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; 1255d4f98a2SYan Zheng }; 1265d4f98a2SYan Zheng 1275d4f98a2SYan Zheng /* 1285d4f98a2SYan Zheng * present a block pointer in the backref cache 1295d4f98a2SYan Zheng */ 1305d4f98a2SYan Zheng struct backref_edge { 1315d4f98a2SYan Zheng struct list_head list[2]; 1325d4f98a2SYan Zheng struct backref_node *node[2]; 1335d4f98a2SYan Zheng }; 1345d4f98a2SYan Zheng 1355d4f98a2SYan Zheng #define LOWER 0 1365d4f98a2SYan Zheng #define UPPER 1 1370647bf56SWang Shilong #define RELOCATION_RESERVED_NODES 256 1385d4f98a2SYan Zheng 1395d4f98a2SYan Zheng struct backref_cache { 1405d4f98a2SYan Zheng /* red black tree of all backref nodes in the cache */ 1415d4f98a2SYan Zheng struct rb_root rb_root; 1423fd0a558SYan, Zheng /* for passing backref nodes to btrfs_reloc_cow_block */ 1433fd0a558SYan, Zheng struct backref_node *path[BTRFS_MAX_LEVEL]; 1443fd0a558SYan, Zheng /* 1453fd0a558SYan, Zheng * list of blocks that have been cowed but some block 1463fd0a558SYan, Zheng * pointers in upper level blocks may not reflect the 1473fd0a558SYan, Zheng * new location 1483fd0a558SYan, Zheng */ 1495d4f98a2SYan Zheng struct list_head pending[BTRFS_MAX_LEVEL]; 1503fd0a558SYan, Zheng /* list of backref nodes with no child node */ 1513fd0a558SYan, Zheng struct list_head leaves; 1523fd0a558SYan, Zheng /* list of blocks that have been cowed in current transaction */ 1533fd0a558SYan, Zheng struct list_head changed; 1543fd0a558SYan, Zheng /* list of detached backref node. */ 1553fd0a558SYan, Zheng struct list_head detached; 1563fd0a558SYan, Zheng 1573fd0a558SYan, Zheng u64 last_trans; 1583fd0a558SYan, Zheng 1593fd0a558SYan, Zheng int nr_nodes; 1603fd0a558SYan, Zheng int nr_edges; 1615d4f98a2SYan Zheng }; 1625d4f98a2SYan Zheng 1635d4f98a2SYan Zheng /* 1645d4f98a2SYan Zheng * map address of tree root to tree 1655d4f98a2SYan Zheng */ 1665d4f98a2SYan Zheng struct mapping_node { 1675d4f98a2SYan Zheng struct rb_node rb_node; 1685d4f98a2SYan Zheng u64 bytenr; 1695d4f98a2SYan Zheng void *data; 1705d4f98a2SYan Zheng }; 1715d4f98a2SYan Zheng 1725d4f98a2SYan Zheng struct mapping_tree { 1735d4f98a2SYan Zheng struct rb_root rb_root; 1745d4f98a2SYan Zheng spinlock_t lock; 1755d4f98a2SYan Zheng }; 1765d4f98a2SYan Zheng 1775d4f98a2SYan Zheng /* 1785d4f98a2SYan Zheng * present a tree block to process 1795d4f98a2SYan Zheng */ 1805d4f98a2SYan Zheng struct tree_block { 1815d4f98a2SYan Zheng struct rb_node rb_node; 1825d4f98a2SYan Zheng u64 bytenr; 1835d4f98a2SYan Zheng struct btrfs_key key; 1845d4f98a2SYan Zheng unsigned int level:8; 1855d4f98a2SYan Zheng unsigned int key_ready:1; 1865d4f98a2SYan Zheng }; 1875d4f98a2SYan Zheng 1880257bb82SYan, Zheng #define MAX_EXTENTS 128 1890257bb82SYan, Zheng 1900257bb82SYan, Zheng struct file_extent_cluster { 1910257bb82SYan, Zheng u64 start; 1920257bb82SYan, Zheng u64 end; 1930257bb82SYan, Zheng u64 boundary[MAX_EXTENTS]; 1940257bb82SYan, Zheng unsigned int nr; 1950257bb82SYan, Zheng }; 1960257bb82SYan, Zheng 1975d4f98a2SYan Zheng struct reloc_control { 1985d4f98a2SYan Zheng /* block group to relocate */ 19932da5386SDavid Sterba struct btrfs_block_group *block_group; 2005d4f98a2SYan Zheng /* extent tree */ 2015d4f98a2SYan Zheng struct btrfs_root *extent_root; 2025d4f98a2SYan Zheng /* inode for moving data */ 2035d4f98a2SYan Zheng struct inode *data_inode; 2043fd0a558SYan, Zheng 2053fd0a558SYan, Zheng struct btrfs_block_rsv *block_rsv; 2063fd0a558SYan, Zheng 2073fd0a558SYan, Zheng struct backref_cache backref_cache; 2083fd0a558SYan, Zheng 2093fd0a558SYan, Zheng struct file_extent_cluster cluster; 2105d4f98a2SYan Zheng /* tree blocks have been processed */ 2115d4f98a2SYan Zheng struct extent_io_tree processed_blocks; 2125d4f98a2SYan Zheng /* map start of tree root to corresponding reloc tree */ 2135d4f98a2SYan Zheng struct mapping_tree reloc_root_tree; 2145d4f98a2SYan Zheng /* list of reloc trees */ 2155d4f98a2SYan Zheng struct list_head reloc_roots; 216d2311e69SQu Wenruo /* list of subvolume trees that get relocated */ 217d2311e69SQu Wenruo struct list_head dirty_subvol_roots; 2183fd0a558SYan, Zheng /* size of metadata reservation for merging reloc trees */ 2193fd0a558SYan, Zheng u64 merging_rsv_size; 2203fd0a558SYan, Zheng /* size of relocated tree nodes */ 2213fd0a558SYan, Zheng u64 nodes_relocated; 2220647bf56SWang Shilong /* reserved size for block group relocation*/ 2230647bf56SWang Shilong u64 reserved_bytes; 2243fd0a558SYan, Zheng 2255d4f98a2SYan Zheng u64 search_start; 2265d4f98a2SYan Zheng u64 extents_found; 2273fd0a558SYan, Zheng 2283fd0a558SYan, Zheng unsigned int stage:8; 2293fd0a558SYan, Zheng unsigned int create_reloc_tree:1; 2303fd0a558SYan, Zheng unsigned int merge_reloc_tree:1; 2315d4f98a2SYan Zheng unsigned int found_file_extent:1; 2325d4f98a2SYan Zheng }; 2335d4f98a2SYan Zheng 2345d4f98a2SYan Zheng /* stages of data relocation */ 2355d4f98a2SYan Zheng #define MOVE_DATA_EXTENTS 0 2365d4f98a2SYan Zheng #define UPDATE_DATA_PTRS 1 2375d4f98a2SYan Zheng 2383fd0a558SYan, Zheng static void remove_backref_node(struct backref_cache *cache, 2393fd0a558SYan, Zheng struct backref_node *node); 2403fd0a558SYan, Zheng static void __mark_block_processed(struct reloc_control *rc, 2413fd0a558SYan, Zheng struct backref_node *node); 2425d4f98a2SYan Zheng 2435d4f98a2SYan Zheng static void mapping_tree_init(struct mapping_tree *tree) 2445d4f98a2SYan Zheng { 2456bef4d31SEric Paris tree->rb_root = RB_ROOT; 2465d4f98a2SYan Zheng spin_lock_init(&tree->lock); 2475d4f98a2SYan Zheng } 2485d4f98a2SYan Zheng 2495d4f98a2SYan Zheng static void backref_cache_init(struct backref_cache *cache) 2505d4f98a2SYan Zheng { 2515d4f98a2SYan Zheng int i; 2526bef4d31SEric Paris cache->rb_root = RB_ROOT; 2535d4f98a2SYan Zheng for (i = 0; i < BTRFS_MAX_LEVEL; i++) 2545d4f98a2SYan Zheng INIT_LIST_HEAD(&cache->pending[i]); 2553fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->changed); 2563fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->detached); 2573fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->leaves); 2585d4f98a2SYan Zheng } 2595d4f98a2SYan Zheng 2603fd0a558SYan, Zheng static void backref_cache_cleanup(struct backref_cache *cache) 2615d4f98a2SYan Zheng { 2623fd0a558SYan, Zheng struct backref_node *node; 2633fd0a558SYan, Zheng int i; 2643fd0a558SYan, Zheng 2653fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 2663fd0a558SYan, Zheng node = list_entry(cache->detached.next, 2673fd0a558SYan, Zheng struct backref_node, list); 2683fd0a558SYan, Zheng remove_backref_node(cache, node); 2693fd0a558SYan, Zheng } 2703fd0a558SYan, Zheng 2713fd0a558SYan, Zheng while (!list_empty(&cache->leaves)) { 2723fd0a558SYan, Zheng node = list_entry(cache->leaves.next, 2733fd0a558SYan, Zheng struct backref_node, lower); 2743fd0a558SYan, Zheng remove_backref_node(cache, node); 2753fd0a558SYan, Zheng } 2763fd0a558SYan, Zheng 2773fd0a558SYan, Zheng cache->last_trans = 0; 2783fd0a558SYan, Zheng 2793fd0a558SYan, Zheng for (i = 0; i < BTRFS_MAX_LEVEL; i++) 280f4907095SLiu Bo ASSERT(list_empty(&cache->pending[i])); 281f4907095SLiu Bo ASSERT(list_empty(&cache->changed)); 282f4907095SLiu Bo ASSERT(list_empty(&cache->detached)); 283f4907095SLiu Bo ASSERT(RB_EMPTY_ROOT(&cache->rb_root)); 284f4907095SLiu Bo ASSERT(!cache->nr_nodes); 285f4907095SLiu Bo ASSERT(!cache->nr_edges); 2863fd0a558SYan, Zheng } 2873fd0a558SYan, Zheng 2883fd0a558SYan, Zheng static struct backref_node *alloc_backref_node(struct backref_cache *cache) 2893fd0a558SYan, Zheng { 2903fd0a558SYan, Zheng struct backref_node *node; 2913fd0a558SYan, Zheng 2923fd0a558SYan, Zheng node = kzalloc(sizeof(*node), GFP_NOFS); 2933fd0a558SYan, Zheng if (node) { 2943fd0a558SYan, Zheng INIT_LIST_HEAD(&node->list); 2955d4f98a2SYan Zheng INIT_LIST_HEAD(&node->upper); 2965d4f98a2SYan Zheng INIT_LIST_HEAD(&node->lower); 2975d4f98a2SYan Zheng RB_CLEAR_NODE(&node->rb_node); 2983fd0a558SYan, Zheng cache->nr_nodes++; 2993fd0a558SYan, Zheng } 3003fd0a558SYan, Zheng return node; 3013fd0a558SYan, Zheng } 3023fd0a558SYan, Zheng 3033fd0a558SYan, Zheng static void free_backref_node(struct backref_cache *cache, 3043fd0a558SYan, Zheng struct backref_node *node) 3053fd0a558SYan, Zheng { 3063fd0a558SYan, Zheng if (node) { 3073fd0a558SYan, Zheng cache->nr_nodes--; 30800246528SJosef Bacik btrfs_put_root(node->root); 3093fd0a558SYan, Zheng kfree(node); 3103fd0a558SYan, Zheng } 3113fd0a558SYan, Zheng } 3123fd0a558SYan, Zheng 3133fd0a558SYan, Zheng static struct backref_edge *alloc_backref_edge(struct backref_cache *cache) 3143fd0a558SYan, Zheng { 3153fd0a558SYan, Zheng struct backref_edge *edge; 3163fd0a558SYan, Zheng 3173fd0a558SYan, Zheng edge = kzalloc(sizeof(*edge), GFP_NOFS); 3183fd0a558SYan, Zheng if (edge) 3193fd0a558SYan, Zheng cache->nr_edges++; 3203fd0a558SYan, Zheng return edge; 3213fd0a558SYan, Zheng } 3223fd0a558SYan, Zheng 3233fd0a558SYan, Zheng static void free_backref_edge(struct backref_cache *cache, 3243fd0a558SYan, Zheng struct backref_edge *edge) 3253fd0a558SYan, Zheng { 3263fd0a558SYan, Zheng if (edge) { 3273fd0a558SYan, Zheng cache->nr_edges--; 3283fd0a558SYan, Zheng kfree(edge); 3293fd0a558SYan, Zheng } 3305d4f98a2SYan Zheng } 3315d4f98a2SYan Zheng 3325d4f98a2SYan Zheng static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr, 3335d4f98a2SYan Zheng struct rb_node *node) 3345d4f98a2SYan Zheng { 3355d4f98a2SYan Zheng struct rb_node **p = &root->rb_node; 3365d4f98a2SYan Zheng struct rb_node *parent = NULL; 3375d4f98a2SYan Zheng struct tree_entry *entry; 3385d4f98a2SYan Zheng 3395d4f98a2SYan Zheng while (*p) { 3405d4f98a2SYan Zheng parent = *p; 3415d4f98a2SYan Zheng entry = rb_entry(parent, struct tree_entry, rb_node); 3425d4f98a2SYan Zheng 3435d4f98a2SYan Zheng if (bytenr < entry->bytenr) 3445d4f98a2SYan Zheng p = &(*p)->rb_left; 3455d4f98a2SYan Zheng else if (bytenr > entry->bytenr) 3465d4f98a2SYan Zheng p = &(*p)->rb_right; 3475d4f98a2SYan Zheng else 3485d4f98a2SYan Zheng return parent; 3495d4f98a2SYan Zheng } 3505d4f98a2SYan Zheng 3515d4f98a2SYan Zheng rb_link_node(node, parent, p); 3525d4f98a2SYan Zheng rb_insert_color(node, root); 3535d4f98a2SYan Zheng return NULL; 3545d4f98a2SYan Zheng } 3555d4f98a2SYan Zheng 3565d4f98a2SYan Zheng static struct rb_node *tree_search(struct rb_root *root, u64 bytenr) 3575d4f98a2SYan Zheng { 3585d4f98a2SYan Zheng struct rb_node *n = root->rb_node; 3595d4f98a2SYan Zheng struct tree_entry *entry; 3605d4f98a2SYan Zheng 3615d4f98a2SYan Zheng while (n) { 3625d4f98a2SYan Zheng entry = rb_entry(n, struct tree_entry, rb_node); 3635d4f98a2SYan Zheng 3645d4f98a2SYan Zheng if (bytenr < entry->bytenr) 3655d4f98a2SYan Zheng n = n->rb_left; 3665d4f98a2SYan Zheng else if (bytenr > entry->bytenr) 3675d4f98a2SYan Zheng n = n->rb_right; 3685d4f98a2SYan Zheng else 3695d4f98a2SYan Zheng return n; 3705d4f98a2SYan Zheng } 3715d4f98a2SYan Zheng return NULL; 3725d4f98a2SYan Zheng } 3735d4f98a2SYan Zheng 37448a3b636SEric Sandeen static void backref_tree_panic(struct rb_node *rb_node, int errno, u64 bytenr) 37543c04fb1SJeff Mahoney { 37643c04fb1SJeff Mahoney 37743c04fb1SJeff Mahoney struct btrfs_fs_info *fs_info = NULL; 37843c04fb1SJeff Mahoney struct backref_node *bnode = rb_entry(rb_node, struct backref_node, 37943c04fb1SJeff Mahoney rb_node); 38043c04fb1SJeff Mahoney if (bnode->root) 38143c04fb1SJeff Mahoney fs_info = bnode->root->fs_info; 3825d163e0eSJeff Mahoney btrfs_panic(fs_info, errno, 3835d163e0eSJeff Mahoney "Inconsistency in backref cache found at offset %llu", 3845d163e0eSJeff Mahoney bytenr); 38543c04fb1SJeff Mahoney } 38643c04fb1SJeff Mahoney 3875d4f98a2SYan Zheng /* 3885d4f98a2SYan Zheng * walk up backref nodes until reach node presents tree root 3895d4f98a2SYan Zheng */ 3905d4f98a2SYan Zheng static struct backref_node *walk_up_backref(struct backref_node *node, 3915d4f98a2SYan Zheng struct backref_edge *edges[], 3925d4f98a2SYan Zheng int *index) 3935d4f98a2SYan Zheng { 3945d4f98a2SYan Zheng struct backref_edge *edge; 3955d4f98a2SYan Zheng int idx = *index; 3965d4f98a2SYan Zheng 3975d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 3985d4f98a2SYan Zheng edge = list_entry(node->upper.next, 3995d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 4005d4f98a2SYan Zheng edges[idx++] = edge; 4015d4f98a2SYan Zheng node = edge->node[UPPER]; 4025d4f98a2SYan Zheng } 4033fd0a558SYan, Zheng BUG_ON(node->detached); 4045d4f98a2SYan Zheng *index = idx; 4055d4f98a2SYan Zheng return node; 4065d4f98a2SYan Zheng } 4075d4f98a2SYan Zheng 4085d4f98a2SYan Zheng /* 4095d4f98a2SYan Zheng * walk down backref nodes to find start of next reference path 4105d4f98a2SYan Zheng */ 4115d4f98a2SYan Zheng static struct backref_node *walk_down_backref(struct backref_edge *edges[], 4125d4f98a2SYan Zheng int *index) 4135d4f98a2SYan Zheng { 4145d4f98a2SYan Zheng struct backref_edge *edge; 4155d4f98a2SYan Zheng struct backref_node *lower; 4165d4f98a2SYan Zheng int idx = *index; 4175d4f98a2SYan Zheng 4185d4f98a2SYan Zheng while (idx > 0) { 4195d4f98a2SYan Zheng edge = edges[idx - 1]; 4205d4f98a2SYan Zheng lower = edge->node[LOWER]; 4215d4f98a2SYan Zheng if (list_is_last(&edge->list[LOWER], &lower->upper)) { 4225d4f98a2SYan Zheng idx--; 4235d4f98a2SYan Zheng continue; 4245d4f98a2SYan Zheng } 4255d4f98a2SYan Zheng edge = list_entry(edge->list[LOWER].next, 4265d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 4275d4f98a2SYan Zheng edges[idx - 1] = edge; 4285d4f98a2SYan Zheng *index = idx; 4295d4f98a2SYan Zheng return edge->node[UPPER]; 4305d4f98a2SYan Zheng } 4315d4f98a2SYan Zheng *index = 0; 4325d4f98a2SYan Zheng return NULL; 4335d4f98a2SYan Zheng } 4345d4f98a2SYan Zheng 4353fd0a558SYan, Zheng static void unlock_node_buffer(struct backref_node *node) 4365d4f98a2SYan Zheng { 4375d4f98a2SYan Zheng if (node->locked) { 4385d4f98a2SYan Zheng btrfs_tree_unlock(node->eb); 4395d4f98a2SYan Zheng node->locked = 0; 4405d4f98a2SYan Zheng } 4413fd0a558SYan, Zheng } 4423fd0a558SYan, Zheng 4433fd0a558SYan, Zheng static void drop_node_buffer(struct backref_node *node) 4443fd0a558SYan, Zheng { 4453fd0a558SYan, Zheng if (node->eb) { 4463fd0a558SYan, Zheng unlock_node_buffer(node); 4475d4f98a2SYan Zheng free_extent_buffer(node->eb); 4485d4f98a2SYan Zheng node->eb = NULL; 4495d4f98a2SYan Zheng } 4505d4f98a2SYan Zheng } 4515d4f98a2SYan Zheng 4525d4f98a2SYan Zheng static void drop_backref_node(struct backref_cache *tree, 4535d4f98a2SYan Zheng struct backref_node *node) 4545d4f98a2SYan Zheng { 4555d4f98a2SYan Zheng BUG_ON(!list_empty(&node->upper)); 4565d4f98a2SYan Zheng 4575d4f98a2SYan Zheng drop_node_buffer(node); 4583fd0a558SYan, Zheng list_del(&node->list); 4595d4f98a2SYan Zheng list_del(&node->lower); 4603fd0a558SYan, Zheng if (!RB_EMPTY_NODE(&node->rb_node)) 4615d4f98a2SYan Zheng rb_erase(&node->rb_node, &tree->rb_root); 4623fd0a558SYan, Zheng free_backref_node(tree, node); 4635d4f98a2SYan Zheng } 4645d4f98a2SYan Zheng 4655d4f98a2SYan Zheng /* 4665d4f98a2SYan Zheng * remove a backref node from the backref cache 4675d4f98a2SYan Zheng */ 4685d4f98a2SYan Zheng static void remove_backref_node(struct backref_cache *cache, 4695d4f98a2SYan Zheng struct backref_node *node) 4705d4f98a2SYan Zheng { 4715d4f98a2SYan Zheng struct backref_node *upper; 4725d4f98a2SYan Zheng struct backref_edge *edge; 4735d4f98a2SYan Zheng 4745d4f98a2SYan Zheng if (!node) 4755d4f98a2SYan Zheng return; 4765d4f98a2SYan Zheng 4773fd0a558SYan, Zheng BUG_ON(!node->lowest && !node->detached); 4785d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 4795d4f98a2SYan Zheng edge = list_entry(node->upper.next, struct backref_edge, 4805d4f98a2SYan Zheng list[LOWER]); 4815d4f98a2SYan Zheng upper = edge->node[UPPER]; 4825d4f98a2SYan Zheng list_del(&edge->list[LOWER]); 4835d4f98a2SYan Zheng list_del(&edge->list[UPPER]); 4843fd0a558SYan, Zheng free_backref_edge(cache, edge); 4853fd0a558SYan, Zheng 4863fd0a558SYan, Zheng if (RB_EMPTY_NODE(&upper->rb_node)) { 4873fd0a558SYan, Zheng BUG_ON(!list_empty(&node->upper)); 4883fd0a558SYan, Zheng drop_backref_node(cache, node); 4893fd0a558SYan, Zheng node = upper; 4903fd0a558SYan, Zheng node->lowest = 1; 4913fd0a558SYan, Zheng continue; 4923fd0a558SYan, Zheng } 4935d4f98a2SYan Zheng /* 4943fd0a558SYan, Zheng * add the node to leaf node list if no other 4955d4f98a2SYan Zheng * child block cached. 4965d4f98a2SYan Zheng */ 4975d4f98a2SYan Zheng if (list_empty(&upper->lower)) { 4983fd0a558SYan, Zheng list_add_tail(&upper->lower, &cache->leaves); 4995d4f98a2SYan Zheng upper->lowest = 1; 5005d4f98a2SYan Zheng } 5015d4f98a2SYan Zheng } 5023fd0a558SYan, Zheng 5035d4f98a2SYan Zheng drop_backref_node(cache, node); 5045d4f98a2SYan Zheng } 5055d4f98a2SYan Zheng 5063fd0a558SYan, Zheng static void update_backref_node(struct backref_cache *cache, 5073fd0a558SYan, Zheng struct backref_node *node, u64 bytenr) 5083fd0a558SYan, Zheng { 5093fd0a558SYan, Zheng struct rb_node *rb_node; 5103fd0a558SYan, Zheng rb_erase(&node->rb_node, &cache->rb_root); 5113fd0a558SYan, Zheng node->bytenr = bytenr; 5123fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node); 51343c04fb1SJeff Mahoney if (rb_node) 51443c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, bytenr); 5153fd0a558SYan, Zheng } 5163fd0a558SYan, Zheng 5173fd0a558SYan, Zheng /* 5183fd0a558SYan, Zheng * update backref cache after a transaction commit 5193fd0a558SYan, Zheng */ 5203fd0a558SYan, Zheng static int update_backref_cache(struct btrfs_trans_handle *trans, 5213fd0a558SYan, Zheng struct backref_cache *cache) 5223fd0a558SYan, Zheng { 5233fd0a558SYan, Zheng struct backref_node *node; 5243fd0a558SYan, Zheng int level = 0; 5253fd0a558SYan, Zheng 5263fd0a558SYan, Zheng if (cache->last_trans == 0) { 5273fd0a558SYan, Zheng cache->last_trans = trans->transid; 5283fd0a558SYan, Zheng return 0; 5293fd0a558SYan, Zheng } 5303fd0a558SYan, Zheng 5313fd0a558SYan, Zheng if (cache->last_trans == trans->transid) 5323fd0a558SYan, Zheng return 0; 5333fd0a558SYan, Zheng 5343fd0a558SYan, Zheng /* 5353fd0a558SYan, Zheng * detached nodes are used to avoid unnecessary backref 5363fd0a558SYan, Zheng * lookup. transaction commit changes the extent tree. 5373fd0a558SYan, Zheng * so the detached nodes are no longer useful. 5383fd0a558SYan, Zheng */ 5393fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 5403fd0a558SYan, Zheng node = list_entry(cache->detached.next, 5413fd0a558SYan, Zheng struct backref_node, list); 5423fd0a558SYan, Zheng remove_backref_node(cache, node); 5433fd0a558SYan, Zheng } 5443fd0a558SYan, Zheng 5453fd0a558SYan, Zheng while (!list_empty(&cache->changed)) { 5463fd0a558SYan, Zheng node = list_entry(cache->changed.next, 5473fd0a558SYan, Zheng struct backref_node, list); 5483fd0a558SYan, Zheng list_del_init(&node->list); 5493fd0a558SYan, Zheng BUG_ON(node->pending); 5503fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 5513fd0a558SYan, Zheng } 5523fd0a558SYan, Zheng 5533fd0a558SYan, Zheng /* 5543fd0a558SYan, Zheng * some nodes can be left in the pending list if there were 5553fd0a558SYan, Zheng * errors during processing the pending nodes. 5563fd0a558SYan, Zheng */ 5573fd0a558SYan, Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 5583fd0a558SYan, Zheng list_for_each_entry(node, &cache->pending[level], list) { 5593fd0a558SYan, Zheng BUG_ON(!node->pending); 5603fd0a558SYan, Zheng if (node->bytenr == node->new_bytenr) 5613fd0a558SYan, Zheng continue; 5623fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 5633fd0a558SYan, Zheng } 5643fd0a558SYan, Zheng } 5653fd0a558SYan, Zheng 5663fd0a558SYan, Zheng cache->last_trans = 0; 5673fd0a558SYan, Zheng return 1; 5683fd0a558SYan, Zheng } 5693fd0a558SYan, Zheng 5706282675eSQu Wenruo static bool reloc_root_is_dead(struct btrfs_root *root) 5716282675eSQu Wenruo { 5726282675eSQu Wenruo /* 5736282675eSQu Wenruo * Pair with set_bit/clear_bit in clean_dirty_subvols and 5746282675eSQu Wenruo * btrfs_update_reloc_root. We need to see the updated bit before 5756282675eSQu Wenruo * trying to access reloc_root 5766282675eSQu Wenruo */ 5776282675eSQu Wenruo smp_rmb(); 5786282675eSQu Wenruo if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state)) 5796282675eSQu Wenruo return true; 5806282675eSQu Wenruo return false; 5816282675eSQu Wenruo } 5826282675eSQu Wenruo 5836282675eSQu Wenruo /* 5846282675eSQu Wenruo * Check if this subvolume tree has valid reloc tree. 5856282675eSQu Wenruo * 5866282675eSQu Wenruo * Reloc tree after swap is considered dead, thus not considered as valid. 5876282675eSQu Wenruo * This is enough for most callers, as they don't distinguish dead reloc root 5886282675eSQu Wenruo * from no reloc root. But should_ignore_root() below is a special case. 5896282675eSQu Wenruo */ 5906282675eSQu Wenruo static bool have_reloc_root(struct btrfs_root *root) 5916282675eSQu Wenruo { 5926282675eSQu Wenruo if (reloc_root_is_dead(root)) 5936282675eSQu Wenruo return false; 5946282675eSQu Wenruo if (!root->reloc_root) 5956282675eSQu Wenruo return false; 5966282675eSQu Wenruo return true; 5976282675eSQu Wenruo } 598f2a97a9dSDavid Sterba 5993fd0a558SYan, Zheng static int should_ignore_root(struct btrfs_root *root) 6003fd0a558SYan, Zheng { 6013fd0a558SYan, Zheng struct btrfs_root *reloc_root; 6023fd0a558SYan, Zheng 60327cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 6043fd0a558SYan, Zheng return 0; 6053fd0a558SYan, Zheng 6066282675eSQu Wenruo /* This root has been merged with its reloc tree, we can ignore it */ 6076282675eSQu Wenruo if (reloc_root_is_dead(root)) 6086282675eSQu Wenruo return 1; 6096282675eSQu Wenruo 6103fd0a558SYan, Zheng reloc_root = root->reloc_root; 6113fd0a558SYan, Zheng if (!reloc_root) 6123fd0a558SYan, Zheng return 0; 6133fd0a558SYan, Zheng 6143fd0a558SYan, Zheng if (btrfs_root_last_snapshot(&reloc_root->root_item) == 6153fd0a558SYan, Zheng root->fs_info->running_transaction->transid - 1) 6163fd0a558SYan, Zheng return 0; 6173fd0a558SYan, Zheng /* 6183fd0a558SYan, Zheng * if there is reloc tree and it was created in previous 6193fd0a558SYan, Zheng * transaction backref lookup can find the reloc tree, 6203fd0a558SYan, Zheng * so backref node for the fs tree root is useless for 6213fd0a558SYan, Zheng * relocation. 6223fd0a558SYan, Zheng */ 6233fd0a558SYan, Zheng return 1; 6243fd0a558SYan, Zheng } 6255d4f98a2SYan Zheng /* 6265d4f98a2SYan Zheng * find reloc tree by address of tree root 6275d4f98a2SYan Zheng */ 6285d4f98a2SYan Zheng static struct btrfs_root *find_reloc_root(struct reloc_control *rc, 6295d4f98a2SYan Zheng u64 bytenr) 6305d4f98a2SYan Zheng { 6315d4f98a2SYan Zheng struct rb_node *rb_node; 6325d4f98a2SYan Zheng struct mapping_node *node; 6335d4f98a2SYan Zheng struct btrfs_root *root = NULL; 6345d4f98a2SYan Zheng 6355d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 6365d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr); 6375d4f98a2SYan Zheng if (rb_node) { 6385d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 6395d4f98a2SYan Zheng root = (struct btrfs_root *)node->data; 6405d4f98a2SYan Zheng } 6415d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 64200246528SJosef Bacik return btrfs_grab_root(root); 6435d4f98a2SYan Zheng } 6445d4f98a2SYan Zheng 6455d4f98a2SYan Zheng static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info, 6465d4f98a2SYan Zheng u64 root_objectid) 6475d4f98a2SYan Zheng { 6485d4f98a2SYan Zheng struct btrfs_key key; 6495d4f98a2SYan Zheng 6505d4f98a2SYan Zheng key.objectid = root_objectid; 6515d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 6525d4f98a2SYan Zheng key.offset = (u64)-1; 6535d4f98a2SYan Zheng 654bc44d7c4SJosef Bacik return btrfs_get_fs_root(fs_info, &key, false); 6555d4f98a2SYan Zheng } 6565d4f98a2SYan Zheng 6575d4f98a2SYan Zheng static noinline_for_stack 6585d4f98a2SYan Zheng int find_inline_backref(struct extent_buffer *leaf, int slot, 6595d4f98a2SYan Zheng unsigned long *ptr, unsigned long *end) 6605d4f98a2SYan Zheng { 6613173a18fSJosef Bacik struct btrfs_key key; 6625d4f98a2SYan Zheng struct btrfs_extent_item *ei; 6635d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 6645d4f98a2SYan Zheng u32 item_size; 6655d4f98a2SYan Zheng 6663173a18fSJosef Bacik btrfs_item_key_to_cpu(leaf, &key, slot); 6673173a18fSJosef Bacik 6685d4f98a2SYan Zheng item_size = btrfs_item_size_nr(leaf, slot); 669ba3c2b19SNikolay Borisov if (item_size < sizeof(*ei)) { 670ba3c2b19SNikolay Borisov btrfs_print_v0_err(leaf->fs_info); 671ba3c2b19SNikolay Borisov btrfs_handle_fs_error(leaf->fs_info, -EINVAL, NULL); 672ba3c2b19SNikolay Borisov return 1; 673ba3c2b19SNikolay Borisov } 6745d4f98a2SYan Zheng ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item); 6755d4f98a2SYan Zheng WARN_ON(!(btrfs_extent_flags(leaf, ei) & 6765d4f98a2SYan Zheng BTRFS_EXTENT_FLAG_TREE_BLOCK)); 6775d4f98a2SYan Zheng 6783173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 6793173a18fSJosef Bacik item_size <= sizeof(*ei) + sizeof(*bi)) { 6805d4f98a2SYan Zheng WARN_ON(item_size < sizeof(*ei) + sizeof(*bi)); 6815d4f98a2SYan Zheng return 1; 6825d4f98a2SYan Zheng } 683d062d13cSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 684d062d13cSJosef Bacik item_size <= sizeof(*ei)) { 685d062d13cSJosef Bacik WARN_ON(item_size < sizeof(*ei)); 686d062d13cSJosef Bacik return 1; 687d062d13cSJosef Bacik } 6885d4f98a2SYan Zheng 6893173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) { 6905d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 6915d4f98a2SYan Zheng *ptr = (unsigned long)(bi + 1); 6923173a18fSJosef Bacik } else { 6933173a18fSJosef Bacik *ptr = (unsigned long)(ei + 1); 6943173a18fSJosef Bacik } 6955d4f98a2SYan Zheng *end = (unsigned long)ei + item_size; 6965d4f98a2SYan Zheng return 0; 6975d4f98a2SYan Zheng } 6985d4f98a2SYan Zheng 6995d4f98a2SYan Zheng /* 7005d4f98a2SYan Zheng * build backref tree for a given tree block. root of the backref tree 7015d4f98a2SYan Zheng * corresponds the tree block, leaves of the backref tree correspond 7025d4f98a2SYan Zheng * roots of b-trees that reference the tree block. 7035d4f98a2SYan Zheng * 7045d4f98a2SYan Zheng * the basic idea of this function is check backrefs of a given block 70501327610SNicholas D Steeves * to find upper level blocks that reference the block, and then check 70601327610SNicholas D Steeves * backrefs of these upper level blocks recursively. the recursion stop 7075d4f98a2SYan Zheng * when tree root is reached or backrefs for the block is cached. 