1 /* 2 * Copyright (C) 2008 Oracle. All rights reserved. 3 * 4 * This program is free software; you can redistribute it and/or 5 * modify it under the terms of the GNU General Public 6 * License v2 as published by the Free Software Foundation. 7 * 8 * This program is distributed in the hope that it will be useful, 9 * but WITHOUT ANY WARRANTY; without even the implied warranty of 10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 11 * General Public License for more details. 12 * 13 * You should have received a copy of the GNU General Public 14 * License along with this program; if not, write to the 15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 16 * Boston, MA 021110-1307, USA. 17 */ 18 #ifndef __DELAYED_REF__ 19 #define __DELAYED_REF__ 20 21 /* these are the possible values of struct btrfs_delayed_ref_node->action */ 22 #define BTRFS_ADD_DELAYED_REF 1 /* add one backref to the tree */ 23 #define BTRFS_DROP_DELAYED_REF 2 /* delete one backref from the tree */ 24 #define BTRFS_ADD_DELAYED_EXTENT 3 /* record a full extent allocation */ 25 #define BTRFS_UPDATE_DELAYED_HEAD 4 /* not changing ref count on head ref */ 26 27 /* 28 * XXX: Qu: I really hate the design that ref_head and tree/data ref shares the 29 * same ref_node structure. 30 * Ref_head is in a higher logic level than tree/data ref, and duplicated 31 * bytenr/num_bytes in ref_node is really a waste or memory, they should be 32 * referred from ref_head. 33 * This gets more disgusting after we use list to store tree/data ref in 34 * ref_head. Must clean this mess up later. 35 */ 36 struct btrfs_delayed_ref_node { 37 /*data/tree ref use list, stored in ref_head->ref_list. */ 38 struct list_head list; 39 /* 40 * If action is BTRFS_ADD_DELAYED_REF, also link this node to 41 * ref_head->ref_add_list, then we do not need to iterate the 42 * whole ref_head->ref_list to find BTRFS_ADD_DELAYED_REF nodes. 43 */ 44 struct list_head add_list; 45 46 /* the starting bytenr of the extent */ 47 u64 bytenr; 48 49 /* the size of the extent */ 50 u64 num_bytes; 51 52 /* seq number to keep track of insertion order */ 53 u64 seq; 54 55 /* ref count on this data structure */ 56 atomic_t refs; 57 58 /* 59 * how many refs is this entry adding or deleting. For 60 * head refs, this may be a negative number because it is keeping 61 * track of the total mods done to the reference count. 62 * For individual refs, this will always be a positive number 63 * 64 * It may be more than one, since it is possible for a single 65 * parent to have more than one ref on an extent 66 */ 67 int ref_mod; 68 69 unsigned int action:8; 70 unsigned int type:8; 71 /* is this node still in the rbtree? */ 72 unsigned int is_head:1; 73 unsigned int in_tree:1; 74 }; 75 76 struct btrfs_delayed_extent_op { 77 struct btrfs_disk_key key; 78 u8 level; 79 bool update_key; 80 bool update_flags; 81 bool is_data; 82 u64 flags_to_set; 83 }; 84 85 /* 86 * the head refs are used to hold a lock on a given extent, which allows us 87 * to make sure that only one process is running the delayed refs 88 * at a time for a single extent. They also store the sum of all the 89 * reference count modifications we've queued up. 90 */ 91 struct btrfs_delayed_ref_head { 92 struct btrfs_delayed_ref_node node; 93 94 /* 95 * the mutex is held while running the refs, and it is also 96 * held when checking the sum of reference modifications. 