1c1d7c514SDavid Sterba // SPDX-License-Identifier: GPL-2.0 25d4f98a2SYan Zheng /* 35d4f98a2SYan Zheng * Copyright (C) 2009 Oracle. All rights reserved. 45d4f98a2SYan Zheng */ 55d4f98a2SYan Zheng 65d4f98a2SYan Zheng #include <linux/sched.h> 75d4f98a2SYan Zheng #include <linux/pagemap.h> 85d4f98a2SYan Zheng #include <linux/writeback.h> 95d4f98a2SYan Zheng #include <linux/blkdev.h> 105d4f98a2SYan Zheng #include <linux/rbtree.h> 115a0e3ad6STejun Heo #include <linux/slab.h> 12726a3421SQu Wenruo #include <linux/error-injection.h> 135d4f98a2SYan Zheng #include "ctree.h" 145d4f98a2SYan Zheng #include "disk-io.h" 155d4f98a2SYan Zheng #include "transaction.h" 165d4f98a2SYan Zheng #include "volumes.h" 175d4f98a2SYan Zheng #include "locking.h" 185d4f98a2SYan Zheng #include "btrfs_inode.h" 195d4f98a2SYan Zheng #include "async-thread.h" 200af3d00bSJosef Bacik #include "free-space-cache.h" 21581bb050SLi Zefan #include "inode-map.h" 2262b99540SQu Wenruo #include "qgroup.h" 23cdccee99SLiu Bo #include "print-tree.h" 2486736342SJosef Bacik #include "delalloc-space.h" 25aac0023cSJosef Bacik #include "block-group.h" 2619b546d7SQu Wenruo #include "backref.h" 275d4f98a2SYan Zheng 285d4f98a2SYan Zheng /* 290c891389SQu Wenruo * Relocation overview 300c891389SQu Wenruo * 310c891389SQu Wenruo * [What does relocation do] 320c891389SQu Wenruo * 330c891389SQu Wenruo * The objective of relocation is to relocate all extents of the target block 340c891389SQu Wenruo * group to other block groups. 350c891389SQu Wenruo * This is utilized by resize (shrink only), profile converting, compacting 360c891389SQu Wenruo * space, or balance routine to spread chunks over devices. 370c891389SQu Wenruo * 380c891389SQu Wenruo * Before | After 390c891389SQu Wenruo * ------------------------------------------------------------------ 400c891389SQu Wenruo * BG A: 10 data extents | BG A: deleted 410c891389SQu Wenruo * BG B: 2 data extents | BG B: 10 data extents (2 old + 8 relocated) 420c891389SQu Wenruo * BG C: 1 extents | BG C: 3 data extents (1 old + 2 relocated) 430c891389SQu Wenruo * 440c891389SQu Wenruo * [How does relocation work] 450c891389SQu Wenruo * 460c891389SQu Wenruo * 1. Mark the target block group read-only 470c891389SQu Wenruo * New extents won't be allocated from the target block group. 480c891389SQu Wenruo * 490c891389SQu Wenruo * 2.1 Record each extent in the target block group 500c891389SQu Wenruo * To build a proper map of extents to be relocated. 510c891389SQu Wenruo * 520c891389SQu Wenruo * 2.2 Build data reloc tree and reloc trees 530c891389SQu Wenruo * Data reloc tree will contain an inode, recording all newly relocated 540c891389SQu Wenruo * data extents. 550c891389SQu Wenruo * There will be only one data reloc tree for one data block group. 560c891389SQu Wenruo * 570c891389SQu Wenruo * Reloc tree will be a special snapshot of its source tree, containing 580c891389SQu Wenruo * relocated tree blocks. 590c891389SQu Wenruo * Each tree referring to a tree block in target block group will get its 600c891389SQu Wenruo * reloc tree built. 610c891389SQu Wenruo * 620c891389SQu Wenruo * 2.3 Swap source tree with its corresponding reloc tree 630c891389SQu Wenruo * Each involved tree only refers to new extents after swap. 640c891389SQu Wenruo * 650c891389SQu Wenruo * 3. Cleanup reloc trees and data reloc tree. 660c891389SQu Wenruo * As old extents in the target block group are still referenced by reloc 670c891389SQu Wenruo * trees, we need to clean them up before really freeing the target block 680c891389SQu Wenruo * group. 690c891389SQu Wenruo * 700c891389SQu Wenruo * The main complexity is in steps 2.2 and 2.3. 710c891389SQu Wenruo * 720c891389SQu Wenruo * The entry point of relocation is relocate_block_group() function. 730c891389SQu Wenruo */ 740c891389SQu Wenruo 750c891389SQu Wenruo /* 765d4f98a2SYan Zheng * backref_node, mapping_node and tree_block start with this 775d4f98a2SYan Zheng */ 785d4f98a2SYan Zheng struct tree_entry { 795d4f98a2SYan Zheng struct rb_node rb_node; 805d4f98a2SYan Zheng u64 bytenr; 815d4f98a2SYan Zheng }; 825d4f98a2SYan Zheng 835d4f98a2SYan Zheng /* 845d4f98a2SYan Zheng * present a tree block in the backref cache 855d4f98a2SYan Zheng */ 865d4f98a2SYan Zheng struct backref_node { 875d4f98a2SYan Zheng struct rb_node rb_node; 885d4f98a2SYan Zheng u64 bytenr; 893fd0a558SYan, Zheng 903fd0a558SYan, Zheng u64 new_bytenr; 913fd0a558SYan, Zheng /* objectid of tree block owner, can be not uptodate */ 925d4f98a2SYan Zheng u64 owner; 933fd0a558SYan, Zheng /* link to pending, changed or detached list */ 943fd0a558SYan, Zheng struct list_head list; 955d4f98a2SYan Zheng /* list of upper level blocks reference this block */ 965d4f98a2SYan Zheng struct list_head upper; 975d4f98a2SYan Zheng /* list of child blocks in the cache */ 985d4f98a2SYan Zheng struct list_head lower; 995d4f98a2SYan Zheng /* NULL if this node is not tree root */ 1005d4f98a2SYan Zheng struct btrfs_root *root; 1015d4f98a2SYan Zheng /* extent buffer got by COW the block */ 1025d4f98a2SYan Zheng struct extent_buffer *eb; 1035d4f98a2SYan Zheng /* level of tree block */ 1045d4f98a2SYan Zheng unsigned int level:8; 1053fd0a558SYan, Zheng /* is the block in non-reference counted tree */ 1063fd0a558SYan, Zheng unsigned int cowonly:1; 1073fd0a558SYan, Zheng /* 1 if no child node in the cache */ 1085d4f98a2SYan Zheng unsigned int lowest:1; 1095d4f98a2SYan Zheng /* is the extent buffer locked */ 1105d4f98a2SYan Zheng unsigned int locked:1; 1115d4f98a2SYan Zheng /* has the block been processed */ 1125d4f98a2SYan Zheng unsigned int processed:1; 1135d4f98a2SYan Zheng /* have backrefs of this block been checked */ 1145d4f98a2SYan Zheng unsigned int checked:1; 1153fd0a558SYan, Zheng /* 1163fd0a558SYan, Zheng * 1 if corresponding block has been cowed but some upper 1173fd0a558SYan, Zheng * level block pointers may not point to the new location 1183fd0a558SYan, Zheng */ 1193fd0a558SYan, Zheng unsigned int pending:1; 1203fd0a558SYan, Zheng /* 1213fd0a558SYan, Zheng * 1 if the backref node isn't connected to any other 1223fd0a558SYan, Zheng * backref node. 1233fd0a558SYan, Zheng */ 1243fd0a558SYan, Zheng unsigned int detached:1; 1252433bea5SQu Wenruo 1262433bea5SQu Wenruo /* 1272433bea5SQu Wenruo * For generic purpose backref cache, where we only care if it's a reloc 1282433bea5SQu Wenruo * root, doesn't care the source subvolid. 1292433bea5SQu Wenruo */ 1302433bea5SQu Wenruo unsigned int is_reloc_root:1; 1315d4f98a2SYan Zheng }; 1325d4f98a2SYan Zheng 1335d4f98a2SYan Zheng /* 1345d4f98a2SYan Zheng * present a block pointer in the backref cache 1355d4f98a2SYan Zheng */ 1365d4f98a2SYan Zheng struct backref_edge { 1375d4f98a2SYan Zheng struct list_head list[2]; 1385d4f98a2SYan Zheng struct backref_node *node[2]; 1395d4f98a2SYan Zheng }; 1405d4f98a2SYan Zheng 1415d4f98a2SYan Zheng #define LOWER 0 1425d4f98a2SYan Zheng #define UPPER 1 1430647bf56SWang Shilong #define RELOCATION_RESERVED_NODES 256 1445d4f98a2SYan Zheng 1455d4f98a2SYan Zheng struct backref_cache { 1465d4f98a2SYan Zheng /* red black tree of all backref nodes in the cache */ 1475d4f98a2SYan Zheng struct rb_root rb_root; 1483fd0a558SYan, Zheng /* for passing backref nodes to btrfs_reloc_cow_block */ 1493fd0a558SYan, Zheng struct backref_node *path[BTRFS_MAX_LEVEL]; 1503fd0a558SYan, Zheng /* 1513fd0a558SYan, Zheng * list of blocks that have been cowed but some block 1523fd0a558SYan, Zheng * pointers in upper level blocks may not reflect the 1533fd0a558SYan, Zheng * new location 1543fd0a558SYan, Zheng */ 1555d4f98a2SYan Zheng struct list_head pending[BTRFS_MAX_LEVEL]; 1563fd0a558SYan, Zheng /* list of backref nodes with no child node */ 1573fd0a558SYan, Zheng struct list_head leaves; 1583fd0a558SYan, Zheng /* list of blocks that have been cowed in current transaction */ 1593fd0a558SYan, Zheng struct list_head changed; 1603fd0a558SYan, Zheng /* list of detached backref node. */ 1613fd0a558SYan, Zheng struct list_head detached; 1623fd0a558SYan, Zheng 1633fd0a558SYan, Zheng u64 last_trans; 1643fd0a558SYan, Zheng 1653fd0a558SYan, Zheng int nr_nodes; 1663fd0a558SYan, Zheng int nr_edges; 16784780289SQu Wenruo 16884780289SQu Wenruo /* The list of unchecked backref edges during backref cache build */ 16984780289SQu Wenruo struct list_head pending_edge; 17084780289SQu Wenruo 17184780289SQu Wenruo /* The list of useless backref nodes during backref cache build */ 17284780289SQu Wenruo struct list_head useless_node; 17333a0f1f7SQu Wenruo 17433a0f1f7SQu Wenruo struct btrfs_fs_info *fs_info; 1752433bea5SQu Wenruo 1762433bea5SQu Wenruo /* 1772433bea5SQu Wenruo * Whether this cache is for relocation 1782433bea5SQu Wenruo * 1792433bea5SQu Wenruo * Reloction backref cache require more info for reloc root compared 1802433bea5SQu Wenruo * to generic backref cache. 1812433bea5SQu Wenruo */ 1822433bea5SQu Wenruo unsigned int is_reloc; 1835d4f98a2SYan Zheng }; 1845d4f98a2SYan Zheng 1855d4f98a2SYan Zheng /* 1865d4f98a2SYan Zheng * map address of tree root to tree 1875d4f98a2SYan Zheng */ 1885d4f98a2SYan Zheng struct mapping_node { 1895d4f98a2SYan Zheng struct rb_node rb_node; 1905d4f98a2SYan Zheng u64 bytenr; 1915d4f98a2SYan Zheng void *data; 1925d4f98a2SYan Zheng }; 1935d4f98a2SYan Zheng 1945d4f98a2SYan Zheng struct mapping_tree { 1955d4f98a2SYan Zheng struct rb_root rb_root; 1965d4f98a2SYan Zheng spinlock_t lock; 1975d4f98a2SYan Zheng }; 1985d4f98a2SYan Zheng 1995d4f98a2SYan Zheng /* 2005d4f98a2SYan Zheng * present a tree block to process 2015d4f98a2SYan Zheng */ 2025d4f98a2SYan Zheng struct tree_block { 2035d4f98a2SYan Zheng struct rb_node rb_node; 2045d4f98a2SYan Zheng u64 bytenr; 2055d4f98a2SYan Zheng struct btrfs_key key; 2065d4f98a2SYan Zheng unsigned int level:8; 2075d4f98a2SYan Zheng unsigned int key_ready:1; 2085d4f98a2SYan Zheng }; 2095d4f98a2SYan Zheng 2100257bb82SYan, Zheng #define MAX_EXTENTS 128 2110257bb82SYan, Zheng 2120257bb82SYan, Zheng struct file_extent_cluster { 2130257bb82SYan, Zheng u64 start; 2140257bb82SYan, Zheng u64 end; 2150257bb82SYan, Zheng u64 boundary[MAX_EXTENTS]; 2160257bb82SYan, Zheng unsigned int nr; 2170257bb82SYan, Zheng }; 2180257bb82SYan, Zheng 2195d4f98a2SYan Zheng struct reloc_control { 2205d4f98a2SYan Zheng /* block group to relocate */ 22132da5386SDavid Sterba struct btrfs_block_group *block_group; 2225d4f98a2SYan Zheng /* extent tree */ 2235d4f98a2SYan Zheng struct btrfs_root *extent_root; 2245d4f98a2SYan Zheng /* inode for moving data */ 2255d4f98a2SYan Zheng struct inode *data_inode; 2263fd0a558SYan, Zheng 2273fd0a558SYan, Zheng struct btrfs_block_rsv *block_rsv; 2283fd0a558SYan, Zheng 2293fd0a558SYan, Zheng struct backref_cache backref_cache; 2303fd0a558SYan, Zheng 2313fd0a558SYan, Zheng struct file_extent_cluster cluster; 2325d4f98a2SYan Zheng /* tree blocks have been processed */ 2335d4f98a2SYan Zheng struct extent_io_tree processed_blocks; 2345d4f98a2SYan Zheng /* map start of tree root to corresponding reloc tree */ 2355d4f98a2SYan Zheng struct mapping_tree reloc_root_tree; 2365d4f98a2SYan Zheng /* list of reloc trees */ 2375d4f98a2SYan Zheng struct list_head reloc_roots; 238d2311e69SQu Wenruo /* list of subvolume trees that get relocated */ 239d2311e69SQu Wenruo struct list_head dirty_subvol_roots; 2403fd0a558SYan, Zheng /* size of metadata reservation for merging reloc trees */ 2413fd0a558SYan, Zheng u64 merging_rsv_size; 2423fd0a558SYan, Zheng /* size of relocated tree nodes */ 2433fd0a558SYan, Zheng u64 nodes_relocated; 2440647bf56SWang Shilong /* reserved size for block group relocation*/ 2450647bf56SWang Shilong u64 reserved_bytes; 2463fd0a558SYan, Zheng 2475d4f98a2SYan Zheng u64 search_start; 2485d4f98a2SYan Zheng u64 extents_found; 2493fd0a558SYan, Zheng 2503fd0a558SYan, Zheng unsigned int stage:8; 2513fd0a558SYan, Zheng unsigned int create_reloc_tree:1; 2523fd0a558SYan, Zheng unsigned int merge_reloc_tree:1; 2535d4f98a2SYan Zheng unsigned int found_file_extent:1; 2545d4f98a2SYan Zheng }; 2555d4f98a2SYan Zheng 2565d4f98a2SYan Zheng /* stages of data relocation */ 2575d4f98a2SYan Zheng #define MOVE_DATA_EXTENTS 0 2585d4f98a2SYan Zheng #define UPDATE_DATA_PTRS 1 2595d4f98a2SYan Zheng 2603fd0a558SYan, Zheng static void remove_backref_node(struct backref_cache *cache, 2613fd0a558SYan, Zheng struct backref_node *node); 2629569cc20SQu Wenruo 2639569cc20SQu Wenruo static void mark_block_processed(struct reloc_control *rc, 2649569cc20SQu Wenruo struct backref_node *node) 2659569cc20SQu Wenruo { 2669569cc20SQu Wenruo u32 blocksize; 2679569cc20SQu Wenruo 2689569cc20SQu Wenruo if (node->level == 0 || 2699569cc20SQu Wenruo in_range(node->bytenr, rc->block_group->start, 2709569cc20SQu Wenruo rc->block_group->length)) { 2719569cc20SQu Wenruo blocksize = rc->extent_root->fs_info->nodesize; 2729569cc20SQu Wenruo set_extent_bits(&rc->processed_blocks, node->bytenr, 2739569cc20SQu Wenruo node->bytenr + blocksize - 1, EXTENT_DIRTY); 2749569cc20SQu Wenruo } 2759569cc20SQu Wenruo node->processed = 1; 2769569cc20SQu Wenruo } 2779569cc20SQu Wenruo 2785d4f98a2SYan Zheng 2795d4f98a2SYan Zheng static void mapping_tree_init(struct mapping_tree *tree) 2805d4f98a2SYan Zheng { 2816bef4d31SEric Paris tree->rb_root = RB_ROOT; 2825d4f98a2SYan Zheng spin_lock_init(&tree->lock); 2835d4f98a2SYan Zheng } 2845d4f98a2SYan Zheng 28533a0f1f7SQu Wenruo static void backref_cache_init(struct btrfs_fs_info *fs_info, 2862433bea5SQu Wenruo struct backref_cache *cache, int is_reloc) 2875d4f98a2SYan Zheng { 2885d4f98a2SYan Zheng int i; 2896bef4d31SEric Paris cache->rb_root = RB_ROOT; 2905d4f98a2SYan Zheng for (i = 0; i < BTRFS_MAX_LEVEL; i++) 2915d4f98a2SYan Zheng INIT_LIST_HEAD(&cache->pending[i]); 2923fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->changed); 2933fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->detached); 2943fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->leaves); 29584780289SQu Wenruo INIT_LIST_HEAD(&cache->pending_edge); 29684780289SQu Wenruo INIT_LIST_HEAD(&cache->useless_node); 29733a0f1f7SQu Wenruo cache->fs_info = fs_info; 2982433bea5SQu Wenruo cache->is_reloc = is_reloc; 2995d4f98a2SYan Zheng } 3005d4f98a2SYan Zheng 3013fd0a558SYan, Zheng static void backref_cache_cleanup(struct backref_cache *cache) 3025d4f98a2SYan Zheng { 3033fd0a558SYan, Zheng struct backref_node *node; 3043fd0a558SYan, Zheng int i; 3053fd0a558SYan, Zheng 3063fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 3073fd0a558SYan, Zheng node = list_entry(cache->detached.next, 3083fd0a558SYan, Zheng struct backref_node, list); 3093fd0a558SYan, Zheng remove_backref_node(cache, node); 3103fd0a558SYan, Zheng } 3113fd0a558SYan, Zheng 3123fd0a558SYan, Zheng while (!list_empty(&cache->leaves)) { 3133fd0a558SYan, Zheng node = list_entry(cache->leaves.next, 3143fd0a558SYan, Zheng struct backref_node, lower); 3153fd0a558SYan, Zheng remove_backref_node(cache, node); 3163fd0a558SYan, Zheng } 3173fd0a558SYan, Zheng 3183fd0a558SYan, Zheng cache->last_trans = 0; 3193fd0a558SYan, Zheng 3203fd0a558SYan, Zheng for (i = 0; i < BTRFS_MAX_LEVEL; i++) 321f4907095SLiu Bo ASSERT(list_empty(&cache->pending[i])); 32284780289SQu Wenruo ASSERT(list_empty(&cache->pending_edge)); 32384780289SQu Wenruo ASSERT(list_empty(&cache->useless_node)); 324f4907095SLiu Bo ASSERT(list_empty(&cache->changed)); 325f4907095SLiu Bo ASSERT(list_empty(&cache->detached)); 326f4907095SLiu Bo ASSERT(RB_EMPTY_ROOT(&cache->rb_root)); 327f4907095SLiu Bo ASSERT(!cache->nr_nodes); 328f4907095SLiu Bo ASSERT(!cache->nr_edges); 3293fd0a558SYan, Zheng } 3303fd0a558SYan, Zheng 3313fd0a558SYan, Zheng static struct backref_node *alloc_backref_node(struct backref_cache *cache) 3323fd0a558SYan, Zheng { 3333fd0a558SYan, Zheng struct backref_node *node; 3343fd0a558SYan, Zheng 3353fd0a558SYan, Zheng node = kzalloc(sizeof(*node), GFP_NOFS); 3363fd0a558SYan, Zheng if (node) { 3373fd0a558SYan, Zheng INIT_LIST_HEAD(&node->list); 3385d4f98a2SYan Zheng INIT_LIST_HEAD(&node->upper); 3395d4f98a2SYan Zheng INIT_LIST_HEAD(&node->lower); 3405d4f98a2SYan Zheng RB_CLEAR_NODE(&node->rb_node); 3413fd0a558SYan, Zheng cache->nr_nodes++; 3423fd0a558SYan, Zheng } 3433fd0a558SYan, Zheng return node; 3443fd0a558SYan, Zheng } 3453fd0a558SYan, Zheng 3463fd0a558SYan, Zheng static void free_backref_node(struct backref_cache *cache, 3473fd0a558SYan, Zheng struct backref_node *node) 3483fd0a558SYan, Zheng { 3493fd0a558SYan, Zheng if (node) { 3503fd0a558SYan, Zheng cache->nr_nodes--; 35100246528SJosef Bacik btrfs_put_root(node->root); 3523fd0a558SYan, Zheng kfree(node); 3533fd0a558SYan, Zheng } 3543fd0a558SYan, Zheng } 3553fd0a558SYan, Zheng 3563fd0a558SYan, Zheng static struct backref_edge *alloc_backref_edge(struct backref_cache *cache) 3573fd0a558SYan, Zheng { 3583fd0a558SYan, Zheng struct backref_edge *edge; 3593fd0a558SYan, Zheng 3603fd0a558SYan, Zheng edge = kzalloc(sizeof(*edge), GFP_NOFS); 3613fd0a558SYan, Zheng if (edge) 3623fd0a558SYan, Zheng cache->nr_edges++; 3633fd0a558SYan, Zheng return edge; 3643fd0a558SYan, Zheng } 3653fd0a558SYan, Zheng 3663fd0a558SYan, Zheng static void free_backref_edge(struct backref_cache *cache, 3673fd0a558SYan, Zheng struct backref_edge *edge) 3683fd0a558SYan, Zheng { 3693fd0a558SYan, Zheng if (edge) { 3703fd0a558SYan, Zheng cache->nr_edges--; 3713fd0a558SYan, Zheng kfree(edge); 3723fd0a558SYan, Zheng } 3735d4f98a2SYan Zheng } 3745d4f98a2SYan Zheng 3755d4f98a2SYan Zheng static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr, 3765d4f98a2SYan Zheng struct rb_node *node) 3775d4f98a2SYan Zheng { 3785d4f98a2SYan Zheng struct rb_node **p = &root->rb_node; 3795d4f98a2SYan Zheng struct rb_node *parent = NULL; 3805d4f98a2SYan Zheng struct tree_entry *entry; 3815d4f98a2SYan Zheng 3825d4f98a2SYan Zheng while (*p) { 3835d4f98a2SYan Zheng parent = *p; 3845d4f98a2SYan Zheng entry = rb_entry(parent, struct tree_entry, rb_node); 3855d4f98a2SYan Zheng 3865d4f98a2SYan Zheng if (bytenr < entry->bytenr) 3875d4f98a2SYan Zheng p = &(*p)->rb_left; 3885d4f98a2SYan Zheng else if (bytenr > entry->bytenr) 3895d4f98a2SYan Zheng p = &(*p)->rb_right; 3905d4f98a2SYan Zheng else 3915d4f98a2SYan Zheng return parent; 3925d4f98a2SYan Zheng } 3935d4f98a2SYan Zheng 3945d4f98a2SYan Zheng rb_link_node(node, parent, p); 3955d4f98a2SYan Zheng rb_insert_color(node, root); 3965d4f98a2SYan Zheng return NULL; 3975d4f98a2SYan Zheng } 3985d4f98a2SYan Zheng 3995d4f98a2SYan Zheng static struct rb_node *tree_search(struct rb_root *root, u64 bytenr) 4005d4f98a2SYan Zheng { 4015d4f98a2SYan Zheng struct rb_node *n = root->rb_node; 4025d4f98a2SYan Zheng struct tree_entry *entry; 4035d4f98a2SYan Zheng 4045d4f98a2SYan Zheng while (n) { 4055d4f98a2SYan Zheng entry = rb_entry(n, struct tree_entry, rb_node); 4065d4f98a2SYan Zheng 4075d4f98a2SYan Zheng if (bytenr < entry->bytenr) 4085d4f98a2SYan Zheng n = n->rb_left; 4095d4f98a2SYan Zheng else if (bytenr > entry->bytenr) 4105d4f98a2SYan Zheng n = n->rb_right; 4115d4f98a2SYan Zheng else 4125d4f98a2SYan Zheng return n; 4135d4f98a2SYan Zheng } 4145d4f98a2SYan Zheng return NULL; 4155d4f98a2SYan Zheng } 4165d4f98a2SYan Zheng 41748a3b636SEric Sandeen static void backref_tree_panic(struct rb_node *rb_node, int errno, u64 bytenr) 41843c04fb1SJeff Mahoney { 41943c04fb1SJeff Mahoney 42043c04fb1SJeff Mahoney struct btrfs_fs_info *fs_info = NULL; 42143c04fb1SJeff Mahoney struct backref_node *bnode = rb_entry(rb_node, struct backref_node, 42243c04fb1SJeff Mahoney rb_node); 42343c04fb1SJeff Mahoney if (bnode->root) 42443c04fb1SJeff Mahoney fs_info = bnode->root->fs_info; 4255d163e0eSJeff Mahoney btrfs_panic(fs_info, errno, 4265d163e0eSJeff Mahoney "Inconsistency in backref cache found at offset %llu", 4275d163e0eSJeff Mahoney bytenr); 42843c04fb1SJeff Mahoney } 42943c04fb1SJeff Mahoney 4305d4f98a2SYan Zheng /* 4315d4f98a2SYan Zheng * walk up backref nodes until reach node presents tree root 4325d4f98a2SYan Zheng */ 4335d4f98a2SYan Zheng static struct backref_node *walk_up_backref(struct backref_node *node, 4345d4f98a2SYan Zheng struct backref_edge *edges[], 4355d4f98a2SYan Zheng int *index) 4365d4f98a2SYan Zheng { 4375d4f98a2SYan Zheng struct backref_edge *edge; 4385d4f98a2SYan Zheng int idx = *index; 4395d4f98a2SYan Zheng 4405d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 4415d4f98a2SYan Zheng edge = list_entry(node->upper.next, 4425d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 4435d4f98a2SYan Zheng edges[idx++] = edge; 4445d4f98a2SYan Zheng node = edge->node[UPPER]; 4455d4f98a2SYan Zheng } 4463fd0a558SYan, Zheng BUG_ON(node->detached); 4475d4f98a2SYan Zheng *index = idx; 4485d4f98a2SYan Zheng return node; 4495d4f98a2SYan Zheng } 4505d4f98a2SYan Zheng 4515d4f98a2SYan Zheng /* 4525d4f98a2SYan Zheng * walk down backref nodes to find start of next reference path 4535d4f98a2SYan Zheng */ 4545d4f98a2SYan Zheng static struct backref_node *walk_down_backref(struct backref_edge *edges[], 4555d4f98a2SYan Zheng int *index) 4565d4f98a2SYan Zheng { 4575d4f98a2SYan Zheng struct backref_edge *edge; 4585d4f98a2SYan Zheng struct backref_node *lower; 4595d4f98a2SYan Zheng int idx = *index; 4605d4f98a2SYan Zheng 4615d4f98a2SYan Zheng while (idx > 0) { 4625d4f98a2SYan Zheng edge = edges[idx - 1]; 4635d4f98a2SYan Zheng lower = edge->node[LOWER]; 4645d4f98a2SYan Zheng if (list_is_last(&edge->list[LOWER], &lower->upper)) { 4655d4f98a2SYan Zheng idx--; 4665d4f98a2SYan Zheng continue; 4675d4f98a2SYan Zheng } 4685d4f98a2SYan Zheng edge = list_entry(edge->list[LOWER].next, 4695d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 4705d4f98a2SYan Zheng edges[idx - 1] = edge; 4715d4f98a2SYan Zheng *index = idx; 4725d4f98a2SYan Zheng return edge->node[UPPER]; 4735d4f98a2SYan Zheng } 4745d4f98a2SYan Zheng *index = 0; 4755d4f98a2SYan Zheng return NULL; 4765d4f98a2SYan Zheng } 4775d4f98a2SYan Zheng 4783fd0a558SYan, Zheng static void unlock_node_buffer(struct backref_node *node) 4795d4f98a2SYan Zheng { 4805d4f98a2SYan Zheng if (node->locked) { 4815d4f98a2SYan Zheng btrfs_tree_unlock(node->eb); 4825d4f98a2SYan Zheng node->locked = 0; 4835d4f98a2SYan Zheng } 4843fd0a558SYan, Zheng } 4853fd0a558SYan, Zheng 4863fd0a558SYan, Zheng static void drop_node_buffer(struct backref_node *node) 4873fd0a558SYan, Zheng { 4883fd0a558SYan, Zheng if (node->eb) { 4893fd0a558SYan, Zheng unlock_node_buffer(node); 4905d4f98a2SYan Zheng free_extent_buffer(node->eb); 4915d4f98a2SYan Zheng node->eb = NULL; 4925d4f98a2SYan Zheng } 4935d4f98a2SYan Zheng } 4945d4f98a2SYan Zheng 4955d4f98a2SYan Zheng static void drop_backref_node(struct backref_cache *tree, 4965d4f98a2SYan Zheng struct backref_node *node) 4975d4f98a2SYan Zheng { 4985d4f98a2SYan Zheng BUG_ON(!list_empty(&node->upper)); 4995d4f98a2SYan Zheng 5005d4f98a2SYan Zheng drop_node_buffer(node); 5013fd0a558SYan, Zheng list_del(&node->list); 5025d4f98a2SYan Zheng list_del(&node->lower); 5033fd0a558SYan, Zheng if (!RB_EMPTY_NODE(&node->rb_node)) 5045d4f98a2SYan Zheng rb_erase(&node->rb_node, &tree->rb_root); 5053fd0a558SYan, Zheng free_backref_node(tree, node); 5065d4f98a2SYan Zheng } 5075d4f98a2SYan Zheng 5085d4f98a2SYan Zheng /* 5095d4f98a2SYan Zheng * remove a backref node from the backref cache 5105d4f98a2SYan Zheng */ 5115d4f98a2SYan Zheng static void remove_backref_node(struct backref_cache *cache, 5125d4f98a2SYan Zheng struct backref_node *node) 5135d4f98a2SYan Zheng { 5145d4f98a2SYan Zheng struct backref_node *upper; 5155d4f98a2SYan Zheng struct backref_edge *edge; 5165d4f98a2SYan Zheng 5175d4f98a2SYan Zheng if (!node) 5185d4f98a2SYan Zheng return; 5195d4f98a2SYan Zheng 5203fd0a558SYan, Zheng BUG_ON(!node->lowest && !node->detached); 5215d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 5225d4f98a2SYan Zheng edge = list_entry(node->upper.next, struct backref_edge, 5235d4f98a2SYan Zheng list[LOWER]); 5245d4f98a2SYan Zheng upper = edge->node[UPPER]; 5255d4f98a2SYan Zheng list_del(&edge->list[LOWER]); 5265d4f98a2SYan Zheng list_del(&edge->list[UPPER]); 5273fd0a558SYan, Zheng free_backref_edge(cache, edge); 5283fd0a558SYan, Zheng 5293fd0a558SYan, Zheng if (RB_EMPTY_NODE(&upper->rb_node)) { 5303fd0a558SYan, Zheng BUG_ON(!list_empty(&node->upper)); 5313fd0a558SYan, Zheng drop_backref_node(cache, node); 5323fd0a558SYan, Zheng node = upper; 5333fd0a558SYan, Zheng node->lowest = 1; 5343fd0a558SYan, Zheng continue; 5353fd0a558SYan, Zheng } 5365d4f98a2SYan Zheng /* 5373fd0a558SYan, Zheng * add the node to leaf node list if no other 5385d4f98a2SYan Zheng * child block cached. 5395d4f98a2SYan Zheng */ 5405d4f98a2SYan Zheng if (list_empty(&upper->lower)) { 5413fd0a558SYan, Zheng list_add_tail(&upper->lower, &cache->leaves); 5425d4f98a2SYan Zheng upper->lowest = 1; 5435d4f98a2SYan Zheng } 5445d4f98a2SYan Zheng } 5453fd0a558SYan, Zheng 5465d4f98a2SYan Zheng drop_backref_node(cache, node); 5475d4f98a2SYan Zheng } 5485d4f98a2SYan Zheng 5493fd0a558SYan, Zheng static void update_backref_node(struct backref_cache *cache, 5503fd0a558SYan, Zheng struct backref_node *node, u64 bytenr) 5513fd0a558SYan, Zheng { 5523fd0a558SYan, Zheng struct rb_node *rb_node; 5533fd0a558SYan, Zheng rb_erase(&node->rb_node, &cache->rb_root); 5543fd0a558SYan, Zheng node->bytenr = bytenr; 5553fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node); 55643c04fb1SJeff Mahoney if (rb_node) 55743c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, bytenr); 5583fd0a558SYan, Zheng } 5593fd0a558SYan, Zheng 5603fd0a558SYan, Zheng /* 5613fd0a558SYan, Zheng * update backref cache after a transaction commit 5623fd0a558SYan, Zheng */ 5633fd0a558SYan, Zheng static int update_backref_cache(struct btrfs_trans_handle *trans, 5643fd0a558SYan, Zheng struct backref_cache *cache) 5653fd0a558SYan, Zheng { 5663fd0a558SYan, Zheng struct backref_node *node; 5673fd0a558SYan, Zheng int level = 0; 5683fd0a558SYan, Zheng 5693fd0a558SYan, Zheng if (cache->last_trans == 0) { 5703fd0a558SYan, Zheng cache->last_trans = trans->transid; 5713fd0a558SYan, Zheng return 0; 5723fd0a558SYan, Zheng } 5733fd0a558SYan, Zheng 5743fd0a558SYan, Zheng if (cache->last_trans == trans->transid) 5753fd0a558SYan, Zheng return 0; 5763fd0a558SYan, Zheng 5773fd0a558SYan, Zheng /* 5783fd0a558SYan, Zheng * detached nodes are used to avoid unnecessary backref 5793fd0a558SYan, Zheng * lookup. transaction commit changes the extent tree. 5803fd0a558SYan, Zheng * so the detached nodes are no longer useful. 5813fd0a558SYan, Zheng */ 5823fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 5833fd0a558SYan, Zheng node = list_entry(cache->detached.next, 5843fd0a558SYan, Zheng struct backref_node, list); 5853fd0a558SYan, Zheng remove_backref_node(cache, node); 5863fd0a558SYan, Zheng } 5873fd0a558SYan, Zheng 5883fd0a558SYan, Zheng while (!list_empty(&cache->changed)) { 5893fd0a558SYan, Zheng node = list_entry(cache->changed.next, 5903fd0a558SYan, Zheng struct backref_node, list); 5913fd0a558SYan, Zheng list_del_init(&node->list); 5923fd0a558SYan, Zheng BUG_ON(node->pending); 5933fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 5943fd0a558SYan, Zheng } 5953fd0a558SYan, Zheng 5963fd0a558SYan, Zheng /* 5973fd0a558SYan, Zheng * some nodes can be left in the pending list if there were 5983fd0a558SYan, Zheng * errors during processing the pending nodes. 