7085d4f98a2SYan Zheng * 7095d4f98a2SYan Zheng * NOTE: if we find backrefs for a block are cached, we know backrefs 7105d4f98a2SYan Zheng * for all upper level blocks that directly/indirectly reference the 7115d4f98a2SYan Zheng * block are also cached. 7125d4f98a2SYan Zheng */ 7133fd0a558SYan, Zheng static noinline_for_stack 7143fd0a558SYan, Zheng struct backref_node *build_backref_tree(struct reloc_control *rc, 7155d4f98a2SYan Zheng struct btrfs_key *node_key, 7165d4f98a2SYan Zheng int level, u64 bytenr) 7175d4f98a2SYan Zheng { 7183fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 719fa6ac715SQu Wenruo struct btrfs_path *path1; /* For searching extent root */ 720fa6ac715SQu Wenruo struct btrfs_path *path2; /* For searching parent of TREE_BLOCK_REF */ 7215d4f98a2SYan Zheng struct extent_buffer *eb; 7225d4f98a2SYan Zheng struct btrfs_root *root; 7235d4f98a2SYan Zheng struct backref_node *cur; 7245d4f98a2SYan Zheng struct backref_node *upper; 7255d4f98a2SYan Zheng struct backref_node *lower; 7265d4f98a2SYan Zheng struct backref_node *node = NULL; 7275d4f98a2SYan Zheng struct backref_node *exist = NULL; 7285d4f98a2SYan Zheng struct backref_edge *edge; 7295d4f98a2SYan Zheng struct rb_node *rb_node; 7305d4f98a2SYan Zheng struct btrfs_key key; 7315d4f98a2SYan Zheng unsigned long end; 7325d4f98a2SYan Zheng unsigned long ptr; 733fa6ac715SQu Wenruo LIST_HEAD(list); /* Pending edge list, upper node needs to be checked */ 7343fd0a558SYan, Zheng LIST_HEAD(useless); 7353fd0a558SYan, Zheng int cowonly; 7365d4f98a2SYan Zheng int ret; 7375d4f98a2SYan Zheng int err = 0; 738b6c60c80SJosef Bacik bool need_check = true; 7395d4f98a2SYan Zheng 7405d4f98a2SYan Zheng path1 = btrfs_alloc_path(); 7415d4f98a2SYan Zheng path2 = btrfs_alloc_path(); 7425d4f98a2SYan Zheng if (!path1 || !path2) { 7435d4f98a2SYan Zheng err = -ENOMEM; 7445d4f98a2SYan Zheng goto out; 7455d4f98a2SYan Zheng } 746e4058b54SDavid Sterba path1->reada = READA_FORWARD; 747e4058b54SDavid Sterba path2->reada = READA_FORWARD; 7485d4f98a2SYan Zheng 7493fd0a558SYan, Zheng node = alloc_backref_node(cache); 7505d4f98a2SYan Zheng if (!node) { 7515d4f98a2SYan Zheng err = -ENOMEM; 7525d4f98a2SYan Zheng goto out; 7535d4f98a2SYan Zheng } 7545d4f98a2SYan Zheng 7555d4f98a2SYan Zheng node->bytenr = bytenr; 7565d4f98a2SYan Zheng node->level = level; 7575d4f98a2SYan Zheng node->lowest = 1; 7585d4f98a2SYan Zheng cur = node; 7595d4f98a2SYan Zheng again: 7605d4f98a2SYan Zheng end = 0; 7615d4f98a2SYan Zheng ptr = 0; 7625d4f98a2SYan Zheng key.objectid = cur->bytenr; 7633173a18fSJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 7645d4f98a2SYan Zheng key.offset = (u64)-1; 7655d4f98a2SYan Zheng 7665d4f98a2SYan Zheng path1->search_commit_root = 1; 7675d4f98a2SYan Zheng path1->skip_locking = 1; 7685d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1, 7695d4f98a2SYan Zheng 0, 0); 7705d4f98a2SYan Zheng if (ret < 0) { 7715d4f98a2SYan Zheng err = ret; 7725d4f98a2SYan Zheng goto out; 7735d4f98a2SYan Zheng } 77475bfb9afSJosef Bacik ASSERT(ret); 77575bfb9afSJosef Bacik ASSERT(path1->slots[0]); 7765d4f98a2SYan Zheng 7775d4f98a2SYan Zheng path1->slots[0]--; 7785d4f98a2SYan Zheng 7795d4f98a2SYan Zheng WARN_ON(cur->checked); 7805d4f98a2SYan Zheng if (!list_empty(&cur->upper)) { 7815d4f98a2SYan Zheng /* 78270f23fd6SJustin P. Mattock * the backref was added previously when processing 7835d4f98a2SYan Zheng * backref of type BTRFS_TREE_BLOCK_REF_KEY 7845d4f98a2SYan Zheng */ 78575bfb9afSJosef Bacik ASSERT(list_is_singular(&cur->upper)); 7865d4f98a2SYan Zheng edge = list_entry(cur->upper.next, struct backref_edge, 7875d4f98a2SYan Zheng list[LOWER]); 78875bfb9afSJosef Bacik ASSERT(list_empty(&edge->list[UPPER])); 7895d4f98a2SYan Zheng exist = edge->node[UPPER]; 7905d4f98a2SYan Zheng /* 7915d4f98a2SYan Zheng * add the upper level block to pending list if we need 7925d4f98a2SYan Zheng * check its backrefs 7935d4f98a2SYan Zheng */ 7945d4f98a2SYan Zheng if (!exist->checked) 7955d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 7965d4f98a2SYan Zheng } else { 7975d4f98a2SYan Zheng exist = NULL; 7985d4f98a2SYan Zheng } 7995d4f98a2SYan Zheng 8005d4f98a2SYan Zheng while (1) { 8015d4f98a2SYan Zheng cond_resched(); 8025d4f98a2SYan Zheng eb = path1->nodes[0]; 8035d4f98a2SYan Zheng 8045d4f98a2SYan Zheng if (ptr >= end) { 8055d4f98a2SYan Zheng if (path1->slots[0] >= btrfs_header_nritems(eb)) { 8065d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path1); 8075d4f98a2SYan Zheng if (ret < 0) { 8085d4f98a2SYan Zheng err = ret; 8095d4f98a2SYan Zheng goto out; 8105d4f98a2SYan Zheng } 8115d4f98a2SYan Zheng if (ret > 0) 8125d4f98a2SYan Zheng break; 8135d4f98a2SYan Zheng eb = path1->nodes[0]; 8145d4f98a2SYan Zheng } 8155d4f98a2SYan Zheng 8165d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &key, path1->slots[0]); 8175d4f98a2SYan Zheng if (key.objectid != cur->bytenr) { 8185d4f98a2SYan Zheng WARN_ON(exist); 8195d4f98a2SYan Zheng break; 8205d4f98a2SYan Zheng } 8215d4f98a2SYan Zheng 8223173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY || 8233173a18fSJosef Bacik key.type == BTRFS_METADATA_ITEM_KEY) { 8245d4f98a2SYan Zheng ret = find_inline_backref(eb, path1->slots[0], 8255d4f98a2SYan Zheng &ptr, &end); 8265d4f98a2SYan Zheng if (ret) 8275d4f98a2SYan Zheng goto next; 8285d4f98a2SYan Zheng } 8295d4f98a2SYan Zheng } 8305d4f98a2SYan Zheng 8315d4f98a2SYan Zheng if (ptr < end) { 8325d4f98a2SYan Zheng /* update key for inline back ref */ 8335d4f98a2SYan Zheng struct btrfs_extent_inline_ref *iref; 8343de28d57SLiu Bo int type; 8355d4f98a2SYan Zheng iref = (struct btrfs_extent_inline_ref *)ptr; 8363de28d57SLiu Bo type = btrfs_get_extent_inline_ref_type(eb, iref, 8373de28d57SLiu Bo BTRFS_REF_TYPE_BLOCK); 8383de28d57SLiu Bo if (type == BTRFS_REF_TYPE_INVALID) { 839af431dcbSSu Yue err = -EUCLEAN; 8403de28d57SLiu Bo goto out; 8413de28d57SLiu Bo } 8423de28d57SLiu Bo key.type = type; 8435d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 8443de28d57SLiu Bo 8455d4f98a2SYan Zheng WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY && 8465d4f98a2SYan Zheng key.type != BTRFS_SHARED_BLOCK_REF_KEY); 8475d4f98a2SYan Zheng } 8485d4f98a2SYan Zheng 849fa6ac715SQu Wenruo /* 850fa6ac715SQu Wenruo * Parent node found and matches current inline ref, no need to 851fa6ac715SQu Wenruo * rebuild this node for this inline ref. 852fa6ac715SQu Wenruo */ 8535d4f98a2SYan Zheng if (exist && 8545d4f98a2SYan Zheng ((key.type == BTRFS_TREE_BLOCK_REF_KEY && 8555d4f98a2SYan Zheng exist->owner == key.offset) || 8565d4f98a2SYan Zheng (key.type == BTRFS_SHARED_BLOCK_REF_KEY && 8575d4f98a2SYan Zheng exist->bytenr == key.offset))) { 8585d4f98a2SYan Zheng exist = NULL; 8595d4f98a2SYan Zheng goto next; 8605d4f98a2SYan Zheng } 8615d4f98a2SYan Zheng 862fa6ac715SQu Wenruo /* SHARED_BLOCK_REF means key.offset is the parent bytenr */ 8635d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) { 8645d4f98a2SYan Zheng if (key.objectid == key.offset) { 8655d4f98a2SYan Zheng /* 866fa6ac715SQu Wenruo * Only root blocks of reloc trees use backref 867fa6ac715SQu Wenruo * pointing to itself. 8685d4f98a2SYan Zheng */ 8695d4f98a2SYan Zheng root = find_reloc_root(rc, cur->bytenr); 87075bfb9afSJosef Bacik ASSERT(root); 8715d4f98a2SYan Zheng cur->root = root; 8725d4f98a2SYan Zheng break; 8735d4f98a2SYan Zheng } 8745d4f98a2SYan Zheng 8753fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 8765d4f98a2SYan Zheng if (!edge) { 8775d4f98a2SYan Zheng err = -ENOMEM; 8785d4f98a2SYan Zheng goto out; 8795d4f98a2SYan Zheng } 8805d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, key.offset); 8815d4f98a2SYan Zheng if (!rb_node) { 8823fd0a558SYan, Zheng upper = alloc_backref_node(cache); 8835d4f98a2SYan Zheng if (!upper) { 8843fd0a558SYan, Zheng free_backref_edge(cache, edge); 8855d4f98a2SYan Zheng err = -ENOMEM; 8865d4f98a2SYan Zheng goto out; 8875d4f98a2SYan Zheng } 8885d4f98a2SYan Zheng upper->bytenr = key.offset; 8895d4f98a2SYan Zheng upper->level = cur->level + 1; 8905d4f98a2SYan Zheng /* 8915d4f98a2SYan Zheng * backrefs for the upper level block isn't 8925d4f98a2SYan Zheng * cached, add the block to pending list 8935d4f98a2SYan Zheng */ 8945d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 8955d4f98a2SYan Zheng } else { 8965d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 8975d4f98a2SYan Zheng rb_node); 89875bfb9afSJosef Bacik ASSERT(upper->checked); 8995d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 9005d4f98a2SYan Zheng } 9013fd0a558SYan, Zheng list_add_tail(&edge->list[LOWER], &cur->upper); 9025d4f98a2SYan Zheng edge->node[LOWER] = cur; 9033fd0a558SYan, Zheng edge->node[UPPER] = upper; 9045d4f98a2SYan Zheng 9055d4f98a2SYan Zheng goto next; 9066d8ff4e4SDavid Sterba } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { 907ba3c2b19SNikolay Borisov err = -EINVAL; 908ba3c2b19SNikolay Borisov btrfs_print_v0_err(rc->extent_root->fs_info); 909ba3c2b19SNikolay Borisov btrfs_handle_fs_error(rc->extent_root->fs_info, err, 910ba3c2b19SNikolay Borisov NULL); 911ba3c2b19SNikolay Borisov goto out; 9125d4f98a2SYan Zheng } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) { 9135d4f98a2SYan Zheng goto next; 9145d4f98a2SYan Zheng } 9155d4f98a2SYan Zheng 916fa6ac715SQu Wenruo /* 917fa6ac715SQu Wenruo * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref offset 918fa6ac715SQu Wenruo * means the root objectid. We need to search the tree to get 919fa6ac715SQu Wenruo * its parent bytenr. 920fa6ac715SQu Wenruo */ 9215d4f98a2SYan Zheng root = read_fs_root(rc->extent_root->fs_info, key.offset); 9225d4f98a2SYan Zheng if (IS_ERR(root)) { 9235d4f98a2SYan Zheng err = PTR_ERR(root); 9245d4f98a2SYan Zheng goto out; 9255d4f98a2SYan Zheng } 9265d4f98a2SYan Zheng 92727cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 9283fd0a558SYan, Zheng cur->cowonly = 1; 9293fd0a558SYan, Zheng 9305d4f98a2SYan Zheng if (btrfs_root_level(&root->root_item) == cur->level) { 9315d4f98a2SYan Zheng /* tree root */ 93275bfb9afSJosef Bacik ASSERT(btrfs_root_bytenr(&root->root_item) == 9335d4f98a2SYan Zheng cur->bytenr); 9340b530bc5SJosef Bacik if (should_ignore_root(root)) { 93500246528SJosef Bacik btrfs_put_root(root); 9363fd0a558SYan, Zheng list_add(&cur->list, &useless); 9370b530bc5SJosef Bacik } else { 9385d4f98a2SYan Zheng cur->root = root; 9390b530bc5SJosef Bacik } 9405d4f98a2SYan Zheng break; 9415d4f98a2SYan Zheng } 9425d4f98a2SYan Zheng 9435d4f98a2SYan Zheng level = cur->level + 1; 9445d4f98a2SYan Zheng 945fa6ac715SQu Wenruo /* Search the tree to find parent blocks referring the block. */ 9465d4f98a2SYan Zheng path2->search_commit_root = 1; 9475d4f98a2SYan Zheng path2->skip_locking = 1; 9485d4f98a2SYan Zheng path2->lowest_level = level; 9495d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0); 9505d4f98a2SYan Zheng path2->lowest_level = 0; 9515d4f98a2SYan Zheng if (ret < 0) { 95200246528SJosef Bacik btrfs_put_root(root); 9535d4f98a2SYan Zheng err = ret; 9545d4f98a2SYan Zheng goto out; 9555d4f98a2SYan Zheng } 95633c66f43SYan Zheng if (ret > 0 && path2->slots[level] > 0) 95733c66f43SYan Zheng path2->slots[level]--; 9585d4f98a2SYan Zheng 9595d4f98a2SYan Zheng eb = path2->nodes[level]; 9603561b9dbSLiu Bo if (btrfs_node_blockptr(eb, path2->slots[level]) != 9613561b9dbSLiu Bo cur->bytenr) { 9623561b9dbSLiu Bo btrfs_err(root->fs_info, 9633561b9dbSLiu Bo "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)", 9644fd786e6SMisono Tomohiro cur->bytenr, level - 1, 9654fd786e6SMisono Tomohiro root->root_key.objectid, 9663561b9dbSLiu Bo node_key->objectid, node_key->type, 9673561b9dbSLiu Bo node_key->offset); 96800246528SJosef Bacik btrfs_put_root(root); 9693561b9dbSLiu Bo err = -ENOENT; 9703561b9dbSLiu Bo goto out; 9713561b9dbSLiu Bo } 9725d4f98a2SYan Zheng lower = cur; 973b6c60c80SJosef Bacik need_check = true; 974fa6ac715SQu Wenruo 975fa6ac715SQu Wenruo /* Add all nodes and edges in the path */ 9765d4f98a2SYan Zheng for (; level < BTRFS_MAX_LEVEL; level++) { 9775d4f98a2SYan Zheng if (!path2->nodes[level]) { 97875bfb9afSJosef Bacik ASSERT(btrfs_root_bytenr(&root->root_item) == 9795d4f98a2SYan Zheng lower->bytenr); 9800b530bc5SJosef Bacik if (should_ignore_root(root)) { 98100246528SJosef Bacik btrfs_put_root(root); 9823fd0a558SYan, Zheng list_add(&lower->list, &useless); 9830b530bc5SJosef Bacik } else { 9845d4f98a2SYan Zheng lower->root = root; 9850b530bc5SJosef Bacik } 9865d4f98a2SYan Zheng break; 9875d4f98a2SYan Zheng } 9885d4f98a2SYan Zheng 9893fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 9905d4f98a2SYan Zheng if (!edge) { 99100246528SJosef Bacik btrfs_put_root(root); 9925d4f98a2SYan Zheng err = -ENOMEM; 9935d4f98a2SYan Zheng goto out; 9945d4f98a2SYan Zheng } 9955d4f98a2SYan Zheng 9965d4f98a2SYan Zheng eb = path2->nodes[level]; 9975d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, eb->start); 9985d4f98a2SYan Zheng if (!rb_node) { 9993fd0a558SYan, Zheng upper = alloc_backref_node(cache); 10005d4f98a2SYan Zheng if (!upper) { 100100246528SJosef Bacik btrfs_put_root(root); 10023fd0a558SYan, Zheng free_backref_edge(cache, edge); 10035d4f98a2SYan Zheng err = -ENOMEM; 10045d4f98a2SYan Zheng goto out; 10055d4f98a2SYan Zheng } 10065d4f98a2SYan Zheng upper->bytenr = eb->start; 10075d4f98a2SYan Zheng upper->owner = btrfs_header_owner(eb); 10085d4f98a2SYan Zheng upper->level = lower->level + 1; 100927cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, 101027cdeb70SMiao Xie &root->state)) 10113fd0a558SYan, Zheng upper->cowonly = 1; 10125d4f98a2SYan Zheng 10135d4f98a2SYan Zheng /* 10145d4f98a2SYan Zheng * if we know the block isn't shared 10155d4f98a2SYan Zheng * we can void checking its backrefs. 10165d4f98a2SYan Zheng */ 10175d4f98a2SYan Zheng if (btrfs_block_can_be_shared(root, eb)) 10185d4f98a2SYan Zheng upper->checked = 0; 10195d4f98a2SYan Zheng else 10205d4f98a2SYan Zheng upper->checked = 1; 10215d4f98a2SYan Zheng 10225d4f98a2SYan Zheng /* 10235d4f98a2SYan Zheng * add the block to pending list if we 1024b6c60c80SJosef Bacik * need check its backrefs, we only do this once 1025b6c60c80SJosef Bacik * while walking up a tree as we will catch 1026b6c60c80SJosef Bacik * anything else later on. 10275d4f98a2SYan Zheng */ 1028b6c60c80SJosef Bacik if (!upper->checked && need_check) { 1029b6c60c80SJosef Bacik need_check = false; 10305d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], 10315d4f98a2SYan Zheng &list); 1032bbe90514SJosef Bacik } else { 1033bbe90514SJosef Bacik if (upper->checked) 1034bbe90514SJosef Bacik need_check = true; 10355d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 1036bbe90514SJosef Bacik } 10375d4f98a2SYan Zheng } else { 10385d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 10395d4f98a2SYan Zheng rb_node); 104075bfb9afSJosef Bacik ASSERT(upper->checked); 10415d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 10423fd0a558SYan, Zheng if (!upper->owner) 10433fd0a558SYan, Zheng upper->owner = btrfs_header_owner(eb); 10445d4f98a2SYan Zheng } 10455d4f98a2SYan Zheng list_add_tail(&edge->list[LOWER], &lower->upper); 10465d4f98a2SYan Zheng edge->node[LOWER] = lower; 10473fd0a558SYan, Zheng edge->node[UPPER] = upper; 10485d4f98a2SYan Zheng 10490b530bc5SJosef Bacik if (rb_node) { 105000246528SJosef Bacik btrfs_put_root(root); 10515d4f98a2SYan Zheng break; 10520b530bc5SJosef Bacik } 10535d4f98a2SYan Zheng lower = upper; 10545d4f98a2SYan Zheng upper = NULL; 10555d4f98a2SYan Zheng } 1056b3b4aa74SDavid Sterba btrfs_release_path(path2); 10575d4f98a2SYan Zheng next: 10585d4f98a2SYan Zheng if (ptr < end) { 10595d4f98a2SYan Zheng ptr += btrfs_extent_inline_ref_size(key.type); 10605d4f98a2SYan Zheng if (ptr >= end) { 10615d4f98a2SYan Zheng WARN_ON(ptr > end); 10625d4f98a2SYan Zheng ptr = 0; 10635d4f98a2SYan Zheng end = 0; 10645d4f98a2SYan Zheng } 10655d4f98a2SYan Zheng } 10665d4f98a2SYan Zheng if (ptr >= end) 10675d4f98a2SYan Zheng path1->slots[0]++; 10685d4f98a2SYan Zheng } 1069b3b4aa74SDavid Sterba btrfs_release_path(path1); 10705d4f98a2SYan Zheng 10715d4f98a2SYan Zheng cur->checked = 1; 10725d4f98a2SYan Zheng WARN_ON(exist); 10735d4f98a2SYan Zheng 10745d4f98a2SYan Zheng /* the pending list isn't empty, take the first block to process */ 10755d4f98a2SYan Zheng if (!list_empty(&list)) { 10765d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, list[UPPER]); 10775d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 10785d4f98a2SYan Zheng cur = edge->node[UPPER]; 10795d4f98a2SYan Zheng goto again; 10805d4f98a2SYan Zheng } 10815d4f98a2SYan Zheng 10825d4f98a2SYan Zheng /* 10835d4f98a2SYan Zheng * everything goes well, connect backref nodes and insert backref nodes 10845d4f98a2SYan Zheng * into the cache. 10855d4f98a2SYan Zheng */ 108675bfb9afSJosef Bacik ASSERT(node->checked); 10873fd0a558SYan, Zheng cowonly = node->cowonly; 10883fd0a558SYan, Zheng if (!cowonly) { 10893fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, 10903fd0a558SYan, Zheng &node->rb_node); 109143c04fb1SJeff Mahoney if (rb_node) 109243c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 10933fd0a558SYan, Zheng list_add_tail(&node->lower, &cache->leaves); 10943fd0a558SYan, Zheng } 10955d4f98a2SYan Zheng 10965d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) 10975d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 10985d4f98a2SYan Zheng 10995d4f98a2SYan Zheng while (!list_empty(&list)) { 11005d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, list[UPPER]); 11015d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 11025d4f98a2SYan Zheng upper = edge->node[UPPER]; 11033fd0a558SYan, Zheng if (upper->detached) { 11043fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 11053fd0a558SYan, Zheng lower = edge->node[LOWER]; 11063fd0a558SYan, Zheng free_backref_edge(cache, edge); 11073fd0a558SYan, Zheng if (list_empty(&lower->upper)) 11083fd0a558SYan, Zheng list_add(&lower->list, &useless); 11093fd0a558SYan, Zheng continue; 11103fd0a558SYan, Zheng } 11115d4f98a2SYan Zheng 11125d4f98a2SYan Zheng if (!RB_EMPTY_NODE(&upper->rb_node)) { 11135d4f98a2SYan Zheng if (upper->lowest) { 11145d4f98a2SYan Zheng list_del_init(&upper->lower); 11155d4f98a2SYan Zheng upper->lowest = 0; 11165d4f98a2SYan Zheng } 11175d4f98a2SYan Zheng 11185d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 11195d4f98a2SYan Zheng continue; 11205d4f98a2SYan Zheng } 11215d4f98a2SYan Zheng 112275bfb9afSJosef Bacik if (!upper->checked) { 112375bfb9afSJosef Bacik /* 112475bfb9afSJosef Bacik * Still want to blow up for developers since this is a 112575bfb9afSJosef Bacik * logic bug. 112675bfb9afSJosef Bacik */ 112775bfb9afSJosef Bacik ASSERT(0); 112875bfb9afSJosef Bacik err = -EINVAL; 112975bfb9afSJosef Bacik goto out; 113075bfb9afSJosef Bacik } 113175bfb9afSJosef Bacik if (cowonly != upper->cowonly) { 113275bfb9afSJosef Bacik ASSERT(0); 113375bfb9afSJosef Bacik err = -EINVAL; 113475bfb9afSJosef Bacik goto out; 113575bfb9afSJosef Bacik } 113675bfb9afSJosef Bacik 11373fd0a558SYan, Zheng if (!cowonly) { 11385d4f98a2SYan Zheng rb_node = tree_insert(&cache->rb_root, upper->bytenr, 11395d4f98a2SYan Zheng &upper->rb_node); 114043c04fb1SJeff Mahoney if (rb_node) 114143c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 114243c04fb1SJeff Mahoney upper->bytenr); 11433fd0a558SYan, Zheng } 11445d4f98a2SYan Zheng 11455d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 11465d4f98a2SYan Zheng 11475d4f98a2SYan Zheng list_for_each_entry(edge, &upper->upper, list[LOWER]) 11485d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 11495d4f98a2SYan Zheng } 11503fd0a558SYan, Zheng /* 11513fd0a558SYan, Zheng * process useless backref nodes. backref nodes for tree leaves 11523fd0a558SYan, Zheng * are deleted from the cache. backref nodes for upper level 11533fd0a558SYan, Zheng * tree blocks are left in the cache to avoid unnecessary backref 11543fd0a558SYan, Zheng * lookup. 11553fd0a558SYan, Zheng */ 11563fd0a558SYan, Zheng while (!list_empty(&useless)) { 11573fd0a558SYan, Zheng upper = list_entry(useless.next, struct backref_node, list); 11583fd0a558SYan, Zheng list_del_init(&upper->list); 115975bfb9afSJosef Bacik ASSERT(list_empty(&upper->upper)); 11603fd0a558SYan, Zheng if (upper == node) 11613fd0a558SYan, Zheng node = NULL; 11623fd0a558SYan, Zheng if (upper->lowest) { 11633fd0a558SYan, Zheng list_del_init(&upper->lower); 11643fd0a558SYan, Zheng upper->lowest = 0; 11653fd0a558SYan, Zheng } 11663fd0a558SYan, Zheng while (!list_empty(&upper->lower)) { 11673fd0a558SYan, Zheng edge = list_entry(upper->lower.next, 11683fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 11693fd0a558SYan, Zheng list_del(&edge->list[UPPER]); 11703fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 11713fd0a558SYan, Zheng lower = edge->node[LOWER]; 11723fd0a558SYan, Zheng free_backref_edge(cache, edge); 11733fd0a558SYan, Zheng 11743fd0a558SYan, Zheng if (list_empty(&lower->upper)) 11753fd0a558SYan, Zheng list_add(&lower->list, &useless); 11763fd0a558SYan, Zheng } 11773fd0a558SYan, Zheng __mark_block_processed(rc, upper); 11783fd0a558SYan, Zheng if (upper->level > 0) { 11793fd0a558SYan, Zheng list_add(&upper->list, &cache->detached); 11803fd0a558SYan, Zheng upper->detached = 1; 11813fd0a558SYan, Zheng } else { 11823fd0a558SYan, Zheng rb_erase(&upper->rb_node, &cache->rb_root); 11833fd0a558SYan, Zheng free_backref_node(cache, upper); 11843fd0a558SYan, Zheng } 11853fd0a558SYan, Zheng } 11865d4f98a2SYan Zheng out: 11875d4f98a2SYan Zheng btrfs_free_path(path1); 11885d4f98a2SYan Zheng btrfs_free_path(path2); 11895d4f98a2SYan Zheng if (err) { 11903fd0a558SYan, Zheng while (!list_empty(&useless)) { 11913fd0a558SYan, Zheng lower = list_entry(useless.next, 119275bfb9afSJosef Bacik struct backref_node, list); 119375bfb9afSJosef Bacik list_del_init(&lower->list); 11943fd0a558SYan, Zheng } 119575bfb9afSJosef Bacik while (!list_empty(&list)) { 119675bfb9afSJosef Bacik edge = list_first_entry(&list, struct backref_edge, 119775bfb9afSJosef Bacik list[UPPER]); 119875bfb9afSJosef Bacik list_del(&edge->list[UPPER]); 11993fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 120075bfb9afSJosef Bacik lower = edge->node[LOWER]; 12015d4f98a2SYan Zheng upper = edge->node[UPPER]; 12023fd0a558SYan, Zheng free_backref_edge(cache, edge); 120375bfb9afSJosef Bacik 120475bfb9afSJosef Bacik /* 120575bfb9afSJosef Bacik * Lower is no longer linked to any upper backref nodes 120675bfb9afSJosef Bacik * and isn't in the cache, we can free it ourselves. 120775bfb9afSJosef Bacik */ 120875bfb9afSJosef Bacik if (list_empty(&lower->upper) && 120975bfb9afSJosef Bacik RB_EMPTY_NODE(&lower->rb_node)) 121075bfb9afSJosef Bacik list_add(&lower->list, &useless); 121175bfb9afSJosef Bacik 121275bfb9afSJosef Bacik if (!RB_EMPTY_NODE(&upper->rb_node)) 121375bfb9afSJosef Bacik continue; 121475bfb9afSJosef Bacik 121501327610SNicholas D Steeves /* Add this guy's upper edges to the list to process */ 121675bfb9afSJosef Bacik list_for_each_entry(edge, &upper->upper, list[LOWER]) 121775bfb9afSJosef Bacik list_add_tail(&edge->list[UPPER], &list); 121875bfb9afSJosef Bacik if (list_empty(&upper->upper)) 121975bfb9afSJosef Bacik list_add(&upper->list, &useless); 122075bfb9afSJosef Bacik } 122175bfb9afSJosef Bacik 122275bfb9afSJosef Bacik while (!list_empty(&useless)) { 122375bfb9afSJosef Bacik lower = list_entry(useless.next, 122475bfb9afSJosef Bacik struct backref_node, list); 122575bfb9afSJosef Bacik list_del_init(&lower->list); 12260fd8c3daSLiu Bo if (lower == node) 12270fd8c3daSLiu Bo node = NULL; 122875bfb9afSJosef Bacik free_backref_node(cache, lower); 12295d4f98a2SYan Zheng } 12300fd8c3daSLiu Bo 12318e19c973SJosef Bacik remove_backref_node(cache, node); 12325d4f98a2SYan Zheng return ERR_PTR(err); 12335d4f98a2SYan Zheng } 123475bfb9afSJosef Bacik ASSERT(!node || !node->detached); 12355d4f98a2SYan Zheng return node; 12365d4f98a2SYan Zheng } 12375d4f98a2SYan Zheng 12385d4f98a2SYan Zheng /* 12393fd0a558SYan, Zheng * helper to add backref node for the newly created snapshot. 