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> 125d4f98a2SYan Zheng #include "ctree.h" 135d4f98a2SYan Zheng #include "disk-io.h" 145d4f98a2SYan Zheng #include "transaction.h" 155d4f98a2SYan Zheng #include "volumes.h" 165d4f98a2SYan Zheng #include "locking.h" 175d4f98a2SYan Zheng #include "btrfs_inode.h" 185d4f98a2SYan Zheng #include "async-thread.h" 190af3d00bSJosef Bacik #include "free-space-cache.h" 20581bb050SLi Zefan #include "inode-map.h" 2162b99540SQu Wenruo #include "qgroup.h" 22cdccee99SLiu Bo #include "print-tree.h" 235d4f98a2SYan Zheng 245d4f98a2SYan Zheng /* 255d4f98a2SYan Zheng * backref_node, mapping_node and tree_block start with this 265d4f98a2SYan Zheng */ 275d4f98a2SYan Zheng struct tree_entry { 285d4f98a2SYan Zheng struct rb_node rb_node; 295d4f98a2SYan Zheng u64 bytenr; 305d4f98a2SYan Zheng }; 315d4f98a2SYan Zheng 325d4f98a2SYan Zheng /* 335d4f98a2SYan Zheng * present a tree block in the backref cache 345d4f98a2SYan Zheng */ 355d4f98a2SYan Zheng struct backref_node { 365d4f98a2SYan Zheng struct rb_node rb_node; 375d4f98a2SYan Zheng u64 bytenr; 383fd0a558SYan, Zheng 393fd0a558SYan, Zheng u64 new_bytenr; 403fd0a558SYan, Zheng /* objectid of tree block owner, can be not uptodate */ 415d4f98a2SYan Zheng u64 owner; 423fd0a558SYan, Zheng /* link to pending, changed or detached list */ 433fd0a558SYan, Zheng struct list_head list; 445d4f98a2SYan Zheng /* list of upper level blocks reference this block */ 455d4f98a2SYan Zheng struct list_head upper; 465d4f98a2SYan Zheng /* list of child blocks in the cache */ 475d4f98a2SYan Zheng struct list_head lower; 485d4f98a2SYan Zheng /* NULL if this node is not tree root */ 495d4f98a2SYan Zheng struct btrfs_root *root; 505d4f98a2SYan Zheng /* extent buffer got by COW the block */ 515d4f98a2SYan Zheng struct extent_buffer *eb; 525d4f98a2SYan Zheng /* level of tree block */ 535d4f98a2SYan Zheng unsigned int level:8; 543fd0a558SYan, Zheng /* is the block in non-reference counted tree */ 553fd0a558SYan, Zheng unsigned int cowonly:1; 563fd0a558SYan, Zheng /* 1 if no child node in the cache */ 575d4f98a2SYan Zheng unsigned int lowest:1; 585d4f98a2SYan Zheng /* is the extent buffer locked */ 595d4f98a2SYan Zheng unsigned int locked:1; 605d4f98a2SYan Zheng /* has the block been processed */ 615d4f98a2SYan Zheng unsigned int processed:1; 625d4f98a2SYan Zheng /* have backrefs of this block been checked */ 635d4f98a2SYan Zheng unsigned int checked:1; 643fd0a558SYan, Zheng /* 653fd0a558SYan, Zheng * 1 if corresponding block has been cowed but some upper 663fd0a558SYan, Zheng * level block pointers may not point to the new location 673fd0a558SYan, Zheng */ 683fd0a558SYan, Zheng unsigned int pending:1; 693fd0a558SYan, Zheng /* 703fd0a558SYan, Zheng * 1 if the backref node isn't connected to any other 713fd0a558SYan, Zheng * backref node. 723fd0a558SYan, Zheng */ 733fd0a558SYan, Zheng unsigned int detached:1; 745d4f98a2SYan Zheng }; 755d4f98a2SYan Zheng 765d4f98a2SYan Zheng /* 775d4f98a2SYan Zheng * present a block pointer in the backref cache 785d4f98a2SYan Zheng */ 795d4f98a2SYan Zheng struct backref_edge { 805d4f98a2SYan Zheng struct list_head list[2]; 815d4f98a2SYan Zheng struct backref_node *node[2]; 825d4f98a2SYan Zheng }; 835d4f98a2SYan Zheng 845d4f98a2SYan Zheng #define LOWER 0 855d4f98a2SYan Zheng #define UPPER 1 860647bf56SWang Shilong #define RELOCATION_RESERVED_NODES 256 875d4f98a2SYan Zheng 885d4f98a2SYan Zheng struct backref_cache { 895d4f98a2SYan Zheng /* red black tree of all backref nodes in the cache */ 905d4f98a2SYan Zheng struct rb_root rb_root; 913fd0a558SYan, Zheng /* for passing backref nodes to btrfs_reloc_cow_block */ 923fd0a558SYan, Zheng struct backref_node *path[BTRFS_MAX_LEVEL]; 933fd0a558SYan, Zheng /* 943fd0a558SYan, Zheng * list of blocks that have been cowed but some block 953fd0a558SYan, Zheng * pointers in upper level blocks may not reflect the 963fd0a558SYan, Zheng * new location 973fd0a558SYan, Zheng */ 985d4f98a2SYan Zheng struct list_head pending[BTRFS_MAX_LEVEL]; 993fd0a558SYan, Zheng /* list of backref nodes with no child node */ 1003fd0a558SYan, Zheng struct list_head leaves; 1013fd0a558SYan, Zheng /* list of blocks that have been cowed in current transaction */ 1023fd0a558SYan, Zheng struct list_head changed; 1033fd0a558SYan, Zheng /* list of detached backref node. */ 1043fd0a558SYan, Zheng struct list_head detached; 1053fd0a558SYan, Zheng 1063fd0a558SYan, Zheng u64 last_trans; 1073fd0a558SYan, Zheng 1083fd0a558SYan, Zheng int nr_nodes; 1093fd0a558SYan, Zheng int nr_edges; 1105d4f98a2SYan Zheng }; 1115d4f98a2SYan Zheng 1125d4f98a2SYan Zheng /* 1135d4f98a2SYan Zheng * map address of tree root to tree 1145d4f98a2SYan Zheng */ 1155d4f98a2SYan Zheng struct mapping_node { 1165d4f98a2SYan Zheng struct rb_node rb_node; 1175d4f98a2SYan Zheng u64 bytenr; 1185d4f98a2SYan Zheng void *data; 1195d4f98a2SYan Zheng }; 1205d4f98a2SYan Zheng 1215d4f98a2SYan Zheng struct mapping_tree { 1225d4f98a2SYan Zheng struct rb_root rb_root; 1235d4f98a2SYan Zheng spinlock_t lock; 1245d4f98a2SYan Zheng }; 1255d4f98a2SYan Zheng 1265d4f98a2SYan Zheng /* 1275d4f98a2SYan Zheng * present a tree block to process 1285d4f98a2SYan Zheng */ 1295d4f98a2SYan Zheng struct tree_block { 1305d4f98a2SYan Zheng struct rb_node rb_node; 1315d4f98a2SYan Zheng u64 bytenr; 1325d4f98a2SYan Zheng struct btrfs_key key; 1335d4f98a2SYan Zheng unsigned int level:8; 1345d4f98a2SYan Zheng unsigned int key_ready:1; 1355d4f98a2SYan Zheng }; 1365d4f98a2SYan Zheng 1370257bb82SYan, Zheng #define MAX_EXTENTS 128 1380257bb82SYan, Zheng 1390257bb82SYan, Zheng struct file_extent_cluster { 1400257bb82SYan, Zheng u64 start; 1410257bb82SYan, Zheng u64 end; 1420257bb82SYan, Zheng u64 boundary[MAX_EXTENTS]; 1430257bb82SYan, Zheng unsigned int nr; 1440257bb82SYan, Zheng }; 1450257bb82SYan, Zheng 1465d4f98a2SYan Zheng struct reloc_control { 1475d4f98a2SYan Zheng /* block group to relocate */ 1485d4f98a2SYan Zheng struct btrfs_block_group_cache *block_group; 1495d4f98a2SYan Zheng /* extent tree */ 1505d4f98a2SYan Zheng struct btrfs_root *extent_root; 1515d4f98a2SYan Zheng /* inode for moving data */ 1525d4f98a2SYan Zheng struct inode *data_inode; 1533fd0a558SYan, Zheng 1543fd0a558SYan, Zheng struct btrfs_block_rsv *block_rsv; 1553fd0a558SYan, Zheng 1563fd0a558SYan, Zheng struct backref_cache backref_cache; 1573fd0a558SYan, Zheng 1583fd0a558SYan, Zheng struct file_extent_cluster cluster; 1595d4f98a2SYan Zheng /* tree blocks have been processed */ 1605d4f98a2SYan Zheng struct extent_io_tree processed_blocks; 1615d4f98a2SYan Zheng /* map start of tree root to corresponding reloc tree */ 1625d4f98a2SYan Zheng struct mapping_tree reloc_root_tree; 1635d4f98a2SYan Zheng /* list of reloc trees */ 1645d4f98a2SYan Zheng struct list_head reloc_roots; 165d2311e69SQu Wenruo /* list of subvolume trees that get relocated */ 166d2311e69SQu Wenruo struct list_head dirty_subvol_roots; 1673fd0a558SYan, Zheng /* size of metadata reservation for merging reloc trees */ 1683fd0a558SYan, Zheng u64 merging_rsv_size; 1693fd0a558SYan, Zheng /* size of relocated tree nodes */ 1703fd0a558SYan, Zheng u64 nodes_relocated; 1710647bf56SWang Shilong /* reserved size for block group relocation*/ 1720647bf56SWang Shilong u64 reserved_bytes; 1733fd0a558SYan, Zheng 1745d4f98a2SYan Zheng u64 search_start; 1755d4f98a2SYan Zheng u64 extents_found; 1763fd0a558SYan, Zheng 1773fd0a558SYan, Zheng unsigned int stage:8; 1783fd0a558SYan, Zheng unsigned int create_reloc_tree:1; 1793fd0a558SYan, Zheng unsigned int merge_reloc_tree:1; 1805d4f98a2SYan Zheng unsigned int found_file_extent:1; 1815d4f98a2SYan Zheng }; 1825d4f98a2SYan Zheng 1835d4f98a2SYan Zheng /* stages of data relocation */ 1845d4f98a2SYan Zheng #define MOVE_DATA_EXTENTS 0 1855d4f98a2SYan Zheng #define UPDATE_DATA_PTRS 1 1865d4f98a2SYan Zheng 1873fd0a558SYan, Zheng static void remove_backref_node(struct backref_cache *cache, 1883fd0a558SYan, Zheng struct backref_node *node); 1893fd0a558SYan, Zheng static void __mark_block_processed(struct reloc_control *rc, 1903fd0a558SYan, Zheng struct backref_node *node); 1915d4f98a2SYan Zheng 1925d4f98a2SYan Zheng static void mapping_tree_init(struct mapping_tree *tree) 1935d4f98a2SYan Zheng { 1946bef4d31SEric Paris tree->rb_root = RB_ROOT; 1955d4f98a2SYan Zheng spin_lock_init(&tree->lock); 1965d4f98a2SYan Zheng } 1975d4f98a2SYan Zheng 1985d4f98a2SYan Zheng static void backref_cache_init(struct backref_cache *cache) 1995d4f98a2SYan Zheng { 2005d4f98a2SYan Zheng int i; 2016bef4d31SEric Paris cache->rb_root = RB_ROOT; 2025d4f98a2SYan Zheng for (i = 0; i < BTRFS_MAX_LEVEL; i++) 2035d4f98a2SYan Zheng INIT_LIST_HEAD(&cache->pending[i]); 2043fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->changed); 2053fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->detached); 2063fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->leaves); 2075d4f98a2SYan Zheng } 2085d4f98a2SYan Zheng 2093fd0a558SYan, Zheng static void backref_cache_cleanup(struct backref_cache *cache) 2105d4f98a2SYan Zheng { 2113fd0a558SYan, Zheng struct backref_node *node; 2123fd0a558SYan, Zheng int i; 2133fd0a558SYan, Zheng 2143fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 2153fd0a558SYan, Zheng node = list_entry(cache->detached.next, 2163fd0a558SYan, Zheng struct backref_node, list); 2173fd0a558SYan, Zheng remove_backref_node(cache, node); 2183fd0a558SYan, Zheng } 2193fd0a558SYan, Zheng 2203fd0a558SYan, Zheng while (!list_empty(&cache->leaves)) { 2213fd0a558SYan, Zheng node = list_entry(cache->leaves.next, 2223fd0a558SYan, Zheng struct backref_node, lower); 2233fd0a558SYan, Zheng remove_backref_node(cache, node); 2243fd0a558SYan, Zheng } 2253fd0a558SYan, Zheng 2263fd0a558SYan, Zheng cache->last_trans = 0; 2273fd0a558SYan, Zheng 2283fd0a558SYan, Zheng for (i = 0; i < BTRFS_MAX_LEVEL; i++) 229f4907095SLiu Bo ASSERT(list_empty(&cache->pending[i])); 230f4907095SLiu Bo ASSERT(list_empty(&cache->changed)); 231f4907095SLiu Bo ASSERT(list_empty(&cache->detached)); 232f4907095SLiu Bo ASSERT(RB_EMPTY_ROOT(&cache->rb_root)); 233f4907095SLiu Bo ASSERT(!cache->nr_nodes); 234f4907095SLiu Bo ASSERT(!cache->nr_edges); 2353fd0a558SYan, Zheng } 2363fd0a558SYan, Zheng 2373fd0a558SYan, Zheng static struct backref_node *alloc_backref_node(struct backref_cache *cache) 2383fd0a558SYan, Zheng { 2393fd0a558SYan, Zheng struct backref_node *node; 2403fd0a558SYan, Zheng 2413fd0a558SYan, Zheng node = kzalloc(sizeof(*node), GFP_NOFS); 2423fd0a558SYan, Zheng if (node) { 2433fd0a558SYan, Zheng INIT_LIST_HEAD(&node->list); 2445d4f98a2SYan Zheng INIT_LIST_HEAD(&node->upper); 2455d4f98a2SYan Zheng INIT_LIST_HEAD(&node->lower); 2465d4f98a2SYan Zheng RB_CLEAR_NODE(&node->rb_node); 2473fd0a558SYan, Zheng cache->nr_nodes++; 2483fd0a558SYan, Zheng } 2493fd0a558SYan, Zheng return node; 2503fd0a558SYan, Zheng } 2513fd0a558SYan, Zheng 2523fd0a558SYan, Zheng static void free_backref_node(struct backref_cache *cache, 2533fd0a558SYan, Zheng struct backref_node *node) 2543fd0a558SYan, Zheng { 2553fd0a558SYan, Zheng if (node) { 2563fd0a558SYan, Zheng cache->nr_nodes--; 2573fd0a558SYan, Zheng kfree(node); 2583fd0a558SYan, Zheng } 2593fd0a558SYan, Zheng } 2603fd0a558SYan, Zheng 2613fd0a558SYan, Zheng static struct backref_edge *alloc_backref_edge(struct backref_cache *cache) 2623fd0a558SYan, Zheng { 2633fd0a558SYan, Zheng struct backref_edge *edge; 2643fd0a558SYan, Zheng 2653fd0a558SYan, Zheng edge = kzalloc(sizeof(*edge), GFP_NOFS); 2663fd0a558SYan, Zheng if (edge) 2673fd0a558SYan, Zheng cache->nr_edges++; 2683fd0a558SYan, Zheng return edge; 2693fd0a558SYan, Zheng } 2703fd0a558SYan, Zheng 2713fd0a558SYan, Zheng static void free_backref_edge(struct backref_cache *cache, 2723fd0a558SYan, Zheng struct backref_edge *edge) 2733fd0a558SYan, Zheng { 2743fd0a558SYan, Zheng if (edge) { 2753fd0a558SYan, Zheng cache->nr_edges--; 2763fd0a558SYan, Zheng kfree(edge); 2773fd0a558SYan, Zheng } 2785d4f98a2SYan Zheng } 2795d4f98a2SYan Zheng 2805d4f98a2SYan Zheng static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr, 2815d4f98a2SYan Zheng struct rb_node *node) 2825d4f98a2SYan Zheng { 2835d4f98a2SYan Zheng struct rb_node **p = &root->rb_node; 2845d4f98a2SYan Zheng struct rb_node *parent = NULL; 2855d4f98a2SYan Zheng struct tree_entry *entry; 2865d4f98a2SYan Zheng 2875d4f98a2SYan Zheng while (*p) { 2885d4f98a2SYan Zheng parent = *p; 2895d4f98a2SYan Zheng entry = rb_entry(parent, struct tree_entry, rb_node); 2905d4f98a2SYan Zheng 2915d4f98a2SYan Zheng if (bytenr < entry->bytenr) 2925d4f98a2SYan Zheng p = &(*p)->rb_left; 2935d4f98a2SYan Zheng else if (bytenr > entry->bytenr) 2945d4f98a2SYan Zheng p = &(*p)->rb_right; 2955d4f98a2SYan Zheng else 2965d4f98a2SYan Zheng return parent; 2975d4f98a2SYan Zheng } 2985d4f98a2SYan Zheng 2995d4f98a2SYan Zheng rb_link_node(node, parent, p); 3005d4f98a2SYan Zheng rb_insert_color(node, root); 3015d4f98a2SYan Zheng return NULL; 3025d4f98a2SYan Zheng } 3035d4f98a2SYan Zheng 3045d4f98a2SYan Zheng static struct rb_node *tree_search(struct rb_root *root, u64 bytenr) 3055d4f98a2SYan Zheng { 3065d4f98a2SYan Zheng struct rb_node *n = root->rb_node; 3075d4f98a2SYan Zheng struct tree_entry *entry; 3085d4f98a2SYan Zheng 3095d4f98a2SYan Zheng while (n) { 3105d4f98a2SYan Zheng entry = rb_entry(n, struct tree_entry, rb_node); 3115d4f98a2SYan Zheng 3125d4f98a2SYan Zheng if (bytenr < entry->bytenr) 3135d4f98a2SYan Zheng n = n->rb_left; 3145d4f98a2SYan Zheng else if (bytenr > entry->bytenr) 3155d4f98a2SYan Zheng n = n->rb_right; 3165d4f98a2SYan Zheng else 3175d4f98a2SYan Zheng return n; 3185d4f98a2SYan Zheng } 3195d4f98a2SYan Zheng return NULL; 3205d4f98a2SYan Zheng } 3215d4f98a2SYan Zheng 32248a3b636SEric Sandeen static void backref_tree_panic(struct rb_node *rb_node, int errno, u64 bytenr) 32343c04fb1SJeff Mahoney { 32443c04fb1SJeff Mahoney 32543c04fb1SJeff Mahoney struct btrfs_fs_info *fs_info = NULL; 32643c04fb1SJeff Mahoney struct backref_node *bnode = rb_entry(rb_node, struct backref_node, 32743c04fb1SJeff Mahoney rb_node); 32843c04fb1SJeff Mahoney if (bnode->root) 32943c04fb1SJeff Mahoney fs_info = bnode->root->fs_info; 3305d163e0eSJeff Mahoney btrfs_panic(fs_info, errno, 3315d163e0eSJeff Mahoney "Inconsistency in backref cache found at