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 15141fac4a54SQu Wenruo /* 15151fac4a54SQu Wenruo * The subvolume has reloc tree but the swap is finished, no need to 15161fac4a54SQu Wenruo * create/update the dead reloc tree 15171fac4a54SQu Wenruo */ 15186282675eSQu Wenruo if (reloc_root_is_dead(root)) 15191fac4a54SQu Wenruo return 0; 15201fac4a54SQu Wenruo 15213fd0a558SYan, Zheng if (root->reloc_root) { 15223fd0a558SYan, Zheng reloc_root = root->reloc_root; 15233fd0a558SYan, Zheng reloc_root->last_trans = trans->transid; 15243fd0a558SYan, Zheng return 0; 15253fd0a558SYan, Zheng } 15263fd0a558SYan, Zheng 15273fd0a558SYan, Zheng if (!rc || !rc->create_reloc_tree || 15283fd0a558SYan, Zheng root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 15293fd0a558SYan, Zheng return 0; 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 } 23729e6a0c52SJosef Bacik trans->block_rsv = rc->block_rsv; 23733fd0a558SYan, Zheng 23743fd0a558SYan, Zheng replaced = 0; 23755d4f98a2SYan Zheng max_level = level; 23765d4f98a2SYan Zheng 23775d4f98a2SYan Zheng ret = walk_down_reloc_tree(reloc_root, path, &level); 23785d4f98a2SYan Zheng if (ret < 0) { 23795d4f98a2SYan Zheng err = ret; 23805d4f98a2SYan Zheng goto out; 23815d4f98a2SYan Zheng } 23825d4f98a2SYan Zheng if (ret > 0) 23835d4f98a2SYan Zheng break; 23845d4f98a2SYan Zheng 23855d4f98a2SYan Zheng if (!find_next_key(path, level, &key) && 23865d4f98a2SYan Zheng btrfs_comp_cpu_keys(&next_key, &key) >= 0) { 23875d4f98a2SYan Zheng ret = 0; 23885d4f98a2SYan Zheng } else { 23893d0174f7SQu Wenruo ret = replace_path(trans, rc, root, reloc_root, path, 23903fd0a558SYan, Zheng &next_key, level, max_level); 23915d4f98a2SYan Zheng } 23925d4f98a2SYan Zheng if (ret < 0) { 23935d4f98a2SYan Zheng err = ret; 23945d4f98a2SYan Zheng goto out; 23955d4f98a2SYan Zheng } 23965d4f98a2SYan Zheng 23975d4f98a2SYan Zheng if (ret > 0) { 23985d4f98a2SYan Zheng level = ret; 23995d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 24005d4f98a2SYan Zheng path->slots[level]); 24015d4f98a2SYan Zheng replaced = 1; 24025d4f98a2SYan Zheng } 24035d4f98a2SYan Zheng 24045d4f98a2SYan Zheng ret = walk_up_reloc_tree(reloc_root, path, &level); 24055d4f98a2SYan Zheng if (ret > 0) 24065d4f98a2SYan Zheng break; 24075d4f98a2SYan Zheng 24085d4f98a2SYan Zheng BUG_ON(level == 0); 24095d4f98a2SYan Zheng /* 24105d4f98a2SYan Zheng * save the merging progress in the drop_progress. 24115d4f98a2SYan Zheng * this is OK since root refs == 1 in this case. 24125d4f98a2SYan Zheng */ 24135d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &root_item->drop_progress, 24145d4f98a2SYan Zheng path->slots[level]); 24155d4f98a2SYan Zheng root_item->drop_level = level; 24165d4f98a2SYan Zheng 24173a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 24189e6a0c52SJosef Bacik trans = NULL; 24195d4f98a2SYan Zheng 24202ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 24215d4f98a2SYan Zheng 24225d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 24235d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 24245d4f98a2SYan Zheng } 24255d4f98a2SYan Zheng 24265d4f98a2SYan Zheng /* 24275d4f98a2SYan Zheng * handle the case only one block in the fs tree need to be 24285d4f98a2SYan Zheng * relocated and the block is tree root. 24295d4f98a2SYan Zheng */ 24305d4f98a2SYan Zheng leaf = btrfs_lock_root_node(root); 24315d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf); 24325d4f98a2SYan Zheng btrfs_tree_unlock(leaf); 24335d4f98a2SYan Zheng free_extent_buffer(leaf); 24345d4f98a2SYan Zheng if (ret < 0) 24355d4f98a2SYan Zheng err = ret; 24365d4f98a2SYan Zheng out: 24375d4f98a2SYan Zheng btrfs_free_path(path); 24385d4f98a2SYan Zheng 2439d2311e69SQu Wenruo if (err == 0) 2440d2311e69SQu Wenruo insert_dirty_subvol(trans, rc, root); 24415d4f98a2SYan Zheng 24429e6a0c52SJosef Bacik if (trans) 24433a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 24445d4f98a2SYan Zheng 24452ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 24465d4f98a2SYan Zheng 24475d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 24485d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 24495d4f98a2SYan Zheng 24505d4f98a2SYan Zheng return err; 24515d4f98a2SYan Zheng } 24525d4f98a2SYan Zheng 24533fd0a558SYan, Zheng static noinline_for_stack 24543fd0a558SYan, Zheng int prepare_to_merge(struct reloc_control *rc, int err) 24555d4f98a2SYan Zheng { 24563fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 24570b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 24583fd0a558SYan, Zheng struct btrfs_root *reloc_root; 24595d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 24603fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 24613fd0a558SYan, Zheng u64 num_bytes = 0; 24623fd0a558SYan, Zheng int ret; 24633fd0a558SYan, Zheng 24640b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 24650b246afaSJeff Mahoney rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 24663fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated * 2; 24670b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 24687585717fSChris Mason 24693fd0a558SYan, Zheng again: 24703fd0a558SYan, Zheng if (!err) { 24713fd0a558SYan, Zheng num_bytes = rc->merging_rsv_size; 247208e007d2SMiao Xie ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes, 247308e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 24743fd0a558SYan, Zheng if (ret) 24753fd0a558SYan, Zheng err = ret; 24763fd0a558SYan, Zheng } 24773fd0a558SYan, Zheng 24787a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 24793612b495STsutomu Itoh if (IS_ERR(trans)) { 24803612b495STsutomu Itoh if (!err) 24812ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 248263f018beSNikolay Borisov num_bytes, NULL); 24833612b495STsutomu Itoh return PTR_ERR(trans); 24843612b495STsutomu Itoh } 24853fd0a558SYan, Zheng 24863fd0a558SYan, Zheng if (!err) { 24873fd0a558SYan, Zheng if (num_bytes != rc->merging_rsv_size) { 24883a45bb20SJeff Mahoney btrfs_end_transaction(trans); 24892ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 249063f018beSNikolay Borisov num_bytes, NULL); 24913fd0a558SYan, Zheng goto again; 24923fd0a558SYan, Zheng } 24933fd0a558SYan, Zheng } 24943fd0a558SYan, Zheng 24953fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 24963fd0a558SYan, Zheng 24973fd0a558SYan, Zheng while (!list_empty(&rc->reloc_roots)) { 24983fd0a558SYan, Zheng reloc_root = list_entry(rc->reloc_roots.next, 24993fd0a558SYan, Zheng struct btrfs_root, root_list); 25003fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 25013fd0a558SYan, Zheng 25020b246afaSJeff Mahoney root = read_fs_root(fs_info, reloc_root->root_key.offset); 25033fd0a558SYan, Zheng BUG_ON(IS_ERR(root)); 25043fd0a558SYan, Zheng BUG_ON(root->reloc_root != reloc_root); 25053fd0a558SYan, Zheng 25063fd0a558SYan, Zheng /* 25073fd0a558SYan, Zheng * set reference count to 1, so btrfs_recover_relocation 25083fd0a558SYan, Zheng * knows it should resumes merging 25093fd0a558SYan, Zheng */ 25103fd0a558SYan, Zheng if (!err) 25113fd0a558SYan, Zheng btrfs_set_root_refs(&reloc_root->root_item, 1); 25123fd0a558SYan, Zheng btrfs_update_reloc_root(trans, root); 25133fd0a558SYan, Zheng 25143fd0a558SYan, Zheng list_add(&reloc_root->root_list, &reloc_roots); 251500246528SJosef Bacik btrfs_put_root(root); 25163fd0a558SYan, Zheng } 25173fd0a558SYan, Zheng 25183fd0a558SYan, Zheng list_splice(&reloc_roots, &rc->reloc_roots); 25193fd0a558SYan, Zheng 25203fd0a558SYan, Zheng if (!err) 25213a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 25223fd0a558SYan, Zheng else 25233a45bb20SJeff Mahoney btrfs_end_transaction(trans); 25243fd0a558SYan, Zheng return err; 25253fd0a558SYan, Zheng } 25263fd0a558SYan, Zheng 25273fd0a558SYan, Zheng static noinline_for_stack 2528aca1bba6SLiu Bo void free_reloc_roots(struct list_head *list) 2529aca1bba6SLiu Bo { 2530aca1bba6SLiu Bo struct btrfs_root *reloc_root; 2531aca1bba6SLiu Bo 2532aca1bba6SLiu Bo while (!list_empty(list)) { 2533aca1bba6SLiu Bo reloc_root = list_entry(list->next, struct btrfs_root, 2534aca1bba6SLiu Bo root_list); 2535bb166d72SNaohiro Aota __del_reloc_root(reloc_root); 25366bdf131fSJosef Bacik free_extent_buffer(reloc_root->node); 25376bdf131fSJosef Bacik free_extent_buffer(reloc_root->commit_root); 25386bdf131fSJosef Bacik reloc_root->node = NULL; 25396bdf131fSJosef Bacik reloc_root->commit_root = NULL; 2540aca1bba6SLiu Bo } 2541aca1bba6SLiu Bo } 2542aca1bba6SLiu Bo 2543aca1bba6SLiu Bo static noinline_for_stack 254494404e82SDavid Sterba void merge_reloc_roots(struct reloc_control *rc) 25453fd0a558SYan, Zheng { 25460b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 25475d4f98a2SYan Zheng struct btrfs_root *root; 25485d4f98a2SYan Zheng struct btrfs_root *reloc_root; 25493fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 25503fd0a558SYan, Zheng int found = 0; 2551aca1bba6SLiu Bo int ret = 0; 25523fd0a558SYan, Zheng again: 25533fd0a558SYan, Zheng root = rc->extent_root; 25547585717fSChris Mason 25557585717fSChris Mason /* 25567585717fSChris Mason * this serializes us with btrfs_record_root_in_transaction, 25577585717fSChris Mason * we have to make sure nobody is in the middle of 25587585717fSChris Mason * adding their roots to the list while we are 25597585717fSChris Mason * doing this splice 25607585717fSChris Mason */ 25610b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 25623fd0a558SYan, Zheng list_splice_init(&rc->reloc_roots, &reloc_roots); 25630b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 25645d4f98a2SYan Zheng 25653fd0a558SYan, Zheng while (!list_empty(&reloc_roots)) { 25663fd0a558SYan, Zheng found = 1; 25673fd0a558SYan, Zheng reloc_root = list_entry(reloc_roots.next, 25683fd0a558SYan, Zheng struct btrfs_root, root_list); 25695d4f98a2SYan Zheng 25705d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 25710b246afaSJeff Mahoney root = read_fs_root(fs_info, 25725d4f98a2SYan Zheng reloc_root->root_key.offset); 25735d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 25745d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 25755d4f98a2SYan Zheng 25763fd0a558SYan, Zheng ret = merge_reloc_root(rc, root); 257700246528SJosef Bacik btrfs_put_root(root); 2578b37b39cdSJosef Bacik if (ret) { 257925e293c2SWang Shilong if (list_empty(&reloc_root->root_list)) 258025e293c2SWang Shilong list_add_tail(&reloc_root->root_list, 258125e293c2SWang Shilong &reloc_roots); 2582aca1bba6SLiu Bo goto out; 2583b37b39cdSJosef Bacik } 25843fd0a558SYan, Zheng } else { 25853fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 258630d40577SQu Wenruo /* Don't forget to queue this reloc root for cleanup */ 258730d40577SQu Wenruo list_add_tail(&reloc_root->reloc_dirty_list, 258830d40577SQu Wenruo &rc->dirty_subvol_roots); 25893fd0a558SYan, Zheng } 25905d4f98a2SYan Zheng } 25915d4f98a2SYan Zheng 25923fd0a558SYan, Zheng if (found) { 25933fd0a558SYan, Zheng found = 0; 25943fd0a558SYan, Zheng goto again; 25955d4f98a2SYan Zheng } 2596aca1bba6SLiu Bo out: 2597aca1bba6SLiu Bo if (ret) { 25980b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 2599aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 2600aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 2601467bb1d2SWang Shilong 2602467bb1d2SWang Shilong /* new reloc root may be added */ 26030b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 2604467bb1d2SWang Shilong list_splice_init(&rc->reloc_roots, &reloc_roots); 26050b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2606467bb1d2SWang Shilong if (!list_empty(&reloc_roots)) 2607467bb1d2SWang Shilong free_reloc_roots(&reloc_roots); 2608aca1bba6SLiu Bo } 2609aca1bba6SLiu Bo 26105d4f98a2SYan Zheng BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); 26115d4f98a2SYan Zheng } 26125d4f98a2SYan Zheng 26135d4f98a2SYan Zheng static void free_block_list(struct rb_root *blocks) 26145d4f98a2SYan Zheng { 26155d4f98a2SYan Zheng struct tree_block *block; 26165d4f98a2SYan Zheng struct rb_node *rb_node; 26175d4f98a2SYan Zheng while ((rb_node = rb_first(blocks))) { 26185d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 26195d4f98a2SYan Zheng rb_erase(rb_node, blocks); 26205d4f98a2SYan Zheng kfree(block); 26215d4f98a2SYan Zheng } 26225d4f98a2SYan Zheng } 26235d4f98a2SYan Zheng 26245d4f98a2SYan Zheng static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans, 26255d4f98a2SYan Zheng struct btrfs_root *reloc_root) 26265d4f98a2SYan Zheng { 26270b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = reloc_root->fs_info; 26285d4f98a2SYan Zheng struct btrfs_root *root; 2629442b1ac5SJosef Bacik int ret; 26305d4f98a2SYan Zheng 26315d4f98a2SYan Zheng if (reloc_root->last_trans == trans->transid) 26325d4f98a2SYan Zheng return 0; 26335d4f98a2SYan Zheng 26340b246afaSJeff Mahoney root = read_fs_root(fs_info, reloc_root->root_key.offset); 26355d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 26365d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 2637442b1ac5SJosef Bacik ret = btrfs_record_root_in_trans(trans, root); 263800246528SJosef Bacik btrfs_put_root(root); 26395d4f98a2SYan Zheng 2640442b1ac5SJosef Bacik return ret; 26415d4f98a2SYan Zheng } 26425d4f98a2SYan Zheng 26433fd0a558SYan, Zheng static noinline_for_stack 26443fd0a558SYan, Zheng struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, 26453fd0a558SYan, Zheng struct reloc_control *rc, 26465d4f98a2SYan Zheng struct backref_node *node, 2647dc4103f9SWang Shilong struct backref_edge *edges[]) 26485d4f98a2SYan Zheng { 26495d4f98a2SYan Zheng struct backref_node *next; 26505d4f98a2SYan Zheng