156bec294SChris Mason /* 256bec294SChris Mason * Copyright (C) 2008 Oracle. All rights reserved. 356bec294SChris Mason * 456bec294SChris Mason * This program is free software; you can redistribute it and/or 556bec294SChris Mason * modify it under the terms of the GNU General Public 656bec294SChris Mason * License v2 as published by the Free Software Foundation. 756bec294SChris Mason * 856bec294SChris Mason * This program is distributed in the hope that it will be useful, 956bec294SChris Mason * but WITHOUT ANY WARRANTY; without even the implied warranty of 1056bec294SChris Mason * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 1156bec294SChris Mason * General Public License for more details. 1256bec294SChris Mason * 1356bec294SChris Mason * You should have received a copy of the GNU General Public 1456bec294SChris Mason * License along with this program; if not, write to the 1556bec294SChris Mason * Free Software Foundation, Inc., 59 Temple Place - Suite 330, 1656bec294SChris Mason * Boston, MA 021110-1307, USA. 1756bec294SChris Mason */ 1856bec294SChris Mason #ifndef __DELAYED_REF__ 1956bec294SChris Mason #define __DELAYED_REF__ 2056bec294SChris Mason 2144a075bdSWang Sheng-Hui /* these are the possible values of struct btrfs_delayed_ref_node->action */ 2256bec294SChris Mason #define BTRFS_ADD_DELAYED_REF 1 /* add one backref to the tree */ 2356bec294SChris Mason #define BTRFS_DROP_DELAYED_REF 2 /* delete one backref from the tree */ 2456bec294SChris Mason #define BTRFS_ADD_DELAYED_EXTENT 3 /* record a full extent allocation */ 251a81af4dSChris Mason #define BTRFS_UPDATE_DELAYED_HEAD 4 /* not changing ref count on head ref */ 2656bec294SChris Mason 27c6fc2454SQu Wenruo /* 28c6fc2454SQu Wenruo * XXX: Qu: I really hate the design that ref_head and tree/data ref shares the 29c6fc2454SQu Wenruo * same ref_node structure. 30c6fc2454SQu Wenruo * Ref_head is in a higher logic level than tree/data ref, and duplicated 31c6fc2454SQu Wenruo * bytenr/num_bytes in ref_node is really a waste or memory, they should be 32c6fc2454SQu Wenruo * referred from ref_head. 33c6fc2454SQu Wenruo * This gets more disgusting after we use list to store tree/data ref in 34c6fc2454SQu Wenruo * ref_head. Must clean this mess up later. 35c6fc2454SQu Wenruo */ 3656bec294SChris Mason struct btrfs_delayed_ref_node { 37c6fc2454SQu Wenruo /* 38c6fc2454SQu Wenruo * ref_head use rb tree, stored in ref_root->href. 39c6fc2454SQu Wenruo * indexed by bytenr 40c6fc2454SQu Wenruo */ 4156bec294SChris Mason struct rb_node rb_node; 4256bec294SChris Mason 43c6fc2454SQu Wenruo /*data/tree ref use list, stored in ref_head->ref_list. */ 44c6fc2454SQu Wenruo struct list_head list; 45c6fc2454SQu Wenruo 4656bec294SChris Mason /* the starting bytenr of the extent */ 4756bec294SChris Mason u64 bytenr; 4856bec294SChris Mason 4956bec294SChris Mason /* the size of the extent */ 5056bec294SChris Mason u64 num_bytes; 5156bec294SChris Mason 5200f04b88SArne Jansen /* seq number to keep track of insertion order */ 5300f04b88SArne Jansen u64 seq; 5400f04b88SArne Jansen 5556bec294SChris Mason /* ref count on this data structure */ 5656bec294SChris Mason atomic_t refs; 5756bec294SChris Mason 5856bec294SChris Mason /* 5956bec294SChris Mason * how many refs is this entry adding or deleting. For 6056bec294SChris Mason * head refs, this may be a negative number because it is keeping 6156bec294SChris Mason * track of the total mods done to the reference count. 