97 */ 98 struct mutex mutex; 99 100 spinlock_t lock; 101 struct list_head ref_list; 102 /* accumulate add BTRFS_ADD_DELAYED_REF nodes to this ref_add_list. */ 103 struct list_head ref_add_list; 104 105 struct rb_node href_node; 106 107 struct btrfs_delayed_extent_op *extent_op; 108 109 /* 110 * This is used to track the final ref_mod from all the refs associated 111 * with this head ref, this is not adjusted as delayed refs are run, 112 * this is meant to track if we need to do the csum accounting or not. 113 */ 114 int total_ref_mod; 115 116 /* 117 * For qgroup reserved space freeing. 118 * 119 * ref_root and reserved will be recorded after 120 * BTRFS_ADD_DELAYED_EXTENT is called. 121 * And will be used to free reserved qgroup space at 122 * run_delayed_refs() time. 123 */ 124 u64 qgroup_ref_root; 125 u64 qgroup_reserved; 126 127 /* 128 * when a new extent is allocated, it is just reserved in memory 129 * The actual extent isn't inserted into the extent allocation tree 130 * until the delayed ref is processed. must_insert_reserved is 131 * used to flag a delayed ref so the accounting can be updated 132 * when a full insert is done. 133 * 134 * It is possible the extent will be freed before it is ever 135 * inserted into the extent allocation tree. In this case 136 * we need to update the in ram accounting to properly reflect 137 * the free has happened. 138 */ 139 unsigned int must_insert_reserved:1; 140 unsigned int is_data:1; 141 unsigned int processing:1; 142 }; 143 144 struct btrfs_delayed_tree_ref { 145 struct btrfs_delayed_ref_node node; 146 u64 root; 147 u64 parent; 148 int level; 149 }; 150 151 struct btrfs_delayed_data_ref { 152 struct btrfs_delayed_ref_node node; 153 u64 root; 154 u64 parent; 155 u64 objectid; 156 u64 offset; 157 }; 158 159 struct btrfs_delayed_ref_root { 160 /* head ref rbtree */ 161 struct rb_root href_root; 162 163 /* dirty extent records */ 164 struct rb_root dirty_extent_root; 165 166 /* this spin lock protects the rbtree and the entries inside */ 167 spinlock_t lock; 168 169 /* how many delayed ref updates we've queued, used by the 170 * throttling code 171 */ 172 atomic_t num_entries; 173 174 /* total number of head nodes in tree */ 175 unsigned long num_heads; 176 177 /* total number of head nodes ready for processing */ 178 unsigned long num_heads_ready; 179 180 u64 pending_csums; 181 182 /* 183 * set when the tree is flushing before a transaction commit, 184 * used by the throttling code to decide if new updates need 185 * to be run right away 186 */ 187 int flushing; 188 189 u64 run_delayed_start; 190 191 /* 192 * To make qgroup to skip given root. 193 * This is for snapshot, as btrfs_qgroup_inherit() will manually 194 * modify counters for snapshot and its source, so we should skip 195 * the snapshot in new_root/old_roots or it will get calculated twice 196 */ 197 u64 qgroup_to_skip; 198 }; 199 200 extern struct kmem_cache *btrfs_delayed_ref_head_cachep; 201 extern struct kmem_cache *btrfs_delayed_tree_ref_cachep; 202 extern struct kmem_cache *btrfs_delayed_data_ref_cachep; 203 extern struct kmem_cache *btrfs_delayed_extent_op_cachep; 204 205 int btrfs_delayed_ref_init(void); 206 void btrfs_delayed_ref_exit(void); 207 208 static inline struct btrfs_delayed_extent_op * 209 btrfs_alloc_delayed_extent_op(void) 210 { 211 return kmem_cache_alloc(btrfs_delayed_extent_op_cachep, GFP_NOFS); 