struct btrfs_root *root; 26513fd0a558SYan, Zheng int index = 0; 26523fd0a558SYan, Zheng 26535d4f98a2SYan Zheng next = node; 26545d4f98a2SYan Zheng while (1) { 26555d4f98a2SYan Zheng cond_resched(); 26565d4f98a2SYan Zheng next = walk_up_backref(next, edges, &index); 26575d4f98a2SYan Zheng root = next->root; 26583fd0a558SYan, Zheng BUG_ON(!root); 265927cdeb70SMiao Xie BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state)); 26605d4f98a2SYan Zheng 26615d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 26625d4f98a2SYan Zheng record_reloc_root_in_trans(trans, root); 26635d4f98a2SYan Zheng break; 26645d4f98a2SYan Zheng } 26655d4f98a2SYan Zheng 26665d4f98a2SYan Zheng btrfs_record_root_in_trans(trans, root); 26673fd0a558SYan, Zheng root = root->reloc_root; 26683fd0a558SYan, Zheng 26693fd0a558SYan, Zheng if (next->new_bytenr != root->node->start) { 26703fd0a558SYan, Zheng BUG_ON(next->new_bytenr); 26713fd0a558SYan, Zheng BUG_ON(!list_empty(&next->list)); 26723fd0a558SYan, Zheng next->new_bytenr = root->node->start; 267300246528SJosef Bacik btrfs_put_root(next->root); 267400246528SJosef Bacik next->root = btrfs_grab_root(root); 26750b530bc5SJosef Bacik ASSERT(next->root); 26763fd0a558SYan, Zheng list_add_tail(&next->list, 26773fd0a558SYan, Zheng &rc->backref_cache.changed); 26783fd0a558SYan, Zheng __mark_block_processed(rc, next); 26795d4f98a2SYan Zheng break; 26805d4f98a2SYan Zheng } 26815d4f98a2SYan Zheng 26823fd0a558SYan, Zheng WARN_ON(1); 26835d4f98a2SYan Zheng root = NULL; 26845d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 26855d4f98a2SYan Zheng if (!next || next->level <= node->level) 26865d4f98a2SYan Zheng break; 26875d4f98a2SYan Zheng } 26883fd0a558SYan, Zheng if (!root) 26893fd0a558SYan, Zheng return NULL; 26905d4f98a2SYan Zheng 26913fd0a558SYan, Zheng next = node; 26923fd0a558SYan, Zheng /* setup backref node path for btrfs_reloc_cow_block */ 26933fd0a558SYan, Zheng while (1) { 26943fd0a558SYan, Zheng rc->backref_cache.path[next->level] = next; 26953fd0a558SYan, Zheng if (--index < 0) 26963fd0a558SYan, Zheng break; 26973fd0a558SYan, Zheng next = edges[index]->node[UPPER]; 26983fd0a558SYan, Zheng } 26995d4f98a2SYan Zheng return root; 27005d4f98a2SYan Zheng } 27015d4f98a2SYan Zheng 27023fd0a558SYan, Zheng /* 27033fd0a558SYan, Zheng * select a tree root for relocation. return NULL if the block 27043fd0a558SYan, Zheng * is reference counted. we should use do_relocation() in this 27053fd0a558SYan, Zheng * case. return a tree root pointer if the block isn't reference 27063fd0a558SYan, Zheng * counted. return -ENOENT if the block is root of reloc tree. 27073fd0a558SYan, Zheng */ 27085d4f98a2SYan Zheng static noinline_for_stack 2709147d256eSZhaolei struct btrfs_root *select_one_root(struct backref_node *node) 27105d4f98a2SYan Zheng { 27113fd0a558SYan, Zheng struct backref_node *next; 27123fd0a558SYan, Zheng struct btrfs_root *root; 27133fd0a558SYan, Zheng struct btrfs_root *fs_root = NULL; 27145d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 27153fd0a558SYan, Zheng int index = 0; 27163fd0a558SYan, Zheng 27173fd0a558SYan, Zheng next = node; 27183fd0a558SYan, Zheng while (1) { 27193fd0a558SYan, Zheng cond_resched(); 27203fd0a558SYan, Zheng next = walk_up_backref(next, edges, &index); 27213fd0a558SYan, Zheng root = next->root; 27223fd0a558SYan, Zheng BUG_ON(!root); 27233fd0a558SYan, Zheng 272425985edcSLucas De Marchi /* no other choice for non-references counted tree */ 272527cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 27263fd0a558SYan, Zheng return root; 27273fd0a558SYan, Zheng 27283fd0a558SYan, Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) 27293fd0a558SYan, Zheng fs_root = root; 27303fd0a558SYan, Zheng 27313fd0a558SYan, Zheng if (next != node) 27323fd0a558SYan, Zheng return NULL; 27333fd0a558SYan, Zheng 27343fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 27353fd0a558SYan, Zheng if (!next || next->level <= node->level) 27363fd0a558SYan, Zheng break; 27373fd0a558SYan, Zheng } 27383fd0a558SYan, Zheng 27393fd0a558SYan, Zheng if (!fs_root) 27403fd0a558SYan, Zheng return ERR_PTR(-ENOENT); 27413fd0a558SYan, Zheng return fs_root; 27425d4f98a2SYan Zheng } 27435d4f98a2SYan Zheng 27445d4f98a2SYan Zheng static noinline_for_stack 27453fd0a558SYan, Zheng u64 calcu_metadata_size(struct reloc_control *rc, 27463fd0a558SYan, Zheng struct backref_node *node, int reserve) 27475d4f98a2SYan Zheng { 27480b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 27493fd0a558SYan, Zheng struct backref_node *next = node; 27503fd0a558SYan, Zheng struct backref_edge *edge; 27513fd0a558SYan, Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 27523fd0a558SYan, Zheng u64 num_bytes = 0; 27533fd0a558SYan, Zheng int index = 0; 27545d4f98a2SYan Zheng 27553fd0a558SYan, Zheng BUG_ON(reserve && node->processed); 27563fd0a558SYan, Zheng 27573fd0a558SYan, Zheng while (next) { 27583fd0a558SYan, Zheng cond_resched(); 27595d4f98a2SYan Zheng while (1) { 27603fd0a558SYan, Zheng if (next->processed && (reserve || next != node)) 27615d4f98a2SYan Zheng break; 27625d4f98a2SYan Zheng 27630b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 27643fd0a558SYan, Zheng 27653fd0a558SYan, Zheng if (list_empty(&next->upper)) 27663fd0a558SYan, Zheng break; 27673fd0a558SYan, Zheng 27683fd0a558SYan, Zheng edge = list_entry(next->upper.next, 27693fd0a558SYan, Zheng struct backref_edge, list[LOWER]); 27703fd0a558SYan, Zheng edges[index++] = edge; 27713fd0a558SYan, Zheng next = edge->node[UPPER]; 27725d4f98a2SYan Zheng } 27733fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 27743fd0a558SYan, Zheng } 27753fd0a558SYan, Zheng return num_bytes; 27763fd0a558SYan, Zheng } 27773fd0a558SYan, Zheng 27783fd0a558SYan, Zheng static int reserve_metadata_space(struct btrfs_trans_handle *trans, 27793fd0a558SYan, Zheng struct reloc_control *rc, 27803fd0a558SYan, Zheng struct backref_node *node) 27813fd0a558SYan, Zheng { 27823fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 2783da17066cSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 27843fd0a558SYan, Zheng u64 num_bytes; 27853fd0a558SYan, Zheng int ret; 27860647bf56SWang Shilong u64 tmp; 27873fd0a558SYan, Zheng 27883fd0a558SYan, Zheng num_bytes = calcu_metadata_size(rc, node, 1) * 2; 27893fd0a558SYan, Zheng 27903fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 27910647bf56SWang Shilong rc->reserved_bytes += num_bytes; 27928ca17f0fSJosef Bacik 27938ca17f0fSJosef Bacik /* 27948ca17f0fSJosef Bacik * We are under a transaction here so we can only do limited flushing. 27958ca17f0fSJosef Bacik * If we get an enospc just kick back -EAGAIN so we know to drop the 27968ca17f0fSJosef Bacik * transaction and try to refill when we can flush all the things. 27978ca17f0fSJosef Bacik */ 27980647bf56SWang Shilong ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes, 27998ca17f0fSJosef Bacik BTRFS_RESERVE_FLUSH_LIMIT); 28003fd0a558SYan, Zheng if (ret) { 2801da17066cSJeff Mahoney tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES; 28020647bf56SWang Shilong while (tmp <= rc->reserved_bytes) 28030647bf56SWang Shilong tmp <<= 1; 28040647bf56SWang Shilong /* 28050647bf56SWang Shilong * only one thread can access block_rsv at this point, 28060647bf56SWang Shilong * so we don't need hold lock to protect block_rsv. 28070647bf56SWang Shilong * we expand more reservation size here to allow enough 280852042d8eSAndrea Gelmini * space for relocation and we will return earlier in 28090647bf56SWang Shilong * enospc case. 28100647bf56SWang Shilong */ 2811da17066cSJeff Mahoney rc->block_rsv->size = tmp + fs_info->nodesize * 28120647bf56SWang Shilong RELOCATION_RESERVED_NODES; 28138ca17f0fSJosef Bacik return -EAGAIN; 28143fd0a558SYan, Zheng } 28153fd0a558SYan, Zheng 28163fd0a558SYan, Zheng return 0; 28173fd0a558SYan, Zheng } 28183fd0a558SYan, Zheng 28195d4f98a2SYan Zheng /* 28205d4f98a2SYan Zheng * relocate a block tree, and then update pointers in upper level 28215d4f98a2SYan Zheng * blocks that reference the block to point to the new location. 28225d4f98a2SYan Zheng * 28235d4f98a2SYan Zheng * if called by link_to_upper, the block has already been relocated. 28245d4f98a2SYan Zheng * in that case this function just updates pointers. 28255d4f98a2SYan Zheng */ 28265d4f98a2SYan Zheng static int do_relocation(struct btrfs_trans_handle *trans, 28273fd0a558SYan, Zheng struct reloc_control *rc, 28285d4f98a2SYan Zheng struct backref_node *node, 28295d4f98a2SYan Zheng struct btrfs_key *key, 28305d4f98a2SYan Zheng struct btrfs_path *path, int lowest) 28315d4f98a2SYan Zheng { 28322ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 28335d4f98a2SYan Zheng struct backref_node *upper; 28345d4f98a2SYan Zheng struct backref_edge *edge; 28355d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 28365d4f98a2SYan Zheng struct btrfs_root *root; 28375d4f98a2SYan Zheng struct extent_buffer *eb; 28385d4f98a2SYan Zheng u32 blocksize; 28395d4f98a2SYan Zheng u64 bytenr; 28405d4f98a2SYan Zheng u64 generation; 28415d4f98a2SYan Zheng int slot; 28425d4f98a2SYan Zheng int ret; 28435d4f98a2SYan Zheng int err = 0; 28445d4f98a2SYan Zheng 28455d4f98a2SYan Zheng BUG_ON(lowest && node->eb); 28465d4f98a2SYan Zheng 28475d4f98a2SYan Zheng path->lowest_level = node->level + 1; 28483fd0a558SYan, Zheng rc->backref_cache.path[node->level] = node; 28495d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) { 2850581c1760SQu Wenruo struct btrfs_key first_key; 285182fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 2852581c1760SQu Wenruo 28535d4f98a2SYan Zheng cond_resched(); 28545d4f98a2SYan Zheng 28555d4f98a2SYan Zheng upper = edge->node[UPPER]; 2856dc4103f9SWang Shilong root = select_reloc_root(trans, rc, upper, edges); 28573fd0a558SYan, Zheng BUG_ON(!root); 28585d4f98a2SYan Zheng 28593fd0a558SYan, Zheng if (upper->eb && !upper->locked) { 28603fd0a558SYan, Zheng if (!lowest) { 28613fd0a558SYan, Zheng ret = btrfs_bin_search(upper->eb, key, 28623fd0a558SYan, Zheng upper->level, &slot); 2863cbca7d59SFilipe Manana if (ret < 0) { 2864cbca7d59SFilipe Manana err = ret; 2865cbca7d59SFilipe Manana goto next; 2866cbca7d59SFilipe Manana } 28673fd0a558SYan, Zheng BUG_ON(ret); 28683fd0a558SYan, Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 28693fd0a558SYan, Zheng if (node->eb->start == bytenr) 28703fd0a558SYan, Zheng goto next; 28713fd0a558SYan, Zheng } 28725d4f98a2SYan Zheng drop_node_buffer(upper); 28733fd0a558SYan, Zheng } 28745d4f98a2SYan Zheng 28755d4f98a2SYan Zheng if (!upper->eb) { 28765d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 28773561b9dbSLiu Bo if (ret) { 28783561b9dbSLiu Bo if (ret < 0) 28795d4f98a2SYan Zheng err = ret; 28803561b9dbSLiu Bo else 28813561b9dbSLiu Bo err = -ENOENT; 28823561b9dbSLiu Bo 28833561b9dbSLiu Bo btrfs_release_path(path); 28845d4f98a2SYan Zheng break; 28855d4f98a2SYan Zheng } 28865d4f98a2SYan Zheng 28873fd0a558SYan, Zheng if (!upper->eb) { 28883fd0a558SYan, Zheng upper->eb = path->nodes[upper->level]; 28893fd0a558SYan, Zheng path->nodes[upper->level] = NULL; 28903fd0a558SYan, Zheng } else { 28913fd0a558SYan, Zheng BUG_ON(upper->eb != path->nodes[upper->level]); 28923fd0a558SYan, Zheng } 28933fd0a558SYan, Zheng 28943fd0a558SYan, Zheng upper->locked = 1; 28953fd0a558SYan, Zheng path->locks[upper->level] = 0; 28963fd0a558SYan, Zheng 28975d4f98a2SYan Zheng slot = path->slots[upper->level]; 2898b3b4aa74SDavid Sterba btrfs_release_path(path); 28995d4f98a2SYan Zheng } else { 29005d4f98a2SYan Zheng ret = btrfs_bin_search(upper->eb, key, upper->level, 29015d4f98a2SYan Zheng &slot); 2902cbca7d59SFilipe Manana if (ret < 0) { 2903cbca7d59SFilipe Manana err = ret; 2904cbca7d59SFilipe Manana goto next; 2905cbca7d59SFilipe Manana } 29065d4f98a2SYan Zheng BUG_ON(ret); 29075d4f98a2SYan Zheng } 29085d4f98a2SYan Zheng 29095d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 29103fd0a558SYan, Zheng if (lowest) { 29114547f4d8SLiu Bo if (bytenr != node->bytenr) { 29124547f4d8SLiu Bo btrfs_err(root->fs_info, 29134547f4d8SLiu Bo "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu", 29144547f4d8SLiu Bo bytenr, node->bytenr, slot, 29154547f4d8SLiu Bo upper->eb->start); 29164547f4d8SLiu Bo err = -EIO; 29174547f4d8SLiu Bo goto next; 29184547f4d8SLiu Bo } 29195d4f98a2SYan Zheng } else { 29203fd0a558SYan, Zheng if (node->eb->start == bytenr) 29213fd0a558SYan, Zheng goto next; 29225d4f98a2SYan Zheng } 29235d4f98a2SYan Zheng 2924da17066cSJeff Mahoney blocksize = root->fs_info->nodesize; 29255d4f98a2SYan Zheng generation = btrfs_node_ptr_generation(upper->eb, slot); 2926581c1760SQu Wenruo btrfs_node_key_to_cpu(upper->eb, &first_key, slot); 2927581c1760SQu Wenruo eb = read_tree_block(fs_info, bytenr, generation, 2928581c1760SQu Wenruo upper->level - 1, &first_key); 292964c043deSLiu Bo if (IS_ERR(eb)) { 293064c043deSLiu Bo err = PTR_ERR(eb); 293164c043deSLiu Bo goto next; 293264c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2933416bc658SJosef Bacik free_extent_buffer(eb); 293497d9a8a4STsutomu Itoh err = -EIO; 293597d9a8a4STsutomu Itoh goto next; 293697d9a8a4STsutomu Itoh } 29375d4f98a2SYan Zheng btrfs_tree_lock(eb); 29388bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 29395d4f98a2SYan Zheng 29405d4f98a2SYan Zheng if (!node->eb) { 29415d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, eb, upper->eb, 29425d4f98a2SYan Zheng slot, &eb); 29433fd0a558SYan, Zheng btrfs_tree_unlock(eb); 29443fd0a558SYan, Zheng free_extent_buffer(eb); 29455d4f98a2SYan Zheng if (ret < 0) { 29465d4f98a2SYan Zheng