le64 reserved[25];
+	__le64 reserved[24];

Or at least inside one of these reserved fields.

#15 lore raw

Re: [PATCH v5 12/19] btrfs-progs: set the number of global roots in the super block

Quoted 12 lines · Johannes Thumshirn
Quoted 11 lines · David Sterba

kernel-shared/ctree.h struct btrfs_super_block { · line 467

Quoted 4 lines
kernel-shared/ctree.h +2 -0
 	u8 metadata_uuid[BTRFS_FSID_SIZE];
 
+	__le64 nr_global_roots;
+

Shouldn't this be added after the last item?

Quoted 4 lines
kernel-shared/ctree.h +0 -0
 	__le64 block_group_root;
 	__le64 block_group_root_generation;
 	u8 block_group_root_level;
Quoted 6 lines
Quoted 4 lines

kernel-shared/ctree.h struct btrfs_super_block { · line 474

Quoted 4 lines
kernel-shared/ctree.h +1 -1
 	/* future expansion */
 	u8 reserved8[7];
-	__le64 reserved[25];
+	__le64 reserved[24];

Or at least inside one of these reserved fields.

OTOH, it's still experimental so we don't expect backward compatibility yet so it should be ok to change for now.

#16 lore raw

Re: [PATCH v5 12/19] btrfs-progs: set the number of global roots in the super block

Quoted 20 lines · David Sterba
Quoted 12 lines · Johannes Thumshirn
Quoted 11 lines · David Sterba

kernel-shared/ctree.h struct btrfs_super_block { · line 467

Quoted 4 lines
kernel-shared/ctree.h +2 -0
 	u8 metadata_uuid[BTRFS_FSID_SIZE];
 
+	__le64 nr_global_roots;
+

Shouldn't this be added after the last item?

Quoted 4 lines
kernel-shared/ctree.h +0 -0
 	__le64 block_group_root;
 	__le64 block_group_root_generation;
 	u8 block_group_root_level;
Quoted 6 lines
Quoted 4 lines

kernel-shared/ctree.h struct btrfs_super_block { · line 474

Quoted 4 lines
kernel-shared/ctree.h +1 -1
 	/* future expansion */
 	u8 reserved8[7];
-	__le64 reserved[25];
+	__le64 reserved[24];

Or at least inside one of these reserved fields.

Quoted 2 lines

OTOH, it's still experimental so we don't expect backward compatibility yet so it should be ok to change for now.

I did it this way because it's all still experimental and it makes more sense for it to be before the new root stuff. Thanks,

Josef

#17 lore raw

Re: [PATCH v5 12/19] btrfs-progs: set the number of global roots in the super block

Quoted 21 lines · Josef Bacik
Quoted 20 lines · David Sterba
Quoted 18 lines · Johannes Thumshirn
Quoted 7 lines · David Sterba

kernel-shared/ctree.h struct btrfs_super_block { · line 467

Quoted 4 lines
  	u8 metadata_uuid[BTRFS_FSID_SIZE];
kernel-shared/ctree.h +2 -0
+	__le64 nr_global_roots;
+

Shouldn't this be added after the last item?

Quoted 8 lines

kernel-shared/ctree.h struct btrfs_super_block { · line 474

Quoted 8 lines

__le64 block_group_root;
__le64 block_group_root_generation;
u8 block_group_root_level;

  	/* future expansion */
  	u8 reserved8[7];
kernel-shared/ctree.h +1 -1
-	__le64 reserved[25];
+	__le64 reserved[24];

Or at least inside one of these reserved fields.

Quoted 5 lines
Quoted 2 lines

OTOH, it's still experimental so we don't expect backward compatibility yet so it should be ok to change for now.

I did it this way because it's all still experimental and it makes more sense for it to be before the new root stuff. Thanks,

I'd say, please don't.

It's making anyone who want to add a new member in super block miserable.

Everyone is going to add the new member from the reserved members, but such insert into the existing members are destructive.

Furthermore, if the new member is going to be merged way before extent tree v2 part, how do we solve the conflicts?

(The new member I want to introduce is just to indicate how many bytes we have reserved at the beginning of each device, with a new RO compat flag).

Thanks,
Qu

Quoted 2 lines

Josef

#18 lore raw

Re: [PATCH v5 12/19] btrfs-progs: set the number of global roots in the super block

Quoted 38 lines · Qu Wenruo
Quoted 19 lines · Josef Bacik
Quoted 18 lines · David Sterba
Quoted 17 lines · Johannes Thumshirn
Quoted 15 lines · David Sterba
Quoted 4 lines

      u8 metadata_uuid[BTRFS_FSID_SIZE];

+    __le64 nr_global_roots;
+

Shouldn't this be added after the last item?

Quoted 8 lines

      __le64 block_group_root;
      __le64 block_group_root_generation;
      u8 block_group_root_level;

      /* future expansion */
      u8 reserved8[7];
-    __le64 reserved[25];
+    __le64 reserved[24];

Quoted 8 lines
Quoted 4 lines
Quoted 1 line

Or at least inside one of these reserved fields.