12403fd0a558SYan, Zheng * the backref node is created by cloning backref node that 12413fd0a558SYan, Zheng * corresponds to root of source tree 12423fd0a558SYan, Zheng */ 12433fd0a558SYan, Zheng static int clone_backref_node(struct btrfs_trans_handle *trans, 12443fd0a558SYan, Zheng struct reloc_control *rc, 12453fd0a558SYan, Zheng struct btrfs_root *src, 12463fd0a558SYan, Zheng struct btrfs_root *dest) 12473fd0a558SYan, Zheng { 12483fd0a558SYan, Zheng struct btrfs_root *reloc_root = src->reloc_root; 12493fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 12503fd0a558SYan, Zheng struct backref_node *node = NULL; 12513fd0a558SYan, Zheng struct backref_node *new_node; 12523fd0a558SYan, Zheng struct backref_edge *edge; 12533fd0a558SYan, Zheng struct backref_edge *new_edge; 12543fd0a558SYan, Zheng struct rb_node *rb_node; 12553fd0a558SYan, Zheng 12563fd0a558SYan, Zheng if (cache->last_trans > 0) 12573fd0a558SYan, Zheng update_backref_cache(trans, cache); 12583fd0a558SYan, Zheng 12593fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, src->commit_root->start); 12603fd0a558SYan, Zheng if (rb_node) { 12613fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, rb_node); 12623fd0a558SYan, Zheng if (node->detached) 12633fd0a558SYan, Zheng node = NULL; 12643fd0a558SYan, Zheng else 12653fd0a558SYan, Zheng BUG_ON(node->new_bytenr != reloc_root->node->start); 12663fd0a558SYan, Zheng } 12673fd0a558SYan, Zheng 12683fd0a558SYan, Zheng if (!node) { 12693fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, 12703fd0a558SYan, Zheng reloc_root->commit_root->start); 12713fd0a558SYan, Zheng if (rb_node) { 12723fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, 12733fd0a558SYan, Zheng rb_node); 12743fd0a558SYan, Zheng BUG_ON(node->detached); 12753fd0a558SYan, Zheng } 12763fd0a558SYan, Zheng } 12773fd0a558SYan, Zheng 12783fd0a558SYan, Zheng if (!node) 12793fd0a558SYan, Zheng return 0; 12803fd0a558SYan, Zheng 12813fd0a558SYan, Zheng new_node = alloc_backref_node(cache); 12823fd0a558SYan, Zheng if (!new_node) 12833fd0a558SYan, Zheng return -ENOMEM; 12843fd0a558SYan, Zheng 12853fd0a558SYan, Zheng new_node->bytenr = dest->node->start; 12863fd0a558SYan, Zheng new_node->level = node->level; 12873fd0a558SYan, Zheng new_node->lowest = node->lowest; 12886848ad64SYan, Zheng new_node->checked = 1; 128900246528SJosef Bacik new_node->root = btrfs_grab_root(dest); 12900b530bc5SJosef Bacik ASSERT(new_node->root); 12913fd0a558SYan, Zheng 12923fd0a558SYan, Zheng if (!node->lowest) { 12933fd0a558SYan, Zheng list_for_each_entry(edge, &node->lower, list[UPPER]) { 12943fd0a558SYan, Zheng new_edge = alloc_backref_edge(cache); 12953fd0a558SYan, Zheng if (!new_edge) 12963fd0a558SYan, Zheng goto fail; 12973fd0a558SYan, Zheng 12983fd0a558SYan, Zheng new_edge->node[UPPER] = new_node; 12993fd0a558SYan, Zheng new_edge->node[LOWER] = edge->node[LOWER]; 13003fd0a558SYan, Zheng list_add_tail(&new_edge->list[UPPER], 13013fd0a558SYan, Zheng &new_node->lower); 13023fd0a558SYan, Zheng } 130376b9e23dSMiao Xie } else { 130476b9e23dSMiao Xie list_add_tail(&new_node->lower, &cache->leaves); 13053fd0a558SYan, Zheng } 13063fd0a558SYan, Zheng 13073fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, new_node->bytenr, 13083fd0a558SYan, Zheng &new_node->rb_node); 130943c04fb1SJeff Mahoney if (rb_node) 131043c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, new_node->bytenr); 13113fd0a558SYan, Zheng 13123fd0a558SYan, Zheng if (!new_node->lowest) { 13133fd0a558SYan, Zheng list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) { 13143fd0a558SYan, Zheng list_add_tail(&new_edge->list[LOWER], 13153fd0a558SYan, Zheng &new_edge->node[LOWER]->upper); 13163fd0a558SYan, Zheng } 13173fd0a558SYan, Zheng } 13183fd0a558SYan, Zheng return 0; 13193fd0a558SYan, Zheng fail: 13203fd0a558SYan, Zheng while (!list_empty(&new_node->lower)) { 13213fd0a558SYan, Zheng new_edge = list_entry(new_node->lower.next, 13223fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 13233fd0a558SYan, Zheng list_del(&new_edge->list[UPPER]); 13243fd0a558SYan, Zheng free_backref_edge(cache, new_edge); 13253fd0a558SYan, Zheng } 13263fd0a558SYan, Zheng free_backref_node(cache, new_node); 13273fd0a558SYan, Zheng return -ENOMEM; 13283fd0a558SYan, Zheng } 13293fd0a558SYan, Zheng 13303fd0a558SYan, Zheng /* 13315d4f98a2SYan Zheng * helper to add 'address of tree root -> reloc tree' mapping 13325d4f98a2SYan Zheng */ 1333ffd7b339SJeff Mahoney static int __must_check __add_reloc_root(struct btrfs_root *root) 13345d4f98a2SYan Zheng { 13350b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13365d4f98a2SYan Zheng struct rb_node *rb_node; 13375d4f98a2SYan Zheng struct mapping_node *node; 13380b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 13395d4f98a2SYan Zheng 13405d4f98a2SYan Zheng node = kmalloc(sizeof(*node), GFP_NOFS); 1341ffd7b339SJeff Mahoney if (!node) 1342ffd7b339SJeff Mahoney return -ENOMEM; 13435d4f98a2SYan Zheng 13445d4f98a2SYan Zheng node->bytenr = root->node->start; 13455d4f98a2SYan Zheng node->data = root; 13465d4f98a2SYan Zheng 13475d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 13485d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 13495d4f98a2SYan Zheng node->bytenr, &node->rb_node); 13505d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 1351ffd7b339SJeff Mahoney if (rb_node) { 13520b246afaSJeff Mahoney btrfs_panic(fs_info, -EEXIST, 13535d163e0eSJeff Mahoney "Duplicate root found for start=%llu while inserting into relocation tree", 13545d163e0eSJeff Mahoney node->bytenr); 1355ffd7b339SJeff Mahoney } 13565d4f98a2SYan Zheng 13575d4f98a2SYan Zheng list_add_tail(&root->root_list, &rc->reloc_roots); 13585d4f98a2SYan Zheng return 0; 13595d4f98a2SYan Zheng } 13605d4f98a2SYan Zheng 13615d4f98a2SYan Zheng /* 1362c974c464SWang Shilong * helper to delete the 'address of tree root -> reloc tree' 13635d4f98a2SYan Zheng * mapping 13645d4f98a2SYan Zheng */ 1365c974c464SWang Shilong static void __del_reloc_root(struct btrfs_root *root) 13665d4f98a2SYan Zheng { 13670b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13685d4f98a2SYan Zheng struct rb_node *rb_node; 13695d4f98a2SYan Zheng struct mapping_node *node = NULL; 13700b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 13715d4f98a2SYan Zheng 137265c6e82bSQu Wenruo if (rc && root->node) { 13735d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 13745d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1375c974c464SWang Shilong root->node->start); 1376c974c464SWang Shilong if (rb_node) { 1377c974c464SWang Shilong node = rb_entry(rb_node, struct mapping_node, rb_node); 1378c974c464SWang Shilong rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 1379c974c464SWang Shilong } 1380c974c464SWang Shilong spin_unlock(&rc->reloc_root_tree.lock); 1381c974c464SWang Shilong if (!node) 1382c974c464SWang Shilong return; 1383c974c464SWang Shilong BUG_ON((struct btrfs_root *)node->data != root); 1384389305b2SQu Wenruo } 1385c974c464SWang Shilong 13860b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1387c974c464SWang Shilong list_del_init(&root->root_list); 13880b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1389c974c464SWang Shilong kfree(node); 1390c974c464SWang Shilong } 1391c974c464SWang Shilong 1392c974c464SWang Shilong /* 1393c974c464SWang Shilong * helper to update the 'address of tree root -> reloc tree' 1394c974c464SWang Shilong * mapping 1395c974c464SWang Shilong */ 1396c974c464SWang Shilong static int __update_reloc_root(struct btrfs_root *root, u64 new_bytenr) 1397c974c464SWang Shilong { 13980b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1399c974c464SWang Shilong struct rb_node *rb_node; 1400c974c464SWang Shilong struct mapping_node *node = NULL; 14010b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 1402c974c464SWang Shilong 1403c974c464SWang Shilong spin_lock(&rc->reloc_root_tree.lock); 1404c974c464SWang Shilong rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1405c974c464SWang Shilong root->node->start); 14065d4f98a2SYan Zheng if (rb_node) { 14075d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 14085d4f98a2SYan Zheng rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 14095d4f98a2SYan Zheng } 14105d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 14115d4f98a2SYan Zheng 14128f71f3e0SLiu Bo if (!node) 14138f71f3e0SLiu Bo return 0; 14145d4f98a2SYan Zheng BUG_ON((struct btrfs_root *)node->data != root); 14155d4f98a2SYan Zheng 14165d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 1417c974c464SWang Shilong node->bytenr = new_bytenr; 14185d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 14195d4f98a2SYan Zheng node->bytenr, &node->rb_node); 14205d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 142143c04fb1SJeff Mahoney if (rb_node) 142243c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 14235d4f98a2SYan Zheng return 0; 14245d4f98a2SYan Zheng } 14255d4f98a2SYan Zheng 14263fd0a558SYan, Zheng static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans, 14273fd0a558SYan, Zheng struct btrfs_root *root, u64 objectid) 14285d4f98a2SYan Zheng { 14290b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 14305d4f98a2SYan Zheng struct btrfs_root *reloc_root; 14315d4f98a2SYan Zheng struct extent_buffer *eb; 14325d4f98a2SYan Zheng struct btrfs_root_item *root_item; 14335d4f98a2SYan Zheng struct btrfs_key root_key; 14345d4f98a2SYan Zheng int ret; 14355d4f98a2SYan Zheng 14365d4f98a2SYan Zheng root_item = kmalloc(sizeof(*root_item), GFP_NOFS); 14375d4f98a2SYan Zheng BUG_ON(!root_item); 14385d4f98a2SYan Zheng 14395d4f98a2SYan Zheng root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; 14405d4f98a2SYan Zheng root_key.type = BTRFS_ROOT_ITEM_KEY; 14413fd0a558SYan, Zheng root_key.offset = objectid; 14425d4f98a2SYan Zheng 14433fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 1444054570a1SFilipe Manana u64 commit_root_gen; 1445054570a1SFilipe Manana 14463fd0a558SYan, Zheng /* called by btrfs_init_reloc_root */ 14475d4f98a2SYan Zheng ret = btrfs_copy_root(trans, root, root->commit_root, &eb, 14485d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID); 14495d4f98a2SYan Zheng BUG_ON(ret); 1450054570a1SFilipe Manana /* 1451054570a1SFilipe Manana * Set the last_snapshot field to the generation of the commit 1452054570a1SFilipe Manana * root - like this ctree.c:btrfs_block_can_be_shared() behaves 1453054570a1SFilipe Manana * correctly (returns true) when the relocation root is created 1454054570a1SFilipe Manana * either inside the critical section of a transaction commit 1455054570a1SFilipe Manana * (through transaction.c:qgroup_account_snapshot()) and when 1456054570a1SFilipe Manana * it's created before the transaction commit is started. 1457054570a1SFilipe Manana */ 1458054570a1SFilipe Manana commit_root_gen = btrfs_header_generation(root->commit_root); 1459054570a1SFilipe Manana btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen); 14603fd0a558SYan, Zheng } else { 14613fd0a558SYan, Zheng /* 14623fd0a558SYan, Zheng * called by btrfs_reloc_post_snapshot_hook. 14633fd0a558SYan, Zheng * the source tree is a reloc tree, all tree blocks 14643fd0a558SYan, Zheng * modified after it was created have RELOC flag 14653fd0a558SYan, Zheng * set in their headers. so it's OK to not update 14663fd0a558SYan, Zheng * the 'last_snapshot'. 14673fd0a558SYan, Zheng */ 14683fd0a558SYan, Zheng ret = btrfs_copy_root(trans, root, root->node, &eb, 14693fd0a558SYan, Zheng BTRFS_TREE_RELOC_OBJECTID); 14703fd0a558SYan, Zheng BUG_ON(ret); 14713fd0a558SYan, Zheng } 14723fd0a558SYan, Zheng 14735d4f98a2SYan Zheng memcpy(root_item, &root->root_item, sizeof(*root_item)); 14745d4f98a2SYan Zheng btrfs_set_root_bytenr(root_item, eb->start); 14755d4f98a2SYan Zheng btrfs_set_root_level(root_item, btrfs_header_level(eb)); 14765d4f98a2SYan Zheng btrfs_set_root_generation(root_item, trans->transid); 14773fd0a558SYan, Zheng 14783fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 14793fd0a558SYan, Zheng btrfs_set_root_refs(root_item, 0); 14803fd0a558SYan, Zheng memset(&root_item->drop_progress, 0, 14813fd0a558SYan, Zheng sizeof(struct btrfs_disk_key)); 14825d4f98a2SYan Zheng root_item->drop_level = 0; 14833fd0a558SYan, Zheng } 14845d4f98a2SYan Zheng 14855d4f98a2SYan Zheng btrfs_tree_unlock(eb); 14865d4f98a2SYan Zheng free_extent_buffer(eb); 14875d4f98a2SYan Zheng 14880b246afaSJeff Mahoney ret = btrfs_insert_root(trans, fs_info->tree_root, 14895d4f98a2SYan Zheng &root_key, root_item); 14905d4f98a2SYan Zheng BUG_ON(ret); 14915d4f98a2SYan Zheng kfree(root_item); 14925d4f98a2SYan Zheng 14933dbf1738SJosef Bacik reloc_root = btrfs_read_tree_root(fs_info->tree_root, &root_key); 14945d4f98a2SYan Zheng BUG_ON(IS_ERR(reloc_root)); 14953dbf1738SJosef Bacik set_bit(BTRFS_ROOT_REF_COWS, &reloc_root->state); 14965d4f98a2SYan Zheng reloc_root->last_trans = trans->transid; 14973fd0a558SYan, Zheng return reloc_root; 14983fd0a558SYan, Zheng } 14993fd0a558SYan, Zheng 15003fd0a558SYan, Zheng /* 15013fd0a558SYan, Zheng * create reloc tree for a given fs tree. reloc tree is just a 15023fd0a558SYan, Zheng * snapshot of the fs tree with special root objectid. 15033fd0a558SYan, Zheng */ 15043fd0a558SYan, Zheng int btrfs_init_reloc_root(struct btrfs_trans_handle *trans, 15053fd0a558SYan, Zheng struct btrfs_root *root) 15063fd0a558SYan, Zheng { 15070b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 15083fd0a558SYan, Zheng struct btrfs_root *reloc_root; 15090b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 151020dd2cbfSMiao Xie struct btrfs_block_rsv *rsv; 15113fd0a558SYan, Zheng int clear_rsv = 0; 1512ffd7b339SJeff Mahoney int ret; 15133fd0a558SYan, Zheng 15142abc726aSJosef Bacik if (!rc || !rc->create_reloc_tree || 15152abc726aSJosef Bacik root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 15162abc726aSJosef Bacik return 0; 15172abc726aSJosef Bacik 15181fac4a54SQu Wenruo /* 15191fac4a54SQu Wenruo * The subvolume has reloc tree but the swap is finished, no need to 15201fac4a54SQu Wenruo * create/update the dead reloc tree 15211fac4a54SQu Wenruo */ 15226282675eSQu Wenruo if (reloc_root_is_dead(root)) 15231fac4a54SQu Wenruo return 0; 15241fac4a54SQu Wenruo 15253fd0a558SYan, Zheng if (root->reloc_root) { 15263fd0a558SYan, Zheng reloc_root = root->reloc_root; 15273fd0a558SYan, Zheng reloc_root->last_trans = trans->transid; 15283fd0a558SYan, Zheng return 0; 15293fd0a558SYan, Zheng } 15303fd0a558SYan, Zheng 153120dd2cbfSMiao Xie if (!trans->reloc_reserved) { 153220dd2cbfSMiao Xie rsv = trans->block_rsv; 15333fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 15343fd0a558SYan, Zheng clear_rsv = 1; 15353fd0a558SYan, Zheng } 15363fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root, root->root_key.objectid); 15373fd0a558SYan, Zheng if (clear_rsv) 153820dd2cbfSMiao Xie trans->block_rsv = rsv; 15395d4f98a2SYan Zheng 1540ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 1541ffd7b339SJeff Mahoney BUG_ON(ret < 0); 15425d4f98a2SYan Zheng root->reloc_root = reloc_root; 15435d4f98a2SYan Zheng return 0; 15445d4f98a2SYan Zheng } 15455d4f98a2SYan Zheng 15465d4f98a2SYan Zheng /* 15475d4f98a2SYan Zheng * update root item of reloc tree 15485d4f98a2SYan Zheng */ 15495d4f98a2SYan Zheng int btrfs_update_reloc_root(struct btrfs_trans_handle *trans, 15505d4f98a2SYan Zheng struct btrfs_root *root) 15515d4f98a2SYan Zheng { 15520b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 15535d4f98a2SYan Zheng struct btrfs_root *reloc_root; 15545d4f98a2SYan Zheng struct btrfs_root_item *root_item; 15555d4f98a2SYan Zheng int ret; 15565d4f98a2SYan Zheng 15576282675eSQu Wenruo if (!have_reloc_root(root)) 15587585717fSChris Mason goto out; 15595d4f98a2SYan Zheng 15605d4f98a2SYan Zheng reloc_root = root->reloc_root; 15615d4f98a2SYan Zheng root_item = &reloc_root->root_item; 15625d4f98a2SYan Zheng 1563d2311e69SQu Wenruo /* root->reloc_root will stay until current relocation finished */ 15640b246afaSJeff Mahoney if (fs_info->reloc_ctl->merge_reloc_tree && 15653fd0a558SYan, Zheng btrfs_root_refs(root_item) == 0) { 1566d2311e69SQu Wenruo set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 15676282675eSQu Wenruo /* 15686282675eSQu Wenruo * Mark the tree as dead before we change reloc_root so 15696282675eSQu Wenruo * have_reloc_root will not touch it from now on. 15706282675eSQu Wenruo */ 15716282675eSQu Wenruo smp_wmb(); 1572c974c464SWang Shilong __del_reloc_root(reloc_root); 15735d4f98a2SYan Zheng } 15745d4f98a2SYan Zheng 15755d4f98a2SYan Zheng if (reloc_root->commit_root != reloc_root->node) { 15765d4f98a2SYan Zheng btrfs_set_root_node(root_item, reloc_root->node); 15775d4f98a2SYan Zheng free_extent_buffer(reloc_root->commit_root); 15785d4f98a2SYan Zheng reloc_root->commit_root = btrfs_root_node(reloc_root); 15795d4f98a2SYan Zheng } 15805d4f98a2SYan Zheng 15810b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 15825d4f98a2SYan Zheng &reloc_root->root_key, root_item); 15835d4f98a2SYan Zheng BUG_ON(ret); 15847585717fSChris Mason 15857585717fSChris Mason out: 15865d4f98a2SYan Zheng return 0; 15875d4f98a2SYan Zheng } 15885d4f98a2SYan Zheng 15895d4f98a2SYan Zheng /* 15905d4f98a2SYan Zheng * helper to find first cached inode with inode number >= objectid 15915d4f98a2SYan Zheng * in a subvolume 15925d4f98a2SYan Zheng */ 15935d4f98a2SYan Zheng static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid) 15945d4f98a2SYan Zheng { 15955d4f98a2SYan Zheng struct rb_node *node; 15965d4f98a2SYan Zheng struct rb_node *prev; 15975d4f98a2SYan Zheng struct btrfs_inode *entry; 15985d4f98a2SYan Zheng struct inode *inode; 15995d4f98a2SYan Zheng 16005d4f98a2SYan Zheng spin_lock(&root->inode_lock); 16015d4f98a2SYan Zheng again: 16025d4f98a2SYan Zheng node = root->inode_tree.rb_node; 16035d4f98a2SYan Zheng prev = NULL; 16045d4f98a2SYan Zheng while (node) { 16055d4f98a2SYan Zheng prev = node; 16065d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 16075d4f98a2SYan Zheng 16084a0cc7caSNikolay Borisov if (objectid < btrfs_ino(entry)) 16095d4f98a2SYan Zheng node = node->rb_left; 16104a0cc7caSNikolay Borisov else if (objectid > btrfs_ino(entry)) 16115d4f98a2SYan Zheng node = node->rb_right; 16125d4f98a2SYan Zheng else 16135d4f98a2SYan Zheng break; 16145d4f98a2SYan Zheng } 16155d4f98a2SYan Zheng if (!node) { 16165d4f98a2SYan Zheng while (prev) { 16175d4f98a2SYan Zheng entry = rb_entry(prev, struct btrfs_inode, rb_node); 16184a0cc7caSNikolay Borisov if (objectid <= btrfs_ino(entry)) { 16195d4f98a2SYan Zheng node = prev; 16205d4f98a2SYan Zheng break; 16215d4f98a2SYan Zheng } 16225d4f98a2SYan Zheng prev = rb_next(prev); 16235d4f98a2SYan Zheng } 16245d4f98a2SYan Zheng } 16255d4f98a2SYan Zheng while (node) { 16265d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 16275d4f98a2SYan Zheng inode = igrab(&entry->vfs_inode); 16285d4f98a2SYan Zheng if (inode) { 16295d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 16305d4f98a2SYan Zheng return inode; 16315d4f98a2SYan Zheng } 16325d4f98a2SYan Zheng 16334a0cc7caSNikolay Borisov objectid = btrfs_ino(entry) + 1; 16345d4f98a2SYan Zheng if (cond_resched_lock(&root->inode_lock)) 16355d4f98a2SYan Zheng goto again; 16365d4f98a2SYan Zheng 16375d4f98a2SYan Zheng node = rb_next(node); 16385d4f98a2SYan Zheng } 16395d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 16405d4f98a2SYan Zheng return NULL; 16415d4f98a2SYan Zheng } 16425d4f98a2SYan Zheng 164332da5386SDavid Sterba static int in_block_group(u64 bytenr, struct btrfs_block_group *block_group) 16445d4f98a2SYan Zheng { 1645b3470b5dSDavid Sterba if (bytenr >= block_group->start && 1646b3470b5dSDavid Sterba bytenr < block_group->start + block_group->length) 16475d4f98a2SYan Zheng return 1; 16485d4f98a2SYan Zheng return 0; 16495d4f98a2SYan Zheng } 16505d4f98a2SYan Zheng 16515d4f98a2SYan Zheng /* 16525d4f98a2SYan Zheng * get new location of data 16535d4f98a2SYan Zheng */ 16545d4f98a2SYan Zheng static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr, 16555d4f98a2SYan Zheng u64 bytenr, u64 num_bytes) 16565d4f98a2SYan Zheng { 16575d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(reloc_inode)->root; 16585d4f98a2SYan Zheng struct btrfs_path *path; 16595d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 16605d4f98a2SYan Zheng struct extent_buffer *leaf; 16615d4f98a2SYan Zheng int ret; 16625d4f98a2SYan Zheng 16635d4f98a2SYan Zheng path = btrfs_alloc_path(); 16645d4f98a2SYan Zheng if (!path) 16655d4f98a2SYan Zheng return -ENOMEM; 16665d4f98a2SYan Zheng 16675d4f98a2SYan Zheng bytenr -= BTRFS_I(reloc_inode)->index_cnt; 1668f85b7379SDavid Sterba ret = btrfs_lookup_file_extent(NULL, root, path, 1669f85b7379SDavid Sterba btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0); 16705d4f98a2SYan Zheng if (ret < 0) 16715d4f98a2SYan Zheng goto out; 16725d4f98a2SYan Zheng if (ret > 0) { 16735d4f98a2SYan Zheng ret = -ENOENT; 16745d4f98a2SYan Zheng goto out; 16755d4f98a2SYan Zheng } 16765d4f98a2SYan Zheng 16775d4f98a2SYan Zheng leaf = path->nodes[0]; 16785d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 16795d4f98a2SYan Zheng struct btrfs_file_extent_item); 16805d4f98a2SYan Zheng 16815d4f98a2SYan Zheng BUG_ON(btrfs_file_extent_offset(leaf, fi) || 16825d4f98a2SYan Zheng btrfs_file_extent_compression(leaf, fi) || 16835d4f98a2SYan Zheng btrfs_file_extent_encryption(leaf, fi) || 16845d4f98a2SYan Zheng btrfs_file_extent_other_encoding(leaf, fi)); 16855d4f98a2SYan Zheng 16865d4f98a2SYan Zheng if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) { 168783d4cfd4SJosef Bacik ret = -EINVAL; 16885d4f98a2SYan Zheng goto out; 16895d4f98a2SYan Zheng } 16905d4f98a2SYan Zheng 16915d4f98a2SYan Zheng *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 16925d4f98a2SYan Zheng ret = 0; 16935d4f98a2SYan Zheng out: 16945d4f98a2SYan Zheng btrfs_free_path(path); 16955d4f98a2SYan Zheng return ret; 16965d4f98a2SYan Zheng } 16975d4f98a2SYan Zheng 16985d4f98a2SYan Zheng /* 16995d4f98a2SYan Zheng * update file extent items in the tree leaf to point to 17005d4f98a2SYan Zheng * the new locations. 17015d4f98a2SYan Zheng */ 17023fd0a558SYan, Zheng static noinline_for_stack 17033fd0a558SYan, Zheng int replace_file_extents(struct btrfs_trans_handle *trans, 17045d4f98a2SYan Zheng struct reloc_control *rc, 17055d4f98a2SYan Zheng struct btrfs_root *root, 17063fd0a558SYan, Zheng struct extent_buffer *leaf) 17075d4f98a2SYan Zheng { 17080b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 17095d4f98a2SYan Zheng struct btrfs_key key; 17105d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 17115d4f98a2SYan Zheng struct inode *inode = NULL; 17125d4f98a2SYan Zheng u64 parent; 17135d4f98a2SYan Zheng u64 bytenr; 17143fd0a558SYan, Zheng u64 new_bytenr = 0; 17155d4f98a2SYan Zheng u64 num_bytes; 17165d4f98a2SYan Zheng u64 end; 17175d4f98a2SYan Zheng u32 nritems; 17185d4f98a2SYan Zheng u32 i; 171983d4cfd4SJosef Bacik int ret = 0; 17205d4f98a2SYan Zheng int first = 1; 17215d4f98a2SYan Zheng int dirty = 0; 17225d4f98a2SYan Zheng 17235d4f98a2SYan Zheng if (rc->stage != UPDATE_DATA_PTRS) 17245d4f98a2SYan Zheng return 0; 17255d4f98a2SYan Zheng 17265d4f98a2SYan Zheng /* reloc trees always use full backref */ 17275d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 17285d4f98a2SYan Zheng parent = leaf->start; 17295d4f98a2SYan Zheng else 17305d4f98a2SYan Zheng parent = 0; 17315d4f98a2SYan Zheng 17325d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 17335d4f98a2SYan Zheng for (i = 0; i < nritems; i++) { 173482fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 173582fa113fSQu Wenruo 17365d4f98a2SYan Zheng cond_resched(); 17375d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, i); 17385d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_DATA_KEY) 17395d4f98a2SYan Zheng continue; 17405d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 17415d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 17425d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 17435d4f98a2SYan Zheng continue; 17445d4f98a2SYan Zheng bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 17455d4f98a2SYan Zheng num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); 17465d4f98a2SYan Zheng if (bytenr == 0) 17475d4f98a2SYan Zheng continue; 17485d4f98a2SYan Zheng if (!in_block_group(bytenr, rc->block_group)) 17495d4f98a2SYan Zheng continue; 17505d4f98a2SYan Zheng 17515d4f98a2SYan Zheng /* 17525d4f98a2SYan Zheng * if we are modifying block in fs tree, wait for readpage 17535d4f98a2SYan Zheng * to complete and drop the extent cache 17545d4f98a2SYan Zheng */ 17555d4f98a2SYan Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 17565d4f98a2SYan Zheng if (first) { 17575d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 17585d4f98a2SYan Zheng first = 0; 17594a0cc7caSNikolay Borisov } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) { 17603fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 17615d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 17625d4f98a2SYan Zheng } 17634a0cc7caSNikolay Borisov if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) { 17645d4f98a2SYan Zheng end = key.offset + 17655d4f98a2SYan Zheng btrfs_file_extent_num_bytes(leaf, fi); 17665d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(key.offset, 17670b246afaSJeff Mahoney fs_info->sectorsize)); 17680b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 17695d4f98a2SYan Zheng end--; 17705d4f98a2SYan Zheng ret = try_lock_extent(&BTRFS_I(inode)->io_tree, 1771d0082371SJeff Mahoney key.offset, end); 17725d4f98a2SYan Zheng if (!ret) 17735d4f98a2SYan Zheng continue; 17745d4f98a2SYan Zheng 1775dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), 1776dcdbc059SNikolay Borisov key.offset, end, 1); 17775d4f98a2SYan Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 1778d0082371SJeff Mahoney key.offset, end); 17795d4f98a2SYan Zheng } 17805d4f98a2SYan Zheng } 17815d4f98a2SYan Zheng 17825d4f98a2SYan Zheng ret = get_new_location(rc->data_inode, &new_bytenr, 17835d4f98a2SYan Zheng bytenr, num_bytes); 178483d4cfd4SJosef Bacik if (ret) { 178583d4cfd4SJosef Bacik /* 178683d4cfd4SJosef Bacik * Don't have to abort since we've not changed anything 178783d4cfd4SJosef Bacik * in the file extent yet. 178883d4cfd4SJosef Bacik */ 178983d4cfd4SJosef Bacik break; 17903fd0a558SYan, Zheng } 17915d4f98a2SYan Zheng 17925d4f98a2SYan Zheng btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr); 17935d4f98a2SYan Zheng dirty = 1; 17945d4f98a2SYan Zheng 17955d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 179682fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 179782fa113fSQu Wenruo num_bytes, parent); 179882fa113fSQu Wenruo ref.real_root = root->root_key.objectid; 179982fa113fSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1800b06c4bf5SFilipe Manana key.objectid, key.offset); 180182fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 180283d4cfd4SJosef Bacik if (ret) { 180366642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 180483d4cfd4SJosef Bacik break; 180583d4cfd4SJosef Bacik } 18065d4f98a2SYan Zheng 1807ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, 1808ffd4bb2aSQu Wenruo num_bytes, parent); 1809ffd4bb2aSQu Wenruo ref.real_root = root->root_key.objectid; 1810ffd4bb2aSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1811b06c4bf5SFilipe Manana key.objectid, key.offset); 1812ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 181383d4cfd4SJosef Bacik if (ret) { 181466642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 181583d4cfd4SJosef Bacik break; 181683d4cfd4SJosef Bacik } 18175d4f98a2SYan Zheng } 18185d4f98a2SYan Zheng if (dirty) 18195d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 18203fd0a558SYan, Zheng if (inode) 18213fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 182283d4cfd4SJosef Bacik return ret; 18235d4f98a2SYan Zheng } 18245d4f98a2SYan Zheng 18255d4f98a2SYan Zheng static noinline_for_stack 18265d4f98a2SYan Zheng int memcmp_node_keys(struct extent_buffer *eb, int slot, 18275d4f98a2SYan Zheng struct btrfs_path *path, int level) 18285d4f98a2SYan Zheng { 18295d4f98a2SYan Zheng struct btrfs_disk_key key1; 18305d4f98a2SYan Zheng struct btrfs_disk_key key2; 18315d4f98a2SYan Zheng btrfs_node_key(eb, &key1, slot); 18325d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &key2, path->slots[level]); 18335d4f98a2SYan Zheng return memcmp(&key1, &key2, sizeof(key1)); 18345d4f98a2SYan Zheng } 18355d4f98a2SYan Zheng 18365d4f98a2SYan Zheng /* 18375d4f98a2SYan Zheng * try to replace tree blocks in fs tree with the new blocks 18385d4f98a2SYan Zheng * in reloc tree. tree blocks haven't been modified since the 18395d4f98a2SYan Zheng * reloc tree was create can be replaced. 18405d4f98a2SYan Zheng * 18415d4f98a2SYan Zheng * if a block was replaced, level of the block + 1 is returned. 