forked from ~ljy/RK356X_SDK_RELEASE

blame | history | patch | commit | commitdiff
hc
2024-05-10 9999e48639b3cecb08ffb37358bcba3b48161b29 
 kernel/fs/btrfs/qgroup.h
....@@ -6,6 +6,9 @@
66 #ifndef BTRFS_QGROUP_H
77 #define BTRFS_QGROUP_H
88 
9
+#include <linux/spinlock.h>
10
+#include <linux/rbtree.h>
11
+#include <linux/kobject.h>
912 #include "ulist.h"
1013 #include "delayed-ref.h"
1114 
....@@ -38,6 +41,66 @@
3841  */
3942 
4043 /*
44
+ * Special performance optimization for balance.
45
+ *
46
+ * For balance, we need to swap subtree of subvolume and reloc trees.
47
+ * In theory, we need to trace all subtree blocks of both subvolume and reloc
48
+ * trees, since their owner has changed during such swap.
49
+ *
50
+ * However since balance has ensured that both subtrees are containing the
51
+ * same contents and have the same tree structures, such swap won't cause
52
+ * qgroup number change.
53
+ *
54
+ * But there is a race window between subtree swap and transaction commit,
55
+ * during that window, if we increase/decrease tree level or merge/split tree
56
+ * blocks, we still need to trace the original subtrees.
57
+ *
58
+ * So for balance, we use a delayed subtree tracing, whose workflow is:
59
+ *
60
+ * 1) Record the subtree root block get swapped.
61
+ *
62
+ *    During subtree swap:
63
+ *    O = Old tree blocks
64
+ *    N = New tree blocks
65
+ *          reloc tree                     subvolume tree X
66
+ *             Root                               Root
67
+ *            /    \                             /    \
68
+ *          NA     OB                          OA      OB
69
+ *        /  |     |  \                      /  |      |  \
70
+ *      NC  ND     OE  OF                   OC  OD     OE  OF
71
+ *
72
+ *   In this case, NA and OA are going to be swapped, record (NA, OA) into
73
+ *   subvolume tree X.
74
+ *
75
+ * 2) After subtree swap.
76
+ *          reloc tree                     subvolume tree X
77
+ *             Root                               Root
78
+ *            /    \                             /    \
79
+ *          OA     OB                          NA      OB
80
+ *        /  |     |  \                      /  |      |  \
81
+ *      OC  OD     OE  OF                   NC  ND     OE  OF
82
+ *
83
+ * 3a) COW happens for OB
84
+ *     If we are going to COW tree block OB, we check OB's bytenr against
85
+ *     tree X's swapped_blocks structure.
86
+ *     If it doesn't fit any, nothing will happen.
87
+ *
88
+ * 3b) COW happens for NA
89
+ *     Check NA's bytenr against tree X's swapped_blocks, and get a hit.
90
+ *     Then we do subtree scan on both subtrees OA and NA.
91
+ *     Resulting 6 tree blocks to be scanned (OA, OC, OD, NA, NC, ND).
92
+ *
93
+ *     Then no matter what we do to subvolume tree X, qgroup numbers will
94
+ *     still be correct.
95
+ *     Then NA's record gets removed from X's swapped_blocks.
96
+ *
97
+ * 4)  Transaction commit
98
+ *     Any record in X's swapped_blocks gets removed, since there is no
99
+ *     modification to the swapped subtrees, no need to trigger heavy qgroup
100
+ *     subtree rescan for them.
101
+ */
102
+
103
+/*
41104  * Record a dirty extent, and info qgroup to update quota on it
42105  * TODO: Use kmem cache to alloc it.
43106  */
....@@ -45,7 +108,36 @@
45108 	struct rb_node node;
46109 	u64 bytenr;
47110 	u64 num_bytes;
111
+
112
+	/*
113
+	 * For qgroup reserved data space freeing.
114
+	 *
115
+	 * @data_rsv_refroot and @data_rsv will be recorded after
116
+	 * BTRFS_ADD_DELAYED_EXTENT is called.
117
+	 * And will be used to free reserved qgroup space at
118
+	 * transaction commit time.
119
+	 */
120
+	u32 data_rsv;		/* reserved data space needs to be freed */
121
+	u64 data_rsv_refroot;	/* which root the reserved data belongs to */
48122 	struct ulist *old_roots;
123
+};
124
+
125
+struct btrfs_qgroup_swapped_block {
126
+	struct rb_node node;
127
+
128
+	int level;
129
+	bool trace_leaf;
130
+
131
+	/* bytenr/generation of the tree block in subvolume tree after swap */
132
+	u64 subvol_bytenr;
133
+	u64 subvol_generation;
134
+
135
+	/* bytenr/generation of the tree block in reloc tree after swap */
136
+	u64 reloc_bytenr;
137
+	u64 reloc_generation;
138
+
139
+	u64 last_snapshot;
140
+	struct btrfs_key first_key;
49141 };
50142 
51143 /*
....@@ -70,7 +162,7 @@
70162  *	be converted into META_PERTRANS.