5993fd0a558SYan, Zheng */ 6003fd0a558SYan, Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 6013fd0a558SYan, Zheng list_for_each_entry(node, &cache->pending[level], list) { 6023fd0a558SYan, Zheng BUG_ON(!node->pending); 6033fd0a558SYan, Zheng if (node->bytenr == node->new_bytenr) 6043fd0a558SYan, Zheng continue; 6053fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 6063fd0a558SYan, Zheng } 6073fd0a558SYan, Zheng } 6083fd0a558SYan, Zheng 6093fd0a558SYan, Zheng cache->last_trans = 0; 6103fd0a558SYan, Zheng return 1; 6113fd0a558SYan, Zheng } 6123fd0a558SYan, Zheng 6136282675eSQu Wenruo static bool reloc_root_is_dead(struct btrfs_root *root) 6146282675eSQu Wenruo { 6156282675eSQu Wenruo /* 6166282675eSQu Wenruo * Pair with set_bit/clear_bit in clean_dirty_subvols and 6176282675eSQu Wenruo * btrfs_update_reloc_root. We need to see the updated bit before 6186282675eSQu Wenruo * trying to access reloc_root 6196282675eSQu Wenruo */ 6206282675eSQu Wenruo smp_rmb(); 6216282675eSQu Wenruo if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)) 6226282675eSQu Wenruo return true; 6236282675eSQu Wenruo return false; 6246282675eSQu Wenruo } 6256282675eSQu Wenruo 6266282675eSQu Wenruo /* 6276282675eSQu Wenruo * Check if this subvolume tree has valid reloc tree. 6286282675eSQu Wenruo * 6296282675eSQu Wenruo * Reloc tree after swap is considered dead, thus not considered as valid. 6306282675eSQu Wenruo * This is enough for most callers, as they don't distinguish dead reloc root 6316282675eSQu Wenruo * from no reloc root. But should_ignore_root() below is a special case. 6326282675eSQu Wenruo */ 6336282675eSQu Wenruo static bool have_reloc_root(struct btrfs_root *root) 6346282675eSQu Wenruo { 6356282675eSQu Wenruo if (reloc_root_is_dead(root)) 6366282675eSQu Wenruo return false; 6376282675eSQu Wenruo if (!root->reloc_root) 6386282675eSQu Wenruo return false; 6396282675eSQu Wenruo return true; 6406282675eSQu Wenruo } 641f2a97a9dSDavid Sterba 6423fd0a558SYan, Zheng static int should_ignore_root(struct btrfs_root *root) 6433fd0a558SYan, Zheng { 6443fd0a558SYan, Zheng struct btrfs_root *reloc_root; 6453fd0a558SYan, Zheng 64627cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 6473fd0a558SYan, Zheng return 0; 6483fd0a558SYan, Zheng 6496282675eSQu Wenruo /* This root has been merged with its reloc tree, we can ignore it */ 6506282675eSQu Wenruo if (reloc_root_is_dead(root)) 6516282675eSQu Wenruo return 1; 6526282675eSQu Wenruo 6533fd0a558SYan, Zheng reloc_root = root->reloc_root; 6543fd0a558SYan, Zheng if (!reloc_root) 6553fd0a558SYan, Zheng return 0; 6563fd0a558SYan, Zheng 6574d4225fcSJosef Bacik if (btrfs_header_generation(reloc_root->commit_root) == 6584d4225fcSJosef Bacik root->fs_info->running_transaction->transid) 6593fd0a558SYan, Zheng return 0; 6603fd0a558SYan, Zheng /* 6613fd0a558SYan, Zheng * if there is reloc tree and it was created in previous 6623fd0a558SYan, Zheng * transaction backref lookup can find the reloc tree, 6633fd0a558SYan, Zheng * so backref node for the fs tree root is useless for 6643fd0a558SYan, Zheng * relocation. 6653fd0a558SYan, Zheng */ 6663fd0a558SYan, Zheng return 1; 6673fd0a558SYan, Zheng } 6685d4f98a2SYan Zheng /* 6695d4f98a2SYan Zheng * find reloc tree by address of tree root 6705d4f98a2SYan Zheng */ 6712433bea5SQu Wenruo struct btrfs_root *find_reloc_root(struct btrfs_fs_info *fs_info, u64 bytenr) 6725d4f98a2SYan Zheng { 6732433bea5SQu Wenruo struct reloc_control *rc = fs_info->reloc_ctl; 6745d4f98a2SYan Zheng struct rb_node *rb_node; 6755d4f98a2SYan Zheng struct mapping_node *node; 6765d4f98a2SYan Zheng struct btrfs_root *root = NULL; 6775d4f98a2SYan Zheng 6782433bea5SQu Wenruo ASSERT(rc); 6795d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 6805d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr); 6815d4f98a2SYan Zheng if (rb_node) { 6825d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 6835d4f98a2SYan Zheng root = (struct btrfs_root *)node->data; 6845d4f98a2SYan Zheng } 6855d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 68600246528SJosef Bacik return btrfs_grab_root(root); 6875d4f98a2SYan Zheng } 6885d4f98a2SYan Zheng 6895d4f98a2SYan Zheng static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info, 6905d4f98a2SYan Zheng u64 root_objectid) 6915d4f98a2SYan Zheng { 6925d4f98a2SYan Zheng struct btrfs_key key; 6935d4f98a2SYan Zheng 6945d4f98a2SYan Zheng key.objectid = root_objectid; 6955d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 6965d4f98a2SYan Zheng key.offset = (u64)-1; 6975d4f98a2SYan Zheng 698bc44d7c4SJosef Bacik return btrfs_get_fs_root(fs_info, &key, false); 6995d4f98a2SYan Zheng } 7005d4f98a2SYan Zheng 7015d4f98a2SYan Zheng /* 7025d4f98a2SYan Zheng * build backref tree for a given tree block. root of the backref tree 7035d4f98a2SYan Zheng * corresponds the tree block, leaves of the backref tree correspond 7045d4f98a2SYan Zheng * roots of b-trees that reference the tree block. 7055d4f98a2SYan Zheng * 7065d4f98a2SYan Zheng * the basic idea of this function is check backrefs of a given block 70701327610SNicholas D Steeves * to find upper level blocks that reference the block, and then check 70801327610SNicholas D Steeves * backrefs of these upper level blocks recursively. the recursion stop 7095d4f98a2SYan Zheng * when tree root is reached or backrefs for the block is cached. 7105d4f98a2SYan Zheng * 7115d4f98a2SYan Zheng * NOTE: if we find backrefs for a block are cached, we know backrefs 7125d4f98a2SYan Zheng * for all upper level blocks that directly/indirectly reference the 7135d4f98a2SYan Zheng * block are also cached. 7145d4f98a2SYan Zheng */ 7153fd0a558SYan, Zheng static noinline_for_stack 7163fd0a558SYan, Zheng struct backref_node *build_backref_tree(struct reloc_control *rc, 7175d4f98a2SYan Zheng struct btrfs_key *node_key, 7185d4f98a2SYan Zheng int level, u64 bytenr) 7195d4f98a2SYan Zheng { 72071f572a9SQu Wenruo struct btrfs_backref_iter *iter; 7213fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 7222433bea5SQu Wenruo struct btrfs_fs_info *fs_info = cache->fs_info; 72371f572a9SQu Wenruo /* For searching parent of TREE_BLOCK_REF */ 72471f572a9SQu Wenruo struct btrfs_path *path; 7255d4f98a2SYan Zheng struct btrfs_root *root; 7265d4f98a2SYan Zheng struct backref_node *cur; 7275d4f98a2SYan Zheng struct backref_node *upper; 7285d4f98a2SYan Zheng struct backref_node *lower; 7295d4f98a2SYan Zheng struct backref_node *node = NULL; 7305d4f98a2SYan Zheng struct backref_node *exist = NULL; 7315d4f98a2SYan Zheng struct backref_edge *edge; 7325d4f98a2SYan Zheng struct rb_node *rb_node; 7333fd0a558SYan, Zheng int cowonly; 7345d4f98a2SYan Zheng int ret; 7355d4f98a2SYan Zheng int err = 0; 736b6c60c80SJosef Bacik bool need_check = true; 7375d4f98a2SYan Zheng 73871f572a9SQu Wenruo iter = btrfs_backref_iter_alloc(rc->extent_root->fs_info, GFP_NOFS); 73971f572a9SQu Wenruo if (!iter) 74071f572a9SQu Wenruo return ERR_PTR(-ENOMEM); 74171f572a9SQu Wenruo path = btrfs_alloc_path(); 74271f572a9SQu Wenruo if (!path) { 7435d4f98a2SYan Zheng err = -ENOMEM; 7445d4f98a2SYan Zheng goto out; 7455d4f98a2SYan Zheng } 7465d4f98a2SYan Zheng 7473fd0a558SYan, Zheng node = alloc_backref_node(cache); 7485d4f98a2SYan Zheng if (!node) { 7495d4f98a2SYan Zheng err = -ENOMEM; 7505d4f98a2SYan Zheng goto out; 7515d4f98a2SYan Zheng } 7525d4f98a2SYan Zheng 7535d4f98a2SYan Zheng node->bytenr = bytenr; 7545d4f98a2SYan Zheng node->level = level; 7555d4f98a2SYan Zheng node->lowest = 1; 7565d4f98a2SYan Zheng cur = node; 7575d4f98a2SYan Zheng again: 75871f572a9SQu Wenruo ret = btrfs_backref_iter_start(iter, cur->bytenr); 7595d4f98a2SYan Zheng if (ret < 0) { 7605d4f98a2SYan Zheng err = ret; 7615d4f98a2SYan Zheng goto out; 7625d4f98a2SYan Zheng } 7635d4f98a2SYan Zheng 76471f572a9SQu Wenruo /* 76571f572a9SQu Wenruo * We skip the first btrfs_tree_block_info, as we don't use the key 76671f572a9SQu Wenruo * stored in it, but fetch it from the tree block 76771f572a9SQu Wenruo */ 76871f572a9SQu Wenruo if (btrfs_backref_has_tree_block_info(iter)) { 76971f572a9SQu Wenruo ret = btrfs_backref_iter_next(iter); 77071f572a9SQu Wenruo if (ret < 0) { 77171f572a9SQu Wenruo err = ret; 77271f572a9SQu Wenruo goto out; 77371f572a9SQu Wenruo } 77471f572a9SQu Wenruo /* No extra backref? This means the tree block is corrupted */ 77571f572a9SQu Wenruo if (ret > 0) { 77671f572a9SQu Wenruo err = -EUCLEAN; 77771f572a9SQu Wenruo goto out; 77871f572a9SQu Wenruo } 77971f572a9SQu Wenruo } 7805d4f98a2SYan Zheng WARN_ON(cur->checked); 7815d4f98a2SYan Zheng if (!list_empty(&cur->upper)) { 7825d4f98a2SYan Zheng /* 78370f23fd6SJustin P. Mattock * the backref was added previously when processing 7845d4f98a2SYan Zheng * backref of type BTRFS_TREE_BLOCK_REF_KEY 7855d4f98a2SYan Zheng */ 78675bfb9afSJosef Bacik ASSERT(list_is_singular(&cur->upper)); 7875d4f98a2SYan Zheng edge = list_entry(cur->upper.next, struct backref_edge, 7885d4f98a2SYan Zheng list[LOWER]); 78975bfb9afSJosef Bacik ASSERT(list_empty(&edge->list[UPPER])); 7905d4f98a2SYan Zheng exist = edge->node[UPPER]; 7915d4f98a2SYan Zheng /* 7925d4f98a2SYan Zheng * add the upper level block to pending list if we need 7935d4f98a2SYan Zheng * check its backrefs 7945d4f98a2SYan Zheng */ 7955d4f98a2SYan Zheng if (!exist->checked) 79684780289SQu Wenruo list_add_tail(&edge->list[UPPER], &cache->pending_edge); 7975d4f98a2SYan Zheng } else { 7985d4f98a2SYan Zheng exist = NULL; 7995d4f98a2SYan Zheng } 8005d4f98a2SYan Zheng 80171f572a9SQu Wenruo for (; ret == 0; ret = btrfs_backref_iter_next(iter)) { 80271f572a9SQu Wenruo struct extent_buffer *eb; 80371f572a9SQu Wenruo struct btrfs_key key; 8043de28d57SLiu Bo int type; 80571f572a9SQu Wenruo 80671f572a9SQu Wenruo cond_resched(); 80771f572a9SQu Wenruo eb = btrfs_backref_get_eb(iter); 80871f572a9SQu Wenruo 80971f572a9SQu Wenruo key.objectid = iter->bytenr; 81071f572a9SQu Wenruo if (btrfs_backref_iter_is_inline_ref(iter)) { 81171f572a9SQu Wenruo struct btrfs_extent_inline_ref *iref; 81271f572a9SQu Wenruo 81371f572a9SQu Wenruo /* update key for inline back ref */ 81471f572a9SQu Wenruo iref = (struct btrfs_extent_inline_ref *) 81571f572a9SQu Wenruo ((unsigned long)iter->cur_ptr); 8163de28d57SLiu Bo type = btrfs_get_extent_inline_ref_type(eb, iref, 8173de28d57SLiu Bo BTRFS_REF_TYPE_BLOCK); 8183de28d57SLiu Bo if (type == BTRFS_REF_TYPE_INVALID) { 819af431dcbSSu Yue err = -EUCLEAN; 8203de28d57SLiu Bo goto out; 8213de28d57SLiu Bo } 8223de28d57SLiu Bo key.type = type; 8235d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 82471f572a9SQu Wenruo } else { 82571f572a9SQu Wenruo key.type = iter->cur_key.type; 82671f572a9SQu Wenruo key.offset = iter->cur_key.offset; 8275d4f98a2SYan Zheng } 8285d4f98a2SYan Zheng 829fa6ac715SQu Wenruo /* 830fa6ac715SQu Wenruo * Parent node found and matches current inline ref, no need to 831fa6ac715SQu Wenruo * rebuild this node for this inline ref. 832fa6ac715SQu Wenruo */ 8335d4f98a2SYan Zheng if (exist && 8345d4f98a2SYan Zheng ((key.type == BTRFS_TREE_BLOCK_REF_KEY && 8355d4f98a2SYan Zheng exist->owner == key.offset) || 8365d4f98a2SYan Zheng (key.type == BTRFS_SHARED_BLOCK_REF_KEY && 8375d4f98a2SYan Zheng exist->bytenr == key.offset))) { 8385d4f98a2SYan Zheng exist = NULL; 83971f572a9SQu Wenruo continue; 8405d4f98a2SYan Zheng } 8415d4f98a2SYan Zheng 842fa6ac715SQu Wenruo /* SHARED_BLOCK_REF means key.offset is the parent bytenr */ 8435d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) { 8445d4f98a2SYan Zheng if (key.objectid == key.offset) { 8452433bea5SQu Wenruo cur->is_reloc_root = 1; 8462433bea5SQu Wenruo /* Only reloc backref cache cares exact root */ 8472433bea5SQu Wenruo if (cache->is_reloc) { 8482433bea5SQu Wenruo root = find_reloc_root(fs_info, 8492433bea5SQu Wenruo cur->bytenr); 8502433bea5SQu Wenruo if (WARN_ON(!root)) { 8512433bea5SQu Wenruo err = -ENOENT; 8522433bea5SQu Wenruo goto out; 8532433bea5SQu Wenruo } 8545d4f98a2SYan Zheng cur->root = root; 8552433bea5SQu Wenruo } else { 8562433bea5SQu Wenruo /* 8572433bea5SQu Wenruo * For generic purpose backref cache, 8582433bea5SQu Wenruo * reloc root node is useless. 8592433bea5SQu Wenruo */ 8602433bea5SQu Wenruo list_add(&cur->list, 8612433bea5SQu Wenruo &cache->useless_node); 8622433bea5SQu Wenruo } 8635d4f98a2SYan Zheng break; 8645d4f98a2SYan Zheng } 8655d4f98a2SYan Zheng 8663fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 8675d4f98a2SYan Zheng if (!edge) { 8685d4f98a2SYan Zheng err = -ENOMEM; 8695d4f98a2SYan Zheng goto out; 8705d4f98a2SYan Zheng } 8715d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, key.offset); 8725d4f98a2SYan Zheng if (!rb_node) { 8733fd0a558SYan, Zheng upper = alloc_backref_node(cache); 8745d4f98a2SYan Zheng if (!upper) { 8753fd0a558SYan, Zheng free_backref_edge(cache, edge); 8765d4f98a2SYan Zheng err = -ENOMEM; 8775d4f98a2SYan Zheng goto out; 8785d4f98a2SYan Zheng } 8795d4f98a2SYan Zheng upper->bytenr = key.offset; 8805d4f98a2SYan Zheng upper->level = cur->level + 1; 8815d4f98a2SYan Zheng /* 8825d4f98a2SYan Zheng * backrefs for the upper level block isn't 8835d4f98a2SYan Zheng * cached, add the block to pending list 8845d4f98a2SYan Zheng */ 88584780289SQu Wenruo list_add_tail(&edge->list[UPPER], 88684780289SQu Wenruo &cache->pending_edge); 8875d4f98a2SYan Zheng } else { 8885d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 8895d4f98a2SYan Zheng rb_node); 89075bfb9afSJosef Bacik ASSERT(upper->checked); 8915d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 8925d4f98a2SYan Zheng } 8933fd0a558SYan, Zheng list_add_tail(&edge->list[LOWER], &cur->upper); 8945d4f98a2SYan Zheng edge->node[LOWER] = cur; 8953fd0a558SYan, Zheng edge->node[UPPER] = upper; 8965d4f98a2SYan Zheng 89771f572a9SQu Wenruo continue; 8986d8ff4e4SDavid Sterba } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { 899ba3c2b19SNikolay Borisov err = -EINVAL; 900ba3c2b19SNikolay Borisov btrfs_print_v0_err(rc->extent_root->fs_info); 901ba3c2b19SNikolay Borisov btrfs_handle_fs_error(rc->extent_root->fs_info, err, 902ba3c2b19SNikolay Borisov NULL); 903ba3c2b19SNikolay Borisov goto out; 9045d4f98a2SYan Zheng } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) { 90571f572a9SQu Wenruo continue; 9065d4f98a2SYan Zheng } 9075d4f98a2SYan Zheng 908fa6ac715SQu Wenruo /* 909fa6ac715SQu Wenruo * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref offset 910fa6ac715SQu Wenruo * means the root objectid. We need to search the tree to get 911fa6ac715SQu Wenruo * its parent bytenr. 912fa6ac715SQu Wenruo */ 9135d4f98a2SYan Zheng root = read_fs_root(rc->extent_root->fs_info, key.offset); 9145d4f98a2SYan Zheng if (IS_ERR(root)) { 9155d4f98a2SYan Zheng err = PTR_ERR(root); 9165d4f98a2SYan Zheng goto out; 9175d4f98a2SYan Zheng } 9185d4f98a2SYan Zheng 91927cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 9203fd0a558SYan, Zheng cur->cowonly = 1; 9213fd0a558SYan, Zheng 9225d4f98a2SYan Zheng if (btrfs_root_level(&root->root_item) == cur->level) { 9235d4f98a2SYan Zheng /* tree root */ 92475bfb9afSJosef Bacik ASSERT(btrfs_root_bytenr(&root->root_item) == 9255d4f98a2SYan Zheng cur->bytenr); 9260b530bc5SJosef Bacik if (should_ignore_root(root)) { 92700246528SJosef Bacik btrfs_put_root(root); 92884780289SQu Wenruo list_add(&cur->list, &cache->useless_node); 9290b530bc5SJosef Bacik } else { 9305d4f98a2SYan Zheng cur->root = root; 9310b530bc5SJosef Bacik } 9325d4f98a2SYan Zheng break; 9335d4f98a2SYan Zheng } 9345d4f98a2SYan Zheng 9355d4f98a2SYan Zheng level = cur->level + 1; 9365d4f98a2SYan Zheng 937fa6ac715SQu Wenruo /* Search the tree to find parent blocks referring the block. */ 93871f572a9SQu Wenruo path->search_commit_root = 1; 93971f572a9SQu Wenruo path->skip_locking = 1; 94071f572a9SQu Wenruo path->lowest_level = level; 94171f572a9SQu Wenruo ret = btrfs_search_slot(NULL, root, node_key, path, 0, 0); 94271f572a9SQu Wenruo path->lowest_level = 0; 9435d4f98a2SYan Zheng if (ret < 0) { 94400246528SJosef Bacik btrfs_put_root(root); 9455d4f98a2SYan Zheng err = ret; 9465d4f98a2SYan Zheng goto out; 9475d4f98a2SYan Zheng } 94871f572a9SQu Wenruo if (ret > 0 && path->slots[level] > 0) 94971f572a9SQu Wenruo path->slots[level]--; 9505d4f98a2SYan Zheng 95171f572a9SQu Wenruo eb = path->nodes[level]; 95271f572a9SQu Wenruo if (btrfs_node_blockptr(eb, path->slots[level]) != 9533561b9dbSLiu Bo cur->bytenr) { 9543561b9dbSLiu Bo btrfs_err(root->fs_info, 9553561b9dbSLiu Bo "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)", 9564fd786e6SMisono Tomohiro cur->bytenr, level - 1, 9574fd786e6SMisono Tomohiro root->root_key.objectid, 9583561b9dbSLiu Bo node_key->objectid, node_key->type, 9593561b9dbSLiu Bo node_key->offset); 96000246528SJosef Bacik btrfs_put_root(root); 9613561b9dbSLiu Bo err = -ENOENT; 9623561b9dbSLiu Bo goto out; 9633561b9dbSLiu Bo } 9645d4f98a2SYan Zheng lower = cur; 965b6c60c80SJosef Bacik need_check = true; 966fa6ac715SQu Wenruo 967fa6ac715SQu Wenruo /* Add all nodes and edges in the path */ 9685d4f98a2SYan Zheng for (; level < BTRFS_MAX_LEVEL; level++) { 96971f572a9SQu Wenruo if (!path->nodes[level]) { 97075bfb9afSJosef Bacik ASSERT(btrfs_root_bytenr(&root->root_item) == 9715d4f98a2SYan Zheng lower->bytenr); 9720b530bc5SJosef Bacik if (should_ignore_root(root)) { 97300246528SJosef Bacik btrfs_put_root(root); 97484780289SQu Wenruo list_add(&lower->list, 97584780289SQu Wenruo &cache->useless_node); 9760b530bc5SJosef Bacik } else { 9775d4f98a2SYan Zheng lower->root = root; 9780b530bc5SJosef Bacik } 9795d4f98a2SYan Zheng break; 9805d4f98a2SYan Zheng } 9815d4f98a2SYan Zheng 9823fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 9835d4f98a2SYan Zheng if (!edge) { 98400246528SJosef Bacik btrfs_put_root(root); 9855d4f98a2SYan Zheng err = -ENOMEM; 9865d4f98a2SYan Zheng goto out; 9875d4f98a2SYan Zheng } 9885d4f98a2SYan Zheng 98971f572a9SQu Wenruo eb = path->nodes[level]; 9905d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, eb->start); 9915d4f98a2SYan Zheng if (!rb_node) { 9923fd0a558SYan, Zheng upper = alloc_backref_node(cache); 9935d4f98a2SYan Zheng if (!upper) { 99400246528SJosef Bacik btrfs_put_root(root); 9953fd0a558SYan, Zheng free_backref_edge(cache, edge); 9965d4f98a2SYan Zheng err = -ENOMEM; 9975d4f98a2SYan Zheng goto out; 9985d4f98a2SYan Zheng } 9995d4f98a2SYan Zheng upper->bytenr = eb->start; 10005d4f98a2SYan Zheng upper->owner = btrfs_header_owner(eb); 10015d4f98a2SYan Zheng upper->level = lower->level + 1; 100227cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, 100327cdeb70SMiao Xie &root->state)) 10043fd0a558SYan, Zheng upper->cowonly = 1; 10055d4f98a2SYan Zheng 10065d4f98a2SYan Zheng /* 10075d4f98a2SYan Zheng * if we know the block isn't shared 10085d4f98a2SYan Zheng * we can void checking its backrefs. 10095d4f98a2SYan Zheng */ 10105d4f98a2SYan Zheng if (btrfs_block_can_be_shared(root, eb)) 10115d4f98a2SYan Zheng upper->checked = 0; 10125d4f98a2SYan Zheng else 10135d4f98a2SYan Zheng upper->checked = 1; 10145d4f98a2SYan Zheng 10155d4f98a2SYan Zheng /* 10165d4f98a2SYan Zheng * add the block to pending list if we 1017b6c60c80SJosef Bacik * need check its backrefs, we only do this once 1018b6c60c80SJosef Bacik * while walking up a tree as we will catch 1019b6c60c80SJosef Bacik * anything else later on. 10205d4f98a2SYan Zheng */ 1021b6c60c80SJosef Bacik if (!upper->checked && need_check) { 1022b6c60c80SJosef Bacik need_check = false; 10235d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], 102484780289SQu Wenruo &cache->pending_edge); 1025bbe90514SJosef Bacik } else { 1026bbe90514SJosef Bacik if (upper->checked) 1027bbe90514SJosef Bacik need_check = true; 10285d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 1029bbe90514SJosef Bacik } 10305d4f98a2SYan Zheng } else { 10315d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 10325d4f98a2SYan Zheng rb_node); 103375bfb9afSJosef Bacik ASSERT(upper->checked); 10345d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 10353fd0a558SYan, Zheng if (!upper->owner) 10363fd0a558SYan, Zheng upper->owner = btrfs_header_owner(eb); 10375d4f98a2SYan Zheng } 10385d4f98a2SYan Zheng list_add_tail(&edge->list[LOWER], &lower->upper); 10395d4f98a2SYan Zheng edge->node[LOWER] = lower; 10403fd0a558SYan, Zheng edge->node[UPPER] = upper; 10415d4f98a2SYan Zheng 10420b530bc5SJosef Bacik if (rb_node) { 104300246528SJosef Bacik btrfs_put_root(root); 10445d4f98a2SYan Zheng break; 10450b530bc5SJosef Bacik } 10465d4f98a2SYan Zheng lower = upper; 10475d4f98a2SYan Zheng upper = NULL; 10485d4f98a2SYan Zheng } 104971f572a9SQu Wenruo btrfs_release_path(path); 10505d4f98a2SYan Zheng } 105171f572a9SQu Wenruo if (ret < 0) { 105271f572a9SQu Wenruo err = ret; 105371f572a9SQu Wenruo goto out; 10545d4f98a2SYan Zheng } 105571f572a9SQu Wenruo ret = 0; 105671f572a9SQu Wenruo btrfs_backref_iter_release(iter); 10575d4f98a2SYan Zheng 10585d4f98a2SYan Zheng cur->checked = 1; 10595d4f98a2SYan Zheng WARN_ON(exist); 10605d4f98a2SYan Zheng 10615d4f98a2SYan Zheng /* the pending list isn't empty, take the first block to process */ 106284780289SQu Wenruo if (!list_empty(&cache->pending_edge)) { 106384780289SQu Wenruo edge = list_first_entry(&cache->pending_edge, 106484780289SQu Wenruo struct backref_edge, list[UPPER]); 10655d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 10665d4f98a2SYan Zheng cur = edge->node[UPPER]; 10675d4f98a2SYan Zheng goto again; 10685d4f98a2SYan Zheng } 10695d4f98a2SYan Zheng 10705d4f98a2SYan Zheng /* 10715d4f98a2SYan Zheng * everything goes well, connect backref nodes and insert backref nodes 10725d4f98a2SYan Zheng * into the cache. 10735d4f98a2SYan Zheng */ 107475bfb9afSJosef Bacik ASSERT(node->checked); 10753fd0a558SYan, Zheng cowonly = node->cowonly; 10763fd0a558SYan, Zheng if (!cowonly) { 10773fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, 10783fd0a558SYan, Zheng &node->rb_node); 107943c04fb1SJeff Mahoney if (rb_node) 108043c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 10813fd0a558SYan, Zheng list_add_tail(&node->lower, &cache->leaves); 10823fd0a558SYan, Zheng } 10835d4f98a2SYan Zheng 10845d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) 108584780289SQu Wenruo list_add_tail(&edge->list[UPPER], &cache->pending_edge); 10865d4f98a2SYan Zheng 108784780289SQu Wenruo while (!list_empty(&cache->pending_edge)) { 108884780289SQu Wenruo edge = list_first_entry(&cache->pending_edge, 108984780289SQu Wenruo struct backref_edge, list[UPPER]); 10905d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 10915d4f98a2SYan Zheng upper = edge->node[UPPER]; 10923fd0a558SYan, Zheng if (upper->detached) { 10933fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 10943fd0a558SYan, Zheng lower = edge->node[LOWER]; 10953fd0a558SYan, Zheng free_backref_edge(cache, edge); 10963fd0a558SYan, Zheng if (list_empty(&lower->upper)) 109784780289SQu Wenruo list_add(&lower->list, &cache->useless_node); 10983fd0a558SYan, Zheng continue; 10993fd0a558SYan, Zheng } 11005d4f98a2SYan Zheng 11015d4f98a2SYan Zheng if (!RB_EMPTY_NODE(&upper->rb_node)) { 11025d4f98a2SYan Zheng if (upper->lowest) { 11035d4f98a2SYan Zheng list_del_init(&upper->lower); 11045d4f98a2SYan Zheng upper->lowest = 0; 11055d4f98a2SYan Zheng } 11065d4f98a2SYan Zheng 11075d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 11085d4f98a2SYan Zheng continue; 11095d4f98a2SYan Zheng } 11105d4f98a2SYan Zheng 111175bfb9afSJosef Bacik if (!upper->checked) { 111275bfb9afSJosef Bacik /* 111375bfb9afSJosef Bacik * Still want to blow up for developers since this is a 111475bfb9afSJosef Bacik * logic bug. 111575bfb9afSJosef Bacik */ 111675bfb9afSJosef Bacik ASSERT(0); 111775bfb9afSJosef Bacik err = -EINVAL; 111875bfb9afSJosef Bacik goto out; 111975bfb9afSJosef Bacik } 112075bfb9afSJosef Bacik if (cowonly != upper->cowonly) { 112175bfb9afSJosef Bacik ASSERT(0); 112275bfb9afSJosef Bacik err = -EINVAL; 112375bfb9afSJosef Bacik goto out; 112475bfb9afSJosef Bacik } 112575bfb9afSJosef Bacik 11263fd0a558SYan, Zheng if (!cowonly) { 11275d4f98a2SYan Zheng rb_node = tree_insert(&cache->rb_root, upper->bytenr, 11285d4f98a2SYan Zheng &upper->rb_node); 112943c04fb1SJeff Mahoney if (rb_node) 113043c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 113143c04fb1SJeff Mahoney upper->bytenr); 11323fd0a558SYan, Zheng } 11335d4f98a2SYan Zheng 11345d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 11355d4f98a2SYan Zheng 11365d4f98a2SYan Zheng list_for_each_entry(edge, &upper->upper, list[LOWER]) 113784780289SQu Wenruo list_add_tail(&edge->list[UPPER], &cache->pending_edge); 11385d4f98a2SYan Zheng } 11393fd0a558SYan, Zheng /* 11403fd0a558SYan, Zheng * process useless backref nodes. backref nodes for tree leaves 11413fd0a558SYan, Zheng * are deleted from the cache. backref nodes for upper level 11423fd0a558SYan, Zheng * tree blocks are left in the cache to avoid unnecessary backref 11433fd0a558SYan, Zheng * lookup. 11443fd0a558SYan, Zheng */ 114584780289SQu Wenruo while (!list_empty(&cache->useless_node)) { 114684780289SQu Wenruo upper = list_first_entry(&cache->useless_node, 114784780289SQu Wenruo struct backref_node, list); 11483fd0a558SYan, Zheng list_del_init(&upper->list); 114975bfb9afSJosef Bacik ASSERT(list_empty(&upper->upper)); 11503fd0a558SYan, Zheng if (upper == node) 11513fd0a558SYan, Zheng node = NULL; 11523fd0a558SYan, Zheng if (upper->lowest) { 11533fd0a558SYan, Zheng list_del_init(&upper->lower); 11543fd0a558SYan, Zheng upper->lowest = 0; 11553fd0a558SYan, Zheng } 11563fd0a558SYan, Zheng while (!list_empty(&upper->lower)) { 115784780289SQu Wenruo edge = list_first_entry(&upper->lower, 11583fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 11593fd0a558SYan, Zheng list_del(&edge->list[UPPER]); 11603fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 11613fd0a558SYan, Zheng lower = edge->node[LOWER]; 11623fd0a558SYan, Zheng free_backref_edge(cache, edge); 11633fd0a558SYan, Zheng 11643fd0a558SYan, Zheng if (list_empty(&lower->upper)) 116584780289SQu Wenruo list_add(&lower->list, &cache->useless_node); 11663fd0a558SYan, Zheng } 11679569cc20SQu Wenruo mark_block_processed(rc, upper); 11683fd0a558SYan, Zheng if (upper->level > 0) { 11693fd0a558SYan, Zheng list_add(&upper->list, &cache->detached); 11703fd0a558SYan, Zheng upper->detached = 1; 11713fd0a558SYan, Zheng } else { 11723fd0a558SYan, Zheng rb_erase(&upper->rb_node, &cache->rb_root); 11733fd0a558SYan, Zheng free_backref_node(cache, upper); 11743fd0a558SYan, Zheng } 11753fd0a558SYan, Zheng } 11765d4f98a2SYan Zheng out: 117771f572a9SQu Wenruo btrfs_backref_iter_free(iter); 117871f572a9SQu Wenruo btrfs_free_path(path); 11795d4f98a2SYan Zheng if (err) { 118084780289SQu Wenruo while (!list_empty(&cache->useless_node)) { 118184780289SQu Wenruo lower = list_first_entry(&cache->useless_node, 118275bfb9afSJosef Bacik struct backref_node, list); 118375bfb9afSJosef Bacik list_del_init(&lower->list); 11843fd0a558SYan, Zheng } 118584780289SQu Wenruo while (!list_empty(&cache->pending_edge)) { 118684780289SQu Wenruo edge = list_first_entry(&cache->pending_edge, 118784780289SQu Wenruo struct backref_edge, list[UPPER]); 118875bfb9afSJosef Bacik list_del(&edge->list[UPPER]); 11893fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 119075bfb9afSJosef Bacik lower = edge->node[LOWER]; 11915d4f98a2SYan Zheng upper = edge->node[UPPER]; 11923fd0a558SYan, Zheng free_backref_edge(cache, edge); 119375bfb9afSJosef Bacik 119475bfb9afSJosef Bacik /* 119575bfb9afSJosef Bacik * Lower is no longer linked to any upper backref nodes 119675bfb9afSJosef Bacik * and isn't in the cache, we can free it ourselves. 