OTOH, it's still experimental so we don't expect backward compatibility yet so it should be ok to change for now.

I did it this way because it's all still experimental and it makes more sense
for it to be before the new root stuff.  Thanks,

I'd say, please don't.

It's making anyone who want to add a new member in super block miserable.

Everyone is going to add the new member from the reserved members, but such insert into the existing members are destructive.

Quoted 6 lines

Furthermore, if the new member is going to be merged way before extent tree v2 part, how do we solve the conflicts?

(The new member I want to introduce is just to indicate how many bytes we have reserved at the beginning of each device, with a new RO compat flag).

My bad, the main problem is not shuffling the members of extent tree v2, but out-of-sync between kernel and btrfs-progs for super block.

Anyway, I'd use the padding[] for my new members, to avoid possible out-of-sync problems.

Thanks,
Qu

Quoted 5 lines

Thanks,
Qu

Quoted 2 lines

Josef

#19 lore raw

PATCH v5 13/19 btrfs-progs: handle the per-block group global root id

In reply to Josef Bacik

We will now be using block_group->chunk_objectid to point at the global root id for this particular block group. For now we'll assign this based on mod'ing the offset of the block group against the number of global root id's and handle the block_group_item updating appropriately.

---
 kernel-shared/ctree.h           |  2 ++
 kernel-shared/disk-io.c         | 24 ++++++++++++++++++++++--
 kernel-shared/disk-io.h         |  1 +
 kernel-shared/extent-tree.c     | 24 ++++++++++++++++++++++--
 kernel-shared/free-space-tree.c |  3 +++
 5 files changed, 50 insertions(+), 4 deletions(-)
kernel-shared/ctree.h +2 -0
index 90de7a65..d79f49c9 100644
@@ -1190,6 +1190,8 @@ struct btrfs_block_group {
 	 */
 	u64 alloc_offset;
 	u64 write_offset;
+
+	u64 global_root_id;
 };
 
 struct btrfs_device;
kernel-shared/disk-io.c +22 -2
index fcef6e97..59c46946 100644
@@ -806,13 +806,33 @@ struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
 	return NULL;
 }
 
+u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr)
+{
+	struct btrfs_block_group *block_group;
+	u64 ret = 0;
+
+	if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
+		return ret;
+
+	/*
+	 * We use this because we won't have this many global roots, and -1 is
+	 * special, so we need something that'll not be found if we have any
+	 * errors from here on.
+	 */
+	ret = BTRFS_LAST_FREE_OBJECTID;
+	block_group = btrfs_lookup_first_block_group(fs_info, bytenr);
+	if (block_group)
+		ret = block_group->global_root_id;
+	return ret;
+}
+
 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info,
 				   u64 bytenr)
 {
 	struct btrfs_key key = {
 		.objectid = BTRFS_CSUM_TREE_OBJECTID,
 		.type = BTRFS_ROOT_ITEM_KEY,
-		.offset = 0,
+		.offset = btrfs_global_root_id(fs_info, bytenr),
 	};
 
 	return btrfs_global_root(fs_info, &key);
@@ -824,7 +844,7 @@ struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_info,
 	struct btrfs_key key = {
 		.objectid = BTRFS_EXTENT_TREE_OBJECTID,
 		.type = BTRFS_ROOT_ITEM_KEY,
-		.offset = 0,
+		.offset = btrfs_global_root_id(fs_info, bytenr),
 	};
 
 	return btrfs_global_root(fs_info, &key);
kernel-shared/disk-io.h +1 -0
index 1e9044f8..81d8670f 100644
@@ -225,6 +225,7 @@ struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr);
 struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_inf, u64 bytenr);
 struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
 				     struct btrfs_key *key);
+u64 btrfs_global_root_id(struct btrfs_fs_info *fs_info, u64 bytenr);
 int btrfs_global_root_insert(struct btrfs_fs_info *fs_info,
 			     struct btrfs_root *root);
 
kernel-shared/extent-tree.c +22 -2
index b2b99d4f..697a8a1e 100644
@@ -1561,7 +1561,7 @@ static int update_block_group_item(struct btrfs_trans_handle *trans,
 	btrfs_set_stack_block_group_used(&bgi, cache->used);
 	btrfs_set_stack_block_group_flags(&bgi, cache->flags);
 	btrfs_set_stack_block_group_chunk_objectid(&bgi,
-			BTRFS_FIRST_CHUNK_TREE_OBJECTID);
+						   cache->global_root_id);
 	write_extent_buffer(leaf, &bgi, bi, sizeof(bgi));
 	btrfs_mark_buffer_dirty(leaf);
 fail:
@@ -2658,6 +2658,7 @@ static int read_block_group_item(struct btrfs_block_group *cache,
 			   sizeof(bgi));
 	cache->used = btrfs_stack_block_group_used(&bgi);
 	cache->flags = btrfs_stack_block_group_flags(&bgi);
+	cache->global_root_id = btrfs_stack_block_group_chunk_objectid(&bgi);
 