err = ret; 29473fd0a558SYan, Zheng goto next; 29485d4f98a2SYan Zheng } 29493fd0a558SYan, Zheng BUG_ON(node->eb != eb); 29505d4f98a2SYan Zheng } else { 29515d4f98a2SYan Zheng btrfs_set_node_blockptr(upper->eb, slot, 29525d4f98a2SYan Zheng node->eb->start); 29535d4f98a2SYan Zheng btrfs_set_node_ptr_generation(upper->eb, slot, 29545d4f98a2SYan Zheng trans->transid); 29555d4f98a2SYan Zheng btrfs_mark_buffer_dirty(upper->eb); 29565d4f98a2SYan Zheng 295782fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, 29585d4f98a2SYan Zheng node->eb->start, blocksize, 295982fa113fSQu Wenruo upper->eb->start); 296082fa113fSQu Wenruo ref.real_root = root->root_key.objectid; 296182fa113fSQu Wenruo btrfs_init_tree_ref(&ref, node->level, 296282fa113fSQu Wenruo btrfs_header_owner(upper->eb)); 296382fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 29645d4f98a2SYan Zheng BUG_ON(ret); 29655d4f98a2SYan Zheng 29665d4f98a2SYan Zheng ret = btrfs_drop_subtree(trans, root, eb, upper->eb); 29675d4f98a2SYan Zheng BUG_ON(ret); 29685d4f98a2SYan Zheng } 29693fd0a558SYan, Zheng next: 29703fd0a558SYan, Zheng if (!upper->pending) 29713fd0a558SYan, Zheng drop_node_buffer(upper); 29723fd0a558SYan, Zheng else 29733fd0a558SYan, Zheng unlock_node_buffer(upper); 29743fd0a558SYan, Zheng if (err) 29753fd0a558SYan, Zheng break; 29765d4f98a2SYan Zheng } 29773fd0a558SYan, Zheng 29783fd0a558SYan, Zheng if (!err && node->pending) { 29793fd0a558SYan, Zheng drop_node_buffer(node); 29803fd0a558SYan, Zheng list_move_tail(&node->list, &rc->backref_cache.changed); 29813fd0a558SYan, Zheng node->pending = 0; 29825d4f98a2SYan Zheng } 29833fd0a558SYan, Zheng 29845d4f98a2SYan Zheng path->lowest_level = 0; 29853fd0a558SYan, Zheng BUG_ON(err == -ENOSPC); 29865d4f98a2SYan Zheng return err; 29875d4f98a2SYan Zheng } 29885d4f98a2SYan Zheng 29895d4f98a2SYan Zheng static int link_to_upper(struct btrfs_trans_handle *trans, 29903fd0a558SYan, Zheng struct reloc_control *rc, 29915d4f98a2SYan Zheng struct backref_node *node, 29925d4f98a2SYan Zheng struct btrfs_path *path) 29935d4f98a2SYan Zheng { 29945d4f98a2SYan Zheng struct btrfs_key key; 29955d4f98a2SYan Zheng 29965d4f98a2SYan Zheng btrfs_node_key_to_cpu(node->eb, &key, 0); 29973fd0a558SYan, Zheng return do_relocation(trans, rc, node, &key, path, 0); 29985d4f98a2SYan Zheng } 29995d4f98a2SYan Zheng 30005d4f98a2SYan Zheng static int finish_pending_nodes(struct btrfs_trans_handle *trans, 30013fd0a558SYan, Zheng struct reloc_control *rc, 30023fd0a558SYan, Zheng struct btrfs_path *path, int err) 30035d4f98a2SYan Zheng { 30043fd0a558SYan, Zheng LIST_HEAD(list); 30053fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 30065d4f98a2SYan Zheng struct backref_node *node; 30075d4f98a2SYan Zheng int level; 30085d4f98a2SYan Zheng int ret; 30095d4f98a2SYan Zheng 30105d4f98a2SYan Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 30115d4f98a2SYan Zheng while (!list_empty(&cache->pending[level])) { 30125d4f98a2SYan Zheng node = list_entry(cache->pending[level].next, 30133fd0a558SYan, Zheng struct backref_node, list); 30143fd0a558SYan, Zheng list_move_tail(&node->list, &list); 30153fd0a558SYan, Zheng BUG_ON(!node->pending); 30165d4f98a2SYan Zheng 30173fd0a558SYan, Zheng if (!err) { 30183fd0a558SYan, Zheng ret = link_to_upper(trans, rc, node, path); 30195d4f98a2SYan Zheng if (ret < 0) 30205d4f98a2SYan Zheng err = ret; 30215d4f98a2SYan Zheng } 30225d4f98a2SYan Zheng } 30233fd0a558SYan, Zheng list_splice_init(&list, &cache->pending[level]); 30243fd0a558SYan, Zheng } 30255d4f98a2SYan Zheng return err; 30265d4f98a2SYan Zheng } 30275d4f98a2SYan Zheng 30285d4f98a2SYan Zheng static void mark_block_processed(struct reloc_control *rc, 30293fd0a558SYan, Zheng u64 bytenr, u32 blocksize) 30303fd0a558SYan, Zheng { 30313fd0a558SYan, Zheng set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1, 3032ceeb0ae7SDavid Sterba EXTENT_DIRTY); 30333fd0a558SYan, Zheng } 30343fd0a558SYan, Zheng 30353fd0a558SYan, Zheng static void __mark_block_processed(struct reloc_control *rc, 30365d4f98a2SYan Zheng struct backref_node *node) 30375d4f98a2SYan Zheng { 30385d4f98a2SYan Zheng u32 blocksize; 30395d4f98a2SYan Zheng if (node->level == 0 || 30405d4f98a2SYan Zheng in_block_group(node->bytenr, rc->block_group)) { 3041da17066cSJeff Mahoney blocksize = rc->extent_root->fs_info->nodesize; 30423fd0a558SYan, Zheng mark_block_processed(rc, node->bytenr, blocksize); 30435d4f98a2SYan Zheng } 30445d4f98a2SYan Zheng node->processed = 1; 30455d4f98a2SYan Zheng } 30465d4f98a2SYan Zheng 30475d4f98a2SYan Zheng /* 30485d4f98a2SYan Zheng * mark a block and all blocks directly/indirectly reference the block 30495d4f98a2SYan Zheng * as processed. 30505d4f98a2SYan Zheng */ 30515d4f98a2SYan Zheng static void update_processed_blocks(struct reloc_control *rc, 30525d4f98a2SYan Zheng struct backref_node *node) 30535d4f98a2SYan Zheng { 30545d4f98a2SYan Zheng struct backref_node *next = node; 30555d4f98a2SYan Zheng struct backref_edge *edge; 30565d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 30575d4f98a2SYan Zheng int index = 0; 30585d4f98a2SYan Zheng 30595d4f98a2SYan Zheng while (next) { 30605d4f98a2SYan Zheng cond_resched(); 30615d4f98a2SYan Zheng while (1) { 30625d4f98a2SYan Zheng if (next->processed) 30635d4f98a2SYan Zheng break; 30645d4f98a2SYan Zheng 30653fd0a558SYan, Zheng __mark_block_processed(rc, next); 30665d4f98a2SYan Zheng 30675d4f98a2SYan Zheng if (list_empty(&next->upper)) 30685d4f98a2SYan Zheng break; 30695d4f98a2SYan Zheng 30705d4f98a2SYan Zheng edge = list_entry(next->upper.next, 30715d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 30725d4f98a2SYan Zheng edges[index++] = edge; 30735d4f98a2SYan Zheng next = edge->node[UPPER]; 30745d4f98a2SYan Zheng } 30755d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 30765d4f98a2SYan Zheng } 30775d4f98a2SYan Zheng } 30785d4f98a2SYan Zheng 30797476dfdaSDavid Sterba static int tree_block_processed(u64 bytenr, struct reloc_control *rc) 30805d4f98a2SYan Zheng { 3081da17066cSJeff Mahoney u32 blocksize = rc->extent_root->fs_info->nodesize; 30827476dfdaSDavid Sterba 30835d4f98a2SYan Zheng if (test_range_bit(&rc->processed_blocks, bytenr, 30849655d298SChris Mason bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL)) 30855d4f98a2SYan Zheng return 1; 30865d4f98a2SYan Zheng return 0; 30875d4f98a2SYan Zheng } 30885d4f98a2SYan Zheng 30892ff7e61eSJeff Mahoney static int get_tree_block_key(struct btrfs_fs_info *fs_info, 30905d4f98a2SYan Zheng struct tree_block *block) 30915d4f98a2SYan Zheng { 30925d4f98a2SYan Zheng struct extent_buffer *eb; 30935d4f98a2SYan Zheng 3094581c1760SQu Wenruo eb = read_tree_block(fs_info, block->bytenr, block->key.offset, 3095581c1760SQu Wenruo block->level, NULL); 309664c043deSLiu Bo if (IS_ERR(eb)) { 309764c043deSLiu Bo return PTR_ERR(eb); 309864c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 3099416bc658SJosef Bacik free_extent_buffer(eb); 3100416bc658SJosef Bacik return -EIO; 3101416bc658SJosef Bacik } 31025d4f98a2SYan Zheng if (block->level == 0) 31035d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &block->key, 0); 31045d4f98a2SYan Zheng else 31055d4f98a2SYan Zheng btrfs_node_key_to_cpu(eb, &block->key, 0); 31065d4f98a2SYan Zheng free_extent_buffer(eb); 31075d4f98a2SYan Zheng block->key_ready = 1; 31085d4f98a2SYan Zheng return 0; 31095d4f98a2SYan Zheng } 31105d4f98a2SYan Zheng 31115d4f98a2SYan Zheng /* 31125d4f98a2SYan Zheng * helper function to relocate a tree block 31135d4f98a2SYan Zheng */ 31145d4f98a2SYan Zheng static int relocate_tree_block(struct btrfs_trans_handle *trans, 31155d4f98a2SYan Zheng struct reloc_control *rc, 31165d4f98a2SYan Zheng struct backref_node *node, 31175d4f98a2SYan Zheng struct btrfs_key *key, 31185d4f98a2SYan Zheng struct btrfs_path *path) 31195d4f98a2SYan Zheng { 31205d4f98a2SYan Zheng struct btrfs_root *root; 31213fd0a558SYan, Zheng int ret = 0; 31225d4f98a2SYan Zheng 31233fd0a558SYan, Zheng if (!node) 31245d4f98a2SYan Zheng return 0; 31253fd0a558SYan, Zheng 31263fd0a558SYan, Zheng BUG_ON(node->processed); 3127147d256eSZhaolei root = select_one_root(node); 31283fd0a558SYan, Zheng if (root == ERR_PTR(-ENOENT)) { 31293fd0a558SYan, Zheng update_processed_blocks(rc, node); 31303fd0a558SYan, Zheng goto out; 31315d4f98a2SYan Zheng } 31325d4f98a2SYan Zheng 313327cdeb70SMiao Xie if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 31343fd0a558SYan, Zheng ret = reserve_metadata_space(trans, rc, node); 31353fd0a558SYan, Zheng if (ret) 31365d4f98a2SYan Zheng goto out; 31375d4f98a2SYan Zheng } 31383fd0a558SYan, Zheng 31393fd0a558SYan, Zheng if (root) { 314027cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 31413fd0a558SYan, Zheng BUG_ON(node->new_bytenr); 31423fd0a558SYan, Zheng BUG_ON(!list_empty(&node->list)); 31433fd0a558SYan, Zheng btrfs_record_root_in_trans(trans, root); 31443fd0a558SYan, Zheng root = root->reloc_root; 31453fd0a558SYan, Zheng node->new_bytenr = root->node->start; 314600246528SJosef Bacik btrfs_put_root(node->root); 314700246528SJosef Bacik node->root = btrfs_grab_root(root); 31480b530bc5SJosef Bacik ASSERT(node->root); 31493fd0a558SYan, Zheng list_add_tail(&node->list, &rc->backref_cache.changed); 31503fd0a558SYan, Zheng } else { 31515d4f98a2SYan Zheng path->lowest_level = node->level; 31525d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 3153b3b4aa74SDavid Sterba btrfs_release_path(path); 31543fd0a558SYan, Zheng if (ret > 0) 31555d4f98a2SYan Zheng ret = 0; 31563fd0a558SYan, Zheng } 31573fd0a558SYan, Zheng if (!ret) 31583fd0a558SYan, Zheng update_processed_blocks(rc, node); 31593fd0a558SYan, Zheng } else { 31603fd0a558SYan, Zheng ret = do_relocation(trans, rc, node, key, path, 1); 31613fd0a558SYan, Zheng } 31625d4f98a2SYan Zheng out: 31630647bf56SWang Shilong if (ret || node->level == 0 || node->cowonly) 31643fd0a558SYan, Zheng remove_backref_node(&rc->backref_cache, node); 31655d4f98a2SYan Zheng return ret; 31665d4f98a2SYan Zheng } 31675d4f98a2SYan Zheng 31685d4f98a2SYan Zheng /* 31695d4f98a2SYan Zheng * relocate a list of blocks 31705d4f98a2SYan Zheng */ 31715d4f98a2SYan Zheng static noinline_for_stack 31725d4f98a2SYan Zheng int relocate_tree_blocks(struct btrfs_trans_handle *trans, 31735d4f98a2SYan Zheng struct reloc_control *rc, struct rb_root *blocks) 31745d4f98a2SYan Zheng { 31752ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 31765d4f98a2SYan Zheng struct backref_node *node; 31775d4f98a2SYan Zheng struct btrfs_path *path; 31785d4f98a2SYan Zheng struct tree_block *block; 317998ff7b94SQu Wenruo struct tree_block *next; 31805d4f98a2SYan Zheng int ret; 31815d4f98a2SYan Zheng int err = 0; 31825d4f98a2SYan Zheng 31835d4f98a2SYan Zheng path = btrfs_alloc_path(); 3184e1a12670SLiu Bo if (!path) { 3185e1a12670SLiu Bo err = -ENOMEM; 318634c2b290SDavid Sterba goto out_free_blocks; 3187e1a12670SLiu Bo } 31885d4f98a2SYan Zheng 318998ff7b94SQu Wenruo /* Kick in readahead for tree blocks with missing keys */ 319098ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 31915d4f98a2SYan Zheng if (!block->key_ready) 31922ff7e61eSJeff Mahoney readahead_tree_block(fs_info, block->bytenr); 31935d4f98a2SYan Zheng } 31945d4f98a2SYan Zheng 319598ff7b94SQu Wenruo /* Get first keys */ 319698ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 319734c2b290SDavid Sterba if (!block->key_ready) { 31982ff7e61eSJeff Mahoney err = get_tree_block_key(fs_info, block); 319934c2b290SDavid Sterba if (err) 320034c2b290SDavid Sterba goto out_free_path; 320134c2b290SDavid Sterba } 32025d4f98a2SYan Zheng } 32035d4f98a2SYan Zheng 320498ff7b94SQu Wenruo /* Do tree relocation */ 320598ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 32063fd0a558SYan, Zheng node = build_backref_tree(rc, &block->key, 32075d4f98a2SYan Zheng block->level, block->bytenr); 32085d4f98a2SYan Zheng if (IS_ERR(node)) { 32095d4f98a2SYan Zheng err = PTR_ERR(node); 32105d4f98a2SYan Zheng goto out; 32115d4f98a2SYan Zheng } 32125d4f98a2SYan Zheng 32135d4f98a2SYan Zheng ret = relocate_tree_block(trans, rc, node, &block->key, 32145d4f98a2SYan Zheng path); 32155d4f98a2SYan Zheng if (ret < 0) { 321698ff7b94SQu Wenruo if (ret != -EAGAIN || &block->rb_node == rb_first(blocks)) 32175d4f98a2SYan Zheng err = ret; 32185d4f98a2SYan Zheng goto out; 32195d4f98a2SYan Zheng } 32205d4f98a2SYan Zheng } 32215d4f98a2SYan Zheng out: 32223fd0a558SYan, Zheng err = finish_pending_nodes(trans, rc, path, err); 32235d4f98a2SYan Zheng 322434c2b290SDavid Sterba out_free_path: 32255d4f98a2SYan Zheng btrfs_free_path(path); 322634c2b290SDavid Sterba out_free_blocks: 3227e1a12670SLiu Bo free_block_list(blocks); 32285d4f98a2SYan Zheng return err; 32295d4f98a2SYan Zheng } 32305d4f98a2SYan Zheng 32315d4f98a2SYan Zheng static noinline_for_stack 3232efa56464SYan, Zheng int prealloc_file_extent_cluster(struct inode *inode, 3233efa56464SYan, Zheng struct file_extent_cluster *cluster) 3234efa56464SYan, Zheng { 3235efa56464SYan, Zheng u64 alloc_hint = 0; 3236efa56464SYan, Zheng u64 start; 3237efa56464SYan, Zheng u64 end; 3238efa56464SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 3239efa56464SYan, Zheng u64 num_bytes; 3240efa56464SYan, Zheng int nr = 0; 3241efa56464SYan, Zheng int ret = 0; 3242dcb40c19SWang Xiaoguang u64 prealloc_start = cluster->start - offset; 3243dcb40c19SWang Xiaoguang u64 prealloc_end = cluster->end - offset; 324418513091SWang Xiaoguang u64 cur_offset; 3245364ecf36SQu Wenruo struct extent_changeset *data_reserved = NULL; 3246efa56464SYan, Zheng 3247efa56464SYan, Zheng BUG_ON(cluster->start != cluster->boundary[0]); 32485955102cSAl Viro inode_lock(inode); 3249efa56464SYan, Zheng 3250364ecf36SQu Wenruo ret = btrfs_check_data_free_space(inode, &data_reserved, prealloc_start, 3251dcb40c19SWang Xiaoguang prealloc_end + 1 - prealloc_start); 3252efa56464SYan, Zheng if (ret) 3253efa56464SYan, Zheng goto out; 3254efa56464SYan, Zheng 325518513091SWang Xiaoguang cur_offset = prealloc_start; 3256efa56464SYan, Zheng while (nr < cluster->nr) { 3257efa56464SYan, Zheng start = cluster->boundary[nr] - offset; 3258efa56464SYan, Zheng if (nr + 1 < cluster->nr) 3259efa56464SYan, Zheng end = cluster->boundary[nr + 1] - 1 - offset; 3260efa56464SYan, Zheng else 3261efa56464SYan, Zheng end = cluster->end - offset; 3262efa56464SYan, Zheng 3263d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 3264efa56464SYan, Zheng num_bytes = end + 1 - start; 326518513091SWang Xiaoguang if (cur_offset < start) 3266bc42bda2SQu Wenruo btrfs_free_reserved_data_space(inode, data_reserved, 3267bc42bda2SQu Wenruo cur_offset, start - cur_offset); 3268efa56464SYan, Zheng ret = btrfs_prealloc_file_range(inode, 0, start, 3269efa56464SYan, Zheng num_bytes, num_bytes, 3270efa56464SYan, Zheng end + 1, &alloc_hint); 327118513091SWang Xiaoguang cur_offset = end + 1; 3272d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 3273efa56464SYan, Zheng if (ret) 3274efa56464SYan, Zheng break; 3275efa56464SYan, Zheng nr++; 3276efa56464SYan, Zheng } 327718513091SWang Xiaoguang if (cur_offset < prealloc_end) 3278bc42bda2SQu Wenruo btrfs_free_reserved_data_space(inode, data_reserved, 3279bc42bda2SQu Wenruo cur_offset, prealloc_end + 1 - cur_offset); 3280efa56464SYan, Zheng out: 32815955102cSAl Viro inode_unlock(inode); 3282364ecf36SQu Wenruo extent_changeset_free(data_reserved); 3283efa56464SYan, Zheng return ret; 3284efa56464SYan, Zheng } 3285efa56464SYan, Zheng 3286efa56464SYan, Zheng static noinline_for_stack 32870257bb82SYan, Zheng int setup_extent_mapping(struct inode *inode, u64 start, u64 end, 32880257bb82SYan, Zheng u64 block_start) 32895d4f98a2SYan Zheng { 32905d4f98a2SYan Zheng struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 32915d4f98a2SYan Zheng struct extent_map *em; 32920257bb82SYan, Zheng int ret = 0; 32935d4f98a2SYan Zheng 3294172ddd60SDavid Sterba em = alloc_extent_map(); 32950257bb82SYan, Zheng if (!em) 32960257bb82SYan, Zheng return -ENOMEM; 32970257bb82SYan, Zheng 32985d4f98a2SYan Zheng em->start = start; 32990257bb82SYan, Zheng em->len = end + 1 - start; 33000257bb82SYan, Zheng em->block_len = em->len; 33010257bb82SYan, Zheng em->block_start = block_start; 33025d4f98a2SYan Zheng set_bit(EXTENT_FLAG_PINNED, &em->flags); 33035d4f98a2SYan Zheng 3304d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 33055d4f98a2SYan Zheng while (1) { 3306890871beSChris Mason write_lock(&em_tree->lock); 330709a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 3308890871beSChris Mason write_unlock(&em_tree->lock); 33095d4f98a2SYan Zheng if (ret != -EEXIST) { 33105d4f98a2SYan Zheng free_extent_map(em); 33115d4f98a2SYan Zheng break; 33125d4f98a2SYan Zheng } 3313dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0); 33145d4f98a2SYan Zheng } 3315d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 33160257bb82SYan, Zheng return ret; 33170257bb82SYan, Zheng } 33185d4f98a2SYan Zheng 3319726a3421SQu Wenruo /* 3320726a3421SQu Wenruo * Allow error injection to test balance cancellation 3321726a3421SQu Wenruo */ 3322726a3421SQu Wenruo int btrfs_should_cancel_balance(struct btrfs_fs_info *fs_info) 3323726a3421SQu Wenruo { 3324726a3421SQu Wenruo return atomic_read(&fs_info->balance_cancel_req); 3325726a3421SQu Wenruo } 3326726a3421SQu Wenruo ALLOW_ERROR_INJECTION(btrfs_should_cancel_balance, TRUE); 3327726a3421SQu Wenruo 33280257bb82SYan, Zheng static int relocate_file_extent_cluster(struct inode *inode, 33290257bb82SYan, Zheng struct file_extent_cluster *cluster) 33300257bb82SYan, Zheng { 33312ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 33320257bb82SYan, Zheng u64 page_start; 33330257bb82SYan, Zheng u64 page_end; 33340257bb82SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 33350257bb82SYan, Zheng unsigned long index; 33360257bb82SYan, Zheng unsigned long last_index; 33370257bb82SYan, Zheng struct page *page; 33380257bb82SYan, Zheng struct file_ra_state *ra; 33393b16a4e3SJosef Bacik gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 33400257bb82SYan, Zheng int nr = 0; 33410257bb82SYan, Zheng int ret = 0; 33420257bb82SYan, Zheng 33430257bb82SYan, Zheng if (!cluster->nr) 33440257bb82SYan, Zheng return 0; 33450257bb82SYan, Zheng 33460257bb82SYan, Zheng ra = kzalloc(sizeof(*ra), GFP_NOFS); 33470257bb82SYan, Zheng if (!ra) 33480257bb82SYan, Zheng return -ENOMEM; 33490257bb82SYan, Zheng 3350efa56464SYan, Zheng ret = prealloc_file_extent_cluster(inode, cluster); 33510257bb82SYan, Zheng if (ret) 3352efa56464SYan, Zheng goto out; 33530257bb82SYan, Zheng 33540257bb82SYan, Zheng file_ra_state_init(ra, inode->i_mapping); 33550257bb82SYan, Zheng 3356efa56464SYan, Zheng ret = setup_extent_mapping(inode, cluster->start - offset, 3357efa56464SYan, Zheng cluster->end - offset, cluster->start); 3358efa56464SYan, Zheng if (ret) 3359efa56464SYan, Zheng goto out; 3360efa56464SYan, Zheng 336109cbfeafSKirill A. Shutemov index = (cluster->start - offset) >> PAGE_SHIFT; 336209cbfeafSKirill A. Shutemov last_index = (cluster->end - offset) >> PAGE_SHIFT; 33630257bb82SYan, Zheng while (index <= last_index) { 33649f3db423SNikolay Borisov ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), 33659f3db423SNikolay Borisov PAGE_SIZE); 3366efa56464SYan, Zheng if (ret) 3367efa56464SYan, Zheng goto out; 3368efa56464SYan, Zheng 33690257bb82SYan, Zheng page = find_lock_page(inode->i_mapping, index); 33700257bb82SYan, Zheng if (!page) { 33710257bb82SYan, Zheng page_cache_sync_readahead(inode->i_mapping, 33720257bb82SYan, Zheng ra, NULL, index, 33730257bb82SYan, Zheng last_index + 1 - index); 3374a94733d0SJosef Bacik page = find_or_create_page(inode->i_mapping, index, 33753b16a4e3SJosef Bacik mask); 33760257bb82SYan, Zheng if (!page) { 3377691fa059SNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 337843b18595SQu Wenruo PAGE_SIZE, true); 337944db1216SFilipe Manana btrfs_delalloc_release_extents(BTRFS_I(inode), 33808702ba93SQu Wenruo PAGE_SIZE); 33810257bb82SYan, Zheng ret = -ENOMEM; 3382efa56464SYan, Zheng goto out; 33830257bb82SYan, Zheng } 33840257bb82SYan, Zheng } 33850257bb82SYan, Zheng 33860257bb82SYan, Zheng if (PageReadahead(page)) { 33870257bb82SYan, Zheng page_cache_async_readahead(inode->i_mapping, 33880257bb82SYan, Zheng ra, NULL, page, index, 33890257bb82SYan, Zheng last_index + 1 - index); 33900257bb82SYan, Zheng } 33910257bb82SYan, Zheng 33920257bb82SYan, Zheng if (!PageUptodate(page)) { 33930257bb82SYan, Zheng btrfs_readpage(NULL, page); 33940257bb82SYan, Zheng lock_page(page); 33950257bb82SYan, Zheng if (!PageUptodate(page)) { 33960257bb82SYan, Zheng unlock_page(page); 339709cbfeafSKirill A. Shutemov put_page(page); 3398691fa059SNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 339943b18595SQu Wenruo PAGE_SIZE, true); 34008b62f87bSJosef Bacik btrfs_delalloc_release_extents(BTRFS_I(inode), 34018702ba93SQu Wenruo PAGE_SIZE); 34020257bb82SYan, Zheng ret = -EIO; 3403efa56464SYan, Zheng goto out; 34040257bb82SYan, Zheng } 34050257bb82SYan, Zheng } 34060257bb82SYan, Zheng 34074eee4fa4SMiao Xie page_start = page_offset(page); 340809cbfeafSKirill A. Shutemov page_end = page_start + PAGE_SIZE - 1; 34090257bb82SYan, Zheng 3410d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end); 34110257bb82SYan, Zheng 34120257bb82SYan, Zheng set_page_extent_mapped(page); 34130257bb82SYan, Zheng 34140257bb82SYan, Zheng if (nr < cluster->nr && 34150257bb82SYan, Zheng page_start + offset == cluster->boundary[nr]) { 34160257bb82SYan, Zheng set_extent_bits(&BTRFS_I(inode)->io_tree, 34170257bb82SYan, Zheng page_start, page_end, 3418ceeb0ae7SDavid Sterba EXTENT_BOUNDARY); 34190257bb82SYan, Zheng nr++; 34200257bb82SYan, Zheng } 34210257bb82SYan, Zheng 3422765f3cebSNikolay Borisov ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0, 3423330a5827SNikolay Borisov NULL); 3424765f3cebSNikolay Borisov if (ret) { 3425765f3cebSNikolay Borisov unlock_page(page); 3426765f3cebSNikolay Borisov put_page(page); 3427765f3cebSNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 342843b18595SQu Wenruo PAGE_SIZE, true); 3429765f3cebSNikolay Borisov btrfs_delalloc_release_extents(BTRFS_I(inode), 34308702ba93SQu Wenruo PAGE_SIZE); 3431765f3cebSNikolay Borisov 3432765f3cebSNikolay Borisov clear_extent_bits(&BTRFS_I(inode)->io_tree, 3433765f3cebSNikolay Borisov page_start, page_end, 3434765f3cebSNikolay Borisov EXTENT_LOCKED | EXTENT_BOUNDARY); 3435765f3cebSNikolay Borisov goto out; 3436765f3cebSNikolay Borisov 3437765f3cebSNikolay Borisov } 34380257bb82SYan, Zheng set_page_dirty(page); 34390257bb82SYan, Zheng 34400257bb82SYan, Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 3441d0082371SJeff Mahoney page_start, page_end); 34420257bb82SYan, Zheng unlock_page(page); 344309cbfeafSKirill A. Shutemov put_page(page); 34440257bb82SYan, Zheng 34450257bb82SYan, Zheng index++; 34468702ba93SQu Wenruo btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE); 3447efa56464SYan, Zheng balance_dirty_pages_ratelimited(inode->i_mapping); 34482ff7e61eSJeff Mahoney btrfs_throttle(fs_info); 34497f913c7cSQu Wenruo if (btrfs_should_cancel_balance(fs_info)) { 34507f913c7cSQu Wenruo ret = -ECANCELED; 34517f913c7cSQu Wenruo goto out; 34527f913c7cSQu Wenruo } 34530257bb82SYan, Zheng } 34540257bb82SYan, Zheng WARN_ON(nr != cluster->nr); 3455efa56464SYan, Zheng out: 34560257bb82SYan, Zheng kfree(ra); 34570257bb82SYan, Zheng return ret; 34580257bb82SYan, Zheng } 34590257bb82SYan, Zheng 34600257bb82SYan, Zheng static noinline_for_stack 34610257bb82SYan, Zheng int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key, 34620257bb82SYan, Zheng struct file_extent_cluster *cluster) 34630257bb82SYan, Zheng { 34640257bb82SYan, Zheng int ret; 34650257bb82SYan, Zheng 34660257bb82SYan, Zheng if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) { 34670257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 34680257bb82SYan, Zheng if (ret) 34690257bb82SYan, Zheng return ret; 34700257bb82SYan, Zheng cluster->nr = 0; 34710257bb82SYan, Zheng } 34720257bb82SYan, Zheng 34730257bb82SYan, Zheng if (!cluster->nr) 34740257bb82SYan, Zheng cluster->start = extent_key->objectid; 34750257bb82SYan, Zheng else 34760257bb82SYan, Zheng BUG_ON(cluster->nr >= MAX_EXTENTS); 34770257bb82SYan, Zheng cluster->end = extent_key->objectid + extent_key->offset - 1; 34780257bb82SYan, Zheng cluster->boundary[cluster->nr] = extent_key->objectid; 34790257bb82SYan, Zheng cluster->nr++; 34800257bb82SYan, Zheng 34810257bb82SYan, Zheng if (cluster->nr >= MAX_EXTENTS) { 34820257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 34830257bb82SYan, Zheng if (ret) 34840257bb82SYan, Zheng return ret; 34850257bb82SYan, Zheng cluster->nr = 0; 34860257bb82SYan, Zheng } 34870257bb82SYan, Zheng return 0; 34885d4f98a2SYan Zheng } 34895d4f98a2SYan Zheng 34905d4f98a2SYan Zheng /* 34915d4f98a2SYan Zheng * helper to add a tree block to the list. 34925d4f98a2SYan Zheng * the major work is getting the generation and level of the block 34935d4f98a2SYan Zheng */ 34945d4f98a2SYan Zheng static int add_tree_block(struct reloc_control *rc, 34955d4f98a2SYan Zheng struct btrfs_key *extent_key, 34965d4f98a2SYan Zheng struct btrfs_path *path, 34975d4f98a2SYan Zheng struct rb_root *blocks) 34985d4f98a2SYan Zheng { 34995d4f98a2SYan Zheng struct extent_buffer *eb; 35005d4f98a2SYan Zheng struct btrfs_extent_item *ei; 35015d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 35025d4f98a2SYan Zheng struct tree_block *block; 35035d4f98a2SYan Zheng struct rb_node *rb_node; 35045d4f98a2SYan Zheng u32 item_size; 35055d4f98a2SYan Zheng int level = -1; 35067fdf4b60SWang Shilong u64 generation; 35075d4f98a2SYan Zheng 35085d4f98a2SYan Zheng eb = path->nodes[0]; 35095d4f98a2SYan Zheng item_size = btrfs_item_size_nr(eb, path->slots[0]); 35105d4f98a2SYan Zheng 35113173a18fSJosef Bacik if (extent_key->type == BTRFS_METADATA_ITEM_KEY || 35123173a18fSJosef Bacik item_size >= sizeof(*ei) + sizeof(*bi)) { 35135d4f98a2SYan Zheng ei = btrfs_item_ptr(eb, path->slots[0], 35145d4f98a2SYan Zheng struct btrfs_extent_item); 35153173a18fSJosef Bacik if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) { 35165d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 35175d4f98a2SYan Zheng level = btrfs_tree_block_level(eb, bi); 35185d4f98a2SYan Zheng } else { 35193173a18fSJosef Bacik level = (int)extent_key->offset; 35203173a18fSJosef Bacik } 35213173a18fSJosef Bacik generation = btrfs_extent_generation(eb, ei); 35226d8ff4e4SDavid Sterba } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) { 3523ba3c2b19SNikolay Borisov btrfs_print_v0_err(eb->fs_info); 3524ba3c2b19SNikolay Borisov btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL); 3525ba3c2b19SNikolay Borisov return -EINVAL; 35263173a18fSJosef Bacik } else { 35275d4f98a2SYan Zheng BUG(); 35285d4f98a2SYan Zheng } 35295d4f98a2SYan Zheng 3530b3b4aa74SDavid Sterba btrfs_release_path(path); 35315d4f98a2SYan Zheng 35325d4f98a2SYan Zheng BUG_ON(level == -1); 35335d4f98a2SYan Zheng 35345d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 35355d4f98a2SYan Zheng if (!block) 35365d4f98a2SYan Zheng return -ENOMEM; 35375d4f98a2SYan Zheng 