1 files changed, 89 insertions, 9 deletions

diff --git a/fs/btrfs/qgroup.h b/fs/btrfs/qgroup.h
index 1bc64c864b62..26932a8a1993 100644
--- a/fs/btrfs/qgroup.h
+++ b/fs/btrfs/qgroup.h
@@ -23,6 +23,34 @@
 #include "delayed-ref.h"
 
 /*
+ * Btrfs qgroup overview
+ *
+ * Btrfs qgroup splits into 3 main part:
+ * 1) Reserve
+ *    Reserve metadata/data space for incoming operations
+ *    Affect how qgroup limit works
+ *
+ * 2) Trace
+ *    Tell btrfs qgroup to trace dirty extents.
+ *
+ *    Dirty extents including:
+ *    - Newly allocated extents
+ *    - Extents going to be deleted (in this trans)
+ *    - Extents whose owner is going to be modified
+ *
+ *    This is the main part affects whether qgroup numbers will stay
+ *    consistent.
+ *    Btrfs qgroup can trace clean extents and won't cause any problem,
+ *    but it will consume extra CPU time, it should be avoided if possible.
+ *
+ * 3) Account
+ *    Btrfs qgroup will updates its numbers, based on dirty extents traced
+ *    in previous step.
+ *
+ *    Normally at qgroup rescan and transaction commit time.
+ */
+
+/*
  * Record a dirty extent, and info qgroup to update quota on it
  * TODO: Use kmem cache to alloc it.
  */
@@ -65,34 +93,85 @@ struct btrfs_delayed_extent_op;
 int btrfs_qgroup_prepare_account_extents(struct btrfs_trans_handle *trans,
 					 struct btrfs_fs_info *fs_info);
 /*
- * Insert one dirty extent record into @delayed_refs, informing qgroup to
- * account that extent at commit trans time.
+ * Inform qgroup to trace one dirty extent, its info is recorded in @record.
+ * So qgroup can account it at transaction committing time.
  *
- * No lock version, caller must acquire delayed ref lock and allocate memory.
+ * No lock version, caller must acquire delayed ref lock and allocated memory,
+ * then call btrfs_qgroup_trace_extent_post() after exiting lock context.
  *
  * Return 0 for success insert
  * Return >0 for existing record, caller can free @record safely.
  * Error is not possible
  */
-int btrfs_qgroup_insert_dirty_extent_nolock(
+int btrfs_qgroup_trace_extent_nolock(
 		struct btrfs_fs_info *fs_info,
 		struct btrfs_delayed_ref_root *delayed_refs,
 		struct btrfs_qgroup_extent_record *record);
 
 /*
- * Insert one dirty extent record into @delayed_refs, informing qgroup to
- * account that extent at commit trans time.
+ * Post handler after qgroup_trace_extent_nolock().
  *
- * Better encapsulated version.
+ * NOTE: Current qgroup does the expensive backref walk at transaction
+ * committing time with TRANS_STATE_COMMIT_DOING, this blocks incoming
+ * new transaction.
+ * This is designed to allow btrfs_find_all_roots() to get correct new_roots
+ * result.
+ *
+ * However for old_roots there is no need to do backref walk at that time,
+ * since we search commit roots to walk backref and result will always be
+ * correct.
+ *
+ * Due to the nature of no lock version, we can't do backref there.
+ * So we must call btrfs_qgroup_trace_extent_post() after exiting
+ * spinlock context.
+ *
+ * TODO: If we can fix and prove btrfs_find_all_roots() can get correct result
+ * using current root, then we can move all expensive backref walk out of
+ * transaction committing, but not now as qgroup accounting will be wrong again.
+ */
+int btrfs_qgroup_trace_extent_post(struct btrfs_fs_info *fs_info,
+				   struct btrfs_qgroup_extent_record *qrecord);
+
+/*
+ * Inform qgroup to trace one dirty extent, specified by @bytenr and
+ * @num_bytes.
+ * So qgroup can account it at commit trans time.
+ *
+ * Better encapsulated version, with memory allocation and backref walk for
+ * commit roots.
+ * So this can sleep.
  *
  * Return 0 if the operation is done.
  * Return <0 for error, like memory allocation failure or invalid parameter
  * (NULL trans)
  */
-int btrfs_qgroup_insert_dirty_extent(struct btrfs_trans_handle *trans,
+int btrfs_qgroup_trace_extent(struct btrfs_trans_handle *trans,
 		struct btrfs_fs_info *fs_info, u64 bytenr, u64 num_bytes,
 		gfp_t gfp_flag);
 
+/*
+ * Inform qgroup to trace all leaf items of data
+ *
+ * Return 0 for success
+ * Return <0 for error(ENOMEM)
+ */
+int btrfs_qgroup_trace_leaf_items(struct btrfs_trans_handle *trans,
+				  struct btrfs_fs_info *fs_info,
+				  struct extent_buffer *eb);
+/*
+ * Inform qgroup to trace a whole subtree, including all its child tree
+ * blocks and data.
+ * The root tree block is specified by @root_eb.
+ *
+ * Normally used by relocation(tree block swap) and subvolume deletion.
+ *
+ * Return 0 for success
+ * Return <0 for error(ENOMEM or tree search error)
+ */
+int btrfs_qgroup_trace_subtree(struct btrfs_trans_handle *trans,
+			       struct btrfs_root *root,
+			       struct extent_buffer *root_eb,
+			       u64 root_gen, int root_level);
 int
 btrfs_qgroup_account_extent(struct btrfs_trans_handle *trans,
 			    struct btrfs_fs_info *fs_info,
@@ -129,7 +208,8 @@ int btrfs_qgroup_reserve_data(struct inode *inode, u64 start, u64 len);
 int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len);
 int btrfs_qgroup_free_data(struct inode *inode, u64 start, u64 len);
 
-int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes);
+int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
+			      bool enforce);
 void btrfs_qgroup_free_meta_all(struct btrfs_root *root);
 void btrfs_qgroup_free_meta(struct btrfs_root *root, int num_bytes);
 void btrfs_qgroup_check_reserved_leak(struct inode *inode);