6256bec294SChris Mason * For individual refs, this will always be a positive number 6356bec294SChris Mason * 6456bec294SChris Mason * It may be more than one, since it is possible for a single 6556bec294SChris Mason * parent to have more than one ref on an extent 6656bec294SChris Mason */ 6756bec294SChris Mason int ref_mod; 6856bec294SChris Mason 695d4f98a2SYan Zheng unsigned int action:8; 705d4f98a2SYan Zheng unsigned int type:8; 7156bec294SChris Mason /* is this node still in the rbtree? */ 725d4f98a2SYan Zheng unsigned int is_head:1; 7356bec294SChris Mason unsigned int in_tree:1; 7456bec294SChris Mason }; 7556bec294SChris Mason 765d4f98a2SYan Zheng struct btrfs_delayed_extent_op { 775d4f98a2SYan Zheng struct btrfs_disk_key key; 7835b3ad50SDavid Sterba u8 level; 7935b3ad50SDavid Sterba bool update_key; 8035b3ad50SDavid Sterba bool update_flags; 8135b3ad50SDavid Sterba bool is_data; 825d4f98a2SYan Zheng u64 flags_to_set; 835d4f98a2SYan Zheng }; 845d4f98a2SYan Zheng 8556bec294SChris Mason /* 8656bec294SChris Mason * the head refs are used to hold a lock on a given extent, which allows us 8756bec294SChris Mason * to make sure that only one process is running the delayed refs 8856bec294SChris Mason * at a time for a single extent. They also store the sum of all the 8956bec294SChris Mason * reference count modifications we've queued up. 9056bec294SChris Mason */ 9156bec294SChris Mason struct btrfs_delayed_ref_head { 9256bec294SChris Mason struct btrfs_delayed_ref_node node; 9356bec294SChris Mason 9456bec294SChris Mason /* 9556bec294SChris Mason * the mutex is held while running the refs, and it is also 9656bec294SChris Mason * held when checking the sum of reference modifications. 9756bec294SChris Mason */ 9856bec294SChris Mason struct mutex mutex; 9956bec294SChris Mason 100d7df2c79SJosef Bacik spinlock_t lock; 101c6fc2454SQu Wenruo struct list_head ref_list; 102c3e69d58SChris Mason 103c46effa6SLiu Bo struct rb_node href_node; 104c46effa6SLiu Bo 1055d4f98a2SYan Zheng struct btrfs_delayed_extent_op *extent_op; 1061262133bSJosef Bacik 1071262133bSJosef Bacik /* 1081262133bSJosef Bacik * This is used to track the final ref_mod from all the refs associated 1091262133bSJosef Bacik * with this head ref, this is not adjusted as delayed refs are run, 1101262133bSJosef Bacik * this is meant to track if we need to do the csum accounting or not. 1111262133bSJosef Bacik */ 1121262133bSJosef Bacik int total_ref_mod; 1131262133bSJosef Bacik 11456bec294SChris Mason /* 115f64d5ca8SQu Wenruo * For qgroup reserved space freeing. 116f64d5ca8SQu Wenruo * 117f64d5ca8SQu Wenruo * ref_root and reserved will be recorded after 118f64d5ca8SQu Wenruo * BTRFS_ADD_DELAYED_EXTENT is called. 