35385d4f98a2SYan Zheng block->bytenr = extent_key->objectid; 3539da17066cSJeff Mahoney block->key.objectid = rc->extent_root->fs_info->nodesize; 35405d4f98a2SYan Zheng block->key.offset = generation; 35415d4f98a2SYan Zheng block->level = level; 35425d4f98a2SYan Zheng block->key_ready = 0; 35435d4f98a2SYan Zheng 35445d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, &block->rb_node); 354543c04fb1SJeff Mahoney if (rb_node) 354643c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, block->bytenr); 35475d4f98a2SYan Zheng 35485d4f98a2SYan Zheng return 0; 35495d4f98a2SYan Zheng } 35505d4f98a2SYan Zheng 35515d4f98a2SYan Zheng /* 35525d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY 35535d4f98a2SYan Zheng */ 35545d4f98a2SYan Zheng static int __add_tree_block(struct reloc_control *rc, 35555d4f98a2SYan Zheng u64 bytenr, u32 blocksize, 35565d4f98a2SYan Zheng struct rb_root *blocks) 35575d4f98a2SYan Zheng { 35580b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 35595d4f98a2SYan Zheng struct btrfs_path *path; 35605d4f98a2SYan Zheng struct btrfs_key key; 35615d4f98a2SYan Zheng int ret; 35620b246afaSJeff Mahoney bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA); 35635d4f98a2SYan Zheng 35647476dfdaSDavid Sterba if (tree_block_processed(bytenr, rc)) 35655d4f98a2SYan Zheng return 0; 35665d4f98a2SYan Zheng 35675d4f98a2SYan Zheng if (tree_search(blocks, bytenr)) 35685d4f98a2SYan Zheng return 0; 35695d4f98a2SYan Zheng 35705d4f98a2SYan Zheng path = btrfs_alloc_path(); 35715d4f98a2SYan Zheng if (!path) 35725d4f98a2SYan Zheng return -ENOMEM; 3573aee68ee5SJosef Bacik again: 35745d4f98a2SYan Zheng key.objectid = bytenr; 3575aee68ee5SJosef Bacik if (skinny) { 3576aee68ee5SJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 3577aee68ee5SJosef Bacik key.offset = (u64)-1; 3578aee68ee5SJosef Bacik } else { 35795d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 35805d4f98a2SYan Zheng key.offset = blocksize; 3581aee68ee5SJosef Bacik } 35825d4f98a2SYan Zheng 35835d4f98a2SYan Zheng path->search_commit_root = 1; 35845d4f98a2SYan Zheng path->skip_locking = 1; 35855d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0); 35865d4f98a2SYan Zheng if (ret < 0) 35875d4f98a2SYan Zheng goto out; 35885d4f98a2SYan Zheng 3589aee68ee5SJosef Bacik if (ret > 0 && skinny) { 3590aee68ee5SJosef Bacik if (path->slots[0]) { 3591aee68ee5SJosef Bacik path->slots[0]--; 3592aee68ee5SJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, 3593aee68ee5SJosef Bacik path->slots[0]); 35943173a18fSJosef Bacik if (key.objectid == bytenr && 3595aee68ee5SJosef Bacik (key.type == BTRFS_METADATA_ITEM_KEY || 3596aee68ee5SJosef Bacik (key.type == BTRFS_EXTENT_ITEM_KEY && 3597aee68ee5SJosef Bacik key.offset == blocksize))) 35983173a18fSJosef Bacik ret = 0; 35993173a18fSJosef Bacik } 3600aee68ee5SJosef Bacik 3601aee68ee5SJosef Bacik if (ret) { 3602aee68ee5SJosef Bacik skinny = false; 3603aee68ee5SJosef Bacik btrfs_release_path(path); 3604aee68ee5SJosef Bacik goto again; 3605aee68ee5SJosef Bacik } 3606aee68ee5SJosef Bacik } 3607cdccee99SLiu Bo if (ret) { 3608cdccee99SLiu Bo ASSERT(ret == 1); 3609cdccee99SLiu Bo btrfs_print_leaf(path->nodes[0]); 3610cdccee99SLiu Bo btrfs_err(fs_info, 3611cdccee99SLiu Bo "tree block extent item (%llu) is not found in extent tree", 3612cdccee99SLiu Bo bytenr); 3613cdccee99SLiu Bo WARN_ON(1); 3614cdccee99SLiu Bo ret = -EINVAL; 3615cdccee99SLiu Bo goto out; 3616cdccee99SLiu Bo } 36173173a18fSJosef Bacik 36185d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, blocks); 36195d4f98a2SYan Zheng out: 36205d4f98a2SYan Zheng btrfs_free_path(path); 36215d4f98a2SYan Zheng return ret; 36225d4f98a2SYan Zheng } 36235d4f98a2SYan Zheng 36240af3d00bSJosef Bacik static int delete_block_group_cache(struct btrfs_fs_info *fs_info, 362532da5386SDavid Sterba struct btrfs_block_group *block_group, 36261bbc621eSChris Mason struct inode *inode, 36271bbc621eSChris Mason u64 ino) 36280af3d00bSJosef Bacik { 36290af3d00bSJosef Bacik struct btrfs_key key; 36300af3d00bSJosef Bacik struct btrfs_root *root = fs_info->tree_root; 36310af3d00bSJosef Bacik struct btrfs_trans_handle *trans; 36320af3d00bSJosef Bacik int ret = 0; 36330af3d00bSJosef Bacik 36340af3d00bSJosef Bacik if (inode) 36350af3d00bSJosef Bacik goto truncate; 36360af3d00bSJosef Bacik 36370af3d00bSJosef Bacik key.objectid = ino; 36380af3d00bSJosef Bacik key.type = BTRFS_INODE_ITEM_KEY; 36390af3d00bSJosef Bacik key.offset = 0; 36400af3d00bSJosef Bacik 36414c66e0d4SDavid Sterba inode = btrfs_iget(fs_info->sb, &key, root); 36422e19f1f9SAl Viro if (IS_ERR(inode)) 36430af3d00bSJosef Bacik return -ENOENT; 36440af3d00bSJosef Bacik 36450af3d00bSJosef Bacik truncate: 36462ff7e61eSJeff Mahoney ret = btrfs_check_trunc_cache_free_space(fs_info, 36477b61cd92SMiao Xie &fs_info->global_block_rsv); 36487b61cd92SMiao Xie if (ret) 36497b61cd92SMiao Xie goto out; 36507b61cd92SMiao Xie 36517a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 36520af3d00bSJosef Bacik if (IS_ERR(trans)) { 36533612b495STsutomu Itoh ret = PTR_ERR(trans); 36540af3d00bSJosef Bacik goto out; 36550af3d00bSJosef Bacik } 36560af3d00bSJosef Bacik 365777ab86bfSJeff Mahoney ret = btrfs_truncate_free_space_cache(trans, block_group, inode); 36580af3d00bSJosef Bacik 36593a45bb20SJeff Mahoney btrfs_end_transaction(trans); 36602ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 36610af3d00bSJosef Bacik out: 36620af3d00bSJosef Bacik iput(inode); 36630af3d00bSJosef Bacik return ret; 36640af3d00bSJosef Bacik } 36650af3d00bSJosef Bacik 36665d4f98a2SYan Zheng /* 366719b546d7SQu Wenruo * Locate the free space cache EXTENT_DATA in root tree leaf and delete the 366819b546d7SQu Wenruo * cache inode, to avoid free space cache data extent blocking data relocation. 36695d4f98a2SYan Zheng */ 367019b546d7SQu Wenruo static int delete_v1_space_cache(struct extent_buffer *leaf, 367119b546d7SQu Wenruo struct btrfs_block_group *block_group, 367219b546d7SQu Wenruo u64 data_bytenr) 36735d4f98a2SYan Zheng { 367419b546d7SQu Wenruo u64 space_cache_ino; 367519b546d7SQu Wenruo struct btrfs_file_extent_item *ei; 36765d4f98a2SYan Zheng struct btrfs_key key; 367719b546d7SQu Wenruo bool found = false; 367819b546d7SQu Wenruo int i; 36795d4f98a2SYan Zheng int ret; 36805d4f98a2SYan Zheng 368119b546d7SQu Wenruo if (btrfs_header_owner(leaf) != BTRFS_ROOT_TREE_OBJECTID) 368219b546d7SQu Wenruo return 0; 36835d4f98a2SYan Zheng 368419b546d7SQu Wenruo for (i = 0; i < btrfs_header_nritems(leaf); i++) { 368519b546d7SQu Wenruo btrfs_item_key_to_cpu(leaf, &key, i); 368619b546d7SQu Wenruo if (key.type != BTRFS_EXTENT_DATA_KEY) 368719b546d7SQu Wenruo continue; 368819b546d7SQu Wenruo ei = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 368919b546d7SQu Wenruo if (btrfs_file_extent_type(leaf, ei) == BTRFS_FILE_EXTENT_REG && 369019b546d7SQu Wenruo btrfs_file_extent_disk_bytenr(leaf, ei) == data_bytenr) { 369119b546d7SQu Wenruo found = true; 369219b546d7SQu Wenruo space_cache_ino = key.objectid; 369319b546d7SQu Wenruo break; 369419b546d7SQu Wenruo } 369519b546d7SQu Wenruo } 369619b546d7SQu Wenruo if (!found) 369719b546d7SQu Wenruo return -ENOENT; 369819b546d7SQu Wenruo ret = delete_block_group_cache(leaf->fs_info, block_group, NULL, 369919b546d7SQu Wenruo space_cache_ino); 37000af3d00bSJosef Bacik return ret; 37015d4f98a2SYan Zheng } 37025d4f98a2SYan Zheng 37035d4f98a2SYan Zheng /* 37042c016dc2SLiu Bo * helper to find all tree blocks that reference a given data extent 37055d4f98a2SYan Zheng */ 37065d4f98a2SYan Zheng static noinline_for_stack 37075d4f98a2SYan Zheng int add_data_references(struct reloc_control *rc, 37085d4f98a2SYan Zheng struct btrfs_key *extent_key, 37095d4f98a2SYan Zheng struct btrfs_path *path, 37105d4f98a2SYan Zheng struct rb_root *blocks) 37115d4f98a2SYan Zheng { 371219b546d7SQu Wenruo struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 371319b546d7SQu Wenruo struct ulist *leaves = NULL; 371419b546d7SQu Wenruo struct ulist_iterator leaf_uiter; 371519b546d7SQu Wenruo struct ulist_node *ref_node = NULL; 371619b546d7SQu Wenruo const u32 blocksize = fs_info->nodesize; 3717647f63bdSFilipe David Borba Manana int ret = 0; 37185d4f98a2SYan Zheng 3719b3b4aa74SDavid Sterba btrfs_release_path(path); 372019b546d7SQu Wenruo ret = btrfs_find_all_leafs(NULL, fs_info, extent_key->objectid, 372119b546d7SQu Wenruo 0, &leaves, NULL, true); 372219b546d7SQu Wenruo if (ret < 0) 372319b546d7SQu Wenruo return ret; 372419b546d7SQu Wenruo 372519b546d7SQu Wenruo ULIST_ITER_INIT(&leaf_uiter); 372619b546d7SQu Wenruo while ((ref_node = ulist_next(leaves, &leaf_uiter))) { 372719b546d7SQu Wenruo struct extent_buffer *eb; 372819b546d7SQu Wenruo 372919b546d7SQu Wenruo eb = read_tree_block(fs_info, ref_node->val, 0, 0, NULL); 373019b546d7SQu Wenruo if (IS_ERR(eb)) { 373119b546d7SQu Wenruo ret = PTR_ERR(eb); 373219b546d7SQu Wenruo break; 373319b546d7SQu Wenruo } 373419b546d7SQu Wenruo ret = delete_v1_space_cache(eb, rc->block_group, 373519b546d7SQu Wenruo extent_key->objectid); 373619b546d7SQu Wenruo free_extent_buffer(eb); 373719b546d7SQu Wenruo if (ret < 0) 373819b546d7SQu Wenruo break; 373919b546d7SQu Wenruo ret = __add_tree_block(rc, ref_node->val, blocksize, blocks); 374019b546d7SQu Wenruo if (ret < 0) 374119b546d7SQu Wenruo break; 374219b546d7SQu Wenruo } 374319b546d7SQu Wenruo if (ret < 0) 37445d4f98a2SYan Zheng free_block_list(blocks); 374519b546d7SQu Wenruo ulist_free(leaves); 374619b546d7SQu Wenruo return ret; 37475d4f98a2SYan Zheng } 37485d4f98a2SYan Zheng 37495d4f98a2SYan Zheng /* 37502c016dc2SLiu Bo * helper to find next unprocessed extent 37515d4f98a2SYan Zheng */ 37525d4f98a2SYan Zheng static noinline_for_stack 3753147d256eSZhaolei int find_next_extent(struct reloc_control *rc, struct btrfs_path *path, 37543fd0a558SYan, Zheng struct btrfs_key *extent_key) 37555d4f98a2SYan Zheng { 37560b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 37575d4f98a2SYan Zheng struct btrfs_key key; 37585d4f98a2SYan Zheng struct extent_buffer *leaf; 37595d4f98a2SYan Zheng u64 start, end, last; 37605d4f98a2SYan Zheng int ret; 37615d4f98a2SYan Zheng 3762b3470b5dSDavid Sterba last = rc->block_group->start + rc->block_group->length; 37635d4f98a2SYan Zheng while (1) { 37645d4f98a2SYan Zheng cond_resched(); 37655d4f98a2SYan Zheng if (rc->search_start >= last) { 37665d4f98a2SYan Zheng ret = 1; 37675d4f98a2SYan Zheng break; 37685d4f98a2SYan Zheng } 37695d4f98a2SYan Zheng 37705d4f98a2SYan Zheng key.objectid = rc->search_start; 37715d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 37725d4f98a2SYan Zheng key.offset = 0; 37735d4f98a2SYan Zheng 37745d4f98a2SYan Zheng path->search_commit_root = 1; 37755d4f98a2SYan Zheng path->skip_locking = 1; 37765d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 37775d4f98a2SYan Zheng 0, 0); 37785d4f98a2SYan Zheng if (ret < 0) 37795d4f98a2SYan Zheng break; 37805d4f98a2SYan Zheng next: 37815d4f98a2SYan Zheng leaf = path->nodes[0]; 37825d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(leaf)) { 37835d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 37845d4f98a2SYan Zheng if (ret != 0) 37855d4f98a2SYan Zheng break; 37865d4f98a2SYan Zheng leaf = path->nodes[0]; 37875d4f98a2SYan Zheng } 37885d4f98a2SYan Zheng 37895d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 37905d4f98a2SYan Zheng if (key.objectid >= last) { 37915d4f98a2SYan Zheng ret = 1; 37925d4f98a2SYan Zheng break; 37935d4f98a2SYan Zheng } 37945d4f98a2SYan Zheng 37953173a18fSJosef Bacik if (key.type != BTRFS_EXTENT_ITEM_KEY && 37963173a18fSJosef Bacik key.type != BTRFS_METADATA_ITEM_KEY) { 37973173a18fSJosef Bacik path->slots[0]++; 37983173a18fSJosef Bacik goto next; 37993173a18fSJosef Bacik } 38003173a18fSJosef Bacik 38013173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 38025d4f98a2SYan Zheng key.objectid + key.offset <= rc->search_start) { 38035d4f98a2SYan Zheng path->slots[0]++; 38045d4f98a2SYan Zheng goto next; 38055d4f98a2SYan Zheng } 38065d4f98a2SYan Zheng 38073173a18fSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 38080b246afaSJeff Mahoney key.objectid + fs_info->nodesize <= 38093173a18fSJosef Bacik rc->search_start) { 38103173a18fSJosef Bacik path->slots[0]++; 38113173a18fSJosef Bacik goto next; 38123173a18fSJosef Bacik } 38133173a18fSJosef Bacik 38145d4f98a2SYan Zheng ret = find_first_extent_bit(&rc->processed_blocks, 38155d4f98a2SYan Zheng key.objectid, &start, &end, 3816e6138876SJosef Bacik EXTENT_DIRTY, NULL); 38175d4f98a2SYan Zheng 38185d4f98a2SYan Zheng if (ret == 0 && start <= key.objectid) { 3819b3b4aa74SDavid Sterba btrfs_release_path(path); 38205d4f98a2SYan Zheng rc->search_start = end + 1; 38215d4f98a2SYan Zheng } else { 38223173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) 38235d4f98a2SYan Zheng rc->search_start = key.objectid + key.offset; 38243173a18fSJosef Bacik else 38253173a18fSJosef Bacik rc->search_start = key.objectid + 38260b246afaSJeff Mahoney fs_info->nodesize; 38273fd0a558SYan, Zheng memcpy(extent_key, &key, sizeof(key)); 38285d4f98a2SYan Zheng return 0; 38295d4f98a2SYan Zheng } 38305d4f98a2SYan Zheng } 3831b3b4aa74SDavid Sterba btrfs_release_path(path); 38325d4f98a2SYan Zheng return ret; 38335d4f98a2SYan Zheng } 38345d4f98a2SYan Zheng 38355d4f98a2SYan Zheng static void set_reloc_control(struct reloc_control *rc) 38365d4f98a2SYan Zheng { 38375d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 38387585717fSChris Mason 38397585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 38405d4f98a2SYan Zheng fs_info->reloc_ctl = rc; 38417585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 38425d4f98a2SYan Zheng } 38435d4f98a2SYan Zheng 38445d4f98a2SYan Zheng static void unset_reloc_control(struct reloc_control *rc) 38455d4f98a2SYan Zheng { 38465d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 38477585717fSChris Mason 38487585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 38495d4f98a2SYan Zheng fs_info->reloc_ctl = NULL; 38507585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 38515d4f98a2SYan Zheng } 38525d4f98a2SYan Zheng 38535d4f98a2SYan Zheng static int check_extent_flags(u64 flags) 38545d4f98a2SYan Zheng { 38555d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 38565d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 38575d4f98a2SYan Zheng return 1; 38585d4f98a2SYan Zheng if (!