18425d4f98a2SYan Zheng * if no block got replaced, 0 is returned. if there are other 18435d4f98a2SYan Zheng * errors, a negative error number is returned. 18445d4f98a2SYan Zheng */ 18453fd0a558SYan, Zheng static noinline_for_stack 18463d0174f7SQu Wenruo int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc, 18475d4f98a2SYan Zheng struct btrfs_root *dest, struct btrfs_root *src, 18485d4f98a2SYan Zheng struct btrfs_path *path, struct btrfs_key *next_key, 18495d4f98a2SYan Zheng int lowest_level, int max_level) 18505d4f98a2SYan Zheng { 18510b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = dest->fs_info; 18525d4f98a2SYan Zheng struct extent_buffer *eb; 18535d4f98a2SYan Zheng struct extent_buffer *parent; 185482fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 18555d4f98a2SYan Zheng struct btrfs_key key; 18565d4f98a2SYan Zheng u64 old_bytenr; 18575d4f98a2SYan Zheng u64 new_bytenr; 18585d4f98a2SYan Zheng u64 old_ptr_gen; 18595d4f98a2SYan Zheng u64 new_ptr_gen; 18605d4f98a2SYan Zheng u64 last_snapshot; 18615d4f98a2SYan Zheng u32 blocksize; 18623fd0a558SYan, Zheng int cow = 0; 18635d4f98a2SYan Zheng int level; 18645d4f98a2SYan Zheng int ret; 18655d4f98a2SYan Zheng int slot; 18665d4f98a2SYan Zheng 18675d4f98a2SYan Zheng BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 18685d4f98a2SYan Zheng BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID); 18695d4f98a2SYan Zheng 18705d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&src->root_item); 18713fd0a558SYan, Zheng again: 18725d4f98a2SYan Zheng slot = path->slots[lowest_level]; 18735d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot); 18745d4f98a2SYan Zheng 18755d4f98a2SYan Zheng eb = btrfs_lock_root_node(dest); 18768bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 18775d4f98a2SYan Zheng level = btrfs_header_level(eb); 18785d4f98a2SYan Zheng 18795d4f98a2SYan Zheng if (level < lowest_level) { 18805d4f98a2SYan Zheng btrfs_tree_unlock(eb); 18815d4f98a2SYan Zheng free_extent_buffer(eb); 18825d4f98a2SYan Zheng return 0; 18835d4f98a2SYan Zheng } 18845d4f98a2SYan Zheng 18853fd0a558SYan, Zheng if (cow) { 18865d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb); 18875d4f98a2SYan Zheng BUG_ON(ret); 18883fd0a558SYan, Zheng } 18898bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 18905d4f98a2SYan Zheng 18915d4f98a2SYan Zheng if (next_key) { 18925d4f98a2SYan Zheng next_key->objectid = (u64)-1; 18935d4f98a2SYan Zheng next_key->type = (u8)-1; 18945d4f98a2SYan Zheng next_key->offset = (u64)-1; 18955d4f98a2SYan Zheng } 18965d4f98a2SYan Zheng 18975d4f98a2SYan Zheng parent = eb; 18985d4f98a2SYan Zheng while (1) { 1899581c1760SQu Wenruo struct btrfs_key first_key; 1900581c1760SQu Wenruo 19015d4f98a2SYan Zheng level = btrfs_header_level(parent); 19025d4f98a2SYan Zheng BUG_ON(level < lowest_level); 19035d4f98a2SYan Zheng 19045d4f98a2SYan Zheng ret = btrfs_bin_search(parent, &key, level, &slot); 1905cbca7d59SFilipe Manana if (ret < 0) 1906cbca7d59SFilipe Manana break; 19075d4f98a2SYan Zheng if (ret && slot > 0) 19085d4f98a2SYan Zheng slot--; 19095d4f98a2SYan Zheng 19105d4f98a2SYan Zheng if (next_key && slot + 1 < btrfs_header_nritems(parent)) 19115d4f98a2SYan Zheng btrfs_node_key_to_cpu(parent, next_key, slot + 1); 19125d4f98a2SYan Zheng 19135d4f98a2SYan Zheng old_bytenr = btrfs_node_blockptr(parent, slot); 19140b246afaSJeff Mahoney blocksize = fs_info->nodesize; 19155d4f98a2SYan Zheng old_ptr_gen = btrfs_node_ptr_generation(parent, slot); 191617515f1bSQu Wenruo btrfs_node_key_to_cpu(parent, &first_key, slot); 19175d4f98a2SYan Zheng 19185d4f98a2SYan Zheng if (level <= max_level) { 19195d4f98a2SYan Zheng eb = path->nodes[level]; 19205d4f98a2SYan Zheng new_bytenr = btrfs_node_blockptr(eb, 19215d4f98a2SYan Zheng path->slots[level]); 19225d4f98a2SYan Zheng new_ptr_gen = btrfs_node_ptr_generation(eb, 19235d4f98a2SYan Zheng path->slots[level]); 19245d4f98a2SYan Zheng } else { 19255d4f98a2SYan Zheng new_bytenr = 0; 19265d4f98a2SYan Zheng new_ptr_gen = 0; 19275d4f98a2SYan Zheng } 19285d4f98a2SYan Zheng 1929fae7f21cSDulshani Gunawardhana if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) { 19305d4f98a2SYan Zheng ret = level; 19315d4f98a2SYan Zheng break; 19325d4f98a2SYan Zheng } 19335d4f98a2SYan Zheng 19345d4f98a2SYan Zheng if (new_bytenr == 0 || old_ptr_gen > last_snapshot || 19355d4f98a2SYan Zheng memcmp_node_keys(parent, slot, path, level)) { 19363fd0a558SYan, Zheng if (level <= lowest_level) { 19375d4f98a2SYan Zheng ret = 0; 19385d4f98a2SYan Zheng break; 19395d4f98a2SYan Zheng } 19405d4f98a2SYan Zheng 1941581c1760SQu Wenruo eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen, 1942581c1760SQu Wenruo level - 1, &first_key); 194364c043deSLiu Bo if (IS_ERR(eb)) { 194464c043deSLiu Bo ret = PTR_ERR(eb); 1945264813acSLiu Bo break; 194664c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 194764c043deSLiu Bo ret = -EIO; 1948416bc658SJosef Bacik free_extent_buffer(eb); 1949379cde74SStefan Behrens break; 1950416bc658SJosef Bacik } 19515d4f98a2SYan Zheng btrfs_tree_lock(eb); 19523fd0a558SYan, Zheng if (cow) { 19535d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, parent, 19545d4f98a2SYan Zheng slot, &eb); 19555d4f98a2SYan Zheng BUG_ON(ret); 19565d4f98a2SYan Zheng } 19578bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 19585d4f98a2SYan Zheng 19595d4f98a2SYan Zheng btrfs_tree_unlock(parent); 19605d4f98a2SYan Zheng free_extent_buffer(parent); 19615d4f98a2SYan Zheng 19625d4f98a2SYan Zheng parent = eb; 19635d4f98a2SYan Zheng continue; 19645d4f98a2SYan Zheng } 19655d4f98a2SYan Zheng 19663fd0a558SYan, Zheng if (!cow) { 19673fd0a558SYan, Zheng btrfs_tree_unlock(parent); 19683fd0a558SYan, Zheng free_extent_buffer(parent); 19693fd0a558SYan, Zheng cow = 1; 19703fd0a558SYan, Zheng goto again; 19713fd0a558SYan, Zheng } 19723fd0a558SYan, Zheng 19735d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 19745d4f98a2SYan Zheng path->slots[level]); 1975b3b4aa74SDavid Sterba btrfs_release_path(path); 19765d4f98a2SYan Zheng 19775d4f98a2SYan Zheng path->lowest_level = level; 19785d4f98a2SYan Zheng ret = btrfs_search_slot(trans, src, &key, path, 0, 1); 19795d4f98a2SYan Zheng path->lowest_level = 0; 19805d4f98a2SYan Zheng BUG_ON(ret); 19815d4f98a2SYan Zheng 19825d4f98a2SYan Zheng /* 1983824d8dffSQu Wenruo * Info qgroup to trace both subtrees. 1984824d8dffSQu Wenruo * 1985824d8dffSQu Wenruo * We must trace both trees. 1986824d8dffSQu Wenruo * 1) Tree reloc subtree 1987824d8dffSQu Wenruo * If not traced, we will leak data numbers 1988824d8dffSQu Wenruo * 2) Fs subtree 1989824d8dffSQu Wenruo * If not traced, we will double count old data 1990f616f5cdSQu Wenruo * 1991f616f5cdSQu Wenruo * We don't scan the subtree right now, but only record 1992f616f5cdSQu Wenruo * the swapped tree blocks. 1993f616f5cdSQu Wenruo * The real subtree rescan is delayed until we have new 1994f616f5cdSQu Wenruo * CoW on the subtree root node before transaction commit. 1995824d8dffSQu Wenruo */ 1996370a11b8SQu Wenruo ret = btrfs_qgroup_add_swapped_blocks(trans, dest, 1997370a11b8SQu Wenruo rc->block_group, parent, slot, 1998370a11b8SQu Wenruo path->nodes[level], path->slots[level], 1999370a11b8SQu Wenruo last_snapshot); 2000370a11b8SQu Wenruo if (ret < 0) 2001370a11b8SQu Wenruo break; 2002824d8dffSQu Wenruo /* 20035d4f98a2SYan Zheng * swap blocks in fs tree and reloc tree. 20045d4f98a2SYan Zheng */ 20055d4f98a2SYan Zheng btrfs_set_node_blockptr(parent, slot, new_bytenr); 20065d4f98a2SYan Zheng btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen); 20075d4f98a2SYan Zheng btrfs_mark_buffer_dirty(parent); 20085d4f98a2SYan Zheng 20095d4f98a2SYan Zheng btrfs_set_node_blockptr(path->nodes[level], 20105d4f98a2SYan Zheng path->slots[level], old_bytenr); 20115d4f98a2SYan Zheng btrfs_set_node_ptr_generation(path->nodes[level], 20125d4f98a2SYan Zheng path->slots[level], old_ptr_gen); 20135d4f98a2SYan Zheng btrfs_mark_buffer_dirty(path->nodes[level]); 20145d4f98a2SYan Zheng 201582fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr, 201682fa113fSQu Wenruo blocksize, path->nodes[level]->start); 201782fa113fSQu Wenruo ref.skip_qgroup = true; 201882fa113fSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid); 201982fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 20205d4f98a2SYan Zheng BUG_ON(ret); 202182fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 202282fa113fSQu Wenruo blocksize, 0); 202382fa113fSQu Wenruo ref.skip_qgroup = true; 202482fa113fSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid); 202582fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 20265d4f98a2SYan Zheng BUG_ON(ret); 20275d4f98a2SYan Zheng 2028ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr, 2029ffd4bb2aSQu Wenruo blocksize, path->nodes[level]->start); 2030ffd4bb2aSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid); 2031ffd4bb2aSQu Wenruo ref.skip_qgroup = true; 2032ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 20335d4f98a2SYan Zheng BUG_ON(ret); 20345d4f98a2SYan Zheng 2035ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr, 2036ffd4bb2aSQu Wenruo blocksize, 0); 2037ffd4bb2aSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid); 2038ffd4bb2aSQu Wenruo ref.skip_qgroup = true; 2039ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 20405d4f98a2SYan Zheng BUG_ON(ret); 20415d4f98a2SYan Zheng 20425d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 20435d4f98a2SYan Zheng 20445d4f98a2SYan Zheng ret = level; 20455d4f98a2SYan Zheng break; 20465d4f98a2SYan Zheng } 20475d4f98a2SYan Zheng btrfs_tree_unlock(parent); 20485d4f98a2SYan Zheng free_extent_buffer(parent); 20495d4f98a2SYan Zheng return ret; 20505d4f98a2SYan Zheng } 20515d4f98a2SYan Zheng 20525d4f98a2SYan Zheng /* 20535d4f98a2SYan Zheng * helper to find next relocated block in reloc tree 20545d4f98a2SYan Zheng */ 20555d4f98a2SYan Zheng static noinline_for_stack 20565d4f98a2SYan Zheng int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 20575d4f98a2SYan Zheng int *level) 20585d4f98a2SYan Zheng { 20595d4f98a2SYan Zheng struct extent_buffer *eb; 20605d4f98a2SYan Zheng int i; 20615d4f98a2SYan Zheng u64 last_snapshot; 20625d4f98a2SYan Zheng u32 nritems; 20635d4f98a2SYan Zheng 20645d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 20655d4f98a2SYan Zheng 20665d4f98a2SYan Zheng for (i = 0; i < *level; i++) { 20675d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 20685d4f98a2SYan Zheng path->nodes[i] = NULL; 20695d4f98a2SYan Zheng } 20705d4f98a2SYan Zheng 20715d4f98a2SYan Zheng for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) { 20725d4f98a2SYan Zheng eb = path->nodes[i]; 20735d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 20745d4f98a2SYan Zheng while (path->slots[i] + 1 < nritems) { 20755d4f98a2SYan Zheng path->slots[i]++; 20765d4f98a2SYan Zheng if (btrfs_node_ptr_generation(eb, path->slots[i]) <= 20775d4f98a2SYan Zheng last_snapshot) 20785d4f98a2SYan Zheng continue; 20795d4f98a2SYan Zheng 20805d4f98a2SYan Zheng *level = i; 20815d4f98a2SYan Zheng return 0; 20825d4f98a2SYan Zheng } 20835d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 20845d4f98a2SYan Zheng path->nodes[i] = NULL; 20855d4f98a2SYan Zheng } 20865d4f98a2SYan Zheng return 1; 20875d4f98a2SYan Zheng } 20885d4f98a2SYan Zheng 20895d4f98a2SYan Zheng /* 20905d4f98a2SYan Zheng * walk down reloc tree to find relocated block of lowest level 20915d4f98a2SYan Zheng */ 20925d4f98a2SYan Zheng static noinline_for_stack 20935d4f98a2SYan Zheng int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 20945d4f98a2SYan Zheng int *level) 20955d4f98a2SYan Zheng { 20962ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 20975d4f98a2SYan Zheng struct extent_buffer *eb = NULL; 20985d4f98a2SYan Zheng int i; 20995d4f98a2SYan Zheng u64 bytenr; 21005d4f98a2SYan Zheng u64 ptr_gen = 0; 21015d4f98a2SYan Zheng u64 last_snapshot; 21025d4f98a2SYan Zheng u32 nritems; 21035d4f98a2SYan Zheng 21045d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 21055d4f98a2SYan Zheng 21065d4f98a2SYan Zheng for (i = *level; i > 0; i--) { 2107581c1760SQu Wenruo struct btrfs_key first_key; 2108581c1760SQu Wenruo 21095d4f98a2SYan Zheng eb = path->nodes[i]; 21105d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 21115d4f98a2SYan Zheng while (path->slots[i] < nritems) { 21125d4f98a2SYan Zheng ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]); 21135d4f98a2SYan Zheng if (ptr_gen > last_snapshot) 21145d4f98a2SYan Zheng break; 21155d4f98a2SYan Zheng path->slots[i]++; 21165d4f98a2SYan Zheng } 21175d4f98a2SYan Zheng if (path->slots[i] >= nritems) { 21185d4f98a2SYan Zheng if (i == *level) 21195d4f98a2SYan Zheng break; 21205d4f98a2SYan Zheng *level = i + 1; 21215d4f98a2SYan Zheng return 0; 21225d4f98a2SYan Zheng } 21235d4f98a2SYan Zheng if (i == 1) { 21245d4f98a2SYan Zheng *level = i; 21255d4f98a2SYan Zheng return 0; 21265d4f98a2SYan Zheng } 21275d4f98a2SYan Zheng 21285d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(eb, path->slots[i]); 2129581c1760SQu Wenruo btrfs_node_key_to_cpu(eb, &first_key, path->slots[i]); 2130581c1760SQu Wenruo eb = read_tree_block(fs_info, bytenr, ptr_gen, i - 1, 2131581c1760SQu Wenruo &first_key); 213264c043deSLiu Bo if (IS_ERR(eb)) { 213364c043deSLiu Bo return PTR_ERR(eb); 213464c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2135416bc658SJosef Bacik free_extent_buffer(eb); 2136416bc658SJosef Bacik return -EIO; 2137416bc658SJosef Bacik } 21385d4f98a2SYan Zheng BUG_ON(btrfs_header_level(eb) != i - 1); 21395d4f98a2SYan Zheng path->nodes[i - 1] = eb; 21405d4f98a2SYan Zheng path->slots[i - 1] = 0; 21415d4f98a2SYan Zheng } 21425d4f98a2SYan Zheng return 1; 21435d4f98a2SYan Zheng } 21445d4f98a2SYan Zheng 21455d4f98a2SYan Zheng /* 21465d4f98a2SYan Zheng * invalidate extent cache for file extents whose key in range of 21475d4f98a2SYan Zheng * [min_key, max_key) 21485d4f98a2SYan Zheng */ 21495d4f98a2SYan Zheng static int invalidate_extent_cache(struct btrfs_root *root, 21505d4f98a2SYan Zheng struct btrfs_key *min_key, 21515d4f98a2SYan Zheng struct btrfs_key *max_key) 21525d4f98a2SYan Zheng { 21530b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 21545d4f98a2SYan Zheng struct inode *inode = NULL; 21555d4f98a2SYan Zheng u64 objectid; 21565d4f98a2SYan Zheng u64 start, end; 215733345d01SLi Zefan u64 ino; 21585d4f98a2SYan Zheng 21595d4f98a2SYan Zheng objectid = min_key->objectid; 21605d4f98a2SYan Zheng while (1) { 21615d4f98a2SYan Zheng cond_resched(); 21625d4f98a2SYan Zheng iput(inode); 21635d4f98a2SYan Zheng 21645d4f98a2SYan Zheng if (objectid > max_key->objectid) 21655d4f98a2SYan Zheng break; 21665d4f98a2SYan Zheng 21675d4f98a2SYan Zheng inode = find_next_inode(root, objectid); 21685d4f98a2SYan Zheng if (!inode) 21695d4f98a2SYan Zheng break; 21704a0cc7caSNikolay Borisov ino = btrfs_ino(BTRFS_I(inode)); 21715d4f98a2SYan Zheng 217233345d01SLi Zefan if (ino > max_key->objectid) { 21735d4f98a2SYan Zheng iput(inode); 21745d4f98a2SYan Zheng break; 21755d4f98a2SYan Zheng } 21765d4f98a2SYan Zheng 217733345d01SLi Zefan objectid = ino + 1; 21785d4f98a2SYan Zheng if (!S_ISREG(inode->i_mode)) 21795d4f98a2SYan Zheng continue; 21805d4f98a2SYan Zheng 218133345d01SLi Zefan if (unlikely(min_key->objectid == ino)) { 21825d4f98a2SYan Zheng if (min_key->type > BTRFS_EXTENT_DATA_KEY) 21835d4f98a2SYan Zheng continue; 21845d4f98a2SYan Zheng if (min_key->type < BTRFS_EXTENT_DATA_KEY) 21855d4f98a2SYan Zheng start = 0; 21865d4f98a2SYan Zheng else { 21875d4f98a2SYan Zheng start = min_key->offset; 21880b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize)); 21895d4f98a2SYan Zheng } 21905d4f98a2SYan Zheng } else { 21915d4f98a2SYan Zheng start = 0; 21925d4f98a2SYan Zheng } 21935d4f98a2SYan Zheng 219433345d01SLi Zefan if (unlikely(max_key->objectid == ino)) { 21955d4f98a2SYan Zheng if (max_key->type < BTRFS_EXTENT_DATA_KEY) 21965d4f98a2SYan Zheng continue; 21975d4f98a2SYan Zheng if (max_key->type > BTRFS_EXTENT_DATA_KEY) { 21985d4f98a2SYan Zheng end = (u64)-1; 21995d4f98a2SYan Zheng } else { 22005d4f98a2SYan Zheng if (max_key->offset == 0) 22015d4f98a2SYan Zheng continue; 22025d4f98a2SYan Zheng end = max_key->offset; 22030b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 22045d4f98a2SYan Zheng end--; 22055d4f98a2SYan Zheng } 22065d4f98a2SYan Zheng } else { 22075d4f98a2SYan Zheng end = (u64)-1; 22085d4f98a2SYan Zheng } 22095d4f98a2SYan Zheng 22105d4f98a2SYan Zheng /* the lock_extent waits for readpage to complete */ 2211d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 2212dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1); 2213d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 22145d4f98a2SYan Zheng } 22155d4f98a2SYan Zheng return 0; 22165d4f98a2SYan Zheng } 22175d4f98a2SYan Zheng 22185d4f98a2SYan Zheng static int find_next_key(struct btrfs_path *path, int level, 22195d4f98a2SYan Zheng struct btrfs_key *key) 22205d4f98a2SYan Zheng 22215d4f98a2SYan Zheng { 22225d4f98a2SYan Zheng while (level < BTRFS_MAX_LEVEL) { 22235d4f98a2SYan Zheng if (!path->nodes[level]) 22245d4f98a2SYan Zheng break; 22255d4f98a2SYan Zheng if (path->slots[level] + 1 < 22265d4f98a2SYan Zheng btrfs_header_nritems(path->nodes[level])) { 22275d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], key, 22285d4f98a2SYan Zheng path->slots[level] + 1); 22295d4f98a2SYan Zheng return 0; 22305d4f98a2SYan Zheng } 22315d4f98a2SYan Zheng level++; 22325d4f98a2SYan Zheng } 22335d4f98a2SYan Zheng return 1; 22345d4f98a2SYan Zheng } 22355d4f98a2SYan Zheng 22365d4f98a2SYan Zheng /* 2237d2311e69SQu Wenruo * Insert current subvolume into reloc_control::dirty_subvol_roots 2238d2311e69SQu Wenruo */ 2239d2311e69SQu Wenruo static void insert_dirty_subvol(struct btrfs_trans_handle *trans, 2240d2311e69SQu Wenruo struct reloc_control *rc, 2241d2311e69SQu Wenruo struct btrfs_root *root) 2242d2311e69SQu Wenruo { 2243d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 2244d2311e69SQu Wenruo struct btrfs_root_item *reloc_root_item; 2245d2311e69SQu Wenruo 2246d2311e69SQu Wenruo /* @root must be a subvolume tree root with a valid reloc tree */ 2247d2311e69SQu Wenruo ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 2248d2311e69SQu Wenruo ASSERT(reloc_root); 2249d2311e69SQu Wenruo 2250d2311e69SQu Wenruo reloc_root_item = &reloc_root->root_item; 2251d2311e69SQu Wenruo memset(&reloc_root_item->drop_progress, 0, 2252d2311e69SQu Wenruo sizeof(reloc_root_item->drop_progress)); 2253d2311e69SQu Wenruo reloc_root_item->drop_level = 0; 2254d2311e69SQu Wenruo btrfs_set_root_refs(reloc_root_item, 0); 2255d2311e69SQu Wenruo btrfs_update_reloc_root(trans, root); 2256d2311e69SQu Wenruo 2257d2311e69SQu Wenruo if (list_empty(&root->reloc_dirty_list)) { 225800246528SJosef Bacik btrfs_grab_root(root); 2259d2311e69SQu Wenruo list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots); 2260d2311e69SQu Wenruo } 2261d2311e69SQu Wenruo } 2262d2311e69SQu Wenruo 2263d2311e69SQu Wenruo static int clean_dirty_subvols(struct reloc_control *rc) 2264d2311e69SQu Wenruo { 2265d2311e69SQu Wenruo struct btrfs_root *root; 2266d2311e69SQu Wenruo struct btrfs_root *next; 2267d2311e69SQu Wenruo int ret = 0; 226830d40577SQu Wenruo int ret2; 2269d2311e69SQu Wenruo 2270d2311e69SQu Wenruo list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots, 2271d2311e69SQu Wenruo reloc_dirty_list) { 227230d40577SQu Wenruo if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 227330d40577SQu Wenruo /* Merged subvolume, cleanup its reloc root */ 2274d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 2275d2311e69SQu Wenruo 2276d2311e69SQu Wenruo list_del_init(&root->reloc_dirty_list); 2277d2311e69SQu Wenruo root->reloc_root = NULL; 2278d2311e69SQu Wenruo if (reloc_root) { 2279d2311e69SQu Wenruo 22800078a9f9SNikolay Borisov ret2 = btrfs_drop_snapshot(reloc_root, 0, 1); 2281d2311e69SQu Wenruo if (ret2 < 0 && !ret) 2282d2311e69SQu Wenruo ret = ret2; 2283d2311e69SQu Wenruo } 22846282675eSQu Wenruo /* 22856282675eSQu Wenruo * Need barrier to ensure clear_bit() only happens after 22866282675eSQu Wenruo * root->reloc_root = NULL. Pairs with have_reloc_root. 22876282675eSQu Wenruo */ 22886282675eSQu Wenruo smp_wmb(); 22891fac4a54SQu Wenruo clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 229000246528SJosef Bacik btrfs_put_root(root); 229130d40577SQu Wenruo } else { 229230d40577SQu Wenruo /* Orphan reloc tree, just clean it up */ 22930078a9f9SNikolay Borisov ret2 = btrfs_drop_snapshot(root, 0, 1); 229430d40577SQu Wenruo if (ret2 < 0 && !ret) 229530d40577SQu Wenruo ret = ret2; 229630d40577SQu Wenruo } 2297d2311e69SQu Wenruo } 2298d2311e69SQu Wenruo return ret; 2299d2311e69SQu Wenruo } 2300d2311e69SQu Wenruo 2301d2311e69SQu Wenruo /* 23025d4f98a2SYan Zheng * merge the relocated tree blocks in reloc tree with corresponding 23035d4f98a2SYan Zheng * fs tree. 23045d4f98a2SYan Zheng */ 23055d4f98a2SYan Zheng static noinline_for_stack int merge_reloc_root(struct reloc_control *rc, 23065d4f98a2SYan Zheng struct btrfs_root *root) 23075d4f98a2SYan Zheng { 23080b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 23095d4f98a2SYan Zheng struct btrfs_key key; 23105d4f98a2SYan Zheng struct btrfs_key next_key; 23119e6a0c52SJosef Bacik struct btrfs_trans_handle *trans = NULL; 23125d4f98a2SYan Zheng struct btrfs_root *reloc_root; 23135d4f98a2SYan Zheng struct btrfs_root_item *root_item; 23145d4f98a2SYan Zheng struct btrfs_path *path; 23153fd0a558SYan, Zheng struct extent_buffer *leaf; 23165d4f98a2SYan Zheng int level; 23175d4f98a2SYan Zheng int max_level; 23185d4f98a2SYan Zheng int replaced = 0; 23195d4f98a2SYan Zheng int ret; 23205d4f98a2SYan Zheng int err = 0; 23213fd0a558SYan, Zheng u32 min_reserved; 23225d4f98a2SYan Zheng 23235d4f98a2SYan Zheng path = btrfs_alloc_path(); 23245d4f98a2SYan Zheng if (!path) 23255d4f98a2SYan Zheng return -ENOMEM; 2326e4058b54SDavid Sterba path->reada = READA_FORWARD; 23275d4f98a2SYan Zheng 23285d4f98a2SYan Zheng reloc_root = root->reloc_root; 23295d4f98a2SYan Zheng root_item = &reloc_root->root_item; 23305d4f98a2SYan Zheng 23315d4f98a2SYan Zheng if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { 23325d4f98a2SYan Zheng level = btrfs_root_level(root_item); 233367439dadSDavid Sterba atomic_inc(&reloc_root->node->refs); 23345d4f98a2SYan Zheng path->nodes[level] = reloc_root->node; 23355d4f98a2SYan Zheng path->slots[level] = 0; 23365d4f98a2SYan Zheng } else { 23375d4f98a2SYan Zheng btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); 23385d4f98a2SYan Zheng 23395d4f98a2SYan Zheng level = root_item->drop_level; 23405d4f98a2SYan Zheng BUG_ON(level == 0); 23415d4f98a2SYan Zheng path->lowest_level = level; 23425d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0); 234333c66f43SYan Zheng path->lowest_level = 0; 23445d4f98a2SYan Zheng if (ret < 0) { 23455d4f98a2SYan Zheng btrfs_free_path(path); 23465d4f98a2SYan Zheng return ret; 23475d4f98a2SYan Zheng } 23485d4f98a2SYan Zheng 23495d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &next_key, 23505d4f98a2SYan Zheng path->slots[level]); 23515d4f98a2SYan Zheng WARN_ON(memcmp(&key, &next_key, sizeof(key))); 23525d4f98a2SYan Zheng 23535d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 23545d4f98a2SYan Zheng } 23555d4f98a2SYan Zheng 23560b246afaSJeff Mahoney min_reserved = fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 23575d4f98a2SYan Zheng memset(&next_key, 0, sizeof(next_key)); 23585d4f98a2SYan Zheng 23595d4f98a2SYan Zheng while (1) { 236008e007d2SMiao Xie ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved, 236108e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 23623fd0a558SYan, Zheng if (ret) { 23639e6a0c52SJosef Bacik err = ret; 23649e6a0c52SJosef Bacik goto out; 23653fd0a558SYan, Zheng } 23669e6a0c52SJosef Bacik trans = btrfs_start_transaction(root, 0); 23679e6a0c52SJosef Bacik if (IS_ERR(trans)) { 23689e6a0c52SJosef Bacik err = PTR_ERR(trans); 23699e6a0c52SJosef Bacik trans = NULL; 23709e6a0c52SJosef Bacik goto out; 23719e6a0c52SJosef Bacik } 23722abc726aSJosef Bacik 23732abc726aSJosef Bacik /* 23742abc726aSJosef Bacik * At this point we no longer have a reloc_control, so we can't 23752abc726aSJosef Bacik * depend on btrfs_init_reloc_root to update our last_trans. 23762abc726aSJosef Bacik * 23772abc726aSJosef Bacik * But that's ok, we started the trans handle on our 23782abc726aSJosef Bacik * corresponding fs_root, which means it's been added to the 23792abc726aSJosef Bacik * dirty list. At commit time we'll still call 23802abc726aSJosef Bacik * btrfs_update_reloc_root() and update our root item 23812abc726aSJosef Bacik * appropriately. 23822abc726aSJosef Bacik */ 23832abc726aSJosef Bacik reloc_root->last_trans = trans->transid; 23849e6a0c52SJosef Bacik trans->block_rsv = rc->block_rsv; 23853fd0a558SYan, Zheng 23863fd0a558SYan, Zheng replaced = 0; 23875d4f98a2SYan Zheng max_level = level; 23885d4f98a2SYan Zheng 23895d4f98a2SYan Zheng ret = walk_down_reloc_tree(reloc_root, path, &level); 23905d4f98a2SYan Zheng if (ret < 0) { 23915d4f98a2SYan Zheng err = ret; 23925d4f98a2SYan Zheng goto out; 23935d4f98a2SYan Zheng } 23945d4f98a2SYan Zheng if (ret > 0) 23955d4f98a2SYan Zheng break; 23965d4f98a2SYan Zheng 23975d4f98a2SYan Zheng if (!find_next_key(path, level, &key) && 23985d4f98a2SYan Zheng btrfs_comp_cpu_keys(&next_key, &key) >= 0) { 23995d4f98a2SYan Zheng ret = 0; 24005d4f98a2SYan Zheng } else { 24013d0174f7SQu Wenruo ret = replace_path(trans, rc, root, reloc_root, path, 24023fd0a558SYan, Zheng &next_key, level, max_level); 24035d4f98a2SYan Zheng } 24045d4f98a2SYan Zheng if (ret < 0) { 24055d4f98a2SYan Zheng err = ret; 24065d4f98a2SYan Zheng goto out; 24075d4f98a2SYan Zheng } 24085d4f98a2SYan Zheng 24095d4f98a2SYan Zheng if (ret > 0) { 24105d4f98a2SYan Zheng level = ret; 24115d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 24125d4f98a2SYan Zheng path->slots[level]); 24135d4f98a2SYan Zheng replaced = 1; 24145d4f98a2SYan Zheng } 24155d4f98a2SYan Zheng 24165d4f98a2SYan Zheng ret = walk_up_reloc_tree(reloc_root, path, &level); 24175d4f98a2SYan Zheng if (ret > 0) 24185d4f98a2SYan Zheng break; 24195d4f98a2SYan Zheng 24205d4f98a2SYan Zheng BUG_ON(level == 0); 24215d4f98a2SYan Zheng /* 24225d4f98a2SYan Zheng * save the merging progress in the drop_progress. 