119775bfb9afSJosef Bacik */ 119875bfb9afSJosef Bacik if (list_empty(&lower->upper) && 119975bfb9afSJosef Bacik RB_EMPTY_NODE(&lower->rb_node)) 120084780289SQu Wenruo list_add(&lower->list, &cache->useless_node); 120175bfb9afSJosef Bacik 120275bfb9afSJosef Bacik if (!RB_EMPTY_NODE(&upper->rb_node)) 120375bfb9afSJosef Bacik continue; 120475bfb9afSJosef Bacik 120501327610SNicholas D Steeves /* Add this guy's upper edges to the list to process */ 120675bfb9afSJosef Bacik list_for_each_entry(edge, &upper->upper, list[LOWER]) 120784780289SQu Wenruo list_add_tail(&edge->list[UPPER], 120884780289SQu Wenruo &cache->pending_edge); 120975bfb9afSJosef Bacik if (list_empty(&upper->upper)) 121084780289SQu Wenruo list_add(&upper->list, &cache->useless_node); 121175bfb9afSJosef Bacik } 121275bfb9afSJosef Bacik 121384780289SQu Wenruo while (!list_empty(&cache->useless_node)) { 121484780289SQu Wenruo lower = list_first_entry(&cache->useless_node, 121575bfb9afSJosef Bacik struct backref_node, list); 121675bfb9afSJosef Bacik list_del_init(&lower->list); 12170fd8c3daSLiu Bo if (lower == node) 12180fd8c3daSLiu Bo node = NULL; 121975bfb9afSJosef Bacik free_backref_node(cache, lower); 12205d4f98a2SYan Zheng } 12210fd8c3daSLiu Bo 12228e19c973SJosef Bacik remove_backref_node(cache, node); 122384780289SQu Wenruo ASSERT(list_empty(&cache->useless_node) && 122484780289SQu Wenruo list_empty(&cache->pending_edge)); 12255d4f98a2SYan Zheng return ERR_PTR(err); 12265d4f98a2SYan Zheng } 122775bfb9afSJosef Bacik ASSERT(!node || !node->detached); 122884780289SQu Wenruo ASSERT(list_empty(&cache->useless_node) && 122984780289SQu Wenruo list_empty(&cache->pending_edge)); 12305d4f98a2SYan Zheng return node; 12315d4f98a2SYan Zheng } 12325d4f98a2SYan Zheng 12335d4f98a2SYan Zheng /* 12343fd0a558SYan, Zheng * helper to add backref node for the newly created snapshot. 12353fd0a558SYan, Zheng * the backref node is created by cloning backref node that 12363fd0a558SYan, Zheng * corresponds to root of source tree 12373fd0a558SYan, Zheng */ 12383fd0a558SYan, Zheng static int clone_backref_node(struct btrfs_trans_handle *trans, 12393fd0a558SYan, Zheng struct reloc_control *rc, 12403fd0a558SYan, Zheng struct btrfs_root *src, 12413fd0a558SYan, Zheng struct btrfs_root *dest) 12423fd0a558SYan, Zheng { 12433fd0a558SYan, Zheng struct btrfs_root *reloc_root = src->reloc_root; 12443fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 12453fd0a558SYan, Zheng struct backref_node *node = NULL; 12463fd0a558SYan, Zheng struct backref_node *new_node; 12473fd0a558SYan, Zheng struct backref_edge *edge; 12483fd0a558SYan, Zheng struct backref_edge *new_edge; 12493fd0a558SYan, Zheng struct rb_node *rb_node; 12503fd0a558SYan, Zheng 12513fd0a558SYan, Zheng if (cache->last_trans > 0) 12523fd0a558SYan, Zheng update_backref_cache(trans, cache); 12533fd0a558SYan, Zheng 12543fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, src->commit_root->start); 12553fd0a558SYan, Zheng if (rb_node) { 12563fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, rb_node); 12573fd0a558SYan, Zheng if (node->detached) 12583fd0a558SYan, Zheng node = NULL; 12593fd0a558SYan, Zheng else 12603fd0a558SYan, Zheng BUG_ON(node->new_bytenr != reloc_root->node->start); 12613fd0a558SYan, Zheng } 12623fd0a558SYan, Zheng 12633fd0a558SYan, Zheng if (!node) { 12643fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, 12653fd0a558SYan, Zheng reloc_root->commit_root->start); 12663fd0a558SYan, Zheng if (rb_node) { 12673fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, 12683fd0a558SYan, Zheng rb_node); 12693fd0a558SYan, Zheng BUG_ON(node->detached); 12703fd0a558SYan, Zheng } 12713fd0a558SYan, Zheng } 12723fd0a558SYan, Zheng 12733fd0a558SYan, Zheng if (!node) 12743fd0a558SYan, Zheng return 0; 12753fd0a558SYan, Zheng 12763fd0a558SYan, Zheng new_node = alloc_backref_node(cache); 12773fd0a558SYan, Zheng if (!new_node) 12783fd0a558SYan, Zheng return -ENOMEM; 12793fd0a558SYan, Zheng 12803fd0a558SYan, Zheng new_node->bytenr = dest->node->start; 12813fd0a558SYan, Zheng new_node->level = node->level; 12823fd0a558SYan, Zheng new_node->lowest = node->lowest; 12836848ad64SYan, Zheng new_node->checked = 1; 128400246528SJosef Bacik new_node->root = btrfs_grab_root(dest); 12850b530bc5SJosef Bacik ASSERT(new_node->root); 12863fd0a558SYan, Zheng 12873fd0a558SYan, Zheng if (!node->lowest) { 12883fd0a558SYan, Zheng list_for_each_entry(edge, &node->lower, list[UPPER]) { 12893fd0a558SYan, Zheng new_edge = alloc_backref_edge(cache); 12903fd0a558SYan, Zheng if (!new_edge) 12913fd0a558SYan, Zheng goto fail; 12923fd0a558SYan, Zheng 12933fd0a558SYan, Zheng new_edge->node[UPPER] = new_node; 12943fd0a558SYan, Zheng new_edge->node[LOWER] = edge->node[LOWER]; 12953fd0a558SYan, Zheng list_add_tail(&new_edge->list[UPPER], 12963fd0a558SYan, Zheng &new_node->lower); 12973fd0a558SYan, Zheng } 129876b9e23dSMiao Xie } else { 129976b9e23dSMiao Xie list_add_tail(&new_node->lower, &cache->leaves); 13003fd0a558SYan, Zheng } 13013fd0a558SYan, Zheng 13023fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, new_node->bytenr, 13033fd0a558SYan, Zheng &new_node->rb_node); 130443c04fb1SJeff Mahoney if (rb_node) 130543c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, new_node->bytenr); 13063fd0a558SYan, Zheng 13073fd0a558SYan, Zheng if (!new_node->lowest) { 13083fd0a558SYan, Zheng list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) { 13093fd0a558SYan, Zheng list_add_tail(&new_edge->list[LOWER], 13103fd0a558SYan, Zheng &new_edge->node[LOWER]->upper); 13113fd0a558SYan, Zheng } 13123fd0a558SYan, Zheng } 13133fd0a558SYan, Zheng return 0; 13143fd0a558SYan, Zheng fail: 13153fd0a558SYan, Zheng while (!list_empty(&new_node->lower)) { 13163fd0a558SYan, Zheng new_edge = list_entry(new_node->lower.next, 13173fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 13183fd0a558SYan, Zheng list_del(&new_edge->list[UPPER]); 13193fd0a558SYan, Zheng free_backref_edge(cache, new_edge); 13203fd0a558SYan, Zheng } 13213fd0a558SYan, Zheng free_backref_node(cache, new_node); 13223fd0a558SYan, Zheng return -ENOMEM; 13233fd0a558SYan, Zheng } 13243fd0a558SYan, Zheng 13253fd0a558SYan, Zheng /* 13265d4f98a2SYan Zheng * helper to add 'address of tree root -> reloc tree' mapping 13275d4f98a2SYan Zheng */ 1328ffd7b339SJeff Mahoney static int __must_check __add_reloc_root(struct btrfs_root *root) 13295d4f98a2SYan Zheng { 13300b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13315d4f98a2SYan Zheng struct rb_node *rb_node; 13325d4f98a2SYan Zheng struct mapping_node *node; 13330b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 13345d4f98a2SYan Zheng 13355d4f98a2SYan Zheng node = kmalloc(sizeof(*node), GFP_NOFS); 1336ffd7b339SJeff Mahoney if (!node) 1337ffd7b339SJeff Mahoney return -ENOMEM; 13385d4f98a2SYan Zheng 1339ea287ab1SJosef Bacik node->bytenr = root->commit_root->start; 13405d4f98a2SYan Zheng node->data = root; 13415d4f98a2SYan Zheng 13425d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 13435d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 13445d4f98a2SYan Zheng node->bytenr, &node->rb_node); 13455d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 1346ffd7b339SJeff Mahoney if (rb_node) { 13470b246afaSJeff Mahoney btrfs_panic(fs_info, -EEXIST, 13485d163e0eSJeff Mahoney "Duplicate root found for start=%llu while inserting into relocation tree", 13495d163e0eSJeff Mahoney node->bytenr); 1350ffd7b339SJeff Mahoney } 13515d4f98a2SYan Zheng 13525d4f98a2SYan Zheng list_add_tail(&root->root_list, &rc->reloc_roots); 13535d4f98a2SYan Zheng return 0; 13545d4f98a2SYan Zheng } 13555d4f98a2SYan Zheng 13565d4f98a2SYan Zheng /* 1357c974c464SWang Shilong * helper to delete the 'address of tree root -> reloc tree' 13585d4f98a2SYan Zheng * mapping 13595d4f98a2SYan Zheng */ 1360c974c464SWang Shilong static void __del_reloc_root(struct btrfs_root *root) 13615d4f98a2SYan Zheng { 13620b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13635d4f98a2SYan Zheng struct rb_node *rb_node; 13645d4f98a2SYan Zheng struct mapping_node *node = NULL; 13650b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 1366f44deb74SJosef Bacik bool put_ref = false; 13675d4f98a2SYan Zheng 136865c6e82bSQu Wenruo if (rc && root->node) { 13695d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 13705d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1371ea287ab1SJosef Bacik root->commit_root->start); 1372c974c464SWang Shilong if (rb_node) { 1373c974c464SWang Shilong node = rb_entry(rb_node, struct mapping_node, rb_node); 1374c974c464SWang Shilong rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 1375ea287ab1SJosef Bacik RB_CLEAR_NODE(&node->rb_node); 1376c974c464SWang Shilong } 1377c974c464SWang Shilong spin_unlock(&rc->reloc_root_tree.lock); 1378c974c464SWang Shilong if (!node) 1379c974c464SWang Shilong return; 1380c974c464SWang Shilong BUG_ON((struct btrfs_root *)node->data != root); 1381389305b2SQu Wenruo } 1382c974c464SWang Shilong 1383f44deb74SJosef Bacik /* 1384f44deb74SJosef Bacik * We only put the reloc root here if it's on the list. There's a lot 1385f44deb74SJosef Bacik * of places where the pattern is to splice the rc->reloc_roots, process 1386f44deb74SJosef Bacik * the reloc roots, and then add the reloc root back onto 1387f44deb74SJosef Bacik * rc->reloc_roots. If we call __del_reloc_root while it's off of the 1388f44deb74SJosef Bacik * list we don't want the reference being dropped, because the guy 1389f44deb74SJosef Bacik * messing with the list is in charge of the reference. 1390f44deb74SJosef Bacik */ 13910b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1392f44deb74SJosef Bacik if (!list_empty(&root->root_list)) { 1393f44deb74SJosef Bacik put_ref = true; 1394c974c464SWang Shilong list_del_init(&root->root_list); 1395f44deb74SJosef Bacik } 13960b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1397f44deb74SJosef Bacik if (put_ref) 1398f44deb74SJosef Bacik btrfs_put_root(root); 1399c974c464SWang Shilong kfree(node); 1400c974c464SWang Shilong } 1401c974c464SWang Shilong 1402c974c464SWang Shilong /* 1403c974c464SWang Shilong * helper to update the 'address of tree root -> reloc tree' 1404c974c464SWang Shilong * mapping 1405c974c464SWang Shilong */ 1406ea287ab1SJosef Bacik static int __update_reloc_root(struct btrfs_root *root) 1407c974c464SWang Shilong { 14080b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1409c974c464SWang Shilong struct rb_node *rb_node; 1410c974c464SWang Shilong struct mapping_node *node = NULL; 14110b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 1412c974c464SWang Shilong 1413c974c464SWang Shilong spin_lock(&rc->reloc_root_tree.lock); 1414c974c464SWang Shilong rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1415ea287ab1SJosef Bacik root->commit_root->start); 14165d4f98a2SYan Zheng if (rb_node) { 14175d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 14185d4f98a2SYan Zheng rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 14195d4f98a2SYan Zheng } 14205d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 14215d4f98a2SYan Zheng 14228f71f3e0SLiu Bo if (!node) 14238f71f3e0SLiu Bo return 0; 14245d4f98a2SYan Zheng BUG_ON((struct btrfs_root *)node->data != root); 14255d4f98a2SYan Zheng 14265d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 1427ea287ab1SJosef Bacik node->bytenr = root->node->start; 14285d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 14295d4f98a2SYan Zheng node->bytenr, &node->rb_node); 14305d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 143143c04fb1SJeff Mahoney if (rb_node) 143243c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 14335d4f98a2SYan Zheng return 0; 14345d4f98a2SYan Zheng } 14355d4f98a2SYan Zheng 14363fd0a558SYan, Zheng static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans, 14373fd0a558SYan, Zheng struct btrfs_root *root, u64 objectid) 14385d4f98a2SYan Zheng { 14390b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 14405d4f98a2SYan Zheng struct btrfs_root *reloc_root; 14415d4f98a2SYan Zheng struct extent_buffer *eb; 14425d4f98a2SYan Zheng struct btrfs_root_item *root_item; 14435d4f98a2SYan Zheng struct btrfs_key root_key; 14445d4f98a2SYan Zheng int ret; 14455d4f98a2SYan Zheng 14465d4f98a2SYan Zheng root_item = kmalloc(sizeof(*root_item), GFP_NOFS); 14475d4f98a2SYan Zheng BUG_ON(!root_item); 14485d4f98a2SYan Zheng 14495d4f98a2SYan Zheng root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; 14505d4f98a2SYan Zheng root_key.type = BTRFS_ROOT_ITEM_KEY; 14513fd0a558SYan, Zheng root_key.offset = objectid; 14525d4f98a2SYan Zheng 14533fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 1454054570a1SFilipe Manana u64 commit_root_gen; 1455054570a1SFilipe Manana 14563fd0a558SYan, Zheng /* called by btrfs_init_reloc_root */ 14575d4f98a2SYan Zheng ret = btrfs_copy_root(trans, root, root->commit_root, &eb, 14585d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID); 14595d4f98a2SYan Zheng BUG_ON(ret); 1460054570a1SFilipe Manana /* 1461054570a1SFilipe Manana * Set the last_snapshot field to the generation of the commit 1462054570a1SFilipe Manana * root - like this ctree.c:btrfs_block_can_be_shared() behaves 1463054570a1SFilipe Manana * correctly (returns true) when the relocation root is created 1464054570a1SFilipe Manana * either inside the critical section of a transaction commit 1465054570a1SFilipe Manana * (through transaction.c:qgroup_account_snapshot()) and when 1466054570a1SFilipe Manana * it's created before the transaction commit is started. 1467054570a1SFilipe Manana */ 1468054570a1SFilipe Manana commit_root_gen = btrfs_header_generation(root->commit_root); 1469054570a1SFilipe Manana btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen); 14703fd0a558SYan, Zheng } else { 14713fd0a558SYan, Zheng /* 14723fd0a558SYan, Zheng * called by btrfs_reloc_post_snapshot_hook. 14733fd0a558SYan, Zheng * the source tree is a reloc tree, all tree blocks 14743fd0a558SYan, Zheng * modified after it was created have RELOC flag 14753fd0a558SYan, Zheng * set in their headers. so it's OK to not update 14763fd0a558SYan, Zheng * the 'last_snapshot'. 14773fd0a558SYan, Zheng */ 14783fd0a558SYan, Zheng ret = btrfs_copy_root(trans, root, root->node, &eb, 14793fd0a558SYan, Zheng BTRFS_TREE_RELOC_OBJECTID); 14803fd0a558SYan, Zheng BUG_ON(ret); 14813fd0a558SYan, Zheng } 14823fd0a558SYan, Zheng 14835d4f98a2SYan Zheng memcpy(root_item, &root->root_item, sizeof(*root_item)); 14845d4f98a2SYan Zheng btrfs_set_root_bytenr(root_item, eb->start); 14855d4f98a2SYan Zheng btrfs_set_root_level(root_item, btrfs_header_level(eb)); 14865d4f98a2SYan Zheng btrfs_set_root_generation(root_item, trans->transid); 14873fd0a558SYan, Zheng 14883fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 14893fd0a558SYan, Zheng btrfs_set_root_refs(root_item, 0); 14903fd0a558SYan, Zheng memset(&root_item->drop_progress, 0, 14913fd0a558SYan, Zheng sizeof(struct btrfs_disk_key)); 14925d4f98a2SYan Zheng root_item->drop_level = 0; 14933fd0a558SYan, Zheng } 14945d4f98a2SYan Zheng 14955d4f98a2SYan Zheng btrfs_tree_unlock(eb); 14965d4f98a2SYan Zheng free_extent_buffer(eb); 14975d4f98a2SYan Zheng 14980b246afaSJeff Mahoney ret = btrfs_insert_root(trans, fs_info->tree_root, 14995d4f98a2SYan Zheng &root_key, root_item); 15005d4f98a2SYan Zheng BUG_ON(ret); 15015d4f98a2SYan Zheng kfree(root_item); 15025d4f98a2SYan Zheng 15033dbf1738SJosef Bacik reloc_root = btrfs_read_tree_root(fs_info->tree_root, &root_key); 15045d4f98a2SYan Zheng BUG_ON(IS_ERR(reloc_root)); 15053dbf1738SJosef Bacik set_bit(BTRFS_ROOT_REF_COWS, &reloc_root->state); 15065d4f98a2SYan Zheng reloc_root->last_trans = trans->transid; 15073fd0a558SYan, Zheng return reloc_root; 15083fd0a558SYan, Zheng } 15093fd0a558SYan, Zheng 15103fd0a558SYan, Zheng /* 15113fd0a558SYan, Zheng * create reloc tree for a given fs tree. reloc tree is just a 15123fd0a558SYan, Zheng * snapshot of the fs tree with special root objectid. 1513f44deb74SJosef Bacik * 1514f44deb74SJosef Bacik * The reloc_root comes out of here with two references, one for 1515f44deb74SJosef Bacik * root->reloc_root, and another for being on the rc->reloc_roots list. 15163fd0a558SYan, Zheng */ 15173fd0a558SYan, Zheng int btrfs_init_reloc_root(struct btrfs_trans_handle *trans, 15183fd0a558SYan, Zheng struct btrfs_root *root) 15193fd0a558SYan, Zheng { 15200b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 15213fd0a558SYan, Zheng struct btrfs_root *reloc_root; 15220b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 152320dd2cbfSMiao Xie struct btrfs_block_rsv *rsv; 15243fd0a558SYan, Zheng int clear_rsv = 0; 1525ffd7b339SJeff Mahoney int ret; 15263fd0a558SYan, Zheng 1527aec7db3bSJosef Bacik if (!rc) 15282abc726aSJosef Bacik return 0; 15292abc726aSJosef Bacik 15301fac4a54SQu Wenruo /* 15311fac4a54SQu Wenruo * The subvolume has reloc tree but the swap is finished, no need to 15321fac4a54SQu Wenruo * create/update the dead reloc tree 15331fac4a54SQu Wenruo */ 15346282675eSQu Wenruo if (reloc_root_is_dead(root)) 15351fac4a54SQu Wenruo return 0; 15361fac4a54SQu Wenruo 1537aec7db3bSJosef Bacik /* 1538aec7db3bSJosef Bacik * This is subtle but important. We do not do 1539aec7db3bSJosef Bacik * record_root_in_transaction for reloc roots, instead we record their 1540aec7db3bSJosef Bacik * corresponding fs root, and then here we update the last trans for the 1541aec7db3bSJosef Bacik * reloc root. This means that we have to do this for the entire life 1542aec7db3bSJosef Bacik * of the reloc root, regardless of which stage of the relocation we are 1543aec7db3bSJosef Bacik * in. 1544aec7db3bSJosef Bacik */ 15453fd0a558SYan, Zheng if (root->reloc_root) { 15463fd0a558SYan, Zheng reloc_root = root->reloc_root; 15473fd0a558SYan, Zheng reloc_root->last_trans = trans->transid; 15483fd0a558SYan, Zheng return 0; 15493fd0a558SYan, Zheng } 15503fd0a558SYan, Zheng 1551aec7db3bSJosef Bacik /* 1552aec7db3bSJosef Bacik * We are merging reloc roots, we do not need new reloc trees. Also 1553aec7db3bSJosef Bacik * reloc trees never need their own reloc tree. 1554aec7db3bSJosef Bacik */ 1555aec7db3bSJosef Bacik if (!rc->create_reloc_tree || 1556aec7db3bSJosef Bacik root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 1557aec7db3bSJosef Bacik return 0; 1558aec7db3bSJosef Bacik 155920dd2cbfSMiao Xie if (!trans->reloc_reserved) { 156020dd2cbfSMiao Xie rsv = trans->block_rsv; 15613fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 15623fd0a558SYan, Zheng clear_rsv = 1; 15633fd0a558SYan, Zheng } 15643fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root, root->root_key.objectid); 15653fd0a558SYan, Zheng if (clear_rsv) 156620dd2cbfSMiao Xie trans->block_rsv = rsv; 15675d4f98a2SYan Zheng 1568ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 1569ffd7b339SJeff Mahoney BUG_ON(ret < 0); 1570f44deb74SJosef Bacik root->reloc_root = btrfs_grab_root(reloc_root); 15715d4f98a2SYan Zheng return 0; 15725d4f98a2SYan Zheng } 15735d4f98a2SYan Zheng 15745d4f98a2SYan Zheng /* 15755d4f98a2SYan Zheng * update root item of reloc tree 15765d4f98a2SYan Zheng */ 15775d4f98a2SYan Zheng int btrfs_update_reloc_root(struct btrfs_trans_handle *trans, 15785d4f98a2SYan Zheng struct btrfs_root *root) 15795d4f98a2SYan Zheng { 15800b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 15815d4f98a2SYan Zheng struct btrfs_root *reloc_root; 15825d4f98a2SYan Zheng struct btrfs_root_item *root_item; 15835d4f98a2SYan Zheng int ret; 15845d4f98a2SYan Zheng 15856282675eSQu Wenruo if (!have_reloc_root(root)) 15867585717fSChris Mason goto out; 15875d4f98a2SYan Zheng 15885d4f98a2SYan Zheng reloc_root = root->reloc_root; 15895d4f98a2SYan Zheng root_item = &reloc_root->root_item; 15905d4f98a2SYan Zheng 1591f44deb74SJosef Bacik /* 1592f44deb74SJosef Bacik * We are probably ok here, but __del_reloc_root() will drop its ref of 1593f44deb74SJosef Bacik * the root. We have the ref for root->reloc_root, but just in case 1594f44deb74SJosef Bacik * hold it while we update the reloc root. 1595f44deb74SJosef Bacik */ 1596f44deb74SJosef Bacik btrfs_grab_root(reloc_root); 1597f44deb74SJosef Bacik 1598d2311e69SQu Wenruo /* root->reloc_root will stay until current relocation finished */ 15990b246afaSJeff Mahoney if (fs_info->reloc_ctl->merge_reloc_tree && 16003fd0a558SYan, Zheng btrfs_root_refs(root_item) == 0) { 1601d2311e69SQu Wenruo set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 16026282675eSQu Wenruo /* 16036282675eSQu Wenruo * Mark the tree as dead before we change reloc_root so 16046282675eSQu Wenruo * have_reloc_root will not touch it from now on. 16056282675eSQu Wenruo */ 16066282675eSQu Wenruo smp_wmb(); 1607c974c464SWang Shilong __del_reloc_root(reloc_root); 16085d4f98a2SYan Zheng } 16095d4f98a2SYan Zheng 16105d4f98a2SYan Zheng if (reloc_root->commit_root != reloc_root->node) { 1611ea287ab1SJosef Bacik __update_reloc_root(reloc_root); 16125d4f98a2SYan Zheng btrfs_set_root_node(root_item, reloc_root->node); 16135d4f98a2SYan Zheng free_extent_buffer(reloc_root->commit_root); 16145d4f98a2SYan Zheng reloc_root->commit_root = btrfs_root_node(reloc_root); 16155d4f98a2SYan Zheng } 16165d4f98a2SYan Zheng 16170b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 16185d4f98a2SYan Zheng &reloc_root->root_key, root_item); 16195d4f98a2SYan Zheng BUG_ON(ret); 1620f44deb74SJosef Bacik btrfs_put_root(reloc_root); 16217585717fSChris Mason out: 16225d4f98a2SYan Zheng return 0; 16235d4f98a2SYan Zheng } 16245d4f98a2SYan Zheng 16255d4f98a2SYan Zheng /* 16265d4f98a2SYan Zheng * helper to find first cached inode with inode number >= objectid 16275d4f98a2SYan Zheng * in a subvolume 16285d4f98a2SYan Zheng */ 16295d4f98a2SYan Zheng static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid) 16305d4f98a2SYan Zheng { 16315d4f98a2SYan Zheng struct rb_node *node; 16325d4f98a2SYan Zheng struct rb_node *prev; 16335d4f98a2SYan Zheng struct btrfs_inode *entry; 16345d4f98a2SYan Zheng struct inode *inode; 16355d4f98a2SYan Zheng 16365d4f98a2SYan Zheng spin_lock(&root->inode_lock); 16375d4f98a2SYan Zheng again: 16385d4f98a2SYan Zheng node = root->inode_tree.rb_node; 16395d4f98a2SYan Zheng prev = NULL; 16405d4f98a2SYan Zheng while (node) { 16415d4f98a2SYan Zheng prev = node; 16425d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 16435d4f98a2SYan Zheng 16444a0cc7caSNikolay Borisov if (objectid < btrfs_ino(entry)) 16455d4f98a2SYan Zheng node = node->rb_left; 16464a0cc7caSNikolay Borisov else if (objectid > btrfs_ino(entry)) 16475d4f98a2SYan Zheng node = node->rb_right; 16485d4f98a2SYan Zheng else 16495d4f98a2SYan Zheng break; 16505d4f98a2SYan Zheng } 16515d4f98a2SYan Zheng if (!node) { 16525d4f98a2SYan Zheng while (prev) { 16535d4f98a2SYan Zheng entry = rb_entry(prev, struct btrfs_inode, rb_node); 16544a0cc7caSNikolay Borisov if (objectid <= btrfs_ino(entry)) { 16555d4f98a2SYan Zheng node = prev; 16565d4f98a2SYan Zheng break; 16575d4f98a2SYan Zheng } 16585d4f98a2SYan Zheng prev = rb_next(prev); 16595d4f98a2SYan Zheng } 16605d4f98a2SYan Zheng } 16615d4f98a2SYan Zheng while (node) { 16625d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 16635d4f98a2SYan Zheng inode = igrab(&entry->vfs_inode); 16645d4f98a2SYan Zheng if (inode) { 16655d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 16665d4f98a2SYan Zheng return inode; 16675d4f98a2SYan Zheng } 16685d4f98a2SYan Zheng 16694a0cc7caSNikolay Borisov objectid = btrfs_ino(entry) + 1; 16705d4f98a2SYan Zheng if (cond_resched_lock(&root->inode_lock)) 16715d4f98a2SYan Zheng goto again; 16725d4f98a2SYan Zheng 16735d4f98a2SYan Zheng node = rb_next(node); 16745d4f98a2SYan Zheng } 16755d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 16765d4f98a2SYan Zheng return NULL; 16775d4f98a2SYan Zheng } 16785d4f98a2SYan Zheng 16795d4f98a2SYan Zheng /* 16805d4f98a2SYan Zheng * get new location of data 16815d4f98a2SYan Zheng */ 16825d4f98a2SYan Zheng static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr, 16835d4f98a2SYan Zheng u64 bytenr, u64 num_bytes) 16845d4f98a2SYan Zheng { 16855d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(reloc_inode)->root; 16865d4f98a2SYan Zheng struct btrfs_path *path; 16875d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 16885d4f98a2SYan Zheng struct extent_buffer *leaf; 16895d4f98a2SYan Zheng int ret; 16905d4f98a2SYan Zheng 16915d4f98a2SYan Zheng path = btrfs_alloc_path(); 16925d4f98a2SYan Zheng if (!path) 16935d4f98a2SYan Zheng return -ENOMEM; 16945d4f98a2SYan Zheng 16955d4f98a2SYan Zheng bytenr -= BTRFS_I(reloc_inode)->index_cnt; 1696f85b7379SDavid Sterba ret = btrfs_lookup_file_extent(NULL, root, path, 1697f85b7379SDavid Sterba btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0); 16985d4f98a2SYan Zheng if (ret < 0) 16995d4f98a2SYan Zheng goto out; 17005d4f98a2SYan Zheng if (ret > 0) { 17015d4f98a2SYan Zheng ret = -ENOENT; 17025d4f98a2SYan Zheng goto out; 17035d4f98a2SYan Zheng } 17045d4f98a2SYan Zheng 17055d4f98a2SYan Zheng leaf = path->nodes[0]; 17065d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 17075d4f98a2SYan Zheng struct btrfs_file_extent_item); 17085d4f98a2SYan Zheng 17095d4f98a2SYan Zheng BUG_ON(btrfs_file_extent_offset(leaf, fi) || 17105d4f98a2SYan Zheng btrfs_file_extent_compression(leaf, fi) || 17115d4f98a2SYan Zheng btrfs_file_extent_encryption(leaf, fi) || 17125d4f98a2SYan Zheng btrfs_file_extent_other_encoding(leaf, fi)); 17135d4f98a2SYan Zheng 17145d4f98a2SYan Zheng if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) { 171583d4cfd4SJosef Bacik ret = -EINVAL; 17165d4f98a2SYan Zheng goto out; 17175d4f98a2SYan Zheng } 17185d4f98a2SYan Zheng 17195d4f98a2SYan Zheng *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 17205d4f98a2SYan Zheng ret = 0; 17215d4f98a2SYan Zheng out: 17225d4f98a2SYan Zheng btrfs_free_path(path); 17235d4f98a2SYan Zheng return ret; 17245d4f98a2SYan Zheng } 17255d4f98a2SYan Zheng 17265d4f98a2SYan Zheng /* 17275d4f98a2SYan Zheng * update file extent items in the tree leaf to point to 17285d4f98a2SYan Zheng * the new locations. 