119f64d5ca8SQu Wenruo * And will be used to free reserved qgroup space at 120f64d5ca8SQu Wenruo * run_delayed_refs() time. 121f64d5ca8SQu Wenruo */ 122f64d5ca8SQu Wenruo u64 qgroup_ref_root; 123f64d5ca8SQu Wenruo u64 qgroup_reserved; 124f64d5ca8SQu Wenruo 125f64d5ca8SQu Wenruo /* 12656bec294SChris Mason * when a new extent is allocated, it is just reserved in memory 12756bec294SChris Mason * The actual extent isn't inserted into the extent allocation tree 12856bec294SChris Mason * until the delayed ref is processed. must_insert_reserved is 12956bec294SChris Mason * used to flag a delayed ref so the accounting can be updated 13056bec294SChris Mason * when a full insert is done. 13156bec294SChris Mason * 13256bec294SChris Mason * It is possible the extent will be freed before it is ever 13356bec294SChris Mason * inserted into the extent allocation tree. In this case 13456bec294SChris Mason * we need to update the in ram accounting to properly reflect 13556bec294SChris Mason * the free has happened. 13656bec294SChris Mason */ 13756bec294SChris Mason unsigned int must_insert_reserved:1; 1385d4f98a2SYan Zheng unsigned int is_data:1; 139d7df2c79SJosef Bacik unsigned int processing:1; 14056bec294SChris Mason }; 14156bec294SChris Mason 1425d4f98a2SYan Zheng struct btrfs_delayed_tree_ref { 14356bec294SChris Mason struct btrfs_delayed_ref_node node; 14456bec294SChris Mason u64 root; 1455d4f98a2SYan Zheng u64 parent; 1465d4f98a2SYan Zheng int level; 1475d4f98a2SYan Zheng }; 14856bec294SChris Mason 1495d4f98a2SYan Zheng struct btrfs_delayed_data_ref { 1505d4f98a2SYan Zheng struct btrfs_delayed_ref_node node; 1515d4f98a2SYan Zheng u64 root; 1525d4f98a2SYan Zheng u64 parent; 1535d4f98a2SYan Zheng u64 objectid; 1545d4f98a2SYan Zheng u64 offset; 15556bec294SChris Mason }; 15656bec294SChris Mason 15756bec294SChris Mason struct btrfs_delayed_ref_root { 158c46effa6SLiu Bo /* head ref rbtree */ 159c46effa6SLiu Bo struct rb_root href_root; 160c46effa6SLiu Bo 1613368d001SQu Wenruo /* dirty extent records */ 1623368d001SQu Wenruo struct rb_root dirty_extent_root; 1633368d001SQu Wenruo 16456bec294SChris Mason /* this spin lock protects the rbtree and the entries inside */ 16556bec294SChris Mason spinlock_t lock; 16656bec294SChris Mason 16756bec294SChris Mason /* how many delayed ref updates we've queued, used by the 16856bec294SChris Mason * throttling code 16956bec294SChris Mason */ 170d7df2c79SJosef Bacik atomic_t num_entries; 17156bec294SChris Mason 172c3e69d58SChris Mason /* total number of head nodes in tree */ 173c3e69d58SChris Mason unsigned long num_heads; 174c3e69d58SChris Mason 175c3e69d58SChris Mason /* total number of head nodes ready for processing */ 176c3e69d58SChris Mason unsigned long num_heads_ready; 177c3e69d58SChris Mason 1781262133bSJosef Bacik u64 pending_csums; 1791262133bSJosef Bacik 18056bec294SChris Mason /* 18156bec294SChris Mason * set when the tree is flushing before a transaction commit, 18256bec294SChris Mason * used by the throttling code to decide if new updates need 18356bec294SChris Mason * to be run right away 18456bec294SChris Mason */ 18556bec294SChris Mason int flushing; 186c3e69d58SChris Mason 187c3e69d58SChris Mason u64 run_delayed_start; 1889086db86SQu Wenruo 1899086db86SQu Wenruo /* 1909086db86SQu Wenruo * To make qgroup to skip given root. 19101327610SNicholas D Steeves * This is for snapshot, as btrfs_qgroup_inherit() will manually 1929086db86SQu Wenruo * modify counters for snapshot and its