24235d4f98a2SYan Zheng * this is OK since root refs == 1 in this case. 24245d4f98a2SYan Zheng */ 24255d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &root_item->drop_progress, 24265d4f98a2SYan Zheng path->slots[level]); 24275d4f98a2SYan Zheng root_item->drop_level = level; 24285d4f98a2SYan Zheng 24293a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 24309e6a0c52SJosef Bacik trans = NULL; 24315d4f98a2SYan Zheng 24322ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 24335d4f98a2SYan Zheng 24345d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 24355d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 24365d4f98a2SYan Zheng } 24375d4f98a2SYan Zheng 24385d4f98a2SYan Zheng /* 24395d4f98a2SYan Zheng * handle the case only one block in the fs tree need to be 24405d4f98a2SYan Zheng * relocated and the block is tree root. 24415d4f98a2SYan Zheng */ 24425d4f98a2SYan Zheng leaf = btrfs_lock_root_node(root); 24435d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf); 24445d4f98a2SYan Zheng btrfs_tree_unlock(leaf); 24455d4f98a2SYan Zheng free_extent_buffer(leaf); 24465d4f98a2SYan Zheng if (ret < 0) 24475d4f98a2SYan Zheng err = ret; 24485d4f98a2SYan Zheng out: 24495d4f98a2SYan Zheng btrfs_free_path(path); 24505d4f98a2SYan Zheng 2451d2311e69SQu Wenruo if (err == 0) 2452d2311e69SQu Wenruo insert_dirty_subvol(trans, rc, root); 24535d4f98a2SYan Zheng 24549e6a0c52SJosef Bacik if (trans) 24553a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 24565d4f98a2SYan Zheng 24572ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 24585d4f98a2SYan Zheng 24595d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 24605d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 24615d4f98a2SYan Zheng 24625d4f98a2SYan Zheng return err; 24635d4f98a2SYan Zheng } 24645d4f98a2SYan Zheng 24653fd0a558SYan, Zheng static noinline_for_stack 24663fd0a558SYan, Zheng int prepare_to_merge(struct reloc_control *rc, int err) 24675d4f98a2SYan Zheng { 24683fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 24690b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 24703fd0a558SYan, Zheng struct btrfs_root *reloc_root; 24715d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 24723fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 24733fd0a558SYan, Zheng u64 num_bytes = 0; 24743fd0a558SYan, Zheng int ret; 24753fd0a558SYan, Zheng 24760b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 24770b246afaSJeff Mahoney rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 24783fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated * 2; 24790b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 24807585717fSChris Mason 24813fd0a558SYan, Zheng again: 24823fd0a558SYan, Zheng if (!err) { 24833fd0a558SYan, Zheng num_bytes = rc->merging_rsv_size; 248408e007d2SMiao Xie ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes, 248508e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 24863fd0a558SYan, Zheng if (ret) 24873fd0a558SYan, Zheng err = ret; 24883fd0a558SYan, Zheng } 24893fd0a558SYan, Zheng 24907a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 24913612b495STsutomu Itoh if (IS_ERR(trans)) { 24923612b495STsutomu Itoh if (!err) 24932ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 249463f018beSNikolay Borisov num_bytes, NULL); 24953612b495STsutomu Itoh return PTR_ERR(trans); 24963612b495STsutomu Itoh } 24973fd0a558SYan, Zheng 24983fd0a558SYan, Zheng if (!err) { 24993fd0a558SYan, Zheng if (num_bytes != rc->merging_rsv_size) { 25003a45bb20SJeff Mahoney btrfs_end_transaction(trans); 25012ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 250263f018beSNikolay Borisov num_bytes, NULL); 25033fd0a558SYan, Zheng goto again; 25043fd0a558SYan, Zheng } 25053fd0a558SYan, Zheng } 25063fd0a558SYan, Zheng 25073fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 25083fd0a558SYan, Zheng 25093fd0a558SYan, Zheng while (!list_empty(&rc->reloc_roots)) { 25103fd0a558SYan, Zheng reloc_root = list_entry(rc->reloc_roots.next, 25113fd0a558SYan, Zheng struct btrfs_root, root_list); 25123fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 25133fd0a558SYan, Zheng 25140b246afaSJeff Mahoney root = read_fs_root(fs_info, reloc_root->root_key.offset); 25153fd0a558SYan, Zheng BUG_ON(IS_ERR(root)); 25163fd0a558SYan, Zheng BUG_ON(root->reloc_root != reloc_root); 25173fd0a558SYan, Zheng 25183fd0a558SYan, Zheng /* 25193fd0a558SYan, Zheng * set reference count to 1, so btrfs_recover_relocation 25203fd0a558SYan, Zheng * knows it should resumes merging 25213fd0a558SYan, Zheng */ 25223fd0a558SYan, Zheng if (!err) 25233fd0a558SYan, Zheng btrfs_set_root_refs(&reloc_root->root_item, 1); 25243fd0a558SYan, Zheng btrfs_update_reloc_root(trans, root); 25253fd0a558SYan, Zheng 25263fd0a558SYan, Zheng list_add(&reloc_root->root_list, &reloc_roots); 252700246528SJosef Bacik btrfs_put_root(root); 25283fd0a558SYan, Zheng } 25293fd0a558SYan, Zheng 25303fd0a558SYan, Zheng list_splice(&reloc_roots, &rc->reloc_roots); 25313fd0a558SYan, Zheng 25323fd0a558SYan, Zheng if (!err) 25333a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 25343fd0a558SYan, Zheng else 25353a45bb20SJeff Mahoney btrfs_end_transaction(trans); 25363fd0a558SYan, Zheng return err; 25373fd0a558SYan, Zheng } 25383fd0a558SYan, Zheng 25393fd0a558SYan, Zheng static noinline_for_stack 2540aca1bba6SLiu Bo void free_reloc_roots(struct list_head *list) 2541aca1bba6SLiu Bo { 2542aca1bba6SLiu Bo struct btrfs_root *reloc_root; 2543aca1bba6SLiu Bo 2544aca1bba6SLiu Bo while (!list_empty(list)) { 2545aca1bba6SLiu Bo reloc_root = list_entry(list->next, struct btrfs_root, 2546aca1bba6SLiu Bo root_list); 2547bb166d72SNaohiro Aota __del_reloc_root(reloc_root); 25486bdf131fSJosef Bacik free_extent_buffer(reloc_root->node); 25496bdf131fSJosef Bacik free_extent_buffer(reloc_root->commit_root); 25506bdf131fSJosef Bacik reloc_root->node = NULL; 25516bdf131fSJosef Bacik reloc_root->commit_root = NULL; 2552aca1bba6SLiu Bo } 2553aca1bba6SLiu Bo } 2554aca1bba6SLiu Bo 2555aca1bba6SLiu Bo static noinline_for_stack 255694404e82SDavid Sterba void merge_reloc_roots(struct reloc_control *rc) 25573fd0a558SYan, Zheng { 25580b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 25595d4f98a2SYan Zheng struct btrfs_root *root; 25605d4f98a2SYan Zheng struct btrfs_root *reloc_root; 25613fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 25623fd0a558SYan, Zheng int found = 0; 2563aca1bba6SLiu Bo int ret = 0; 25643fd0a558SYan, Zheng again: 25653fd0a558SYan, Zheng root = rc->extent_root; 25667585717fSChris Mason 25677585717fSChris Mason /* 25687585717fSChris Mason * this serializes us with btrfs_record_root_in_transaction, 25697585717fSChris Mason * we have to make sure nobody is in the middle of 25707585717fSChris Mason * adding their roots to the list while we are 25717585717fSChris Mason * doing this splice 25727585717fSChris Mason */ 25730b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 25743fd0a558SYan, Zheng list_splice_init(&rc->reloc_roots, &reloc_roots); 25750b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 25765d4f98a2SYan Zheng 25773fd0a558SYan, Zheng while (!list_empty(&reloc_roots)) { 25783fd0a558SYan, Zheng found = 1; 25793fd0a558SYan, Zheng reloc_root = list_entry(reloc_roots.next, 25803fd0a558SYan, Zheng struct btrfs_root, root_list); 25815d4f98a2SYan Zheng 25825d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 25830b246afaSJeff Mahoney root = read_fs_root(fs_info, 25845d4f98a2SYan Zheng reloc_root->root_key.offset); 25855d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 25865d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 25875d4f98a2SYan Zheng 25883fd0a558SYan, Zheng ret = merge_reloc_root(rc, root); 258900246528SJosef Bacik btrfs_put_root(root); 2590b37b39cdSJosef Bacik if (ret) { 259125e293c2SWang Shilong if (list_empty(&reloc_root->root_list)) 259225e293c2SWang Shilong list_add_tail(&reloc_root->root_list, 259325e293c2SWang Shilong &reloc_roots); 2594aca1bba6SLiu Bo goto out; 2595b37b39cdSJosef Bacik } 25963fd0a558SYan, Zheng } else { 25973fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 259830d40577SQu Wenruo /* Don't forget to queue this reloc root for cleanup */ 259930d40577SQu Wenruo list_add_tail(&reloc_root->reloc_dirty_list, 260030d40577SQu Wenruo &rc->dirty_subvol_roots); 26013fd0a558SYan, Zheng } 26025d4f98a2SYan Zheng } 26035d4f98a2SYan Zheng 26043fd0a558SYan, Zheng if (found) { 26053fd0a558SYan, Zheng found = 0; 26063fd0a558SYan, Zheng goto again; 26075d4f98a2SYan Zheng } 2608aca1bba6SLiu Bo out: 2609aca1bba6SLiu Bo if (ret) { 26100b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 2611aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 2612aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 2613467bb1d2SWang Shilong 2614467bb1d2SWang Shilong /* new reloc root may be added */ 26150b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 2616467bb1d2SWang Shilong list_splice_init(&rc->reloc_roots, &reloc_roots); 26170b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2618467bb1d2SWang Shilong if (!list_empty(&reloc_roots)) 2619467bb1d2SWang Shilong free_reloc_roots(&reloc_roots); 2620aca1bba6SLiu Bo } 2621aca1bba6SLiu Bo 26225d4f98a2SYan Zheng BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); 26235d4f98a2SYan Zheng } 26245d4f98a2SYan Zheng 26255d4f98a2SYan Zheng static void free_block_list(struct rb_root *blocks) 26265d4f98a2SYan Zheng { 26275d4f98a2SYan Zheng struct tree_block *block; 26285d4f98a2SYan Zheng struct rb_node *rb_node; 26295d4f98a2SYan Zheng while ((rb_node = rb_first(blocks))) { 26305d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 26315d4f98a2SYan Zheng rb_erase(rb_node, blocks); 26325d4f98a2SYan Zheng kfree(block); 26335d4f98a2SYan Zheng } 26345d4f98a2SYan Zheng } 26355d4f98a2SYan Zheng 26365d4f98a2SYan Zheng static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans, 26375d4f98a2SYan Zheng struct btrfs_root *reloc_root) 26385d4f98a2SYan Zheng { 26390b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = reloc_root->fs_info; 26405d4f98a2SYan Zheng struct btrfs_root *root; 2641442b1ac5SJosef Bacik int ret; 26425d4f98a2SYan Zheng 26435d4f98a2SYan Zheng if (reloc_root->last_trans == trans->transid) 26445d4f98a2SYan Zheng return 0; 26455d4f98a2SYan Zheng 26460b246afaSJeff Mahoney root = read_fs_root(fs_info, reloc_root->root_key.offset); 26475d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 26485d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 2649442b1ac5SJosef Bacik ret = btrfs_record_root_in_trans(trans, root); 265000246528SJosef Bacik btrfs_put_root(root); 26515d4f98a2SYan Zheng 2652442b1ac5SJosef Bacik return ret; 26535d4f98a2SYan Zheng } 26545d4f98a2SYan Zheng 26553fd0a558SYan, Zheng static noinline_for_stack 26563fd0a558SYan, Zheng struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, 26573fd0a558SYan, Zheng struct reloc_control *rc, 26585d4f98a2SYan Zheng struct backref_node *node, 2659dc4103f9SWang Shilong struct backref_edge *edges[]) 26605d4f98a2SYan Zheng { 26615d4f98a2SYan Zheng struct backref_node *next; 26625d4f98a2SYan Zheng struct btrfs_root *root; 26633fd0a558SYan, Zheng int index = 0; 26643fd0a558SYan, Zheng 26655d4f98a2SYan Zheng next = node; 26665d4f98a2SYan Zheng while (1) { 26675d4f98a2SYan Zheng cond_resched(); 26685d4f98a2SYan Zheng next = walk_up_backref(next, edges, &index); 26695d4f98a2SYan Zheng root = next->root; 26703fd0a558SYan, Zheng BUG_ON(!root); 267127cdeb70SMiao Xie BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state)); 26725d4f98a2SYan Zheng 26735d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 26745d4f98a2SYan Zheng record_reloc_root_in_trans(trans, root); 26755d4f98a2SYan Zheng break; 26765d4f98a2SYan Zheng } 26775d4f98a2SYan Zheng 26785d4f98a2SYan Zheng btrfs_record_root_in_trans(trans, root); 26793fd0a558SYan, Zheng root = root->reloc_root; 26803fd0a558SYan, Zheng 26813fd0a558SYan, Zheng if (next->new_bytenr != root->node->start) { 26823fd0a558SYan, Zheng BUG_ON(next->new_bytenr); 26833fd0a558SYan, Zheng BUG_ON(!list_empty(&next->list)); 26843fd0a558SYan, Zheng next->new_bytenr = root->node->start; 268500246528SJosef Bacik btrfs_put_root(next->root); 268600246528SJosef Bacik next->root = btrfs_grab_root(root); 26870b530bc5SJosef Bacik ASSERT(next->root); 26883fd0a558SYan, Zheng list_add_tail(&next->list, 26893fd0a558SYan, Zheng &rc->backref_cache.changed); 26903fd0a558SYan, Zheng __mark_block_processed(rc, next); 26915d4f98a2SYan Zheng break; 26925d4f98a2SYan Zheng } 26935d4f98a2SYan Zheng 26943fd0a558SYan, Zheng WARN_ON(1); 26955d4f98a2SYan Zheng root = NULL; 26965d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 26975d4f98a2SYan Zheng if (!next || next->level <= node->level) 26985d4f98a2SYan Zheng break; 26995d4f98a2SYan Zheng } 27003fd0a558SYan, Zheng if (!root) 27013fd0a558SYan, Zheng return NULL; 27025d4f98a2SYan Zheng 27033fd0a558SYan, Zheng next = node; 27043fd0a558SYan, Zheng /* setup backref node path for btrfs_reloc_cow_block */ 27053fd0a558SYan, Zheng while (1) { 27063fd0a558SYan, Zheng rc->backref_cache.path[next->level] = next; 27073fd0a558SYan, Zheng if (--index < 0) 27083fd0a558SYan, Zheng break; 27093fd0a558SYan, Zheng next = edges[index]->node[UPPER]; 27103fd0a558SYan, Zheng } 27115d4f98a2SYan Zheng return root; 27125d4f98a2SYan Zheng } 27135d4f98a2SYan Zheng 27143fd0a558SYan, Zheng /* 27153fd0a558SYan, Zheng * select a tree root for relocation. return NULL if the block 27163fd0a558SYan, Zheng * is reference counted. we should use do_relocation() in this 27173fd0a558SYan, Zheng * case. return a tree root pointer if the block isn't reference 27183fd0a558SYan, Zheng * counted. return -ENOENT if the block is root of reloc tree. 27193fd0a558SYan, Zheng */ 27205d4f98a2SYan Zheng static noinline_for_stack 2721147d256eSZhaolei struct btrfs_root *select_one_root(struct backref_node *node) 27225d4f98a2SYan Zheng { 27233fd0a558SYan, Zheng struct backref_node *next; 27243fd0a558SYan, Zheng struct btrfs_root *root; 27253fd0a558SYan, Zheng struct btrfs_root *fs_root = NULL; 27265d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 27273fd0a558SYan, Zheng int index = 0; 27283fd0a558SYan, Zheng 27293fd0a558SYan, Zheng next = node; 27303fd0a558SYan, Zheng while (1) { 27313fd0a558SYan, Zheng cond_resched(); 27323fd0a558SYan, Zheng next = walk_up_backref(next, edges, &index); 27333fd0a558SYan, Zheng root = next->root; 27343fd0a558SYan, Zheng BUG_ON(!root); 27353fd0a558SYan, Zheng 273625985edcSLucas De Marchi /* no other choice for non-references counted tree */ 273727cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 27383fd0a558SYan, Zheng return root; 27393fd0a558SYan, Zheng 27403fd0a558SYan, Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) 27413fd0a558SYan, Zheng fs_root = root; 27423fd0a558SYan, Zheng 27433fd0a558SYan, Zheng if (next != node) 27443fd0a558SYan, Zheng return NULL; 27453fd0a558SYan, Zheng 27463fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 27473fd0a558SYan, Zheng if (!next || next->level <= node->level) 27483fd0a558SYan, Zheng break; 27493fd0a558SYan, Zheng } 27503fd0a558SYan, Zheng 27513fd0a558SYan, Zheng if (!fs_root) 27523fd0a558SYan, Zheng return ERR_PTR(-ENOENT); 27533fd0a558SYan, Zheng return fs_root; 27545d4f98a2SYan Zheng } 27555d4f98a2SYan Zheng 27565d4f98a2SYan Zheng static noinline_for_stack 27573fd0a558SYan, Zheng u64 calcu_metadata_size(struct reloc_control *rc, 27583fd0a558SYan, Zheng struct backref_node *node, int reserve) 27595d4f98a2SYan Zheng { 27600b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 27613fd0a558SYan, Zheng struct backref_node *next = node; 27623fd0a558SYan, Zheng struct backref_edge *edge; 27633fd0a558SYan, Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 27643fd0a558SYan, Zheng u64 num_bytes = 0; 27653fd0a558SYan, Zheng int index = 0; 27665d4f98a2SYan Zheng 27673fd0a558SYan, Zheng BUG_ON(reserve && node->processed); 27683fd0a558SYan, Zheng 27693fd0a558SYan, Zheng while (next) { 27703fd0a558SYan, Zheng cond_resched(); 27715d4f98a2SYan Zheng while (1) { 27723fd0a558SYan, Zheng if (next->processed && (reserve || next != node)) 27735d4f98a2SYan Zheng break; 27745d4f98a2SYan Zheng 27750b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 27763fd0a558SYan, Zheng 27773fd0a558SYan, Zheng if (list_empty(&next->upper)) 27783fd0a558SYan, Zheng break; 27793fd0a558SYan, Zheng 27803fd0a558SYan, Zheng edge = list_entry(next->upper.next, 27813fd0a558SYan, Zheng struct backref_edge, list[LOWER]); 27823fd0a558SYan, Zheng edges[index++] = edge; 27833fd0a558SYan, Zheng next = edge->node[UPPER]; 27845d4f98a2SYan Zheng } 27853fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 27863fd0a558SYan, Zheng } 27873fd0a558SYan, Zheng return num_bytes; 27883fd0a558SYan, Zheng } 27893fd0a558SYan, Zheng 27903fd0a558SYan, Zheng static int reserve_metadata_space(struct btrfs_trans_handle *trans, 27913fd0a558SYan, Zheng struct reloc_control *rc, 27923fd0a558SYan, Zheng struct backref_node *node) 27933fd0a558SYan, Zheng { 27943fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 2795da17066cSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 27963fd0a558SYan, Zheng u64 num_bytes; 27973fd0a558SYan, Zheng int ret; 27980647bf56SWang Shilong u64 tmp; 27993fd0a558SYan, Zheng 28003fd0a558SYan, Zheng num_bytes = calcu_metadata_size(rc, node, 1) * 2; 28013fd0a558SYan, Zheng 28023fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 28030647bf56SWang Shilong rc->reserved_bytes += num_bytes; 28048ca17f0fSJosef Bacik 28058ca17f0fSJosef Bacik /* 28068ca17f0fSJosef Bacik * We are under a transaction here so we can only do limited flushing. 28078ca17f0fSJosef Bacik * If we get an enospc just kick back -EAGAIN so we know to drop the 28088ca17f0fSJosef Bacik * transaction and try to refill when we can flush all the things. 28098ca17f0fSJosef Bacik */ 28100647bf56SWang Shilong ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes, 28118ca17f0fSJosef Bacik BTRFS_RESERVE_FLUSH_LIMIT); 28123fd0a558SYan, Zheng if (ret) { 2813da17066cSJeff Mahoney tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES; 28140647bf56SWang Shilong while (tmp <= rc->reserved_bytes) 28150647bf56SWang Shilong tmp <<= 1; 28160647bf56SWang Shilong /* 28170647bf56SWang Shilong * only one thread can access block_rsv at this point, 28180647bf56SWang Shilong * so we don't need hold lock to protect block_rsv. 28190647bf56SWang Shilong * we expand more reservation size here to allow enough 282052042d8eSAndrea Gelmini * space for relocation and we will return earlier in 28210647bf56SWang Shilong * enospc case. 28220647bf56SWang Shilong */ 2823da17066cSJeff Mahoney rc->block_rsv->size = tmp + fs_info->nodesize * 28240647bf56SWang Shilong RELOCATION_RESERVED_NODES; 28258ca17f0fSJosef Bacik return -EAGAIN; 28263fd0a558SYan, Zheng } 28273fd0a558SYan, Zheng 28283fd0a558SYan, Zheng return 0; 28293fd0a558SYan, Zheng } 28303fd0a558SYan, Zheng 28315d4f98a2SYan Zheng /* 28325d4f98a2SYan Zheng * relocate a block tree, and then update pointers in upper level 28335d4f98a2SYan Zheng * blocks that reference the block to point to the new location. 28345d4f98a2SYan Zheng * 28355d4f98a2SYan Zheng * if called by link_to_upper, the block has already been relocated. 28365d4f98a2SYan Zheng * in that case this function just updates pointers. 28375d4f98a2SYan Zheng */ 28385d4f98a2SYan Zheng static int do_relocation(struct btrfs_trans_handle *trans, 28393fd0a558SYan, Zheng struct reloc_control *rc, 28405d4f98a2SYan Zheng struct backref_node *node, 28415d4f98a2SYan Zheng struct btrfs_key *key, 28425d4f98a2SYan Zheng struct btrfs_path *path, int lowest) 28435d4f98a2SYan Zheng { 28442ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 28455d4f98a2SYan Zheng struct backref_node *upper; 28465d4f98a2SYan Zheng struct backref_edge *edge; 28475d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 28485d4f98a2SYan Zheng struct btrfs_root *root; 28495d4f98a2SYan Zheng struct extent_buffer *eb; 28505d4f98a2SYan Zheng u32 blocksize; 28515d4f98a2SYan Zheng u64 bytenr; 28525d4f98a2SYan Zheng u64 generation; 28535d4f98a2SYan Zheng int slot; 28545d4f98a2SYan Zheng int ret; 28555d4f98a2SYan Zheng int err = 0; 28565d4f98a2SYan Zheng 28575d4f98a2SYan Zheng BUG_ON(lowest && node->eb); 28585d4f98a2SYan Zheng 28595d4f98a2SYan Zheng path->lowest_level = node->level + 1; 28603fd0a558SYan, Zheng rc->backref_cache.path[node->level] = node; 28615d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) { 2862581c1760SQu Wenruo struct btrfs_key first_key; 286382fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 2864581c1760SQu Wenruo 28655d4f98a2SYan Zheng cond_resched(); 28665d4f98a2SYan Zheng 28675d4f98a2SYan Zheng upper = edge->node[UPPER]; 2868dc4103f9SWang Shilong root = select_reloc_root(trans, rc, upper, edges); 28693fd0a558SYan, Zheng BUG_ON(!root); 28705d4f98a2SYan Zheng 28713fd0a558SYan, Zheng if (upper->eb && !upper->locked) { 28723fd0a558SYan, Zheng if (!lowest) { 28733fd0a558SYan, Zheng ret = btrfs_bin_search(upper->eb, key, 28743fd0a558SYan, Zheng upper->level, &slot); 2875cbca7d59SFilipe Manana if (ret < 0) { 2876cbca7d59SFilipe Manana err = ret; 2877cbca7d59SFilipe Manana goto next; 2878cbca7d59SFilipe Manana } 28793fd0a558SYan, Zheng BUG_ON(ret); 28803fd0a558SYan, Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 28813fd0a558SYan, Zheng if (node->eb->start == bytenr) 28823fd0a558SYan, Zheng goto next; 28833fd0a558SYan, Zheng } 28845d4f98a2SYan Zheng drop_node_buffer(upper); 28853fd0a558SYan, Zheng } 28865d4f98a2SYan Zheng 28875d4f98a2SYan Zheng if (!upper->eb) { 28885d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 28893561b9dbSLiu Bo if (ret) { 28903561b9dbSLiu Bo if (ret < 0) 28915d4f98a2SYan Zheng err = ret; 28923561b9dbSLiu Bo else 28933561b9dbSLiu Bo err = -ENOENT; 28943561b9dbSLiu Bo 28953561b9dbSLiu Bo btrfs_release_path(path); 28965d4f98a2SYan Zheng break; 28975d4f98a2SYan Zheng } 28985d4f98a2SYan Zheng 28993fd0a558SYan, Zheng if (!upper->eb) { 29003fd0a558SYan, Zheng upper->eb = path->nodes[upper->level]; 29013fd0a558SYan, Zheng path->nodes[upper->level] = NULL; 29023fd0a558SYan, Zheng } else { 29033fd0a558SYan, Zheng BUG_ON(upper->eb != path->nodes[upper->level]); 29043fd0a558SYan, Zheng } 29053fd0a558SYan, Zheng 29063fd0a558SYan, Zheng upper->locked = 1; 29073fd0a558SYan, Zheng path->locks[upper->level] = 0; 29083fd0a558SYan, Zheng 29095d4f98a2SYan Zheng slot = path->slots[upper->level]; 2910b3b4aa74SDavid Sterba btrfs_release_path(path); 29115d4f98a2SYan Zheng } else { 29125d4f98a2SYan Zheng ret = btrfs_bin_search(upper->eb, key, upper->level, 29135d4f98a2SYan Zheng &slot); 2914cbca7d59SFilipe Manana if (ret < 0) { 2915cbca7d59SFilipe Manana err = ret; 2916cbca7d59SFilipe Manana goto next; 2917cbca7d59SFilipe Manana } 29185d4f98a2SYan Zheng BUG_ON(ret); 29195d4f98a2SYan Zheng } 29205d4f98a2SYan Zheng 29215d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 29223fd0a558SYan, Zheng if (lowest) { 29234547f4d8SLiu Bo if (bytenr != node->bytenr) { 29244547f4d8SLiu Bo btrfs_err(root->fs_info, 29254547f4d8SLiu Bo "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu", 29264547f4d8SLiu Bo bytenr, node->bytenr, slot, 29274547f4d8SLiu Bo upper->eb->start); 29284547f4d8SLiu Bo err = -EIO; 29294547f4d8SLiu Bo goto next; 29304547f4d8SLiu Bo } 29315d4f98a2SYan Zheng } else { 29323fd0a558SYan, Zheng if (node->eb->start == bytenr) 29333fd0a558SYan, Zheng goto next; 29345d4f98a2SYan Zheng } 29355d4f98a2SYan Zheng 2936da17066cSJeff Mahoney blocksize = root->fs_info->nodesize; 29375d4f98a2SYan Zheng generation = btrfs_node_ptr_generation(upper->eb, slot); 2938581c1760SQu Wenruo btrfs_node_key_to_cpu(upper->eb, &first_key, slot); 2939581c1760SQu Wenruo eb = read_tree_block(fs_info, bytenr, generation, 2940581c1760SQu Wenruo upper->level - 1, &first_key); 294164c043deSLiu Bo if (IS_ERR(eb)) { 294264c043deSLiu Bo err = PTR_ERR(eb); 294364c043deSLiu Bo goto next; 294464c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2945416bc658SJosef Bacik free_extent_buffer(eb); 294697d9a8a4STsutomu Itoh err = -EIO; 294797d9a8a4STsutomu Itoh goto next; 294897d9a8a4STsutomu Itoh } 29495d4f98a2SYan Zheng btrfs_tree_lock(eb); 29508bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 29515d4f98a2SYan Zheng 29525d4f98a2SYan Zheng if (!node->eb) { 29535d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, eb, upper->eb, 29545d4f98a2SYan Zheng slot, &eb); 29553fd0a558SYan, Zheng btrfs_tree_unlock(eb); 29563fd0a558SYan, Zheng free_extent_buffer(eb); 29575d4f98a2SYan Zheng if (ret < 0) { 29585d4f98a2SYan Zheng err = ret; 29593fd0a558SYan, Zheng goto next; 29605d4f98a2SYan Zheng } 29613fd0a558SYan, Zheng BUG_ON(node->eb != eb); 29625d4f98a2SYan Zheng } else { 29635d4f98a2SYan Zheng btrfs_set_node_blockptr(upper->eb, slot, 29645d4f98a2SYan Zheng node->eb->start); 29655d4f98a2SYan Zheng btrfs_set_node_ptr_generation(upper->eb, slot, 29665d4f98a2SYan Zheng trans->transid); 29675d4f98a2SYan Zheng btrfs_mark_buffer_dirty(upper->eb); 29685d4f98a2SYan Zheng 296982fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, 29705d4f98a2SYan Zheng node->eb->start, blocksize, 297182fa113fSQu Wenruo upper->eb->start); 297282fa113fSQu Wenruo ref.real_root = root->root_key.objectid; 297382fa113fSQu Wenruo btrfs_init_tree_ref(&ref, node->level, 297482fa113fSQu Wenruo btrfs_header_owner(upper->eb)); 297582fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 29765d4f98a2SYan Zheng BUG_ON(ret); 29775d4f98a2SYan Zheng 29785d4f98a2SYan Zheng ret = btrfs_drop_subtree(trans, root, eb, upper->eb); 29795d4f98a2SYan Zheng BUG_ON(ret); 29805d4f98a2SYan Zheng } 29813fd0a558SYan, Zheng next: 29823fd0a558SYan, Zheng if (!upper->pending) 29833fd0a558SYan, Zheng drop_node_buffer(upper); 29843fd0a558SYan, Zheng else 29853fd0a558SYan, Zheng unlock_node_buffer(upper); 29863fd0a558SYan, Zheng if (err) 29873fd0a558SYan, Zheng break; 29885d4f98a2SYan Zheng } 29893fd0a558SYan, Zheng 29903fd0a558SYan, Zheng if (!err && node->pending) { 29913fd0a558SYan, Zheng drop_node_buffer(node); 29923fd0a558SYan, Zheng list_move_tail(&node->list, &rc->backref_cache.changed); 29933fd0a558SYan, Zheng node->pending = 0; 29945d4f98a2SYan Zheng } 29953fd0a558SYan, Zheng 29965d4f98a2SYan Zheng path->lowest_level = 0; 29973fd0a558SYan, Zheng BUG_ON(err == -ENOSPC); 29985d4f98a2SYan Zheng return err; 29995d4f98a2SYan Zheng } 30005d4f98a2SYan Zheng 30015d4f98a2SYan Zheng static int link_to_upper(struct btrfs_trans_handle *trans, 30023fd0a558SYan, Zheng struct reloc_control *rc, 30035d4f98a2SYan Zheng struct backref_node *node, 30045d4f98a2SYan Zheng struct btrfs_path *path) 30055d4f98a2SYan Zheng { 30065d4f98a2SYan Zheng struct btrfs_key key; 30075d4f98a2SYan Zheng 30085d4f98a2SYan Zheng btrfs_node_key_to_cpu(node->eb, &key, 0); 30093fd0a558SYan, Zheng return do_relocation(trans, rc, node, &key, path, 0); 30105d4f98a2SYan Zheng } 30115d4f98a2SYan Zheng 30125d4f98a2SYan Zheng static int finish_pending_nodes(struct btrfs_trans_handle *trans, 30133fd0a558SYan, Zheng struct reloc_control *rc, 30143fd0a558SYan, Zheng struct btrfs_path *path, int err) 30155d4f98a2SYan Zheng { 30163fd0a558SYan, Zheng LIST_HEAD(list); 30173fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 30185d4f98a2SYan Zheng struct backref_node *node; 30195d4f98a2SYan Zheng int level; 30205d4f98a2SYan Zheng int ret; 30215d4f98a2SYan Zheng 30225d4f98a2SYan Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 30235d4f98a2SYan Zheng while (!list_empty(&cache->pending[level])) { 30245d4f98a2SYan Zheng node = list_entry(cache->pending[level].next, 30253fd0a558SYan, Zheng struct backref_node, list); 30263fd0a558SYan, Zheng list_move_tail(&node->list, &list); 30273fd0a558SYan, Zheng BUG_ON(!node->pending); 30285d4f98a2SYan Zheng 30293fd0a558SYan, Zheng if (!err) { 30303fd0a558SYan, Zheng ret = link_to_upper(trans, rc, node, path); 30315d4f98a2SYan Zheng if (ret < 0) 30325d4f98a2SYan Zheng err = ret; 30335d4f98a2SYan Zheng } 30345d4f98a2SYan Zheng } 30353fd0a558SYan, Zheng list_splice_init(&list, &cache->pending[level]); 30363fd0a558SYan, Zheng } 30375d4f98a2SYan Zheng return err; 30385d4f98a2SYan Zheng } 30395d4f98a2SYan Zheng 30405d4f98a2SYan Zheng static void mark_block_processed(struct reloc_control *rc, 30413fd0a558SYan, Zheng u64 bytenr, u32 blocksize) 30423fd0a558SYan, Zheng { 30433fd0a558SYan, Zheng set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1, 3044ceeb0ae7SDavid Sterba EXTENT_DIRTY); 30453fd0a558SYan, Zheng } 30463fd0a558SYan, Zheng 30473fd0a558SYan, Zheng static void __mark_block_processed(struct reloc_control *rc, 30485d4f98a2SYan Zheng struct backref_node *node) 30495d4f98a2SYan Zheng { 30505d4f98a2SYan Zheng u32 blocksize; 30515d4f98a2SYan Zheng if (node->level == 0 || 30525d4f98a2SYan Zheng in_block_group(node->bytenr, rc->block_group)) { 3053da17066cSJeff Mahoney blocksize = rc->extent_root->fs_info->nodesize; 30543fd0a558SYan, Zheng mark_block_processed(rc, node->bytenr, blocksize); 30555d4f98a2SYan Zheng } 30565d4f98a2SYan Zheng node->processed = 1; 30575d4f98a2SYan Zheng } 30585d4f98a2SYan Zheng 30595d4f98a2SYan Zheng /* 30605d4f98a2SYan Zheng * mark a block and all blocks directly/indirectly reference the block 30615d4f98a2SYan Zheng * as processed. 