17295d4f98a2SYan Zheng */ 17303fd0a558SYan, Zheng static noinline_for_stack 17313fd0a558SYan, Zheng int replace_file_extents(struct btrfs_trans_handle *trans, 17325d4f98a2SYan Zheng struct reloc_control *rc, 17335d4f98a2SYan Zheng struct btrfs_root *root, 17343fd0a558SYan, Zheng struct extent_buffer *leaf) 17355d4f98a2SYan Zheng { 17360b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 17375d4f98a2SYan Zheng struct btrfs_key key; 17385d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 17395d4f98a2SYan Zheng struct inode *inode = NULL; 17405d4f98a2SYan Zheng u64 parent; 17415d4f98a2SYan Zheng u64 bytenr; 17423fd0a558SYan, Zheng u64 new_bytenr = 0; 17435d4f98a2SYan Zheng u64 num_bytes; 17445d4f98a2SYan Zheng u64 end; 17455d4f98a2SYan Zheng u32 nritems; 17465d4f98a2SYan Zheng u32 i; 174783d4cfd4SJosef Bacik int ret = 0; 17485d4f98a2SYan Zheng int first = 1; 17495d4f98a2SYan Zheng int dirty = 0; 17505d4f98a2SYan Zheng 17515d4f98a2SYan Zheng if (rc->stage != UPDATE_DATA_PTRS) 17525d4f98a2SYan Zheng return 0; 17535d4f98a2SYan Zheng 17545d4f98a2SYan Zheng /* reloc trees always use full backref */ 17555d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 17565d4f98a2SYan Zheng parent = leaf->start; 17575d4f98a2SYan Zheng else 17585d4f98a2SYan Zheng parent = 0; 17595d4f98a2SYan Zheng 17605d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 17615d4f98a2SYan Zheng for (i = 0; i < nritems; i++) { 176282fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 176382fa113fSQu Wenruo 17645d4f98a2SYan Zheng cond_resched(); 17655d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, i); 17665d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_DATA_KEY) 17675d4f98a2SYan Zheng continue; 17685d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 17695d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 17705d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 17715d4f98a2SYan Zheng continue; 17725d4f98a2SYan Zheng bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 17735d4f98a2SYan Zheng num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); 17745d4f98a2SYan Zheng if (bytenr == 0) 17755d4f98a2SYan Zheng continue; 17769569cc20SQu Wenruo if (!in_range(bytenr, rc->block_group->start, 17779569cc20SQu Wenruo rc->block_group->length)) 17785d4f98a2SYan Zheng continue; 17795d4f98a2SYan Zheng 17805d4f98a2SYan Zheng /* 17815d4f98a2SYan Zheng * if we are modifying block in fs tree, wait for readpage 17825d4f98a2SYan Zheng * to complete and drop the extent cache 17835d4f98a2SYan Zheng */ 17845d4f98a2SYan Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 17855d4f98a2SYan Zheng if (first) { 17865d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 17875d4f98a2SYan Zheng first = 0; 17884a0cc7caSNikolay Borisov } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) { 17893fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 17905d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 17915d4f98a2SYan Zheng } 17924a0cc7caSNikolay Borisov if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) { 17935d4f98a2SYan Zheng end = key.offset + 17945d4f98a2SYan Zheng btrfs_file_extent_num_bytes(leaf, fi); 17955d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(key.offset, 17960b246afaSJeff Mahoney fs_info->sectorsize)); 17970b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 17985d4f98a2SYan Zheng end--; 17995d4f98a2SYan Zheng ret = try_lock_extent(&BTRFS_I(inode)->io_tree, 1800d0082371SJeff Mahoney key.offset, end); 18015d4f98a2SYan Zheng if (!ret) 18025d4f98a2SYan Zheng continue; 18035d4f98a2SYan Zheng 1804dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), 1805dcdbc059SNikolay Borisov key.offset, end, 1); 18065d4f98a2SYan Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 1807d0082371SJeff Mahoney key.offset, end); 18085d4f98a2SYan Zheng } 18095d4f98a2SYan Zheng } 18105d4f98a2SYan Zheng 18115d4f98a2SYan Zheng ret = get_new_location(rc->data_inode, &new_bytenr, 18125d4f98a2SYan Zheng bytenr, num_bytes); 181383d4cfd4SJosef Bacik if (ret) { 181483d4cfd4SJosef Bacik /* 181583d4cfd4SJosef Bacik * Don't have to abort since we've not changed anything 181683d4cfd4SJosef Bacik * in the file extent yet. 181783d4cfd4SJosef Bacik */ 181883d4cfd4SJosef Bacik break; 18193fd0a558SYan, Zheng } 18205d4f98a2SYan Zheng 18215d4f98a2SYan Zheng btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr); 18225d4f98a2SYan Zheng dirty = 1; 18235d4f98a2SYan Zheng 18245d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 182582fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 182682fa113fSQu Wenruo num_bytes, parent); 182782fa113fSQu Wenruo ref.real_root = root->root_key.objectid; 182882fa113fSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1829b06c4bf5SFilipe Manana key.objectid, key.offset); 183082fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 183183d4cfd4SJosef Bacik if (ret) { 183266642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 183383d4cfd4SJosef Bacik break; 183483d4cfd4SJosef Bacik } 18355d4f98a2SYan Zheng 1836ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, 1837ffd4bb2aSQu Wenruo num_bytes, parent); 1838ffd4bb2aSQu Wenruo ref.real_root = root->root_key.objectid; 1839ffd4bb2aSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1840b06c4bf5SFilipe Manana key.objectid, key.offset); 1841ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 184283d4cfd4SJosef Bacik if (ret) { 184366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 184483d4cfd4SJosef Bacik break; 184583d4cfd4SJosef Bacik } 18465d4f98a2SYan Zheng } 18475d4f98a2SYan Zheng if (dirty) 18485d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 18493fd0a558SYan, Zheng if (inode) 18503fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 185183d4cfd4SJosef Bacik return ret; 18525d4f98a2SYan Zheng } 18535d4f98a2SYan Zheng 18545d4f98a2SYan Zheng static noinline_for_stack 18555d4f98a2SYan Zheng int memcmp_node_keys(struct extent_buffer *eb, int slot, 18565d4f98a2SYan Zheng struct btrfs_path *path, int level) 18575d4f98a2SYan Zheng { 18585d4f98a2SYan Zheng struct btrfs_disk_key key1; 18595d4f98a2SYan Zheng struct btrfs_disk_key key2; 18605d4f98a2SYan Zheng btrfs_node_key(eb, &key1, slot); 18615d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &key2, path->slots[level]); 18625d4f98a2SYan Zheng return memcmp(&key1, &key2, sizeof(key1)); 18635d4f98a2SYan Zheng } 18645d4f98a2SYan Zheng 18655d4f98a2SYan Zheng /* 18665d4f98a2SYan Zheng * try to replace tree blocks in fs tree with the new blocks 18675d4f98a2SYan Zheng * in reloc tree. tree blocks haven't been modified since the 18685d4f98a2SYan Zheng * reloc tree was create can be replaced. 18695d4f98a2SYan Zheng * 18705d4f98a2SYan Zheng * if a block was replaced, level of the block + 1 is returned. 18715d4f98a2SYan Zheng * if no block got replaced, 0 is returned. if there are other 18725d4f98a2SYan Zheng * errors, a negative error number is returned. 18735d4f98a2SYan Zheng */ 18743fd0a558SYan, Zheng static noinline_for_stack 18753d0174f7SQu Wenruo int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc, 18765d4f98a2SYan Zheng struct btrfs_root *dest, struct btrfs_root *src, 18775d4f98a2SYan Zheng struct btrfs_path *path, struct btrfs_key *next_key, 18785d4f98a2SYan Zheng int lowest_level, int max_level) 18795d4f98a2SYan Zheng { 18800b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = dest->fs_info; 18815d4f98a2SYan Zheng struct extent_buffer *eb; 18825d4f98a2SYan Zheng struct extent_buffer *parent; 188382fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 18845d4f98a2SYan Zheng struct btrfs_key key; 18855d4f98a2SYan Zheng u64 old_bytenr; 18865d4f98a2SYan Zheng u64 new_bytenr; 18875d4f98a2SYan Zheng u64 old_ptr_gen; 18885d4f98a2SYan Zheng u64 new_ptr_gen; 18895d4f98a2SYan Zheng u64 last_snapshot; 18905d4f98a2SYan Zheng u32 blocksize; 18913fd0a558SYan, Zheng int cow = 0; 18925d4f98a2SYan Zheng int level; 18935d4f98a2SYan Zheng int ret; 18945d4f98a2SYan Zheng int slot; 18955d4f98a2SYan Zheng 18965d4f98a2SYan Zheng BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 18975d4f98a2SYan Zheng BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID); 18985d4f98a2SYan Zheng 18995d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&src->root_item); 19003fd0a558SYan, Zheng again: 19015d4f98a2SYan Zheng slot = path->slots[lowest_level]; 19025d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot); 19035d4f98a2SYan Zheng 19045d4f98a2SYan Zheng eb = btrfs_lock_root_node(dest); 19058bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 19065d4f98a2SYan Zheng level = btrfs_header_level(eb); 19075d4f98a2SYan Zheng 19085d4f98a2SYan Zheng if (level < lowest_level) { 19095d4f98a2SYan Zheng btrfs_tree_unlock(eb); 19105d4f98a2SYan Zheng free_extent_buffer(eb); 19115d4f98a2SYan Zheng return 0; 19125d4f98a2SYan Zheng } 19135d4f98a2SYan Zheng 19143fd0a558SYan, Zheng if (cow) { 19155d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb); 19165d4f98a2SYan Zheng BUG_ON(ret); 19173fd0a558SYan, Zheng } 19188bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 19195d4f98a2SYan Zheng 19205d4f98a2SYan Zheng if (next_key) { 19215d4f98a2SYan Zheng next_key->objectid = (u64)-1; 19225d4f98a2SYan Zheng next_key->type = (u8)-1; 19235d4f98a2SYan Zheng next_key->offset = (u64)-1; 19245d4f98a2SYan Zheng } 19255d4f98a2SYan Zheng 19265d4f98a2SYan Zheng parent = eb; 19275d4f98a2SYan Zheng while (1) { 1928581c1760SQu Wenruo struct btrfs_key first_key; 1929581c1760SQu Wenruo 19305d4f98a2SYan Zheng level = btrfs_header_level(parent); 19315d4f98a2SYan Zheng BUG_ON(level < lowest_level); 19325d4f98a2SYan Zheng 19335d4f98a2SYan Zheng ret = btrfs_bin_search(parent, &key, level, &slot); 1934cbca7d59SFilipe Manana if (ret < 0) 1935cbca7d59SFilipe Manana break; 19365d4f98a2SYan Zheng if (ret && slot > 0) 19375d4f98a2SYan Zheng slot--; 19385d4f98a2SYan Zheng 19395d4f98a2SYan Zheng if (next_key && slot + 1 < btrfs_header_nritems(parent)) 19405d4f98a2SYan Zheng btrfs_node_key_to_cpu(parent, next_key, slot + 1); 19415d4f98a2SYan Zheng 19425d4f98a2SYan Zheng old_bytenr = btrfs_node_blockptr(parent, slot); 19430b246afaSJeff Mahoney blocksize = fs_info->nodesize; 19445d4f98a2SYan Zheng old_ptr_gen = btrfs_node_ptr_generation(parent, slot); 194517515f1bSQu Wenruo btrfs_node_key_to_cpu(parent, &first_key, slot); 19465d4f98a2SYan Zheng 19475d4f98a2SYan Zheng if (level <= max_level) { 19485d4f98a2SYan Zheng eb = path->nodes[level]; 19495d4f98a2SYan Zheng new_bytenr = btrfs_node_blockptr(eb, 19505d4f98a2SYan Zheng path->slots[level]); 19515d4f98a2SYan Zheng new_ptr_gen = btrfs_node_ptr_generation(eb, 19525d4f98a2SYan Zheng path->slots[level]); 19535d4f98a2SYan Zheng } else { 19545d4f98a2SYan Zheng new_bytenr = 0; 19555d4f98a2SYan Zheng new_ptr_gen = 0; 19565d4f98a2SYan Zheng } 19575d4f98a2SYan Zheng 1958fae7f21cSDulshani Gunawardhana if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) { 19595d4f98a2SYan Zheng ret = level; 19605d4f98a2SYan Zheng break; 19615d4f98a2SYan Zheng } 19625d4f98a2SYan Zheng 19635d4f98a2SYan Zheng if (new_bytenr == 0 || old_ptr_gen > last_snapshot || 19645d4f98a2SYan Zheng memcmp_node_keys(parent, slot, path, level)) { 19653fd0a558SYan, Zheng if (level <= lowest_level) { 19665d4f98a2SYan Zheng ret = 0; 19675d4f98a2SYan Zheng break; 19685d4f98a2SYan Zheng } 19695d4f98a2SYan Zheng 1970581c1760SQu Wenruo eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen, 1971581c1760SQu Wenruo level - 1, &first_key); 197264c043deSLiu Bo if (IS_ERR(eb)) { 197364c043deSLiu Bo ret = PTR_ERR(eb); 1974264813acSLiu Bo break; 197564c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 197664c043deSLiu Bo ret = -EIO; 1977416bc658SJosef Bacik free_extent_buffer(eb); 1978379cde74SStefan Behrens break; 1979416bc658SJosef Bacik } 19805d4f98a2SYan Zheng btrfs_tree_lock(eb); 19813fd0a558SYan, Zheng if (cow) { 19825d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, parent, 19835d4f98a2SYan Zheng slot, &eb); 19845d4f98a2SYan Zheng BUG_ON(ret); 19855d4f98a2SYan Zheng } 19868bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 19875d4f98a2SYan Zheng 19885d4f98a2SYan Zheng btrfs_tree_unlock(parent); 19895d4f98a2SYan Zheng free_extent_buffer(parent); 19905d4f98a2SYan Zheng 19915d4f98a2SYan Zheng parent = eb; 19925d4f98a2SYan Zheng continue; 19935d4f98a2SYan Zheng } 19945d4f98a2SYan Zheng 19953fd0a558SYan, Zheng if (!cow) { 19963fd0a558SYan, Zheng btrfs_tree_unlock(parent); 19973fd0a558SYan, Zheng free_extent_buffer(parent); 19983fd0a558SYan, Zheng cow = 1; 19993fd0a558SYan, Zheng goto again; 20003fd0a558SYan, Zheng } 20013fd0a558SYan, Zheng 20025d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 20035d4f98a2SYan Zheng path->slots[level]); 2004b3b4aa74SDavid Sterba btrfs_release_path(path); 20055d4f98a2SYan Zheng 20065d4f98a2SYan Zheng path->lowest_level = level; 20075d4f98a2SYan Zheng ret = btrfs_search_slot(trans, src, &key, path, 0, 1); 20085d4f98a2SYan Zheng path->lowest_level = 0; 20095d4f98a2SYan Zheng BUG_ON(ret); 20105d4f98a2SYan Zheng 20115d4f98a2SYan Zheng /* 2012824d8dffSQu Wenruo * Info qgroup to trace both subtrees. 2013824d8dffSQu Wenruo * 2014824d8dffSQu Wenruo * We must trace both trees. 2015824d8dffSQu Wenruo * 1) Tree reloc subtree 2016824d8dffSQu Wenruo * If not traced, we will leak data numbers 2017824d8dffSQu Wenruo * 2) Fs subtree 2018824d8dffSQu Wenruo * If not traced, we will double count old data 2019f616f5cdSQu Wenruo * 2020f616f5cdSQu Wenruo * We don't scan the subtree right now, but only record 2021f616f5cdSQu Wenruo * the swapped tree blocks. 2022f616f5cdSQu Wenruo * The real subtree rescan is delayed until we have new 2023f616f5cdSQu Wenruo * CoW on the subtree root node before transaction commit. 2024824d8dffSQu Wenruo */ 2025370a11b8SQu Wenruo ret = btrfs_qgroup_add_swapped_blocks(trans, dest, 2026370a11b8SQu Wenruo rc->block_group, parent, slot, 2027370a11b8SQu Wenruo path->nodes[level], path->slots[level], 2028370a11b8SQu Wenruo last_snapshot); 2029370a11b8SQu Wenruo if (ret < 0) 2030370a11b8SQu Wenruo break; 2031824d8dffSQu Wenruo /* 20325d4f98a2SYan Zheng * swap blocks in fs tree and reloc tree. 20335d4f98a2SYan Zheng */ 20345d4f98a2SYan Zheng btrfs_set_node_blockptr(parent, slot, new_bytenr); 20355d4f98a2SYan Zheng btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen); 20365d4f98a2SYan Zheng btrfs_mark_buffer_dirty(parent); 20375d4f98a2SYan Zheng 20385d4f98a2SYan Zheng btrfs_set_node_blockptr(path->nodes[level], 20395d4f98a2SYan Zheng path->slots[level], old_bytenr); 20405d4f98a2SYan Zheng btrfs_set_node_ptr_generation(path->nodes[level], 20415d4f98a2SYan Zheng path->slots[level], old_ptr_gen); 20425d4f98a2SYan Zheng btrfs_mark_buffer_dirty(path->nodes[level]); 20435d4f98a2SYan Zheng 204482fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr, 204582fa113fSQu Wenruo blocksize, path->nodes[level]->start); 204682fa113fSQu Wenruo ref.skip_qgroup = true; 204782fa113fSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid); 204882fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 20495d4f98a2SYan Zheng BUG_ON(ret); 205082fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 205182fa113fSQu Wenruo blocksize, 0); 205282fa113fSQu Wenruo ref.skip_qgroup = true; 205382fa113fSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid); 205482fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 20555d4f98a2SYan Zheng BUG_ON(ret); 20565d4f98a2SYan Zheng 2057ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr, 2058ffd4bb2aSQu Wenruo blocksize, path->nodes[level]->start); 2059ffd4bb2aSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid); 2060ffd4bb2aSQu Wenruo ref.skip_qgroup = true; 2061ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 20625d4f98a2SYan Zheng BUG_ON(ret); 20635d4f98a2SYan Zheng 2064ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr, 2065ffd4bb2aSQu Wenruo blocksize, 0); 2066ffd4bb2aSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid); 2067ffd4bb2aSQu Wenruo ref.skip_qgroup = true; 2068ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 20695d4f98a2SYan Zheng BUG_ON(ret); 20705d4f98a2SYan Zheng 20715d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 20725d4f98a2SYan Zheng 20735d4f98a2SYan Zheng ret = level; 20745d4f98a2SYan Zheng break; 20755d4f98a2SYan Zheng } 20765d4f98a2SYan Zheng btrfs_tree_unlock(parent); 20775d4f98a2SYan Zheng free_extent_buffer(parent); 20785d4f98a2SYan Zheng return ret; 20795d4f98a2SYan Zheng } 20805d4f98a2SYan Zheng 20815d4f98a2SYan Zheng /* 20825d4f98a2SYan Zheng * helper to find next relocated block in reloc tree 20835d4f98a2SYan Zheng */ 20845d4f98a2SYan Zheng static noinline_for_stack 20855d4f98a2SYan Zheng int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 20865d4f98a2SYan Zheng int *level) 20875d4f98a2SYan Zheng { 20885d4f98a2SYan Zheng struct extent_buffer *eb; 20895d4f98a2SYan Zheng int i; 20905d4f98a2SYan Zheng u64 last_snapshot; 20915d4f98a2SYan Zheng u32 nritems; 20925d4f98a2SYan Zheng 20935d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 20945d4f98a2SYan Zheng 20955d4f98a2SYan Zheng for (i = 0; i < *level; i++) { 20965d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 20975d4f98a2SYan Zheng path->nodes[i] = NULL; 20985d4f98a2SYan Zheng } 20995d4f98a2SYan Zheng 21005d4f98a2SYan Zheng for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) { 21015d4f98a2SYan Zheng eb = path->nodes[i]; 21025d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 21035d4f98a2SYan Zheng while (path->slots[i] + 1 < nritems) { 21045d4f98a2SYan Zheng path->slots[i]++; 21055d4f98a2SYan Zheng if (btrfs_node_ptr_generation(eb, path->slots[i]) <= 21065d4f98a2SYan Zheng last_snapshot) 21075d4f98a2SYan Zheng continue; 21085d4f98a2SYan Zheng 21095d4f98a2SYan Zheng *level = i; 21105d4f98a2SYan Zheng return 0; 21115d4f98a2SYan Zheng } 21125d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 21135d4f98a2SYan Zheng path->nodes[i] = NULL; 21145d4f98a2SYan Zheng } 21155d4f98a2SYan Zheng return 1; 21165d4f98a2SYan Zheng } 21175d4f98a2SYan Zheng 21185d4f98a2SYan Zheng /* 21195d4f98a2SYan Zheng * walk down reloc tree to find relocated block of lowest level 21205d4f98a2SYan Zheng */ 21215d4f98a2SYan Zheng static noinline_for_stack 21225d4f98a2SYan Zheng int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 21235d4f98a2SYan Zheng int *level) 21245d4f98a2SYan Zheng { 21252ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 21265d4f98a2SYan Zheng struct extent_buffer *eb = NULL; 21275d4f98a2SYan Zheng int i; 21285d4f98a2SYan Zheng u64 bytenr; 21295d4f98a2SYan Zheng u64 ptr_gen = 0; 21305d4f98a2SYan Zheng u64 last_snapshot; 21315d4f98a2SYan Zheng u32 nritems; 21325d4f98a2SYan Zheng 21335d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 21345d4f98a2SYan Zheng 21355d4f98a2SYan Zheng for (i = *level; i > 0; i--) { 2136581c1760SQu Wenruo struct btrfs_key first_key; 2137581c1760SQu Wenruo 21385d4f98a2SYan Zheng eb = path->nodes[i]; 21395d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 21405d4f98a2SYan Zheng while (path->slots[i] < nritems) { 21415d4f98a2SYan Zheng ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]); 21425d4f98a2SYan Zheng if (ptr_gen > last_snapshot) 21435d4f98a2SYan Zheng break; 21445d4f98a2SYan Zheng path->slots[i]++; 21455d4f98a2SYan Zheng } 21465d4f98a2SYan Zheng if (path->slots[i] >= nritems) { 21475d4f98a2SYan Zheng if (i == *level) 21485d4f98a2SYan Zheng break; 21495d4f98a2SYan Zheng *level = i + 1; 21505d4f98a2SYan Zheng return 0; 21515d4f98a2SYan Zheng } 21525d4f98a2SYan Zheng if (i == 1) { 21535d4f98a2SYan Zheng *level = i; 21545d4f98a2SYan Zheng return 0; 21555d4f98a2SYan Zheng } 21565d4f98a2SYan Zheng 21575d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(eb, path->slots[i]); 2158581c1760SQu Wenruo btrfs_node_key_to_cpu(eb, &first_key, path->slots[i]); 2159581c1760SQu Wenruo eb = read_tree_block(fs_info, bytenr, ptr_gen, i - 1, 2160581c1760SQu Wenruo &first_key); 216164c043deSLiu Bo if (IS_ERR(eb)) { 216264c043deSLiu Bo return PTR_ERR(eb); 216364c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2164416bc658SJosef Bacik free_extent_buffer(eb); 2165416bc658SJosef Bacik return -EIO; 2166416bc658SJosef Bacik } 21675d4f98a2SYan Zheng BUG_ON(btrfs_header_level(eb) != i - 1); 21685d4f98a2SYan Zheng path->nodes[i - 1] = eb; 21695d4f98a2SYan Zheng path->slots[i - 1] = 0; 21705d4f98a2SYan Zheng } 21715d4f98a2SYan Zheng return 1; 21725d4f98a2SYan Zheng } 21735d4f98a2SYan Zheng 21745d4f98a2SYan Zheng /* 21755d4f98a2SYan Zheng * invalidate extent cache for file extents whose key in range of 21765d4f98a2SYan Zheng * [min_key, max_key) 21775d4f98a2SYan Zheng */ 21785d4f98a2SYan Zheng static int invalidate_extent_cache(struct btrfs_root *root, 21795d4f98a2SYan Zheng struct btrfs_key *min_key, 21805d4f98a2SYan Zheng struct btrfs_key *max_key) 21815d4f98a2SYan Zheng { 21820b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 21835d4f98a2SYan Zheng struct inode *inode = NULL; 21845d4f98a2SYan Zheng u64 objectid; 21855d4f98a2SYan Zheng u64 start, end; 218633345d01SLi Zefan u64 ino; 21875d4f98a2SYan Zheng 21885d4f98a2SYan Zheng objectid = min_key->objectid; 21895d4f98a2SYan Zheng while (1) { 21905d4f98a2SYan Zheng cond_resched(); 21915d4f98a2SYan Zheng iput(inode); 21925d4f98a2SYan Zheng 21935d4f98a2SYan Zheng if (objectid > max_key->objectid) 21945d4f98a2SYan Zheng break; 21955d4f98a2SYan Zheng 21965d4f98a2SYan Zheng inode = find_next_inode(root, objectid); 21975d4f98a2SYan Zheng if (!inode) 21985d4f98a2SYan Zheng break; 21994a0cc7caSNikolay Borisov ino = btrfs_ino(BTRFS_I(inode)); 22005d4f98a2SYan Zheng 220133345d01SLi Zefan if (ino > max_key->objectid) { 22025d4f98a2SYan Zheng iput(inode); 22035d4f98a2SYan Zheng break; 22045d4f98a2SYan Zheng } 22055d4f98a2SYan Zheng 220633345d01SLi Zefan objectid = ino + 1; 22075d4f98a2SYan Zheng if (!S_ISREG(inode->i_mode)) 22085d4f98a2SYan Zheng continue; 22095d4f98a2SYan Zheng 221033345d01SLi Zefan if (unlikely(min_key->objectid == ino)) { 22115d4f98a2SYan Zheng if (min_key->type > BTRFS_EXTENT_DATA_KEY) 22125d4f98a2SYan Zheng continue; 22135d4f98a2SYan Zheng if (min_key->type < BTRFS_EXTENT_DATA_KEY) 22145d4f98a2SYan Zheng start = 0; 22155d4f98a2SYan Zheng else { 22165d4f98a2SYan Zheng start = min_key->offset; 22170b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize)); 22185d4f98a2SYan Zheng } 22195d4f98a2SYan Zheng } else { 22205d4f98a2SYan Zheng start = 0; 22215d4f98a2SYan Zheng } 22225d4f98a2SYan Zheng 222333345d01SLi Zefan if (unlikely(max_key->objectid == ino)) { 22245d4f98a2SYan Zheng if (max_key->type < BTRFS_EXTENT_DATA_KEY) 22255d4f98a2SYan Zheng continue; 22265d4f98a2SYan Zheng if (max_key->type > BTRFS_EXTENT_DATA_KEY) { 22275d4f98a2SYan Zheng end = (u64)-1; 22285d4f98a2SYan Zheng } else { 22295d4f98a2SYan Zheng if (max_key->offset == 0) 22305d4f98a2SYan Zheng continue; 22315d4f98a2SYan Zheng end = max_key->offset; 22320b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 22335d4f98a2SYan Zheng end--; 22345d4f98a2SYan Zheng } 22355d4f98a2SYan Zheng } else { 22365d4f98a2SYan Zheng end = (u64)-1; 22375d4f98a2SYan Zheng } 22385d4f98a2SYan Zheng 22395d4f98a2SYan Zheng /* the lock_extent waits for readpage to complete */ 2240d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 2241dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1); 2242d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 22435d4f98a2SYan Zheng } 22445d4f98a2SYan Zheng return 0; 22455d4f98a2SYan Zheng } 22465d4f98a2SYan Zheng 22475d4f98a2SYan Zheng static int find_next_key(struct btrfs_path *path, int level, 22485d4f98a2SYan Zheng struct btrfs_key *key) 22495d4f98a2SYan Zheng 22505d4f98a2SYan Zheng { 22515d4f98a2SYan Zheng while (level < BTRFS_MAX_LEVEL) { 22525d4f98a2SYan Zheng if (!path->nodes[level]) 22535d4f98a2SYan Zheng break; 22545d4f98a2SYan Zheng if (path->slots[level] + 1 < 22555d4f98a2SYan Zheng btrfs_header_nritems(path->nodes[level])) { 22565d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], key, 22575d4f98a2SYan Zheng path->slots[level] + 1); 22585d4f98a2SYan Zheng return 0; 22595d4f98a2SYan Zheng } 22605d4f98a2SYan Zheng level++; 22615d4f98a2SYan Zheng } 22625d4f98a2SYan Zheng return 1; 22635d4f98a2SYan Zheng } 22645d4f98a2SYan Zheng 22655d4f98a2SYan Zheng /* 2266d2311e69SQu Wenruo * Insert current subvolume into reloc_control::dirty_subvol_roots 2267d2311e69SQu Wenruo */ 2268d2311e69SQu Wenruo static void insert_dirty_subvol(struct btrfs_trans_handle *trans, 2269d2311e69SQu Wenruo struct reloc_control *rc, 2270d2311e69SQu Wenruo struct btrfs_root *root) 2271d2311e69SQu Wenruo { 2272d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 2273d2311e69SQu Wenruo struct btrfs_root_item *reloc_root_item; 2274d2311e69SQu Wenruo 2275d2311e69SQu Wenruo /* @root must be a subvolume tree root with a valid reloc tree */ 2276d2311e69SQu Wenruo ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 2277d2311e69SQu Wenruo ASSERT(reloc_root); 2278d2311e69SQu Wenruo 2279d2311e69SQu Wenruo reloc_root_item = &reloc_root->root_item; 2280d2311e69SQu Wenruo memset(&reloc_root_item->drop_progress, 0, 2281d2311e69SQu Wenruo sizeof(reloc_root_item->drop_progress)); 2282d2311e69SQu Wenruo reloc_root_item->drop_level = 0; 2283d2311e69SQu Wenruo btrfs_set_root_refs(reloc_root_item, 0); 2284d2311e69SQu Wenruo btrfs_update_reloc_root(trans, root); 2285d2311e69SQu Wenruo 2286d2311e69SQu Wenruo if (list_empty(&root->reloc_dirty_list)) { 228700246528SJosef Bacik btrfs_grab_root(root); 2288d2311e69SQu Wenruo list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots); 2289d2311e69SQu Wenruo } 2290d2311e69SQu Wenruo } 2291d2311e69SQu Wenruo 2292d2311e69SQu Wenruo static int clean_dirty_subvols(struct reloc_control *rc) 2293d2311e69SQu Wenruo { 2294d2311e69SQu Wenruo struct btrfs_root *root; 2295d2311e69SQu Wenruo struct btrfs_root *next; 2296d2311e69SQu Wenruo int ret = 0; 229730d40577SQu Wenruo int ret2; 2298d2311e69SQu Wenruo 2299d2311e69SQu Wenruo list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots, 2300d2311e69SQu Wenruo reloc_dirty_list) { 230130d40577SQu Wenruo if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 230230d40577SQu Wenruo /* Merged subvolume, cleanup its reloc root */ 2303d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 2304d2311e69SQu Wenruo 2305d2311e69SQu Wenruo list_del_init(&root->reloc_dirty_list); 2306d2311e69SQu Wenruo root->reloc_root = NULL; 23076282675eSQu Wenruo /* 23086282675eSQu Wenruo * Need barrier to ensure clear_bit() only happens after 23096282675eSQu Wenruo * root->reloc_root = NULL. Pairs with have_reloc_root. 23106282675eSQu Wenruo */ 23116282675eSQu Wenruo smp_wmb(); 23121fac4a54SQu Wenruo clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 2313f28de8d8SJosef Bacik if (reloc_root) { 2314f44deb74SJosef Bacik /* 2315f44deb74SJosef Bacik * btrfs_drop_snapshot drops our ref we hold for 2316f44deb74SJosef Bacik * ->reloc_root. If it fails however we must 2317f44deb74SJosef Bacik * drop the ref ourselves. 2318f44deb74SJosef Bacik */ 2319f28de8d8SJosef Bacik ret2 = btrfs_drop_snapshot(reloc_root, 0, 1); 2320f44deb74SJosef Bacik if (ret2 < 0) { 2321f44deb74SJosef Bacik btrfs_put_root(reloc_root); 2322f44deb74SJosef Bacik if (!ret) 2323f28de8d8SJosef Bacik ret = ret2; 2324f28de8d8SJosef Bacik } 2325f44deb74SJosef Bacik } 232600246528SJosef Bacik btrfs_put_root(root); 232730d40577SQu Wenruo } else { 232830d40577SQu Wenruo /* Orphan reloc tree, just clean it up */ 23290078a9f9SNikolay Borisov ret2 = btrfs_drop_snapshot(root, 0, 1); 2330f44deb74SJosef Bacik if (ret2 < 0) { 2331f44deb74SJosef Bacik btrfs_put_root(root); 2332f44deb74SJosef Bacik if (!ret) 233330d40577SQu Wenruo ret = ret2; 233430d40577SQu Wenruo } 2335d2311e69SQu Wenruo } 2336f44deb74SJosef Bacik } 2337d2311e69SQu Wenruo return ret; 2338d2311e69SQu Wenruo } 2339d2311e69SQu Wenruo 2340d2311e69SQu Wenruo /* 23415d4f98a2SYan Zheng * merge the relocated tree blocks in reloc tree with corresponding 23425d4f98a2SYan Zheng * fs tree. 23435d4f98a2SYan Zheng */ 23445d4f98a2SYan Zheng static noinline_for_stack int merge_reloc_root(struct reloc_control *rc, 23455d4f98a2SYan Zheng struct btrfs_root *root) 23465d4f98a2SYan Zheng { 23470b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 23485d4f98a2SYan Zheng struct btrfs_key key; 23495d4f98a2SYan Zheng struct btrfs_key next_key; 23509e6a0c52SJosef Bacik struct btrfs_trans_handle *trans = NULL; 23515d4f98a2SYan Zheng struct btrfs_root *reloc_root; 23525d4f98a2SYan Zheng struct btrfs_root_item *root_item; 23535d4f98a2SYan Zheng struct btrfs_path *path; 23543fd0a558SYan, Zheng struct extent_buffer *leaf; 23555d4f98a2SYan Zheng int level; 23565d4f98a2SYan Zheng int max_level; 23575d4f98a2SYan Zheng int replaced = 0; 23585d4f98a2SYan Zheng int ret; 23595d4f98a2SYan Zheng int err = 0; 23603fd0a558SYan, Zheng u32 min_reserved; 23615d4f98a2SYan Zheng 23625d4f98a2SYan Zheng path = btrfs_alloc_path(); 23635d4f98a2SYan Zheng if (!path) 23645d4f98a2SYan Zheng return -ENOMEM; 2365e4058b54SDavid Sterba path->reada = READA_FORWARD; 23665d4f98a2SYan Zheng 23675d4f98a2SYan Zheng reloc_root = root->reloc_root; 23685d4f98a2SYan Zheng root_item = &reloc_root->root_item; 23695d4f98a2SYan Zheng 23705d4f98a2SYan Zheng if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { 23715d4f98a2SYan Zheng level = btrfs_root_level(root_item); 237267439dadSDavid Sterba atomic_inc(&reloc_root->node->refs); 23735d4f98a2SYan Zheng path->nodes[level] = reloc_root->node; 23745d4f98a2SYan Zheng path->slots[level] = 0; 23755d4f98a2SYan Zheng } else { 23765d4f98a2SYan Zheng btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); 23775d4f98a2SYan Zheng 23785d4f98a2SYan Zheng level = root_item->drop_level; 23795d4f98a2SYan Zheng BUG_ON(level == 0); 23805d4f98a2SYan Zheng path->lowest_level = level; 23815d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0); 238233c66f43SYan Zheng path->lowest_level = 0; 23835d4f98a2SYan Zheng if (ret < 0) { 23845d4f98a2SYan Zheng btrfs_free_path(path); 23855d4f98a2SYan Zheng return ret; 23865d4f98a2SYan Zheng } 23875d4f98a2SYan Zheng 23885d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &next_key, 23895d4f98a2SYan Zheng path->slots[level]); 23905d4f98a2SYan Zheng WARN_ON(memcmp(&key, &next_key, sizeof(key))); 23915d4f98a2SYan Zheng 23925d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 23935d4f98a2SYan Zheng } 23945d4f98a2SYan Zheng 23950b246afaSJeff Mahoney min_reserved = fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 23965d4f98a2SYan Zheng memset(&next_key, 0, sizeof(next_key)); 23975d4f98a2SYan Zheng 23985d4f98a2SYan Zheng while (1) { 239908e007d2SMiao Xie ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved, 240008e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 24013fd0a558SYan, Zheng if (ret) { 24029e6a0c52SJosef Bacik err = ret; 24039e6a0c52SJosef Bacik goto out; 24043fd0a558SYan, Zheng } 24059e6a0c52SJosef Bacik trans = btrfs_start_transaction(root, 0); 24069e6a0c52SJosef Bacik if (IS_ERR(trans)) { 24079e6a0c52SJosef Bacik err = PTR_ERR(trans); 24089e6a0c52SJosef Bacik trans = NULL; 24099e6a0c52SJosef Bacik goto out; 24109e6a0c52SJosef Bacik } 24112abc726aSJosef Bacik 24122abc726aSJosef Bacik /* 24132abc726aSJosef Bacik * At this point we no longer have a reloc_control, so we can't 24142abc726aSJosef Bacik * depend on btrfs_init_reloc_root to update our last_trans. 