offset %llu", 3325d163e0eSJeff Mahoney bytenr); 33343c04fb1SJeff Mahoney } 33443c04fb1SJeff Mahoney 3355d4f98a2SYan Zheng /* 3365d4f98a2SYan Zheng * walk up backref nodes until reach node presents tree root 3375d4f98a2SYan Zheng */ 3385d4f98a2SYan Zheng static struct backref_node *walk_up_backref(struct backref_node *node, 3395d4f98a2SYan Zheng struct backref_edge *edges[], 3405d4f98a2SYan Zheng int *index) 3415d4f98a2SYan Zheng { 3425d4f98a2SYan Zheng struct backref_edge *edge; 3435d4f98a2SYan Zheng int idx = *index; 3445d4f98a2SYan Zheng 3455d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 3465d4f98a2SYan Zheng edge = list_entry(node->upper.next, 3475d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 3485d4f98a2SYan Zheng edges[idx++] = edge; 3495d4f98a2SYan Zheng node = edge->node[UPPER]; 3505d4f98a2SYan Zheng } 3513fd0a558SYan, Zheng BUG_ON(node->detached); 3525d4f98a2SYan Zheng *index = idx; 3535d4f98a2SYan Zheng return node; 3545d4f98a2SYan Zheng } 3555d4f98a2SYan Zheng 3565d4f98a2SYan Zheng /* 3575d4f98a2SYan Zheng * walk down backref nodes to find start of next reference path 3585d4f98a2SYan Zheng */ 3595d4f98a2SYan Zheng static struct backref_node *walk_down_backref(struct backref_edge *edges[], 3605d4f98a2SYan Zheng int *index) 3615d4f98a2SYan Zheng { 3625d4f98a2SYan Zheng struct backref_edge *edge; 3635d4f98a2SYan Zheng struct backref_node *lower; 3645d4f98a2SYan Zheng int idx = *index; 3655d4f98a2SYan Zheng 3665d4f98a2SYan Zheng while (idx > 0) { 3675d4f98a2SYan Zheng edge = edges[idx - 1]; 3685d4f98a2SYan Zheng lower = edge->node[LOWER]; 3695d4f98a2SYan Zheng if (list_is_last(&edge->list[LOWER], &lower->upper)) { 3705d4f98a2SYan Zheng idx--; 3715d4f98a2SYan Zheng continue; 3725d4f98a2SYan Zheng } 3735d4f98a2SYan Zheng edge = list_entry(edge->list[LOWER].next, 3745d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 3755d4f98a2SYan Zheng edges[idx - 1] = edge; 3765d4f98a2SYan Zheng *index = idx; 3775d4f98a2SYan Zheng return edge->node[UPPER]; 3785d4f98a2SYan Zheng } 3795d4f98a2SYan Zheng *index = 0; 3805d4f98a2SYan Zheng return NULL; 3815d4f98a2SYan Zheng } 3825d4f98a2SYan Zheng 3833fd0a558SYan, Zheng static void unlock_node_buffer(struct backref_node *node) 3845d4f98a2SYan Zheng { 3855d4f98a2SYan Zheng if (node->locked) { 3865d4f98a2SYan Zheng btrfs_tree_unlock(node->eb); 3875d4f98a2SYan Zheng node->locked = 0; 3885d4f98a2SYan Zheng } 3893fd0a558SYan, Zheng } 3903fd0a558SYan, Zheng 3913fd0a558SYan, Zheng static void drop_node_buffer(struct backref_node *node) 3923fd0a558SYan, Zheng { 3933fd0a558SYan, Zheng if (node->eb) { 3943fd0a558SYan, Zheng unlock_node_buffer(node); 3955d4f98a2SYan Zheng free_extent_buffer(node->eb); 3965d4f98a2SYan Zheng node->eb = NULL; 3975d4f98a2SYan Zheng } 3985d4f98a2SYan Zheng } 3995d4f98a2SYan Zheng 4005d4f98a2SYan Zheng static void drop_backref_node(struct backref_cache *tree, 4015d4f98a2SYan Zheng struct backref_node *node) 4025d4f98a2SYan Zheng { 4035d4f98a2SYan Zheng BUG_ON(!list_empty(&node->upper)); 4045d4f98a2SYan Zheng 4055d4f98a2SYan Zheng drop_node_buffer(node); 4063fd0a558SYan, Zheng list_del(&node->list); 4075d4f98a2SYan Zheng list_del(&node->lower); 4083fd0a558SYan, Zheng if (!RB_EMPTY_NODE(&node->rb_node)) 4095d4f98a2SYan Zheng rb_erase(&node->rb_node, &tree->rb_root); 4103fd0a558SYan, Zheng free_backref_node(tree, node); 4115d4f98a2SYan Zheng } 4125d4f98a2SYan Zheng 4135d4f98a2SYan Zheng /* 4145d4f98a2SYan Zheng * remove a backref node from the backref cache 4155d4f98a2SYan Zheng */ 4165d4f98a2SYan Zheng static void remove_backref_node(struct backref_cache *cache, 4175d4f98a2SYan Zheng struct backref_node *node) 4185d4f98a2SYan Zheng { 4195d4f98a2SYan Zheng struct backref_node *upper; 4205d4f98a2SYan Zheng struct backref_edge *edge; 4215d4f98a2SYan Zheng 4225d4f98a2SYan Zheng if (!node) 4235d4f98a2SYan Zheng return; 4245d4f98a2SYan Zheng 4253fd0a558SYan, Zheng BUG_ON(!node->lowest && !node->detached); 4265d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 4275d4f98a2SYan Zheng edge = list_entry(node->upper.next, struct backref_edge, 4285d4f98a2SYan Zheng list[LOWER]); 4295d4f98a2SYan Zheng upper = edge->node[UPPER]; 4305d4f98a2SYan Zheng list_del(&edge->list[LOWER]); 4315d4f98a2SYan Zheng list_del(&edge->list[UPPER]); 4323fd0a558SYan, Zheng free_backref_edge(cache, edge); 4333fd0a558SYan, Zheng 4343fd0a558SYan, Zheng if (RB_EMPTY_NODE(&upper->rb_node)) { 4353fd0a558SYan, Zheng BUG_ON(!list_empty(&node->upper)); 4363fd0a558SYan, Zheng drop_backref_node(cache, node); 4373fd0a558SYan, Zheng node = upper; 4383fd0a558SYan, Zheng node->lowest = 1; 4393fd0a558SYan, Zheng continue; 4403fd0a558SYan, Zheng } 4415d4f98a2SYan Zheng /* 4423fd0a558SYan, Zheng * add the node to leaf node list if no other 4435d4f98a2SYan Zheng * child block cached. 4445d4f98a2SYan Zheng */ 4455d4f98a2SYan Zheng if (list_empty(&upper->lower)) { 4463fd0a558SYan, Zheng list_add_tail(&upper->lower, &cache->leaves); 4475d4f98a2SYan Zheng upper->lowest = 1; 4485d4f98a2SYan Zheng } 4495d4f98a2SYan Zheng } 4503fd0a558SYan, Zheng 4515d4f98a2SYan Zheng drop_backref_node(cache, node); 4525d4f98a2SYan Zheng } 4535d4f98a2SYan Zheng 4543fd0a558SYan, Zheng static void update_backref_node(struct backref_cache *cache, 4553fd0a558SYan, Zheng struct backref_node *node, u64 bytenr) 4563fd0a558SYan, Zheng { 4573fd0a558SYan, Zheng struct rb_node *rb_node; 4583fd0a558SYan, Zheng rb_erase(&node->rb_node, &cache->rb_root); 4593fd0a558SYan, Zheng node->bytenr = bytenr; 4603fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node); 46143c04fb1SJeff Mahoney if (rb_node) 46243c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, bytenr); 4633fd0a558SYan, Zheng } 4643fd0a558SYan, Zheng 4653fd0a558SYan, Zheng /* 4663fd0a558SYan, Zheng * update backref cache after a transaction commit 4673fd0a558SYan, Zheng */ 4683fd0a558SYan, Zheng static int update_backref_cache(struct btrfs_trans_handle *trans, 4693fd0a558SYan, Zheng struct backref_cache *cache) 4703fd0a558SYan, Zheng { 4713fd0a558SYan, Zheng struct backref_node *node; 4723fd0a558SYan, Zheng int level = 0; 4733fd0a558SYan, Zheng 4743fd0a558SYan, Zheng if (cache->last_trans == 0) { 4753fd0a558SYan, Zheng cache->last_trans = trans->transid; 4763fd0a558SYan, Zheng return 0; 4773fd0a558SYan, Zheng } 4783fd0a558SYan, Zheng 4793fd0a558SYan, Zheng if (cache->last_trans == trans->transid) 4803fd0a558SYan, Zheng return 0; 4813fd0a558SYan, Zheng 4823fd0a558SYan, Zheng /* 4833fd0a558SYan, Zheng * detached nodes are used to avoid unnecessary backref 4843fd0a558SYan, Zheng * lookup. transaction commit changes the extent tree. 4853fd0a558SYan, Zheng * so the detached nodes are no longer useful. 4863fd0a558SYan, Zheng */ 4873fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 4883fd0a558SYan, Zheng node = list_entry(cache->detached.next, 4893fd0a558SYan, Zheng struct backref_node, list); 4903fd0a558SYan, Zheng remove_backref_node(cache, node); 4913fd0a558SYan, Zheng } 4923fd0a558SYan, Zheng 4933fd0a558SYan, Zheng while (!list_empty(&cache->changed)) { 4943fd0a558SYan, Zheng node = list_entry(cache->changed.next, 4953fd0a558SYan, Zheng struct backref_node, list); 4963fd0a558SYan, Zheng list_del_init(&node->list); 4973fd0a558SYan, Zheng BUG_ON(node->pending); 4983fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 4993fd0a558SYan, Zheng } 5003fd0a558SYan, Zheng 5013fd0a558SYan, Zheng /* 5023fd0a558SYan, Zheng * some nodes can be left in the pending list if there were 5033fd0a558SYan, Zheng * errors during processing the pending nodes. 5043fd0a558SYan, Zheng */ 5053fd0a558SYan, Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 5063fd0a558SYan, Zheng list_for_each_entry(node, &cache->pending[level], list) { 5073fd0a558SYan, Zheng BUG_ON(!node->pending); 5083fd0a558SYan, Zheng if (node->bytenr == node->new_bytenr) 5093fd0a558SYan, Zheng continue; 5103fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 5113fd0a558SYan, Zheng } 5123fd0a558SYan, Zheng } 5133fd0a558SYan, Zheng 5143fd0a558SYan, Zheng cache->last_trans = 0; 5153fd0a558SYan, Zheng return 1; 5163fd0a558SYan, Zheng } 5173fd0a558SYan, Zheng 518f2a97a9dSDavid Sterba 5193fd0a558SYan, Zheng static int should_ignore_root(struct btrfs_root *root) 5203fd0a558SYan, Zheng { 5213fd0a558SYan, Zheng struct btrfs_root *reloc_root; 5223fd0a558SYan, Zheng 52327cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 5243fd0a558SYan, Zheng return 0; 5253fd0a558SYan, Zheng 5263fd0a558SYan, Zheng reloc_root = root->reloc_root; 5273fd0a558SYan, Zheng if (!reloc_root) 5283fd0a558SYan, Zheng return 0; 5293fd0a558SYan, Zheng 5303fd0a558SYan, Zheng if (btrfs_root_last_snapshot(&reloc_root->root_item) == 5313fd0a558SYan, Zheng root->fs_info->running_transaction->transid - 1) 5323fd0a558SYan, Zheng return 0; 5333fd0a558SYan, Zheng /* 5343fd0a558SYan, Zheng * if there is reloc tree and it was created in previous 5353fd0a558SYan, Zheng * transaction backref lookup can find the reloc tree, 5363fd0a558SYan, Zheng * so backref node for the fs tree root is useless for 5373fd0a558SYan, Zheng * relocation. 5383fd0a558SYan, Zheng */ 5393fd0a558SYan, Zheng return 1; 5403fd0a558SYan, Zheng } 5415d4f98a2SYan Zheng /* 5425d4f98a2SYan Zheng * find reloc tree by address of tree root 5435d4f98a2SYan Zheng */ 5445d4f98a2SYan Zheng static struct btrfs_root *find_reloc_root(struct reloc_control *rc, 5455d4f98a2SYan Zheng u64 bytenr) 5465d4f98a2SYan Zheng { 5475d4f98a2SYan Zheng struct rb_node *rb_node; 5485d4f98a2SYan Zheng struct mapping_node *node; 5495d4f98a2SYan Zheng struct btrfs_root *root = NULL; 5505d4f98a2SYan Zheng 5515d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 5525d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr); 5535d4f98a2SYan Zheng if (rb_node) { 5545d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 5555d4f98a2SYan Zheng root = (struct btrfs_root *)node->data; 5565d4f98a2SYan Zheng } 5575d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 5585d4f98a2SYan Zheng return root; 5595d4f98a2SYan Zheng } 5605d4f98a2SYan Zheng 5615d4f98a2SYan Zheng static int is_cowonly_root(u64 root_objectid) 5625d4f98a2SYan Zheng { 5635d4f98a2SYan Zheng if (root_objectid == BTRFS_ROOT_TREE_OBJECTID || 5645d4f98a2SYan Zheng root_objectid == BTRFS_EXTENT_TREE_OBJECTID || 5655d4f98a2SYan Zheng root_objectid == BTRFS_CHUNK_TREE_OBJECTID || 5665d4f98a2SYan Zheng root_objectid == BTRFS_DEV_TREE_OBJECTID || 5675d4f98a2SYan Zheng root_objectid == BTRFS_TREE_LOG_OBJECTID || 56866463748SWang Shilong root_objectid == BTRFS_CSUM_TREE_OBJECTID || 56966463748SWang Shilong root_objectid == BTRFS_UUID_TREE_OBJECTID || 5703e4c5efbSDavid Sterba root_objectid == BTRFS_QUOTA_TREE_OBJECTID || 5713e4c5efbSDavid Sterba root_objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) 5725d4f98a2SYan Zheng return 1; 5735d4f98a2SYan Zheng return 0; 5745d4f98a2SYan Zheng } 5755d4f98a2SYan Zheng 5765d4f98a2SYan Zheng static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info, 5775d4f98a2SYan Zheng u64 root_objectid) 5785d4f98a2SYan Zheng { 5795d4f98a2SYan Zheng struct btrfs_key key; 5805d4f98a2SYan Zheng 5815d4f98a2SYan Zheng key.objectid = root_objectid; 5825d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 5835d4f98a2SYan Zheng if (is_cowonly_root(root_objectid)) 5845d4f98a2SYan Zheng key.offset = 0; 5855d4f98a2SYan Zheng else 5865d4f98a2SYan Zheng key.offset = (u64)-1; 5875d4f98a2SYan Zheng 588c00869f1SMiao Xie return btrfs_get_fs_root(fs_info, &key, false); 5895d4f98a2SYan Zheng } 5905d4f98a2SYan Zheng 5915d4f98a2SYan Zheng static noinline_for_stack 5925d4f98a2SYan Zheng int find_inline_backref(struct extent_buffer *leaf, int slot, 5935d4f98a2SYan Zheng unsigned long *ptr, unsigned long *end) 5945d4f98a2SYan Zheng { 5953173a18fSJosef Bacik struct btrfs_key key; 5965d4f98a2SYan Zheng struct btrfs_extent_item *ei; 5975d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 5985d4f98a2SYan Zheng u32 item_size; 5995d4f98a2SYan Zheng 6003173a18fSJosef Bacik btrfs_item_key_to_cpu(leaf, &key, slot); 6013173a18fSJosef Bacik 6025d4f98a2SYan Zheng item_size = btrfs_item_size_nr(leaf, slot); 603ba3c2b19SNikolay Borisov if (item_size < sizeof(*ei)) { 604ba3c2b19SNikolay Borisov btrfs_print_v0_err(leaf->fs_info); 605ba3c2b19SNikolay Borisov btrfs_handle_fs_error(leaf->fs_info, -EINVAL, NULL); 606ba3c2b19SNikolay Borisov return 1; 607ba3c2b19SNikolay Borisov } 6085d4f98a2SYan Zheng ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item); 6095d4f98a2SYan Zheng WARN_ON(!(btrfs_extent_flags(leaf, ei) & 6105d4f98a2SYan Zheng BTRFS_EXTENT_FLAG_TREE_BLOCK)); 6115d4f98a2SYan Zheng 6123173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 6133173a18fSJosef Bacik item_size <= sizeof(*ei) + sizeof(*bi)) { 6145d4f98a2SYan Zheng WARN_ON(item_size < sizeof(*ei) + sizeof(*bi)); 6155d4f98a2SYan Zheng return 1; 6165d4f98a2SYan Zheng } 617d062d13cSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 618d062d13cSJosef Bacik item_size <= sizeof(*ei)) { 619d062d13cSJosef Bacik WARN_ON(item_size < sizeof(*ei)); 620d062d13cSJosef Bacik return 1; 621d062d13cSJosef Bacik } 6225d4f98a2SYan Zheng 6233173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) { 6245d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 6255d4f98a2SYan Zheng *ptr = (unsigned long)(bi + 1); 6263173a18fSJosef Bacik } else { 6273173a18fSJosef Bacik *ptr = (unsigned long)(ei + 1); 6283173a18fSJosef Bacik } 6295d4f98a2SYan Zheng *end = (unsigned long)ei + item_size; 6305d4f98a2SYan Zheng return 0; 6315d4f98a2SYan Zheng } 6325d4f98a2SYan Zheng 6335d4f98a2SYan Zheng /* 6345d4f98a2SYan Zheng * build backref tree for a given tree block. root of the backref tree 6355d4f98a2SYan Zheng * corresponds the tree block, leaves of the backref tree correspond 6365d4f98a2SYan Zheng * roots of b-trees that reference the tree block. 6375d4f98a2SYan Zheng * 6385d4f98a2SYan Zheng * the basic idea of this function is check backrefs of a given block 63901327610SNicholas D Steeves * to find upper level blocks that reference the block, and then check 64001327610SNicholas D Steeves * backrefs of these upper level blocks recursively. the recursion stop 6415d4f98a2SYan Zheng * when tree root is reached or backrefs for the block is cached. 