 	return 0;
 }
@@ -2765,6 +2766,24 @@ error:
 	return ret;
 }
 
+/*
+ * For extent tree v2 we use the block_group_item->chunk_offset to point at our
+ * global root id.  For v1 it's always set to BTRFS_FIRST_CHUNK_TREE_OBJECTID.
+ */
+static u64 calculate_global_root_id(struct btrfs_fs_info *fs_info, u64 offset)
+{
+	u64 div = SZ_1G;
+
+	if (!btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
+		return BTRFS_FIRST_CHUNK_TREE_OBJECTID;
+
+	/* If we have a smaller fs index based on 128m. */
+	if (btrfs_super_total_bytes(fs_info->super_copy) <= (SZ_1G * 10ULL))
+		div = SZ_128M;
+
+	return (div_u64(offset, div) % fs_info->nr_global_roots);
+}
+
 struct btrfs_block_group *
 btrfs_add_block_group(struct btrfs_fs_info *fs_info, u64 bytes_used, u64 type,
 		      u64 chunk_offset, u64 size)
@@ -2776,6 +2795,7 @@ btrfs_add_block_group(struct btrfs_fs_info *fs_info, u64 bytes_used, u64 type,
 	BUG_ON(!cache);
 	cache->start = chunk_offset;
 	cache->length = size;
+	cache->global_root_id = calculate_global_root_id(fs_info, chunk_offset);
 
 	ret = btrfs_load_block_group_zone_info(fs_info, cache);
 	BUG_ON(ret);
@@ -2806,7 +2826,7 @@ static int insert_block_group_item(struct btrfs_trans_handle *trans,
 
 	btrfs_set_stack_block_group_used(&bgi, block_group->used);
 	btrfs_set_stack_block_group_chunk_objectid(&bgi,
-				BTRFS_FIRST_CHUNK_TREE_OBJECTID);
+						   block_group->global_root_id);
 	btrfs_set_stack_block_group_flags(&bgi, block_group->flags);
 	key.objectid = block_group->start;
 	key.type = BTRFS_BLOCK_GROUP_ITEM_KEY;
kernel-shared/free-space-tree.c +3 -0
index 896bd3a2..a82865d3 100644
@@ -34,6 +34,9 @@ static struct btrfs_root *btrfs_free_space_root(struct btrfs_fs_info *fs_info,
 		.offset = 0,
 	};
 
+	if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2))
+		key.offset = block_group->global_root_id;
+
 	return btrfs_global_root(fs_info, &key);
 }
 
Signature · 1 line
2.26.3
#20 lore raw

PATCH v5 14/19 btrfs-progs: add a btrfs_delete_and_free_root helper

In reply to Josef Bacik

The free space tree code already does this, but we need it for cleaning up per block group roots. Abstract this code out into a helper so that we can use it in multiple places in the future.

---
 kernel-shared/disk-io.c         | 25 +++++++++++++++++++++++++
 kernel-shared/disk-io.h         |  2 ++
 kernel-shared/free-space-tree.c | 24 +++---------------------
 3 files changed, 30 insertions(+), 21 deletions(-)
kernel-shared/disk-io.c +25 -0
index 59c46946..f3ddf9e3 100644
@@ -2399,6 +2399,31 @@ int btrfs_set_buffer_uptodate(struct extent_buffer *eb)
 	return set_extent_buffer_uptodate(eb);
 }
 
+int btrfs_delete_and_free_root(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *root)
+{
+	struct btrfs_fs_info *fs_info = root->fs_info;
+	struct btrfs_root *tree_root = fs_info->tree_root;
+	int ret;
+
+	ret = btrfs_del_root(trans, tree_root, &root->root_key);
+	if (ret)
+		return ret;
+
+	list_del(&root->dirty_list);
+	ret = clean_tree_block(root->node);
+	if (ret)
+		return ret;
+	ret = btrfs_free_tree_block(trans, root, root->node, 0, 1);
+	if (ret)
+		return ret;
+	rb_erase(&root->rb_node, &fs_info->global_roots_tree);
+	free_extent_buffer(root->node);
+	free_extent_buffer(root->commit_root);
+	kfree(root);
+	return 0;
+}
+
 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
 				     struct btrfs_fs_info *fs_info,
 				     u64 objectid)
kernel-shared/disk-io.h +2 -0
index 81d8670f..1e97b9ac 100644
@@ -221,6 +221,8 @@ int btrfs_fs_roots_compare_roots(struct rb_node *node1, struct rb_node *node2);
 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
 				     struct btrfs_fs_info *fs_info,
 				     u64 objectid);
+int btrfs_delete_and_free_root(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *root);
 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr);
 struct btrfs_root *btrfs_extent_root(struct btrfs_fs_info *fs_inf, u64 bytenr);
 struct btrfs_root *btrfs_global_root(struct btrfs_fs_info *fs_info,
kernel-shared/free-space-tree.c +3 -21
index a82865d3..0a13b1d6 100644
@@ -1257,27 +1257,9 @@ int btrfs_clear_free_space_tree(struct btrfs_fs_info *fs_info)
 	if (ret)
 		goto abort;
 