24152abc726aSJosef Bacik * 24162abc726aSJosef Bacik * But that's ok, we started the trans handle on our 24172abc726aSJosef Bacik * corresponding fs_root, which means it's been added to the 24182abc726aSJosef Bacik * dirty list. At commit time we'll still call 24192abc726aSJosef Bacik * btrfs_update_reloc_root() and update our root item 24202abc726aSJosef Bacik * appropriately. 24212abc726aSJosef Bacik */ 24222abc726aSJosef Bacik reloc_root->last_trans = trans->transid; 24239e6a0c52SJosef Bacik trans->block_rsv = rc->block_rsv; 24243fd0a558SYan, Zheng 24253fd0a558SYan, Zheng replaced = 0; 24265d4f98a2SYan Zheng max_level = level; 24275d4f98a2SYan Zheng 24285d4f98a2SYan Zheng ret = walk_down_reloc_tree(reloc_root, path, &level); 24295d4f98a2SYan Zheng if (ret < 0) { 24305d4f98a2SYan Zheng err = ret; 24315d4f98a2SYan Zheng goto out; 24325d4f98a2SYan Zheng } 24335d4f98a2SYan Zheng if (ret > 0) 24345d4f98a2SYan Zheng break; 24355d4f98a2SYan Zheng 24365d4f98a2SYan Zheng if (!find_next_key(path, level, &key) && 24375d4f98a2SYan Zheng btrfs_comp_cpu_keys(&next_key, &key) >= 0) { 24385d4f98a2SYan Zheng ret = 0; 24395d4f98a2SYan Zheng } else { 24403d0174f7SQu Wenruo ret = replace_path(trans, rc, root, reloc_root, path, 24413fd0a558SYan, Zheng &next_key, level, max_level); 24425d4f98a2SYan Zheng } 24435d4f98a2SYan Zheng if (ret < 0) { 24445d4f98a2SYan Zheng err = ret; 24455d4f98a2SYan Zheng goto out; 24465d4f98a2SYan Zheng } 24475d4f98a2SYan Zheng 24485d4f98a2SYan Zheng if (ret > 0) { 24495d4f98a2SYan Zheng level = ret; 24505d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 24515d4f98a2SYan Zheng path->slots[level]); 24525d4f98a2SYan Zheng replaced = 1; 24535d4f98a2SYan Zheng } 24545d4f98a2SYan Zheng 24555d4f98a2SYan Zheng ret = walk_up_reloc_tree(reloc_root, path, &level); 24565d4f98a2SYan Zheng if (ret > 0) 24575d4f98a2SYan Zheng break; 24585d4f98a2SYan Zheng 24595d4f98a2SYan Zheng BUG_ON(level == 0); 24605d4f98a2SYan Zheng /* 24615d4f98a2SYan Zheng * save the merging progress in the drop_progress. 24625d4f98a2SYan Zheng * this is OK since root refs == 1 in this case. 24635d4f98a2SYan Zheng */ 24645d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &root_item->drop_progress, 24655d4f98a2SYan Zheng path->slots[level]); 24665d4f98a2SYan Zheng root_item->drop_level = level; 24675d4f98a2SYan Zheng 24683a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 24699e6a0c52SJosef Bacik trans = NULL; 24705d4f98a2SYan Zheng 24712ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 24725d4f98a2SYan Zheng 24735d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 24745d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 24755d4f98a2SYan Zheng } 24765d4f98a2SYan Zheng 24775d4f98a2SYan Zheng /* 24785d4f98a2SYan Zheng * handle the case only one block in the fs tree need to be 24795d4f98a2SYan Zheng * relocated and the block is tree root. 24805d4f98a2SYan Zheng */ 24815d4f98a2SYan Zheng leaf = btrfs_lock_root_node(root); 24825d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf); 24835d4f98a2SYan Zheng btrfs_tree_unlock(leaf); 24845d4f98a2SYan Zheng free_extent_buffer(leaf); 24855d4f98a2SYan Zheng if (ret < 0) 24865d4f98a2SYan Zheng err = ret; 24875d4f98a2SYan Zheng out: 24885d4f98a2SYan Zheng btrfs_free_path(path); 24895d4f98a2SYan Zheng 2490d2311e69SQu Wenruo if (err == 0) 2491d2311e69SQu Wenruo insert_dirty_subvol(trans, rc, root); 24925d4f98a2SYan Zheng 24939e6a0c52SJosef Bacik if (trans) 24943a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 24955d4f98a2SYan Zheng 24962ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 24975d4f98a2SYan Zheng 24985d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 24995d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 25005d4f98a2SYan Zheng 25015d4f98a2SYan Zheng return err; 25025d4f98a2SYan Zheng } 25035d4f98a2SYan Zheng 25043fd0a558SYan, Zheng static noinline_for_stack 25053fd0a558SYan, Zheng int prepare_to_merge(struct reloc_control *rc, int err) 25065d4f98a2SYan Zheng { 25073fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 25080b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 25093fd0a558SYan, Zheng struct btrfs_root *reloc_root; 25105d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 25113fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 25123fd0a558SYan, Zheng u64 num_bytes = 0; 25133fd0a558SYan, Zheng int ret; 25143fd0a558SYan, Zheng 25150b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 25160b246afaSJeff Mahoney rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 25173fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated * 2; 25180b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 25197585717fSChris Mason 25203fd0a558SYan, Zheng again: 25213fd0a558SYan, Zheng if (!err) { 25223fd0a558SYan, Zheng num_bytes = rc->merging_rsv_size; 252308e007d2SMiao Xie ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes, 252408e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 25253fd0a558SYan, Zheng if (ret) 25263fd0a558SYan, Zheng err = ret; 25273fd0a558SYan, Zheng } 25283fd0a558SYan, Zheng 25297a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 25303612b495STsutomu Itoh if (IS_ERR(trans)) { 25313612b495STsutomu Itoh if (!err) 25322ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 253363f018beSNikolay Borisov num_bytes, NULL); 25343612b495STsutomu Itoh return PTR_ERR(trans); 25353612b495STsutomu Itoh } 25363fd0a558SYan, Zheng 25373fd0a558SYan, Zheng if (!err) { 25383fd0a558SYan, Zheng if (num_bytes != rc->merging_rsv_size) { 25393a45bb20SJeff Mahoney btrfs_end_transaction(trans); 25402ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 254163f018beSNikolay Borisov num_bytes, NULL); 25423fd0a558SYan, Zheng goto again; 25433fd0a558SYan, Zheng } 25443fd0a558SYan, Zheng } 25453fd0a558SYan, Zheng 25463fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 25473fd0a558SYan, Zheng 25483fd0a558SYan, Zheng while (!list_empty(&rc->reloc_roots)) { 25493fd0a558SYan, Zheng reloc_root = list_entry(rc->reloc_roots.next, 25503fd0a558SYan, Zheng struct btrfs_root, root_list); 25513fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 25523fd0a558SYan, Zheng 25530b246afaSJeff Mahoney root = read_fs_root(fs_info, reloc_root->root_key.offset); 25543fd0a558SYan, Zheng BUG_ON(IS_ERR(root)); 25553fd0a558SYan, Zheng BUG_ON(root->reloc_root != reloc_root); 25563fd0a558SYan, Zheng 25573fd0a558SYan, Zheng /* 25583fd0a558SYan, Zheng * set reference count to 1, so btrfs_recover_relocation 25593fd0a558SYan, Zheng * knows it should resumes merging 25603fd0a558SYan, Zheng */ 25613fd0a558SYan, Zheng if (!err) 25623fd0a558SYan, Zheng btrfs_set_root_refs(&reloc_root->root_item, 1); 25633fd0a558SYan, Zheng btrfs_update_reloc_root(trans, root); 25643fd0a558SYan, Zheng 25653fd0a558SYan, Zheng list_add(&reloc_root->root_list, &reloc_roots); 256600246528SJosef Bacik btrfs_put_root(root); 25673fd0a558SYan, Zheng } 25683fd0a558SYan, Zheng 25693fd0a558SYan, Zheng list_splice(&reloc_roots, &rc->reloc_roots); 25703fd0a558SYan, Zheng 25713fd0a558SYan, Zheng if (!err) 25723a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 25733fd0a558SYan, Zheng else 25743a45bb20SJeff Mahoney btrfs_end_transaction(trans); 25753fd0a558SYan, Zheng return err; 25763fd0a558SYan, Zheng } 25773fd0a558SYan, Zheng 25783fd0a558SYan, Zheng static noinline_for_stack 2579aca1bba6SLiu Bo void free_reloc_roots(struct list_head *list) 2580aca1bba6SLiu Bo { 2581aca1bba6SLiu Bo struct btrfs_root *reloc_root; 2582aca1bba6SLiu Bo 2583aca1bba6SLiu Bo while (!list_empty(list)) { 2584aca1bba6SLiu Bo reloc_root = list_entry(list->next, struct btrfs_root, 2585aca1bba6SLiu Bo root_list); 2586bb166d72SNaohiro Aota __del_reloc_root(reloc_root); 2587aca1bba6SLiu Bo } 2588aca1bba6SLiu Bo } 2589aca1bba6SLiu Bo 2590aca1bba6SLiu Bo static noinline_for_stack 259194404e82SDavid Sterba void merge_reloc_roots(struct reloc_control *rc) 25923fd0a558SYan, Zheng { 25930b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 25945d4f98a2SYan Zheng struct btrfs_root *root; 25955d4f98a2SYan Zheng struct btrfs_root *reloc_root; 25963fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 25973fd0a558SYan, Zheng int found = 0; 2598aca1bba6SLiu Bo int ret = 0; 25993fd0a558SYan, Zheng again: 26003fd0a558SYan, Zheng root = rc->extent_root; 26017585717fSChris Mason 26027585717fSChris Mason /* 26037585717fSChris Mason * this serializes us with btrfs_record_root_in_transaction, 26047585717fSChris Mason * we have to make sure nobody is in the middle of 26057585717fSChris Mason * adding their roots to the list while we are 26067585717fSChris Mason * doing this splice 26077585717fSChris Mason */ 26080b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 26093fd0a558SYan, Zheng list_splice_init(&rc->reloc_roots, &reloc_roots); 26100b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 26115d4f98a2SYan Zheng 26123fd0a558SYan, Zheng while (!list_empty(&reloc_roots)) { 26133fd0a558SYan, Zheng found = 1; 26143fd0a558SYan, Zheng reloc_root = list_entry(reloc_roots.next, 26153fd0a558SYan, Zheng struct btrfs_root, root_list); 26165d4f98a2SYan Zheng 26175d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 26180b246afaSJeff Mahoney root = read_fs_root(fs_info, 26195d4f98a2SYan Zheng reloc_root->root_key.offset); 26205d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 26215d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 26225d4f98a2SYan Zheng 26233fd0a558SYan, Zheng ret = merge_reloc_root(rc, root); 262400246528SJosef Bacik btrfs_put_root(root); 2625b37b39cdSJosef Bacik if (ret) { 262625e293c2SWang Shilong if (list_empty(&reloc_root->root_list)) 262725e293c2SWang Shilong list_add_tail(&reloc_root->root_list, 262825e293c2SWang Shilong &reloc_roots); 2629aca1bba6SLiu Bo goto out; 2630b37b39cdSJosef Bacik } 26313fd0a558SYan, Zheng } else { 26323fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 263330d40577SQu Wenruo /* Don't forget to queue this reloc root for cleanup */ 263430d40577SQu Wenruo list_add_tail(&reloc_root->reloc_dirty_list, 263530d40577SQu Wenruo &rc->dirty_subvol_roots); 26363fd0a558SYan, Zheng } 26375d4f98a2SYan Zheng } 26385d4f98a2SYan Zheng 26393fd0a558SYan, Zheng if (found) { 26403fd0a558SYan, Zheng found = 0; 26413fd0a558SYan, Zheng goto again; 26425d4f98a2SYan Zheng } 2643aca1bba6SLiu Bo out: 2644aca1bba6SLiu Bo if (ret) { 26450b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 2646aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 2647aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 2648467bb1d2SWang Shilong 2649467bb1d2SWang Shilong /* new reloc root may be added */ 26500b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 2651467bb1d2SWang Shilong list_splice_init(&rc->reloc_roots, &reloc_roots); 26520b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2653467bb1d2SWang Shilong if (!list_empty(&reloc_roots)) 2654467bb1d2SWang Shilong free_reloc_roots(&reloc_roots); 2655aca1bba6SLiu Bo } 2656aca1bba6SLiu Bo 26577b7b7431SJosef Bacik /* 26587b7b7431SJosef Bacik * We used to have 26597b7b7431SJosef Bacik * 26607b7b7431SJosef Bacik * BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); 26617b7b7431SJosef Bacik * 26627b7b7431SJosef Bacik * here, but it's wrong. If we fail to start the transaction in 26637b7b7431SJosef Bacik * prepare_to_merge() we will have only 0 ref reloc roots, none of which 26647b7b7431SJosef Bacik * have actually been removed from the reloc_root_tree rb tree. This is 26657b7b7431SJosef Bacik * fine because we're bailing here, and we hold a reference on the root 26667b7b7431SJosef Bacik * for the list that holds it, so these roots will be cleaned up when we 26677b7b7431SJosef Bacik * do the reloc_dirty_list afterwards. Meanwhile the root->reloc_root 26687b7b7431SJosef Bacik * will be cleaned up on unmount. 26697b7b7431SJosef Bacik * 26707b7b7431SJosef Bacik * The remaining nodes will be cleaned up by free_reloc_control. 26717b7b7431SJosef Bacik */ 26725d4f98a2SYan Zheng } 26735d4f98a2SYan Zheng 26745d4f98a2SYan Zheng static void free_block_list(struct rb_root *blocks) 26755d4f98a2SYan Zheng { 26765d4f98a2SYan Zheng struct tree_block *block; 26775d4f98a2SYan Zheng struct rb_node *rb_node; 26785d4f98a2SYan Zheng while ((rb_node = rb_first(blocks))) { 26795d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 26805d4f98a2SYan Zheng rb_erase(rb_node, blocks); 26815d4f98a2SYan Zheng kfree(block); 26825d4f98a2SYan Zheng } 26835d4f98a2SYan Zheng } 26845d4f98a2SYan Zheng 26855d4f98a2SYan Zheng static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans, 26865d4f98a2SYan Zheng struct btrfs_root *reloc_root) 26875d4f98a2SYan Zheng { 26880b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = reloc_root->fs_info; 26895d4f98a2SYan Zheng struct btrfs_root *root; 2690442b1ac5SJosef Bacik int ret; 26915d4f98a2SYan Zheng 26925d4f98a2SYan Zheng if (reloc_root->last_trans == trans->transid) 26935d4f98a2SYan Zheng return 0; 26945d4f98a2SYan Zheng 26950b246afaSJeff Mahoney root = read_fs_root(fs_info, reloc_root->root_key.offset); 26965d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 26975d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 2698442b1ac5SJosef Bacik ret = btrfs_record_root_in_trans(trans, root); 269900246528SJosef Bacik btrfs_put_root(root); 27005d4f98a2SYan Zheng 2701442b1ac5SJosef Bacik return ret; 27025d4f98a2SYan Zheng } 27035d4f98a2SYan Zheng 27043fd0a558SYan, Zheng static noinline_for_stack 27053fd0a558SYan, Zheng struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, 27063fd0a558SYan, Zheng struct reloc_control *rc, 27075d4f98a2SYan Zheng struct backref_node *node, 2708dc4103f9SWang Shilong struct backref_edge *edges[]) 27095d4f98a2SYan Zheng { 27105d4f98a2SYan Zheng struct backref_node *next; 27115d4f98a2SYan Zheng struct btrfs_root *root; 27123fd0a558SYan, Zheng int index = 0; 27133fd0a558SYan, Zheng 27145d4f98a2SYan Zheng next = node; 27155d4f98a2SYan Zheng while (1) { 27165d4f98a2SYan Zheng cond_resched(); 27175d4f98a2SYan Zheng next = walk_up_backref(next, edges, &index); 27185d4f98a2SYan Zheng root = next->root; 27193fd0a558SYan, Zheng BUG_ON(!root); 272027cdeb70SMiao Xie BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state)); 27215d4f98a2SYan Zheng 27225d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 27235d4f98a2SYan Zheng record_reloc_root_in_trans(trans, root); 27245d4f98a2SYan Zheng break; 27255d4f98a2SYan Zheng } 27265d4f98a2SYan Zheng 27275d4f98a2SYan Zheng btrfs_record_root_in_trans(trans, root); 27283fd0a558SYan, Zheng root = root->reloc_root; 27293fd0a558SYan, Zheng 27303fd0a558SYan, Zheng if (next->new_bytenr != root->node->start) { 27313fd0a558SYan, Zheng BUG_ON(next->new_bytenr); 27323fd0a558SYan, Zheng BUG_ON(!list_empty(&next->list)); 27333fd0a558SYan, Zheng next->new_bytenr = root->node->start; 273400246528SJosef Bacik btrfs_put_root(next->root); 273500246528SJosef Bacik next->root = btrfs_grab_root(root); 27360b530bc5SJosef Bacik ASSERT(next->root); 27373fd0a558SYan, Zheng list_add_tail(&next->list, 27383fd0a558SYan, Zheng &rc->backref_cache.changed); 27399569cc20SQu Wenruo mark_block_processed(rc, next); 27405d4f98a2SYan Zheng break; 27415d4f98a2SYan Zheng } 27425d4f98a2SYan Zheng 27433fd0a558SYan, Zheng WARN_ON(1); 27445d4f98a2SYan Zheng root = NULL; 27455d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 27465d4f98a2SYan Zheng if (!next || next->level <= node->level) 27475d4f98a2SYan Zheng break; 27485d4f98a2SYan Zheng } 27493fd0a558SYan, Zheng if (!root) 27503fd0a558SYan, Zheng return NULL; 27515d4f98a2SYan Zheng 27523fd0a558SYan, Zheng next = node; 27533fd0a558SYan, Zheng /* setup backref node path for btrfs_reloc_cow_block */ 27543fd0a558SYan, Zheng while (1) { 27553fd0a558SYan, Zheng rc->backref_cache.path[next->level] = next; 27563fd0a558SYan, Zheng if (--index < 0) 27573fd0a558SYan, Zheng break; 27583fd0a558SYan, Zheng next = edges[index]->node[UPPER]; 27593fd0a558SYan, Zheng } 27605d4f98a2SYan Zheng return root; 27615d4f98a2SYan Zheng } 27625d4f98a2SYan Zheng 27633fd0a558SYan, Zheng /* 27643fd0a558SYan, Zheng * select a tree root for relocation. return NULL if the block 27653fd0a558SYan, Zheng * is reference counted. we should use do_relocation() in this 27663fd0a558SYan, Zheng * case. return a tree root pointer if the block isn't reference 27673fd0a558SYan, Zheng * counted. return -ENOENT if the block is root of reloc tree. 27683fd0a558SYan, Zheng */ 27695d4f98a2SYan Zheng static noinline_for_stack 2770147d256eSZhaolei struct btrfs_root *select_one_root(struct backref_node *node) 27715d4f98a2SYan Zheng { 27723fd0a558SYan, Zheng struct backref_node *next; 27733fd0a558SYan, Zheng struct btrfs_root *root; 27743fd0a558SYan, Zheng struct btrfs_root *fs_root = NULL; 27755d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 27763fd0a558SYan, Zheng int index = 0; 27773fd0a558SYan, Zheng 27783fd0a558SYan, Zheng next = node; 27793fd0a558SYan, Zheng while (1) { 27803fd0a558SYan, Zheng cond_resched(); 27813fd0a558SYan, Zheng next = walk_up_backref(next, edges, &index); 27823fd0a558SYan, Zheng root = next->root; 27833fd0a558SYan, Zheng BUG_ON(!root); 27843fd0a558SYan, Zheng 278525985edcSLucas De Marchi /* no other choice for non-references counted tree */ 278627cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 27873fd0a558SYan, Zheng return root; 27883fd0a558SYan, Zheng 27893fd0a558SYan, Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) 27903fd0a558SYan, Zheng fs_root = root; 27913fd0a558SYan, Zheng 27923fd0a558SYan, Zheng if (next != node) 27933fd0a558SYan, Zheng return NULL; 27943fd0a558SYan, Zheng 27953fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 27963fd0a558SYan, Zheng if (!next || next->level <= node->level) 27973fd0a558SYan, Zheng break; 27983fd0a558SYan, Zheng } 27993fd0a558SYan, Zheng 28003fd0a558SYan, Zheng if (!fs_root) 28013fd0a558SYan, Zheng return ERR_PTR(-ENOENT); 28023fd0a558SYan, Zheng return fs_root; 28035d4f98a2SYan Zheng } 28045d4f98a2SYan Zheng 28055d4f98a2SYan Zheng static noinline_for_stack 28063fd0a558SYan, Zheng u64 calcu_metadata_size(struct reloc_control *rc, 28073fd0a558SYan, Zheng struct backref_node *node, int reserve) 28085d4f98a2SYan Zheng { 28090b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 28103fd0a558SYan, Zheng struct backref_node *next = node; 28113fd0a558SYan, Zheng struct backref_edge *edge; 28123fd0a558SYan, Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 28133fd0a558SYan, Zheng u64 num_bytes = 0; 28143fd0a558SYan, Zheng int index = 0; 28155d4f98a2SYan Zheng 28163fd0a558SYan, Zheng BUG_ON(reserve && node->processed); 28173fd0a558SYan, Zheng 28183fd0a558SYan, Zheng while (next) { 28193fd0a558SYan, Zheng cond_resched(); 28205d4f98a2SYan Zheng while (1) { 28213fd0a558SYan, Zheng if (next->processed && (reserve || next != node)) 28225d4f98a2SYan Zheng break; 28235d4f98a2SYan Zheng 28240b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 28253fd0a558SYan, Zheng 28263fd0a558SYan, Zheng if (list_empty(&next->upper)) 28273fd0a558SYan, Zheng break; 28283fd0a558SYan, Zheng 28293fd0a558SYan, Zheng edge = list_entry(next->upper.next, 28303fd0a558SYan, Zheng struct backref_edge, list[LOWER]); 28313fd0a558SYan, Zheng edges[index++] = edge; 28323fd0a558SYan, Zheng next = edge->node[UPPER]; 28335d4f98a2SYan Zheng } 28343fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 28353fd0a558SYan, Zheng } 28363fd0a558SYan, Zheng return num_bytes; 28373fd0a558SYan, Zheng } 28383fd0a558SYan, Zheng 28393fd0a558SYan, Zheng static int reserve_metadata_space(struct btrfs_trans_handle *trans, 28403fd0a558SYan, Zheng struct reloc_control *rc, 28413fd0a558SYan, Zheng struct backref_node *node) 28423fd0a558SYan, Zheng { 28433fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 2844da17066cSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 28453fd0a558SYan, Zheng u64 num_bytes; 28463fd0a558SYan, Zheng int ret; 28470647bf56SWang Shilong u64 tmp; 28483fd0a558SYan, Zheng 28493fd0a558SYan, Zheng num_bytes = calcu_metadata_size(rc, node, 1) * 2; 28503fd0a558SYan, Zheng 28513fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 28520647bf56SWang Shilong rc->reserved_bytes += num_bytes; 28538ca17f0fSJosef Bacik 28548ca17f0fSJosef Bacik /* 28558ca17f0fSJosef Bacik * We are under a transaction here so we can only do limited flushing. 28568ca17f0fSJosef Bacik * If we get an enospc just kick back -EAGAIN so we know to drop the 28578ca17f0fSJosef Bacik * transaction and try to refill when we can flush all the things. 28588ca17f0fSJosef Bacik */ 28590647bf56SWang Shilong ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes, 28608ca17f0fSJosef Bacik BTRFS_RESERVE_FLUSH_LIMIT); 28613fd0a558SYan, Zheng if (ret) { 2862da17066cSJeff Mahoney tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES; 28630647bf56SWang Shilong while (tmp <= rc->reserved_bytes) 28640647bf56SWang Shilong tmp <<= 1; 28650647bf56SWang Shilong /* 28660647bf56SWang Shilong * only one thread can access block_rsv at this point, 28670647bf56SWang Shilong * so we don't need hold lock to protect block_rsv. 28680647bf56SWang Shilong * we expand more reservation size here to allow enough 286952042d8eSAndrea Gelmini * space for relocation and we will return earlier in 28700647bf56SWang Shilong * enospc case. 28710647bf56SWang Shilong */ 2872da17066cSJeff Mahoney rc->block_rsv->size = tmp + fs_info->nodesize * 28730647bf56SWang Shilong RELOCATION_RESERVED_NODES; 28748ca17f0fSJosef Bacik return -EAGAIN; 28753fd0a558SYan, Zheng } 28763fd0a558SYan, Zheng 28773fd0a558SYan, Zheng return 0; 28783fd0a558SYan, Zheng } 28793fd0a558SYan, Zheng 28805d4f98a2SYan Zheng /* 28815d4f98a2SYan Zheng * relocate a block tree, and then update pointers in upper level 28825d4f98a2SYan Zheng * blocks that reference the block to point to the new location. 28835d4f98a2SYan Zheng * 28845d4f98a2SYan Zheng * if called by link_to_upper, the block has already been relocated. 28855d4f98a2SYan Zheng * in that case this function just updates pointers. 