6425d4f98a2SYan Zheng * 6435d4f98a2SYan Zheng * NOTE: if we find backrefs for a block are cached, we know backrefs 6445d4f98a2SYan Zheng * for all upper level blocks that directly/indirectly reference the 6455d4f98a2SYan Zheng * block are also cached. 6465d4f98a2SYan Zheng */ 6473fd0a558SYan, Zheng static noinline_for_stack 6483fd0a558SYan, Zheng struct backref_node *build_backref_tree(struct reloc_control *rc, 6495d4f98a2SYan Zheng struct btrfs_key *node_key, 6505d4f98a2SYan Zheng int level, u64 bytenr) 6515d4f98a2SYan Zheng { 6523fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 653fa6ac715SQu Wenruo struct btrfs_path *path1; /* For searching extent root */ 654fa6ac715SQu Wenruo struct btrfs_path *path2; /* For searching parent of TREE_BLOCK_REF */ 6555d4f98a2SYan Zheng struct extent_buffer *eb; 6565d4f98a2SYan Zheng struct btrfs_root *root; 6575d4f98a2SYan Zheng struct backref_node *cur; 6585d4f98a2SYan Zheng struct backref_node *upper; 6595d4f98a2SYan Zheng struct backref_node *lower; 6605d4f98a2SYan Zheng struct backref_node *node = NULL; 6615d4f98a2SYan Zheng struct backref_node *exist = NULL; 6625d4f98a2SYan Zheng struct backref_edge *edge; 6635d4f98a2SYan Zheng struct rb_node *rb_node; 6645d4f98a2SYan Zheng struct btrfs_key key; 6655d4f98a2SYan Zheng unsigned long end; 6665d4f98a2SYan Zheng unsigned long ptr; 667fa6ac715SQu Wenruo LIST_HEAD(list); /* Pending edge list, upper node needs to be checked */ 6683fd0a558SYan, Zheng LIST_HEAD(useless); 6693fd0a558SYan, Zheng int cowonly; 6705d4f98a2SYan Zheng int ret; 6715d4f98a2SYan Zheng int err = 0; 672b6c60c80SJosef Bacik bool need_check = true; 6735d4f98a2SYan Zheng 6745d4f98a2SYan Zheng path1 = btrfs_alloc_path(); 6755d4f98a2SYan Zheng path2 = btrfs_alloc_path(); 6765d4f98a2SYan Zheng if (!path1 || !path2) { 6775d4f98a2SYan Zheng err = -ENOMEM; 6785d4f98a2SYan Zheng goto out; 6795d4f98a2SYan Zheng } 680e4058b54SDavid Sterba path1->reada = READA_FORWARD; 681e4058b54SDavid Sterba path2->reada = READA_FORWARD; 6825d4f98a2SYan Zheng 6833fd0a558SYan, Zheng node = alloc_backref_node(cache); 6845d4f98a2SYan Zheng if (!node) { 6855d4f98a2SYan Zheng err = -ENOMEM; 6865d4f98a2SYan Zheng goto out; 6875d4f98a2SYan Zheng } 6885d4f98a2SYan Zheng 6895d4f98a2SYan Zheng node->bytenr = bytenr; 6905d4f98a2SYan Zheng node->level = level; 6915d4f98a2SYan Zheng node->lowest = 1; 6925d4f98a2SYan Zheng cur = node; 6935d4f98a2SYan Zheng again: 6945d4f98a2SYan Zheng end = 0; 6955d4f98a2SYan Zheng ptr = 0; 6965d4f98a2SYan Zheng key.objectid = cur->bytenr; 6973173a18fSJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 6985d4f98a2SYan Zheng key.offset = (u64)-1; 6995d4f98a2SYan Zheng 7005d4f98a2SYan Zheng path1->search_commit_root = 1; 7015d4f98a2SYan Zheng path1->skip_locking = 1; 7025d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1, 7035d4f98a2SYan Zheng 0, 0); 7045d4f98a2SYan Zheng if (ret < 0) { 7055d4f98a2SYan Zheng err = ret; 7065d4f98a2SYan Zheng goto out; 7075d4f98a2SYan Zheng } 70875bfb9afSJosef Bacik ASSERT(ret); 70975bfb9afSJosef Bacik ASSERT(path1->slots[0]); 7105d4f98a2SYan Zheng 7115d4f98a2SYan Zheng path1->slots[0]--; 7125d4f98a2SYan Zheng 7135d4f98a2SYan Zheng WARN_ON(cur->checked); 7145d4f98a2SYan Zheng if (!list_empty(&cur->upper)) { 7155d4f98a2SYan Zheng /* 71670f23fd6SJustin P. Mattock * the backref was added previously when processing 7175d4f98a2SYan Zheng * backref of type BTRFS_TREE_BLOCK_REF_KEY 7185d4f98a2SYan Zheng */ 71975bfb9afSJosef Bacik ASSERT(list_is_singular(&cur->upper)); 7205d4f98a2SYan Zheng edge = list_entry(cur->upper.next, struct backref_edge, 7215d4f98a2SYan Zheng list[LOWER]); 72275bfb9afSJosef Bacik ASSERT(list_empty(&edge->list[UPPER])); 7235d4f98a2SYan Zheng exist = edge->node[UPPER]; 7245d4f98a2SYan Zheng /* 7255d4f98a2SYan Zheng * add the upper level block to pending list if we need 7265d4f98a2SYan Zheng * check its backrefs 7275d4f98a2SYan Zheng */ 7285d4f98a2SYan Zheng if (!exist->checked) 7295d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 7305d4f98a2SYan Zheng } else { 7315d4f98a2SYan Zheng exist = NULL; 7325d4f98a2SYan Zheng } 7335d4f98a2SYan Zheng 7345d4f98a2SYan Zheng while (1) { 7355d4f98a2SYan Zheng cond_resched(); 7365d4f98a2SYan Zheng eb = path1->nodes[0]; 7375d4f98a2SYan Zheng 7385d4f98a2SYan Zheng if (ptr >= end) { 7395d4f98a2SYan Zheng if (path1->slots[0] >= btrfs_header_nritems(eb)) { 7405d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path1); 7415d4f98a2SYan Zheng if (ret < 0) { 7425d4f98a2SYan Zheng err = ret; 7435d4f98a2SYan Zheng goto out; 7445d4f98a2SYan Zheng } 7455d4f98a2SYan Zheng if (ret > 0) 7465d4f98a2SYan Zheng break; 7475d4f98a2SYan Zheng eb = path1->nodes[0]; 7485d4f98a2SYan Zheng } 7495d4f98a2SYan Zheng 7505d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &key, path1->slots[0]); 7515d4f98a2SYan Zheng if (key.objectid != cur->bytenr) { 7525d4f98a2SYan Zheng WARN_ON(exist); 7535d4f98a2SYan Zheng break; 7545d4f98a2SYan Zheng } 7555d4f98a2SYan Zheng 7563173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY || 7573173a18fSJosef Bacik key.type == BTRFS_METADATA_ITEM_KEY) { 7585d4f98a2SYan Zheng ret = find_inline_backref(eb, path1->slots[0], 7595d4f98a2SYan Zheng &ptr, &end); 7605d4f98a2SYan Zheng if (ret) 7615d4f98a2SYan Zheng goto next; 7625d4f98a2SYan Zheng } 7635d4f98a2SYan Zheng } 7645d4f98a2SYan Zheng 7655d4f98a2SYan Zheng if (ptr < end) { 7665d4f98a2SYan Zheng /* update key for inline back ref */ 7675d4f98a2SYan Zheng struct btrfs_extent_inline_ref *iref; 7683de28d57SLiu Bo int type; 7695d4f98a2SYan Zheng iref = (struct btrfs_extent_inline_ref *)ptr; 7703de28d57SLiu Bo type = btrfs_get_extent_inline_ref_type(eb, iref, 7713de28d57SLiu Bo BTRFS_REF_TYPE_BLOCK); 7723de28d57SLiu Bo if (type == BTRFS_REF_TYPE_INVALID) { 773af431dcbSSu Yue err = -EUCLEAN; 7743de28d57SLiu Bo goto out; 7753de28d57SLiu Bo } 7763de28d57SLiu Bo key.type = type; 7775d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 7783de28d57SLiu Bo 7795d4f98a2SYan Zheng WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY && 7805d4f98a2SYan Zheng key.type != BTRFS_SHARED_BLOCK_REF_KEY); 7815d4f98a2SYan Zheng } 7825d4f98a2SYan Zheng 783fa6ac715SQu Wenruo /* 784fa6ac715SQu Wenruo * Parent node found and matches current inline ref, no need to 785fa6ac715SQu Wenruo * rebuild this node for this inline ref. 786fa6ac715SQu Wenruo */ 7875d4f98a2SYan Zheng if (exist && 7885d4f98a2SYan Zheng ((key.type == BTRFS_TREE_BLOCK_REF_KEY && 7895d4f98a2SYan Zheng exist->owner == key.offset) || 7905d4f98a2SYan Zheng (key.type == BTRFS_SHARED_BLOCK_REF_KEY && 7915d4f98a2SYan Zheng exist->bytenr == key.offset))) { 7925d4f98a2SYan Zheng exist = NULL; 7935d4f98a2SYan Zheng goto next; 7945d4f98a2SYan Zheng } 7955d4f98a2SYan Zheng 796fa6ac715SQu Wenruo /* SHARED_BLOCK_REF means key.offset is the parent bytenr */ 7975d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) { 7985d4f98a2SYan Zheng if (key.objectid == key.offset) { 7995d4f98a2SYan Zheng /* 800fa6ac715SQu Wenruo * Only root blocks of reloc trees use backref 801fa6ac715SQu Wenruo * pointing to itself. 8025d4f98a2SYan Zheng */ 8035d4f98a2SYan Zheng root = find_reloc_root(rc, cur->bytenr); 80475bfb9afSJosef Bacik ASSERT(root); 8055d4f98a2SYan Zheng cur->root = root; 8065d4f98a2SYan Zheng break; 8075d4f98a2SYan Zheng } 8085d4f98a2SYan Zheng 8093fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 8105d4f98a2SYan Zheng if (!edge) { 8115d4f98a2SYan Zheng err = -ENOMEM; 8125d4f98a2SYan Zheng goto out; 8135d4f98a2SYan Zheng } 8145d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, key.offset); 8155d4f98a2SYan Zheng if (!rb_node) { 8163fd0a558SYan, Zheng upper = alloc_backref_node(cache); 8175d4f98a2SYan Zheng if (!upper) { 8183fd0a558SYan, Zheng free_backref_edge(cache, edge); 8195d4f98a2SYan Zheng err = -ENOMEM; 8205d4f98a2SYan Zheng goto out; 8215d4f98a2SYan Zheng } 8225d4f98a2SYan Zheng upper->bytenr = key.offset; 8235d4f98a2SYan Zheng upper->level = cur->level + 1; 8245d4f98a2SYan Zheng /* 8255d4f98a2SYan Zheng * backrefs for the upper level block isn't 8265d4f98a2SYan Zheng * cached, add the block to pending list 8275d4f98a2SYan Zheng */ 8285d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 8295d4f98a2SYan Zheng } else { 8305d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 8315d4f98a2SYan Zheng rb_node); 83275bfb9afSJosef Bacik ASSERT(upper->checked); 8335d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 8345d4f98a2SYan Zheng } 8353fd0a558SYan, Zheng list_add_tail(&edge->list[LOWER], &cur->upper); 8365d4f98a2SYan Zheng edge->node[LOWER] = cur; 8373fd0a558SYan, Zheng edge->node[UPPER] = upper; 8385d4f98a2SYan Zheng 8395d4f98a2SYan Zheng goto next; 8406d8ff4e4SDavid Sterba } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { 841ba3c2b19SNikolay Borisov err = -EINVAL; 842ba3c2b19SNikolay Borisov btrfs_print_v0_err(rc->extent_root->fs_info); 843ba3c2b19SNikolay Borisov btrfs_handle_fs_error(rc->extent_root->fs_info, err, 844ba3c2b19SNikolay Borisov NULL); 845ba3c2b19SNikolay Borisov goto out; 8465d4f98a2SYan Zheng } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) { 8475d4f98a2SYan Zheng goto next; 8485d4f98a2SYan Zheng } 8495d4f98a2SYan Zheng 850fa6ac715SQu Wenruo /* 851fa6ac715SQu Wenruo * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref offset 852fa6ac715SQu Wenruo * means the root objectid. We need to search the tree to get 853fa6ac715SQu Wenruo * its parent bytenr. 854fa6ac715SQu Wenruo */ 8555d4f98a2SYan Zheng root = read_fs_root(rc->extent_root->fs_info, key.offset); 8565d4f98a2SYan Zheng if (IS_ERR(root)) { 8575d4f98a2SYan Zheng err = PTR_ERR(root); 8585d4f98a2SYan Zheng goto out; 8595d4f98a2SYan Zheng } 8605d4f98a2SYan Zheng 86127cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 8623fd0a558SYan, Zheng cur->cowonly = 1; 8633fd0a558SYan, Zheng 8645d4f98a2SYan Zheng if (btrfs_root_level(&root->root_item) == cur->level) { 8655d4f98a2SYan Zheng /* tree root */ 86675bfb9afSJosef Bacik ASSERT(btrfs_root_bytenr(&root->root_item) == 8675d4f98a2SYan Zheng cur->bytenr); 8683fd0a558SYan, Zheng if (should_ignore_root(root)) 8693fd0a558SYan, Zheng list_add(&cur->list, &useless); 8703fd0a558SYan, Zheng else 8715d4f98a2SYan Zheng cur->root = root; 8725d4f98a2SYan Zheng break; 8735d4f98a2SYan Zheng } 8745d4f98a2SYan Zheng 8755d4f98a2SYan Zheng level = cur->level + 1; 8765d4f98a2SYan Zheng 877fa6ac715SQu Wenruo /* Search the tree to find parent blocks referring the block. */ 8785d4f98a2SYan Zheng path2->search_commit_root = 1; 8795d4f98a2SYan Zheng path2->skip_locking = 1; 8805d4f98a2SYan Zheng path2->lowest_level = level; 8815d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0); 8825d4f98a2SYan Zheng path2->lowest_level = 0; 8835d4f98a2SYan Zheng if (ret < 0) { 8845d4f98a2SYan Zheng err = ret; 8855d4f98a2SYan Zheng goto out; 8865d4f98a2SYan Zheng } 88733c66f43SYan Zheng if (ret > 0 && path2->slots[level] > 0) 88833c66f43SYan Zheng path2->slots[level]--; 8895d4f98a2SYan Zheng 8905d4f98a2SYan Zheng eb = path2->nodes[level]; 8913561b9dbSLiu Bo if (btrfs_node_blockptr(eb, path2->slots[level]) != 8923561b9dbSLiu Bo cur->bytenr) { 8933561b9dbSLiu Bo btrfs_err(root->fs_info, 8943561b9dbSLiu Bo "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)", 8954fd786e6SMisono Tomohiro cur->bytenr, level - 1, 8964fd786e6SMisono Tomohiro root->root_key.objectid, 8973561b9dbSLiu Bo node_key->objectid, node_key->type, 8983561b9dbSLiu Bo node_key->offset); 8993561b9dbSLiu Bo err = -ENOENT; 9003561b9dbSLiu Bo goto out; 9013561b9dbSLiu Bo } 9025d4f98a2SYan Zheng lower = cur; 903b6c60c80SJosef Bacik need_check = true; 904fa6ac715SQu Wenruo 905fa6ac715SQu Wenruo /* Add all nodes and edges in the path */ 9065d4f98a2SYan Zheng for (; level < BTRFS_MAX_LEVEL; level++) { 9075d4f98a2SYan Zheng if (!path2->nodes[level]) { 90875bfb9afSJosef Bacik ASSERT(btrfs_root_bytenr(&root->root_item) == 9095d4f98a2SYan Zheng lower->bytenr); 9103fd0a558SYan, Zheng if (should_ignore_root(root)) 9113fd0a558SYan, Zheng list_add(&lower->list, &useless); 9123fd0a558SYan, Zheng else 9135d4f98a2SYan Zheng lower->root = root; 9145d4f98a2SYan Zheng break; 9155d4f98a2SYan Zheng } 9165d4f98a2SYan Zheng 9173fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 9185d4f98a2SYan Zheng if (!edge) { 9195d4f98a2SYan Zheng err = -ENOMEM; 9205d4f98a2SYan Zheng goto out; 9215d4f98a2SYan Zheng } 9225d4f98a2SYan Zheng 9235d4f98a2SYan Zheng eb = path2->nodes[level]; 9245d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, eb->start); 9255d4f98a2SYan Zheng if (!rb_node) { 9263fd0a558SYan, Zheng upper = alloc_backref_node(cache); 9275d4f98a2SYan Zheng if (!upper) { 9283fd0a558SYan, Zheng free_backref_edge(cache, edge); 9295d4f98a2SYan Zheng err = -ENOMEM; 9305d4f98a2SYan Zheng goto out; 9315d4f98a2SYan Zheng } 9325d4f98a2SYan Zheng upper->bytenr = eb->start; 9335d4f98a2SYan Zheng upper->owner = btrfs_header_owner(eb); 9345d4f98a2SYan Zheng upper->level = lower->level + 1; 93527cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, 93627cdeb70SMiao Xie &root->state)) 9373fd0a558SYan, Zheng upper->cowonly = 1; 9385d4f98a2SYan Zheng 9395d4f98a2SYan Zheng /* 9405d4f98a2SYan Zheng * if we know the block isn't shared 9415d4f98a2SYan Zheng * we can void checking its backrefs. 