source, so we should skip 1939086db86SQu Wenruo * the snapshot in new_root/old_roots or it will get calculated twice 1949086db86SQu Wenruo */ 1959086db86SQu Wenruo u64 qgroup_to_skip; 19656bec294SChris Mason }; 19756bec294SChris Mason 19878a6184aSMiao Xie extern struct kmem_cache *btrfs_delayed_ref_head_cachep; 19978a6184aSMiao Xie extern struct kmem_cache *btrfs_delayed_tree_ref_cachep; 20078a6184aSMiao Xie extern struct kmem_cache *btrfs_delayed_data_ref_cachep; 20178a6184aSMiao Xie extern struct kmem_cache *btrfs_delayed_extent_op_cachep; 20278a6184aSMiao Xie 20378a6184aSMiao Xie int btrfs_delayed_ref_init(void); 20478a6184aSMiao Xie void btrfs_delayed_ref_exit(void); 20578a6184aSMiao Xie 20678a6184aSMiao Xie static inline struct btrfs_delayed_extent_op * 20778a6184aSMiao Xie btrfs_alloc_delayed_extent_op(void) 20878a6184aSMiao Xie { 20978a6184aSMiao Xie return kmem_cache_alloc(btrfs_delayed_extent_op_cachep, GFP_NOFS); 21078a6184aSMiao Xie } 21178a6184aSMiao Xie 21278a6184aSMiao Xie static inline void 21378a6184aSMiao Xie btrfs_free_delayed_extent_op(struct btrfs_delayed_extent_op *op) 21478a6184aSMiao Xie { 21578a6184aSMiao Xie if (op) 21678a6184aSMiao Xie kmem_cache_free(btrfs_delayed_extent_op_cachep, op); 21778a6184aSMiao Xie } 21878a6184aSMiao Xie 21956bec294SChris Mason static inline void btrfs_put_delayed_ref(struct btrfs_delayed_ref_node *ref) 22056bec294SChris Mason { 22156bec294SChris Mason WARN_ON(atomic_read(&ref->refs) == 0); 22256bec294SChris Mason if (atomic_dec_and_test(&ref->refs)) { 22356bec294SChris Mason WARN_ON(ref->in_tree); 22478a6184aSMiao Xie switch (ref->type) { 22578a6184aSMiao Xie case BTRFS_TREE_BLOCK_REF_KEY: 22678a6184aSMiao Xie case BTRFS_SHARED_BLOCK_REF_KEY: 22778a6184aSMiao Xie kmem_cache_free(btrfs_delayed_tree_ref_cachep, ref); 22878a6184aSMiao Xie break; 22978a6184aSMiao Xie case BTRFS_EXTENT_DATA_REF_KEY: 23078a6184aSMiao Xie case BTRFS_SHARED_DATA_REF_KEY: 23178a6184aSMiao Xie kmem_cache_free(btrfs_delayed_data_ref_cachep, ref); 23278a6184aSMiao Xie break; 23378a6184aSMiao Xie case 0: 23478a6184aSMiao Xie kmem_cache_free(btrfs_delayed_ref_head_cachep, ref); 23578a6184aSMiao Xie break; 23678a6184aSMiao Xie default: 23778a6184aSMiao Xie BUG(); 23878a6184aSMiao Xie } 23956bec294SChris Mason } 24056bec294SChris Mason } 24156bec294SChris Mason 24266d7e7f0SArne Jansen int btrfs_add_delayed_tree_ref(struct btrfs_fs_info *fs_info, 24366d7e7f0SArne Jansen struct btrfs_trans_handle *trans, 2445d4f98a2SYan Zheng u64 bytenr, u64 num_bytes, u64 parent, 2455d4f98a2SYan Zheng u64 ref_root, int level, int action, 246b06c4bf5SFilipe Manana struct btrfs_delayed_extent_op *extent_op); 24766d7e7f0SArne Jansen int btrfs_add_delayed_data_ref(struct btrfs_fs_info *fs_info, 24866d7e7f0SArne Jansen struct btrfs_trans_handle *trans, 2495d4f98a2SYan Zheng u64 bytenr, u64 num_bytes, 2505d4f98a2SYan Zheng u64 parent, u64 ref_root, 2515846a3c2SQu Wenruo u64 owner, u64 offset, u64 reserved, int action, 252b06c4bf5SFilipe Manana struct btrfs_delayed_extent_op *extent_op); 253f64d5ca8SQu Wenruo int btrfs_add_delayed_qgroup_reserve(struct btrfs_fs_info *fs_info, 254f64d5ca8SQu Wenruo struct btrfs_trans_handle *trans, 255f64d5ca8SQu Wenruo u64 ref_root, u64 