28865d4f98a2SYan Zheng */ 28875d4f98a2SYan Zheng static int do_relocation(struct btrfs_trans_handle *trans, 28883fd0a558SYan, Zheng struct reloc_control *rc, 28895d4f98a2SYan Zheng struct backref_node *node, 28905d4f98a2SYan Zheng struct btrfs_key *key, 28915d4f98a2SYan Zheng struct btrfs_path *path, int lowest) 28925d4f98a2SYan Zheng { 28932ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 28945d4f98a2SYan Zheng struct backref_node *upper; 28955d4f98a2SYan Zheng struct backref_edge *edge; 28965d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 28975d4f98a2SYan Zheng struct btrfs_root *root; 28985d4f98a2SYan Zheng struct extent_buffer *eb; 28995d4f98a2SYan Zheng u32 blocksize; 29005d4f98a2SYan Zheng u64 bytenr; 29015d4f98a2SYan Zheng u64 generation; 29025d4f98a2SYan Zheng int slot; 29035d4f98a2SYan Zheng int ret; 29045d4f98a2SYan Zheng int err = 0; 29055d4f98a2SYan Zheng 29065d4f98a2SYan Zheng BUG_ON(lowest && node->eb); 29075d4f98a2SYan Zheng 29085d4f98a2SYan Zheng path->lowest_level = node->level + 1; 29093fd0a558SYan, Zheng rc->backref_cache.path[node->level] = node; 29105d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) { 2911581c1760SQu Wenruo struct btrfs_key first_key; 291282fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 2913581c1760SQu Wenruo 29145d4f98a2SYan Zheng cond_resched(); 29155d4f98a2SYan Zheng 29165d4f98a2SYan Zheng upper = edge->node[UPPER]; 2917dc4103f9SWang Shilong root = select_reloc_root(trans, rc, upper, edges); 29183fd0a558SYan, Zheng BUG_ON(!root); 29195d4f98a2SYan Zheng 29203fd0a558SYan, Zheng if (upper->eb && !upper->locked) { 29213fd0a558SYan, Zheng if (!lowest) { 29223fd0a558SYan, Zheng ret = btrfs_bin_search(upper->eb, key, 29233fd0a558SYan, Zheng upper->level, &slot); 2924cbca7d59SFilipe Manana if (ret < 0) { 2925cbca7d59SFilipe Manana err = ret; 2926cbca7d59SFilipe Manana goto next; 2927cbca7d59SFilipe Manana } 29283fd0a558SYan, Zheng BUG_ON(ret); 29293fd0a558SYan, Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 29303fd0a558SYan, Zheng if (node->eb->start == bytenr) 29313fd0a558SYan, Zheng goto next; 29323fd0a558SYan, Zheng } 29335d4f98a2SYan Zheng drop_node_buffer(upper); 29343fd0a558SYan, Zheng } 29355d4f98a2SYan Zheng 29365d4f98a2SYan Zheng if (!upper->eb) { 29375d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 29383561b9dbSLiu Bo if (ret) { 29393561b9dbSLiu Bo if (ret < 0) 29405d4f98a2SYan Zheng err = ret; 29413561b9dbSLiu Bo else 29423561b9dbSLiu Bo err = -ENOENT; 29433561b9dbSLiu Bo 29443561b9dbSLiu Bo btrfs_release_path(path); 29455d4f98a2SYan Zheng break; 29465d4f98a2SYan Zheng } 29475d4f98a2SYan Zheng 29483fd0a558SYan, Zheng if (!upper->eb) { 29493fd0a558SYan, Zheng upper->eb = path->nodes[upper->level]; 29503fd0a558SYan, Zheng path->nodes[upper->level] = NULL; 29513fd0a558SYan, Zheng } else { 29523fd0a558SYan, Zheng BUG_ON(upper->eb != path->nodes[upper->level]); 29533fd0a558SYan, Zheng } 29543fd0a558SYan, Zheng 29553fd0a558SYan, Zheng upper->locked = 1; 29563fd0a558SYan, Zheng path->locks[upper->level] = 0; 29573fd0a558SYan, Zheng 29585d4f98a2SYan Zheng slot = path->slots[upper->level]; 2959b3b4aa74SDavid Sterba btrfs_release_path(path); 29605d4f98a2SYan Zheng } else { 29615d4f98a2SYan Zheng ret = btrfs_bin_search(upper->eb, key, upper->level, 29625d4f98a2SYan Zheng &slot); 2963cbca7d59SFilipe Manana if (ret < 0) { 2964cbca7d59SFilipe Manana err = ret; 2965cbca7d59SFilipe Manana goto next; 2966cbca7d59SFilipe Manana } 29675d4f98a2SYan Zheng BUG_ON(ret); 29685d4f98a2SYan Zheng } 29695d4f98a2SYan Zheng 29705d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 29713fd0a558SYan, Zheng if (lowest) { 29724547f4d8SLiu Bo if (bytenr != node->bytenr) { 29734547f4d8SLiu Bo btrfs_err(root->fs_info, 29744547f4d8SLiu Bo "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu", 29754547f4d8SLiu Bo bytenr, node->bytenr, slot, 29764547f4d8SLiu Bo upper->eb->start); 29774547f4d8SLiu Bo err = -EIO; 29784547f4d8SLiu Bo goto next; 29794547f4d8SLiu Bo } 29805d4f98a2SYan Zheng } else { 29813fd0a558SYan, Zheng if (node->eb->start == bytenr) 29823fd0a558SYan, Zheng goto next; 29835d4f98a2SYan Zheng } 29845d4f98a2SYan Zheng 2985da17066cSJeff Mahoney blocksize = root->fs_info->nodesize; 29865d4f98a2SYan Zheng generation = btrfs_node_ptr_generation(upper->eb, slot); 2987581c1760SQu Wenruo btrfs_node_key_to_cpu(upper->eb, &first_key, slot); 2988581c1760SQu Wenruo eb = read_tree_block(fs_info, bytenr, generation, 2989581c1760SQu Wenruo upper->level - 1, &first_key); 299064c043deSLiu Bo if (IS_ERR(eb)) { 299164c043deSLiu Bo err = PTR_ERR(eb); 299264c043deSLiu Bo goto next; 299364c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2994416bc658SJosef Bacik free_extent_buffer(eb); 299597d9a8a4STsutomu Itoh err = -EIO; 299697d9a8a4STsutomu Itoh goto next; 299797d9a8a4STsutomu Itoh } 29985d4f98a2SYan Zheng btrfs_tree_lock(eb); 29998bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 30005d4f98a2SYan Zheng 30015d4f98a2SYan Zheng if (!node->eb) { 30025d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, eb, upper->eb, 30035d4f98a2SYan Zheng slot, &eb); 30043fd0a558SYan, Zheng btrfs_tree_unlock(eb); 30053fd0a558SYan, Zheng free_extent_buffer(eb); 30065d4f98a2SYan Zheng if (ret < 0) { 30075d4f98a2SYan Zheng err = ret; 30083fd0a558SYan, Zheng goto next; 30095d4f98a2SYan Zheng } 30103fd0a558SYan, Zheng BUG_ON(node->eb != eb); 30115d4f98a2SYan Zheng } else { 30125d4f98a2SYan Zheng btrfs_set_node_blockptr(upper->eb, slot, 30135d4f98a2SYan Zheng node->eb->start); 30145d4f98a2SYan Zheng btrfs_set_node_ptr_generation(upper->eb, slot, 30155d4f98a2SYan Zheng trans->transid); 30165d4f98a2SYan Zheng btrfs_mark_buffer_dirty(upper->eb); 30175d4f98a2SYan Zheng 301882fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, 30195d4f98a2SYan Zheng node->eb->start, blocksize, 302082fa113fSQu Wenruo upper->eb->start); 302182fa113fSQu Wenruo ref.real_root = root->root_key.objectid; 302282fa113fSQu Wenruo btrfs_init_tree_ref(&ref, node->level, 302382fa113fSQu Wenruo btrfs_header_owner(upper->eb)); 302482fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 30255d4f98a2SYan Zheng BUG_ON(ret); 30265d4f98a2SYan Zheng 30275d4f98a2SYan Zheng ret = btrfs_drop_subtree(trans, root, eb, upper->eb); 30285d4f98a2SYan Zheng BUG_ON(ret); 30295d4f98a2SYan Zheng } 30303fd0a558SYan, Zheng next: 30313fd0a558SYan, Zheng if (!upper->pending) 30323fd0a558SYan, Zheng drop_node_buffer(upper); 30333fd0a558SYan, Zheng else 30343fd0a558SYan, Zheng unlock_node_buffer(upper); 30353fd0a558SYan, Zheng if (err) 30363fd0a558SYan, Zheng break; 30375d4f98a2SYan Zheng } 30383fd0a558SYan, Zheng 30393fd0a558SYan, Zheng if (!err && node->pending) { 30403fd0a558SYan, Zheng drop_node_buffer(node); 30413fd0a558SYan, Zheng list_move_tail(&node->list, &rc->backref_cache.changed); 30423fd0a558SYan, Zheng node->pending = 0; 30435d4f98a2SYan Zheng } 30443fd0a558SYan, Zheng 30455d4f98a2SYan Zheng path->lowest_level = 0; 30463fd0a558SYan, Zheng BUG_ON(err == -ENOSPC); 30475d4f98a2SYan Zheng return err; 30485d4f98a2SYan Zheng } 30495d4f98a2SYan Zheng 30505d4f98a2SYan Zheng static int link_to_upper(struct btrfs_trans_handle *trans, 30513fd0a558SYan, Zheng struct reloc_control *rc, 30525d4f98a2SYan Zheng struct backref_node *node, 30535d4f98a2SYan Zheng struct btrfs_path *path) 30545d4f98a2SYan Zheng { 30555d4f98a2SYan Zheng struct btrfs_key key; 30565d4f98a2SYan Zheng 30575d4f98a2SYan Zheng btrfs_node_key_to_cpu(node->eb, &key, 0); 30583fd0a558SYan, Zheng return do_relocation(trans, rc, node, &key, path, 0); 30595d4f98a2SYan Zheng } 30605d4f98a2SYan Zheng 30615d4f98a2SYan Zheng static int finish_pending_nodes(struct btrfs_trans_handle *trans, 30623fd0a558SYan, Zheng struct reloc_control *rc, 30633fd0a558SYan, Zheng struct btrfs_path *path, int err) 30645d4f98a2SYan Zheng { 30653fd0a558SYan, Zheng LIST_HEAD(list); 30663fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 30675d4f98a2SYan Zheng struct backref_node *node; 30685d4f98a2SYan Zheng int level; 30695d4f98a2SYan Zheng int ret; 30705d4f98a2SYan Zheng 30715d4f98a2SYan Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 30725d4f98a2SYan Zheng while (!list_empty(&cache->pending[level])) { 30735d4f98a2SYan Zheng node = list_entry(cache->pending[level].next, 30743fd0a558SYan, Zheng struct backref_node, list); 30753fd0a558SYan, Zheng list_move_tail(&node->list, &list); 30763fd0a558SYan, Zheng BUG_ON(!node->pending); 30775d4f98a2SYan Zheng 30783fd0a558SYan, Zheng if (!err) { 30793fd0a558SYan, Zheng ret = link_to_upper(trans, rc, node, path); 30805d4f98a2SYan Zheng if (ret < 0) 30815d4f98a2SYan Zheng err = ret; 30825d4f98a2SYan Zheng } 30835d4f98a2SYan Zheng } 30843fd0a558SYan, Zheng list_splice_init(&list, &cache->pending[level]); 30853fd0a558SYan, Zheng } 30865d4f98a2SYan Zheng return err; 30875d4f98a2SYan Zheng } 30885d4f98a2SYan Zheng 30895d4f98a2SYan Zheng /* 30905d4f98a2SYan Zheng * mark a block and all blocks directly/indirectly reference the block 30915d4f98a2SYan Zheng * as processed. 30925d4f98a2SYan Zheng */ 30935d4f98a2SYan Zheng static void update_processed_blocks(struct reloc_control *rc, 30945d4f98a2SYan Zheng struct backref_node *node) 30955d4f98a2SYan Zheng { 30965d4f98a2SYan Zheng struct backref_node *next = node; 30975d4f98a2SYan Zheng struct backref_edge *edge; 30985d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 30995d4f98a2SYan Zheng int index = 0; 31005d4f98a2SYan Zheng 31015d4f98a2SYan Zheng while (next) { 31025d4f98a2SYan Zheng cond_resched(); 31035d4f98a2SYan Zheng while (1) { 31045d4f98a2SYan Zheng if (next->processed) 31055d4f98a2SYan Zheng break; 31065d4f98a2SYan Zheng 31079569cc20SQu Wenruo mark_block_processed(rc, next); 31085d4f98a2SYan Zheng 31095d4f98a2SYan Zheng if (list_empty(&next->upper)) 31105d4f98a2SYan Zheng break; 31115d4f98a2SYan Zheng 31125d4f98a2SYan Zheng edge = list_entry(next->upper.next, 31135d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 31145d4f98a2SYan Zheng edges[index++] = edge; 31155d4f98a2SYan Zheng next = edge->node[UPPER]; 31165d4f98a2SYan Zheng } 31175d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 31185d4f98a2SYan Zheng } 31195d4f98a2SYan Zheng } 31205d4f98a2SYan Zheng 31217476dfdaSDavid Sterba static int tree_block_processed(u64 bytenr, struct reloc_control *rc) 31225d4f98a2SYan Zheng { 3123da17066cSJeff Mahoney u32 blocksize = rc->extent_root->fs_info->nodesize; 31247476dfdaSDavid Sterba 31255d4f98a2SYan Zheng if (test_range_bit(&rc->processed_blocks, bytenr, 31269655d298SChris Mason bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL)) 31275d4f98a2SYan Zheng return 1; 31285d4f98a2SYan Zheng return 0; 31295d4f98a2SYan Zheng } 31305d4f98a2SYan Zheng 31312ff7e61eSJeff Mahoney static int get_tree_block_key(struct btrfs_fs_info *fs_info, 31325d4f98a2SYan Zheng struct tree_block *block) 31335d4f98a2SYan Zheng { 31345d4f98a2SYan Zheng struct extent_buffer *eb; 31355d4f98a2SYan Zheng 3136581c1760SQu Wenruo eb = read_tree_block(fs_info, block->bytenr, block->key.offset, 3137581c1760SQu Wenruo block->level, NULL); 313864c043deSLiu Bo if (IS_ERR(eb)) { 313964c043deSLiu Bo return PTR_ERR(eb); 314064c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 3141416bc658SJosef Bacik free_extent_buffer(eb); 3142416bc658SJosef Bacik return -EIO; 3143416bc658SJosef Bacik } 31445d4f98a2SYan Zheng if (block->level == 0) 31455d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &block->key, 0); 31465d4f98a2SYan Zheng else 31475d4f98a2SYan Zheng btrfs_node_key_to_cpu(eb, &block->key, 0); 31485d4f98a2SYan Zheng free_extent_buffer(eb); 31495d4f98a2SYan Zheng block->key_ready = 1; 31505d4f98a2SYan Zheng return 0; 31515d4f98a2SYan Zheng } 31525d4f98a2SYan Zheng 31535d4f98a2SYan Zheng /* 31545d4f98a2SYan Zheng * helper function to relocate a tree block 31555d4f98a2SYan Zheng */ 31565d4f98a2SYan Zheng static int relocate_tree_block(struct btrfs_trans_handle *trans, 31575d4f98a2SYan Zheng struct reloc_control *rc, 31585d4f98a2SYan Zheng struct backref_node *node, 31595d4f98a2SYan Zheng struct btrfs_key *key, 31605d4f98a2SYan Zheng struct btrfs_path *path) 31615d4f98a2SYan Zheng { 31625d4f98a2SYan Zheng struct btrfs_root *root; 31633fd0a558SYan, Zheng int ret = 0; 31645d4f98a2SYan Zheng 31653fd0a558SYan, Zheng if (!node) 31665d4f98a2SYan Zheng return 0; 31673fd0a558SYan, Zheng 31685f6b2e5cSJosef Bacik /* 31695f6b2e5cSJosef Bacik * If we fail here we want to drop our backref_node because we are going 31705f6b2e5cSJosef Bacik * to start over and regenerate the tree for it. 31715f6b2e5cSJosef Bacik */ 31725f6b2e5cSJosef Bacik ret = reserve_metadata_space(trans, rc, node); 31735f6b2e5cSJosef Bacik if (ret) 31745f6b2e5cSJosef Bacik goto out; 31755f6b2e5cSJosef Bacik 31763fd0a558SYan, Zheng BUG_ON(node->processed); 3177147d256eSZhaolei root = select_one_root(node); 31783fd0a558SYan, Zheng if (root == ERR_PTR(-ENOENT)) { 31793fd0a558SYan, Zheng update_processed_blocks(rc, node); 31803fd0a558SYan, Zheng goto out; 31815d4f98a2SYan Zheng } 31825d4f98a2SYan Zheng 31833fd0a558SYan, Zheng if (root) { 318427cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 31853fd0a558SYan, Zheng BUG_ON(node->new_bytenr); 31863fd0a558SYan, Zheng BUG_ON(!list_empty(&node->list)); 31873fd0a558SYan, Zheng btrfs_record_root_in_trans(trans, root); 31883fd0a558SYan, Zheng root = root->reloc_root; 31893fd0a558SYan, Zheng node->new_bytenr = root->node->start; 319000246528SJosef Bacik btrfs_put_root(node->root); 319100246528SJosef Bacik node->root = btrfs_grab_root(root); 31920b530bc5SJosef Bacik ASSERT(node->root); 31933fd0a558SYan, Zheng list_add_tail(&node->list, &rc->backref_cache.changed); 31943fd0a558SYan, Zheng } else { 31955d4f98a2SYan Zheng path->lowest_level = node->level; 31965d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 3197b3b4aa74SDavid Sterba btrfs_release_path(path); 31983fd0a558SYan, Zheng if (ret > 0) 31995d4f98a2SYan Zheng ret = 0; 32003fd0a558SYan, Zheng } 32013fd0a558SYan, Zheng if (!ret) 32023fd0a558SYan, Zheng update_processed_blocks(rc, node); 32033fd0a558SYan, Zheng } else { 32043fd0a558SYan, Zheng ret = do_relocation(trans, rc, node, key, path, 1); 32053fd0a558SYan, Zheng } 32065d4f98a2SYan Zheng out: 32070647bf56SWang Shilong if (ret || node->level == 0 || node->cowonly) 32083fd0a558SYan, Zheng remove_backref_node(&rc->backref_cache, node); 32095d4f98a2SYan Zheng return ret; 32105d4f98a2SYan Zheng } 32115d4f98a2SYan Zheng 32125d4f98a2SYan Zheng /* 32135d4f98a2SYan Zheng * relocate a list of blocks 32145d4f98a2SYan Zheng */ 32155d4f98a2SYan Zheng static noinline_for_stack 32165d4f98a2SYan Zheng int relocate_tree_blocks(struct btrfs_trans_handle *trans, 32175d4f98a2SYan Zheng struct reloc_control *rc, struct rb_root *blocks) 32185d4f98a2SYan Zheng { 32192ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 32205d4f98a2SYan Zheng struct backref_node *node; 32215d4f98a2SYan Zheng struct btrfs_path *path; 32225d4f98a2SYan Zheng struct tree_block *block; 322398ff7b94SQu Wenruo struct tree_block *next; 32245d4f98a2SYan Zheng int ret; 32255d4f98a2SYan Zheng int err = 0; 32265d4f98a2SYan Zheng 32275d4f98a2SYan Zheng path = btrfs_alloc_path(); 3228e1a12670SLiu Bo if (!path) { 3229e1a12670SLiu Bo err = -ENOMEM; 323034c2b290SDavid Sterba goto out_free_blocks; 3231e1a12670SLiu Bo } 32325d4f98a2SYan Zheng 323398ff7b94SQu Wenruo /* Kick in readahead for tree blocks with missing keys */ 323498ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 32355d4f98a2SYan Zheng if (!block->key_ready) 32362ff7e61eSJeff Mahoney readahead_tree_block(fs_info, block->bytenr); 32375d4f98a2SYan Zheng } 32385d4f98a2SYan Zheng 323998ff7b94SQu Wenruo /* Get first keys */ 324098ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 324134c2b290SDavid Sterba if (!block->key_ready) { 32422ff7e61eSJeff Mahoney err = get_tree_block_key(fs_info, block); 324334c2b290SDavid Sterba if (err) 324434c2b290SDavid Sterba goto out_free_path; 324534c2b290SDavid Sterba } 32465d4f98a2SYan Zheng } 32475d4f98a2SYan Zheng 324898ff7b94SQu Wenruo /* Do tree relocation */ 324998ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 32503fd0a558SYan, Zheng node = build_backref_tree(rc, &block->key, 32515d4f98a2SYan Zheng block->level, block->bytenr); 32525d4f98a2SYan Zheng if (IS_ERR(node)) { 32535d4f98a2SYan Zheng err = PTR_ERR(node); 32545d4f98a2SYan Zheng goto out; 32555d4f98a2SYan Zheng } 32565d4f98a2SYan Zheng 32575d4f98a2SYan Zheng ret = relocate_tree_block(trans, rc, node, &block->key, 32585d4f98a2SYan Zheng path); 32595d4f98a2SYan Zheng if (ret < 0) { 32605d4f98a2SYan Zheng err = ret; 326150dbbb71SJosef Bacik break; 32625d4f98a2SYan Zheng } 32635d4f98a2SYan Zheng } 32645d4f98a2SYan Zheng out: 32653fd0a558SYan, Zheng err = finish_pending_nodes(trans, rc, path, err); 32665d4f98a2SYan Zheng 326734c2b290SDavid Sterba out_free_path: 32685d4f98a2SYan Zheng btrfs_free_path(path); 326934c2b290SDavid Sterba out_free_blocks: 3270e1a12670SLiu Bo free_block_list(blocks); 32715d4f98a2SYan Zheng return err; 32725d4f98a2SYan Zheng } 32735d4f98a2SYan Zheng 32745d4f98a2SYan Zheng static noinline_for_stack 3275efa56464SYan, Zheng int prealloc_file_extent_cluster(struct inode *inode, 3276efa56464SYan, Zheng struct file_extent_cluster *cluster) 3277efa56464SYan, Zheng { 3278efa56464SYan, Zheng u64 alloc_hint = 0; 3279efa56464SYan, Zheng u64 start; 3280efa56464SYan, Zheng u64 end; 3281efa56464SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 3282efa56464SYan, Zheng u64 num_bytes; 3283efa56464SYan, Zheng int nr = 0; 3284efa56464SYan, Zheng int ret = 0; 3285dcb40c19SWang Xiaoguang u64 prealloc_start = cluster->start - offset; 3286dcb40c19SWang Xiaoguang u64 prealloc_end = cluster->end - offset; 328718513091SWang Xiaoguang u64 cur_offset; 3288364ecf36SQu Wenruo struct extent_changeset *data_reserved = NULL; 3289efa56464SYan, Zheng 3290efa56464SYan, Zheng BUG_ON(cluster->start != cluster->boundary[0]); 32915955102cSAl Viro inode_lock(inode); 3292efa56464SYan, Zheng 3293364ecf36SQu Wenruo ret = btrfs_check_data_free_space(inode, &data_reserved, prealloc_start, 3294dcb40c19SWang Xiaoguang prealloc_end + 1 - prealloc_start); 3295efa56464SYan, Zheng if (ret) 3296efa56464SYan, Zheng goto out; 3297efa56464SYan, Zheng 329818513091SWang Xiaoguang cur_offset = prealloc_start; 3299efa56464SYan, Zheng while (nr < cluster->nr) { 3300efa56464SYan, Zheng start = cluster->boundary[nr] - offset; 3301efa56464SYan, Zheng if (nr + 1 < cluster->nr) 3302efa56464SYan, Zheng end = cluster->boundary[nr + 1] - 1 - offset; 3303efa56464SYan, Zheng else 3304efa56464SYan, Zheng end = cluster->end - offset; 3305efa56464SYan, Zheng 3306d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 3307efa56464SYan, Zheng num_bytes = end + 1 - start; 330818513091SWang Xiaoguang if (cur_offset < start) 3309bc42bda2SQu Wenruo btrfs_free_reserved_data_space(inode, data_reserved, 3310bc42bda2SQu Wenruo cur_offset, start - cur_offset); 3311efa56464SYan, Zheng ret = btrfs_prealloc_file_range(inode, 0, start, 3312efa56464SYan, Zheng num_bytes, num_bytes, 3313efa56464SYan, Zheng end + 1, &alloc_hint); 331418513091SWang Xiaoguang cur_offset = end + 1; 3315d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 3316efa56464SYan, Zheng if (ret) 3317efa56464SYan, Zheng break; 3318efa56464SYan, Zheng nr++; 3319efa56464SYan, Zheng } 332018513091SWang Xiaoguang if (cur_offset < prealloc_end) 3321bc42bda2SQu Wenruo btrfs_free_reserved_data_space(inode, data_reserved, 3322bc42bda2SQu Wenruo cur_offset, prealloc_end + 1 - cur_offset); 3323efa56464SYan, Zheng out: 33245955102cSAl Viro inode_unlock(inode); 3325364ecf36SQu Wenruo extent_changeset_free(data_reserved); 3326efa56464SYan, Zheng return ret; 3327efa56464SYan, Zheng } 3328efa56464SYan, Zheng 3329efa56464SYan, Zheng static noinline_for_stack 33300257bb82SYan, Zheng int setup_extent_mapping(struct inode *inode, u64 start, u64 end, 33310257bb82SYan, Zheng u64 block_start) 33325d4f98a2SYan Zheng { 33335d4f98a2SYan Zheng struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 33345d4f98a2SYan Zheng struct extent_map *em; 33350257bb82SYan, Zheng int ret = 0; 33365d4f98a2SYan Zheng 3337172ddd60SDavid Sterba em = alloc_extent_map(); 33380257bb82SYan, Zheng if (!em) 33390257bb82SYan, Zheng return -ENOMEM; 33400257bb82SYan, Zheng 33415d4f98a2SYan Zheng em->start = start; 33420257bb82SYan, Zheng em->len = end + 1 - start; 33430257bb82SYan, Zheng em->block_len = em->len; 33440257bb82SYan, Zheng em->block_start = block_start; 33455d4f98a2SYan Zheng set_bit(EXTENT_FLAG_PINNED, &em->flags); 33465d4f98a2SYan Zheng 3347d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 33485d4f98a2SYan Zheng while (1) { 3349890871beSChris Mason write_lock(&em_tree->lock); 335009a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 3351890871beSChris Mason write_unlock(&em_tree->lock); 33525d4f98a2SYan Zheng if (ret != -EEXIST) { 33535d4f98a2SYan Zheng free_extent_map(em); 33545d4f98a2SYan Zheng break; 33555d4f98a2SYan Zheng } 3356dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0); 33575d4f98a2SYan Zheng } 3358d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 33590257bb82SYan, Zheng return ret; 33600257bb82SYan, Zheng } 33615d4f98a2SYan Zheng 3362726a3421SQu Wenruo /* 3363726a3421SQu Wenruo * Allow error injection to test balance cancellation 3364726a3421SQu Wenruo */ 3365726a3421SQu Wenruo int btrfs_should_cancel_balance(struct btrfs_fs_info *fs_info) 3366726a3421SQu Wenruo { 3367726a3421SQu Wenruo return atomic_read(&fs_info->balance_cancel_req); 3368726a3421SQu Wenruo } 3369726a3421SQu Wenruo ALLOW_ERROR_INJECTION(btrfs_should_cancel_balance, TRUE); 3370726a3421SQu Wenruo 33710257bb82SYan, Zheng static int relocate_file_extent_cluster(struct inode *inode, 33720257bb82SYan, Zheng struct file_extent_cluster *cluster) 33730257bb82SYan, Zheng { 33742ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 33750257bb82SYan, Zheng u64 page_start; 33760257bb82SYan, Zheng u64 page_end; 33770257bb82SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 33780257bb82SYan, Zheng unsigned long index; 33790257bb82SYan, Zheng unsigned long last_index; 33800257bb82SYan, Zheng struct page *page; 33810257bb82SYan, Zheng struct file_ra_state *ra; 33823b16a4e3SJosef Bacik gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 33830257bb82SYan, Zheng int nr = 0; 33840257bb82SYan, Zheng int ret = 0; 33850257bb82SYan, Zheng 33860257bb82SYan, Zheng if (!cluster->nr) 33870257bb82SYan, Zheng return 0; 33880257bb82SYan, Zheng 33890257bb82SYan, Zheng ra = kzalloc(sizeof(*ra), GFP_NOFS); 33900257bb82SYan, Zheng if (!ra) 33910257bb82SYan, Zheng return -ENOMEM; 33920257bb82SYan, Zheng 3393efa56464SYan, Zheng ret = prealloc_file_extent_cluster(inode, cluster); 33940257bb82SYan, Zheng if (ret) 3395efa56464SYan, Zheng goto out; 33960257bb82SYan, Zheng 33970257bb82SYan, Zheng file_ra_state_init(ra, inode->i_mapping); 33980257bb82SYan, Zheng 3399efa56464SYan, Zheng ret = setup_extent_mapping(inode, cluster->start - offset, 3400efa56464SYan, Zheng cluster->end - offset, cluster->start); 3401efa56464SYan, Zheng if (ret) 3402efa56464SYan, Zheng goto out; 3403efa56464SYan, Zheng 340409cbfeafSKirill A. Shutemov index = (cluster->start - offset) >> PAGE_SHIFT; 340509cbfeafSKirill A. Shutemov last_index = (cluster->end - offset) >> PAGE_SHIFT; 34060257bb82SYan, Zheng while (index <= last_index) { 34079f3db423SNikolay Borisov ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), 34089f3db423SNikolay Borisov PAGE_SIZE); 3409efa56464SYan, Zheng if (ret) 3410efa56464SYan, Zheng goto out; 3411efa56464SYan, Zheng 34120257bb82SYan, Zheng page = find_lock_page(inode->i_mapping, index); 34130257bb82SYan, Zheng if (!page) { 34140257bb82SYan, Zheng page_cache_sync_readahead(inode->i_mapping, 34150257bb82SYan, Zheng ra, NULL, index, 34160257bb82SYan, Zheng last_index + 1 - index); 3417a94733d0SJosef Bacik page = find_or_create_page(inode->i_mapping, index, 34183b16a4e3SJosef Bacik mask); 34190257bb82SYan, Zheng if (!page) { 3420691fa059SNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 342143b18595SQu Wenruo PAGE_SIZE, true); 342244db1216SFilipe Manana btrfs_delalloc_release_extents(BTRFS_I(inode), 34238702ba93SQu Wenruo PAGE_SIZE); 34240257bb82SYan, Zheng ret = -ENOMEM; 3425efa56464SYan, Zheng goto out; 34260257bb82SYan, Zheng } 34270257bb82SYan, Zheng } 34280257bb82SYan, Zheng 34290257bb82SYan, Zheng if (PageReadahead(page)) { 34300257bb82SYan, Zheng page_cache_async_readahead(inode->i_mapping, 34310257bb82SYan, Zheng ra, NULL, page, index, 34320257bb82SYan, Zheng last_index + 1 - index); 34330257bb82SYan, Zheng } 34340257bb82SYan, Zheng 34350257bb82SYan, Zheng if (!PageUptodate(page)) { 34360257bb82SYan, Zheng btrfs_readpage(NULL, page); 34370257bb82SYan, Zheng lock_page(page); 34380257bb82SYan, Zheng if (!PageUptodate(page)) { 34390257bb82SYan, Zheng unlock_page(page); 344009cbfeafSKirill A. Shutemov put_page(page); 3441691fa059SNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 344243b18595SQu Wenruo PAGE_SIZE, true); 34438b62f87bSJosef Bacik btrfs_delalloc_release_extents(BTRFS_I(inode), 34448702ba93SQu Wenruo PAGE_SIZE); 34450257bb82SYan, Zheng ret = -EIO; 3446efa56464SYan, Zheng goto out; 34470257bb82SYan, Zheng } 34480257bb82SYan, Zheng } 34490257bb82SYan, Zheng 34504eee4fa4SMiao Xie page_start = page_offset(page); 345109cbfeafSKirill A. Shutemov page_end = page_start + PAGE_SIZE - 1; 34520257bb82SYan, Zheng 3453d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end); 34540257bb82SYan, Zheng 34550257bb82SYan, Zheng set_page_extent_mapped(page); 34560257bb82SYan, Zheng 34570257bb82SYan, Zheng if (nr < cluster->nr && 34580257bb82SYan, Zheng page_start + offset == cluster->boundary[nr]) { 34590257bb82SYan, Zheng set_extent_bits(&BTRFS_I(inode)->io_tree, 34600257bb82SYan, Zheng page_start, page_end, 3461ceeb0ae7SDavid Sterba EXTENT_BOUNDARY); 34620257bb82SYan, Zheng nr++; 34630257bb82SYan, Zheng } 34640257bb82SYan, Zheng 3465765f3cebSNikolay Borisov ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0, 3466330a5827SNikolay Borisov NULL); 3467765f3cebSNikolay Borisov if (ret) { 3468765f3cebSNikolay Borisov unlock_page(page); 3469765f3cebSNikolay Borisov put_page(page); 3470765f3cebSNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 347143b18595SQu Wenruo PAGE_SIZE, true); 3472765f3cebSNikolay Borisov btrfs_delalloc_release_extents(BTRFS_I(inode), 34738702ba93SQu Wenruo PAGE_SIZE); 3474765f3cebSNikolay Borisov 3475765f3cebSNikolay Borisov clear_extent_bits(&BTRFS_I(inode)->io_tree, 3476765f3cebSNikolay Borisov page_start, page_end, 3477765f3cebSNikolay Borisov EXTENT_LOCKED | EXTENT_BOUNDARY); 3478765f3cebSNikolay Borisov goto out; 3479765f3cebSNikolay Borisov 3480765f3cebSNikolay Borisov } 34810257bb82SYan, Zheng set_page_dirty(page); 34820257bb82SYan, Zheng 34830257bb82SYan, Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 3484d0082371SJeff Mahoney page_start, page_end); 34850257bb82SYan, Zheng unlock_page(page); 348609cbfeafSKirill A. Shutemov put_page(page); 34870257bb82SYan, Zheng 34880257bb82SYan, Zheng index++; 34898702ba93SQu Wenruo btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); 3490efa56464SYan, Zheng balance_dirty_pages_ratelimited(inode->i_mapping); 34912ff7e61eSJeff Mahoney btrfs_throttle(fs_info); 34927f913c7cSQu Wenruo if (btrfs_should_cancel_balance(fs_info)) { 34937f913c7cSQu Wenruo ret = -ECANCELED; 34947f913c7cSQu Wenruo goto out; 34957f913c7cSQu Wenruo } 34960257bb82SYan, Zheng } 34970257bb82SYan, Zheng WARN_ON(nr != cluster->nr); 3498efa56464SYan, Zheng out: 34990257bb82SYan, Zheng kfree(ra); 35000257bb82SYan, Zheng return ret; 35010257bb82SYan, Zheng } 35020257bb82SYan, Zheng 35030257bb82SYan, Zheng static noinline_for_stack 35040257bb82SYan, Zheng int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key, 35050257bb82SYan, Zheng struct file_extent_cluster *cluster) 35060257bb82SYan, Zheng { 35070257bb82SYan, Zheng int ret; 35080257bb82SYan, Zheng 35090257bb82SYan, Zheng if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) { 35100257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 35110257bb82SYan, Zheng if (ret) 35120257bb82SYan, Zheng return ret; 35130257bb82SYan, Zheng cluster->nr = 0; 35140257bb82SYan, Zheng } 35150257bb82SYan, Zheng 35160257bb82SYan, Zheng if (!cluster->nr) 35170257bb82SYan, Zheng cluster->start = extent_key->objectid; 35180257bb82SYan, Zheng else 35190257bb82SYan, Zheng BUG_ON(cluster->nr >= MAX_EXTENTS); 35200257bb82SYan, Zheng cluster->end = extent_key->objectid + extent_key->offset - 1; 35210257bb82SYan, Zheng cluster->boundary[cluster->nr] = extent_key->objectid; 35220257bb82SYan, Zheng cluster->nr++; 35230257bb82SYan, Zheng 35240257bb82SYan, Zheng if (cluster->nr >= MAX_EXTENTS) { 35250257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 35260257bb82SYan, Zheng if (ret) 35270257bb82SYan, Zheng return ret; 35280257bb82SYan, Zheng cluster->nr = 0; 35290257bb82SYan, Zheng } 35300257bb82SYan, Zheng return 0; 35315d4f98a2SYan Zheng } 35325d4f98a2SYan Zheng 35335d4f98a2SYan Zheng /* 35345d4f98a2SYan Zheng * helper to add a tree block to the list. 35355d4f98a2SYan Zheng * the major work is getting the generation and level of the block 35365d4f98a2SYan Zheng */ 35375d4f98a2SYan Zheng static int add_tree_block(struct reloc_control *rc, 35385d4f98a2SYan Zheng struct btrfs_key *extent_key, 35395d4f98a2SYan Zheng struct btrfs_path *path, 35405d4f98a2SYan Zheng struct rb_root *blocks) 35415d4f98a2SYan Zheng { 35425d4f98a2SYan Zheng struct extent_buffer *eb; 35435d4f98a2SYan Zheng struct btrfs_extent_item *ei; 35445d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 35455d4f98a2SYan Zheng struct tree_block *block; 35465d4f98a2SYan Zheng struct rb_node *rb_node; 35475d4f98a2SYan Zheng u32 item_size; 35485d4f98a2SYan Zheng int level = -1; 35497fdf4b60SWang Shilong u64 generation; 35505d4f98a2SYan Zheng 35515d4f98a2SYan Zheng eb = path->nodes[0]; 35525d4f98a2SYan Zheng item_size = btrfs_item_size_nr(eb, path->slots[0]); 35535d4f98a2SYan Zheng 35543173a18fSJosef Bacik if (extent_key->type == BTRFS_METADATA_ITEM_KEY || 35553173a18fSJosef Bacik item_size >= sizeof(*ei) + sizeof(*bi)) { 35565d4f98a2SYan Zheng ei = btrfs_item_ptr(eb, path->slots[0], 35575d4f98a2SYan Zheng struct btrfs_extent_item); 35583173a18fSJosef Bacik if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) { 35595d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 35605d4f98a2SYan Zheng level = btrfs_tree_block_level(eb, bi); 35615d4f98a2SYan Zheng } else { 35623173a18fSJosef Bacik level = (int)extent_key->offset; 35633173a18fSJosef Bacik } 35643173a18fSJosef Bacik generation = btrfs_extent_generation(eb, ei); 35656d8ff4e4SDavid Sterba } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) { 3566ba3c2b19SNikolay