9425d4f98a2SYan Zheng */ 9435d4f98a2SYan Zheng if (btrfs_block_can_be_shared(root, eb)) 9445d4f98a2SYan Zheng upper->checked = 0; 9455d4f98a2SYan Zheng else 9465d4f98a2SYan Zheng upper->checked = 1; 9475d4f98a2SYan Zheng 9485d4f98a2SYan Zheng /* 9495d4f98a2SYan Zheng * add the block to pending list if we 950b6c60c80SJosef Bacik * need check its backrefs, we only do this once 951b6c60c80SJosef Bacik * while walking up a tree as we will catch 952b6c60c80SJosef Bacik * anything else later on. 9535d4f98a2SYan Zheng */ 954b6c60c80SJosef Bacik if (!upper->checked && need_check) { 955b6c60c80SJosef Bacik need_check = false; 9565d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], 9575d4f98a2SYan Zheng &list); 958bbe90514SJosef Bacik } else { 959bbe90514SJosef Bacik if (upper->checked) 960bbe90514SJosef Bacik need_check = true; 9615d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 962bbe90514SJosef Bacik } 9635d4f98a2SYan Zheng } else { 9645d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 9655d4f98a2SYan Zheng rb_node); 96675bfb9afSJosef Bacik ASSERT(upper->checked); 9675d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 9683fd0a558SYan, Zheng if (!upper->owner) 9693fd0a558SYan, Zheng upper->owner = btrfs_header_owner(eb); 9705d4f98a2SYan Zheng } 9715d4f98a2SYan Zheng list_add_tail(&edge->list[LOWER], &lower->upper); 9725d4f98a2SYan Zheng edge->node[LOWER] = lower; 9733fd0a558SYan, Zheng edge->node[UPPER] = upper; 9745d4f98a2SYan Zheng 9755d4f98a2SYan Zheng if (rb_node) 9765d4f98a2SYan Zheng break; 9775d4f98a2SYan Zheng lower = upper; 9785d4f98a2SYan Zheng upper = NULL; 9795d4f98a2SYan Zheng } 980b3b4aa74SDavid Sterba btrfs_release_path(path2); 9815d4f98a2SYan Zheng next: 9825d4f98a2SYan Zheng if (ptr < end) { 9835d4f98a2SYan Zheng ptr += btrfs_extent_inline_ref_size(key.type); 9845d4f98a2SYan Zheng if (ptr >= end) { 9855d4f98a2SYan Zheng WARN_ON(ptr > end); 9865d4f98a2SYan Zheng ptr = 0; 9875d4f98a2SYan Zheng end = 0; 9885d4f98a2SYan Zheng } 9895d4f98a2SYan Zheng } 9905d4f98a2SYan Zheng if (ptr >= end) 9915d4f98a2SYan Zheng path1->slots[0]++; 9925d4f98a2SYan Zheng } 993b3b4aa74SDavid Sterba btrfs_release_path(path1); 9945d4f98a2SYan Zheng 9955d4f98a2SYan Zheng cur->checked = 1; 9965d4f98a2SYan Zheng WARN_ON(exist); 9975d4f98a2SYan Zheng 9985d4f98a2SYan Zheng /* the pending list isn't empty, take the first block to process */ 9995d4f98a2SYan Zheng if (!list_empty(&list)) { 10005d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, list[UPPER]); 10015d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 10025d4f98a2SYan Zheng cur = edge->node[UPPER]; 10035d4f98a2SYan Zheng goto again; 10045d4f98a2SYan Zheng } 10055d4f98a2SYan Zheng 10065d4f98a2SYan Zheng /* 10075d4f98a2SYan Zheng * everything goes well, connect backref nodes and insert backref nodes 10085d4f98a2SYan Zheng * into the cache. 10095d4f98a2SYan Zheng */ 101075bfb9afSJosef Bacik ASSERT(node->checked); 10113fd0a558SYan, Zheng cowonly = node->cowonly; 10123fd0a558SYan, Zheng if (!cowonly) { 10133fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, 10143fd0a558SYan, Zheng &node->rb_node); 101543c04fb1SJeff Mahoney if (rb_node) 101643c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 10173fd0a558SYan, Zheng list_add_tail(&node->lower, &cache->leaves); 10183fd0a558SYan, Zheng } 10195d4f98a2SYan Zheng 10205d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) 10215d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 10225d4f98a2SYan Zheng 10235d4f98a2SYan Zheng while (!list_empty(&list)) { 10245d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, list[UPPER]); 10255d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 10265d4f98a2SYan Zheng upper = edge->node[UPPER]; 10273fd0a558SYan, Zheng if (upper->detached) { 10283fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 10293fd0a558SYan, Zheng lower = edge->node[LOWER]; 10303fd0a558SYan, Zheng free_backref_edge(cache, edge); 10313fd0a558SYan, Zheng if (list_empty(&lower->upper)) 10323fd0a558SYan, Zheng list_add(&lower->list, &useless); 10333fd0a558SYan, Zheng continue; 10343fd0a558SYan, Zheng } 10355d4f98a2SYan Zheng 10365d4f98a2SYan Zheng if (!RB_EMPTY_NODE(&upper->rb_node)) { 10375d4f98a2SYan Zheng if (upper->lowest) { 10385d4f98a2SYan Zheng list_del_init(&upper->lower); 10395d4f98a2SYan Zheng upper->lowest = 0; 10405d4f98a2SYan Zheng } 10415d4f98a2SYan Zheng 10425d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 10435d4f98a2SYan Zheng continue; 10445d4f98a2SYan Zheng } 10455d4f98a2SYan Zheng 104675bfb9afSJosef Bacik if (!upper->checked) { 104775bfb9afSJosef Bacik /* 104875bfb9afSJosef Bacik * Still want to blow up for developers since this is a 104975bfb9afSJosef Bacik * logic bug. 105075bfb9afSJosef Bacik */ 105175bfb9afSJosef Bacik ASSERT(0); 105275bfb9afSJosef Bacik err = -EINVAL; 105375bfb9afSJosef Bacik goto out; 105475bfb9afSJosef Bacik } 105575bfb9afSJosef Bacik if (cowonly != upper->cowonly) { 105675bfb9afSJosef Bacik ASSERT(0); 105775bfb9afSJosef Bacik err = -EINVAL; 105875bfb9afSJosef Bacik goto out; 105975bfb9afSJosef Bacik } 106075bfb9afSJosef Bacik 10613fd0a558SYan, Zheng if (!cowonly) { 10625d4f98a2SYan Zheng rb_node = tree_insert(&cache->rb_root, upper->bytenr, 10635d4f98a2SYan Zheng &upper->rb_node); 106443c04fb1SJeff Mahoney if (rb_node) 106543c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 106643c04fb1SJeff Mahoney upper->bytenr); 10673fd0a558SYan, Zheng } 10685d4f98a2SYan Zheng 10695d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 10705d4f98a2SYan Zheng 10715d4f98a2SYan Zheng list_for_each_entry(edge, &upper->upper, list[LOWER]) 10725d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 10735d4f98a2SYan Zheng } 10743fd0a558SYan, Zheng /* 10753fd0a558SYan, Zheng * process useless backref nodes. backref nodes for tree leaves 10763fd0a558SYan, Zheng * are deleted from the cache. backref nodes for upper level 10773fd0a558SYan, Zheng * tree blocks are left in the cache to avoid unnecessary backref 10783fd0a558SYan, Zheng * lookup. 10793fd0a558SYan, Zheng */ 10803fd0a558SYan, Zheng while (!list_empty(&useless)) { 10813fd0a558SYan, Zheng upper = list_entry(useless.next, struct backref_node, list); 10823fd0a558SYan, Zheng list_del_init(&upper->list); 108375bfb9afSJosef Bacik ASSERT(list_empty(&upper->upper)); 10843fd0a558SYan, Zheng if (upper == node) 10853fd0a558SYan, Zheng node = NULL; 10863fd0a558SYan, Zheng if (upper->lowest) { 10873fd0a558SYan, Zheng list_del_init(&upper->lower); 10883fd0a558SYan, Zheng upper->lowest = 0; 10893fd0a558SYan, Zheng } 10903fd0a558SYan, Zheng while (!list_empty(&upper->lower)) { 10913fd0a558SYan, Zheng edge = list_entry(upper->lower.next, 10923fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 10933fd0a558SYan, Zheng list_del(&edge->list[UPPER]); 10943fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 10953fd0a558SYan, Zheng lower = edge->node[LOWER]; 10963fd0a558SYan, Zheng free_backref_edge(cache, edge); 10973fd0a558SYan, Zheng 10983fd0a558SYan, Zheng if (list_empty(&lower->upper)) 10993fd0a558SYan, Zheng list_add(&lower->list, &useless); 11003fd0a558SYan, Zheng } 11013fd0a558SYan, Zheng __mark_block_processed(rc, upper); 11023fd0a558SYan, Zheng if (upper->level > 0) { 11033fd0a558SYan, Zheng list_add(&upper->list, &cache->detached); 11043fd0a558SYan, Zheng upper->detached = 1; 11053fd0a558SYan, Zheng } else { 11063fd0a558SYan, Zheng rb_erase(&upper->rb_node, &cache->rb_root); 11073fd0a558SYan, Zheng free_backref_node(cache, upper); 11083fd0a558SYan, Zheng } 11093fd0a558SYan, Zheng } 11105d4f98a2SYan Zheng out: 11115d4f98a2SYan Zheng btrfs_free_path(path1); 11125d4f98a2SYan Zheng btrfs_free_path(path2); 11135d4f98a2SYan Zheng if (err) { 11143fd0a558SYan, Zheng while (!list_empty(&useless)) { 11153fd0a558SYan, Zheng lower = list_entry(useless.next, 111675bfb9afSJosef Bacik struct backref_node, list); 111775bfb9afSJosef Bacik list_del_init(&lower->list); 11183fd0a558SYan, Zheng } 111975bfb9afSJosef Bacik while (!list_empty(&list)) { 112075bfb9afSJosef Bacik edge = list_first_entry(&list, struct backref_edge, 112175bfb9afSJosef Bacik list[UPPER]); 112275bfb9afSJosef Bacik list_del(&edge->list[UPPER]); 11233fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 112475bfb9afSJosef Bacik lower = edge->node[LOWER]; 11255d4f98a2SYan Zheng upper = edge->node[UPPER]; 11263fd0a558SYan, Zheng free_backref_edge(cache, edge); 112775bfb9afSJosef Bacik 112875bfb9afSJosef Bacik /* 112975bfb9afSJosef Bacik * Lower is no longer linked to any upper backref nodes 113075bfb9afSJosef Bacik * and isn't in the cache, we can free it ourselves. 113175bfb9afSJosef Bacik */ 113275bfb9afSJosef Bacik if (list_empty(&lower->upper) && 113375bfb9afSJosef Bacik RB_EMPTY_NODE(&lower->rb_node)) 113475bfb9afSJosef Bacik list_add(&lower->list, &useless); 113575bfb9afSJosef Bacik 113675bfb9afSJosef Bacik if (!RB_EMPTY_NODE(&upper->rb_node)) 113775bfb9afSJosef Bacik continue; 113875bfb9afSJosef Bacik 113901327610SNicholas D Steeves /* Add this guy's upper edges to the list to process */ 114075bfb9afSJosef Bacik list_for_each_entry(edge, &upper->upper, list[LOWER]) 114175bfb9afSJosef Bacik list_add_tail(&edge->list[UPPER], &list); 114275bfb9afSJosef Bacik if (list_empty(&upper->upper)) 114375bfb9afSJosef Bacik list_add(&upper->list, &useless); 114475bfb9afSJosef Bacik } 114575bfb9afSJosef Bacik 114675bfb9afSJosef Bacik while (!list_empty(&useless)) { 114775bfb9afSJosef Bacik lower = list_entry(useless.next, 114875bfb9afSJosef Bacik struct backref_node, list); 114975bfb9afSJosef Bacik list_del_init(&lower->list); 11500fd8c3daSLiu Bo if (lower == node) 11510fd8c3daSLiu Bo node = NULL; 115275bfb9afSJosef Bacik free_backref_node(cache, lower); 11535d4f98a2SYan Zheng } 11540fd8c3daSLiu Bo 11550fd8c3daSLiu Bo free_backref_node(cache, node); 11565d4f98a2SYan Zheng return ERR_PTR(err); 11575d4f98a2SYan Zheng } 115875bfb9afSJosef Bacik ASSERT(!node || !node->detached); 11595d4f98a2SYan Zheng return node; 11605d4f98a2SYan Zheng } 11615d4f98a2SYan Zheng 11625d4f98a2SYan Zheng /* 11633fd0a558SYan, Zheng * helper to add backref node for the newly created snapshot. 11643fd0a558SYan, Zheng * the backref node is created by cloning backref node that 11653fd0a558SYan, Zheng * corresponds to root of source tree 11663fd0a558SYan, Zheng */ 11673fd0a558SYan, Zheng static int clone_backref_node(struct btrfs_trans_handle *trans, 11683fd0a558SYan, Zheng struct reloc_control *rc, 11693fd0a558SYan, Zheng struct btrfs_root *src, 11703fd0a558SYan, Zheng struct btrfs_root *dest) 11713fd0a558SYan, Zheng { 11723fd0a558SYan, Zheng struct btrfs_root *reloc_root = src->reloc_root; 11733fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 11743fd0a558SYan, Zheng struct backref_node *node = NULL; 11753fd0a558SYan, Zheng struct backref_node *new_node; 11763fd0a558SYan, Zheng struct backref_edge *edge; 11773fd0a558SYan, Zheng struct backref_edge *new_edge; 11783fd0a558SYan, Zheng struct rb_node *rb_node; 11793fd0a558SYan, Zheng 11803fd0a558SYan, Zheng if (cache->last_trans > 0) 11813fd0a558SYan, Zheng update_backref_cache(trans, cache); 11823fd0a558SYan, Zheng 11833fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, src->commit_root->start); 11843fd0a558SYan, Zheng if (rb_node) { 11853fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, rb_node); 11863fd0a558SYan, Zheng if (node->detached) 11873fd0a558SYan, Zheng node = NULL; 11883fd0a558SYan, Zheng else 11893fd0a558SYan, Zheng BUG_ON(node->new_bytenr != reloc_root->node->start); 11903fd0a558SYan, Zheng } 11913fd0a558SYan, Zheng 11923fd0a558SYan, Zheng if (!node) { 11933fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, 11943fd0a558SYan, Zheng reloc_root->commit_root->start); 11953fd0a558SYan, Zheng if (rb_node) { 11963fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, 11973fd0a558SYan, Zheng rb_node); 11983fd0a558SYan, Zheng BUG_ON(node->detached); 11993fd0a558SYan, Zheng } 12003fd0a558SYan, Zheng } 12013fd0a558SYan, Zheng 12023fd0a558SYan, Zheng if (!node) 12033fd0a558SYan, Zheng return 0; 12043fd0a558SYan, Zheng 12053fd0a558SYan, Zheng new_node = alloc_backref_node(cache); 12063fd0a558SYan, Zheng if (!new_node) 12073fd0a558SYan, Zheng return -ENOMEM; 12083fd0a558SYan, Zheng 12093fd0a558SYan, Zheng new_node->bytenr = dest->node->start; 12103fd0a558SYan, Zheng new_node->level = node->level; 12113fd0a558SYan, Zheng new_node->lowest = node->lowest; 12126848ad64SYan, Zheng new_node->checked = 1; 12133fd0a558SYan, Zheng new_node->root = dest; 12143fd0a558SYan, Zheng 12153fd0a558SYan, Zheng if (!node->lowest) { 12163fd0a558SYan, Zheng list_for_each_entry(edge, &node->lower, list[UPPER]) { 12173fd0a558SYan, Zheng new_edge = alloc_backref_edge(cache); 12183fd0a558SYan, Zheng if (!new_edge) 12193fd0a558SYan, Zheng goto fail; 12203fd0a558SYan, Zheng 12213fd0a558SYan, Zheng new_edge->node[UPPER] = new_node; 12223fd0a558SYan, Zheng new_edge->node[LOWER] = edge->node[LOWER]; 12233fd0a558SYan, Zheng list_add_tail(&new_edge->list[UPPER], 12243fd0a558SYan, Zheng &new_node->lower); 12253fd0a558SYan, Zheng } 122676b9e23dSMiao Xie } else { 122776b9e23dSMiao Xie list_add_tail(&new_node->lower, &cache->leaves); 12283fd0a558SYan, Zheng } 12293fd0a558SYan, Zheng 12303fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, new_node->bytenr, 12313fd0a558SYan, Zheng &new_node->rb_node); 123243c04fb1SJeff Mahoney if (rb_node) 123343c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, new_node->bytenr); 12343fd0a558SYan, Zheng 12353fd0a558SYan, Zheng if (!new_node->lowest) { 12363fd0a558SYan, Zheng list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) { 12373fd0a558SYan, Zheng list_add_tail(&new_edge->list[LOWER], 12383fd0a558SYan, Zheng &new_edge->node[LOWER]->upper); 12393fd0a558SYan, Zheng } 12403fd0a558SYan, Zheng } 12413fd0a558SYan, Zheng return 0; 12423fd0a558SYan, Zheng fail: 12433fd0a558SYan, Zheng while (!list_empty(&new_node->lower)) { 12443fd0a558SYan, Zheng new_edge = list_entry(new_node->lower.next, 12453fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 12463fd0a558SYan, Zheng list_del(&new_edge->list[UPPER]); 12473fd0a558SYan, Zheng free_backref_edge(cache, new_edge); 12483fd0a558SYan, Zheng } 12493fd0a558SYan, Zheng free_backref_node(cache, new_node); 12503fd0a558SYan, Zheng return -ENOMEM; 12513fd0a558SYan, Zheng } 12523fd0a558SYan, Zheng 12533fd0a558SYan, Zheng /* 12545d4f98a2SYan Zheng * helper to add 'address of tree root -> reloc tree' mapping 12555d4f98a2SYan Zheng */ 1256ffd7b339SJeff Mahoney static int __must_check __add_reloc_root(struct btrfs_root *root) 12575d4f98a2SYan Zheng { 12580b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 12595d4f98a2SYan Zheng struct rb_node *rb_node; 12605d4f98a2SYan Zheng struct mapping_node *node; 12610b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 12625d4f98a2SYan Zheng 12635d4f98a2SYan Zheng node = kmalloc(sizeof(*node), GFP_NOFS); 1264ffd7b339SJeff Mahoney if (!node) 1265ffd7b339SJeff Mahoney return -ENOMEM; 12665d4f98a2SYan Zheng 12675d4f98a2SYan Zheng node->bytenr = root->node->start; 12685d4f98a2SYan Zheng node->data = root; 12695d4f98a2SYan Zheng 12705d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 12715d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 