-	ret = btrfs_del_root(trans, tree_root, &free_space_root->root_key);
-	if (ret)
-		goto abort;
-
-	list_del(&free_space_root->dirty_list);
-
-	ret = clean_tree_block(free_space_root->node);
-	if (ret)
-		goto abort;
-	ret = btrfs_free_tree_block(trans, free_space_root,
-				    free_space_root->node, 0, 1);
-	if (ret)
-		goto abort;
-
-	rb_erase(&free_space_root->rb_node, &fs_info->global_roots_tree);
-	free_extent_buffer(free_space_root->node);
-	free_extent_buffer(free_space_root->commit_root);
-	kfree(free_space_root);
-
-	ret = btrfs_commit_transaction(trans, tree_root);
-
+	ret = btrfs_delete_and_free_root(trans, free_space_root);
+	if (!ret)
+		ret = btrfs_commit_transaction(trans, tree_root);
 abort:
 	return ret;
 }
Signature · 1 line
2.26.3
#21 lore raw

PATCH v5 15/19 btrfs-progs: make btrfs_clear_free_space_tree extent tree v2 aware

In reply to Josef Bacik

With extent tree v2 we'll have multiple free space trees, and we can't just unset the feature flags for the free space tree. Fix this to loop through all of the free space trees and clear them out properly.

---
 kernel-shared/free-space-tree.c | 37 ++++++++++++++++++++++++---------
 1 file changed, 27 insertions(+), 10 deletions(-)
kernel-shared/free-space-tree.c +27 -10
index 0a13b1d6..7ac75c20 100644
@@ -1248,18 +1248,35 @@ int btrfs_clear_free_space_tree(struct btrfs_fs_info *fs_info)
 	if (IS_ERR(trans))
 		return PTR_ERR(trans);
 
-	features = btrfs_super_compat_ro_flags(fs_info->super_copy);
-	features &= ~(BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE_VALID |
-		      BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE);
-	btrfs_set_super_compat_ro_flags(fs_info->super_copy, features);
+	if (btrfs_fs_incompat(fs_info, EXTENT_TREE_V2)) {
+		struct btrfs_key key = {
+			.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID,
+			.type = BTRFS_ROOT_ITEM_KEY,
+			.offset = 0,
+		};
+
+		while (key.offset < fs_info->nr_global_roots) {
+			free_space_root = btrfs_global_root(fs_info, &key);
+			ret = clear_free_space_tree(trans, free_space_root);
+			if (ret)
+				goto abort;
+			key.offset++;
+		}
+	} else {
+		features = btrfs_super_compat_ro_flags(fs_info->super_copy);
+		features &= ~(BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE_VALID |
+			      BTRFS_FEATURE_COMPAT_RO_FREE_SPACE_TREE);
+		btrfs_set_super_compat_ro_flags(fs_info->super_copy, features);
 
-	ret = clear_free_space_tree(trans, free_space_root);
-	if (ret)
-		goto abort;
+		ret = clear_free_space_tree(trans, free_space_root);
+		if (ret)
+			goto abort;
 
-	ret = btrfs_delete_and_free_root(trans, free_space_root);
-	if (!ret)
-		ret = btrfs_commit_transaction(trans, tree_root);
+		ret = btrfs_delete_and_free_root(trans, free_space_root);
+		if (ret)
+			goto abort;
+	}
+	ret = btrfs_commit_transaction(trans, tree_root);
 abort:
 	return ret;
 }
Signature · 1 line
2.26.3
#22 lore raw

PATCH v5 16/19 btrfs-progs: make btrfs_create_tree take a key for the root key

In reply to Josef Bacik

We're going to start create global roots from mkfs, and we need to have a offset set for the root key. Make the btrfs_create_tree() take a key for the root_key instead of just the objectid so we can setup these new style roots properly.

---
 kernel-shared/disk-io.c         | 21 ++++++++-------------
 kernel-shared/disk-io.h         |  2 +-
 kernel-shared/free-space-tree.c |  7 +++++--
 mkfs/main.c                     | 13 ++++++++++---
 4 files changed, 24 insertions(+), 19 deletions(-)
kernel-shared/disk-io.c +8 -13
index f3ddf9e3..4964cd38 100644
@@ -2426,25 +2426,22 @@ int btrfs_delete_and_free_root(struct btrfs_trans_handle *trans,
 
 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
 				     struct btrfs_fs_info *fs_info,
-				     u64 objectid)
+				     struct btrfs_key *key)
 {
 	struct extent_buffer *leaf;
 	struct btrfs_root *tree_root = fs_info->tree_root;
 	struct btrfs_root *root;
-	struct btrfs_key key;
 	int ret = 0;
 
 	root = kzalloc(sizeof(*root), GFP_KERNEL);
 	if (!root)
 		return ERR_PTR(-ENOMEM);
 