Borisov btrfs_print_v0_err(eb->fs_info); 3567ba3c2b19SNikolay Borisov btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL); 3568ba3c2b19SNikolay Borisov return -EINVAL; 35693173a18fSJosef Bacik } else { 35705d4f98a2SYan Zheng BUG(); 35715d4f98a2SYan Zheng } 35725d4f98a2SYan Zheng 3573b3b4aa74SDavid Sterba btrfs_release_path(path); 35745d4f98a2SYan Zheng 35755d4f98a2SYan Zheng BUG_ON(level == -1); 35765d4f98a2SYan Zheng 35775d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 35785d4f98a2SYan Zheng if (!block) 35795d4f98a2SYan Zheng return -ENOMEM; 35805d4f98a2SYan Zheng 35815d4f98a2SYan Zheng block->bytenr = extent_key->objectid; 3582da17066cSJeff Mahoney block->key.objectid = rc->extent_root->fs_info->nodesize; 35835d4f98a2SYan Zheng block->key.offset = generation; 35845d4f98a2SYan Zheng block->level = level; 35855d4f98a2SYan Zheng block->key_ready = 0; 35865d4f98a2SYan Zheng 35875d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, &block->rb_node); 358843c04fb1SJeff Mahoney if (rb_node) 358943c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, block->bytenr); 35905d4f98a2SYan Zheng 35915d4f98a2SYan Zheng return 0; 35925d4f98a2SYan Zheng } 35935d4f98a2SYan Zheng 35945d4f98a2SYan Zheng /* 35955d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY 35965d4f98a2SYan Zheng */ 35975d4f98a2SYan Zheng static int __add_tree_block(struct reloc_control *rc, 35985d4f98a2SYan Zheng u64 bytenr, u32 blocksize, 35995d4f98a2SYan Zheng struct rb_root *blocks) 36005d4f98a2SYan Zheng { 36010b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 36025d4f98a2SYan Zheng struct btrfs_path *path; 36035d4f98a2SYan Zheng struct btrfs_key key; 36045d4f98a2SYan Zheng int ret; 36050b246afaSJeff Mahoney bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA); 36065d4f98a2SYan Zheng 36077476dfdaSDavid Sterba if (tree_block_processed(bytenr, rc)) 36085d4f98a2SYan Zheng return 0; 36095d4f98a2SYan Zheng 36105d4f98a2SYan Zheng if (tree_search(blocks, bytenr)) 36115d4f98a2SYan Zheng return 0; 36125d4f98a2SYan Zheng 36135d4f98a2SYan Zheng path = btrfs_alloc_path(); 36145d4f98a2SYan Zheng if (!path) 36155d4f98a2SYan Zheng return -ENOMEM; 3616aee68ee5SJosef Bacik again: 36175d4f98a2SYan Zheng key.objectid = bytenr; 3618aee68ee5SJosef Bacik if (skinny) { 3619aee68ee5SJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 3620aee68ee5SJosef Bacik key.offset = (u64)-1; 3621aee68ee5SJosef Bacik } else { 36225d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 36235d4f98a2SYan Zheng key.offset = blocksize; 3624aee68ee5SJosef Bacik } 36255d4f98a2SYan Zheng 36265d4f98a2SYan Zheng path->search_commit_root = 1; 36275d4f98a2SYan Zheng path->skip_locking = 1; 36285d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0); 36295d4f98a2SYan Zheng if (ret < 0) 36305d4f98a2SYan Zheng goto out; 36315d4f98a2SYan Zheng 3632aee68ee5SJosef Bacik if (ret > 0 && skinny) { 3633aee68ee5SJosef Bacik if (path->slots[0]) { 3634aee68ee5SJosef Bacik path->slots[0]--; 3635aee68ee5SJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, 3636aee68ee5SJosef Bacik path->slots[0]); 36373173a18fSJosef Bacik if (key.objectid == bytenr && 3638aee68ee5SJosef Bacik (key.type == BTRFS_METADATA_ITEM_KEY || 3639aee68ee5SJosef Bacik (key.type == BTRFS_EXTENT_ITEM_KEY && 3640aee68ee5SJosef Bacik key.offset == blocksize))) 36413173a18fSJosef Bacik ret = 0; 36423173a18fSJosef Bacik } 3643aee68ee5SJosef Bacik 3644aee68ee5SJosef Bacik if (ret) { 3645aee68ee5SJosef Bacik skinny = false; 3646aee68ee5SJosef Bacik btrfs_release_path(path); 3647aee68ee5SJosef Bacik goto again; 3648aee68ee5SJosef Bacik } 3649aee68ee5SJosef Bacik } 3650cdccee99SLiu Bo if (ret) { 3651cdccee99SLiu Bo ASSERT(ret == 1); 3652cdccee99SLiu Bo btrfs_print_leaf(path->nodes[0]); 3653cdccee99SLiu Bo btrfs_err(fs_info, 3654cdccee99SLiu Bo "tree block extent item (%llu) is not found in extent tree", 3655cdccee99SLiu Bo bytenr); 3656cdccee99SLiu Bo WARN_ON(1); 3657cdccee99SLiu Bo ret = -EINVAL; 3658cdccee99SLiu Bo goto out; 3659cdccee99SLiu Bo } 36603173a18fSJosef Bacik 36615d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, blocks); 36625d4f98a2SYan Zheng out: 36635d4f98a2SYan Zheng btrfs_free_path(path); 36645d4f98a2SYan Zheng return ret; 36655d4f98a2SYan Zheng } 36665d4f98a2SYan Zheng 36670af3d00bSJosef Bacik static int delete_block_group_cache(struct btrfs_fs_info *fs_info, 366832da5386SDavid Sterba struct btrfs_block_group *block_group, 36691bbc621eSChris Mason struct inode *inode, 36701bbc621eSChris Mason u64 ino) 36710af3d00bSJosef Bacik { 36720af3d00bSJosef Bacik struct btrfs_key key; 36730af3d00bSJosef Bacik struct btrfs_root *root = fs_info->tree_root; 36740af3d00bSJosef Bacik struct btrfs_trans_handle *trans; 36750af3d00bSJosef Bacik int ret = 0; 36760af3d00bSJosef Bacik 36770af3d00bSJosef Bacik if (inode) 36780af3d00bSJosef Bacik goto truncate; 36790af3d00bSJosef Bacik 36800af3d00bSJosef Bacik key.objectid = ino; 36810af3d00bSJosef Bacik key.type = BTRFS_INODE_ITEM_KEY; 36820af3d00bSJosef Bacik key.offset = 0; 36830af3d00bSJosef Bacik 36844c66e0d4SDavid Sterba inode = btrfs_iget(fs_info->sb, &key, root); 36852e19f1f9SAl Viro if (IS_ERR(inode)) 36860af3d00bSJosef Bacik return -ENOENT; 36870af3d00bSJosef Bacik 36880af3d00bSJosef Bacik truncate: 36892ff7e61eSJeff Mahoney ret = btrfs_check_trunc_cache_free_space(fs_info, 36907b61cd92SMiao Xie &fs_info->global_block_rsv); 36917b61cd92SMiao Xie if (ret) 36927b61cd92SMiao Xie goto out; 36937b61cd92SMiao Xie 36947a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 36950af3d00bSJosef Bacik if (IS_ERR(trans)) { 36963612b495STsutomu Itoh ret = PTR_ERR(trans); 36970af3d00bSJosef Bacik goto out; 36980af3d00bSJosef Bacik } 36990af3d00bSJosef Bacik 370077ab86bfSJeff Mahoney ret = btrfs_truncate_free_space_cache(trans, block_group, inode); 37010af3d00bSJosef Bacik 37023a45bb20SJeff Mahoney btrfs_end_transaction(trans); 37032ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 37040af3d00bSJosef Bacik out: 37050af3d00bSJosef Bacik iput(inode); 37060af3d00bSJosef Bacik return ret; 37070af3d00bSJosef Bacik } 37080af3d00bSJosef Bacik 37095d4f98a2SYan Zheng /* 371019b546d7SQu Wenruo * Locate the free space cache EXTENT_DATA in root tree leaf and delete the 371119b546d7SQu Wenruo * cache inode, to avoid free space cache data extent blocking data relocation. 37125d4f98a2SYan Zheng */ 371319b546d7SQu Wenruo static int delete_v1_space_cache(struct extent_buffer *leaf, 371419b546d7SQu Wenruo struct btrfs_block_group *block_group, 371519b546d7SQu Wenruo u64 data_bytenr) 37165d4f98a2SYan Zheng { 371719b546d7SQu Wenruo u64 space_cache_ino; 371819b546d7SQu Wenruo struct btrfs_file_extent_item *ei; 37195d4f98a2SYan Zheng struct btrfs_key key; 372019b546d7SQu Wenruo bool found = false; 372119b546d7SQu Wenruo int i; 37225d4f98a2SYan Zheng int ret; 37235d4f98a2SYan Zheng 372419b546d7SQu Wenruo if (btrfs_header_owner(leaf) != BTRFS_ROOT_TREE_OBJECTID) 372519b546d7SQu Wenruo return 0; 37265d4f98a2SYan Zheng 372719b546d7SQu Wenruo for (i = 0; i < btrfs_header_nritems(leaf); i++) { 372819b546d7SQu Wenruo btrfs_item_key_to_cpu(leaf, &key, i); 372919b546d7SQu Wenruo if (key.type != BTRFS_EXTENT_DATA_KEY) 373019b546d7SQu Wenruo continue; 373119b546d7SQu Wenruo ei = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 373219b546d7SQu Wenruo if (btrfs_file_extent_type(leaf, ei) == BTRFS_FILE_EXTENT_REG && 373319b546d7SQu Wenruo btrfs_file_extent_disk_bytenr(leaf, ei) == data_bytenr) { 373419b546d7SQu Wenruo found = true; 373519b546d7SQu Wenruo space_cache_ino = key.objectid; 373619b546d7SQu Wenruo break; 373719b546d7SQu Wenruo } 373819b546d7SQu Wenruo } 373919b546d7SQu Wenruo if (!found) 374019b546d7SQu Wenruo return -ENOENT; 374119b546d7SQu Wenruo ret = delete_block_group_cache(leaf->fs_info, block_group, NULL, 374219b546d7SQu Wenruo space_cache_ino); 37430af3d00bSJosef Bacik return ret; 37445d4f98a2SYan Zheng } 37455d4f98a2SYan Zheng 37465d4f98a2SYan Zheng /* 37472c016dc2SLiu Bo * helper to find all tree blocks that reference a given data extent 37485d4f98a2SYan Zheng */ 37495d4f98a2SYan Zheng static noinline_for_stack 37505d4f98a2SYan Zheng int add_data_references(struct reloc_control *rc, 37515d4f98a2SYan Zheng struct btrfs_key *extent_key, 37525d4f98a2SYan Zheng struct btrfs_path *path, 37535d4f98a2SYan Zheng struct rb_root *blocks) 37545d4f98a2SYan Zheng { 375519b546d7SQu Wenruo struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 375619b546d7SQu Wenruo struct ulist *leaves = NULL; 375719b546d7SQu Wenruo struct ulist_iterator leaf_uiter; 375819b546d7SQu Wenruo struct ulist_node *ref_node = NULL; 375919b546d7SQu Wenruo const u32 blocksize = fs_info->nodesize; 3760647f63bdSFilipe David Borba Manana int ret = 0; 37615d4f98a2SYan Zheng 3762b3b4aa74SDavid Sterba btrfs_release_path(path); 376319b546d7SQu Wenruo ret = btrfs_find_all_leafs(NULL, fs_info, extent_key->objectid, 376419b546d7SQu Wenruo 0, &leaves, NULL, true); 376519b546d7SQu Wenruo if (ret < 0) 376619b546d7SQu Wenruo return ret; 376719b546d7SQu Wenruo 376819b546d7SQu Wenruo ULIST_ITER_INIT(&leaf_uiter); 376919b546d7SQu Wenruo while ((ref_node = ulist_next(leaves, &leaf_uiter))) { 377019b546d7SQu Wenruo struct extent_buffer *eb; 377119b546d7SQu Wenruo 377219b546d7SQu Wenruo eb = read_tree_block(fs_info, ref_node->val, 0, 0, NULL); 377319b546d7SQu Wenruo if (IS_ERR(eb)) { 377419b546d7SQu Wenruo ret = PTR_ERR(eb); 377519b546d7SQu Wenruo break; 377619b546d7SQu Wenruo } 377719b546d7SQu Wenruo ret = delete_v1_space_cache(eb, rc->block_group, 377819b546d7SQu Wenruo extent_key->objectid); 377919b546d7SQu Wenruo free_extent_buffer(eb); 378019b546d7SQu Wenruo if (ret < 0) 378119b546d7SQu Wenruo break; 378219b546d7SQu Wenruo ret = __add_tree_block(rc, ref_node->val, blocksize, blocks); 378319b546d7SQu Wenruo if (ret < 0) 378419b546d7SQu Wenruo break; 378519b546d7SQu Wenruo } 378619b546d7SQu Wenruo if (ret < 0) 37875d4f98a2SYan Zheng free_block_list(blocks); 378819b546d7SQu Wenruo ulist_free(leaves); 378919b546d7SQu Wenruo return ret; 37905d4f98a2SYan Zheng } 37915d4f98a2SYan Zheng 37925d4f98a2SYan Zheng /* 37932c016dc2SLiu Bo * helper to find next unprocessed extent 37945d4f98a2SYan Zheng */ 37955d4f98a2SYan Zheng static noinline_for_stack 3796147d256eSZhaolei int find_next_extent(struct reloc_control *rc, struct btrfs_path *path, 37973fd0a558SYan, Zheng struct btrfs_key *extent_key) 37985d4f98a2SYan Zheng { 37990b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 38005d4f98a2SYan Zheng struct btrfs_key key; 38015d4f98a2SYan Zheng struct extent_buffer *leaf; 38025d4f98a2SYan Zheng u64 start, end, last; 38035d4f98a2SYan Zheng int ret; 38045d4f98a2SYan Zheng 3805b3470b5dSDavid Sterba last = rc->block_group->start + rc->block_group->length; 38065d4f98a2SYan Zheng while (1) { 38075d4f98a2SYan Zheng cond_resched(); 38085d4f98a2SYan Zheng if (rc->search_start >= last) { 38095d4f98a2SYan Zheng ret = 1; 38105d4f98a2SYan Zheng break; 38115d4f98a2SYan Zheng } 38125d4f98a2SYan Zheng 38135d4f98a2SYan Zheng key.objectid = rc->search_start; 38145d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 38155d4f98a2SYan Zheng key.offset = 0; 38165d4f98a2SYan Zheng 38175d4f98a2SYan Zheng path->search_commit_root = 1; 38185d4f98a2SYan Zheng path->skip_locking = 1; 38195d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 38205d4f98a2SYan Zheng 0, 0); 38215d4f98a2SYan Zheng if (ret < 0) 38225d4f98a2SYan Zheng break; 38235d4f98a2SYan Zheng next: 38245d4f98a2SYan Zheng leaf = path->nodes[0]; 38255d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(leaf)) { 38265d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 38275d4f98a2SYan Zheng if (ret != 0) 38285d4f98a2SYan Zheng break; 38295d4f98a2SYan Zheng leaf = path->nodes[0]; 38305d4f98a2SYan Zheng } 38315d4f98a2SYan Zheng 38325d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 38335d4f98a2SYan Zheng if (key.objectid >= last) { 38345d4f98a2SYan Zheng ret = 1; 38355d4f98a2SYan Zheng break; 38365d4f98a2SYan Zheng } 38375d4f98a2SYan Zheng 38383173a18fSJosef Bacik if (key.type != BTRFS_EXTENT_ITEM_KEY && 38393173a18fSJosef Bacik key.type != BTRFS_METADATA_ITEM_KEY) { 38403173a18fSJosef Bacik path->slots[0]++; 38413173a18fSJosef Bacik goto next; 38423173a18fSJosef Bacik } 38433173a18fSJosef Bacik 38443173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 38455d4f98a2SYan Zheng key.objectid + key.offset <= rc->search_start) { 38465d4f98a2SYan Zheng path->slots[0]++; 38475d4f98a2SYan Zheng goto next; 38485d4f98a2SYan Zheng } 38495d4f98a2SYan Zheng 38503173a18fSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 38510b246afaSJeff Mahoney key.objectid + fs_info->nodesize <= 38523173a18fSJosef Bacik rc->search_start) { 38533173a18fSJosef Bacik path->slots[0]++; 38543173a18fSJosef Bacik goto next; 38553173a18fSJosef Bacik } 38563173a18fSJosef Bacik 38575d4f98a2SYan Zheng ret = find_first_extent_bit(&rc->processed_blocks, 38585d4f98a2SYan Zheng key.objectid, &start, &end, 3859e6138876SJosef Bacik EXTENT_DIRTY, NULL); 38605d4f98a2SYan Zheng 38615d4f98a2SYan Zheng if (ret == 0 && start <= key.objectid) { 3862b3b4aa74SDavid Sterba btrfs_release_path(path); 38635d4f98a2SYan Zheng rc->search_start = end + 1; 38645d4f98a2SYan Zheng } else { 38653173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) 38665d4f98a2SYan Zheng rc->search_start = key.objectid + key.offset; 38673173a18fSJosef Bacik else 38683173a18fSJosef Bacik rc->search_start = key.objectid + 38690b246afaSJeff Mahoney fs_info->nodesize; 38703fd0a558SYan, Zheng memcpy(extent_key, &key, sizeof(key)); 38715d4f98a2SYan Zheng return 0; 38725d4f98a2SYan Zheng } 38735d4f98a2SYan Zheng } 3874b3b4aa74SDavid Sterba btrfs_release_path(path); 38755d4f98a2SYan Zheng return ret; 38765d4f98a2SYan Zheng } 38775d4f98a2SYan Zheng 38785d4f98a2SYan Zheng static void set_reloc_control(struct reloc_control *rc) 38795d4f98a2SYan Zheng { 38805d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 38817585717fSChris Mason 38827585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 38835d4f98a2SYan Zheng fs_info->reloc_ctl = rc; 38847585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 38855d4f98a2SYan Zheng } 38865d4f98a2SYan Zheng 38875d4f98a2SYan Zheng static void unset_reloc_control(struct reloc_control *rc) 38885d4f98a2SYan Zheng { 38895d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 38907585717fSChris Mason 38917585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 38925d4f98a2SYan Zheng fs_info->reloc_ctl = NULL; 38937585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 38945d4f98a2SYan Zheng } 38955d4f98a2SYan Zheng 38965d4f98a2SYan Zheng static int check_extent_flags(u64 flags) 38975d4f98a2SYan Zheng { 38985d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 38995d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 39005d4f98a2SYan Zheng return 1; 39015d4f98a2SYan Zheng if (!(flags & BTRFS_EXTENT_FLAG_DATA) && 39025d4f98a2SYan Zheng !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 39035d4f98a2SYan Zheng return 1; 39045d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 39055d4f98a2SYan Zheng (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) 39065d4f98a2SYan Zheng return 1; 39075d4f98a2SYan Zheng return 0; 39085d4f98a2SYan Zheng } 39095d4f98a2SYan Zheng 39103fd0a558SYan, Zheng static noinline_for_stack 39113fd0a558SYan, Zheng int prepare_to_relocate(struct reloc_control *rc) 39123fd0a558SYan, Zheng { 39133fd0a558SYan, Zheng struct btrfs_trans_handle *trans; 3914ac2fabacSJosef Bacik int ret; 39153fd0a558SYan, Zheng 39162ff7e61eSJeff Mahoney rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info, 391766d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 39183fd0a558SYan, Zheng if (!rc->block_rsv) 39193fd0a558SYan, Zheng return -ENOMEM; 39203fd0a558SYan, Zheng 39213fd0a558SYan, Zheng memset(&rc->cluster, 0, sizeof(rc->cluster)); 3922b3470b5dSDavid Sterba rc->search_start = rc->block_group->start; 39233fd0a558SYan, Zheng rc->extents_found = 0; 39243fd0a558SYan, Zheng rc->nodes_relocated = 0; 39253fd0a558SYan, Zheng rc->merging_rsv_size = 0; 39260647bf56SWang Shilong rc->reserved_bytes = 0; 3927da17066cSJeff Mahoney rc->block_rsv->size = rc->extent_root->fs_info->nodesize * 39280647bf56SWang Shilong RELOCATION_RESERVED_NODES; 3929ac2fabacSJosef Bacik ret = btrfs_block_rsv_refill(rc->extent_root, 3930ac2fabacSJosef Bacik rc->block_rsv, rc->block_rsv->size, 3931ac2fabacSJosef Bacik BTRFS_RESERVE_FLUSH_ALL); 3932ac2fabacSJosef Bacik if (ret) 3933ac2fabacSJosef Bacik return ret; 39343fd0a558SYan, Zheng 39353fd0a558SYan, Zheng rc->create_reloc_tree = 1; 39363fd0a558SYan, Zheng set_reloc_control(rc); 39373fd0a558SYan, Zheng 39387a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 393928818947SLiu Bo if (IS_ERR(trans)) { 394028818947SLiu Bo unset_reloc_control(rc); 394128818947SLiu Bo /* 394228818947SLiu Bo * extent tree is not a ref_cow tree and has no reloc_root to 394328818947SLiu Bo * cleanup. And callers are responsible to free the above 394428818947SLiu Bo * block rsv. 394528818947SLiu Bo */ 394628818947SLiu Bo return PTR_ERR(trans); 394728818947SLiu Bo } 39483a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 39493fd0a558SYan, Zheng return 0; 39503fd0a558SYan, Zheng } 395176dda93cSYan, Zheng 39525d4f98a2SYan Zheng static noinline_for_stack int relocate_block_group(struct reloc_control *rc) 39535d4f98a2SYan Zheng { 39542ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 39555d4f98a2SYan Zheng struct rb_root blocks = RB_ROOT; 39565d4f98a2SYan Zheng struct btrfs_key key; 39575d4f98a2SYan Zheng struct btrfs_trans_handle *trans = NULL; 39585d4f98a2SYan Zheng struct btrfs_path *path; 39595d4f98a2SYan Zheng struct btrfs_extent_item *ei; 39605d4f98a2SYan Zheng u64 flags; 39615d4f98a2SYan Zheng u32 item_size; 39625d4f98a2SYan Zheng int ret; 39635d4f98a2SYan Zheng int err = 0; 3964c87f08caSChris Mason int progress = 0; 39655d4f98a2SYan Zheng 39665d4f98a2SYan Zheng path = btrfs_alloc_path(); 39673fd0a558SYan, Zheng if (!path) 39685d4f98a2SYan Zheng return -ENOMEM; 3969e4058b54SDavid Sterba path->reada = READA_FORWARD; 39703fd0a558SYan, Zheng 39713fd0a558SYan, Zheng ret = prepare_to_relocate(rc); 39723fd0a558SYan, Zheng if (ret) { 39733fd0a558SYan, Zheng err = ret; 39743fd0a558SYan, Zheng goto out_free; 39752423fdfbSJiri Slaby } 39765d4f98a2SYan Zheng 39775d4f98a2SYan Zheng while (1) { 39780647bf56SWang Shilong rc->reserved_bytes = 0; 39790647bf56SWang Shilong ret = btrfs_block_rsv_refill(rc->extent_root, 39800647bf56SWang Shilong rc->block_rsv, rc->block_rsv->size, 39810647bf56SWang Shilong BTRFS_RESERVE_FLUSH_ALL); 39820647bf56SWang Shilong if (ret) { 39830647bf56SWang Shilong err = ret; 39840647bf56SWang Shilong break; 39850647bf56SWang Shilong } 3986c87f08caSChris Mason progress++; 3987a22285a6SYan, Zheng trans = btrfs_start_transaction(rc->extent_root, 0); 39880f788c58SLiu Bo if (IS_ERR(trans)) { 39890f788c58SLiu Bo err = PTR_ERR(trans); 39900f788c58SLiu Bo trans = NULL; 39910f788c58SLiu Bo break; 39920f788c58SLiu Bo } 3993c87f08caSChris Mason restart: 39943fd0a558SYan, Zheng if (update_backref_cache(trans, &rc->backref_cache)) { 39953a45bb20SJeff Mahoney btrfs_end_transaction(trans); 399642a657f5SPan Bian trans = NULL; 39973fd0a558SYan, Zheng continue; 39983fd0a558SYan, Zheng } 39993fd0a558SYan, Zheng 4000147d256eSZhaolei ret = find_next_extent(rc, path, &key); 40015d4f98a2SYan Zheng if (ret < 0) 40025d4f98a2SYan Zheng err = ret; 40035d4f98a2SYan Zheng if (ret != 0) 40045d4f98a2SYan Zheng break; 40055d4f98a2SYan Zheng 40065d4f98a2SYan Zheng rc->extents_found++; 40075d4f98a2SYan Zheng 40085d4f98a2SYan Zheng ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 40095d4f98a2SYan Zheng struct btrfs_extent_item); 40103fd0a558SYan, Zheng item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); 40115d4f98a2SYan Zheng if (item_size >= sizeof(*ei)) { 40125d4f98a2SYan Zheng flags = btrfs_extent_flags(path->nodes[0], ei); 40135d4f98a2SYan Zheng ret = check_extent_flags(flags); 40145d4f98a2SYan Zheng BUG_ON(ret); 40156d8ff4e4SDavid Sterba } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) { 4016ba3c2b19SNikolay Borisov err = -EINVAL; 4017ba3c2b19SNikolay Borisov btrfs_print_v0_err(trans->fs_info); 4018ba3c2b19SNikolay Borisov btrfs_abort_transaction(trans, err); 4019ba3c2b19SNikolay Borisov break; 40205d4f98a2SYan Zheng } else { 40215d4f98a2SYan Zheng BUG(); 40225d4f98a2SYan Zheng } 40235d4f98a2SYan Zheng 40245d4f98a2SYan Zheng if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 40255d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, &blocks); 40265d4f98a2SYan Zheng } else if (rc->stage == UPDATE_DATA_PTRS && 40275d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 40285d4f98a2SYan Zheng ret = add_data_references(rc, &key, path, &blocks); 40295d4f98a2SYan Zheng } else { 4030b3b4aa74SDavid Sterba btrfs_release_path(path); 40315d4f98a2SYan Zheng ret = 0; 40325d4f98a2SYan Zheng } 40335d4f98a2SYan Zheng if (ret < 0) { 40343fd0a558SYan, Zheng err = ret; 40355d4f98a2SYan Zheng break; 40365d4f98a2SYan Zheng } 40375d4f98a2SYan Zheng 40385d4f98a2SYan Zheng if (!RB_EMPTY_ROOT(&blocks)) { 40395d4f98a2SYan Zheng ret = relocate_tree_blocks(trans, rc, &blocks); 40405d4f98a2SYan Zheng if (ret < 0) { 40413fd0a558SYan, Zheng if (ret != -EAGAIN) { 40425d4f98a2SYan Zheng err = ret; 40435d4f98a2SYan Zheng break; 40445d4f98a2SYan Zheng } 40453fd0a558SYan, Zheng rc->extents_found--; 40463fd0a558SYan, Zheng rc->search_start = key.objectid; 40473fd0a558SYan, Zheng } 40485d4f98a2SYan Zheng } 40495d4f98a2SYan Zheng 40503a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 40512ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 40523fd0a558SYan, Zheng trans = NULL; 40535d4f98a2SYan Zheng 40545d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && 40555d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 40565d4f98a2SYan Zheng rc->found_file_extent = 1; 40570257bb82SYan, Zheng ret = relocate_data_extent(rc->data_inode, 40583fd0a558SYan, Zheng &key, &rc->cluster); 40595d4f98a2SYan Zheng if (ret < 0) { 40605d4f98a2SYan Zheng err = ret; 40615d4f98a2SYan Zheng break; 40625d4f98a2SYan Zheng } 40635d4f98a2SYan Zheng } 4064f31ea088SQu Wenruo if (btrfs_should_cancel_balance(fs_info)) { 4065f31ea088SQu Wenruo err = -ECANCELED; 4066f31ea088SQu Wenruo break; 4067f31ea088SQu Wenruo } 40685d4f98a2SYan Zheng } 4069c87f08caSChris Mason if (trans && progress && err == -ENOSPC) { 407043a7e99dSNikolay Borisov ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags); 40719689457bSShilong Wang if (ret == 1) { 4072c87f08caSChris Mason err = 0; 4073c87f08caSChris Mason progress = 0; 4074c87f08caSChris Mason goto restart; 4075c87f08caSChris Mason } 4076c87f08caSChris Mason } 40773fd0a558SYan, Zheng 4078b3b4aa74SDavid Sterba btrfs_release_path(path); 407991166212SDavid Sterba clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY); 40805d4f98a2SYan Zheng 40815d4f98a2SYan Zheng if (trans) { 40823a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 40832ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 40845d4f98a2SYan Zheng } 40855d4f98a2SYan Zheng 40860257bb82SYan, Zheng if (!err) { 40873fd0a558SYan, Zheng ret = relocate_file_extent_cluster(rc->data_inode, 40883fd0a558SYan, Zheng &rc->cluster); 40890257bb82SYan, Zheng if (ret < 0) 40900257bb82SYan, Zheng err = ret; 40910257bb82SYan, Zheng } 40920257bb82SYan, Zheng 40933fd0a558SYan, Zheng rc->create_reloc_tree = 0; 40943fd0a558SYan, Zheng set_reloc_control(rc); 40950257bb82SYan, Zheng 40963fd0a558SYan, Zheng backref_cache_cleanup(&rc->backref_cache); 409763f018beSNikolay Borisov btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL); 40985d4f98a2SYan Zheng 40997f913c7cSQu Wenruo /* 41007f913c7cSQu Wenruo * Even in the case when the relocation is cancelled, we should all go 41017f913c7cSQu Wenruo * through prepare_to_merge() and merge_reloc_roots(). 41027f913c7cSQu Wenruo * 41037f913c7cSQu Wenruo * For error (including cancelled balance), prepare_to_merge() will 41047f913c7cSQu Wenruo * mark all reloc trees orphan, then queue them for cleanup in 41057f913c7cSQu Wenruo * merge_reloc_roots() 41067f913c7cSQu Wenruo */ 41073fd0a558SYan, Zheng err = prepare_to_merge(rc, err); 41085d4f98a2SYan Zheng 41095d4f98a2SYan Zheng merge_reloc_roots(rc); 41105d4f98a2SYan Zheng 41113fd0a558SYan, Zheng rc->merge_reloc_tree = 0; 41125d4f98a2SYan Zheng unset_reloc_control(rc); 411363f018beSNikolay Borisov btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL); 41145d4f98a2SYan Zheng 41155d4f98a2SYan Zheng /* get rid of pinned extents */ 41167a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 411762b99540SQu Wenruo if (IS_ERR(trans)) { 41183612b495STsutomu Itoh err = PTR_ERR(trans); 411962b99540SQu Wenruo goto out_free; 412062b99540SQu Wenruo } 41213a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 41226217b0faSJosef Bacik out_free: 4123d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 4124d2311e69SQu Wenruo if (ret < 0 && !err) 4125d2311e69SQu Wenruo err = ret; 41262ff7e61eSJeff Mahoney btrfs_free_block_rsv(fs_info, rc->block_rsv); 41273fd0a558SYan, Zheng btrfs_free_path(path); 41285d4f98a2SYan Zheng return err; 41295d4f98a2SYan Zheng } 41305d4f98a2SYan Zheng 41315d4f98a2SYan Zheng static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 41320257bb82SYan, Zheng struct btrfs_root *root, u64 objectid) 41335d4f98a2SYan Zheng { 41345d4f98a2SYan Zheng struct btrfs_path *path; 41355d4f98a2SYan Zheng struct btrfs_inode_item *item; 41365d4f98a2SYan Zheng struct extent_buffer *leaf; 41375d4f98a2SYan Zheng int ret; 41385d4f98a2SYan Zheng 41395d4f98a2SYan Zheng path = btrfs_alloc_path(); 41405d4f98a2SYan Zheng if (!path) 41415d4f98a2SYan Zheng return -ENOMEM; 41425d4f98a2SYan Zheng 41435d4f98a2SYan Zheng ret = btrfs_insert_empty_inode(trans, root, path, objectid); 41445d4f98a2SYan Zheng if (ret) 41455d4f98a2SYan Zheng goto out; 41465d4f98a2SYan Zheng 41475d4f98a2SYan Zheng leaf = path->nodes[0]; 41485d4f98a2SYan Zheng item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); 4149b159fa28SDavid Sterba memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 41505d4f98a2SYan Zheng btrfs_set_inode_generation(leaf, item, 1); 41510257bb82SYan, Zheng btrfs_set_inode_size(leaf, item, 0); 41525d4f98a2SYan Zheng btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); 41533fd0a558SYan, Zheng btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS | 41543fd0a558SYan, Zheng BTRFS_INODE_PREALLOC); 41555d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 41565d4f98a2SYan Zheng out: 41575d4f98a2SYan Zheng btrfs_free_path(path); 41585d4f98a2SYan Zheng return ret; 41595d4f98a2SYan Zheng } 41605d4f98a2SYan Zheng 41615d4f98a2SYan Zheng /* 41625d4f98a2SYan Zheng * helper to create inode for data relocation. 