(flags & BTRFS_EXTENT_FLAG_DATA) && 38595d4f98a2SYan Zheng !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 38605d4f98a2SYan Zheng return 1; 38615d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 38625d4f98a2SYan Zheng (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) 38635d4f98a2SYan Zheng return 1; 38645d4f98a2SYan Zheng return 0; 38655d4f98a2SYan Zheng } 38665d4f98a2SYan Zheng 38673fd0a558SYan, Zheng static noinline_for_stack 38683fd0a558SYan, Zheng int prepare_to_relocate(struct reloc_control *rc) 38693fd0a558SYan, Zheng { 38703fd0a558SYan, Zheng struct btrfs_trans_handle *trans; 3871ac2fabacSJosef Bacik int ret; 38723fd0a558SYan, Zheng 38732ff7e61eSJeff Mahoney rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info, 387466d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 38753fd0a558SYan, Zheng if (!rc->block_rsv) 38763fd0a558SYan, Zheng return -ENOMEM; 38773fd0a558SYan, Zheng 38783fd0a558SYan, Zheng memset(&rc->cluster, 0, sizeof(rc->cluster)); 3879b3470b5dSDavid Sterba rc->search_start = rc->block_group->start; 38803fd0a558SYan, Zheng rc->extents_found = 0; 38813fd0a558SYan, Zheng rc->nodes_relocated = 0; 38823fd0a558SYan, Zheng rc->merging_rsv_size = 0; 38830647bf56SWang Shilong rc->reserved_bytes = 0; 3884da17066cSJeff Mahoney rc->block_rsv->size = rc->extent_root->fs_info->nodesize * 38850647bf56SWang Shilong RELOCATION_RESERVED_NODES; 3886ac2fabacSJosef Bacik ret = btrfs_block_rsv_refill(rc->extent_root, 3887ac2fabacSJosef Bacik rc->block_rsv, rc->block_rsv->size, 3888ac2fabacSJosef Bacik BTRFS_RESERVE_FLUSH_ALL); 3889ac2fabacSJosef Bacik if (ret) 3890ac2fabacSJosef Bacik return ret; 38913fd0a558SYan, Zheng 38923fd0a558SYan, Zheng rc->create_reloc_tree = 1; 38933fd0a558SYan, Zheng set_reloc_control(rc); 38943fd0a558SYan, Zheng 38957a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 389628818947SLiu Bo if (IS_ERR(trans)) { 389728818947SLiu Bo unset_reloc_control(rc); 389828818947SLiu Bo /* 389928818947SLiu Bo * extent tree is not a ref_cow tree and has no reloc_root to 390028818947SLiu Bo * cleanup. And callers are responsible to free the above 390128818947SLiu Bo * block rsv. 390228818947SLiu Bo */ 390328818947SLiu Bo return PTR_ERR(trans); 390428818947SLiu Bo } 39053a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 39063fd0a558SYan, Zheng return 0; 39073fd0a558SYan, Zheng } 390876dda93cSYan, Zheng 39095d4f98a2SYan Zheng static noinline_for_stack int relocate_block_group(struct reloc_control *rc) 39105d4f98a2SYan Zheng { 39112ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 39125d4f98a2SYan Zheng struct rb_root blocks = RB_ROOT; 39135d4f98a2SYan Zheng struct btrfs_key key; 39145d4f98a2SYan Zheng struct btrfs_trans_handle *trans = NULL; 39155d4f98a2SYan Zheng struct btrfs_path *path; 39165d4f98a2SYan Zheng struct btrfs_extent_item *ei; 39175d4f98a2SYan Zheng u64 flags; 39185d4f98a2SYan Zheng u32 item_size; 39195d4f98a2SYan Zheng int ret; 39205d4f98a2SYan Zheng int err = 0; 3921c87f08caSChris Mason int progress = 0; 39225d4f98a2SYan Zheng 39235d4f98a2SYan Zheng path = btrfs_alloc_path(); 39243fd0a558SYan, Zheng if (!path) 39255d4f98a2SYan Zheng return -ENOMEM; 3926e4058b54SDavid Sterba path->reada = READA_FORWARD; 39273fd0a558SYan, Zheng 39283fd0a558SYan, Zheng ret = prepare_to_relocate(rc); 39293fd0a558SYan, Zheng if (ret) { 39303fd0a558SYan, Zheng err = ret; 39313fd0a558SYan, Zheng goto out_free; 39322423fdfbSJiri Slaby } 39335d4f98a2SYan Zheng 39345d4f98a2SYan Zheng while (1) { 39350647bf56SWang Shilong rc->reserved_bytes = 0; 39360647bf56SWang Shilong ret = btrfs_block_rsv_refill(rc->extent_root, 39370647bf56SWang Shilong rc->block_rsv, rc->block_rsv->size, 39380647bf56SWang Shilong BTRFS_RESERVE_FLUSH_ALL); 39390647bf56SWang Shilong if (ret) { 39400647bf56SWang Shilong err = ret; 39410647bf56SWang Shilong break; 39420647bf56SWang Shilong } 3943c87f08caSChris Mason progress++; 3944a22285a6SYan, Zheng trans = btrfs_start_transaction(rc->extent_root, 0); 39450f788c58SLiu Bo if (IS_ERR(trans)) { 39460f788c58SLiu Bo err = PTR_ERR(trans); 39470f788c58SLiu Bo trans = NULL; 39480f788c58SLiu Bo break; 39490f788c58SLiu Bo } 3950c87f08caSChris Mason restart: 39513fd0a558SYan, Zheng if (update_backref_cache(trans, &rc->backref_cache)) { 39523a45bb20SJeff Mahoney btrfs_end_transaction(trans); 395342a657f5SPan Bian trans = NULL; 39543fd0a558SYan, Zheng continue; 39553fd0a558SYan, Zheng } 39563fd0a558SYan, Zheng 3957147d256eSZhaolei ret = find_next_extent(rc, path, &key); 39585d4f98a2SYan Zheng if (ret < 0) 39595d4f98a2SYan Zheng err = ret; 39605d4f98a2SYan Zheng if (ret != 0) 39615d4f98a2SYan Zheng break; 39625d4f98a2SYan Zheng 39635d4f98a2SYan Zheng rc->extents_found++; 39645d4f98a2SYan Zheng 39655d4f98a2SYan Zheng ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 39665d4f98a2SYan Zheng struct btrfs_extent_item); 39673fd0a558SYan, Zheng item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); 39685d4f98a2SYan Zheng if (item_size >= sizeof(*ei)) { 39695d4f98a2SYan Zheng flags = btrfs_extent_flags(path->nodes[0], ei); 39705d4f98a2SYan Zheng ret = check_extent_flags(flags); 39715d4f98a2SYan Zheng BUG_ON(ret); 39726d8ff4e4SDavid Sterba } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) { 3973ba3c2b19SNikolay Borisov err = -EINVAL; 3974ba3c2b19SNikolay Borisov btrfs_print_v0_err(trans->fs_info); 3975ba3c2b19SNikolay Borisov btrfs_abort_transaction(trans, err); 3976ba3c2b19SNikolay Borisov break; 39775d4f98a2SYan Zheng } else { 39785d4f98a2SYan Zheng BUG(); 39795d4f98a2SYan Zheng } 39805d4f98a2SYan Zheng 39815d4f98a2SYan Zheng if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 39825d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, &blocks); 39835d4f98a2SYan Zheng } else if (rc->stage == UPDATE_DATA_PTRS && 39845d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 39855d4f98a2SYan Zheng ret = add_data_references(rc, &key, path, &blocks); 39865d4f98a2SYan Zheng } else { 3987b3b4aa74SDavid Sterba btrfs_release_path(path); 39885d4f98a2SYan Zheng ret = 0; 39895d4f98a2SYan Zheng } 39905d4f98a2SYan Zheng if (ret < 0) { 39913fd0a558SYan, Zheng err = ret; 39925d4f98a2SYan Zheng break; 39935d4f98a2SYan Zheng } 39945d4f98a2SYan Zheng 39955d4f98a2SYan Zheng if (!RB_EMPTY_ROOT(&blocks)) { 39965d4f98a2SYan Zheng ret = relocate_tree_blocks(trans, rc, &blocks); 39975d4f98a2SYan Zheng if (ret < 0) { 39981708cc57SWang Shilong /* 39991708cc57SWang Shilong * if we fail to relocate tree blocks, force to update 40001708cc57SWang Shilong * backref cache when committing transaction. 40011708cc57SWang Shilong */ 40021708cc57SWang Shilong rc->backref_cache.last_trans = trans->transid - 1; 40031708cc57SWang Shilong 40043fd0a558SYan, Zheng if (ret != -EAGAIN) { 40055d4f98a2SYan Zheng err = ret; 40065d4f98a2SYan Zheng break; 40075d4f98a2SYan Zheng } 40083fd0a558SYan, Zheng rc->extents_found--; 40093fd0a558SYan, Zheng rc->search_start = key.objectid; 40103fd0a558SYan, Zheng } 40115d4f98a2SYan Zheng } 40125d4f98a2SYan Zheng 40133a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 40142ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 40153fd0a558SYan, Zheng trans = NULL; 40165d4f98a2SYan Zheng 40175d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && 40185d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 40195d4f98a2SYan Zheng rc->found_file_extent = 1; 40200257bb82SYan, Zheng ret = relocate_data_extent(rc->data_inode, 40213fd0a558SYan, Zheng &key, &rc->cluster); 40225d4f98a2SYan Zheng if (ret < 0) { 40235d4f98a2SYan Zheng err = ret; 40245d4f98a2SYan Zheng break; 40255d4f98a2SYan Zheng } 40265d4f98a2SYan Zheng } 4027f31ea088SQu Wenruo if (btrfs_should_cancel_balance(fs_info)) { 4028f31ea088SQu Wenruo err = -ECANCELED; 4029f31ea088SQu Wenruo break; 4030f31ea088SQu Wenruo } 40315d4f98a2SYan Zheng } 4032c87f08caSChris Mason if (trans && progress && err == -ENOSPC) { 403343a7e99dSNikolay Borisov ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags); 40349689457bSShilong Wang if (ret == 1) { 4035c87f08caSChris Mason err = 0; 4036c87f08caSChris Mason progress = 0; 4037c87f08caSChris Mason goto restart; 4038c87f08caSChris Mason } 4039c87f08caSChris Mason } 40403fd0a558SYan, Zheng 4041b3b4aa74SDavid Sterba btrfs_release_path(path); 404291166212SDavid Sterba clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY); 40435d4f98a2SYan Zheng 40445d4f98a2SYan Zheng if (trans) { 40453a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 40462ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 40475d4f98a2SYan Zheng } 40485d4f98a2SYan Zheng 40490257bb82SYan, Zheng if (!err) { 40503fd0a558SYan, Zheng ret = relocate_file_extent_cluster(rc->data_inode, 40513fd0a558SYan, Zheng &rc->cluster); 40520257bb82SYan, Zheng if (ret < 0) 40530257bb82SYan, Zheng err = ret; 40540257bb82SYan, Zheng } 40550257bb82SYan, Zheng 40563fd0a558SYan, Zheng rc->create_reloc_tree = 0; 40573fd0a558SYan, Zheng set_reloc_control(rc); 40580257bb82SYan, Zheng 40593fd0a558SYan, Zheng backref_cache_cleanup(&rc->backref_cache); 406063f018beSNikolay Borisov btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL); 40615d4f98a2SYan Zheng 40627f913c7cSQu Wenruo /* 40637f913c7cSQu Wenruo * Even in the case when the relocation is cancelled, we should all go 40647f913c7cSQu Wenruo * through prepare_to_merge() and merge_reloc_roots(). 40657f913c7cSQu Wenruo * 40667f913c7cSQu Wenruo * For error (including cancelled balance), prepare_to_merge() will 40677f913c7cSQu Wenruo * mark all reloc trees orphan, then queue them for cleanup in 40687f913c7cSQu Wenruo * merge_reloc_roots() 40697f913c7cSQu Wenruo */ 40703fd0a558SYan, Zheng err = prepare_to_merge(rc, err); 40715d4f98a2SYan Zheng 40725d4f98a2SYan Zheng merge_reloc_roots(rc); 40735d4f98a2SYan Zheng 40743fd0a558SYan, Zheng rc->merge_reloc_tree = 0; 40755d4f98a2SYan Zheng unset_reloc_control(rc); 407663f018beSNikolay Borisov btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1, NULL); 40775d4f98a2SYan Zheng 40785d4f98a2SYan Zheng /* get rid of pinned extents */ 40797a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 408062b99540SQu Wenruo if (IS_ERR(trans)) { 40813612b495STsutomu Itoh err = PTR_ERR(trans); 408262b99540SQu Wenruo goto out_free; 408362b99540SQu Wenruo } 40843a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 4085*6217b0faSJosef Bacik out_free: 4086d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 4087d2311e69SQu Wenruo if (ret < 0 && !err) 4088d2311e69SQu Wenruo err = ret; 40892ff7e61eSJeff Mahoney btrfs_free_block_rsv(fs_info, rc->block_rsv); 40903fd0a558SYan, Zheng btrfs_free_path(path); 40915d4f98a2SYan Zheng return err; 40925d4f98a2SYan Zheng } 40935d4f98a2SYan Zheng 40945d4f98a2SYan Zheng static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 40950257bb82SYan, Zheng struct btrfs_root *root, u64 objectid) 40965d4f98a2SYan Zheng { 40975d4f98a2SYan Zheng struct btrfs_path *path; 40985d4f98a2SYan Zheng struct btrfs_inode_item *item; 40995d4f98a2SYan Zheng struct extent_buffer *leaf; 41005d4f98a2SYan Zheng int ret; 41015d4f98a2SYan Zheng 41025d4f98a2SYan Zheng path = btrfs_alloc_path(); 41035d4f98a2SYan Zheng if (!path) 41045d4f98a2SYan Zheng return -ENOMEM; 41055d4f98a2SYan Zheng 41065d4f98a2SYan Zheng ret = btrfs_insert_empty_inode(trans, root, path, objectid); 41075d4f98a2SYan Zheng if (ret) 41085d4f98a2SYan Zheng goto out; 41095d4f98a2SYan Zheng 41105d4f98a2SYan Zheng leaf = path->nodes[0]; 41115d4f98a2SYan Zheng item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); 4112b159fa28SDavid Sterba memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 41135d4f98a2SYan Zheng btrfs_set_inode_generation(leaf, item, 1); 41140257bb82SYan, Zheng btrfs_set_inode_size(leaf, item, 0); 41155d4f98a2SYan Zheng btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); 41163fd0a558SYan, Zheng btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS | 41173fd0a558SYan, Zheng BTRFS_INODE_PREALLOC); 41185d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 41195d4f98a2SYan Zheng out: 41205d4f98a2SYan Zheng btrfs_free_path(path); 41215d4f98a2SYan Zheng return ret; 41225d4f98a2SYan Zheng } 41235d4f98a2SYan Zheng 41245d4f98a2SYan Zheng /* 41255d4f98a2SYan Zheng * helper to create inode for data relocation. 