30625d4f98a2SYan Zheng */ 30635d4f98a2SYan Zheng static void update_processed_blocks(struct reloc_control *rc, 30645d4f98a2SYan Zheng struct backref_node *node) 30655d4f98a2SYan Zheng { 30665d4f98a2SYan Zheng struct backref_node *next = node; 30675d4f98a2SYan Zheng struct backref_edge *edge; 30685d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 30695d4f98a2SYan Zheng int index = 0; 30705d4f98a2SYan Zheng 30715d4f98a2SYan Zheng while (next) { 30725d4f98a2SYan Zheng cond_resched(); 30735d4f98a2SYan Zheng while (1) { 30745d4f98a2SYan Zheng if (next->processed) 30755d4f98a2SYan Zheng break; 30765d4f98a2SYan Zheng 30773fd0a558SYan, Zheng __mark_block_processed(rc, next); 30785d4f98a2SYan Zheng 30795d4f98a2SYan Zheng if (list_empty(&next->upper)) 30805d4f98a2SYan Zheng break; 30815d4f98a2SYan Zheng 30825d4f98a2SYan Zheng edge = list_entry(next->upper.next, 30835d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 30845d4f98a2SYan Zheng edges[index++] = edge; 30855d4f98a2SYan Zheng next = edge->node[UPPER]; 30865d4f98a2SYan Zheng } 30875d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 30885d4f98a2SYan Zheng } 30895d4f98a2SYan Zheng } 30905d4f98a2SYan Zheng 30917476dfdaSDavid Sterba static int tree_block_processed(u64 bytenr, struct reloc_control *rc) 30925d4f98a2SYan Zheng { 3093da17066cSJeff Mahoney u32 blocksize = rc->extent_root->fs_info->nodesize; 30947476dfdaSDavid Sterba 30955d4f98a2SYan Zheng if (test_range_bit(&rc->processed_blocks, bytenr, 30969655d298SChris Mason bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL)) 30975d4f98a2SYan Zheng return 1; 30985d4f98a2SYan Zheng return 0; 30995d4f98a2SYan Zheng } 31005d4f98a2SYan Zheng 31012ff7e61eSJeff Mahoney static int get_tree_block_key(struct btrfs_fs_info *fs_info, 31025d4f98a2SYan Zheng struct tree_block *block) 31035d4f98a2SYan Zheng { 31045d4f98a2SYan Zheng struct extent_buffer *eb; 31055d4f98a2SYan Zheng 3106581c1760SQu Wenruo eb = read_tree_block(fs_info, block->bytenr, block->key.offset, 3107581c1760SQu Wenruo block->level, NULL); 310864c043deSLiu Bo if (IS_ERR(eb)) { 310964c043deSLiu Bo return PTR_ERR(eb); 311064c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 3111416bc658SJosef Bacik free_extent_buffer(eb); 3112416bc658SJosef Bacik return -EIO; 3113416bc658SJosef Bacik } 31145d4f98a2SYan Zheng if (block->level == 0) 31155d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &block->key, 0); 31165d4f98a2SYan Zheng else 31175d4f98a2SYan Zheng btrfs_node_key_to_cpu(eb, &block->key, 0); 31185d4f98a2SYan Zheng free_extent_buffer(eb); 31195d4f98a2SYan Zheng block->key_ready = 1; 31205d4f98a2SYan Zheng return 0; 31215d4f98a2SYan Zheng } 31225d4f98a2SYan Zheng 31235d4f98a2SYan Zheng /* 31245d4f98a2SYan Zheng * helper function to relocate a tree block 31255d4f98a2SYan Zheng */ 31265d4f98a2SYan Zheng static int relocate_tree_block(struct btrfs_trans_handle *trans, 31275d4f98a2SYan Zheng struct reloc_control *rc, 31285d4f98a2SYan Zheng struct backref_node *node, 31295d4f98a2SYan Zheng struct btrfs_key *key, 31305d4f98a2SYan Zheng struct btrfs_path *path) 31315d4f98a2SYan Zheng { 31325d4f98a2SYan Zheng struct btrfs_root *root; 31333fd0a558SYan, Zheng int ret = 0; 31345d4f98a2SYan Zheng 31353fd0a558SYan, Zheng if (!node) 31365d4f98a2SYan Zheng return 0; 31373fd0a558SYan, Zheng 31383fd0a558SYan, Zheng BUG_ON(node->processed); 3139147d256eSZhaolei root = select_one_root(node); 31403fd0a558SYan, Zheng if (root == ERR_PTR(-ENOENT)) { 31413fd0a558SYan, Zheng update_processed_blocks(rc, node); 31423fd0a558SYan, Zheng goto out; 31435d4f98a2SYan Zheng } 31445d4f98a2SYan Zheng 314527cdeb70SMiao Xie if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 31463fd0a558SYan, Zheng ret = reserve_metadata_space(trans, rc, node); 31473fd0a558SYan, Zheng if (ret) 31485d4f98a2SYan Zheng goto out; 31495d4f98a2SYan Zheng } 31503fd0a558SYan, Zheng 31513fd0a558SYan, Zheng if (root) { 315227cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 31533fd0a558SYan, Zheng BUG_ON(node->new_bytenr); 31543fd0a558SYan, Zheng BUG_ON(!list_empty(&node->list)); 31553fd0a558SYan, Zheng btrfs_record_root_in_trans(trans, root); 31563fd0a558SYan, Zheng root = root->reloc_root; 31573fd0a558SYan, Zheng node->new_bytenr = root->node->start; 315800246528SJosef Bacik btrfs_put_root(node->root); 315900246528SJosef Bacik node->root = btrfs_grab_root(root); 31600b530bc5SJosef Bacik ASSERT(node->root); 31613fd0a558SYan, Zheng list_add_tail(&node->list, &rc->backref_cache.changed); 31623fd0a558SYan, Zheng } else { 31635d4f98a2SYan Zheng path->lowest_level = node->level; 31645d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 3165b3b4aa74SDavid Sterba btrfs_release_path(path); 31663fd0a558SYan, Zheng if (ret > 0) 31675d4f98a2SYan Zheng ret = 0; 31683fd0a558SYan, Zheng } 31693fd0a558SYan, Zheng if (!ret) 31703fd0a558SYan, Zheng update_processed_blocks(rc, node); 31713fd0a558SYan, Zheng } else { 31723fd0a558SYan, Zheng ret = do_relocation(trans, rc, node, key, path, 1); 31733fd0a558SYan, Zheng } 31745d4f98a2SYan Zheng out: 31750647bf56SWang Shilong if (ret || node->level == 0 || node->cowonly) 31763fd0a558SYan, Zheng remove_backref_node(&rc->backref_cache, node); 31775d4f98a2SYan Zheng return ret; 31785d4f98a2SYan Zheng } 31795d4f98a2SYan Zheng 31805d4f98a2SYan Zheng /* 31815d4f98a2SYan Zheng * relocate a list of blocks 31825d4f98a2SYan Zheng */ 31835d4f98a2SYan Zheng static noinline_for_stack 31845d4f98a2SYan Zheng int relocate_tree_blocks(struct btrfs_trans_handle *trans, 31855d4f98a2SYan Zheng struct reloc_control *rc, struct rb_root *blocks) 31865d4f98a2SYan Zheng { 31872ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 31885d4f98a2SYan Zheng struct backref_node *node; 31895d4f98a2SYan Zheng struct btrfs_path *path; 31905d4f98a2SYan Zheng struct tree_block *block; 319198ff7b94SQu Wenruo struct tree_block *next; 31925d4f98a2SYan Zheng int ret; 31935d4f98a2SYan Zheng int err = 0; 31945d4f98a2SYan Zheng 31955d4f98a2SYan Zheng path = btrfs_alloc_path(); 3196e1a12670SLiu Bo if (!path) { 3197e1a12670SLiu Bo err = -ENOMEM; 319834c2b290SDavid Sterba goto out_free_blocks; 3199e1a12670SLiu Bo } 32005d4f98a2SYan Zheng 320198ff7b94SQu Wenruo /* Kick in readahead for tree blocks with missing keys */ 320298ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 32035d4f98a2SYan Zheng if (!block->key_ready) 32042ff7e61eSJeff Mahoney readahead_tree_block(fs_info, block->bytenr); 32055d4f98a2SYan Zheng } 32065d4f98a2SYan Zheng 320798ff7b94SQu Wenruo /* Get first keys */ 320898ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 320934c2b290SDavid Sterba if (!block->key_ready) { 32102ff7e61eSJeff Mahoney err = get_tree_block_key(fs_info, block); 321134c2b290SDavid Sterba if (err) 321234c2b290SDavid Sterba goto out_free_path; 321334c2b290SDavid Sterba } 32145d4f98a2SYan Zheng } 32155d4f98a2SYan Zheng 321698ff7b94SQu Wenruo /* Do tree relocation */ 321798ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 32183fd0a558SYan, Zheng node = build_backref_tree(rc, &block->key, 32195d4f98a2SYan Zheng block->level, block->bytenr); 32205d4f98a2SYan Zheng if (IS_ERR(node)) { 32215d4f98a2SYan Zheng err = PTR_ERR(node); 32225d4f98a2SYan Zheng goto out; 32235d4f98a2SYan Zheng } 32245d4f98a2SYan Zheng 32255d4f98a2SYan Zheng ret = relocate_tree_block(trans, rc, node, &block->key, 32265d4f98a2SYan Zheng path); 32275d4f98a2SYan Zheng if (ret < 0) { 322898ff7b94SQu Wenruo if (ret != -EAGAIN || &block->rb_node == rb_first(blocks)) 32295d4f98a2SYan Zheng err = ret; 32305d4f98a2SYan Zheng goto out; 32315d4f98a2SYan Zheng } 32325d4f98a2SYan Zheng } 32335d4f98a2SYan Zheng out: 32343fd0a558SYan, Zheng err = finish_pending_nodes(trans, rc, path, err); 32355d4f98a2SYan Zheng 323634c2b290SDavid Sterba out_free_path: 32375d4f98a2SYan Zheng btrfs_free_path(path); 323834c2b290SDavid Sterba out_free_blocks: 3239e1a12670SLiu Bo free_block_list(blocks); 32405d4f98a2SYan Zheng return err; 32415d4f98a2SYan Zheng } 32425d4f98a2SYan Zheng 32435d4f98a2SYan Zheng static noinline_for_stack 3244efa56464SYan, Zheng int prealloc_file_extent_cluster(struct inode *inode, 3245efa56464SYan, Zheng struct file_extent_cluster *cluster) 3246efa56464SYan, Zheng { 3247efa56464SYan, Zheng u64 alloc_hint = 0; 3248efa56464SYan, Zheng u64 start; 3249efa56464SYan, Zheng u64 end; 3250efa56464SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 3251efa56464SYan, Zheng u64 num_bytes; 3252efa56464SYan, Zheng int nr = 0; 3253efa56464SYan, Zheng int ret = 0; 3254dcb40c19SWang Xiaoguang u64 prealloc_start = cluster->start - offset; 3255dcb40c19SWang Xiaoguang u64 prealloc_end = cluster->end - offset; 325618513091SWang Xiaoguang u64 cur_offset; 3257364ecf36SQu Wenruo struct extent_changeset *data_reserved = NULL; 3258efa56464SYan, Zheng 3259efa56464SYan, Zheng BUG_ON(cluster->start != cluster->boundary[0]); 32605955102cSAl Viro inode_lock(inode); 3261efa56464SYan, Zheng 3262364ecf36SQu Wenruo ret = btrfs_check_data_free_space(inode, &data_reserved, prealloc_start, 3263dcb40c19SWang Xiaoguang prealloc_end + 1 - prealloc_start); 3264efa56464SYan, Zheng if (ret) 3265efa56464SYan, Zheng goto out; 3266efa56464SYan, Zheng 326718513091SWang Xiaoguang cur_offset = prealloc_start; 3268efa56464SYan, Zheng while (nr < cluster->nr) { 3269efa56464SYan, Zheng start = cluster->boundary[nr] - offset; 3270efa56464SYan, Zheng if (nr + 1 < cluster->nr) 3271efa56464SYan, Zheng end = cluster->boundary[nr + 1] - 1 - offset; 3272efa56464SYan, Zheng else 3273efa56464SYan, Zheng end = cluster->end - offset; 3274efa56464SYan, Zheng 3275d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 3276efa56464SYan, Zheng num_bytes = end + 1 - start; 327718513091SWang Xiaoguang if (cur_offset < start) 3278bc42bda2SQu Wenruo btrfs_free_reserved_data_space(inode, data_reserved, 3279bc42bda2SQu Wenruo cur_offset, start - cur_offset); 3280efa56464SYan, Zheng ret = btrfs_prealloc_file_range(inode, 0, start, 3281efa56464SYan, Zheng num_bytes, num_bytes, 3282efa56464SYan, Zheng end + 1, &alloc_hint); 328318513091SWang Xiaoguang cur_offset = end + 1; 3284d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 3285efa56464SYan, Zheng if (ret) 3286efa56464SYan, Zheng break; 3287efa56464SYan, Zheng nr++; 3288efa56464SYan, Zheng } 328918513091SWang Xiaoguang if (cur_offset < prealloc_end) 3290bc42bda2SQu Wenruo btrfs_free_reserved_data_space(inode, data_reserved, 3291bc42bda2SQu Wenruo cur_offset, prealloc_end + 1 - cur_offset); 3292efa56464SYan, Zheng out: 32935955102cSAl Viro inode_unlock(inode); 3294364ecf36SQu Wenruo extent_changeset_free(data_reserved); 3295efa56464SYan, Zheng return ret; 3296efa56464SYan, Zheng } 3297efa56464SYan, Zheng 3298efa56464SYan, Zheng static noinline_for_stack 32990257bb82SYan, Zheng int setup_extent_mapping(struct inode *inode, u64 start, u64 end, 33000257bb82SYan, Zheng u64 block_start) 33015d4f98a2SYan Zheng { 33025d4f98a2SYan Zheng struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 33035d4f98a2SYan Zheng struct extent_map *em; 33040257bb82SYan, Zheng int ret = 0; 33055d4f98a2SYan Zheng 3306172ddd60SDavid Sterba em = alloc_extent_map(); 33070257bb82SYan, Zheng if (!em) 33080257bb82SYan, Zheng return -ENOMEM; 33090257bb82SYan, Zheng 33105d4f98a2SYan Zheng em->start = start; 33110257bb82SYan, Zheng em->len = end + 1 - start; 33120257bb82SYan, Zheng em->block_len = em->len; 33130257bb82SYan, Zheng em->block_start = block_start; 33145d4f98a2SYan Zheng set_bit(EXTENT_FLAG_PINNED, &em->flags); 33155d4f98a2SYan Zheng 3316d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 33175d4f98a2SYan Zheng while (1) { 3318890871beSChris Mason write_lock(&em_tree->lock); 331909a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 3320890871beSChris Mason write_unlock(&em_tree->lock); 33215d4f98a2SYan Zheng if (ret != -EEXIST) { 33225d4f98a2SYan Zheng free_extent_map(em); 33235d4f98a2SYan Zheng break; 33245d4f98a2SYan Zheng } 3325dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0); 33265d4f98a2SYan Zheng } 3327d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 33280257bb82SYan, Zheng return ret; 33290257bb82SYan, Zheng } 33305d4f98a2SYan Zheng 3331726a3421SQu Wenruo /* 3332726a3421SQu Wenruo * Allow error injection to test balance cancellation 3333726a3421SQu Wenruo */ 3334726a3421SQu Wenruo int btrfs_should_cancel_balance(struct btrfs_fs_info *fs_info) 3335726a3421SQu Wenruo { 3336726a3421SQu Wenruo return atomic_read(&fs_info->balance_cancel_req); 3337726a3421SQu Wenruo } 3338726a3421SQu Wenruo ALLOW_ERROR_INJECTION(btrfs_should_cancel_balance, TRUE); 3339726a3421SQu Wenruo 33400257bb82SYan, Zheng static int relocate_file_extent_cluster(struct inode *inode, 33410257bb82SYan, Zheng struct file_extent_cluster *cluster) 33420257bb82SYan, Zheng { 33432ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 33440257bb82SYan, Zheng u64 page_start; 33450257bb82SYan, Zheng u64 page_end; 33460257bb82SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 33470257bb82SYan, Zheng unsigned long index; 33480257bb82SYan, Zheng unsigned long last_index; 33490257bb82SYan, Zheng struct page *page; 33500257bb82SYan, Zheng struct file_ra_state *ra; 33513b16a4e3SJosef Bacik gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 33520257bb82SYan, Zheng int nr = 0; 33530257bb82SYan, Zheng int ret = 0; 33540257bb82SYan, Zheng 33550257bb82SYan, Zheng if (!cluster->nr) 33560257bb82SYan, Zheng return 0; 33570257bb82SYan, Zheng 33580257bb82SYan, Zheng ra = kzalloc(sizeof(*ra), GFP_NOFS); 33590257bb82SYan, Zheng if (!ra) 33600257bb82SYan, Zheng return -ENOMEM; 33610257bb82SYan, Zheng 3362efa56464SYan, Zheng ret = prealloc_file_extent_cluster(inode, cluster); 33630257bb82SYan, Zheng if (ret) 3364efa56464SYan, Zheng goto out; 33650257bb82SYan, Zheng 33660257bb82SYan, Zheng file_ra_state_init(ra, inode->i_mapping); 33670257bb82SYan, Zheng 3368efa56464SYan, Zheng ret = setup_extent_mapping(inode, cluster->start - offset, 3369efa56464SYan, Zheng cluster->end - offset, cluster->start); 3370efa56464SYan, Zheng if (ret) 3371efa56464SYan, Zheng goto out; 3372efa56464SYan, Zheng 337309cbfeafSKirill A. Shutemov index = (cluster->start - offset) >> PAGE_SHIFT; 337409cbfeafSKirill A. Shutemov last_index = (cluster->end - offset) >> PAGE_SHIFT; 33750257bb82SYan, Zheng while (index <= last_index) { 33769f3db423SNikolay Borisov ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), 33779f3db423SNikolay Borisov PAGE_SIZE); 3378efa56464SYan, Zheng if (ret) 3379efa56464SYan, Zheng goto out; 3380efa56464SYan, Zheng 33810257bb82SYan, Zheng page = find_lock_page(inode->i_mapping, index); 33820257bb82SYan, Zheng if (!page) { 33830257bb82SYan, Zheng page_cache_sync_readahead(inode->i_mapping, 33840257bb82SYan, Zheng ra, NULL, index, 33850257bb82SYan, Zheng last_index + 1 - index); 3386a94733d0SJosef Bacik page = find_or_create_page(inode->i_mapping, index, 33873b16a4e3SJosef Bacik mask); 33880257bb82SYan, Zheng if (!page) { 3389691fa059SNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 339043b18595SQu Wenruo PAGE_SIZE, true); 339144db1216SFilipe Manana btrfs_delalloc_release_extents(BTRFS_I(inode), 33928702ba93SQu Wenruo PAGE_SIZE); 33930257bb82SYan, Zheng ret = -ENOMEM; 3394efa56464SYan, Zheng goto out; 33950257bb82SYan, Zheng } 33960257bb82SYan, Zheng } 33970257bb82SYan, Zheng 33980257bb82SYan, Zheng if (PageReadahead(page)) { 33990257bb82SYan, Zheng page_cache_async_readahead(inode->i_mapping, 34000257bb82SYan, Zheng ra, NULL, page, index, 34010257bb82SYan, Zheng last_index + 1 - index); 34020257bb82SYan, Zheng } 34030257bb82SYan, Zheng 34040257bb82SYan, Zheng if (!PageUptodate(page)) { 34050257bb82SYan, Zheng btrfs_readpage(NULL, page); 34060257bb82SYan, Zheng lock_page(page); 34070257bb82SYan, Zheng if (!PageUptodate(page)) { 34080257bb82SYan, Zheng unlock_page(page); 340909cbfeafSKirill A. Shutemov put_page(page); 3410691fa059SNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 341143b18595SQu Wenruo PAGE_SIZE, true); 34128b62f87bSJosef Bacik btrfs_delalloc_release_extents(BTRFS_I(inode), 34138702ba93SQu Wenruo PAGE_SIZE); 34140257bb82SYan, Zheng ret = -EIO; 3415efa56464SYan, Zheng goto out; 34160257bb82SYan, Zheng } 34170257bb82SYan, Zheng } 34180257bb82SYan, Zheng 34194eee4fa4SMiao Xie page_start = page_offset(page); 342009cbfeafSKirill A. Shutemov page_end = page_start + PAGE_SIZE - 1; 34210257bb82SYan, Zheng 3422d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end); 34230257bb82SYan, Zheng 34240257bb82SYan, Zheng set_page_extent_mapped(page); 34250257bb82SYan, Zheng 34260257bb82SYan, Zheng if (nr < cluster->nr && 34270257bb82SYan, Zheng page_start + offset == cluster->boundary[nr]) { 34280257bb82SYan, Zheng set_extent_bits(&BTRFS_I(inode)->io_tree, 34290257bb82SYan, Zheng page_start, page_end, 3430ceeb0ae7SDavid Sterba EXTENT_BOUNDARY); 34310257bb82SYan, Zheng nr++; 34320257bb82SYan, Zheng } 34330257bb82SYan, Zheng 3434765f3cebSNikolay Borisov ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0, 3435330a5827SNikolay Borisov NULL); 3436765f3cebSNikolay Borisov if (ret) { 3437765f3cebSNikolay Borisov unlock_page(page); 3438765f3cebSNikolay Borisov put_page(page); 3439765f3cebSNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 344043b18595SQu Wenruo PAGE_SIZE, true); 3441765f3cebSNikolay Borisov btrfs_delalloc_release_extents(BTRFS_I(inode), 34428702ba93SQu Wenruo PAGE_SIZE); 3443765f3cebSNikolay Borisov 3444765f3cebSNikolay Borisov clear_extent_bits(&BTRFS_I(inode)->io_tree, 3445765f3cebSNikolay Borisov page_start, page_end, 3446765f3cebSNikolay Borisov EXTENT_LOCKED | EXTENT_BOUNDARY); 3447765f3cebSNikolay Borisov goto out; 3448765f3cebSNikolay Borisov 3449765f3cebSNikolay Borisov } 34500257bb82SYan, Zheng set_page_dirty(page); 34510257bb82SYan, Zheng 34520257bb82SYan, Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 3453d0082371SJeff Mahoney page_start, page_end); 34540257bb82SYan, Zheng unlock_page(page); 345509cbfeafSKirill A. Shutemov put_page(page); 34560257bb82SYan, Zheng 34570257bb82SYan, Zheng index++; 34588702ba93SQu Wenruo btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); 3459efa56464SYan, Zheng balance_dirty_pages_ratelimited(inode->i_mapping); 34602ff7e61eSJeff Mahoney btrfs_throttle(fs_info); 34617f913c7cSQu Wenruo if (btrfs_should_cancel_balance(fs_info)) { 34627f913c7cSQu Wenruo ret = -ECANCELED; 34637f913c7cSQu Wenruo goto out; 34647f913c7cSQu Wenruo } 34650257bb82SYan, Zheng } 34660257bb82SYan, Zheng WARN_ON(nr != cluster->nr); 3467efa56464SYan, Zheng out: 34680257bb82SYan, Zheng kfree(ra); 34690257bb82SYan, Zheng return ret; 34700257bb82SYan, Zheng } 34710257bb82SYan, Zheng 34720257bb82SYan, Zheng static noinline_for_stack 34730257bb82SYan, Zheng int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key, 34740257bb82SYan, Zheng struct file_extent_cluster *cluster) 34750257bb82SYan, Zheng { 34760257bb82SYan, Zheng int ret; 34770257bb82SYan, Zheng 34780257bb82SYan, Zheng if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) { 34790257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 34800257bb82SYan, Zheng if (ret) 34810257bb82SYan, Zheng return ret; 34820257bb82SYan, Zheng cluster->nr = 0; 34830257bb82SYan, Zheng } 34840257bb82SYan, Zheng 34850257bb82SYan, Zheng if (!cluster->nr) 34860257bb82SYan, Zheng cluster->start = extent_key->objectid; 34870257bb82SYan, Zheng else 34880257bb82SYan, Zheng BUG_ON(cluster->nr >= MAX_EXTENTS); 34890257bb82SYan, Zheng cluster->end = extent_key->objectid + extent_key->offset - 1; 34900257bb82SYan, Zheng cluster->boundary[cluster->nr] = extent_key->objectid; 34910257bb82SYan, Zheng cluster->nr++; 34920257bb82SYan, Zheng 34930257bb82SYan, Zheng if (cluster->nr >= MAX_EXTENTS) { 34940257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 34950257bb82SYan, Zheng if (ret) 34960257bb82SYan, Zheng return ret; 34970257bb82SYan, Zheng cluster->nr = 0; 34980257bb82SYan, Zheng } 34990257bb82SYan, Zheng return 0; 35005d4f98a2SYan Zheng } 35015d4f98a2SYan Zheng 35025d4f98a2SYan Zheng /* 35035d4f98a2SYan Zheng * helper to add a tree block to the list. 35045d4f98a2SYan Zheng * the major work is getting the generation and level of the block 35055d4f98a2SYan Zheng */ 35065d4f98a2SYan Zheng static int add_tree_block(struct reloc_control *rc, 35075d4f98a2SYan Zheng struct btrfs_key *extent_key, 35085d4f98a2SYan Zheng struct btrfs_path *path, 35095d4f98a2SYan Zheng struct rb_root *blocks) 35105d4f98a2SYan Zheng { 35115d4f98a2SYan Zheng struct extent_buffer *eb; 35125d4f98a2SYan Zheng struct btrfs_extent_item *ei; 35135d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 35145d4f98a2SYan Zheng struct tree_block *block; 35155d4f98a2SYan Zheng struct rb_node *rb_node; 35165d4f98a2SYan Zheng u32 item_size; 35175d4f98a2SYan Zheng int level = -1; 35187fdf4b60SWang Shilong u64 generation; 35195d4f98a2SYan Zheng 35205d4f98a2SYan Zheng eb = path->nodes[0]; 35215d4f98a2SYan Zheng item_size = btrfs_item_size_nr(eb, path->slots[0]); 35225d4f98a2SYan Zheng 35233173a18fSJosef Bacik if (extent_key->type == BTRFS_METADATA_ITEM_KEY || 35243173a18fSJosef Bacik item_size >= sizeof(*ei) + sizeof(*bi)) { 35255d4f98a2SYan Zheng ei = btrfs_item_ptr(eb, path->slots[0], 35265d4f98a2SYan Zheng struct btrfs_extent_item); 35273173a18fSJosef Bacik if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) { 35285d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 35295d4f98a2SYan Zheng level = btrfs_tree_block_level(eb, bi); 35305d4f98a2SYan Zheng } else { 35313173a18fSJosef Bacik level = (int)extent_key->offset; 35323173a18fSJosef Bacik } 35333173a18fSJosef Bacik generation = btrfs_extent_generation(eb, ei); 35346d8ff4e4SDavid Sterba } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) { 3535ba3c2b19SNikolay Borisov btrfs_print_v0_err(eb->fs_info); 3536ba3c2b19SNikolay Borisov btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL); 3537ba3c2b19SNikolay Borisov return -EINVAL; 35383173a18fSJosef Bacik } else { 35395d4f98a2SYan Zheng BUG(); 35405d4f98a2SYan Zheng } 35415d4f98a2SYan Zheng 3542b3b4aa74SDavid Sterba btrfs_release_path(path); 35435d4f98a2SYan Zheng 35445d4f98a2SYan Zheng BUG_ON(level == -1); 35455d4f98a2SYan Zheng 35465d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 35475d4f98a2SYan Zheng if (!block) 35485d4f98a2SYan Zheng return -ENOMEM; 35495d4f98a2SYan Zheng 35505d4f98a2SYan Zheng block->bytenr = extent_key->objectid; 3551da17066cSJeff Mahoney block->key.objectid = rc->extent_root->fs_info->nodesize; 35525d4f98a2SYan Zheng block->key.offset = generation; 35535d4f98a2SYan Zheng block->level = level; 35545d4f98a2SYan Zheng block->key_ready = 0; 35555d4f98a2SYan Zheng 35565d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, &block->rb_node); 355743c04fb1SJeff Mahoney if (rb_node) 355843c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, block->bytenr); 35595d4f98a2SYan Zheng 35605d4f98a2SYan Zheng return 0; 35615d4f98a2SYan Zheng } 35625d4f98a2SYan Zheng 35635d4f98a2SYan Zheng /* 35645d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY 35655d4f98a2SYan Zheng */ 35665d4f98a2SYan Zheng static int __add_tree_block(struct reloc_control *rc, 35675d4f98a2SYan Zheng u64 bytenr, u32 blocksize, 35685d4f98a2SYan Zheng struct rb_root *blocks) 35695d4f98a2SYan Zheng { 35700b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 35715d4f98a2SYan Zheng struct btrfs_path *path; 35725d4f98a2SYan Zheng struct btrfs_key key; 35735d4f98a2SYan Zheng int ret; 35740b246afaSJeff Mahoney bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA); 35755d4f98a2SYan Zheng 35767476dfdaSDavid Sterba if (tree_block_processed(bytenr, rc)) 35775d4f98a2SYan