-	btrfs_setup_root(root, fs_info, objectid);
-	root->root_key.objectid = objectid;
-	root->root_key.type = BTRFS_ROOT_ITEM_KEY;
-	root->root_key.offset = 0;
+	btrfs_setup_root(root, fs_info, key->objectid);
+	memcpy(&root->root_key, key, sizeof(struct btrfs_key));
 
-	leaf = btrfs_alloc_free_block(trans, root, fs_info->nodesize, objectid,
-			NULL, 0, 0, 0);
+	leaf = btrfs_alloc_free_block(trans, root, fs_info->nodesize,
+				      root->root_key.objectid, NULL, 0, 0, 0);
 	if (IS_ERR(leaf)) {
 		ret = PTR_ERR(leaf);
 		leaf = NULL;
@@ -2455,7 +2452,7 @@ struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
 	btrfs_set_header_bytenr(leaf, leaf->start);
 	btrfs_set_header_generation(leaf, trans->transid);
 	btrfs_set_header_backref_rev(leaf, BTRFS_MIXED_BACKREF_REV);
-	btrfs_set_header_owner(leaf, objectid);
+	btrfs_set_header_owner(leaf, root->root_key.objectid);
 	root->node = leaf;
 	write_extent_buffer(leaf, fs_info->fs_devices->metadata_uuid,
 			    btrfs_header_fsid(), BTRFS_FSID_SIZE);
@@ -2480,10 +2477,8 @@ struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
 	memset(root->root_item.uuid, 0, BTRFS_UUID_SIZE);
 	root->root_item.drop_level = 0;
 
-	key.objectid = objectid;
-	key.type = BTRFS_ROOT_ITEM_KEY;
-	key.offset = 0;
-	ret = btrfs_insert_root(trans, tree_root, &key, &root->root_item);
+	ret = btrfs_insert_root(trans, tree_root, &root->root_key,
+				&root->root_item);
 	if (ret)
 		goto fail;
 
kernel-shared/disk-io.h +1 -1
index 1e97b9ac..e07141a9 100644
@@ -220,7 +220,7 @@ int write_and_map_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb);
 int btrfs_fs_roots_compare_roots(struct rb_node *node1, struct rb_node *node2);
 struct btrfs_root *btrfs_create_tree(struct btrfs_trans_handle *trans,
 				     struct btrfs_fs_info *fs_info,
-				     u64 objectid);
+				     struct btrfs_key *key);
 int btrfs_delete_and_free_root(struct btrfs_trans_handle *trans,
 			       struct btrfs_root *root);
 struct btrfs_root *btrfs_csum_root(struct btrfs_fs_info *fs_info, u64 bytenr);
kernel-shared/free-space-tree.c +5 -2
index 7ac75c20..03eb0ed2 100644
@@ -1475,14 +1475,17 @@ int btrfs_create_free_space_tree(struct btrfs_fs_info *fs_info)
 	struct btrfs_root *free_space_root;
 	struct btrfs_block_group *block_group;
 	u64 start = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
+	struct btrfs_key root_key = {
+		.objectid = BTRFS_FREE_SPACE_TREE_OBJECTID,
+		.type = BTRFS_ROOT_ITEM_KEY,
+	};
 	int ret;
 
 	trans = btrfs_start_transaction(tree_root, 0);
 	if (IS_ERR(trans))
 		return PTR_ERR(trans);
 
-	free_space_root = btrfs_create_tree(trans, fs_info,
-					    BTRFS_FREE_SPACE_TREE_OBJECTID);
+	free_space_root = btrfs_create_tree(trans, fs_info, &root_key);
 	if (IS_ERR(free_space_root)) {
 		ret = PTR_ERR(free_space_root);
 		goto abort;
mkfs/main.c +10 -3
index 7f79ba1a..19535604 100644
@@ -717,12 +717,15 @@ static int create_data_reloc_tree(struct btrfs_trans_handle *trans)
 	struct btrfs_inode_item *inode;
 	struct btrfs_root *root;
 	struct btrfs_path path;
-	struct btrfs_key key;
+	struct btrfs_key key = {
+		.objectid = BTRFS_DATA_RELOC_TREE_OBJECTID,
+		.type = BTRFS_ROOT_ITEM_KEY,
+	};
 	u64 ino = BTRFS_FIRST_FREE_OBJECTID;
 	char *name = "..";
 	int ret;
 
-	root = btrfs_create_tree(trans, fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID);
+	root = btrfs_create_tree(trans, fs_info, &key);
 	if (IS_ERR(root)) {
 		ret = PTR_ERR(root);
 		goto out;
@@ -782,10 +785,14 @@ static int create_uuid_tree(struct btrfs_trans_handle *trans)
 {
 	struct btrfs_fs_info *fs_info = trans->fs_info;
 	struct btrfs_root *root;
+	struct btrfs_key key = {
+		.objectid = BTRFS_UUID_TREE_OBJECTID,
+		.type = BTRFS_ROOT_ITEM_KEY,
+	};
 	int ret = 0;
 