212 } 213 214 static inline void 215 btrfs_free_delayed_extent_op(struct btrfs_delayed_extent_op *op) 216 { 217 if (op) 218 kmem_cache_free(btrfs_delayed_extent_op_cachep, op); 219 } 220 221 static inline void btrfs_put_delayed_ref(struct btrfs_delayed_ref_node *ref) 222 { 223 WARN_ON(atomic_read(&ref->refs) == 0); 224 if (atomic_dec_and_test(&ref->refs)) { 225 WARN_ON(ref->in_tree); 226 switch (ref->type) { 227 case BTRFS_TREE_BLOCK_REF_KEY: 228 case BTRFS_SHARED_BLOCK_REF_KEY: 229 kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref); 230 break; 231 case BTRFS_EXTENT_DATA_REF_KEY: 232 case BTRFS_SHARED_DATA_REF_KEY: 233 kmem_cache_free(btrfs_delayed_data_ref_cachep, ref); 234 break; 235 case 0: 236 kmem_cache_free(btrfs_delayed_ref_head_cachep, ref); 237 break; 238 default: 239 BUG(); 240 } 241 } 242 } 243 244 int btrfs_add_delayed_tree_ref(struct btrfs_fs_info *fs_info, 245 struct btrfs_trans_handle *trans, 246 u64 bytenr, u64 num_bytes, u64 parent, 247 u64 ref_root, int level, int action, 248 struct btrfs_delayed_extent_op *extent_op); 249 int btrfs_add_delayed_data_ref(struct btrfs_fs_info *fs_info, 250 struct btrfs_trans_handle *trans, 251 u64 bytenr, u64 num_bytes, 252 u64 parent, u64 ref_root, 253 u64 owner, u64 offset, u64 reserved, int action, 254 struct btrfs_delayed_extent_op *extent_op); 255 int btrfs_add_delayed_extent_op(struct btrfs_fs_info *fs_info, 256 struct btrfs_trans_handle *trans, 257 u64 bytenr, u64 num_bytes, 258 struct btrfs_delayed_extent_op *extent_op); 259 void btrfs_merge_delayed_refs(struct btrfs_trans_handle *trans, 260 struct btrfs_fs_info *fs_info, 261 struct btrfs_delayed_ref_root *delayed_refs, 262 struct btrfs_delayed_ref_head *head); 263 264 struct btrfs_delayed_ref_head * 265 btrfs_find_delayed_ref_head(struct btrfs_trans_handle *trans, u64 bytenr); 266 int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans, 267 struct btrfs_delayed_ref_head *head); 268 static inline void btrfs_delayed_ref_unlock(struct btrfs_delayed_ref_head *head) 269 { 270 mutex_unlock(&head->mutex); 271 } 272 273 274 struct btrfs_delayed_ref_head * 275 btrfs_select_ref_head(struct btrfs_trans_handle *trans); 276 277 int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info, 278 struct btrfs_delayed_ref_root *delayed_refs, 279 u64 seq); 280 281 /* 282 * a node might live in a head or a regular ref, this lets you 283 * test for the proper type to use. 284 */ 285 static int btrfs_delayed_ref_is_head(struct btrfs_delayed_ref_node *node) 286 { 287 return node->is_head; 288 } 289 290 /* 291 * helper functions to cast a node into its container 292 */ 293 static inline struct btrfs_delayed_tree_ref * 294 btrfs_delayed_node_to_tree_ref(struct btrfs_delayed_ref_node *node) 295 { 296 WARN_ON(btrfs_delayed_ref_is_head(node)); 297 return container_of(node, struct btrfs_delayed_tree_ref, node); 298 } 299 300 static inline struct btrfs_delayed_data_ref * 301 btrfs_delayed_node_to_data_ref(struct btrfs_delayed_ref_node *node) 302 { 303 WARN_ON(btrfs_delayed_ref_is_head(node)); 304 return container_of(node, struct btrfs_delayed_data_ref, node); 305 } 306 307 static inline struct btrfs_delayed_ref_head * 308 btrfs_delayed_node_to_head(struct btrfs_delayed_ref_node *node) 309 { 310 WARN_ON(!btrfs_delayed_ref_is_head(node)); 311 return container_of(node, struct btrfs_delayed_ref_head, node); 312 } 313 #endif 314