Zheng return 0; 35785d4f98a2SYan Zheng 35795d4f98a2SYan Zheng if (tree_search(blocks, bytenr)) 35805d4f98a2SYan Zheng return 0; 35815d4f98a2SYan Zheng 35825d4f98a2SYan Zheng path = btrfs_alloc_path(); 35835d4f98a2SYan Zheng if (!path) 35845d4f98a2SYan Zheng return -ENOMEM; 3585aee68ee5SJosef Bacik again: 35865d4f98a2SYan Zheng key.objectid = bytenr; 3587aee68ee5SJosef Bacik if (skinny) { 3588aee68ee5SJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 3589aee68ee5SJosef Bacik key.offset = (u64)-1; 3590aee68ee5SJosef Bacik } else { 35915d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 35925d4f98a2SYan Zheng key.offset = blocksize; 3593aee68ee5SJosef Bacik } 35945d4f98a2SYan Zheng 35955d4f98a2SYan Zheng path->search_commit_root = 1; 35965d4f98a2SYan Zheng path->skip_locking = 1; 35975d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0); 35985d4f98a2SYan Zheng if (ret < 0) 35995d4f98a2SYan Zheng goto out; 36005d4f98a2SYan Zheng 3601aee68ee5SJosef Bacik if (ret > 0 && skinny) { 3602aee68ee5SJosef Bacik if (path->slots[0]) { 3603aee68ee5SJosef Bacik path->slots[0]--; 3604aee68ee5SJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, 3605aee68ee5SJosef Bacik path->slots[0]); 36063173a18fSJosef Bacik if (key.objectid == bytenr && 3607aee68ee5SJosef Bacik (key.type == BTRFS_METADATA_ITEM_KEY || 3608aee68ee5SJosef Bacik (key.type == BTRFS_EXTENT_ITEM_KEY && 3609aee68ee5SJosef Bacik key.offset == blocksize))) 36103173a18fSJosef Bacik ret = 0; 36113173a18fSJosef Bacik } 3612aee68ee5SJosef Bacik 3613aee68ee5SJosef Bacik if (ret) { 3614aee68ee5SJosef Bacik skinny = false; 3615aee68ee5SJosef Bacik btrfs_release_path(path); 3616aee68ee5SJosef Bacik goto again; 3617aee68ee5SJosef Bacik } 3618aee68ee5SJosef Bacik } 3619cdccee99SLiu Bo if (ret) { 3620cdccee99SLiu Bo ASSERT(ret == 1); 3621cdccee99SLiu Bo btrfs_print_leaf(path->nodes[0]); 3622cdccee99SLiu Bo btrfs_err(fs_info, 3623cdccee99SLiu Bo "tree block extent item (%llu) is not found in extent tree", 3624cdccee99SLiu Bo bytenr); 3625cdccee99SLiu Bo WARN_ON(1); 3626cdccee99SLiu Bo ret = -EINVAL; 3627cdccee99SLiu Bo goto out; 3628cdccee99SLiu Bo } 36293173a18fSJosef Bacik 36305d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, blocks); 36315d4f98a2SYan Zheng out: 36325d4f98a2SYan Zheng btrfs_free_path(path); 36335d4f98a2SYan Zheng return ret; 36345d4f98a2SYan Zheng } 36355d4f98a2SYan Zheng 36360af3d00bSJosef Bacik static int delete_block_group_cache(struct btrfs_fs_info *fs_info, 363732da5386SDavid Sterba struct btrfs_block_group *block_group, 36381bbc621eSChris Mason struct inode *inode, 36391bbc621eSChris Mason u64 ino) 36400af3d00bSJosef Bacik { 36410af3d00bSJosef Bacik struct btrfs_key key; 36420af3d00bSJosef Bacik struct btrfs_root *root = fs_info->tree_root; 36430af3d00bSJosef Bacik struct btrfs_trans_handle *trans; 36440af3d00bSJosef Bacik int ret = 0; 36450af3d00bSJosef Bacik 36460af3d00bSJosef Bacik if (inode) 36470af3d00bSJosef Bacik goto truncate; 36480af3d00bSJosef Bacik 36490af3d00bSJosef Bacik key.objectid = ino; 36500af3d00bSJosef Bacik key.type = BTRFS_INODE_ITEM_KEY; 36510af3d00bSJosef Bacik key.offset = 0; 36520af3d00bSJosef Bacik 36534c66e0d4SDavid Sterba inode = btrfs_iget(fs_info->sb, &key, root); 36542e19f1f9SAl Viro if (IS_ERR(inode)) 36550af3d00bSJosef Bacik return -ENOENT; 36560af3d00bSJosef Bacik 36570af3d00bSJosef Bacik truncate: 36582ff7e61eSJeff Mahoney ret = btrfs_check_trunc_cache_free_space(fs_info, 36597b61cd92SMiao Xie &fs_info->global_block_rsv); 36607b61cd92SMiao Xie if (ret) 36617b61cd92SMiao Xie goto out; 36627b61cd92SMiao Xie 36637a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 36640af3d00bSJosef Bacik if (IS_ERR(trans)) { 36653612b495STsutomu Itoh ret = PTR_ERR(trans); 36660af3d00bSJosef Bacik goto out; 36670af3d00bSJosef Bacik } 36680af3d00bSJosef Bacik 366977ab86bfSJeff Mahoney ret = btrfs_truncate_free_space_cache(trans, block_group, inode); 36700af3d00bSJosef Bacik 36713a45bb20SJeff Mahoney btrfs_end_transaction(trans); 36722ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 36730af3d00bSJosef Bacik out: 36740af3d00bSJosef Bacik iput(inode); 36750af3d00bSJosef Bacik return ret; 36760af3d00bSJosef Bacik } 36770af3d00bSJosef Bacik 36785d4f98a2SYan Zheng /* 367919b546d7SQu Wenruo * Locate the free space cache EXTENT_DATA in root tree leaf and delete the 368019b546d7SQu Wenruo * cache inode, to avoid free space cache data extent blocking data relocation. 36815d4f98a2SYan Zheng */ 368219b546d7SQu Wenruo static int delete_v1_space_cache(struct extent_buffer *leaf, 368319b546d7SQu Wenruo struct btrfs_block_group *block_group, 368419b546d7SQu Wenruo u64 data_bytenr) 36855d4f98a2SYan Zheng { 368619b546d7SQu Wenruo u64 space_cache_ino; 368719b546d7SQu Wenruo struct btrfs_file_extent_item *ei; 36885d4f98a2SYan Zheng struct btrfs_key key; 368919b546d7SQu Wenruo bool found = false; 369019b546d7SQu Wenruo int i; 36915d4f98a2SYan Zheng int ret; 36925d4f98a2SYan Zheng 369319b546d7SQu Wenruo if (btrfs_header_owner(leaf) != BTRFS_ROOT_TREE_OBJECTID) 369419b546d7SQu Wenruo return 0; 36955d4f98a2SYan Zheng 369619b546d7SQu Wenruo for (i = 0; i < btrfs_header_nritems(leaf); i++) { 369719b546d7SQu Wenruo btrfs_item_key_to_cpu(leaf, &key, i); 369819b546d7SQu Wenruo if (key.type != BTRFS_EXTENT_DATA_KEY) 369919b546d7SQu Wenruo continue; 370019b546d7SQu Wenruo ei = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 370119b546d7SQu Wenruo if (btrfs_file_extent_type(leaf, ei) == BTRFS_FILE_EXTENT_REG && 370219b546d7SQu Wenruo btrfs_file_extent_disk_bytenr(leaf, ei) == data_bytenr) { 370319b546d7SQu Wenruo found = true; 370419b546d7SQu Wenruo space_cache_ino = key.objectid; 370519b546d7SQu Wenruo break; 370619b546d7SQu Wenruo } 370719b546d7SQu Wenruo } 370819b546d7SQu Wenruo if (!found) 370919b546d7SQu Wenruo return -ENOENT; 371019b546d7SQu Wenruo ret = delete_block_group_cache(leaf->fs_info, block_group, NULL, 371119b546d7SQu Wenruo space_cache_ino); 37120af3d00bSJosef Bacik return ret; 37135d4f98a2SYan Zheng } 37145d4f98a2SYan Zheng 37155d4f98a2SYan Zheng /* 37162c016dc2SLiu Bo * helper to find all tree blocks that reference a given data extent 37175d4f98a2SYan Zheng */ 37185d4f98a2SYan Zheng static noinline_for_stack 37195d4f98a2SYan Zheng int add_data_references(struct reloc_control *rc, 37205d4f98a2SYan Zheng struct btrfs_key *extent_key, 37215d4f98a2SYan Zheng struct btrfs_path *path, 37225d4f98a2SYan Zheng struct rb_root *blocks) 37235d4f98a2SYan Zheng { 372419b546d7SQu Wenruo struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 372519b546d7SQu Wenruo struct ulist *leaves = NULL; 372619b546d7SQu Wenruo struct ulist_iterator leaf_uiter; 372719b546d7SQu Wenruo struct ulist_node *ref_node = NULL; 372819b546d7SQu Wenruo const u32 blocksize = fs_info->nodesize; 3729647f63bdSFilipe David Borba Manana int ret = 0; 37305d4f98a2SYan Zheng 3731b3b4aa74SDavid Sterba btrfs_release_path(path); 373219b546d7SQu Wenruo ret = btrfs_find_all_leafs(NULL, fs_info, extent_key->objectid, 373319b546d7SQu Wenruo 0, &leaves, NULL, true); 373419b546d7SQu Wenruo if (ret < 0) 373519b546d7SQu Wenruo return ret; 373619b546d7SQu Wenruo 373719b546d7SQu Wenruo ULIST_ITER_INIT(&leaf_uiter); 373819b546d7SQu Wenruo while ((ref_node = ulist_next(leaves, &leaf_uiter))) { 373919b546d7SQu Wenruo struct extent_buffer *eb; 374019b546d7SQu Wenruo 374119b546d7SQu Wenruo eb = read_tree_block(fs_info, ref_node->val, 0, 0, NULL); 374219b546d7SQu Wenruo if (IS_ERR(eb)) { 374319b546d7SQu Wenruo ret = PTR_ERR(eb); 374419b546d7SQu Wenruo break; 374519b546d7SQu Wenruo } 374619b546d7SQu Wenruo ret = delete_v1_space_cache(eb, rc->block_group, 374719b546d7SQu Wenruo extent_key->objectid); 374819b546d7SQu Wenruo free_extent_buffer(eb); 374919b546d7SQu Wenruo if (ret < 0) 375019b546d7SQu Wenruo break; 375119b546d7SQu Wenruo ret = __add_tree_block(rc, ref_node->val, blocksize, blocks); 375219b546d7SQu Wenruo if (ret < 0) 375319b546d7SQu Wenruo break; 375419b546d7SQu Wenruo } 375519b546d7SQu Wenruo if (ret < 0) 37565d4f98a2SYan Zheng free_block_list(blocks); 375719b546d7SQu Wenruo ulist_free(leaves); 375819b546d7SQu Wenruo return ret; 37595d4f98a2SYan Zheng } 37605d4f98a2SYan Zheng 37615d4f98a2SYan Zheng /* 37622c016dc2SLiu Bo * helper to find next unprocessed extent 37635d4f98a2SYan Zheng */ 37645d4f98a2SYan Zheng static noinline_for_stack 3765147d256eSZhaolei int find_next_extent(struct reloc_control *rc, struct btrfs_path *path, 37663fd0a558SYan, Zheng struct btrfs_key *extent_key) 37675d4f98a2SYan Zheng { 37680b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 37695d4f98a2SYan Zheng struct btrfs_key key; 37705d4f98a2SYan Zheng struct extent_buffer *leaf; 37715d4f98a2SYan Zheng u64 start, end, last; 37725d4f98a2SYan Zheng int ret; 37735d4f98a2SYan Zheng 3774b3470b5dSDavid Sterba last = rc->block_group->start + rc->block_group->length; 37755d4f98a2SYan Zheng while (1) { 37765d4f98a2SYan Zheng cond_resched(); 37775d4f98a2SYan Zheng if (rc->search_start >= last) { 37785d4f98a2SYan Zheng ret = 1; 37795d4f98a2SYan Zheng break; 37805d4f98a2SYan Zheng } 37815d4f98a2SYan Zheng 37825d4f98a2SYan Zheng key.objectid = rc->search_start; 37835d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 37845d4f98a2SYan Zheng key.offset = 0; 37855d4f98a2SYan Zheng 37865d4f98a2SYan Zheng path->search_commit_root = 1; 37875d4f98a2SYan Zheng path->skip_locking = 1; 37885d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 37895d4f98a2SYan Zheng 0, 0); 37905d4f98a2SYan Zheng if (ret < 0) 37915d4f98a2SYan Zheng break; 37925d4f98a2SYan Zheng next: 37935d4f98a2SYan Zheng leaf = path->nodes[0]; 37945d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(leaf)) { 37955d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 37965d4f98a2SYan Zheng if (ret != 0) 37975d4f98a2SYan Zheng break; 37985d4f98a2SYan Zheng leaf = path->nodes[0]; 37995d4f98a2SYan Zheng } 38005d4f98a2SYan Zheng 38015d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 38025d4f98a2SYan Zheng if (key.objectid >= last) { 38035d4f98a2SYan Zheng ret = 1; 38045d4f98a2SYan Zheng break; 38055d4f98a2SYan Zheng } 38065d4f98a2SYan Zheng 38073173a18fSJosef Bacik if (key.type != BTRFS_EXTENT_ITEM_KEY && 38083173a18fSJosef Bacik key.type != BTRFS_METADATA_ITEM_KEY) { 38093173a18fSJosef Bacik path->slots[0]++; 38103173a18fSJosef Bacik goto next; 38113173a18fSJosef Bacik } 38123173a18fSJosef Bacik 38133173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 38145d4f98a2SYan Zheng key.objectid + key.offset <= rc->search_start) { 38155d4f98a2SYan Zheng path->slots[0]++; 38165d4f98a2SYan Zheng goto next; 38175d4f98a2SYan Zheng } 38185d4f98a2SYan Zheng 38193173a18fSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 38200b246afaSJeff Mahoney key.objectid + fs_info->nodesize <= 38213173a18fSJosef Bacik rc->search_start) { 38223173a18fSJosef Bacik path->slots[0]++; 38233173a18fSJosef Bacik goto next; 38243173a18fSJosef Bacik } 38253173a18fSJosef Bacik 38265d4f98a2SYan Zheng ret = find_first_extent_bit(&rc->processed_blocks, 38275d4f98a2SYan Zheng key.objectid, &start, &end, 3828e6138876SJosef Bacik EXTENT_DIRTY, NULL); 38295d4f98a2SYan Zheng 38305d4f98a2SYan Zheng if (ret == 0 && start <= key.objectid) { 3831b3b4aa74SDavid Sterba btrfs_release_path(path); 38325d4f98a2SYan Zheng rc->search_start = end + 1; 38335d4f98a2SYan Zheng } else { 38343173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) 38355d4f98a2SYan Zheng rc->search_start = key.objectid + key.offset; 38363173a18fSJosef Bacik else 38373173a18fSJosef Bacik rc->search_start = key.objectid + 38380b246afaSJeff Mahoney fs_info->nodesize; 38393fd0a558SYan, Zheng memcpy(extent_key, &key, sizeof(key)); 38405d4f98a2SYan Zheng return 0; 38415d4f98a2SYan Zheng } 38425d4f98a2SYan Zheng } 3843b3b4aa74SDavid Sterba btrfs_release_path(path); 38445d4f98a2SYan Zheng return ret; 38455d4f98a2SYan Zheng } 38465d4f98a2SYan Zheng 38475d4f98a2SYan Zheng static void set_reloc_control(struct reloc_control *rc) 38485d4f98a2SYan Zheng { 38495d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 38507585717fSChris Mason 38517585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 38525d4f98a2SYan Zheng fs_info->reloc_ctl = rc; 38537585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 38545d4f98a2SYan Zheng } 38555d4f98a2SYan Zheng 38565d4f98a2SYan Zheng static void unset_reloc_control(struct reloc_control *rc) 38575d4f98a2SYan Zheng { 38585d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 38597585717fSChris Mason 38607585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 38615d4f98a2SYan Zheng fs_info->reloc_ctl = NULL; 38627585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 38635d4f98a2SYan Zheng } 38645d4f98a2SYan Zheng 38655d4f98a2SYan Zheng static int check_extent_flags(u64 flags) 38665d4f98a2SYan Zheng { 38675d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 38685d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 38695d4f98a2SYan Zheng return 1; 38705d4f98a2SYan Zheng if (!(flags & BTRFS_EXTENT_FLAG_DATA) && 38715d4f98a2SYan Zheng !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 38725d4f98a2SYan Zheng return 1; 38735d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 38745d4f98a2SYan Zheng (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) 38755d4f98a2SYan Zheng return 1; 38765d4f98a2SYan Zheng return 0; 38775d4f98a2SYan Zheng } 38785d4f98a2SYan Zheng 38793fd0a558SYan, Zheng static noinline_for_stack 38803fd0a558SYan, Zheng int prepare_to_relocate(struct reloc_control *rc) 38813fd0a558SYan, Zheng { 38823fd0a558SYan, Zheng struct btrfs_trans_handle *trans; 3883ac2fabacSJosef Bacik int ret; 38843fd0a558SYan, Zheng 38852ff7e61eSJeff Mahoney rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info, 388666d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 38873fd0a558SYan, Zheng if (!rc->block_rsv) 38883fd0a558SYan, Zheng return -ENOMEM; 38893fd0a558SYan, Zheng 38903fd0a558SYan, Zheng memset(&rc->cluster, 0, sizeof(rc->cluster)); 3891b3470b5dSDavid Sterba rc->search_start = rc->block_group->start; 38923fd0a558SYan, Zheng rc->extents_found = 0; 38933fd0a558SYan, Zheng rc->nodes_relocated = 0; 38943fd0a558SYan, Zheng rc->merging_rsv_size = 0; 38950647bf56SWang Shilong rc->reserved_bytes = 0; 3896da17066cSJeff Mahoney rc->block_rsv->size = rc->extent_root->fs_info->nodesize * 38970647bf56SWang Shilong RELOCATION_RESERVED_NODES; 3898ac2fabacSJosef Bacik ret = btrfs_block_rsv_refill(rc->extent_root, 3899ac2fabacSJosef Bacik rc->block_rsv, rc->block_rsv->size, 3900ac2fabacSJosef Bacik BTRFS_RESERVE_FLUSH_ALL); 3901ac2fabacSJosef Bacik if (ret) 3902ac2fabacSJosef Bacik return ret; 39033fd0a558SYan, Zheng 39043fd0a558SYan, Zheng rc->create_reloc_tree = 1; 39053fd0a558SYan, Zheng set_reloc_control(rc); 39063fd0a558SYan, Zheng 39077a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 390828818947SLiu Bo if (IS_ERR(trans)) { 390928818947SLiu Bo unset_reloc_control(rc); 391028818947SLiu Bo /* 391128818947SLiu Bo * extent tree is not a ref_cow tree and has no reloc_root to 391228818947SLiu Bo * cleanup. And callers are responsible to free the above 391328818947SLiu Bo * block rsv. 391428818947SLiu Bo */ 391528818947SLiu Bo return PTR_ERR(trans); 391628818947SLiu Bo } 39173a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 39183fd0a558SYan, Zheng return 0; 39193fd0a558SYan, Zheng } 392076dda93cSYan, Zheng 39215d4f98a2SYan Zheng static noinline_for_stack int relocate_block_group(struct reloc_control *rc) 39225d4f98a2SYan Zheng { 39232ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 39245d4f98a2SYan Zheng struct rb_root blocks = RB_ROOT; 39255d4f98a2SYan Zheng struct btrfs_key key; 39265d4f98a2SYan Zheng struct btrfs_trans_handle *trans = NULL; 39275d4f98a2SYan Zheng struct btrfs_path *path; 39285d4f98a2SYan Zheng struct btrfs_extent_item *ei; 39295d4f98a2SYan Zheng u64 flags; 39305d4f98a2SYan Zheng u32 item_size; 39315d4f98a2SYan Zheng int ret; 39325d4f98a2SYan Zheng int err = 0; 3933c87f08caSChris Mason int progress = 0; 39345d4f98a2SYan Zheng 39355d4f98a2SYan Zheng path = btrfs_alloc_path(); 39363fd0a558SYan, Zheng if (!path) 39375d4f98a2SYan Zheng return -ENOMEM; 3938e4058b54SDavid Sterba path->reada = READA_FORWARD; 39393fd0a558SYan, Zheng 39403fd0a558SYan, Zheng ret = prepare_to_relocate(rc); 39413fd0a558SYan, Zheng if (ret) { 39423fd0a558SYan, Zheng err = ret; 39433fd0a558SYan, Zheng goto out_free; 39442423fdfbSJiri Slaby } 39455d4f98a2SYan Zheng 39465d4f98a2SYan Zheng while (1) { 39470647bf56SWang Shilong rc->reserved_bytes = 0; 39480647bf56SWang Shilong ret = btrfs_block_rsv_refill(rc->extent_root, 39490647bf56SWang Shilong rc->block_rsv, rc->block_rsv->size, 39500647bf56SWang Shilong BTRFS_RESERVE_FLUSH_ALL); 39510647bf56SWang Shilong if (ret) { 39520647bf56SWang Shilong err = ret; 39530647bf56SWang Shilong break; 39540647bf56SWang Shilong } 3955c87f08caSChris Mason progress++; 3956a22285a6SYan, Zheng trans = btrfs_start_transaction(rc->extent_root, 0); 39570f788c58SLiu Bo if (IS_ERR(trans)) { 39580f788c58SLiu Bo err = PTR_ERR(trans); 39590f788c58SLiu Bo trans = NULL; 39600f788c58SLiu Bo break; 39610f788c58SLiu Bo } 3962c87f08caSChris Mason restart: 39633fd0a558SYan, Zheng if (update_backref_cache(trans, &rc->backref_cache)) { 39643a45bb20SJeff Mahoney btrfs_end_transaction(trans); 396542a657f5SPan Bian trans = NULL; 39663fd0a558SYan, Zheng continue; 39673fd0a558SYan, Zheng } 39683fd0a558SYan, Zheng 3969147d256eSZhaolei ret = find_next_extent(rc, path, &key); 39705d4f98a2SYan Zheng if (ret < 0) 39715d4f98a2SYan Zheng err = ret; 39725d4f98a2SYan Zheng if (ret != 0) 39735d4f98a2SYan Zheng break; 39745d4f98a2SYan Zheng 39755d4f98a2SYan Zheng rc->extents_found++; 39765d4f98a2SYan Zheng 39775d4f98a2SYan Zheng ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 39785d4f98a2SYan Zheng struct btrfs_extent_item); 39793fd0a558SYan, Zheng item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); 39805d4f98a2SYan Zheng if (item_size >= sizeof(*ei)) { 39815d4f98a2SYan Zheng flags = btrfs_extent_flags(path->nodes[0], ei); 39825d4f98a2SYan Zheng ret = check_extent_flags(flags); 39835d4f98a2SYan Zheng BUG_ON(ret); 39846d8ff4e4SDavid Sterba } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) { 3985ba3c2b19SNikolay Borisov err = -EINVAL; 3986ba3c2b19SNikolay Borisov btrfs_print_v0_err(trans->fs_info); 3987ba3c2b19SNikolay Borisov btrfs_abort_transaction(trans, err); 3988ba3c2b19SNikolay Borisov break; 39895d4f98a2SYan Zheng } else { 39905d4f98a2SYan Zheng BUG(); 39915d4f98a2SYan Zheng } 39925d4f98a2SYan Zheng 39935d4f98a2SYan Zheng if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 39945d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, &blocks); 39955d4f98a2SYan Zheng } else if (rc->stage == UPDATE_DATA_PTRS && 39965d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 39975d4f98a2SYan Zheng ret = add_data_references(rc, &key, path, &blocks); 39985d4f98a2SYan Zheng } else { 3999b3b4aa74SDavid Sterba btrfs_release_path(path); 40005d4f98a2SYan Zheng ret = 0; 40015d4f98a2SYan Zheng } 40025d4f98a2SYan Zheng if (ret < 0) { 40033fd0a558SYan, Zheng err = ret; 40045d4f98a2SYan Zheng break; 40055d4f98a2SYan Zheng } 40065d4f98a2SYan Zheng 40075d4f98a2SYan Zheng if (!RB_EMPTY_ROOT(&blocks)) { 40085d4f98a2SYan Zheng ret = relocate_tree_blocks(trans, rc, &blocks); 40095d4f98a2SYan Zheng if (ret < 0) { 40101708cc57SWang Shilong /* 40111708cc57SWang Shilong * if we fail to relocate tree blocks, force to update 40121708cc57SWang Shilong * backref cache when committing transaction. 40131708cc57SWang Shilong */ 40141708cc57SWang Shilong rc->backref_cache.last_trans = trans->transid - 1; 40151708cc57SWang Shilong 40163fd0a558SYan, Zheng if (ret != -EAGAIN) { 40175d4f98a2SYan Zheng err = ret; 40185d4f98a2SYan Zheng break; 40195d4f98a2SYan Zheng } 40203fd0a558SYan, Zheng rc->extents_found--; 40213fd0a558SYan, Zheng rc->search_start = key.objectid; 40223fd0a558SYan, Zheng } 40235d4f98a2SYan Zheng } 40245d4f98a2SYan Zheng 40253a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 40262ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 40273fd0a558SYan, Zheng trans = NULL; 40285d4f98a2SYan Zheng 40295d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && 40305d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 40315d4f98a2SYan Zheng rc->found_file_extent = 1; 40320257bb82SYan, Zheng ret = relocate_data_extent(rc->data_inode, 40333fd0a558SYan, Zheng &key, &rc->cluster); 40345d4f98a2SYan Zheng if (ret < 0) { 40355d4f98a2SYan Zheng err = ret; 40365d4f98a2SYan Zheng break; 40375d4f98a2SYan Zheng } 40385d4f98a2SYan Zheng } 4039f31ea088SQu Wenruo if (btrfs_should_cancel_balance(fs_info)) { 4040f31ea088SQu Wenruo err = -ECANCELED; 4041f31ea088SQu Wenruo break; 4042f31ea088SQu Wenruo } 40435d4f98a2SYan Zheng } 4044c87f08caSChris Mason if (trans && progress && err == -ENOSPC) { 404543a7e99dSNikolay Borisov ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags); 40469689457bSShilong Wang if (ret == 1) { 4047c87f08caSChris Mason err = 0; 4048c87f08caSChris Mason progress = 0; 4049c87f08caSChris Mason goto restart; 4050c87f08caSChris Mason } 4051c87f08caSChris Mason } 40523fd0a558SYan, Zheng 4053b3b4aa74SDavid Sterba btrfs_release_path(path); 405491166212SDavid Sterba clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY); 40555d4f98a2SYan Zheng 40565d4f98a2SYan Zheng if (trans) { 40573a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 40582ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 40595d4f98a2SYan Zheng } 40605d4f98a2SYan Zheng 40610257bb82SYan, Zheng if (!err) { 40623fd0a558SYan, Zheng ret = relocate_file_extent_cluster(rc->data_inode, 40633fd0a558SYan, Zheng &rc->cluster); 40640257bb82SYan, Zheng if (ret < 0) 40650257bb82SYan, Zheng err = ret; 40660257bb82SYan, Zheng } 40670257bb82SYan, Zheng 40683fd0a558SYan, Zheng rc->create_reloc_tree = 0; 40693fd0a558SYan, Zheng set_reloc_control(rc); 40700257bb82SYan, Zheng 40713fd0a558SYan, Zheng backref_cache_cleanup(&rc->backref_cache); 407263f018beSNikolay Borisov btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL); 40735d4f98a2SYan Zheng 40747f913c7cSQu Wenruo /* 40757f913c7cSQu Wenruo * Even in the case when the relocation is cancelled, we should all go 40767f913c7cSQu Wenruo * through prepare_to_merge() and merge_reloc_roots(). 40777f913c7cSQu Wenruo * 40787f913c7cSQu Wenruo * For error (including cancelled balance), prepare_to_merge() will 40797f913c7cSQu Wenruo * mark all reloc trees orphan, then queue them for cleanup in 40807f913c7cSQu Wenruo * merge_reloc_roots() 40817f913c7cSQu Wenruo */ 40823fd0a558SYan, Zheng err = prepare_to_merge(rc, err); 40835d4f98a2SYan Zheng 40845d4f98a2SYan Zheng merge_reloc_roots(rc); 40855d4f98a2SYan Zheng 40863fd0a558SYan, Zheng rc->merge_reloc_tree = 0; 40875d4f98a2SYan Zheng unset_reloc_control(rc); 408863f018beSNikolay Borisov btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL); 40895d4f98a2SYan Zheng 40905d4f98a2SYan Zheng /* get rid of pinned extents */ 40917a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 409262b99540SQu Wenruo if (IS_ERR(trans)) { 40933612b495STsutomu Itoh err = PTR_ERR(trans); 409462b99540SQu Wenruo goto out_free; 409562b99540SQu Wenruo } 40963a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 40976217b0faSJosef Bacik out_free: 4098d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 4099d2311e69SQu Wenruo if (ret < 0 && !err) 4100d2311e69SQu Wenruo err = ret; 41012ff7e61eSJeff Mahoney btrfs_free_block_rsv(fs_info, rc->block_rsv); 41023fd0a558SYan, Zheng btrfs_free_path(path); 41035d4f98a2SYan Zheng return err; 41045d4f98a2SYan Zheng } 41055d4f98a2SYan Zheng 41065d4f98a2SYan Zheng static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 41070257bb82SYan, Zheng struct btrfs_root *root, u64 objectid) 41085d4f98a2SYan Zheng { 41095d4f98a2SYan Zheng struct btrfs_path *path; 41105d4f98a2SYan Zheng struct btrfs_inode_item *item; 41115d4f98a2SYan Zheng struct extent_buffer *leaf; 41125d4f98a2SYan Zheng int ret; 41135d4f98a2SYan Zheng 41145d4f98a2SYan Zheng path = btrfs_alloc_path(); 41155d4f98a2SYan Zheng if (!path) 41165d4f98a2SYan Zheng return -ENOMEM; 41175d4f98a2SYan Zheng 41185d4f98a2SYan Zheng ret = btrfs_insert_empty_inode(trans, root, path, objectid); 41195d4f98a2SYan Zheng if (ret) 41205d4f98a2SYan Zheng goto out; 41215d4f98a2SYan Zheng 41225d4f98a2SYan Zheng leaf = path->nodes[0]; 41235d4f98a2SYan Zheng item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); 4124b159fa28SDavid Sterba memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 41255d4f98a2SYan Zheng btrfs_set_inode_generation(leaf, item, 1); 41260257bb82SYan, Zheng btrfs_set_inode_size(leaf, item, 0); 41275d4f98a2SYan Zheng btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); 41283fd0a558SYan, Zheng btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS | 41293fd0a558SYan, Zheng BTRFS_INODE_PREALLOC); 41305d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 41315d4f98a2SYan Zheng out: 41325d4f98a2SYan Zheng btrfs_free_path(path); 41335d4f98a2SYan Zheng return ret; 41345d4f98a2SYan Zheng } 41355d4f98a2SYan Zheng 41365d4f98a2SYan Zheng /* 41375d4f98a2SYan Zheng * helper to create inode for data relocation. 