12725d4f98a2SYan Zheng node->bytenr, &node->rb_node); 12735d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 1274ffd7b339SJeff Mahoney if (rb_node) { 12750b246afaSJeff Mahoney btrfs_panic(fs_info, -EEXIST, 12765d163e0eSJeff Mahoney "Duplicate root found for start=%llu while inserting into relocation tree", 12775d163e0eSJeff Mahoney node->bytenr); 1278ffd7b339SJeff Mahoney } 12795d4f98a2SYan Zheng 12805d4f98a2SYan Zheng list_add_tail(&root->root_list, &rc->reloc_roots); 12815d4f98a2SYan Zheng return 0; 12825d4f98a2SYan Zheng } 12835d4f98a2SYan Zheng 12845d4f98a2SYan Zheng /* 1285c974c464SWang Shilong * helper to delete the 'address of tree root -> reloc tree' 12865d4f98a2SYan Zheng * mapping 12875d4f98a2SYan Zheng */ 1288c974c464SWang Shilong static void __del_reloc_root(struct btrfs_root *root) 12895d4f98a2SYan Zheng { 12900b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 12915d4f98a2SYan Zheng struct rb_node *rb_node; 12925d4f98a2SYan Zheng struct mapping_node *node = NULL; 12930b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 12945d4f98a2SYan Zheng 129565c6e82bSQu Wenruo if (rc && root->node) { 12965d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 12975d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1298c974c464SWang Shilong root->node->start); 1299c974c464SWang Shilong if (rb_node) { 1300c974c464SWang Shilong node = rb_entry(rb_node, struct mapping_node, rb_node); 1301c974c464SWang Shilong rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 1302c974c464SWang Shilong } 1303c974c464SWang Shilong spin_unlock(&rc->reloc_root_tree.lock); 1304c974c464SWang Shilong if (!node) 1305c974c464SWang Shilong return; 1306c974c464SWang Shilong BUG_ON((struct btrfs_root *)node->data != root); 1307389305b2SQu Wenruo } 1308c974c464SWang Shilong 13090b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1310c974c464SWang Shilong list_del_init(&root->root_list); 13110b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1312c974c464SWang Shilong kfree(node); 1313c974c464SWang Shilong } 1314c974c464SWang Shilong 1315c974c464SWang Shilong /* 1316c974c464SWang Shilong * helper to update the 'address of tree root -> reloc tree' 1317c974c464SWang Shilong * mapping 1318c974c464SWang Shilong */ 1319c974c464SWang Shilong static int __update_reloc_root(struct btrfs_root *root, u64 new_bytenr) 1320c974c464SWang Shilong { 13210b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1322c974c464SWang Shilong struct rb_node *rb_node; 1323c974c464SWang Shilong struct mapping_node *node = NULL; 13240b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 1325c974c464SWang Shilong 1326c974c464SWang Shilong spin_lock(&rc->reloc_root_tree.lock); 1327c974c464SWang Shilong rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1328c974c464SWang Shilong root->node->start); 13295d4f98a2SYan Zheng if (rb_node) { 13305d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 13315d4f98a2SYan Zheng rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 13325d4f98a2SYan Zheng } 13335d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 13345d4f98a2SYan Zheng 13358f71f3e0SLiu Bo if (!node) 13368f71f3e0SLiu Bo return 0; 13375d4f98a2SYan Zheng BUG_ON((struct btrfs_root *)node->data != root); 13385d4f98a2SYan Zheng 13395d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 1340c974c464SWang Shilong node->bytenr = new_bytenr; 13415d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 13425d4f98a2SYan Zheng node->bytenr, &node->rb_node); 13435d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 134443c04fb1SJeff Mahoney if (rb_node) 134543c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 13465d4f98a2SYan Zheng return 0; 13475d4f98a2SYan Zheng } 13485d4f98a2SYan Zheng 13493fd0a558SYan, Zheng static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans, 13503fd0a558SYan, Zheng struct btrfs_root *root, u64 objectid) 13515d4f98a2SYan Zheng { 13520b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13535d4f98a2SYan Zheng struct btrfs_root *reloc_root; 13545d4f98a2SYan Zheng struct extent_buffer *eb; 13555d4f98a2SYan Zheng struct btrfs_root_item *root_item; 13565d4f98a2SYan Zheng struct btrfs_key root_key; 13575d4f98a2SYan Zheng int ret; 13585d4f98a2SYan Zheng 13595d4f98a2SYan Zheng root_item = kmalloc(sizeof(*root_item), GFP_NOFS); 13605d4f98a2SYan Zheng BUG_ON(!root_item); 13615d4f98a2SYan Zheng 13625d4f98a2SYan Zheng root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; 13635d4f98a2SYan Zheng root_key.type = BTRFS_ROOT_ITEM_KEY; 13643fd0a558SYan, Zheng root_key.offset = objectid; 13655d4f98a2SYan Zheng 13663fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 1367054570a1SFilipe Manana u64 commit_root_gen; 1368054570a1SFilipe Manana 13693fd0a558SYan, Zheng /* called by btrfs_init_reloc_root */ 13705d4f98a2SYan Zheng ret = btrfs_copy_root(trans, root, root->commit_root, &eb, 13715d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID); 13725d4f98a2SYan Zheng BUG_ON(ret); 1373054570a1SFilipe Manana /* 1374054570a1SFilipe Manana * Set the last_snapshot field to the generation of the commit 1375054570a1SFilipe Manana * root - like this ctree.c:btrfs_block_can_be_shared() behaves 1376054570a1SFilipe Manana * correctly (returns true) when the relocation root is created 1377054570a1SFilipe Manana * either inside the critical section of a transaction commit 1378054570a1SFilipe Manana * (through transaction.c:qgroup_account_snapshot()) and when 1379054570a1SFilipe Manana * it's created before the transaction commit is started. 1380054570a1SFilipe Manana */ 1381054570a1SFilipe Manana commit_root_gen = btrfs_header_generation(root->commit_root); 1382054570a1SFilipe Manana btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen); 13833fd0a558SYan, Zheng } else { 13843fd0a558SYan, Zheng /* 13853fd0a558SYan, Zheng * called by btrfs_reloc_post_snapshot_hook. 13863fd0a558SYan, Zheng * the source tree is a reloc tree, all tree blocks 13873fd0a558SYan, Zheng * modified after it was created have RELOC flag 13883fd0a558SYan, Zheng * set in their headers. so it's OK to not update 13893fd0a558SYan, Zheng * the 'last_snapshot'. 13903fd0a558SYan, Zheng */ 13913fd0a558SYan, Zheng ret = btrfs_copy_root(trans, root, root->node, &eb, 13923fd0a558SYan, Zheng BTRFS_TREE_RELOC_OBJECTID); 13933fd0a558SYan, Zheng BUG_ON(ret); 13943fd0a558SYan, Zheng } 13953fd0a558SYan, Zheng 13965d4f98a2SYan Zheng memcpy(root_item, &root->root_item, sizeof(*root_item)); 13975d4f98a2SYan Zheng btrfs_set_root_bytenr(root_item, eb->start); 13985d4f98a2SYan Zheng btrfs_set_root_level(root_item, btrfs_header_level(eb)); 13995d4f98a2SYan Zheng btrfs_set_root_generation(root_item, trans->transid); 14003fd0a558SYan, Zheng 14013fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 14023fd0a558SYan, Zheng btrfs_set_root_refs(root_item, 0); 14033fd0a558SYan, Zheng memset(&root_item->drop_progress, 0, 14043fd0a558SYan, Zheng sizeof(struct btrfs_disk_key)); 14055d4f98a2SYan Zheng root_item->drop_level = 0; 14063fd0a558SYan, Zheng } 14075d4f98a2SYan Zheng 14085d4f98a2SYan Zheng btrfs_tree_unlock(eb); 14095d4f98a2SYan Zheng free_extent_buffer(eb); 14105d4f98a2SYan Zheng 14110b246afaSJeff Mahoney ret = btrfs_insert_root(trans, fs_info->tree_root, 14125d4f98a2SYan Zheng &root_key, root_item); 14135d4f98a2SYan Zheng BUG_ON(ret); 14145d4f98a2SYan Zheng kfree(root_item); 14155d4f98a2SYan Zheng 14160b246afaSJeff Mahoney reloc_root = btrfs_read_fs_root(fs_info->tree_root, &root_key); 14175d4f98a2SYan Zheng BUG_ON(IS_ERR(reloc_root)); 14185d4f98a2SYan Zheng reloc_root->last_trans = trans->transid; 14193fd0a558SYan, Zheng return reloc_root; 14203fd0a558SYan, Zheng } 14213fd0a558SYan, Zheng 14223fd0a558SYan, Zheng /* 14233fd0a558SYan, Zheng * create reloc tree for a given fs tree. reloc tree is just a 14243fd0a558SYan, Zheng * snapshot of the fs tree with special root objectid. 14253fd0a558SYan, Zheng */ 14263fd0a558SYan, Zheng int btrfs_init_reloc_root(struct btrfs_trans_handle *trans, 14273fd0a558SYan, Zheng struct btrfs_root *root) 14283fd0a558SYan, Zheng { 14290b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 14303fd0a558SYan, Zheng struct btrfs_root *reloc_root; 14310b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 143220dd2cbfSMiao Xie struct btrfs_block_rsv *rsv; 14333fd0a558SYan, Zheng int clear_rsv = 0; 1434ffd7b339SJeff Mahoney int ret; 14353fd0a558SYan, Zheng 14363fd0a558SYan, Zheng if (root->reloc_root) { 14373fd0a558SYan, Zheng reloc_root = root->reloc_root; 14383fd0a558SYan, Zheng reloc_root->last_trans = trans->transid; 14393fd0a558SYan, Zheng return 0; 14403fd0a558SYan, Zheng } 14413fd0a558SYan, Zheng 14423fd0a558SYan, Zheng if (!rc || !rc->create_reloc_tree || 14433fd0a558SYan, Zheng root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 14443fd0a558SYan, Zheng return 0; 14453fd0a558SYan, Zheng 144620dd2cbfSMiao Xie if (!trans->reloc_reserved) { 144720dd2cbfSMiao Xie rsv = trans->block_rsv; 14483fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 14493fd0a558SYan, Zheng clear_rsv = 1; 14503fd0a558SYan, Zheng } 14513fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root, root->root_key.objectid); 14523fd0a558SYan, Zheng if (clear_rsv) 145320dd2cbfSMiao Xie trans->block_rsv = rsv; 14545d4f98a2SYan Zheng 1455ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 1456ffd7b339SJeff Mahoney BUG_ON(ret < 0); 14575d4f98a2SYan Zheng root->reloc_root = reloc_root; 14585d4f98a2SYan Zheng return 0; 14595d4f98a2SYan Zheng } 14605d4f98a2SYan Zheng 14615d4f98a2SYan Zheng /* 14625d4f98a2SYan Zheng * update root item of reloc tree 14635d4f98a2SYan Zheng */ 14645d4f98a2SYan Zheng int btrfs_update_reloc_root(struct btrfs_trans_handle *trans, 14655d4f98a2SYan Zheng struct btrfs_root *root) 14665d4f98a2SYan Zheng { 14670b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 14685d4f98a2SYan Zheng struct btrfs_root *reloc_root; 14695d4f98a2SYan Zheng struct btrfs_root_item *root_item; 14705d4f98a2SYan Zheng int ret; 14715d4f98a2SYan Zheng 1472d2311e69SQu Wenruo if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state) || 1473d2311e69SQu Wenruo !root->reloc_root) 14747585717fSChris Mason goto out; 14755d4f98a2SYan Zheng 14765d4f98a2SYan Zheng reloc_root = root->reloc_root; 14775d4f98a2SYan Zheng root_item = &reloc_root->root_item; 14785d4f98a2SYan Zheng 1479d2311e69SQu Wenruo /* root->reloc_root will stay until current relocation finished */ 14800b246afaSJeff Mahoney if (fs_info->reloc_ctl->merge_reloc_tree && 14813fd0a558SYan, Zheng btrfs_root_refs(root_item) == 0) { 1482d2311e69SQu Wenruo set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 1483c974c464SWang Shilong __del_reloc_root(reloc_root); 14845d4f98a2SYan Zheng } 14855d4f98a2SYan Zheng 14865d4f98a2SYan Zheng if (reloc_root->commit_root != reloc_root->node) { 14875d4f98a2SYan Zheng btrfs_set_root_node(root_item, reloc_root->node); 14885d4f98a2SYan Zheng free_extent_buffer(reloc_root->commit_root); 14895d4f98a2SYan Zheng reloc_root->commit_root = btrfs_root_node(reloc_root); 14905d4f98a2SYan Zheng } 14915d4f98a2SYan Zheng 14920b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 14935d4f98a2SYan Zheng &reloc_root->root_key, root_item); 14945d4f98a2SYan Zheng BUG_ON(ret); 14957585717fSChris Mason 14967585717fSChris Mason out: 14975d4f98a2SYan Zheng return 0; 14985d4f98a2SYan Zheng } 14995d4f98a2SYan Zheng 15005d4f98a2SYan Zheng /* 15015d4f98a2SYan Zheng * helper to find first cached inode with inode number >= objectid 15025d4f98a2SYan Zheng * in a subvolume 15035d4f98a2SYan Zheng */ 15045d4f98a2SYan Zheng static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid) 15055d4f98a2SYan Zheng { 15065d4f98a2SYan Zheng struct rb_node *node; 15075d4f98a2SYan Zheng struct rb_node *prev; 15085d4f98a2SYan Zheng struct btrfs_inode *entry; 15095d4f98a2SYan Zheng struct inode *inode; 15105d4f98a2SYan Zheng 15115d4f98a2SYan Zheng spin_lock(&root->inode_lock); 15125d4f98a2SYan Zheng again: 15135d4f98a2SYan Zheng node = root->inode_tree.rb_node; 15145d4f98a2SYan Zheng prev = NULL; 15155d4f98a2SYan Zheng while (node) { 15165d4f98a2SYan Zheng prev = node; 15175d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 15185d4f98a2SYan Zheng 15194a0cc7caSNikolay Borisov if (objectid < btrfs_ino(entry)) 15205d4f98a2SYan Zheng node = node->rb_left; 15214a0cc7caSNikolay Borisov else if (objectid > btrfs_ino(entry)) 15225d4f98a2SYan Zheng node = node->rb_right; 15235d4f98a2SYan Zheng else 15245d4f98a2SYan Zheng break; 15255d4f98a2SYan Zheng } 15265d4f98a2SYan Zheng if (!node) { 15275d4f98a2SYan Zheng while (prev) { 15285d4f98a2SYan Zheng entry = rb_entry(prev, struct btrfs_inode, rb_node); 15294a0cc7caSNikolay Borisov if (objectid <= btrfs_ino(entry)) { 15305d4f98a2SYan Zheng node = prev; 15315d4f98a2SYan Zheng break; 15325d4f98a2SYan Zheng } 15335d4f98a2SYan Zheng prev = rb_next(prev); 15345d4f98a2SYan Zheng } 15355d4f98a2SYan Zheng } 15365d4f98a2SYan Zheng while (node) { 15375d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 15385d4f98a2SYan Zheng inode = igrab(&entry->vfs_inode); 15395d4f98a2SYan Zheng if (inode) { 15405d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 15415d4f98a2SYan Zheng return inode; 15425d4f98a2SYan Zheng } 15435d4f98a2SYan Zheng 15444a0cc7caSNikolay Borisov objectid = btrfs_ino(entry) + 1; 15455d4f98a2SYan Zheng if (cond_resched_lock(&root->inode_lock)) 15465d4f98a2SYan Zheng goto again; 15475d4f98a2SYan Zheng 15485d4f98a2SYan Zheng node = rb_next(node); 15495d4f98a2SYan Zheng } 15505d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 15515d4f98a2SYan Zheng return NULL; 15525d4f98a2SYan Zheng } 15535d4f98a2SYan Zheng 15545d4f98a2SYan Zheng static int in_block_group(u64 bytenr, 15555d4f98a2SYan Zheng struct btrfs_block_group_cache *block_group) 15565d4f98a2SYan Zheng { 15575d4f98a2SYan Zheng if (bytenr >= block_group->key.objectid && 15585d4f98a2SYan Zheng bytenr < block_group->key.objectid + block_group->key.offset) 15595d4f98a2SYan Zheng return 1; 15605d4f98a2SYan Zheng return 0; 15615d4f98a2SYan Zheng } 15625d4f98a2SYan Zheng 15635d4f98a2SYan Zheng /* 15645d4f98a2SYan Zheng * get new location of data 15655d4f98a2SYan Zheng */ 15665d4f98a2SYan Zheng static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr, 15675d4f98a2SYan Zheng u64 bytenr, u64 num_bytes) 15685d4f98a2SYan Zheng { 15695d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(reloc_inode)->root; 15705d4f98a2SYan Zheng struct btrfs_path *path; 15715d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 15725d4f98a2SYan Zheng struct extent_buffer *leaf; 15735d4f98a2SYan Zheng int ret; 15745d4f98a2SYan Zheng 15755d4f98a2SYan Zheng path = btrfs_alloc_path(); 15765d4f98a2SYan Zheng if (!path) 15775d4f98a2SYan Zheng return -ENOMEM; 15785d4f98a2SYan Zheng 15795d4f98a2SYan Zheng bytenr -= BTRFS_I(reloc_inode)->index_cnt; 1580f85b7379SDavid Sterba ret = btrfs_lookup_file_extent(NULL, root, path, 1581f85b7379SDavid Sterba btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0); 15825d4f98a2SYan Zheng if (ret < 0) 15835d4f98a2SYan Zheng goto out; 15845d4f98a2SYan Zheng if (ret > 0) { 15855d4f98a2SYan Zheng ret = -ENOENT; 15865d4f98a2SYan Zheng goto out; 15875d4f98a2SYan Zheng } 15885d4f98a2SYan Zheng 15895d4f98a2SYan Zheng leaf = path->nodes[0]; 15905d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 15915d4f98a2SYan Zheng struct btrfs_file_extent_item); 15925d4f98a2SYan Zheng 15935d4f98a2SYan Zheng BUG_ON(btrfs_file_extent_offset(leaf, fi) || 15945d4f98a2SYan Zheng btrfs_file_extent_compression(leaf, fi) || 15955d4f98a2SYan Zheng btrfs_file_extent_encryption(leaf, fi) || 15965d4f98a2SYan Zheng btrfs_file_extent_other_encoding(leaf, fi)); 15975d4f98a2SYan Zheng 15985d4f98a2SYan Zheng if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) { 159983d4cfd4SJosef Bacik ret = -EINVAL; 16005d4f98a2SYan Zheng goto out; 16015d4f98a2SYan Zheng } 16025d4f98a2SYan Zheng 16035d4f98a2SYan Zheng *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 16045d4f98a2SYan Zheng ret = 0; 16055d4f98a2SYan Zheng out: 16065d4f98a2SYan Zheng btrfs_free_path(path); 16075d4f98a2SYan Zheng return ret; 16085d4f98a2SYan Zheng } 16095d4f98a2SYan Zheng 16105d4f98a2SYan Zheng /* 16115d4f98a2SYan Zheng * update file extent items in the tree leaf to point to 16125d4f98a2SYan Zheng * the new locations. 