71163  */
72164 enum btrfs_qgroup_rsv_type {
73
-	BTRFS_QGROUP_RSV_DATA = 0,
165
+	BTRFS_QGROUP_RSV_DATA,
74166 	BTRFS_QGROUP_RSV_META_PERTRANS,
75167 	BTRFS_QGROUP_RSV_META_PREALLOC,
76168 	BTRFS_QGROUP_RSV_LAST,
....@@ -81,10 +173,10 @@
81173  *
82174  * Each type should have different reservation behavior.
83175  * E.g, data follows its io_tree flag modification, while
84
- * *currently* meta is just reserve-and-clear during transcation.
176
+ * *currently* meta is just reserve-and-clear during transaction.
85177  *
86178  * TODO: Add new type for reservation which can survive transaction commit.
87
- * Currect metadata reservation behavior is not suitable for such case.
179
+ * Current metadata reservation behavior is not suitable for such case.
88180  */
89181 struct btrfs_qgroup_rsv {
90182 	u64 values[BTRFS_QGROUP_RSV_LAST];
....@@ -132,7 +224,17 @@
132224 	 */
133225 	u64 old_refcnt;
134226 	u64 new_refcnt;
227
+
228
+	/*
229
+	 * Sysfs kobjectid
230
+	 */
231
+	struct kobject kobj;
135232 };
233
+
234
+static inline u64 btrfs_qgroup_subvolid(u64 qgroupid)
235
+{
236
+	return (qgroupid & ((1ULL << BTRFS_QGROUP_LEVEL_SHIFT) - 1));
237
+}
136238 
137239 /*
138240  * For qgroup event trace points only
....@@ -246,15 +348,6 @@
246348 void btrfs_qgroup_free_refroot(struct btrfs_fs_info *fs_info,
247349 			       u64 ref_root, u64 num_bytes,
248350 			       enum btrfs_qgroup_rsv_type type);
249
-static inline void btrfs_qgroup_free_delayed_ref(struct btrfs_fs_info *fs_info,
250
-						 u64 ref_root, u64 num_bytes)
251
-{
252
-	if (!test_bit(BTRFS_FS_QUOTA_ENABLED, &fs_info->flags))
253
-		return;
254
-	trace_btrfs_qgroup_free_delayed_ref(fs_info, ref_root, num_bytes);
255
-	btrfs_qgroup_free_refroot(fs_info, ref_root, num_bytes,
256
-				  BTRFS_QGROUP_RSV_DATA);
257
-}
258351 
259352 #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS
260353 int btrfs_verify_qgroup_counts(struct btrfs_fs_info *fs_info, u64 qgroupid,
....@@ -262,12 +355,14 @@
262355 #endif
263356 
264357 /* New io_tree based accurate qgroup reserve API */
265
-int btrfs_qgroup_reserve_data(struct inode *inode,
358
+int btrfs_qgroup_reserve_data(struct btrfs_inode *inode,
266359 			struct extent_changeset **reserved, u64 start, u64 len);
267
-int btrfs_qgroup_release_data(struct inode *inode, u64 start, u64 len);
268
-int btrfs_qgroup_free_data(struct inode *inode,
269
-			struct extent_changeset *reserved, u64 start, u64 len);
270
-
360
+int btrfs_qgroup_release_data(struct btrfs_inode *inode, u64 start, u64 len);
361
+int btrfs_qgroup_free_data(struct btrfs_inode *inode,
362
+			   struct extent_changeset *reserved, u64 start,
363
+			   u64 len);
364
+int btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
365
+			      enum btrfs_qgroup_rsv_type type, bool enforce);
271366 int __btrfs_qgroup_reserve_meta(struct btrfs_root *root, int num_bytes,
272367 				enum btrfs_qgroup_rsv_type type, bool enforce);
273368 /* Reserve metadata space for pertrans and prealloc type */
....@@ -317,6 +412,22 @@
317412  */
318413 void btrfs_qgroup_convert_reserved_meta(struct btrfs_root *root, int num_bytes);
319414 
320
-void btrfs_qgroup_check_reserved_leak(struct inode *inode);
415
+void btrfs_qgroup_check_reserved_leak(struct btrfs_inode *inode);
416
+
417
+/* btrfs_qgroup_swapped_blocks related functions */
418
+void btrfs_qgroup_init_swapped_blocks(
419
+	struct btrfs_qgroup_swapped_blocks *swapped_blocks);
420
+
421
+void btrfs_qgroup_clean_swapped_blocks(struct btrfs_root *root);
422
+int btrfs_qgroup_add_swapped_blocks(struct btrfs_trans_handle *trans,
423
+		struct btrfs_root *subvol_root,
424
+		struct btrfs_block_group *bg,
425
+		struct extent_buffer *subvol_parent, int subvol_slot,
426
+		struct extent_buffer *reloc_parent, int reloc_slot,
427
+		u64 last_snapshot);
428
+int btrfs_qgroup_trace_subtree_after_cow(struct btrfs_trans_handle *trans,
429
+		struct btrfs_root *root, struct extent_buffer *eb);
430
+void btrfs_qgroup_destroy_extent_records(struct btrfs_transaction *trans);
431
+bool btrfs_check_quota_leak(struct btrfs_fs_info *fs_info);
321432 
322433 #endif