41385d4f98a2SYan Zheng * the inode is in data relocation tree and its link count is 0 41395d4f98a2SYan Zheng */ 41403fd0a558SYan, Zheng static noinline_for_stack 41413fd0a558SYan, Zheng struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, 414232da5386SDavid Sterba struct btrfs_block_group *group) 41435d4f98a2SYan Zheng { 41445d4f98a2SYan Zheng struct inode *inode = NULL; 41455d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 41465d4f98a2SYan Zheng struct btrfs_root *root; 41475d4f98a2SYan Zheng struct btrfs_key key; 41484624900dSZhaolei u64 objectid; 41495d4f98a2SYan Zheng int err = 0; 41505d4f98a2SYan Zheng 41515d4f98a2SYan Zheng root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 41525d4f98a2SYan Zheng if (IS_ERR(root)) 41535d4f98a2SYan Zheng return ERR_CAST(root); 41545d4f98a2SYan Zheng 4155a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 6); 415676deacf0SJosef Bacik if (IS_ERR(trans)) { 415700246528SJosef Bacik btrfs_put_root(root); 41583fd0a558SYan, Zheng return ERR_CAST(trans); 415976deacf0SJosef Bacik } 41605d4f98a2SYan Zheng 4161581bb050SLi Zefan err = btrfs_find_free_objectid(root, &objectid); 41625d4f98a2SYan Zheng if (err) 41635d4f98a2SYan Zheng goto out; 41645d4f98a2SYan Zheng 41650257bb82SYan, Zheng err = __insert_orphan_inode(trans, root, objectid); 41665d4f98a2SYan Zheng BUG_ON(err); 41675d4f98a2SYan Zheng 41685d4f98a2SYan Zheng key.objectid = objectid; 41695d4f98a2SYan Zheng key.type = BTRFS_INODE_ITEM_KEY; 41705d4f98a2SYan Zheng key.offset = 0; 41714c66e0d4SDavid Sterba inode = btrfs_iget(fs_info->sb, &key, root); 41722e19f1f9SAl Viro BUG_ON(IS_ERR(inode)); 4173b3470b5dSDavid Sterba BTRFS_I(inode)->index_cnt = group->start; 41745d4f98a2SYan Zheng 417573f2e545SNikolay Borisov err = btrfs_orphan_add(trans, BTRFS_I(inode)); 41765d4f98a2SYan Zheng out: 417700246528SJosef Bacik btrfs_put_root(root); 41783a45bb20SJeff Mahoney btrfs_end_transaction(trans); 41792ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 41805d4f98a2SYan Zheng if (err) { 41815d4f98a2SYan Zheng if (inode) 41825d4f98a2SYan Zheng iput(inode); 41835d4f98a2SYan Zheng inode = ERR_PTR(err); 41845d4f98a2SYan Zheng } 41855d4f98a2SYan Zheng return inode; 41865d4f98a2SYan Zheng } 41875d4f98a2SYan Zheng 4188c258d6e3SQu Wenruo static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info) 41893fd0a558SYan, Zheng { 41903fd0a558SYan, Zheng struct reloc_control *rc; 41913fd0a558SYan, Zheng 41923fd0a558SYan, Zheng rc = kzalloc(sizeof(*rc), GFP_NOFS); 41933fd0a558SYan, Zheng if (!rc) 41943fd0a558SYan, Zheng return NULL; 41953fd0a558SYan, Zheng 41963fd0a558SYan, Zheng INIT_LIST_HEAD(&rc->reloc_roots); 4197d2311e69SQu Wenruo INIT_LIST_HEAD(&rc->dirty_subvol_roots); 41983fd0a558SYan, Zheng backref_cache_init(&rc->backref_cache); 41993fd0a558SYan, Zheng mapping_tree_init(&rc->reloc_root_tree); 420043eb5f29SQu Wenruo extent_io_tree_init(fs_info, &rc->processed_blocks, 420143eb5f29SQu Wenruo IO_TREE_RELOC_BLOCKS, NULL); 42023fd0a558SYan, Zheng return rc; 42033fd0a558SYan, Zheng } 42043fd0a558SYan, Zheng 42055d4f98a2SYan Zheng /* 4206ebce0e01SAdam Borowski * Print the block group being relocated 4207ebce0e01SAdam Borowski */ 4208ebce0e01SAdam Borowski static void describe_relocation(struct btrfs_fs_info *fs_info, 420932da5386SDavid Sterba struct btrfs_block_group *block_group) 4210ebce0e01SAdam Borowski { 4211f89e09cfSAnand Jain char buf[128] = {'\0'}; 4212ebce0e01SAdam Borowski 4213f89e09cfSAnand Jain btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf)); 4214ebce0e01SAdam Borowski 4215ebce0e01SAdam Borowski btrfs_info(fs_info, 4216ebce0e01SAdam Borowski "relocating block group %llu flags %s", 4217b3470b5dSDavid Sterba block_group->start, buf); 4218ebce0e01SAdam Borowski } 4219ebce0e01SAdam Borowski 4220430640e3SQu Wenruo static const char *stage_to_string(int stage) 4221430640e3SQu Wenruo { 4222430640e3SQu Wenruo if (stage == MOVE_DATA_EXTENTS) 4223430640e3SQu Wenruo return "move data extents"; 4224430640e3SQu Wenruo if (stage == UPDATE_DATA_PTRS) 4225430640e3SQu Wenruo return "update data pointers"; 4226430640e3SQu Wenruo return "unknown"; 4227430640e3SQu Wenruo } 4228430640e3SQu Wenruo 4229ebce0e01SAdam Borowski /* 42305d4f98a2SYan Zheng * function to relocate all extents in a block group. 42315d4f98a2SYan Zheng */ 42326bccf3abSJeff Mahoney int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) 42335d4f98a2SYan Zheng { 423432da5386SDavid Sterba struct btrfs_block_group *bg; 42356bccf3abSJeff Mahoney struct btrfs_root *extent_root = fs_info->extent_root; 42365d4f98a2SYan Zheng struct reloc_control *rc; 42370af3d00bSJosef Bacik struct inode *inode; 42380af3d00bSJosef Bacik struct btrfs_path *path; 42395d4f98a2SYan Zheng int ret; 4240f0486c68SYan, Zheng int rw = 0; 42415d4f98a2SYan Zheng int err = 0; 42425d4f98a2SYan Zheng 4243eede2bf3SOmar Sandoval bg = btrfs_lookup_block_group(fs_info, group_start); 4244eede2bf3SOmar Sandoval if (!bg) 4245eede2bf3SOmar Sandoval return -ENOENT; 4246eede2bf3SOmar Sandoval 4247eede2bf3SOmar Sandoval if (btrfs_pinned_by_swapfile(fs_info, bg)) { 4248eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 4249eede2bf3SOmar Sandoval return -ETXTBSY; 4250eede2bf3SOmar Sandoval } 4251eede2bf3SOmar Sandoval 4252c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 4253eede2bf3SOmar Sandoval if (!rc) { 4254eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 42555d4f98a2SYan Zheng return -ENOMEM; 4256eede2bf3SOmar Sandoval } 42575d4f98a2SYan Zheng 4258f0486c68SYan, Zheng rc->extent_root = extent_root; 4259eede2bf3SOmar Sandoval rc->block_group = bg; 42605d4f98a2SYan Zheng 4261b12de528SQu Wenruo ret = btrfs_inc_block_group_ro(rc->block_group, true); 4262f0486c68SYan, Zheng if (ret) { 4263f0486c68SYan, Zheng err = ret; 4264f0486c68SYan, Zheng goto out; 4265f0486c68SYan, Zheng } 4266f0486c68SYan, Zheng rw = 1; 4267f0486c68SYan, Zheng 42680af3d00bSJosef Bacik path = btrfs_alloc_path(); 42690af3d00bSJosef Bacik if (!path) { 42700af3d00bSJosef Bacik err = -ENOMEM; 42710af3d00bSJosef Bacik goto out; 42720af3d00bSJosef Bacik } 42730af3d00bSJosef Bacik 42747949f339SDavid Sterba inode = lookup_free_space_inode(rc->block_group, path); 42750af3d00bSJosef Bacik btrfs_free_path(path); 42760af3d00bSJosef Bacik 42770af3d00bSJosef Bacik if (!IS_ERR(inode)) 42781bbc621eSChris Mason ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0); 42790af3d00bSJosef Bacik else 42800af3d00bSJosef Bacik ret = PTR_ERR(inode); 42810af3d00bSJosef Bacik 42820af3d00bSJosef Bacik if (ret && ret != -ENOENT) { 42830af3d00bSJosef Bacik err = ret; 42840af3d00bSJosef Bacik goto out; 42850af3d00bSJosef Bacik } 42860af3d00bSJosef Bacik 42875d4f98a2SYan Zheng rc->data_inode = create_reloc_inode(fs_info, rc->block_group); 42885d4f98a2SYan Zheng if (IS_ERR(rc->data_inode)) { 42895d4f98a2SYan Zheng err = PTR_ERR(rc->data_inode); 42905d4f98a2SYan Zheng rc->data_inode = NULL; 42915d4f98a2SYan Zheng goto out; 42925d4f98a2SYan Zheng } 42935d4f98a2SYan Zheng 42940b246afaSJeff Mahoney describe_relocation(fs_info, rc->block_group); 42955d4f98a2SYan Zheng 42969cfa3e34SFilipe Manana btrfs_wait_block_group_reservations(rc->block_group); 4297f78c436cSFilipe Manana btrfs_wait_nocow_writers(rc->block_group); 42986374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 4299b3470b5dSDavid Sterba rc->block_group->start, 4300b3470b5dSDavid Sterba rc->block_group->length); 43015d4f98a2SYan Zheng 43025d4f98a2SYan Zheng while (1) { 4303430640e3SQu Wenruo int finishes_stage; 4304430640e3SQu Wenruo 430576dda93cSYan, Zheng mutex_lock(&fs_info->cleaner_mutex); 43065d4f98a2SYan Zheng ret = relocate_block_group(rc); 430776dda93cSYan, Zheng mutex_unlock(&fs_info->cleaner_mutex); 4308ff612ba7SJosef Bacik if (ret < 0) 43095d4f98a2SYan Zheng err = ret; 4310ff612ba7SJosef Bacik 4311430640e3SQu Wenruo finishes_stage = rc->stage; 4312ff612ba7SJosef Bacik /* 4313ff612ba7SJosef Bacik * We may have gotten ENOSPC after we already dirtied some 4314ff612ba7SJosef Bacik * extents. If writeout happens while we're relocating a 4315ff612ba7SJosef Bacik * different block group we could end up hitting the 4316ff612ba7SJosef Bacik * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in 4317ff612ba7SJosef Bacik * btrfs_reloc_cow_block. Make sure we write everything out 4318ff612ba7SJosef Bacik * properly so we don't trip over this problem, and then break 4319ff612ba7SJosef Bacik * out of the loop if we hit an error. 4320ff612ba7SJosef Bacik */ 4321ff612ba7SJosef Bacik if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) { 4322ff612ba7SJosef Bacik ret = btrfs_wait_ordered_range(rc->data_inode, 0, 4323ff612ba7SJosef Bacik (u64)-1); 4324ff612ba7SJosef Bacik if (ret) 4325ff612ba7SJosef Bacik err = ret; 4326ff612ba7SJosef Bacik invalidate_mapping_pages(rc->data_inode->i_mapping, 4327ff612ba7SJosef Bacik 0, -1); 4328ff612ba7SJosef Bacik rc->stage = UPDATE_DATA_PTRS; 43295d4f98a2SYan Zheng } 43305d4f98a2SYan Zheng 4331ff612ba7SJosef Bacik if (err < 0) 4332ff612ba7SJosef Bacik goto out; 4333ff612ba7SJosef Bacik 43345d4f98a2SYan Zheng if (rc->extents_found == 0) 43355d4f98a2SYan Zheng break; 43365d4f98a2SYan Zheng 4337430640e3SQu Wenruo btrfs_info(fs_info, "found %llu extents, stage: %s", 4338430640e3SQu Wenruo rc->extents_found, stage_to_string(finishes_stage)); 43395d4f98a2SYan Zheng } 43405d4f98a2SYan Zheng 43415d4f98a2SYan Zheng WARN_ON(rc->block_group->pinned > 0); 43425d4f98a2SYan Zheng WARN_ON(rc->block_group->reserved > 0); 4343bf38be65SDavid Sterba WARN_ON(rc->block_group->used > 0); 43445d4f98a2SYan Zheng out: 4345f0486c68SYan, Zheng if (err && rw) 43462ff7e61eSJeff Mahoney btrfs_dec_block_group_ro(rc->block_group); 43475d4f98a2SYan Zheng iput(rc->data_inode); 43485d4f98a2SYan Zheng btrfs_put_block_group(rc->block_group); 43495d4f98a2SYan Zheng kfree(rc); 43505d4f98a2SYan Zheng return err; 43515d4f98a2SYan Zheng } 43525d4f98a2SYan Zheng 435376dda93cSYan, Zheng static noinline_for_stack int mark_garbage_root(struct btrfs_root *root) 435476dda93cSYan, Zheng { 43550b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 435676dda93cSYan, Zheng struct btrfs_trans_handle *trans; 435779787eaaSJeff Mahoney int ret, err; 435876dda93cSYan, Zheng 43590b246afaSJeff Mahoney trans = btrfs_start_transaction(fs_info->tree_root, 0); 436079787eaaSJeff Mahoney if (IS_ERR(trans)) 436179787eaaSJeff Mahoney return PTR_ERR(trans); 436276dda93cSYan, Zheng 436376dda93cSYan, Zheng memset(&root->root_item.drop_progress, 0, 436476dda93cSYan, Zheng sizeof(root->root_item.drop_progress)); 436576dda93cSYan, Zheng root->root_item.drop_level = 0; 436676dda93cSYan, Zheng btrfs_set_root_refs(&root->root_item, 0); 43670b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 436876dda93cSYan, Zheng &root->root_key, &root->root_item); 436976dda93cSYan, Zheng 43703a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 437179787eaaSJeff Mahoney if (err) 437279787eaaSJeff Mahoney return err; 437379787eaaSJeff Mahoney return ret; 437476dda93cSYan, Zheng } 437576dda93cSYan, Zheng 43765d4f98a2SYan Zheng /* 43775d4f98a2SYan Zheng * recover relocation interrupted by system crash. 43785d4f98a2SYan Zheng * 43795d4f98a2SYan Zheng * this function resumes merging reloc trees with corresponding fs trees. 43805d4f98a2SYan Zheng * this is important for keeping the sharing of tree blocks 43815d4f98a2SYan Zheng */ 43825d4f98a2SYan Zheng int btrfs_recover_relocation(struct btrfs_root *root) 43835d4f98a2SYan Zheng { 43840b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 43855d4f98a2SYan Zheng LIST_HEAD(reloc_roots); 43865d4f98a2SYan Zheng struct btrfs_key key; 43875d4f98a2SYan Zheng struct btrfs_root *fs_root; 43885d4f98a2SYan Zheng struct btrfs_root *reloc_root; 43895d4f98a2SYan Zheng struct btrfs_path *path; 43905d4f98a2SYan Zheng struct extent_buffer *leaf; 43915d4f98a2SYan Zheng struct reloc_control *rc = NULL; 43925d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 43935d4f98a2SYan Zheng int ret; 43945d4f98a2SYan Zheng int err = 0; 43955d4f98a2SYan Zheng 43965d4f98a2SYan Zheng path = btrfs_alloc_path(); 43975d4f98a2SYan Zheng if (!path) 43985d4f98a2SYan Zheng return -ENOMEM; 4399e4058b54SDavid Sterba path->reada = READA_BACK; 44005d4f98a2SYan Zheng 44015d4f98a2SYan Zheng key.objectid = BTRFS_TREE_RELOC_OBJECTID; 44025d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 44035d4f98a2SYan Zheng key.offset = (u64)-1; 44045d4f98a2SYan Zheng 44055d4f98a2SYan Zheng while (1) { 44060b246afaSJeff Mahoney ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, 44075d4f98a2SYan Zheng path, 0, 0); 44085d4f98a2SYan Zheng if (ret < 0) { 44095d4f98a2SYan Zheng err = ret; 44105d4f98a2SYan Zheng goto out; 44115d4f98a2SYan Zheng } 44125d4f98a2SYan Zheng if (ret > 0) { 44135d4f98a2SYan Zheng if (path->slots[0] == 0) 44145d4f98a2SYan Zheng break; 44155d4f98a2SYan Zheng path->slots[0]--; 44165d4f98a2SYan Zheng } 44175d4f98a2SYan Zheng leaf = path->nodes[0]; 44185d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 4419b3b4aa74SDavid Sterba btrfs_release_path(path); 44205d4f98a2SYan Zheng 44215d4f98a2SYan Zheng if (key.objectid != BTRFS_TREE_RELOC_OBJECTID || 44225d4f98a2SYan Zheng key.type != BTRFS_ROOT_ITEM_KEY) 44235d4f98a2SYan Zheng break; 44245d4f98a2SYan Zheng 44253dbf1738SJosef Bacik reloc_root = btrfs_read_tree_root(root, &key); 44265d4f98a2SYan Zheng if (IS_ERR(reloc_root)) { 44275d4f98a2SYan Zheng err = PTR_ERR(reloc_root); 44285d4f98a2SYan Zheng goto out; 44295d4f98a2SYan Zheng } 44305d4f98a2SYan Zheng 44313dbf1738SJosef Bacik set_bit(BTRFS_ROOT_REF_COWS, &reloc_root->state); 44325d4f98a2SYan Zheng list_add(&reloc_root->root_list, &reloc_roots); 44335d4f98a2SYan Zheng 44345d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 44350b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, 44365d4f98a2SYan Zheng reloc_root->root_key.offset); 44375d4f98a2SYan Zheng if (IS_ERR(fs_root)) { 443876dda93cSYan, Zheng ret = PTR_ERR(fs_root); 443976dda93cSYan, Zheng if (ret != -ENOENT) { 444076dda93cSYan, Zheng err = ret; 44415d4f98a2SYan Zheng goto out; 44425d4f98a2SYan Zheng } 444379787eaaSJeff Mahoney ret = mark_garbage_root(reloc_root); 444479787eaaSJeff Mahoney if (ret < 0) { 444579787eaaSJeff Mahoney err = ret; 444679787eaaSJeff Mahoney goto out; 444779787eaaSJeff Mahoney } 4448932fd26dSJosef Bacik } else { 444900246528SJosef Bacik btrfs_put_root(fs_root); 445076dda93cSYan, Zheng } 44515d4f98a2SYan Zheng } 44525d4f98a2SYan Zheng 44535d4f98a2SYan Zheng if (key.offset == 0) 44545d4f98a2SYan Zheng break; 44555d4f98a2SYan Zheng 44565d4f98a2SYan Zheng key.offset--; 44575d4f98a2SYan Zheng } 4458b3b4aa74SDavid Sterba btrfs_release_path(path); 44595d4f98a2SYan Zheng 44605d4f98a2SYan Zheng if (list_empty(&reloc_roots)) 44615d4f98a2SYan Zheng goto out; 44625d4f98a2SYan Zheng 4463c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 44645d4f98a2SYan Zheng if (!rc) { 44655d4f98a2SYan Zheng err = -ENOMEM; 44665d4f98a2SYan Zheng goto out; 44675d4f98a2SYan Zheng } 44685d4f98a2SYan Zheng 44690b246afaSJeff Mahoney rc->extent_root = fs_info->extent_root; 44705d4f98a2SYan Zheng 44715d4f98a2SYan Zheng set_reloc_control(rc); 44725d4f98a2SYan Zheng 44737a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 44743612b495STsutomu Itoh if (IS_ERR(trans)) { 44753612b495STsutomu Itoh err = PTR_ERR(trans); 4476fb2d83eeSJosef Bacik goto out_unset; 44773612b495STsutomu Itoh } 44783fd0a558SYan, Zheng 44793fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 44803fd0a558SYan, Zheng 44815d4f98a2SYan Zheng while (!list_empty(&reloc_roots)) { 44825d4f98a2SYan Zheng reloc_root = list_entry(reloc_roots.next, 44835d4f98a2SYan Zheng struct btrfs_root, root_list); 44845d4f98a2SYan Zheng list_del(&reloc_root->root_list); 44855d4f98a2SYan Zheng 44865d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) == 0) { 44875d4f98a2SYan Zheng list_add_tail(&reloc_root->root_list, 44885d4f98a2SYan Zheng &rc->reloc_roots); 44895d4f98a2SYan Zheng continue; 44905d4f98a2SYan Zheng } 44915d4f98a2SYan Zheng 44920b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, reloc_root->root_key.offset); 449379787eaaSJeff Mahoney if (IS_ERR(fs_root)) { 449479787eaaSJeff Mahoney err = PTR_ERR(fs_root); 4495ca1aa281SJosef Bacik list_add_tail(&reloc_root->root_list, &reloc_roots); 4496fb2d83eeSJosef Bacik goto out_unset; 449779787eaaSJeff Mahoney } 44985d4f98a2SYan Zheng 4499ffd7b339SJeff Mahoney err = __add_reloc_root(reloc_root); 450079787eaaSJeff Mahoney BUG_ON(err < 0); /* -ENOMEM or logic error */ 45015d4f98a2SYan Zheng fs_root->reloc_root = reloc_root; 450200246528SJosef Bacik btrfs_put_root(fs_root); 45035d4f98a2SYan Zheng } 45045d4f98a2SYan Zheng 45053a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 450679787eaaSJeff Mahoney if (err) 4507fb2d83eeSJosef Bacik goto out_unset; 45085d4f98a2SYan Zheng 45095d4f98a2SYan Zheng merge_reloc_roots(rc); 45105d4f98a2SYan Zheng 45115d4f98a2SYan Zheng unset_reloc_control(rc); 45125d4f98a2SYan Zheng 45137a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 451462b99540SQu Wenruo if (IS_ERR(trans)) { 45153612b495STsutomu Itoh err = PTR_ERR(trans); 45166217b0faSJosef Bacik goto out_clean; 451762b99540SQu Wenruo } 45183a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 45196217b0faSJosef Bacik out_clean: 4520d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 4521d2311e69SQu Wenruo if (ret < 0 && !err) 4522d2311e69SQu Wenruo err = ret; 4523fb2d83eeSJosef Bacik out_unset: 4524fb2d83eeSJosef Bacik unset_reloc_control(rc); 45255d4f98a2SYan Zheng kfree(rc); 45263612b495STsutomu Itoh out: 4527aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 4528aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 4529aca1bba6SLiu Bo 45305d4f98a2SYan Zheng btrfs_free_path(path); 45315d4f98a2SYan Zheng 45325d4f98a2SYan Zheng if (err == 0) { 45335d4f98a2SYan Zheng /* cleanup orphan inode in data relocation tree */ 45340b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 4535932fd26dSJosef Bacik if (IS_ERR(fs_root)) { 45365d4f98a2SYan Zheng err = PTR_ERR(fs_root); 4537932fd26dSJosef Bacik } else { 453866b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_root); 453900246528SJosef Bacik btrfs_put_root(fs_root); 4540932fd26dSJosef Bacik } 4541932fd26dSJosef Bacik } 45425d4f98a2SYan Zheng return err; 45435d4f98a2SYan Zheng } 45445d4f98a2SYan Zheng 45455d4f98a2SYan Zheng /* 45465d4f98a2SYan Zheng * helper to add ordered checksum for data relocation. 45475d4f98a2SYan Zheng * 45485d4f98a2SYan Zheng * cloning checksum properly handles the nodatasum extents. 45495d4f98a2SYan Zheng * it also saves CPU time to re-calculate the checksum. 45505d4f98a2SYan Zheng */ 45515d4f98a2SYan Zheng int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len) 45525d4f98a2SYan Zheng { 45530b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 45545d4f98a2SYan Zheng struct btrfs_ordered_sum *sums; 45555d4f98a2SYan Zheng struct btrfs_ordered_extent *ordered; 45565d4f98a2SYan Zheng int ret; 45575d4f98a2SYan Zheng u64 disk_bytenr; 45584577b014SJosef Bacik u64 new_bytenr; 45595d4f98a2SYan Zheng LIST_HEAD(list); 45605d4f98a2SYan Zheng 45615d4f98a2SYan Zheng ordered = btrfs_lookup_ordered_extent(inode, file_pos); 4562bffe633eSOmar Sandoval BUG_ON(ordered->file_offset != file_pos || ordered->num_bytes != len); 45635d4f98a2SYan Zheng 45645d4f98a2SYan Zheng disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt; 45650b246afaSJeff Mahoney ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr, 4566a2de733cSArne Jansen disk_bytenr + len - 1, &list, 0); 456779787eaaSJeff Mahoney if (ret) 456879787eaaSJeff Mahoney goto out; 45695d4f98a2SYan Zheng 45705d4f98a2SYan Zheng while (!list_empty(&list)) { 45715d4f98a2SYan Zheng sums = list_entry(list.next, struct btrfs_ordered_sum, list); 45725d4f98a2SYan Zheng list_del_init(&sums->list); 45735d4f98a2SYan Zheng 45744577b014SJosef Bacik /* 45754577b014SJosef Bacik * We need to offset the new_bytenr based on where the csum is. 45764577b014SJosef Bacik * We need to do this because we will read in entire prealloc 45774577b014SJosef Bacik * extents but we may have written to say the middle of the 45784577b014SJosef Bacik * prealloc extent, so we need to make sure the csum goes with 45794577b014SJosef Bacik * the right disk offset. 45804577b014SJosef Bacik * 45814577b014SJosef Bacik * We can do this because the data reloc inode refers strictly 45824577b014SJosef Bacik * to the on disk bytes, so we don't have to worry about 45834577b014SJosef Bacik * disk_len vs real len like with real inodes since it's all 45844577b014SJosef Bacik * disk length. 45854577b014SJosef Bacik */ 4586bffe633eSOmar Sandoval new_bytenr = ordered->disk_bytenr + sums->bytenr - disk_bytenr; 45874577b014SJosef Bacik sums->bytenr = new_bytenr; 45885d4f98a2SYan Zheng 4589f9756261SNikolay Borisov btrfs_add_ordered_sum(ordered, sums); 45905d4f98a2SYan Zheng } 459179787eaaSJeff Mahoney out: 45925d4f98a2SYan Zheng btrfs_put_ordered_extent(ordered); 4593411fc6bcSAndi Kleen return ret; 45945d4f98a2SYan Zheng } 45953fd0a558SYan, Zheng 459683d4cfd4SJosef Bacik int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans, 45973fd0a558SYan, Zheng struct btrfs_root *root, struct extent_buffer *buf, 45983fd0a558SYan, Zheng struct extent_buffer *cow) 45993fd0a558SYan, Zheng { 46000b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 46013fd0a558SYan, Zheng struct reloc_control *rc; 46023fd0a558SYan, Zheng struct backref_node *node; 46033fd0a558SYan, Zheng int first_cow = 0; 46043fd0a558SYan, Zheng int level; 460583d4cfd4SJosef Bacik int ret = 0; 46063fd0a558SYan, Zheng 46070b246afaSJeff Mahoney rc = fs_info->reloc_ctl; 46083fd0a558SYan, Zheng if (!rc) 460983d4cfd4SJosef Bacik return 0; 46103fd0a558SYan, Zheng 46113fd0a558SYan, Zheng BUG_ON(rc->stage == UPDATE_DATA_PTRS && 46123fd0a558SYan, Zheng root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID); 46133fd0a558SYan, Zheng 4614c974c464SWang Shilong if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 4615c974c464SWang Shilong if (buf == root->node) 4616c974c464SWang Shilong __update_reloc_root(root, cow->start); 4617c974c464SWang Shilong } 4618c974c464SWang Shilong 46193fd0a558SYan, Zheng level = btrfs_header_level(buf); 46203fd0a558SYan, Zheng if (btrfs_header_generation(buf) <= 46213fd0a558SYan, Zheng btrfs_root_last_snapshot(&root->root_item)) 46223fd0a558SYan, Zheng first_cow = 1; 46233fd0a558SYan, Zheng 46243fd0a558SYan, Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID && 46253fd0a558SYan, Zheng rc->create_reloc_tree) { 46263fd0a558SYan, Zheng WARN_ON(!first_cow && level == 0); 46273fd0a558SYan, Zheng 46283fd0a558SYan, Zheng node = rc->backref_cache.path[level]; 46293fd0a558SYan, Zheng BUG_ON(node->bytenr != buf->start && 46303fd0a558SYan, Zheng node->new_bytenr != buf->start); 46313fd0a558SYan, Zheng 46323fd0a558SYan, Zheng drop_node_buffer(node); 463367439dadSDavid Sterba atomic_inc(&cow->refs); 46343fd0a558SYan, Zheng node->eb = cow; 46353fd0a558SYan, Zheng node->new_bytenr = cow->start; 46363fd0a558SYan, Zheng 46373fd0a558SYan, Zheng if (!node->pending) { 46383fd0a558SYan, Zheng list_move_tail(&node->list, 46393fd0a558SYan, Zheng &rc->backref_cache.pending[level]); 46403fd0a558SYan, Zheng node->pending = 1; 46413fd0a558SYan, Zheng } 46423fd0a558SYan, Zheng 46433fd0a558SYan, Zheng if (first_cow) 46443fd0a558SYan, Zheng __mark_block_processed(rc, node); 46453fd0a558SYan, Zheng 46463fd0a558SYan, Zheng if (first_cow && level > 0) 46473fd0a558SYan, Zheng rc->nodes_relocated += buf->len; 46483fd0a558SYan, Zheng } 46493fd0a558SYan, Zheng 465083d4cfd4SJosef Bacik if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) 46513fd0a558SYan, Zheng ret = replace_file_extents(trans, rc, root, cow); 465283d4cfd4SJosef Bacik return ret; 46533fd0a558SYan, Zheng } 46543fd0a558SYan, Zheng 46553fd0a558SYan, Zheng /* 46563fd0a558SYan, Zheng * called before creating snapshot. it calculates metadata reservation 465701327610SNicholas D Steeves * required for relocating tree blocks in the snapshot 46583fd0a558SYan, Zheng */ 4659147d256eSZhaolei void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending, 46603fd0a558SYan, Zheng u64 *bytes_to_reserve) 46613fd0a558SYan, Zheng { 466210995c04SQu Wenruo struct btrfs_root *root = pending->root; 466310995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 46643fd0a558SYan, Zheng 46656282675eSQu Wenruo if (!rc || !have_reloc_root(root)) 46663fd0a558SYan, Zheng return; 46673fd0a558SYan, Zheng 46683fd0a558SYan, Zheng if (!rc->merge_reloc_tree) 46693fd0a558SYan, Zheng return; 46703fd0a558SYan, Zheng 46713fd0a558SYan, Zheng root = root->reloc_root; 46723fd0a558SYan, Zheng BUG_ON(btrfs_root_refs(&root->root_item) == 0); 46733fd0a558SYan, Zheng /* 46743fd0a558SYan, Zheng * relocation is in the stage of merging trees. the space 46753fd0a558SYan, Zheng * used by merging a reloc tree is twice the size of 46763fd0a558SYan, Zheng * relocated tree nodes in the worst case. half for cowing 46773fd0a558SYan, Zheng * the reloc tree, half for cowing the fs tree. the space 46783fd0a558SYan, Zheng * used by cowing the reloc tree will be freed after the 46793fd0a558SYan, Zheng * tree is dropped. if we create snapshot, cowing the fs 46803fd0a558SYan, Zheng * tree may use more space than it frees. so we need 46813fd0a558SYan, Zheng * reserve extra space. 46823fd0a558SYan, Zheng */ 46833fd0a558SYan, Zheng *bytes_to_reserve += rc->nodes_relocated; 46843fd0a558SYan, Zheng } 46853fd0a558SYan, Zheng 46863fd0a558SYan, Zheng /* 46873fd0a558SYan, Zheng * called after snapshot is created. migrate block reservation 46883fd0a558SYan, Zheng * and create reloc root for the newly created snapshot 46893fd0a558SYan, Zheng */ 469049b25e05SJeff Mahoney int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans, 46913fd0a558SYan, Zheng struct btrfs_pending_snapshot *pending) 46923fd0a558SYan, Zheng { 46933fd0a558SYan, Zheng struct btrfs_root *root = pending->root; 46943fd0a558SYan, Zheng struct btrfs_root *reloc_root; 46953fd0a558SYan, Zheng struct btrfs_root *new_root; 469610995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 46973fd0a558SYan, Zheng int ret; 46983fd0a558SYan, Zheng 46996282675eSQu Wenruo if (!rc || !have_reloc_root(root)) 470049b25e05SJeff Mahoney return 0; 47013fd0a558SYan, Zheng 47023fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 47033fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated; 47043fd0a558SYan, Zheng 47053fd0a558SYan, Zheng if (rc->merge_reloc_tree) { 47063fd0a558SYan, Zheng ret = btrfs_block_rsv_migrate(&pending->block_rsv, 47073fd0a558SYan, Zheng rc->block_rsv, 47083a584174SLu Fengqi rc->nodes_relocated, true); 470949b25e05SJeff Mahoney if (ret) 471049b25e05SJeff Mahoney return ret; 47113fd0a558SYan, Zheng } 47123fd0a558SYan, Zheng 47133fd0a558SYan, Zheng new_root = pending->snap; 47143fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root->reloc_root, 47153fd0a558SYan, Zheng new_root->root_key.objectid); 471649b25e05SJeff Mahoney if (IS_ERR(reloc_root)) 471749b25e05SJeff Mahoney return PTR_ERR(reloc_root); 47183fd0a558SYan, Zheng 4719ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 4720ffd7b339SJeff Mahoney BUG_ON(ret < 0); 47213fd0a558SYan, Zheng new_root->reloc_root = reloc_root; 47223fd0a558SYan, Zheng 472349b25e05SJeff Mahoney if (rc->create_reloc_tree) 47243fd0a558SYan, Zheng ret = clone_backref_node(trans, rc, root, reloc_root); 472549b25e05SJeff Mahoney return ret; 47263fd0a558SYan, Zheng } 4727