41265d4f98a2SYan Zheng * the inode is in data relocation tree and its link count is 0 41275d4f98a2SYan Zheng */ 41283fd0a558SYan, Zheng static noinline_for_stack 41293fd0a558SYan, Zheng struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, 413032da5386SDavid Sterba struct btrfs_block_group *group) 41315d4f98a2SYan Zheng { 41325d4f98a2SYan Zheng struct inode *inode = NULL; 41335d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 41345d4f98a2SYan Zheng struct btrfs_root *root; 41355d4f98a2SYan Zheng struct btrfs_key key; 41364624900dSZhaolei u64 objectid; 41375d4f98a2SYan Zheng int err = 0; 41385d4f98a2SYan Zheng 41395d4f98a2SYan Zheng root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 41405d4f98a2SYan Zheng if (IS_ERR(root)) 41415d4f98a2SYan Zheng return ERR_CAST(root); 41425d4f98a2SYan Zheng 4143a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 6); 414476deacf0SJosef Bacik if (IS_ERR(trans)) { 414500246528SJosef Bacik btrfs_put_root(root); 41463fd0a558SYan, Zheng return ERR_CAST(trans); 414776deacf0SJosef Bacik } 41485d4f98a2SYan Zheng 4149581bb050SLi Zefan err = btrfs_find_free_objectid(root, &objectid); 41505d4f98a2SYan Zheng if (err) 41515d4f98a2SYan Zheng goto out; 41525d4f98a2SYan Zheng 41530257bb82SYan, Zheng err = __insert_orphan_inode(trans, root, objectid); 41545d4f98a2SYan Zheng BUG_ON(err); 41555d4f98a2SYan Zheng 41565d4f98a2SYan Zheng key.objectid = objectid; 41575d4f98a2SYan Zheng key.type = BTRFS_INODE_ITEM_KEY; 41585d4f98a2SYan Zheng key.offset = 0; 41594c66e0d4SDavid Sterba inode = btrfs_iget(fs_info->sb, &key, root); 41602e19f1f9SAl Viro BUG_ON(IS_ERR(inode)); 4161b3470b5dSDavid Sterba BTRFS_I(inode)->index_cnt = group->start; 41625d4f98a2SYan Zheng 416373f2e545SNikolay Borisov err = btrfs_orphan_add(trans, BTRFS_I(inode)); 41645d4f98a2SYan Zheng out: 416500246528SJosef Bacik btrfs_put_root(root); 41663a45bb20SJeff Mahoney btrfs_end_transaction(trans); 41672ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 41685d4f98a2SYan Zheng if (err) { 41695d4f98a2SYan Zheng if (inode) 41705d4f98a2SYan Zheng iput(inode); 41715d4f98a2SYan Zheng inode = ERR_PTR(err); 41725d4f98a2SYan Zheng } 41735d4f98a2SYan Zheng return inode; 41745d4f98a2SYan Zheng } 41755d4f98a2SYan Zheng 4176c258d6e3SQu Wenruo static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info) 41773fd0a558SYan, Zheng { 41783fd0a558SYan, Zheng struct reloc_control *rc; 41793fd0a558SYan, Zheng 41803fd0a558SYan, Zheng rc = kzalloc(sizeof(*rc), GFP_NOFS); 41813fd0a558SYan, Zheng if (!rc) 41823fd0a558SYan, Zheng return NULL; 41833fd0a558SYan, Zheng 41843fd0a558SYan, Zheng INIT_LIST_HEAD(&rc->reloc_roots); 4185d2311e69SQu Wenruo INIT_LIST_HEAD(&rc->dirty_subvol_roots); 41863fd0a558SYan, Zheng backref_cache_init(&rc->backref_cache); 41873fd0a558SYan, Zheng mapping_tree_init(&rc->reloc_root_tree); 418843eb5f29SQu Wenruo extent_io_tree_init(fs_info, &rc->processed_blocks, 418943eb5f29SQu Wenruo IO_TREE_RELOC_BLOCKS, NULL); 41903fd0a558SYan, Zheng return rc; 41913fd0a558SYan, Zheng } 41923fd0a558SYan, Zheng 41935d4f98a2SYan Zheng /* 4194ebce0e01SAdam Borowski * Print the block group being relocated 4195ebce0e01SAdam Borowski */ 4196ebce0e01SAdam Borowski static void describe_relocation(struct btrfs_fs_info *fs_info, 419732da5386SDavid Sterba struct btrfs_block_group *block_group) 4198ebce0e01SAdam Borowski { 4199f89e09cfSAnand Jain char buf[128] = {'\0'}; 4200ebce0e01SAdam Borowski 4201f89e09cfSAnand Jain btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf)); 4202ebce0e01SAdam Borowski 4203ebce0e01SAdam Borowski btrfs_info(fs_info, 4204ebce0e01SAdam Borowski "relocating block group %llu flags %s", 4205b3470b5dSDavid Sterba block_group->start, buf); 4206ebce0e01SAdam Borowski } 4207ebce0e01SAdam Borowski 4208430640e3SQu Wenruo static const char *stage_to_string(int stage) 4209430640e3SQu Wenruo { 4210430640e3SQu Wenruo if (stage == MOVE_DATA_EXTENTS) 4211430640e3SQu Wenruo return "move data extents"; 4212430640e3SQu Wenruo if (stage == UPDATE_DATA_PTRS) 4213430640e3SQu Wenruo return "update data pointers"; 4214430640e3SQu Wenruo return "unknown"; 4215430640e3SQu Wenruo } 4216430640e3SQu Wenruo 4217ebce0e01SAdam Borowski /* 42185d4f98a2SYan Zheng * function to relocate all extents in a block group. 42195d4f98a2SYan Zheng */ 42206bccf3abSJeff Mahoney int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) 42215d4f98a2SYan Zheng { 422232da5386SDavid Sterba struct btrfs_block_group *bg; 42236bccf3abSJeff Mahoney struct btrfs_root *extent_root = fs_info->extent_root; 42245d4f98a2SYan Zheng struct reloc_control *rc; 42250af3d00bSJosef Bacik struct inode *inode; 42260af3d00bSJosef Bacik struct btrfs_path *path; 42275d4f98a2SYan Zheng int ret; 4228f0486c68SYan, Zheng int rw = 0; 42295d4f98a2SYan Zheng int err = 0; 42305d4f98a2SYan Zheng 4231eede2bf3SOmar Sandoval bg = btrfs_lookup_block_group(fs_info, group_start); 4232eede2bf3SOmar Sandoval if (!bg) 4233eede2bf3SOmar Sandoval return -ENOENT; 4234eede2bf3SOmar Sandoval 4235eede2bf3SOmar Sandoval if (btrfs_pinned_by_swapfile(fs_info, bg)) { 4236eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 4237eede2bf3SOmar Sandoval return -ETXTBSY; 4238eede2bf3SOmar Sandoval } 4239eede2bf3SOmar Sandoval 4240c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 4241eede2bf3SOmar Sandoval if (!rc) { 4242eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 42435d4f98a2SYan Zheng return -ENOMEM; 4244eede2bf3SOmar Sandoval } 42455d4f98a2SYan Zheng 4246f0486c68SYan, Zheng rc->extent_root = extent_root; 4247eede2bf3SOmar Sandoval rc->block_group = bg; 42485d4f98a2SYan Zheng 4249b12de528SQu Wenruo ret = btrfs_inc_block_group_ro(rc->block_group, true); 4250f0486c68SYan, Zheng if (ret) { 4251f0486c68SYan, Zheng err = ret; 4252f0486c68SYan, Zheng goto out; 4253f0486c68SYan, Zheng } 4254f0486c68SYan, Zheng rw = 1; 4255f0486c68SYan, Zheng 42560af3d00bSJosef Bacik path = btrfs_alloc_path(); 42570af3d00bSJosef Bacik if (!path) { 42580af3d00bSJosef Bacik err = -ENOMEM; 42590af3d00bSJosef Bacik goto out; 42600af3d00bSJosef Bacik } 42610af3d00bSJosef Bacik 42627949f339SDavid Sterba inode = lookup_free_space_inode(rc->block_group, path); 42630af3d00bSJosef Bacik btrfs_free_path(path); 42640af3d00bSJosef Bacik 42650af3d00bSJosef Bacik if (!IS_ERR(inode)) 42661bbc621eSChris Mason ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0); 42670af3d00bSJosef Bacik else 42680af3d00bSJosef Bacik ret = PTR_ERR(inode); 42690af3d00bSJosef Bacik 42700af3d00bSJosef Bacik if (ret && ret != -ENOENT) { 42710af3d00bSJosef Bacik err = ret; 42720af3d00bSJosef Bacik goto out; 42730af3d00bSJosef Bacik } 42740af3d00bSJosef Bacik 42755d4f98a2SYan Zheng rc->data_inode = create_reloc_inode(fs_info, rc->block_group); 42765d4f98a2SYan Zheng if (IS_ERR(rc->data_inode)) { 42775d4f98a2SYan Zheng err = PTR_ERR(rc->data_inode); 42785d4f98a2SYan Zheng rc->data_inode = NULL; 42795d4f98a2SYan Zheng goto out; 42805d4f98a2SYan Zheng } 42815d4f98a2SYan Zheng 42820b246afaSJeff Mahoney describe_relocation(fs_info, rc->block_group); 42835d4f98a2SYan Zheng 42849cfa3e34SFilipe Manana btrfs_wait_block_group_reservations(rc->block_group); 4285f78c436cSFilipe Manana btrfs_wait_nocow_writers(rc->block_group); 42866374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 4287b3470b5dSDavid Sterba rc->block_group->start, 4288b3470b5dSDavid Sterba rc->block_group->length); 42895d4f98a2SYan Zheng 42905d4f98a2SYan Zheng while (1) { 4291430640e3SQu Wenruo int finishes_stage; 4292430640e3SQu Wenruo 429376dda93cSYan, Zheng mutex_lock(&fs_info->cleaner_mutex); 42945d4f98a2SYan Zheng ret = relocate_block_group(rc); 429576dda93cSYan, Zheng mutex_unlock(&fs_info->cleaner_mutex); 4296ff612ba7SJosef Bacik if (ret < 0) 42975d4f98a2SYan Zheng err = ret; 4298ff612ba7SJosef Bacik 4299430640e3SQu Wenruo finishes_stage = rc->stage; 4300ff612ba7SJosef Bacik /* 4301ff612ba7SJosef Bacik * We may have gotten ENOSPC after we already dirtied some 4302ff612ba7SJosef Bacik * extents. If writeout happens while we're relocating a 4303ff612ba7SJosef Bacik * different block group we could end up hitting the 4304ff612ba7SJosef Bacik * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in 4305ff612ba7SJosef Bacik * btrfs_reloc_cow_block. Make sure we write everything out 4306ff612ba7SJosef Bacik * properly so we don't trip over this problem, and then break 4307ff612ba7SJosef Bacik * out of the loop if we hit an error. 4308ff612ba7SJosef Bacik */ 4309ff612ba7SJosef Bacik if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) { 4310ff612ba7SJosef Bacik ret = btrfs_wait_ordered_range(rc->data_inode, 0, 4311ff612ba7SJosef Bacik (u64)-1); 4312ff612ba7SJosef Bacik if (ret) 4313ff612ba7SJosef Bacik err = ret; 4314ff612ba7SJosef Bacik invalidate_mapping_pages(rc->data_inode->i_mapping, 4315ff612ba7SJosef Bacik 0, -1); 4316ff612ba7SJosef Bacik rc->stage = UPDATE_DATA_PTRS; 43175d4f98a2SYan Zheng } 43185d4f98a2SYan Zheng 4319ff612ba7SJosef Bacik if (err < 0) 4320ff612ba7SJosef Bacik goto out; 4321ff612ba7SJosef Bacik 43225d4f98a2SYan Zheng if (rc->extents_found == 0) 43235d4f98a2SYan Zheng break; 43245d4f98a2SYan Zheng 4325430640e3SQu Wenruo btrfs_info(fs_info, "found %llu extents, stage: %s", 4326430640e3SQu Wenruo rc->extents_found, stage_to_string(finishes_stage)); 43275d4f98a2SYan Zheng } 43285d4f98a2SYan Zheng 43295d4f98a2SYan Zheng WARN_ON(rc->block_group->pinned > 0); 43305d4f98a2SYan Zheng WARN_ON(rc->block_group->reserved > 0); 4331bf38be65SDavid Sterba WARN_ON(rc->block_group->used > 0); 43325d4f98a2SYan Zheng out: 4333f0486c68SYan, Zheng if (err && rw) 43342ff7e61eSJeff Mahoney btrfs_dec_block_group_ro(rc->block_group); 43355d4f98a2SYan Zheng iput(rc->data_inode); 43365d4f98a2SYan Zheng btrfs_put_block_group(rc->block_group); 43375d4f98a2SYan Zheng kfree(rc); 43385d4f98a2SYan Zheng return err; 43395d4f98a2SYan Zheng } 43405d4f98a2SYan Zheng 434176dda93cSYan, Zheng static noinline_for_stack int mark_garbage_root(struct btrfs_root *root) 434276dda93cSYan, Zheng { 43430b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 434476dda93cSYan, Zheng struct btrfs_trans_handle *trans; 434579787eaaSJeff Mahoney int ret, err; 434676dda93cSYan, Zheng 43470b246afaSJeff Mahoney trans = btrfs_start_transaction(fs_info->tree_root, 0); 434879787eaaSJeff Mahoney if (IS_ERR(trans)) 434979787eaaSJeff Mahoney return PTR_ERR(trans); 435076dda93cSYan, Zheng 435176dda93cSYan, Zheng memset(&root->root_item.drop_progress, 0, 435276dda93cSYan, Zheng sizeof(root->root_item.drop_progress)); 435376dda93cSYan, Zheng root->root_item.drop_level = 0; 435476dda93cSYan, Zheng btrfs_set_root_refs(&root->root_item, 0); 43550b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 435676dda93cSYan, Zheng &root->root_key, &root->root_item); 435776dda93cSYan, Zheng 43583a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 435979787eaaSJeff Mahoney if (err) 436079787eaaSJeff Mahoney return err; 436179787eaaSJeff Mahoney return ret; 436276dda93cSYan, Zheng } 436376dda93cSYan, Zheng 43645d4f98a2SYan Zheng /* 43655d4f98a2SYan Zheng * recover relocation interrupted by system crash. 43665d4f98a2SYan Zheng * 43675d4f98a2SYan Zheng * this function resumes merging reloc trees with corresponding fs trees. 43685d4f98a2SYan Zheng * this is important for keeping the sharing of tree blocks 43695d4f98a2SYan Zheng */ 43705d4f98a2SYan Zheng int btrfs_recover_relocation(struct btrfs_root *root) 43715d4f98a2SYan Zheng { 43720b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 43735d4f98a2SYan Zheng LIST_HEAD(reloc_roots); 43745d4f98a2SYan Zheng struct btrfs_key key; 43755d4f98a2SYan Zheng struct btrfs_root *fs_root; 43765d4f98a2SYan Zheng struct btrfs_root *reloc_root; 43775d4f98a2SYan Zheng struct btrfs_path *path; 43785d4f98a2SYan Zheng struct extent_buffer *leaf; 43795d4f98a2SYan Zheng struct reloc_control *rc = NULL; 43805d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 43815d4f98a2SYan Zheng int ret; 43825d4f98a2SYan Zheng int err = 0; 43835d4f98a2SYan Zheng 43845d4f98a2SYan Zheng path = btrfs_alloc_path(); 43855d4f98a2SYan Zheng if (!path) 43865d4f98a2SYan Zheng return -ENOMEM; 4387e4058b54SDavid Sterba path->reada = READA_BACK; 43885d4f98a2SYan Zheng 43895d4f98a2SYan Zheng key.objectid = BTRFS_TREE_RELOC_OBJECTID; 43905d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 43915d4f98a2SYan Zheng key.offset = (u64)-1; 43925d4f98a2SYan Zheng 43935d4f98a2SYan Zheng while (1) { 43940b246afaSJeff Mahoney ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, 43955d4f98a2SYan Zheng path, 0, 0); 43965d4f98a2SYan Zheng if (ret < 0) { 43975d4f98a2SYan Zheng err = ret; 43985d4f98a2SYan Zheng goto out; 43995d4f98a2SYan Zheng } 44005d4f98a2SYan Zheng if (ret > 0) { 44015d4f98a2SYan Zheng if (path->slots[0] == 0) 44025d4f98a2SYan Zheng break; 44035d4f98a2SYan Zheng path->slots[0]--; 44045d4f98a2SYan Zheng } 44055d4f98a2SYan Zheng leaf = path->nodes[0]; 44065d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 4407b3b4aa74SDavid Sterba btrfs_release_path(path); 44085d4f98a2SYan Zheng 44095d4f98a2SYan Zheng if (key.objectid != BTRFS_TREE_RELOC_OBJECTID || 44105d4f98a2SYan Zheng key.type != BTRFS_ROOT_ITEM_KEY) 44115d4f98a2SYan Zheng break; 44125d4f98a2SYan Zheng 44133dbf1738SJosef Bacik reloc_root = btrfs_read_tree_root(root, &key); 44145d4f98a2SYan Zheng if (IS_ERR(reloc_root)) { 44155d4f98a2SYan Zheng err = PTR_ERR(reloc_root); 44165d4f98a2SYan Zheng goto out; 44175d4f98a2SYan Zheng } 44185d4f98a2SYan Zheng 44193dbf1738SJosef Bacik set_bit(BTRFS_ROOT_REF_COWS, &reloc_root->state); 44205d4f98a2SYan Zheng list_add(&reloc_root->root_list, &reloc_roots); 44215d4f98a2SYan Zheng 44225d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 44230b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, 44245d4f98a2SYan Zheng reloc_root->root_key.offset); 44255d4f98a2SYan Zheng if (IS_ERR(fs_root)) { 442676dda93cSYan, Zheng ret = PTR_ERR(fs_root); 442776dda93cSYan, Zheng if (ret != -ENOENT) { 442876dda93cSYan, Zheng err = ret; 44295d4f98a2SYan Zheng goto out; 44305d4f98a2SYan Zheng } 443179787eaaSJeff Mahoney ret = mark_garbage_root(reloc_root); 443279787eaaSJeff Mahoney if (ret < 0) { 443379787eaaSJeff Mahoney err = ret; 443479787eaaSJeff Mahoney goto out; 443579787eaaSJeff Mahoney } 4436932fd26dSJosef Bacik } else { 443700246528SJosef Bacik btrfs_put_root(fs_root); 443876dda93cSYan, Zheng } 44395d4f98a2SYan Zheng } 44405d4f98a2SYan Zheng 44415d4f98a2SYan Zheng if (key.offset == 0) 44425d4f98a2SYan Zheng break; 44435d4f98a2SYan Zheng 44445d4f98a2SYan Zheng key.offset--; 44455d4f98a2SYan Zheng } 4446b3b4aa74SDavid Sterba btrfs_release_path(path); 44475d4f98a2SYan Zheng 44485d4f98a2SYan Zheng if (list_empty(&reloc_roots)) 44495d4f98a2SYan Zheng goto out; 44505d4f98a2SYan Zheng 4451c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 44525d4f98a2SYan Zheng if (!rc) { 44535d4f98a2SYan Zheng err = -ENOMEM; 44545d4f98a2SYan Zheng goto out; 44555d4f98a2SYan Zheng } 44565d4f98a2SYan Zheng 44570b246afaSJeff Mahoney rc->extent_root = fs_info->extent_root; 44585d4f98a2SYan Zheng 44595d4f98a2SYan Zheng set_reloc_control(rc); 44605d4f98a2SYan Zheng 44617a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 44623612b495STsutomu Itoh if (IS_ERR(trans)) { 44633612b495STsutomu Itoh err = PTR_ERR(trans); 4464fb2d83eeSJosef Bacik goto out_unset; 44653612b495STsutomu Itoh } 44663fd0a558SYan, Zheng 44673fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 44683fd0a558SYan, Zheng 44695d4f98a2SYan Zheng while (!list_empty(&reloc_roots)) { 44705d4f98a2SYan Zheng reloc_root = list_entry(reloc_roots.next, 44715d4f98a2SYan Zheng struct btrfs_root, root_list); 44725d4f98a2SYan Zheng list_del(&reloc_root->root_list); 44735d4f98a2SYan Zheng 44745d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) == 0) { 44755d4f98a2SYan Zheng list_add_tail(&reloc_root->root_list, 44765d4f98a2SYan Zheng &rc->reloc_roots); 44775d4f98a2SYan Zheng continue; 44785d4f98a2SYan Zheng } 44795d4f98a2SYan Zheng 44800b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, reloc_root->root_key.offset); 448179787eaaSJeff Mahoney if (IS_ERR(fs_root)) { 448279787eaaSJeff Mahoney err = PTR_ERR(fs_root); 4483ca1aa281SJosef Bacik list_add_tail(&reloc_root->root_list, &reloc_roots); 4484fb2d83eeSJosef Bacik goto out_unset; 448579787eaaSJeff Mahoney } 44865d4f98a2SYan Zheng 4487ffd7b339SJeff Mahoney err = __add_reloc_root(reloc_root); 448879787eaaSJeff Mahoney BUG_ON(err < 0); /* -ENOMEM or logic error */ 44895d4f98a2SYan Zheng fs_root->reloc_root = reloc_root; 449000246528SJosef Bacik btrfs_put_root(fs_root); 44915d4f98a2SYan Zheng } 44925d4f98a2SYan Zheng 44933a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 449479787eaaSJeff Mahoney if (err) 4495fb2d83eeSJosef Bacik goto out_unset; 44965d4f98a2SYan Zheng 44975d4f98a2SYan Zheng merge_reloc_roots(rc); 44985d4f98a2SYan Zheng 44995d4f98a2SYan Zheng unset_reloc_control(rc); 45005d4f98a2SYan Zheng 45017a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 450262b99540SQu Wenruo if (IS_ERR(trans)) { 45033612b495STsutomu Itoh err = PTR_ERR(trans); 4504*6217b0faSJosef Bacik goto out_clean; 450562b99540SQu Wenruo } 45063a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 4507*6217b0faSJosef Bacik out_clean: 4508d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 4509d2311e69SQu Wenruo if (ret < 0 && !err) 4510d2311e69SQu Wenruo err = ret; 4511fb2d83eeSJosef Bacik out_unset: 4512fb2d83eeSJosef Bacik unset_reloc_control(rc); 45135d4f98a2SYan Zheng kfree(rc); 45143612b495STsutomu Itoh out: 4515aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 4516aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 4517aca1bba6SLiu Bo 45185d4f98a2SYan Zheng btrfs_free_path(path); 45195d4f98a2SYan Zheng 45205d4f98a2SYan Zheng if (err == 0) { 45215d4f98a2SYan Zheng /* cleanup orphan inode in data relocation tree */ 45220b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 4523932fd26dSJosef Bacik if (IS_ERR(fs_root)) { 45245d4f98a2SYan Zheng err = PTR_ERR(fs_root); 4525932fd26dSJosef Bacik } else { 452666b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_root); 452700246528SJosef Bacik btrfs_put_root(fs_root); 4528932fd26dSJosef Bacik } 4529932fd26dSJosef Bacik } 45305d4f98a2SYan Zheng return err; 45315d4f98a2SYan Zheng } 45325d4f98a2SYan Zheng 45335d4f98a2SYan Zheng /* 45345d4f98a2SYan Zheng * helper to add ordered checksum for data relocation. 45355d4f98a2SYan Zheng * 45365d4f98a2SYan Zheng * cloning checksum properly handles the nodatasum extents. 45375d4f98a2SYan Zheng * it also saves CPU time to re-calculate the checksum. 45385d4f98a2SYan Zheng */ 45395d4f98a2SYan Zheng int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len) 45405d4f98a2SYan Zheng { 45410b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 45425d4f98a2SYan Zheng struct btrfs_ordered_sum *sums; 45435d4f98a2SYan Zheng struct btrfs_ordered_extent *ordered; 45445d4f98a2SYan Zheng int ret; 45455d4f98a2SYan Zheng u64 disk_bytenr; 45464577b014SJosef Bacik u64 new_bytenr; 45475d4f98a2SYan Zheng LIST_HEAD(list); 45485d4f98a2SYan Zheng 45495d4f98a2SYan Zheng ordered = btrfs_lookup_ordered_extent(inode, file_pos); 4550bffe633eSOmar Sandoval BUG_ON(ordered->file_offset != file_pos || ordered->num_bytes != len); 45515d4f98a2SYan Zheng 45525d4f98a2SYan Zheng disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt; 45530b246afaSJeff Mahoney ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr, 4554a2de733cSArne Jansen disk_bytenr + len - 1, &list, 0); 455579787eaaSJeff Mahoney if (ret) 455679787eaaSJeff Mahoney goto out; 45575d4f98a2SYan Zheng 45585d4f98a2SYan Zheng while (!list_empty(&list)) { 45595d4f98a2SYan Zheng sums = list_entry(list.next, struct btrfs_ordered_sum, list); 45605d4f98a2SYan Zheng list_del_init(&sums->list); 45615d4f98a2SYan Zheng 45624577b014SJosef Bacik /* 45634577b014SJosef Bacik * We need to offset the new_bytenr based on where the csum is. 45644577b014SJosef Bacik * We need to do this because we will read in entire prealloc 45654577b014SJosef Bacik * extents but we may have written to say the middle of the 45664577b014SJosef Bacik * prealloc extent, so we need to make sure the csum goes with 45674577b014SJosef Bacik * the right disk offset. 45684577b014SJosef Bacik * 45694577b014SJosef Bacik * We can do this because the data reloc inode refers strictly 45704577b014SJosef Bacik * to the on disk bytes, so we don't have to worry about 45714577b014SJosef Bacik * disk_len vs real len like with real inodes since it's all 45724577b014SJosef Bacik * disk length. 45734577b014SJosef Bacik */ 4574bffe633eSOmar Sandoval new_bytenr = ordered->disk_bytenr + sums->bytenr - disk_bytenr; 45754577b014SJosef Bacik sums->bytenr = new_bytenr; 45765d4f98a2SYan Zheng 4577f9756261SNikolay Borisov btrfs_add_ordered_sum(ordered, sums); 45785d4f98a2SYan Zheng } 457979787eaaSJeff Mahoney out: 45805d4f98a2SYan Zheng btrfs_put_ordered_extent(ordered); 4581411fc6bcSAndi Kleen return ret; 45825d4f98a2SYan Zheng } 45833fd0a558SYan, Zheng 458483d4cfd4SJosef Bacik int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans, 45853fd0a558SYan, Zheng struct btrfs_root *root, struct extent_buffer *buf, 45863fd0a558SYan, Zheng struct extent_buffer *cow) 45873fd0a558SYan, Zheng { 45880b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 45893fd0a558SYan, Zheng struct reloc_control *rc; 45903fd0a558SYan, Zheng struct backref_node *node; 45913fd0a558SYan, Zheng int first_cow = 0; 45923fd0a558SYan, Zheng int level; 459383d4cfd4SJosef Bacik int ret = 0; 45943fd0a558SYan, Zheng 45950b246afaSJeff Mahoney rc = fs_info->reloc_ctl; 45963fd0a558SYan, Zheng if (!rc) 459783d4cfd4SJosef Bacik return 0; 45983fd0a558SYan, Zheng 45993fd0a558SYan, Zheng BUG_ON(rc->stage == UPDATE_DATA_PTRS && 46003fd0a558SYan, Zheng root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID); 46013fd0a558SYan, Zheng 4602c974c464SWang Shilong if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 4603c974c464SWang Shilong if (buf == root->node) 4604c974c464SWang Shilong __update_reloc_root(root, cow->start); 4605c974c464SWang Shilong } 4606c974c464SWang Shilong 46073fd0a558SYan, Zheng level = btrfs_header_level(buf); 46083fd0a558SYan, Zheng if (btrfs_header_generation(buf) <= 46093fd0a558SYan, Zheng btrfs_root_last_snapshot(&root->root_item)) 46103fd0a558SYan, Zheng first_cow = 1; 46113fd0a558SYan, Zheng 46123fd0a558SYan, Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID && 46133fd0a558SYan, Zheng rc->create_reloc_tree) { 46143fd0a558SYan, Zheng WARN_ON(!first_cow && level == 0); 46153fd0a558SYan, Zheng 46163fd0a558SYan, Zheng node = rc->backref_cache.path[level]; 46173fd0a558SYan, Zheng BUG_ON(node->bytenr != buf->start && 46183fd0a558SYan, Zheng node->new_bytenr != buf->start); 46193fd0a558SYan, Zheng 46203fd0a558SYan, Zheng drop_node_buffer(node); 462167439dadSDavid Sterba atomic_inc(&cow->refs); 46223fd0a558SYan, Zheng node->eb = cow; 46233fd0a558SYan, Zheng node->new_bytenr = cow->start; 46243fd0a558SYan, Zheng 46253fd0a558SYan, Zheng if (!node->pending) { 46263fd0a558SYan, Zheng list_move_tail(&node->list, 46273fd0a558SYan, Zheng &rc->backref_cache.pending[level]); 46283fd0a558SYan, Zheng node->pending = 1; 46293fd0a558SYan, Zheng } 46303fd0a558SYan, Zheng 46313fd0a558SYan, Zheng if (first_cow) 46323fd0a558SYan, Zheng __mark_block_processed(rc, node); 46333fd0a558SYan, Zheng 46343fd0a558SYan, Zheng if (first_cow && level > 0) 46353fd0a558SYan, Zheng rc->nodes_relocated += buf->len; 46363fd0a558SYan, Zheng } 46373fd0a558SYan, Zheng 463883d4cfd4SJosef Bacik if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) 46393fd0a558SYan, Zheng ret = replace_file_extents(trans, rc, root, cow); 464083d4cfd4SJosef Bacik return ret; 46413fd0a558SYan, Zheng } 46423fd0a558SYan, Zheng 46433fd0a558SYan, Zheng /* 46443fd0a558SYan, Zheng * called before creating snapshot. it calculates metadata reservation 464501327610SNicholas D Steeves * required for relocating tree blocks in the snapshot 46463fd0a558SYan, Zheng */ 4647147d256eSZhaolei void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending, 46483fd0a558SYan, Zheng u64 *bytes_to_reserve) 46493fd0a558SYan, Zheng { 465010995c04SQu Wenruo struct btrfs_root *root = pending->root; 465110995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 46523fd0a558SYan, Zheng 46536282675eSQu Wenruo if (!rc || !have_reloc_root(root)) 46543fd0a558SYan, Zheng return; 46553fd0a558SYan, Zheng 46563fd0a558SYan, Zheng if (!rc->merge_reloc_tree) 46573fd0a558SYan, Zheng return; 46583fd0a558SYan, Zheng 46593fd0a558SYan, Zheng root = root->reloc_root; 46603fd0a558SYan, Zheng BUG_ON(btrfs_root_refs(&root->root_item) == 0); 46613fd0a558SYan, Zheng /* 46623fd0a558SYan, Zheng * relocation is in the stage of merging trees. the space 46633fd0a558SYan, Zheng * used by merging a reloc tree is twice the size of 46643fd0a558SYan, Zheng * relocated tree nodes in the worst case. half for cowing 46653fd0a558SYan, Zheng * the reloc tree, half for cowing the fs tree. the space 46663fd0a558SYan, Zheng * used by cowing the reloc tree will be freed after the 46673fd0a558SYan, Zheng * tree is dropped. if we create snapshot, cowing the fs 46683fd0a558SYan, Zheng * tree may use more space than it frees. so we need 46693fd0a558SYan, Zheng * reserve extra space. 46703fd0a558SYan, Zheng */ 46713fd0a558SYan, Zheng *bytes_to_reserve += rc->nodes_relocated; 46723fd0a558SYan, Zheng } 46733fd0a558SYan, Zheng 46743fd0a558SYan, Zheng /* 46753fd0a558SYan, Zheng * called after snapshot is created. migrate block reservation 46763fd0a558SYan, Zheng * and create reloc root for the newly created snapshot 46773fd0a558SYan, Zheng */ 467849b25e05SJeff Mahoney int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans, 46793fd0a558SYan, Zheng struct btrfs_pending_snapshot *pending) 46803fd0a558SYan, Zheng { 46813fd0a558SYan, Zheng struct btrfs_root *root = pending->root; 46823fd0a558SYan, Zheng struct btrfs_root *reloc_root; 46833fd0a558SYan, Zheng struct btrfs_root *new_root; 468410995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 46853fd0a558SYan, Zheng int ret; 46863fd0a558SYan, Zheng 46876282675eSQu Wenruo if (!rc || !have_reloc_root(root)) 468849b25e05SJeff Mahoney return 0; 46893fd0a558SYan, Zheng 46903fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 46913fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated; 46923fd0a558SYan, Zheng 46933fd0a558SYan, Zheng if (rc->merge_reloc_tree) { 46943fd0a558SYan, Zheng ret = btrfs_block_rsv_migrate(&pending->block_rsv, 46953fd0a558SYan, Zheng rc->block_rsv, 46963a584174SLu Fengqi rc->nodes_relocated, true); 469749b25e05SJeff Mahoney if (ret) 469849b25e05SJeff Mahoney return ret; 46993fd0a558SYan, Zheng } 47003fd0a558SYan, Zheng 47013fd0a558SYan, Zheng new_root = pending->snap; 47023fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root->reloc_root, 47033fd0a558SYan, Zheng new_root->root_key.objectid); 470449b25e05SJeff Mahoney if (IS_ERR(reloc_root)) 470549b25e05SJeff Mahoney return PTR_ERR(reloc_root); 47063fd0a558SYan, Zheng 4707ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 4708ffd7b339SJeff Mahoney BUG_ON(ret < 0); 47093fd0a558SYan, Zheng new_root->reloc_root = reloc_root; 47103fd0a558SYan, Zheng 471149b25e05SJeff Mahoney if (rc->create_reloc_tree) 47123fd0a558SYan, Zheng ret = clone_backref_node(trans, rc, root, reloc_root); 471349b25e05SJeff Mahoney return ret; 47143fd0a558SYan, Zheng } 4715