16135d4f98a2SYan Zheng */ 16143fd0a558SYan, Zheng static noinline_for_stack 16153fd0a558SYan, Zheng int replace_file_extents(struct btrfs_trans_handle *trans, 16165d4f98a2SYan Zheng struct reloc_control *rc, 16175d4f98a2SYan Zheng struct btrfs_root *root, 16183fd0a558SYan, Zheng struct extent_buffer *leaf) 16195d4f98a2SYan Zheng { 16200b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 16215d4f98a2SYan Zheng struct btrfs_key key; 16225d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 16235d4f98a2SYan Zheng struct inode *inode = NULL; 16245d4f98a2SYan Zheng u64 parent; 16255d4f98a2SYan Zheng u64 bytenr; 16263fd0a558SYan, Zheng u64 new_bytenr = 0; 16275d4f98a2SYan Zheng u64 num_bytes; 16285d4f98a2SYan Zheng u64 end; 16295d4f98a2SYan Zheng u32 nritems; 16305d4f98a2SYan Zheng u32 i; 163183d4cfd4SJosef Bacik int ret = 0; 16325d4f98a2SYan Zheng int first = 1; 16335d4f98a2SYan Zheng int dirty = 0; 16345d4f98a2SYan Zheng 16355d4f98a2SYan Zheng if (rc->stage != UPDATE_DATA_PTRS) 16365d4f98a2SYan Zheng return 0; 16375d4f98a2SYan Zheng 16385d4f98a2SYan Zheng /* reloc trees always use full backref */ 16395d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 16405d4f98a2SYan Zheng parent = leaf->start; 16415d4f98a2SYan Zheng else 16425d4f98a2SYan Zheng parent = 0; 16435d4f98a2SYan Zheng 16445d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 16455d4f98a2SYan Zheng for (i = 0; i < nritems; i++) { 16465d4f98a2SYan Zheng cond_resched(); 16475d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, i); 16485d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_DATA_KEY) 16495d4f98a2SYan Zheng continue; 16505d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 16515d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 16525d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 16535d4f98a2SYan Zheng continue; 16545d4f98a2SYan Zheng bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 16555d4f98a2SYan Zheng num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); 16565d4f98a2SYan Zheng if (bytenr == 0) 16575d4f98a2SYan Zheng continue; 16585d4f98a2SYan Zheng if (!in_block_group(bytenr, rc->block_group)) 16595d4f98a2SYan Zheng continue; 16605d4f98a2SYan Zheng 16615d4f98a2SYan Zheng /* 16625d4f98a2SYan Zheng * if we are modifying block in fs tree, wait for readpage 16635d4f98a2SYan Zheng * to complete and drop the extent cache 16645d4f98a2SYan Zheng */ 16655d4f98a2SYan Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 16665d4f98a2SYan Zheng if (first) { 16675d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 16685d4f98a2SYan Zheng first = 0; 16694a0cc7caSNikolay Borisov } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) { 16703fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 16715d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 16725d4f98a2SYan Zheng } 16734a0cc7caSNikolay Borisov if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) { 16745d4f98a2SYan Zheng end = key.offset + 16755d4f98a2SYan Zheng btrfs_file_extent_num_bytes(leaf, fi); 16765d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(key.offset, 16770b246afaSJeff Mahoney fs_info->sectorsize)); 16780b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 16795d4f98a2SYan Zheng end--; 16805d4f98a2SYan Zheng ret = try_lock_extent(&BTRFS_I(inode)->io_tree, 1681d0082371SJeff Mahoney key.offset, end); 16825d4f98a2SYan Zheng if (!ret) 16835d4f98a2SYan Zheng continue; 16845d4f98a2SYan Zheng 1685dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), 1686dcdbc059SNikolay Borisov key.offset, end, 1); 16875d4f98a2SYan Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 1688d0082371SJeff Mahoney key.offset, end); 16895d4f98a2SYan Zheng } 16905d4f98a2SYan Zheng } 16915d4f98a2SYan Zheng 16925d4f98a2SYan Zheng ret = get_new_location(rc->data_inode, &new_bytenr, 16935d4f98a2SYan Zheng bytenr, num_bytes); 169483d4cfd4SJosef Bacik if (ret) { 169583d4cfd4SJosef Bacik /* 169683d4cfd4SJosef Bacik * Don't have to abort since we've not changed anything 169783d4cfd4SJosef Bacik * in the file extent yet. 169883d4cfd4SJosef Bacik */ 169983d4cfd4SJosef Bacik break; 17003fd0a558SYan, Zheng } 17015d4f98a2SYan Zheng 17025d4f98a2SYan Zheng btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr); 17035d4f98a2SYan Zheng dirty = 1; 17045d4f98a2SYan Zheng 17055d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 170684f7d8e6SJosef Bacik ret = btrfs_inc_extent_ref(trans, root, new_bytenr, 17075d4f98a2SYan Zheng num_bytes, parent, 17085d4f98a2SYan Zheng btrfs_header_owner(leaf), 1709b06c4bf5SFilipe Manana key.objectid, key.offset); 171083d4cfd4SJosef Bacik if (ret) { 171166642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 171283d4cfd4SJosef Bacik break; 171383d4cfd4SJosef Bacik } 17145d4f98a2SYan Zheng 171584f7d8e6SJosef Bacik ret = btrfs_free_extent(trans, root, bytenr, num_bytes, 17165d4f98a2SYan Zheng parent, btrfs_header_owner(leaf), 1717b06c4bf5SFilipe Manana key.objectid, key.offset); 171883d4cfd4SJosef Bacik if (ret) { 171966642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 172083d4cfd4SJosef Bacik break; 172183d4cfd4SJosef Bacik } 17225d4f98a2SYan Zheng } 17235d4f98a2SYan Zheng if (dirty) 17245d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 17253fd0a558SYan, Zheng if (inode) 17263fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 172783d4cfd4SJosef Bacik return ret; 17285d4f98a2SYan Zheng } 17295d4f98a2SYan Zheng 17305d4f98a2SYan Zheng static noinline_for_stack 17315d4f98a2SYan Zheng int memcmp_node_keys(struct extent_buffer *eb, int slot, 17325d4f98a2SYan Zheng struct btrfs_path *path, int level) 17335d4f98a2SYan Zheng { 17345d4f98a2SYan Zheng struct btrfs_disk_key key1; 17355d4f98a2SYan Zheng struct btrfs_disk_key key2; 17365d4f98a2SYan Zheng btrfs_node_key(eb, &key1, slot); 17375d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &key2, path->slots[level]); 17385d4f98a2SYan Zheng return memcmp(&key1, &key2, sizeof(key1)); 17395d4f98a2SYan Zheng } 17405d4f98a2SYan Zheng 17415d4f98a2SYan Zheng /* 17425d4f98a2SYan Zheng * try to replace tree blocks in fs tree with the new blocks 17435d4f98a2SYan Zheng * in reloc tree. tree blocks haven't been modified since the 17445d4f98a2SYan Zheng * reloc tree was create can be replaced. 17455d4f98a2SYan Zheng * 17465d4f98a2SYan Zheng * if a block was replaced, level of the block + 1 is returned. 17475d4f98a2SYan Zheng * if no block got replaced, 0 is returned. if there are other 17485d4f98a2SYan Zheng * errors, a negative error number is returned. 17495d4f98a2SYan Zheng */ 17503fd0a558SYan, Zheng static noinline_for_stack 17513d0174f7SQu Wenruo int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc, 17525d4f98a2SYan Zheng struct btrfs_root *dest, struct btrfs_root *src, 17535d4f98a2SYan Zheng struct btrfs_path *path, struct btrfs_key *next_key, 17545d4f98a2SYan Zheng int lowest_level, int max_level) 17555d4f98a2SYan Zheng { 17560b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = dest->fs_info; 17575d4f98a2SYan Zheng struct extent_buffer *eb; 17585d4f98a2SYan Zheng struct extent_buffer *parent; 17595d4f98a2SYan Zheng struct btrfs_key key; 17605d4f98a2SYan Zheng u64 old_bytenr; 17615d4f98a2SYan Zheng u64 new_bytenr; 17625d4f98a2SYan Zheng u64 old_ptr_gen; 17635d4f98a2SYan Zheng u64 new_ptr_gen; 17645d4f98a2SYan Zheng u64 last_snapshot; 17655d4f98a2SYan Zheng u32 blocksize; 17663fd0a558SYan, Zheng int cow = 0; 17675d4f98a2SYan Zheng int level; 17685d4f98a2SYan Zheng int ret; 17695d4f98a2SYan Zheng int slot; 17705d4f98a2SYan Zheng 17715d4f98a2SYan Zheng BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 17725d4f98a2SYan Zheng BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID); 17735d4f98a2SYan Zheng 17745d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&src->root_item); 17753fd0a558SYan, Zheng again: 17765d4f98a2SYan Zheng slot = path->slots[lowest_level]; 17775d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot); 17785d4f98a2SYan Zheng 17795d4f98a2SYan Zheng eb = btrfs_lock_root_node(dest); 17808bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 17815d4f98a2SYan Zheng level = btrfs_header_level(eb); 17825d4f98a2SYan Zheng 17835d4f98a2SYan Zheng if (level < lowest_level) { 17845d4f98a2SYan Zheng btrfs_tree_unlock(eb); 17855d4f98a2SYan Zheng free_extent_buffer(eb); 17865d4f98a2SYan Zheng return 0; 17875d4f98a2SYan Zheng } 17885d4f98a2SYan Zheng 17893fd0a558SYan, Zheng if (cow) { 17905d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb); 17915d4f98a2SYan Zheng BUG_ON(ret); 17923fd0a558SYan, Zheng } 17938bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 17945d4f98a2SYan Zheng 17955d4f98a2SYan Zheng if (next_key) { 17965d4f98a2SYan Zheng next_key->objectid = (u64)-1; 17975d4f98a2SYan Zheng next_key->type = (u8)-1; 17985d4f98a2SYan Zheng next_key->offset = (u64)-1; 17995d4f98a2SYan Zheng } 18005d4f98a2SYan Zheng 18015d4f98a2SYan Zheng parent = eb; 18025d4f98a2SYan Zheng while (1) { 1803581c1760SQu Wenruo struct btrfs_key first_key; 1804581c1760SQu Wenruo 18055d4f98a2SYan Zheng level = btrfs_header_level(parent); 18065d4f98a2SYan Zheng BUG_ON(level < lowest_level); 18075d4f98a2SYan Zheng 18085d4f98a2SYan Zheng ret = btrfs_bin_search(parent, &key, level, &slot); 1809cbca7d59SFilipe Manana if (ret < 0) 1810cbca7d59SFilipe Manana break; 18115d4f98a2SYan Zheng if (ret && slot > 0) 18125d4f98a2SYan Zheng slot--; 18135d4f98a2SYan Zheng 18145d4f98a2SYan Zheng if (next_key && slot + 1 < btrfs_header_nritems(parent)) 18155d4f98a2SYan Zheng btrfs_node_key_to_cpu(parent, next_key, slot + 1); 18165d4f98a2SYan Zheng 18175d4f98a2SYan Zheng old_bytenr = btrfs_node_blockptr(parent, slot); 18180b246afaSJeff Mahoney blocksize = fs_info->nodesize; 18195d4f98a2SYan Zheng old_ptr_gen = btrfs_node_ptr_generation(parent, slot); 182017515f1bSQu Wenruo btrfs_node_key_to_cpu(parent, &first_key, slot); 18215d4f98a2SYan Zheng 18225d4f98a2SYan Zheng if (level <= max_level) { 18235d4f98a2SYan Zheng eb = path->nodes[level]; 18245d4f98a2SYan Zheng new_bytenr = btrfs_node_blockptr(eb, 18255d4f98a2SYan Zheng path->slots[level]); 18265d4f98a2SYan Zheng new_ptr_gen = btrfs_node_ptr_generation(eb, 18275d4f98a2SYan Zheng path->slots[level]); 18285d4f98a2SYan Zheng } else { 18295d4f98a2SYan Zheng new_bytenr = 0; 18305d4f98a2SYan Zheng new_ptr_gen = 0; 18315d4f98a2SYan Zheng } 18325d4f98a2SYan Zheng 1833fae7f21cSDulshani Gunawardhana if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) { 18345d4f98a2SYan Zheng ret = level; 18355d4f98a2SYan Zheng break; 18365d4f98a2SYan Zheng } 18375d4f98a2SYan Zheng 18385d4f98a2SYan Zheng if (new_bytenr == 0 || old_ptr_gen > last_snapshot || 18395d4f98a2SYan Zheng memcmp_node_keys(parent, slot, path, level)) { 18403fd0a558SYan, Zheng if (level <= lowest_level) { 18415d4f98a2SYan Zheng ret = 0; 18425d4f98a2SYan Zheng break; 18435d4f98a2SYan Zheng } 18445d4f98a2SYan Zheng 1845581c1760SQu Wenruo eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen, 1846581c1760SQu Wenruo level - 1, &first_key); 184764c043deSLiu Bo if (IS_ERR(eb)) { 184864c043deSLiu Bo ret = PTR_ERR(eb); 1849264813acSLiu Bo break; 185064c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 185164c043deSLiu Bo ret = -EIO; 1852416bc658SJosef Bacik free_extent_buffer(eb); 1853379cde74SStefan Behrens break; 1854416bc658SJosef Bacik } 18555d4f98a2SYan Zheng btrfs_tree_lock(eb); 18563fd0a558SYan, Zheng if (cow) { 18575d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, parent, 18585d4f98a2SYan Zheng slot, &eb); 18595d4f98a2SYan Zheng BUG_ON(ret); 18605d4f98a2SYan Zheng } 18618bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 18625d4f98a2SYan Zheng 18635d4f98a2SYan Zheng btrfs_tree_unlock(parent); 18645d4f98a2SYan Zheng free_extent_buffer(parent); 18655d4f98a2SYan Zheng 18665d4f98a2SYan Zheng parent = eb; 18675d4f98a2SYan Zheng continue; 18685d4f98a2SYan Zheng } 18695d4f98a2SYan Zheng 18703fd0a558SYan, Zheng if (!cow) { 18713fd0a558SYan, Zheng btrfs_tree_unlock(parent); 18723fd0a558SYan, Zheng free_extent_buffer(parent); 18733fd0a558SYan, Zheng cow = 1; 18743fd0a558SYan, Zheng goto again; 18753fd0a558SYan, Zheng } 18763fd0a558SYan, Zheng 18775d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 18785d4f98a2SYan Zheng path->slots[level]); 1879b3b4aa74SDavid Sterba btrfs_release_path(path); 18805d4f98a2SYan Zheng 18815d4f98a2SYan Zheng path->lowest_level = level; 18825d4f98a2SYan Zheng ret = btrfs_search_slot(trans, src, &key, path, 0, 1); 18835d4f98a2SYan Zheng path->lowest_level = 0; 18845d4f98a2SYan Zheng BUG_ON(ret); 18855d4f98a2SYan Zheng 18865d4f98a2SYan Zheng /* 1887824d8dffSQu Wenruo * Info qgroup to trace both subtrees. 