 	ASSERT(fs_info->uuid_root == NULL);
-	root = btrfs_create_tree(trans, fs_info, BTRFS_UUID_TREE_OBJECTID);
+	root = btrfs_create_tree(trans, fs_info, &key);
 	if (IS_ERR(root)) {
 		ret = PTR_ERR(root);
 		goto out;
Signature · 1 line
2.26.3
#23 lore raw

PATCH v5 17/19 btrfs-progs: mkfs: set chunk_item_objectid properly for extent tree v2

In reply to Josef Bacik

Our initial block group will use global root id 0 with extent tree v2, so adjust the helper to take the chunk_objectid as an argument, as we'll set this to 0 for extent tree v2 and then BTRFS_FIRST_CHUNK_TREE_OBJECTID for extent tree v1.

---
 mkfs/common.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)
mkfs/common.c +4 -4
index eac8c46c..75680d03 100644
@@ -191,7 +191,7 @@ static int create_free_space_tree(int fd, struct btrfs_mkfs_config *cfg,
 
 static void write_block_group_item(struct extent_buffer *buf, u32 nr,
 				   u64 objectid, u64 offset, u64 used,
-				   u32 itemoff)
+				   u64 chunk_objectid, u32 itemoff)
 {
 	struct btrfs_block_group_item *bg_item;
 	struct btrfs_disk_key disk_key;
@@ -206,8 +206,7 @@ static void write_block_group_item(struct extent_buffer *buf, u32 nr,
 	bg_item = btrfs_item_ptr(buf, nr, struct btrfs_block_group_item);
 	btrfs_set_block_group_used(buf, bg_item, used);
 	btrfs_set_block_group_flags(buf, bg_item, BTRFS_BLOCK_GROUP_SYSTEM);
-	btrfs_set_block_group_chunk_objectid(buf, bg_item,
-					     BTRFS_FIRST_CHUNK_TREE_OBJECTID);
+	btrfs_set_block_group_chunk_objectid(buf, bg_item, chunk_objectid);
 }
 
 static int create_block_group_tree(int fd, struct btrfs_mkfs_config *cfg,
@@ -218,7 +217,7 @@ static int create_block_group_tree(int fd, struct btrfs_mkfs_config *cfg,
 
 	memset(buf->data + sizeof(struct btrfs_header), 0,
 		cfg->nodesize - sizeof(struct btrfs_header));
-	write_block_group_item(buf, 0, bg_offset, bg_size, bg_used,
+	write_block_group_item(buf, 0, bg_offset, bg_size, bg_used, 0,
 			       cfg->leaf_data_size -
 			       sizeof(struct btrfs_block_group_item));
 	btrfs_set_header_bytenr(buf, cfg->blocks[MKFS_BLOCK_GROUP_TREE]);
@@ -396,6 +395,7 @@ int make_btrfs(int fd, struct btrfs_mkfs_config *cfg)
 			write_block_group_item(buf, nritems,
 					       system_group_offset,
 					       system_group_size, total_used,
+					       BTRFS_FIRST_CHUNK_TREE_OBJECTID,
 					       itemoff);
 			add_block_group = false;
 			nritems++;
Signature · 1 line
2.26.3
#24 lore raw

PATCH v5 18/19 btrfs-progs: mkfs: create the global root's

In reply to Josef Bacik

Now that we have all of the supporting code, add the ability to create all of the global roots for an extent tree v2 fs. This will default to nr_cpu's, but also allow the user to specify how many global roots they would like.

---
 mkfs/main.c | 73 ++++++++++++++++++++++++++++++++++++++++++++++++++++-
 1 file changed, 72 insertions(+), 1 deletion(-)
mkfs/main.c +72 -1
index 19535604..a603ec58 100644
@@ -810,6 +810,53 @@ out:
 	return ret;
 }
 