bytenr, u64 num_bytes); 25666d7e7f0SArne Jansen int btrfs_add_delayed_extent_op(struct btrfs_fs_info *fs_info, 25766d7e7f0SArne Jansen struct btrfs_trans_handle *trans, 2585d4f98a2SYan Zheng u64 bytenr, u64 num_bytes, 2595d4f98a2SYan Zheng struct btrfs_delayed_extent_op *extent_op); 260ae1e206bSJosef Bacik void btrfs_merge_delayed_refs(struct btrfs_trans_handle *trans, 261ae1e206bSJosef Bacik struct btrfs_fs_info *fs_info, 262ae1e206bSJosef Bacik struct btrfs_delayed_ref_root *delayed_refs, 263ae1e206bSJosef Bacik struct btrfs_delayed_ref_head *head); 26456bec294SChris Mason 2651887be66SChris Mason struct btrfs_delayed_ref_head * 2661887be66SChris Mason btrfs_find_delayed_ref_head(struct btrfs_trans_handle *trans, u64 bytenr); 267c3e69d58SChris Mason int btrfs_delayed_ref_lock(struct btrfs_trans_handle *trans, 268c3e69d58SChris Mason struct btrfs_delayed_ref_head *head); 269093486c4SMiao Xie static inline void btrfs_delayed_ref_unlock(struct btrfs_delayed_ref_head *head) 270093486c4SMiao Xie { 271093486c4SMiao Xie mutex_unlock(&head->mutex); 272093486c4SMiao Xie } 273093486c4SMiao Xie 274d7df2c79SJosef Bacik 275d7df2c79SJosef Bacik struct btrfs_delayed_ref_head * 276d7df2c79SJosef Bacik btrfs_select_ref_head(struct btrfs_trans_handle *trans); 27700f04b88SArne Jansen 278097b8a7cSJan Schmidt int btrfs_check_delayed_seq(struct btrfs_fs_info *fs_info, 279097b8a7cSJan Schmidt struct btrfs_delayed_ref_root *delayed_refs, 28000f04b88SArne Jansen u64 seq); 28100f04b88SArne Jansen 28200f04b88SArne Jansen /* 28356bec294SChris Mason * a node might live in a head or a regular ref, this lets you 28456bec294SChris Mason * test for the proper type to use. 28556bec294SChris Mason */ 28656bec294SChris Mason static int btrfs_delayed_ref_is_head(struct btrfs_delayed_ref_node *node) 28756bec294SChris Mason { 2885d4f98a2SYan Zheng return node->is_head; 28956bec294SChris Mason } 29056bec294SChris Mason 29156bec294SChris Mason /* 29256bec294SChris Mason * helper functions to cast a node into its container 29356bec294SChris Mason */ 2945d4f98a2SYan Zheng static inline struct btrfs_delayed_tree_ref * 2955d4f98a2SYan Zheng btrfs_delayed_node_to_tree_ref(struct btrfs_delayed_ref_node *node) 29656bec294SChris Mason { 29756bec294SChris Mason WARN_ON(btrfs_delayed_ref_is_head(node)); 2985d4f98a2SYan Zheng return container_of(node, struct btrfs_delayed_tree_ref, node); 2995d4f98a2SYan Zheng } 30056bec294SChris Mason 3015d4f98a2SYan Zheng static inline struct btrfs_delayed_data_ref * 3025d4f98a2SYan Zheng btrfs_delayed_node_to_data_ref(struct btrfs_delayed_ref_node *node) 3035d4f98a2SYan Zheng { 3045d4f98a2SYan Zheng WARN_ON(btrfs_delayed_ref_is_head(node)); 3055d4f98a2SYan Zheng return container_of(node, struct btrfs_delayed_data_ref, node); 30656bec294SChris Mason } 30756bec294SChris Mason 30856bec294SChris Mason static inline struct btrfs_delayed_ref_head * 30956bec294SChris Mason btrfs_delayed_node_to_head(struct btrfs_delayed_ref_node *node) 31056bec294SChris Mason { 31156bec294SChris Mason WARN_ON(!btrfs_delayed_ref_is_head(node)); 31256bec294SChris Mason return container_of(node, struct btrfs_delayed_ref_head, node); 31356bec294SChris Mason } 31456bec294SChris Mason #endif 315