41635d4f98a2SYan Zheng * the inode is in data relocation tree and its link count is 0 41645d4f98a2SYan Zheng */ 41653fd0a558SYan, Zheng static noinline_for_stack 41663fd0a558SYan, Zheng struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, 416732da5386SDavid Sterba struct btrfs_block_group *group) 41685d4f98a2SYan Zheng { 41695d4f98a2SYan Zheng struct inode *inode = NULL; 41705d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 41715d4f98a2SYan Zheng struct btrfs_root *root; 41725d4f98a2SYan Zheng struct btrfs_key key; 41734624900dSZhaolei u64 objectid; 41745d4f98a2SYan Zheng int err = 0; 41755d4f98a2SYan Zheng 41765d4f98a2SYan Zheng root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 41775d4f98a2SYan Zheng if (IS_ERR(root)) 41785d4f98a2SYan Zheng return ERR_CAST(root); 41795d4f98a2SYan Zheng 4180a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 6); 418176deacf0SJosef Bacik if (IS_ERR(trans)) { 418200246528SJosef Bacik btrfs_put_root(root); 41833fd0a558SYan, Zheng return ERR_CAST(trans); 418476deacf0SJosef Bacik } 41855d4f98a2SYan Zheng 4186581bb050SLi Zefan err = btrfs_find_free_objectid(root, &objectid); 41875d4f98a2SYan Zheng if (err) 41885d4f98a2SYan Zheng goto out; 41895d4f98a2SYan Zheng 41900257bb82SYan, Zheng err = __insert_orphan_inode(trans, root, objectid); 41915d4f98a2SYan Zheng BUG_ON(err); 41925d4f98a2SYan Zheng 41935d4f98a2SYan Zheng key.objectid = objectid; 41945d4f98a2SYan Zheng key.type = BTRFS_INODE_ITEM_KEY; 41955d4f98a2SYan Zheng key.offset = 0; 41964c66e0d4SDavid Sterba inode = btrfs_iget(fs_info->sb, &key, root); 41972e19f1f9SAl Viro BUG_ON(IS_ERR(inode)); 4198b3470b5dSDavid Sterba BTRFS_I(inode)->index_cnt = group->start; 41995d4f98a2SYan Zheng 420073f2e545SNikolay Borisov err = btrfs_orphan_add(trans, BTRFS_I(inode)); 42015d4f98a2SYan Zheng out: 420200246528SJosef Bacik btrfs_put_root(root); 42033a45bb20SJeff Mahoney btrfs_end_transaction(trans); 42042ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 42055d4f98a2SYan Zheng if (err) { 42065d4f98a2SYan Zheng if (inode) 42075d4f98a2SYan Zheng iput(inode); 42085d4f98a2SYan Zheng inode = ERR_PTR(err); 42095d4f98a2SYan Zheng } 42105d4f98a2SYan Zheng return inode; 42115d4f98a2SYan Zheng } 42125d4f98a2SYan Zheng 4213c258d6e3SQu Wenruo static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info) 42143fd0a558SYan, Zheng { 42153fd0a558SYan, Zheng struct reloc_control *rc; 42163fd0a558SYan, Zheng 42173fd0a558SYan, Zheng rc = kzalloc(sizeof(*rc), GFP_NOFS); 42183fd0a558SYan, Zheng if (!rc) 42193fd0a558SYan, Zheng return NULL; 42203fd0a558SYan, Zheng 42213fd0a558SYan, Zheng INIT_LIST_HEAD(&rc->reloc_roots); 4222d2311e69SQu Wenruo INIT_LIST_HEAD(&rc->dirty_subvol_roots); 42232433bea5SQu Wenruo backref_cache_init(fs_info, &rc->backref_cache, 1); 42243fd0a558SYan, Zheng mapping_tree_init(&rc->reloc_root_tree); 422543eb5f29SQu Wenruo extent_io_tree_init(fs_info, &rc->processed_blocks, 422643eb5f29SQu Wenruo IO_TREE_RELOC_BLOCKS, NULL); 42273fd0a558SYan, Zheng return rc; 42283fd0a558SYan, Zheng } 42293fd0a558SYan, Zheng 42301a0afa0eSJosef Bacik static void free_reloc_control(struct reloc_control *rc) 42311a0afa0eSJosef Bacik { 42321a0afa0eSJosef Bacik struct mapping_node *node, *tmp; 42331a0afa0eSJosef Bacik 42341a0afa0eSJosef Bacik free_reloc_roots(&rc->reloc_roots); 42351a0afa0eSJosef Bacik rbtree_postorder_for_each_entry_safe(node, tmp, 42361a0afa0eSJosef Bacik &rc->reloc_root_tree.rb_root, rb_node) 42371a0afa0eSJosef Bacik kfree(node); 42381a0afa0eSJosef Bacik 42391a0afa0eSJosef Bacik kfree(rc); 42401a0afa0eSJosef Bacik } 42411a0afa0eSJosef Bacik 42425d4f98a2SYan Zheng /* 4243ebce0e01SAdam Borowski * Print the block group being relocated 4244ebce0e01SAdam Borowski */ 4245ebce0e01SAdam Borowski static void describe_relocation(struct btrfs_fs_info *fs_info, 424632da5386SDavid Sterba struct btrfs_block_group *block_group) 4247ebce0e01SAdam Borowski { 4248f89e09cfSAnand Jain char buf[128] = {'\0'}; 4249ebce0e01SAdam Borowski 4250f89e09cfSAnand Jain btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf)); 4251ebce0e01SAdam Borowski 4252ebce0e01SAdam Borowski btrfs_info(fs_info, 4253ebce0e01SAdam Borowski "relocating block group %llu flags %s", 4254b3470b5dSDavid Sterba block_group->start, buf); 4255ebce0e01SAdam Borowski } 4256ebce0e01SAdam Borowski 4257430640e3SQu Wenruo static const char *stage_to_string(int stage) 4258430640e3SQu Wenruo { 4259430640e3SQu Wenruo if (stage == MOVE_DATA_EXTENTS) 4260430640e3SQu Wenruo return "move data extents"; 4261430640e3SQu Wenruo if (stage == UPDATE_DATA_PTRS) 4262430640e3SQu Wenruo return "update data pointers"; 4263430640e3SQu Wenruo return "unknown"; 4264430640e3SQu Wenruo } 4265430640e3SQu Wenruo 4266ebce0e01SAdam Borowski /* 42675d4f98a2SYan Zheng * function to relocate all extents in a block group. 42685d4f98a2SYan Zheng */ 42696bccf3abSJeff Mahoney int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) 42705d4f98a2SYan Zheng { 427132da5386SDavid Sterba struct btrfs_block_group *bg; 42726bccf3abSJeff Mahoney struct btrfs_root *extent_root = fs_info->extent_root; 42735d4f98a2SYan Zheng struct reloc_control *rc; 42740af3d00bSJosef Bacik struct inode *inode; 42750af3d00bSJosef Bacik struct btrfs_path *path; 42765d4f98a2SYan Zheng int ret; 4277f0486c68SYan, Zheng int rw = 0; 42785d4f98a2SYan Zheng int err = 0; 42795d4f98a2SYan Zheng 4280eede2bf3SOmar Sandoval bg = btrfs_lookup_block_group(fs_info, group_start); 4281eede2bf3SOmar Sandoval if (!bg) 4282eede2bf3SOmar Sandoval return -ENOENT; 4283eede2bf3SOmar Sandoval 4284eede2bf3SOmar Sandoval if (btrfs_pinned_by_swapfile(fs_info, bg)) { 4285eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 4286eede2bf3SOmar Sandoval return -ETXTBSY; 4287eede2bf3SOmar Sandoval } 4288eede2bf3SOmar Sandoval 4289c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 4290eede2bf3SOmar Sandoval if (!rc) { 4291eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 42925d4f98a2SYan Zheng return -ENOMEM; 4293eede2bf3SOmar Sandoval } 42945d4f98a2SYan Zheng 4295f0486c68SYan, Zheng rc->extent_root = extent_root; 4296eede2bf3SOmar Sandoval rc->block_group = bg; 42975d4f98a2SYan Zheng 4298b12de528SQu Wenruo ret = btrfs_inc_block_group_ro(rc->block_group, true); 4299f0486c68SYan, Zheng if (ret) { 4300f0486c68SYan, Zheng err = ret; 4301f0486c68SYan, Zheng goto out; 4302f0486c68SYan, Zheng } 4303f0486c68SYan, Zheng rw = 1; 4304f0486c68SYan, Zheng 43050af3d00bSJosef Bacik path = btrfs_alloc_path(); 43060af3d00bSJosef Bacik if (!path) { 43070af3d00bSJosef Bacik err = -ENOMEM; 43080af3d00bSJosef Bacik goto out; 43090af3d00bSJosef Bacik } 43100af3d00bSJosef Bacik 43117949f339SDavid Sterba inode = lookup_free_space_inode(rc->block_group, path); 43120af3d00bSJosef Bacik btrfs_free_path(path); 43130af3d00bSJosef Bacik 43140af3d00bSJosef Bacik if (!IS_ERR(inode)) 43151bbc621eSChris Mason ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0); 43160af3d00bSJosef Bacik else 43170af3d00bSJosef Bacik ret = PTR_ERR(inode); 43180af3d00bSJosef Bacik 43190af3d00bSJosef Bacik if (ret && ret != -ENOENT) { 43200af3d00bSJosef Bacik err = ret; 43210af3d00bSJosef Bacik goto out; 43220af3d00bSJosef Bacik } 43230af3d00bSJosef Bacik 43245d4f98a2SYan Zheng rc->data_inode = create_reloc_inode(fs_info, rc->block_group); 43255d4f98a2SYan Zheng if (IS_ERR(rc->data_inode)) { 43265d4f98a2SYan Zheng err = PTR_ERR(rc->data_inode); 43275d4f98a2SYan Zheng rc->data_inode = NULL; 43285d4f98a2SYan Zheng goto out; 43295d4f98a2SYan Zheng } 43305d4f98a2SYan Zheng 43310b246afaSJeff Mahoney describe_relocation(fs_info, rc->block_group); 43325d4f98a2SYan Zheng 43339cfa3e34SFilipe Manana btrfs_wait_block_group_reservations(rc->block_group); 4334f78c436cSFilipe Manana btrfs_wait_nocow_writers(rc->block_group); 43356374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 4336b3470b5dSDavid Sterba rc->block_group->start, 4337b3470b5dSDavid Sterba rc->block_group->length); 43385d4f98a2SYan Zheng 43395d4f98a2SYan Zheng while (1) { 4340430640e3SQu Wenruo int finishes_stage; 4341430640e3SQu Wenruo 434276dda93cSYan, Zheng mutex_lock(&fs_info->cleaner_mutex); 43435d4f98a2SYan Zheng ret = relocate_block_group(rc); 434476dda93cSYan, Zheng mutex_unlock(&fs_info->cleaner_mutex); 4345ff612ba7SJosef Bacik if (ret < 0) 43465d4f98a2SYan Zheng err = ret; 4347ff612ba7SJosef Bacik 4348430640e3SQu Wenruo finishes_stage = rc->stage; 4349ff612ba7SJosef Bacik /* 4350ff612ba7SJosef Bacik * We may have gotten ENOSPC after we already dirtied some 4351ff612ba7SJosef Bacik * extents. If writeout happens while we're relocating a 4352ff612ba7SJosef Bacik * different block group we could end up hitting the 4353ff612ba7SJosef Bacik * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in 4354ff612ba7SJosef Bacik * btrfs_reloc_cow_block. Make sure we write everything out 4355ff612ba7SJosef Bacik * properly so we don't trip over this problem, and then break 4356ff612ba7SJosef Bacik * out of the loop if we hit an error. 4357ff612ba7SJosef Bacik */ 4358ff612ba7SJosef Bacik if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) { 4359ff612ba7SJosef Bacik ret = btrfs_wait_ordered_range(rc->data_inode, 0, 4360ff612ba7SJosef Bacik (u64)-1); 4361ff612ba7SJosef Bacik if (ret) 4362ff612ba7SJosef Bacik err = ret; 4363ff612ba7SJosef Bacik invalidate_mapping_pages(rc->data_inode->i_mapping, 4364ff612ba7SJosef Bacik 0, -1); 4365ff612ba7SJosef Bacik rc->stage = UPDATE_DATA_PTRS; 43665d4f98a2SYan Zheng } 43675d4f98a2SYan Zheng 4368ff612ba7SJosef Bacik if (err < 0) 4369ff612ba7SJosef Bacik goto out; 4370ff612ba7SJosef Bacik 43715d4f98a2SYan Zheng if (rc->extents_found == 0) 43725d4f98a2SYan Zheng break; 43735d4f98a2SYan Zheng 4374430640e3SQu Wenruo btrfs_info(fs_info, "found %llu extents, stage: %s", 4375430640e3SQu Wenruo rc->extents_found, stage_to_string(finishes_stage)); 43765d4f98a2SYan Zheng } 43775d4f98a2SYan Zheng 43785d4f98a2SYan Zheng WARN_ON(rc->block_group->pinned > 0); 43795d4f98a2SYan Zheng WARN_ON(rc->block_group->reserved > 0); 4380bf38be65SDavid Sterba WARN_ON(rc->block_group->used > 0); 43815d4f98a2SYan Zheng out: 4382f0486c68SYan, Zheng if (err && rw) 43832ff7e61eSJeff Mahoney btrfs_dec_block_group_ro(rc->block_group); 43845d4f98a2SYan Zheng iput(rc->data_inode); 43855d4f98a2SYan Zheng btrfs_put_block_group(rc->block_group); 43861a0afa0eSJosef Bacik free_reloc_control(rc); 43875d4f98a2SYan Zheng return err; 43885d4f98a2SYan Zheng } 43895d4f98a2SYan Zheng 439076dda93cSYan, Zheng static noinline_for_stack int mark_garbage_root(struct btrfs_root *root) 439176dda93cSYan, Zheng { 43920b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 439376dda93cSYan, Zheng struct btrfs_trans_handle *trans; 439479787eaaSJeff Mahoney int ret, err; 439576dda93cSYan, Zheng 43960b246afaSJeff Mahoney trans = btrfs_start_transaction(fs_info->tree_root, 0); 439779787eaaSJeff Mahoney if (IS_ERR(trans)) 439879787eaaSJeff Mahoney return PTR_ERR(trans); 439976dda93cSYan, Zheng 440076dda93cSYan, Zheng memset(&root->root_item.drop_progress, 0, 440176dda93cSYan, Zheng sizeof(root->root_item.drop_progress)); 440276dda93cSYan, Zheng root->root_item.drop_level = 0; 440376dda93cSYan, Zheng btrfs_set_root_refs(&root->root_item, 0); 44040b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 440576dda93cSYan, Zheng &root->root_key, &root->root_item); 440676dda93cSYan, Zheng 44073a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 440879787eaaSJeff Mahoney if (err) 440979787eaaSJeff Mahoney return err; 441079787eaaSJeff Mahoney return ret; 441176dda93cSYan, Zheng } 441276dda93cSYan, Zheng 44135d4f98a2SYan Zheng /* 44145d4f98a2SYan Zheng * recover relocation interrupted by system crash. 44155d4f98a2SYan Zheng * 44165d4f98a2SYan Zheng * this function resumes merging reloc trees with corresponding fs trees. 44175d4f98a2SYan Zheng * this is important for keeping the sharing of tree blocks 44185d4f98a2SYan Zheng */ 44195d4f98a2SYan Zheng int btrfs_recover_relocation(struct btrfs_root *root) 44205d4f98a2SYan Zheng { 44210b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 44225d4f98a2SYan Zheng LIST_HEAD(reloc_roots); 44235d4f98a2SYan Zheng struct btrfs_key key; 44245d4f98a2SYan Zheng struct btrfs_root *fs_root; 44255d4f98a2SYan Zheng struct btrfs_root *reloc_root; 44265d4f98a2SYan Zheng struct btrfs_path *path; 44275d4f98a2SYan Zheng struct extent_buffer *leaf; 44285d4f98a2SYan Zheng struct reloc_control *rc = NULL; 44295d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 44305d4f98a2SYan Zheng int ret; 44315d4f98a2SYan Zheng int err = 0; 44325d4f98a2SYan Zheng 44335d4f98a2SYan Zheng path = btrfs_alloc_path(); 44345d4f98a2SYan Zheng if (!path) 44355d4f98a2SYan Zheng return -ENOMEM; 4436e4058b54SDavid Sterba path->reada = READA_BACK; 44375d4f98a2SYan Zheng 44385d4f98a2SYan Zheng key.objectid = BTRFS_TREE_RELOC_OBJECTID; 44395d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 44405d4f98a2SYan Zheng key.offset = (u64)-1; 44415d4f98a2SYan Zheng 44425d4f98a2SYan Zheng while (1) { 44430b246afaSJeff Mahoney ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, 44445d4f98a2SYan Zheng path, 0, 0); 44455d4f98a2SYan Zheng if (ret < 0) { 44465d4f98a2SYan Zheng err = ret; 44475d4f98a2SYan Zheng goto out; 44485d4f98a2SYan Zheng } 44495d4f98a2SYan Zheng if (ret > 0) { 44505d4f98a2SYan Zheng if (path->slots[0] == 0) 44515d4f98a2SYan Zheng break; 44525d4f98a2SYan Zheng path->slots[0]--; 44535d4f98a2SYan Zheng } 44545d4f98a2SYan Zheng leaf = path->nodes[0]; 44555d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 4456b3b4aa74SDavid Sterba btrfs_release_path(path); 44575d4f98a2SYan Zheng 44585d4f98a2SYan Zheng if (key.objectid != BTRFS_TREE_RELOC_OBJECTID || 44595d4f98a2SYan Zheng key.type != BTRFS_ROOT_ITEM_KEY) 44605d4f98a2SYan Zheng break; 44615d4f98a2SYan Zheng 44623dbf1738SJosef Bacik reloc_root = btrfs_read_tree_root(root, &key); 44635d4f98a2SYan Zheng if (IS_ERR(reloc_root)) { 44645d4f98a2SYan Zheng err = PTR_ERR(reloc_root); 44655d4f98a2SYan Zheng goto out; 44665d4f98a2SYan Zheng } 44675d4f98a2SYan Zheng 44683dbf1738SJosef Bacik set_bit(BTRFS_ROOT_REF_COWS, &reloc_root->state); 44695d4f98a2SYan Zheng list_add(&reloc_root->root_list, &reloc_roots); 44705d4f98a2SYan Zheng 44715d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 44720b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, 44735d4f98a2SYan Zheng reloc_root->root_key.offset); 44745d4f98a2SYan Zheng if (IS_ERR(fs_root)) { 447576dda93cSYan, Zheng ret = PTR_ERR(fs_root); 447676dda93cSYan, Zheng if (ret != -ENOENT) { 447776dda93cSYan, Zheng err = ret; 44785d4f98a2SYan Zheng goto out; 44795d4f98a2SYan Zheng } 448079787eaaSJeff Mahoney ret = mark_garbage_root(reloc_root); 448179787eaaSJeff Mahoney if (ret < 0) { 448279787eaaSJeff Mahoney err = ret; 448379787eaaSJeff Mahoney goto out; 448479787eaaSJeff Mahoney } 4485932fd26dSJosef Bacik } else { 448600246528SJosef Bacik btrfs_put_root(fs_root); 448776dda93cSYan, Zheng } 44885d4f98a2SYan Zheng } 44895d4f98a2SYan Zheng 44905d4f98a2SYan Zheng if (key.offset == 0) 44915d4f98a2SYan Zheng break; 44925d4f98a2SYan Zheng 44935d4f98a2SYan Zheng key.offset--; 44945d4f98a2SYan Zheng } 4495b3b4aa74SDavid Sterba btrfs_release_path(path); 44965d4f98a2SYan Zheng 44975d4f98a2SYan Zheng if (list_empty(&reloc_roots)) 44985d4f98a2SYan Zheng goto out; 44995d4f98a2SYan Zheng 4500c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 45015d4f98a2SYan Zheng if (!rc) { 45025d4f98a2SYan Zheng err = -ENOMEM; 45035d4f98a2SYan Zheng goto out; 45045d4f98a2SYan Zheng } 45055d4f98a2SYan Zheng 45060b246afaSJeff Mahoney rc->extent_root = fs_info->extent_root; 45075d4f98a2SYan Zheng 45085d4f98a2SYan Zheng set_reloc_control(rc); 45095d4f98a2SYan Zheng 45107a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 45113612b495STsutomu Itoh if (IS_ERR(trans)) { 45123612b495STsutomu Itoh err = PTR_ERR(trans); 4513fb2d83eeSJosef Bacik goto out_unset; 45143612b495STsutomu Itoh } 45153fd0a558SYan, Zheng 45163fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 45173fd0a558SYan, Zheng 45185d4f98a2SYan Zheng while (!list_empty(&reloc_roots)) { 45195d4f98a2SYan Zheng reloc_root = list_entry(reloc_roots.next, 45205d4f98a2SYan Zheng struct btrfs_root, root_list); 45215d4f98a2SYan Zheng list_del(&reloc_root->root_list); 45225d4f98a2SYan Zheng 45235d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) == 0) { 45245d4f98a2SYan Zheng list_add_tail(&reloc_root->root_list, 45255d4f98a2SYan Zheng &rc->reloc_roots); 45265d4f98a2SYan Zheng continue; 45275d4f98a2SYan Zheng } 45285d4f98a2SYan Zheng 45290b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, reloc_root->root_key.offset); 453079787eaaSJeff Mahoney if (IS_ERR(fs_root)) { 453179787eaaSJeff Mahoney err = PTR_ERR(fs_root); 4532ca1aa281SJosef Bacik list_add_tail(&reloc_root->root_list, &reloc_roots); 45331402d17dSXiyu Yang btrfs_end_transaction(trans); 4534fb2d83eeSJosef Bacik goto out_unset; 453579787eaaSJeff Mahoney } 45365d4f98a2SYan Zheng 4537ffd7b339SJeff Mahoney err = __add_reloc_root(reloc_root); 453879787eaaSJeff Mahoney BUG_ON(err < 0); /* -ENOMEM or logic error */ 4539f44deb74SJosef Bacik fs_root->reloc_root = btrfs_grab_root(reloc_root); 454000246528SJosef Bacik btrfs_put_root(fs_root); 45415d4f98a2SYan Zheng } 45425d4f98a2SYan Zheng 45433a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 454479787eaaSJeff Mahoney if (err) 4545fb2d83eeSJosef Bacik goto out_unset; 45465d4f98a2SYan Zheng 45475d4f98a2SYan Zheng merge_reloc_roots(rc); 45485d4f98a2SYan Zheng 45495d4f98a2SYan Zheng unset_reloc_control(rc); 45505d4f98a2SYan Zheng 45517a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 455262b99540SQu Wenruo if (IS_ERR(trans)) { 45533612b495STsutomu Itoh err = PTR_ERR(trans); 45546217b0faSJosef Bacik goto out_clean; 455562b99540SQu Wenruo } 45563a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 45576217b0faSJosef Bacik out_clean: 4558d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 4559d2311e69SQu Wenruo if (ret < 0 && !err) 4560d2311e69SQu Wenruo err = ret; 4561fb2d83eeSJosef Bacik out_unset: 4562fb2d83eeSJosef Bacik unset_reloc_control(rc); 45631a0afa0eSJosef Bacik free_reloc_control(rc); 45643612b495STsutomu Itoh out: 4565aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 4566aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 4567aca1bba6SLiu Bo 45685d4f98a2SYan Zheng btrfs_free_path(path); 45695d4f98a2SYan Zheng 45705d4f98a2SYan Zheng if (err == 0) { 45715d4f98a2SYan Zheng /* cleanup orphan inode in data relocation tree */ 45720b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 4573932fd26dSJosef Bacik if (IS_ERR(fs_root)) { 45745d4f98a2SYan Zheng err = PTR_ERR(fs_root); 4575932fd26dSJosef Bacik } else { 457666b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_root); 457700246528SJosef Bacik btrfs_put_root(fs_root); 4578932fd26dSJosef Bacik } 4579932fd26dSJosef Bacik } 45805d4f98a2SYan Zheng return err; 45815d4f98a2SYan Zheng } 45825d4f98a2SYan Zheng 45835d4f98a2SYan Zheng /* 45845d4f98a2SYan Zheng * helper to add ordered checksum for data relocation. 45855d4f98a2SYan Zheng * 45865d4f98a2SYan Zheng * cloning checksum properly handles the nodatasum extents. 45875d4f98a2SYan Zheng * it also saves CPU time to re-calculate the checksum. 45885d4f98a2SYan Zheng */ 45895d4f98a2SYan Zheng int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len) 45905d4f98a2SYan Zheng { 45910b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 45925d4f98a2SYan Zheng struct btrfs_ordered_sum *sums; 45935d4f98a2SYan Zheng struct btrfs_ordered_extent *ordered; 45945d4f98a2SYan Zheng int ret; 45955d4f98a2SYan Zheng u64 disk_bytenr; 45964577b014SJosef Bacik u64 new_bytenr; 45975d4f98a2SYan Zheng LIST_HEAD(list); 45985d4f98a2SYan Zheng 45995d4f98a2SYan Zheng ordered = btrfs_lookup_ordered_extent(inode, file_pos); 4600bffe633eSOmar Sandoval BUG_ON(ordered->file_offset != file_pos || ordered->num_bytes != len); 46015d4f98a2SYan Zheng 46025d4f98a2SYan Zheng disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt; 46030b246afaSJeff Mahoney ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr, 4604a2de733cSArne Jansen disk_bytenr + len - 1, &list, 0); 460579787eaaSJeff Mahoney if (ret) 460679787eaaSJeff Mahoney goto out; 46075d4f98a2SYan Zheng 46085d4f98a2SYan Zheng while (!list_empty(&list)) { 46095d4f98a2SYan Zheng sums = list_entry(list.next, struct btrfs_ordered_sum, list); 46105d4f98a2SYan Zheng list_del_init(&sums->list); 46115d4f98a2SYan Zheng 46124577b014SJosef Bacik /* 46134577b014SJosef Bacik * We need to offset the new_bytenr based on where the csum is. 46144577b014SJosef Bacik * We need to do this because we will read in entire prealloc 46154577b014SJosef Bacik * extents but we may have written to say the middle of the 46164577b014SJosef Bacik * prealloc extent, so we need to make sure the csum goes with 46174577b014SJosef Bacik * the right disk offset. 46184577b014SJosef Bacik * 46194577b014SJosef Bacik * We can do this because the data reloc inode refers strictly 46204577b014SJosef Bacik * to the on disk bytes, so we don't have to worry about 46214577b014SJosef Bacik * disk_len vs real len like with real inodes since it's all 46224577b014SJosef Bacik * disk length. 46234577b014SJosef Bacik */ 4624bffe633eSOmar Sandoval new_bytenr = ordered->disk_bytenr + sums->bytenr - disk_bytenr; 46254577b014SJosef Bacik sums->bytenr = new_bytenr; 46265d4f98a2SYan Zheng 4627f9756261SNikolay Borisov btrfs_add_ordered_sum(ordered, sums); 46285d4f98a2SYan Zheng } 462979787eaaSJeff Mahoney out: 46305d4f98a2SYan Zheng btrfs_put_ordered_extent(ordered); 4631411fc6bcSAndi Kleen return ret; 46325d4f98a2SYan Zheng } 46333fd0a558SYan, Zheng 463483d4cfd4SJosef Bacik int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans, 46353fd0a558SYan, Zheng struct btrfs_root *root, struct extent_buffer *buf, 46363fd0a558SYan, Zheng struct extent_buffer *cow) 46373fd0a558SYan, Zheng { 46380b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 46393fd0a558SYan, Zheng struct reloc_control *rc; 46403fd0a558SYan, Zheng struct backref_node *node; 46413fd0a558SYan, Zheng int first_cow = 0; 46423fd0a558SYan, Zheng int level; 464383d4cfd4SJosef Bacik int ret = 0; 46443fd0a558SYan, Zheng 46450b246afaSJeff Mahoney rc = fs_info->reloc_ctl; 46463fd0a558SYan, Zheng if (!rc) 464783d4cfd4SJosef Bacik return 0; 46483fd0a558SYan, Zheng 46493fd0a558SYan, Zheng BUG_ON(rc->stage == UPDATE_DATA_PTRS && 46503fd0a558SYan, Zheng root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID); 46513fd0a558SYan, Zheng 46523fd0a558SYan, Zheng level = btrfs_header_level(buf); 46533fd0a558SYan, Zheng if (btrfs_header_generation(buf) <= 46543fd0a558SYan, Zheng btrfs_root_last_snapshot(&root->root_item)) 46553fd0a558SYan, Zheng first_cow = 1; 46563fd0a558SYan, Zheng 46573fd0a558SYan, Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID && 46583fd0a558SYan, Zheng rc->create_reloc_tree) { 46593fd0a558SYan, Zheng WARN_ON(!first_cow && level == 0); 46603fd0a558SYan, Zheng 46613fd0a558SYan, Zheng node = rc->backref_cache.path[level]; 46623fd0a558SYan, Zheng BUG_ON(node->bytenr != buf->start && 46633fd0a558SYan, Zheng node->new_bytenr != buf->start); 46643fd0a558SYan, Zheng 46653fd0a558SYan, Zheng drop_node_buffer(node); 466667439dadSDavid Sterba atomic_inc(&cow->refs); 46673fd0a558SYan, Zheng node->eb = cow; 46683fd0a558SYan, Zheng node->new_bytenr = cow->start; 46693fd0a558SYan, Zheng 46703fd0a558SYan, Zheng if (!node->pending) { 46713fd0a558SYan, Zheng list_move_tail(&node->list, 46723fd0a558SYan, Zheng &rc->backref_cache.pending[level]); 46733fd0a558SYan, Zheng node->pending = 1; 46743fd0a558SYan, Zheng } 46753fd0a558SYan, Zheng 46763fd0a558SYan, Zheng if (first_cow) 46779569cc20SQu Wenruo mark_block_processed(rc, node); 46783fd0a558SYan, Zheng 46793fd0a558SYan, Zheng if (first_cow && level > 0) 46803fd0a558SYan, Zheng rc->nodes_relocated += buf->len; 46813fd0a558SYan, Zheng } 46823fd0a558SYan, Zheng 468383d4cfd4SJosef Bacik if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) 46843fd0a558SYan, Zheng ret = replace_file_extents(trans, rc, root, cow); 468583d4cfd4SJosef Bacik return ret; 46863fd0a558SYan, Zheng } 46873fd0a558SYan, Zheng 46883fd0a558SYan, Zheng /* 46893fd0a558SYan, Zheng * called before creating snapshot. it calculates metadata reservation 469001327610SNicholas D Steeves * required for relocating tree blocks in the snapshot 46913fd0a558SYan, Zheng */ 4692147d256eSZhaolei void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending, 46933fd0a558SYan, Zheng u64 *bytes_to_reserve) 46943fd0a558SYan, Zheng { 469510995c04SQu Wenruo struct btrfs_root *root = pending->root; 469610995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 46973fd0a558SYan, Zheng 46986282675eSQu Wenruo if (!rc || !have_reloc_root(root)) 46993fd0a558SYan, Zheng return; 47003fd0a558SYan, Zheng 47013fd0a558SYan, Zheng if (!rc->merge_reloc_tree) 47023fd0a558SYan, Zheng return; 47033fd0a558SYan, Zheng 47043fd0a558SYan, Zheng root = root->reloc_root; 47053fd0a558SYan, Zheng BUG_ON(btrfs_root_refs(&root->root_item) == 0); 47063fd0a558SYan, Zheng /* 47073fd0a558SYan, Zheng * relocation is in the stage of merging trees. the space 47083fd0a558SYan, Zheng * used by merging a reloc tree is twice the size of 47093fd0a558SYan, Zheng * relocated tree nodes in the worst case. half for cowing 47103fd0a558SYan, Zheng * the reloc tree, half for cowing the fs tree. the space 47113fd0a558SYan, Zheng * used by cowing the reloc tree will be freed after the 47123fd0a558SYan, Zheng * tree is dropped. if we create snapshot, cowing the fs 47133fd0a558SYan, Zheng * tree may use more space than it frees. so we need 47143fd0a558SYan, Zheng * reserve extra space. 47153fd0a558SYan, Zheng */ 47163fd0a558SYan, Zheng *bytes_to_reserve += rc->nodes_relocated; 47173fd0a558SYan, Zheng } 47183fd0a558SYan, Zheng 47193fd0a558SYan, Zheng /* 47203fd0a558SYan, Zheng * called after snapshot is created. migrate block reservation 47213fd0a558SYan, Zheng * and create reloc root for the newly created snapshot 4722f44deb74SJosef Bacik * 4723f44deb74SJosef Bacik * This is similar to btrfs_init_reloc_root(), we come out of here with two 4724f44deb74SJosef Bacik * references held on the reloc_root, one for root->reloc_root and one for 4725f44deb74SJosef Bacik * rc->reloc_roots. 47263fd0a558SYan, Zheng */ 472749b25e05SJeff Mahoney int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans, 47283fd0a558SYan, Zheng struct btrfs_pending_snapshot *pending) 47293fd0a558SYan, Zheng { 47303fd0a558SYan, Zheng struct btrfs_root *root = pending->root; 47313fd0a558SYan, Zheng struct btrfs_root *reloc_root; 47323fd0a558SYan, Zheng struct btrfs_root *new_root; 473310995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 47343fd0a558SYan, Zheng int ret; 47353fd0a558SYan, Zheng 47366282675eSQu Wenruo if (!rc || !have_reloc_root(root)) 473749b25e05SJeff Mahoney return 0; 47383fd0a558SYan, Zheng 47393fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 47403fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated; 47413fd0a558SYan, Zheng 47423fd0a558SYan, Zheng if (rc->merge_reloc_tree) { 47433fd0a558SYan, Zheng ret = btrfs_block_rsv_migrate(&pending->block_rsv, 47443fd0a558SYan, Zheng rc->block_rsv, 47453a584174SLu Fengqi rc->nodes_relocated, true); 474649b25e05SJeff Mahoney if (ret) 474749b25e05SJeff Mahoney return ret; 47483fd0a558SYan, Zheng } 47493fd0a558SYan, Zheng 47503fd0a558SYan, Zheng new_root = pending->snap; 47513fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root->reloc_root, 47523fd0a558SYan, Zheng new_root->root_key.objectid); 475349b25e05SJeff Mahoney if (IS_ERR(reloc_root)) 475449b25e05SJeff Mahoney return PTR_ERR(reloc_root); 47553fd0a558SYan, Zheng 4756ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 4757ffd7b339SJeff Mahoney BUG_ON(ret < 0); 4758f44deb74SJosef Bacik new_root->reloc_root = btrfs_grab_root(reloc_root); 47593fd0a558SYan, Zheng 476049b25e05SJeff Mahoney if (rc->create_reloc_tree) 47613fd0a558SYan, Zheng ret = clone_backref_node(trans, rc, root, reloc_root); 476249b25e05SJeff Mahoney return ret; 47633fd0a558SYan, Zheng } 4764