+static int create_global_root(struct btrfs_trans_handle *trans, u64 objectid,
+			      int root_id)
+{
+	struct btrfs_fs_info *fs_info = trans->fs_info;
+	struct btrfs_root *root;
+	struct btrfs_key key = {
+		.objectid = objectid,
+		.type = BTRFS_ROOT_ITEM_KEY,
+		.offset = root_id,
+	};
+	int ret = 0;
+
+	root = btrfs_create_tree(trans, fs_info, &key);
+	if (IS_ERR(root)) {
+		ret = PTR_ERR(root);
+		goto out;
+	}
+	ret = btrfs_global_root_insert(fs_info, root);
+out:
+	if (ret)
+		btrfs_abort_transaction(trans, ret);
+	return ret;
+}
+
+static int create_global_roots(struct btrfs_trans_handle *trans,
+			       int nr_global_roots)
+{
+	int ret, i;
+
+	for (i = 1; i < nr_global_roots; i++) {
+		ret = create_global_root(trans, BTRFS_EXTENT_TREE_OBJECTID, i);
+		if (ret)
+			return ret;
+		ret = create_global_root(trans, BTRFS_CSUM_TREE_OBJECTID, i);
+		if (ret)
+			return ret;
+		ret = create_global_root(trans, BTRFS_FREE_SPACE_TREE_OBJECTID, i);
+		if (ret)
+			return ret;
+	}
+
+	btrfs_set_super_nr_global_roots(trans->fs_info->super_copy,
+					nr_global_roots);
+
+	return 0;
+}
+
 static int insert_qgroup_items(struct btrfs_trans_handle *trans,
 			       struct btrfs_fs_info *fs_info,
 			       u64 qgroupid)
@@ -966,13 +1013,18 @@ int BOX_MAIN(mkfs)(int argc, char **argv)
 	struct btrfs_mkfs_config mkfs_cfg;
 	enum btrfs_csum_type csum_type = BTRFS_CSUM_TYPE_CRC32;
 	u64 system_group_size;
+	int nr_global_roots = sysconf(_SC_NPROCESSORS_ONLN);
 
 	crc32c_optimization_init();
 	btrfs_config_init();
 
 	while(1) {
 		int c;
-		enum { GETOPT_VAL_SHRINK = 257, GETOPT_VAL_CHECKSUM };
+		enum {
+			GETOPT_VAL_SHRINK = 257,
+			GETOPT_VAL_CHECKSUM,
+			GETOPT_VAL_GLOBAL_ROOTS,
+		};
 		static const struct option long_options[] = {
 			{ "byte-count", required_argument, NULL, 'b' },
 			{ "csum", required_argument, NULL,
@@ -996,6 +1048,9 @@ int BOX_MAIN(mkfs)(int argc, char **argv)
 			{ "quiet", 0, NULL, 'q' },
 			{ "verbose", 0, NULL, 'v' },
 			{ "shrink", no_argument, NULL, GETOPT_VAL_SHRINK },
+#if EXPERIMENTAL
+			{ "num-global-roots", required_argument, NULL, GETOPT_VAL_GLOBAL_ROOTS },
+#endif
 			{ "help", no_argument, NULL, GETOPT_VAL_HELP },
 			{ NULL, 0, NULL, 0}
 		};
@@ -1100,6 +1155,9 @@ int BOX_MAIN(mkfs)(int argc, char **argv)
 			case GETOPT_VAL_CHECKSUM:
 				csum_type = parse_csum_type(optarg);
 				break;
+			case GETOPT_VAL_GLOBAL_ROOTS:
+				nr_global_roots = (int)arg_strtou64(optarg);
+				break;
 			case GETOPT_VAL_HELP:
 			default:
 				print_usage(c != GETOPT_VAL_HELP);
@@ -1239,6 +1297,11 @@ int BOX_MAIN(mkfs)(int argc, char **argv)
 	if (features & BTRFS_FEATURE_INCOMPAT_EXTENT_TREE_V2) {
 		features |= BTRFS_FEATURE_INCOMPAT_NO_HOLES;
 		runtime_features |= BTRFS_RUNTIME_FEATURE_FREE_SPACE_TREE;
+
+		if (!nr_global_roots) {
+			error("you must set a non-zero num-global-roots value");
+			exit(1);
+		}
 	}
 
 	if (zoned) {
@@ -1467,6 +1530,14 @@ int BOX_MAIN(mkfs)(int argc, char **argv)
 		goto error;
 	}
 
+	if (features & BTRFS_FEATURE_INCOMPAT_EXTENT_TREE_V2) {
+		ret = create_global_roots(trans, nr_global_roots);
+		if (ret) {
+			error("failed to create global roots: %d", ret);
+			goto error;
+		}
+	}
+
 	ret = make_root_dir(trans, root);
 	if (ret) {
 		error("failed to setup the root directory: %d", ret);
Signature · 1 line
2.26.3
#25 lore raw

Re: [PATCH v5 18/19] btrfs-progs: mkfs: create the global root's

Quoted 4 lines · Josef Bacik

Now that we have all of the supporting code, add the ability to create all of the global roots for an extent tree v2 fs. This will default to nr_cpu's, but also allow the user to specify how many global roots they would like.

Why is number of online cpus a good default? Or how a user should know what's a good number? It resembles the allocation groups on xfs that are set at mkfs time and once the filesystem is grown the size remains but the number explodes and becomes problematic if the the old/new sizes are disproportionate. We have more flexibility in btrfs with the resize so we could afford to set the intial number based rather on the device size and then a rebalance after resize can adjust that again. Maybe there's something in kernel taking care of that, I don't know.

#26 lore raw

Re: [PATCH v5 18/19] btrfs-progs: mkfs: create the global root's

Quoted 6 lines · David Sterba
Quoted 4 lines · Josef Bacik

Now that we have all of the supporting code, add the ability to create all of the global roots for an extent tree v2 fs. This will default to nr_cpu's, but also allow the user to specify how many global roots they would like.

Quoted 8 lines

Why is number of online cpus a good default? Or how a user should know what's a good number? It resembles the allocation groups on xfs that are set at mkfs time and once the filesystem is grown the size remains but the number explodes and becomes problematic if the the old/new sizes are disproportionate. We have more flexibility in btrfs with the resize so we could afford to set the intial number based rather on the device size and then a rebalance after resize can adjust that again. Maybe there's something in kernel taking care of that, I don't know.

Right now I have no idea what a good number is, so I'm defaulting to NR_CPUS. I *think* this is a good idea because generally we want to spread the locking pain, so hopefully NR_CPU's is good enough for most people?

We allow setting your own number if a user does the benchmarking to find their ideal global roots number (*cough*Zygo*cough*). And it'll be easy enough to add new global roots since we tie the block group to the global root at create time. Thanks,

Josef

#27 lore raw

PATCH v5 19/19 btrfs-progs: check: don't do the root item check for extent tree v2

In reply to Josef Bacik

With the current set of changes we could probably do this check, but it would involve changing the code quite a bit, and in the future we're not going to track the metadata in the extent tree at all. Since this check was for a very old kernel just skip it for extent tree v2.

---
 check/main.c | 3 +++
 1 file changed, 3 insertions(+)
check/main.c +3 -0
index 45065989..6bedd648 100644
@@ -9860,6 +9860,9 @@ static int repair_root_items(void)
 	int bad_roots = 0;
 	int need_trans = 0;
 
+	if (btrfs_fs_incompat(gfs_info, EXTENT_TREE_V2))
+		return 0;
+
 	btrfs_init_path(&path);
 
 	ret = build_roots_info_cache();
Signature · 1 line
2.26.3
#28 lore raw
In reply to Josef Bacik
Quoted 66 lines · Josef Bacik

v4->v3:
- Rebase onto devel, depends on "btrfs-progs: cleanup btrfs_item* accessors".
- Dropped the various patches that have already been merged into -progs.

v3->v4:
- Rebase onto devel, depends on the v3 prep patches that were sent on December 1st which has the rest of the "don't access ->*_root" patches.
- I think I screwed up the versioning of this, but I lost the other submission, so call this v3.

v1->v2:
- These depend on the v3 of the prep patches (it's marked as v2 because I'm stupid, but the second v2 posting I sent.)
- I've moved the global root rb tree patches into this series to differentiate them from the actual fixes in the prep series.

--- Original email ---
Hello,

These patches are the first chunk of the extent tree v2 format changes. This includes the separate block group root which will hold all of the block group items. This also includes the global root support, which is the work to allow us to have multiple extent, csum, and free space trees in the same file system.

The goal of these two changes are straightforward. For the block group root, on very large file systems the block group items are very widely separated, which means it takes a very long time to mount the file system on large, slow disks. Putting the block group items in their own root will allow us to densely populate the tree and dramatically increase mount times in these cases.

The global roots change is motivated by lock contention on the root nodes of these global roots. I've had to make many changes to how we run delayed refs to speed up things like the transaction commit because of all the delayed refs going into one tree and contending on the root node of the extent tree. In the same token you can have heavy lock contention on the csum roots when writing to many files. Allowing for multiple roots will let us spread the lock contention load around.

I have disabled a few key features, namely balance and qgroups. There will be more to come as I make more and more invasive changes, and then they will slowly be re-enabled as the work is added. These are disabled to avoid a bunch of work that would be thrown away by future changes.

These patches have passed xfstests without panicing, but clearly failing a lot of tests because of the disabled features. I've also run it through fsperf to validate that there are no major performance regressions.

WARNING: there are many more format changes planned, this is just the first batch. If you want to test then please feel free, but know that the format is still in flux. Thanks,

Josef

Josef Bacik (19):
btrfs-progs: add support for loading the block group root
btrfs-progs: add print support for the block group tree
btrfs-progs: mkfs: use the btrfs_block_group_root helper
btrfs-progs: check-lowmem: use the btrfs_block_group_root helper
btrfs-progs: handle no bg item in extent tree for free space tree
btrfs-progs: mkfs: add support for the block group tree
btrfs-progs: check: add block group tree support
btrfs-progs: qgroup-verify: scan extents based on block groups
btrfs-progs: check: make free space tree validation extent tree v2
aware
btrfs-progs: check: add helper to reinit the root based on a key
btrfs-progs: check: handle the block group tree properly

Quoted 8 lines

btrfs-progs: set the number of global roots in the super block
btrfs-progs: handle the per-block group global root id
btrfs-progs: add a btrfs_delete_and_free_root helper
btrfs-progs: make btrfs_clear_free_space_tree extent tree v2 aware
btrfs-progs: make btrfs_create_tree take a key for the root key
btrfs-progs: mkfs: set chunk_item_objectid properly for extent tree v2
btrfs-progs: mkfs: create the global root's
btrfs-progs: check: don't do the root item check for extent tree v2

Added to devel, thanks.