1888824d8dffSQu Wenruo * 1889824d8dffSQu Wenruo * We must trace both trees. 1890824d8dffSQu Wenruo * 1) Tree reloc subtree 1891824d8dffSQu Wenruo * If not traced, we will leak data numbers 1892824d8dffSQu Wenruo * 2) Fs subtree 1893824d8dffSQu Wenruo * If not traced, we will double count old data 1894f616f5cdSQu Wenruo * 1895f616f5cdSQu Wenruo * We don't scan the subtree right now, but only record 1896f616f5cdSQu Wenruo * the swapped tree blocks. 1897f616f5cdSQu Wenruo * The real subtree rescan is delayed until we have new 1898f616f5cdSQu Wenruo * CoW on the subtree root node before transaction commit. 1899824d8dffSQu Wenruo */ 1900370a11b8SQu Wenruo ret = btrfs_qgroup_add_swapped_blocks(trans, dest, 1901370a11b8SQu Wenruo rc->block_group, parent, slot, 1902370a11b8SQu Wenruo path->nodes[level], path->slots[level], 1903370a11b8SQu Wenruo last_snapshot); 1904370a11b8SQu Wenruo if (ret < 0) 1905370a11b8SQu Wenruo break; 1906824d8dffSQu Wenruo /* 19075d4f98a2SYan Zheng * swap blocks in fs tree and reloc tree. 19085d4f98a2SYan Zheng */ 19095d4f98a2SYan Zheng btrfs_set_node_blockptr(parent, slot, new_bytenr); 19105d4f98a2SYan Zheng btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen); 19115d4f98a2SYan Zheng btrfs_mark_buffer_dirty(parent); 19125d4f98a2SYan Zheng 19135d4f98a2SYan Zheng btrfs_set_node_blockptr(path->nodes[level], 19145d4f98a2SYan Zheng path->slots[level], old_bytenr); 19155d4f98a2SYan Zheng btrfs_set_node_ptr_generation(path->nodes[level], 19165d4f98a2SYan Zheng path->slots[level], old_ptr_gen); 19175d4f98a2SYan Zheng btrfs_mark_buffer_dirty(path->nodes[level]); 19185d4f98a2SYan Zheng 191984f7d8e6SJosef Bacik ret = btrfs_inc_extent_ref(trans, src, old_bytenr, 19202ff7e61eSJeff Mahoney blocksize, path->nodes[level]->start, 1921b06c4bf5SFilipe Manana src->root_key.objectid, level - 1, 0); 19225d4f98a2SYan Zheng BUG_ON(ret); 192384f7d8e6SJosef Bacik ret = btrfs_inc_extent_ref(trans, dest, new_bytenr, 19242ff7e61eSJeff Mahoney blocksize, 0, dest->root_key.objectid, 19252ff7e61eSJeff Mahoney level - 1, 0); 19265d4f98a2SYan Zheng BUG_ON(ret); 19275d4f98a2SYan Zheng 192884f7d8e6SJosef Bacik ret = btrfs_free_extent(trans, src, new_bytenr, blocksize, 19295d4f98a2SYan Zheng path->nodes[level]->start, 1930b06c4bf5SFilipe Manana src->root_key.objectid, level - 1, 0); 19315d4f98a2SYan Zheng BUG_ON(ret); 19325d4f98a2SYan Zheng 193384f7d8e6SJosef Bacik ret = btrfs_free_extent(trans, dest, old_bytenr, blocksize, 19345d4f98a2SYan Zheng 0, dest->root_key.objectid, level - 1, 1935b06c4bf5SFilipe Manana 0); 19365d4f98a2SYan Zheng BUG_ON(ret); 19375d4f98a2SYan Zheng 19385d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 19395d4f98a2SYan Zheng 19405d4f98a2SYan Zheng ret = level; 19415d4f98a2SYan Zheng break; 19425d4f98a2SYan Zheng } 19435d4f98a2SYan Zheng btrfs_tree_unlock(parent); 19445d4f98a2SYan Zheng free_extent_buffer(parent); 19455d4f98a2SYan Zheng return ret; 19465d4f98a2SYan Zheng } 19475d4f98a2SYan Zheng 19485d4f98a2SYan Zheng /* 19495d4f98a2SYan Zheng * helper to find next relocated block in reloc tree 19505d4f98a2SYan Zheng */ 19515d4f98a2SYan Zheng static noinline_for_stack 19525d4f98a2SYan Zheng int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 19535d4f98a2SYan Zheng int *level) 19545d4f98a2SYan Zheng { 19555d4f98a2SYan Zheng struct extent_buffer *eb; 19565d4f98a2SYan Zheng int i; 19575d4f98a2SYan Zheng u64 last_snapshot; 19585d4f98a2SYan Zheng u32 nritems; 19595d4f98a2SYan Zheng 19605d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 19615d4f98a2SYan Zheng 19625d4f98a2SYan Zheng for (i = 0; i < *level; i++) { 19635d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 19645d4f98a2SYan Zheng path->nodes[i] = NULL; 19655d4f98a2SYan Zheng } 19665d4f98a2SYan Zheng 19675d4f98a2SYan Zheng for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) { 19685d4f98a2SYan Zheng eb = path->nodes[i]; 19695d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 19705d4f98a2SYan Zheng while (path->slots[i] + 1 < nritems) { 19715d4f98a2SYan Zheng path->slots[i]++; 19725d4f98a2SYan Zheng if (btrfs_node_ptr_generation(eb, path->slots[i]) <= 19735d4f98a2SYan Zheng last_snapshot) 19745d4f98a2SYan Zheng continue; 19755d4f98a2SYan Zheng 19765d4f98a2SYan Zheng *level = i; 19775d4f98a2SYan Zheng return 0; 19785d4f98a2SYan Zheng } 19795d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 19805d4f98a2SYan Zheng path->nodes[i] = NULL; 19815d4f98a2SYan Zheng } 19825d4f98a2SYan Zheng return 1; 19835d4f98a2SYan Zheng } 19845d4f98a2SYan Zheng 19855d4f98a2SYan Zheng /* 19865d4f98a2SYan Zheng * walk down reloc tree to find relocated block of lowest level 19875d4f98a2SYan Zheng */ 19885d4f98a2SYan Zheng static noinline_for_stack 19895d4f98a2SYan Zheng int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 19905d4f98a2SYan Zheng int *level) 19915d4f98a2SYan Zheng { 19922ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 19935d4f98a2SYan Zheng struct extent_buffer *eb = NULL; 19945d4f98a2SYan Zheng int i; 19955d4f98a2SYan Zheng u64 bytenr; 19965d4f98a2SYan Zheng u64 ptr_gen = 0; 19975d4f98a2SYan Zheng u64 last_snapshot; 19985d4f98a2SYan Zheng u32 nritems; 19995d4f98a2SYan Zheng 20005d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 20015d4f98a2SYan Zheng 20025d4f98a2SYan Zheng for (i = *level; i > 0; i--) { 2003581c1760SQu Wenruo struct btrfs_key first_key; 2004581c1760SQu Wenruo 20055d4f98a2SYan Zheng eb = path->nodes[i]; 20065d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 20075d4f98a2SYan Zheng while (path->slots[i] < nritems) { 20085d4f98a2SYan Zheng ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]); 20095d4f98a2SYan Zheng if (ptr_gen > last_snapshot) 20105d4f98a2SYan Zheng break; 20115d4f98a2SYan Zheng path->slots[i]++; 20125d4f98a2SYan Zheng } 20135d4f98a2SYan Zheng if (path->slots[i] >= nritems) { 20145d4f98a2SYan Zheng if (i == *level) 20155d4f98a2SYan Zheng break; 20165d4f98a2SYan Zheng *level = i + 1; 20175d4f98a2SYan Zheng return 0; 20185d4f98a2SYan Zheng } 20195d4f98a2SYan Zheng if (i == 1) { 20205d4f98a2SYan Zheng *level = i; 20215d4f98a2SYan Zheng return 0; 20225d4f98a2SYan Zheng } 20235d4f98a2SYan Zheng 20245d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(eb, path->slots[i]); 2025581c1760SQu Wenruo btrfs_node_key_to_cpu(eb, &first_key, path->slots[i]); 2026581c1760SQu Wenruo eb = read_tree_block(fs_info, bytenr, ptr_gen, i - 1, 2027581c1760SQu Wenruo &first_key); 202864c043deSLiu Bo if (IS_ERR(eb)) { 202964c043deSLiu Bo return PTR_ERR(eb); 203064c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2031416bc658SJosef Bacik free_extent_buffer(eb); 2032416bc658SJosef Bacik return -EIO; 2033416bc658SJosef Bacik } 20345d4f98a2SYan Zheng BUG_ON(btrfs_header_level(eb) != i - 1); 20355d4f98a2SYan Zheng path->nodes[i - 1] = eb; 20365d4f98a2SYan Zheng path->slots[i - 1] = 0; 20375d4f98a2SYan Zheng } 20385d4f98a2SYan Zheng return 1; 20395d4f98a2SYan Zheng } 20405d4f98a2SYan Zheng 20415d4f98a2SYan Zheng /* 20425d4f98a2SYan Zheng * invalidate extent cache for file extents whose key in range of 20435d4f98a2SYan Zheng * [min_key, max_key) 20445d4f98a2SYan Zheng */ 20455d4f98a2SYan Zheng static int invalidate_extent_cache(struct btrfs_root *root, 20465d4f98a2SYan Zheng struct btrfs_key *min_key, 20475d4f98a2SYan Zheng struct btrfs_key *max_key) 20485d4f98a2SYan Zheng { 20490b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 20505d4f98a2SYan Zheng struct inode *inode = NULL; 20515d4f98a2SYan Zheng u64 objectid; 20525d4f98a2SYan Zheng u64 start, end; 205333345d01SLi Zefan u64 ino; 20545d4f98a2SYan Zheng 20555d4f98a2SYan Zheng objectid = min_key->objectid; 20565d4f98a2SYan Zheng while (1) { 20575d4f98a2SYan Zheng cond_resched(); 20585d4f98a2SYan Zheng iput(inode); 20595d4f98a2SYan Zheng 20605d4f98a2SYan Zheng if (objectid > max_key->objectid) 20615d4f98a2SYan Zheng break; 20625d4f98a2SYan Zheng 20635d4f98a2SYan Zheng inode = find_next_inode(root, objectid); 20645d4f98a2SYan Zheng if (!inode) 20655d4f98a2SYan Zheng break; 20664a0cc7caSNikolay Borisov ino = btrfs_ino(BTRFS_I(inode)); 20675d4f98a2SYan Zheng 206833345d01SLi Zefan if (ino > max_key->objectid) { 20695d4f98a2SYan Zheng iput(inode); 20705d4f98a2SYan Zheng break; 20715d4f98a2SYan Zheng } 20725d4f98a2SYan Zheng 207333345d01SLi Zefan objectid = ino + 1; 20745d4f98a2SYan Zheng if (!S_ISREG(inode->i_mode)) 20755d4f98a2SYan Zheng continue; 20765d4f98a2SYan Zheng 207733345d01SLi Zefan if (unlikely(min_key->objectid == ino)) { 20785d4f98a2SYan Zheng if (min_key->type > BTRFS_EXTENT_DATA_KEY) 20795d4f98a2SYan Zheng continue; 20805d4f98a2SYan Zheng if (min_key->type < BTRFS_EXTENT_DATA_KEY) 20815d4f98a2SYan Zheng start = 0; 20825d4f98a2SYan Zheng else { 20835d4f98a2SYan Zheng start = min_key->offset; 20840b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize)); 20855d4f98a2SYan Zheng } 20865d4f98a2SYan Zheng } else { 20875d4f98a2SYan Zheng start = 0; 20885d4f98a2SYan Zheng } 20895d4f98a2SYan Zheng 209033345d01SLi Zefan if (unlikely(max_key->objectid == ino)) { 20915d4f98a2SYan Zheng if (max_key->type < BTRFS_EXTENT_DATA_KEY) 20925d4f98a2SYan Zheng continue; 20935d4f98a2SYan Zheng if (max_key->type > BTRFS_EXTENT_DATA_KEY) { 20945d4f98a2SYan Zheng end = (u64)-1; 20955d4f98a2SYan Zheng } else { 20965d4f98a2SYan Zheng if (max_key->offset == 0) 20975d4f98a2SYan Zheng continue; 20985d4f98a2SYan Zheng end = max_key->offset; 20990b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 21005d4f98a2SYan Zheng end--; 21015d4f98a2SYan Zheng } 21025d4f98a2SYan Zheng } else { 21035d4f98a2SYan Zheng end = (u64)-1; 21045d4f98a2SYan Zheng } 21055d4f98a2SYan Zheng 21065d4f98a2SYan Zheng /* the lock_extent waits for readpage to complete */ 2107d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 2108dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1); 2109d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 21105d4f98a2SYan Zheng } 21115d4f98a2SYan Zheng return 0; 21125d4f98a2SYan Zheng } 21135d4f98a2SYan Zheng 21145d4f98a2SYan Zheng static int find_next_key(struct btrfs_path *path, int level, 21155d4f98a2SYan Zheng struct btrfs_key *key) 21165d4f98a2SYan Zheng 21175d4f98a2SYan Zheng { 21185d4f98a2SYan Zheng while (level < BTRFS_MAX_LEVEL) { 21195d4f98a2SYan Zheng if (!path->nodes[level]) 21205d4f98a2SYan Zheng break; 21215d4f98a2SYan Zheng if (path->slots[level] + 1 < 21225d4f98a2SYan Zheng btrfs_header_nritems(path->nodes[level])) { 21235d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], key, 21245d4f98a2SYan Zheng path->slots[level] + 1); 21255d4f98a2SYan Zheng return 0; 21265d4f98a2SYan Zheng } 21275d4f98a2SYan Zheng level++; 21285d4f98a2SYan Zheng } 21295d4f98a2SYan Zheng return 1; 21305d4f98a2SYan Zheng } 21315d4f98a2SYan Zheng 21325d4f98a2SYan Zheng /* 2133d2311e69SQu Wenruo * Insert current subvolume into reloc_control::dirty_subvol_roots 2134d2311e69SQu Wenruo */ 2135d2311e69SQu Wenruo static void insert_dirty_subvol(struct btrfs_trans_handle *trans, 2136d2311e69SQu Wenruo struct reloc_control *rc, 2137d2311e69SQu Wenruo struct btrfs_root *root) 2138d2311e69SQu Wenruo { 2139d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 2140d2311e69SQu Wenruo struct btrfs_root_item *reloc_root_item; 2141d2311e69SQu Wenruo 2142d2311e69SQu Wenruo /* @root must be a subvolume tree root with a valid reloc tree */ 2143d2311e69SQu Wenruo ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 2144d2311e69SQu Wenruo ASSERT(reloc_root); 2145d2311e69SQu Wenruo 2146d2311e69SQu Wenruo reloc_root_item = &reloc_root->root_item; 2147d2311e69SQu Wenruo memset(&reloc_root_item->drop_progress, 0, 2148d2311e69SQu Wenruo sizeof(reloc_root_item->drop_progress)); 2149d2311e69SQu Wenruo reloc_root_item->drop_level = 0; 2150d2311e69SQu Wenruo btrfs_set_root_refs(reloc_root_item, 0); 2151d2311e69SQu Wenruo btrfs_update_reloc_root(trans, root); 2152d2311e69SQu Wenruo 2153d2311e69SQu Wenruo if (list_empty(&root->reloc_dirty_list)) { 2154d2311e69SQu Wenruo btrfs_grab_fs_root(root); 2155d2311e69SQu Wenruo list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots); 2156d2311e69SQu Wenruo } 2157d2311e69SQu Wenruo } 2158d2311e69SQu Wenruo 2159d2311e69SQu Wenruo static int clean_dirty_subvols(struct reloc_control *rc) 2160d2311e69SQu Wenruo { 2161d2311e69SQu Wenruo struct btrfs_root *root; 2162d2311e69SQu Wenruo struct btrfs_root *next; 2163d2311e69SQu Wenruo int ret = 0; 2164d2311e69SQu Wenruo 2165d2311e69SQu Wenruo list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots, 2166d2311e69SQu Wenruo reloc_dirty_list) { 2167d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 2168d2311e69SQu Wenruo 2169d2311e69SQu Wenruo clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 2170d2311e69SQu Wenruo list_del_init(&root->reloc_dirty_list); 2171d2311e69SQu Wenruo root->reloc_root = NULL; 2172d2311e69SQu Wenruo if (reloc_root) { 2173d2311e69SQu Wenruo int ret2; 2174d2311e69SQu Wenruo 2175d2311e69SQu Wenruo ret2 = btrfs_drop_snapshot(reloc_root, NULL, 0, 1); 2176d2311e69SQu Wenruo if (ret2 < 0 && !ret) 2177d2311e69SQu Wenruo ret = ret2; 2178d2311e69SQu Wenruo } 2179d2311e69SQu Wenruo btrfs_put_fs_root(root); 2180d2311e69SQu Wenruo } 2181d2311e69SQu Wenruo return ret; 2182d2311e69SQu Wenruo } 2183d2311e69SQu Wenruo 2184d2311e69SQu Wenruo /* 21855d4f98a2SYan Zheng * merge the relocated tree blocks in reloc tree with corresponding 21865d4f98a2SYan Zheng * fs tree. 21875d4f98a2SYan Zheng */ 21885d4f98a2SYan Zheng static noinline_for_stack int merge_reloc_root(struct reloc_control *rc, 21895d4f98a2SYan Zheng struct btrfs_root *root) 21905d4f98a2SYan Zheng { 21910b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 21925d4f98a2SYan Zheng struct btrfs_key key; 21935d4f98a2SYan Zheng struct btrfs_key next_key; 21949e6a0c52SJosef Bacik struct btrfs_trans_handle *trans = NULL; 21955d4f98a2SYan Zheng struct btrfs_root *reloc_root; 21965d4f98a2SYan Zheng struct btrfs_root_item *root_item; 21975d4f98a2SYan Zheng struct btrfs_path *path; 21983fd0a558SYan, Zheng struct extent_buffer *leaf; 21995d4f98a2SYan Zheng int level; 22005d4f98a2SYan Zheng int max_level; 22015d4f98a2SYan Zheng int replaced = 0; 22025d4f98a2SYan Zheng int ret; 22035d4f98a2SYan Zheng int err = 0; 22043fd0a558SYan, Zheng u32 min_reserved; 22055d4f98a2SYan Zheng 22065d4f98a2SYan Zheng path = btrfs_alloc_path(); 22075d4f98a2SYan Zheng if (!path) 22085d4f98a2SYan Zheng return -ENOMEM; 2209e4058b54SDavid Sterba path->reada = READA_FORWARD; 22105d4f98a2SYan Zheng 22115d4f98a2SYan Zheng reloc_root = root->reloc_root; 22125d4f98a2SYan Zheng root_item = &reloc_root->root_item; 22135d4f98a2SYan Zheng 22145d4f98a2SYan Zheng if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { 22155d4f98a2SYan Zheng level = btrfs_root_level(root_item); 22165d4f98a2SYan Zheng extent_buffer_get(reloc_root->node); 22175d4f98a2SYan Zheng path->nodes[level] = reloc_root->node; 22185d4f98a2SYan Zheng path->slots[level] = 0; 22195d4f98a2SYan Zheng } else { 22205d4f98a2SYan Zheng btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); 22215d4f98a2SYan Zheng 22225d4f98a2SYan Zheng level = root_item->drop_level; 22235d4f98a2SYan Zheng BUG_ON(level == 0); 22245d4f98a2SYan Zheng path->lowest_level = level; 22255d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0); 222633c66f43SYan Zheng path->lowest_level = 0; 22275d4f98a2SYan Zheng if (ret < 0) { 22285d4f98a2SYan Zheng btrfs_free_path(path); 22295d4f98a2SYan Zheng return ret; 22305d4f98a2SYan Zheng } 22315d4f98a2SYan Zheng 22325d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &next_key, 22335d4f98a2SYan Zheng path->slots[level]); 22345d4f98a2SYan Zheng WARN_ON(memcmp(&key, &next_key, sizeof(key))); 22355d4f98a2SYan Zheng 22365d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 22375d4f98a2SYan Zheng } 22385d4f98a2SYan Zheng 22390b246afaSJeff Mahoney min_reserved = fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 22405d4f98a2SYan Zheng memset(&next_key, 0, sizeof(next_key)); 22415d4f98a2SYan Zheng 22425d4f98a2SYan Zheng while (1) { 224308e007d2SMiao Xie ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved, 224408e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 22453fd0a558SYan, Zheng if (ret) { 22469e6a0c52SJosef Bacik err = ret; 22479e6a0c52SJosef Bacik goto out; 22483fd0a558SYan, Zheng } 22499e6a0c52SJosef Bacik trans = btrfs_start_transaction(root, 0); 22509e6a0c52SJosef Bacik if (IS_ERR(trans)) { 22519e6a0c52SJosef Bacik err = PTR_ERR(trans); 22529e6a0c52SJosef Bacik trans = NULL; 22539e6a0c52SJosef Bacik goto out; 22549e6a0c52SJosef Bacik } 22559e6a0c52SJosef Bacik trans->block_rsv = rc->block_rsv; 22563fd0a558SYan, Zheng 22573fd0a558SYan, Zheng replaced = 0; 22585d4f98a2SYan Zheng max_level = level; 22595d4f98a2SYan Zheng 22605d4f98a2SYan Zheng ret = walk_down_reloc_tree(reloc_root, path, &level); 22615d4f98a2SYan Zheng if (ret < 0) { 22625d4f98a2SYan Zheng err = ret; 22635d4f98a2SYan Zheng goto out; 22645d4f98a2SYan Zheng } 22655d4f98a2SYan Zheng if (ret > 0) 22665d4f98a2SYan Zheng break; 22675d4f98a2SYan Zheng 22685d4f98a2SYan Zheng if (!find_next_key(path, level, &key) && 22695d4f98a2SYan Zheng btrfs_comp_cpu_keys(&next_key, &key) >= 0) { 22705d4f98a2SYan Zheng ret = 0; 22715d4f98a2SYan Zheng } else { 22723d0174f7SQu Wenruo ret = replace_path(trans, rc, root, reloc_root, path, 22733fd0a558SYan, Zheng &next_key, level, max_level); 22745d4f98a2SYan Zheng } 22755d4f98a2SYan Zheng if (ret < 0) { 22765d4f98a2SYan Zheng err = ret; 22775d4f98a2SYan Zheng goto out; 22785d4f98a2SYan Zheng } 22795d4f98a2SYan Zheng 22805d4f98a2SYan Zheng if (ret > 0) { 22815d4f98a2SYan Zheng level = ret; 22825d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 22835d4f98a2SYan Zheng path->slots[level]); 22845d4f98a2SYan Zheng replaced = 1; 22855d4f98a2SYan Zheng } 22865d4f98a2SYan Zheng 22875d4f98a2SYan Zheng ret = walk_up_reloc_tree(reloc_root, path, &level); 22885d4f98a2SYan Zheng if (ret > 0) 22895d4f98a2SYan Zheng break; 22905d4f98a2SYan Zheng 22915d4f98a2SYan Zheng BUG_ON(level == 0); 22925d4f98a2SYan Zheng /* 22935d4f98a2SYan Zheng * save the merging progress in the drop_progress. 22945d4f98a2SYan Zheng * this is OK since root refs == 1 in this case. 22955d4f98a2SYan Zheng */ 22965d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &root_item->drop_progress, 22975d4f98a2SYan Zheng path->slots[level]); 22985d4f98a2SYan Zheng root_item->drop_level = level; 22995d4f98a2SYan Zheng 23003a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 23019e6a0c52SJosef Bacik trans = NULL; 23025d4f98a2SYan Zheng 23032ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 23045d4f98a2SYan Zheng 23055d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 23065d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 23075d4f98a2SYan Zheng } 23085d4f98a2SYan Zheng 23095d4f98a2SYan Zheng /* 23105d4f98a2SYan Zheng * handle the case only one block in the fs tree need to be 23115d4f98a2SYan Zheng * relocated and the block is tree root. 23125d4f98a2SYan Zheng */ 23135d4f98a2SYan Zheng leaf = btrfs_lock_root_node(root); 23145d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf); 23155d4f98a2SYan Zheng btrfs_tree_unlock(leaf); 23165d4f98a2SYan Zheng free_extent_buffer(leaf); 23175d4f98a2SYan Zheng if (ret < 0) 23185d4f98a2SYan Zheng err = ret; 23195d4f98a2SYan Zheng out: 23205d4f98a2SYan Zheng btrfs_free_path(path); 23215d4f98a2SYan Zheng 2322d2311e69SQu Wenruo if (err == 0) 2323d2311e69SQu Wenruo insert_dirty_subvol(trans, rc, root); 23245d4f98a2SYan Zheng 23259e6a0c52SJosef Bacik if (trans) 23263a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 23275d4f98a2SYan Zheng 23282ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 23295d4f98a2SYan Zheng 23305d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 23315d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 23325d4f98a2SYan Zheng 23335d4f98a2SYan Zheng return err; 23345d4f98a2SYan Zheng } 23355d4f98a2SYan Zheng 23363fd0a558SYan, Zheng static noinline_for_stack 23373fd0a558SYan, Zheng int prepare_to_merge(struct reloc_control *rc, int err) 23385d4f98a2SYan Zheng { 23393fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 23400b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 23413fd0a558SYan, Zheng struct btrfs_root *reloc_root; 23425d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 23433fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 23443fd0a558SYan, Zheng u64 num_bytes = 0; 23453fd0a558SYan, Zheng int ret; 23463fd0a558SYan, Zheng 23470b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 23480b246afaSJeff Mahoney rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 23493fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated * 2; 23500b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 23517585717fSChris Mason 23523fd0a558SYan, Zheng again: 23533fd0a558SYan, Zheng if (!err) { 23543fd0a558SYan, Zheng num_bytes = rc->merging_rsv_size; 235508e007d2SMiao Xie ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes, 235608e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 23573fd0a558SYan, Zheng if (ret) 23583fd0a558SYan, Zheng err = ret; 23593fd0a558SYan, Zheng } 23603fd0a558SYan, Zheng 23617a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 23623612b495STsutomu Itoh if (IS_ERR(trans)) { 23633612b495STsutomu Itoh if (!err) 23642ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 23652ff7e61eSJeff Mahoney num_bytes); 23663612b495STsutomu Itoh return PTR_ERR(trans); 23673612b495STsutomu Itoh } 23683fd0a558SYan, Zheng 23693fd0a558SYan, Zheng if (!err) { 23703fd0a558SYan, Zheng if (num_bytes != rc->merging_rsv_size) { 23713a45bb20SJeff Mahoney btrfs_end_transaction(trans); 23722ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 23732ff7e61eSJeff Mahoney num_bytes); 23743fd0a558SYan, Zheng goto again; 23753fd0a558SYan, Zheng } 23763fd0a558SYan, Zheng } 23773fd0a558SYan, Zheng 23783fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 23793fd0a558SYan, Zheng 23803fd0a558SYan, Zheng while (!list_empty(&rc->reloc_roots)) { 23813fd0a558SYan, Zheng reloc_root = list_entry(rc->reloc_roots.next, 23823fd0a558SYan, Zheng struct btrfs_root, root_list); 23833fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 23843fd0a558SYan, Zheng 23850b246afaSJeff Mahoney root = read_fs_root(fs_info, reloc_root->root_key.offset); 23863fd0a558SYan, Zheng BUG_ON(IS_ERR(root)); 23873fd0a558SYan, Zheng BUG_ON(root->reloc_root != reloc_root); 23883fd0a558SYan, Zheng 23893fd0a558SYan, Zheng /* 23903fd0a558SYan, Zheng * set reference count to 1, so btrfs_recover_relocation 23913fd0a558SYan, Zheng * knows it should resumes merging 23923fd0a558SYan, Zheng */ 23933fd0a558SYan, Zheng if (!err) 23943fd0a558SYan, Zheng btrfs_set_root_refs(&reloc_root->root_item, 1); 23953fd0a558SYan, Zheng btrfs_update_reloc_root(trans, root); 23963fd0a558SYan, Zheng 23973fd0a558SYan, Zheng list_add(&reloc_root->root_list, &reloc_roots); 23983fd0a558SYan, Zheng } 23993fd0a558SYan, Zheng 24003fd0a558SYan, Zheng list_splice(&reloc_roots, &rc->reloc_roots); 24013fd0a558SYan, Zheng 24023fd0a558SYan, Zheng if (!err) 24033a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 24043fd0a558SYan, Zheng else 24053a45bb20SJeff Mahoney btrfs_end_transaction(trans); 24063fd0a558SYan, Zheng return err; 24073fd0a558SYan, Zheng } 24083fd0a558SYan, Zheng 24093fd0a558SYan, Zheng static noinline_for_stack 2410aca1bba6SLiu Bo void free_reloc_roots(struct list_head *list) 2411aca1bba6SLiu Bo { 2412aca1bba6SLiu Bo struct btrfs_root *reloc_root; 2413aca1bba6SLiu Bo 2414aca1bba6SLiu Bo while (!list_empty(list)) { 2415aca1bba6SLiu Bo reloc_root = list_entry(list->next, struct btrfs_root, 2416aca1bba6SLiu Bo root_list); 2417bb166d72SNaohiro Aota __del_reloc_root(reloc_root); 24186bdf131fSJosef Bacik free_extent_buffer(reloc_root->node); 24196bdf131fSJosef Bacik free_extent_buffer(reloc_root->commit_root); 24206bdf131fSJosef Bacik reloc_root->node = NULL; 24216bdf131fSJosef Bacik reloc_root->commit_root = NULL; 2422aca1bba6SLiu Bo } 2423aca1bba6SLiu Bo } 2424aca1bba6SLiu Bo 2425aca1bba6SLiu Bo static noinline_for_stack 242694404e82SDavid Sterba void merge_reloc_roots(struct reloc_control *rc) 24273fd0a558SYan, Zheng { 24280b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 24295d4f98a2SYan Zheng struct btrfs_root *root; 24305d4f98a2SYan Zheng struct btrfs_root *reloc_root; 24313fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 24323fd0a558SYan, Zheng int found = 0; 2433aca1bba6SLiu Bo int ret = 0; 24343fd0a558SYan, Zheng again: 24353fd0a558SYan, Zheng root = rc->extent_root; 24367585717fSChris Mason 24377585717fSChris Mason /* 24387585717fSChris Mason * this serializes us with btrfs_record_root_in_transaction, 24397585717fSChris Mason * we have to make sure nobody is in the middle of 24407585717fSChris Mason * adding their roots to the list while we are 24417585717fSChris Mason * doing this splice 24427585717fSChris Mason */ 24430b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 24443fd0a558SYan, Zheng list_splice_init(&rc->reloc_roots, &reloc_roots); 24450b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 24465d4f98a2SYan Zheng 24473fd0a558SYan, Zheng while (!list_empty(&reloc_roots)) { 24483fd0a558SYan, Zheng found = 1; 24493fd0a558SYan, Zheng reloc_root = list_entry(reloc_roots.next, 24503fd0a558SYan, Zheng struct btrfs_root, root_list); 24515d4f98a2SYan Zheng 24525d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 24530b246afaSJeff Mahoney root = read_fs_root(fs_info, 24545d4f98a2SYan Zheng reloc_root->root_key.offset); 24555d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 24565d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 24575d4f98a2SYan Zheng 24583fd0a558SYan, Zheng ret = merge_reloc_root(rc, root); 2459b37b39cdSJosef Bacik if (ret) { 246025e293c2SWang Shilong if (list_empty(&reloc_root->root_list)) 246125e293c2SWang Shilong list_add_tail(&reloc_root->root_list, 246225e293c2SWang Shilong &reloc_roots); 2463aca1bba6SLiu Bo goto out; 2464b37b39cdSJosef Bacik } 24653fd0a558SYan, Zheng } else { 24663fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 24673fd0a558SYan, Zheng } 24685d4f98a2SYan Zheng } 24695d4f98a2SYan Zheng 24703fd0a558SYan, Zheng if (found) { 24713fd0a558SYan, Zheng found = 0; 24723fd0a558SYan, Zheng goto again; 24735d4f98a2SYan Zheng } 2474aca1bba6SLiu Bo out: 2475aca1bba6SLiu Bo if (ret) { 24760b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 2477aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 2478aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 2479467bb1d2SWang Shilong 2480467bb1d2SWang Shilong /* new reloc root may be added */ 24810b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 2482467bb1d2SWang Shilong list_splice_init(&rc->reloc_roots, &reloc_roots); 24830b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2484467bb1d2SWang Shilong if (!list_empty(&reloc_roots)) 2485467bb1d2SWang Shilong free_reloc_roots(&reloc_roots); 2486aca1bba6SLiu Bo } 2487aca1bba6SLiu Bo 24885d4f98a2SYan Zheng BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); 24895d4f98a2SYan Zheng } 24905d4f98a2SYan Zheng 24915d4f98a2SYan Zheng static void free_block_list(struct rb_root *blocks) 24925d4f98a2SYan Zheng { 24935d4f98a2SYan Zheng struct tree_block *block; 24945d4f98a2SYan Zheng struct rb_node *rb_node; 24955d4f98a2SYan Zheng while ((rb_node = rb_first(blocks))) { 24965d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 24975d4f98a2SYan Zheng rb_erase(rb_node, blocks); 24985d4f98a2SYan Zheng kfree(block); 24995d4f98a2SYan Zheng } 25005d4f98a2SYan Zheng } 25015d4f98a2SYan Zheng 25025d4f98a2SYan Zheng static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans, 25035d4f98a2SYan Zheng struct btrfs_root *reloc_root) 25045d4f98a2SYan Zheng { 25050b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = reloc_root->fs_info; 25065d4f98a2SYan Zheng struct btrfs_root *root; 25075d4f98a2SYan Zheng 25085d4f98a2SYan Zheng if (reloc_root->last_trans == trans->transid) 25095d4f98a2SYan Zheng return 0; 25105d4f98a2SYan Zheng 25110b246afaSJeff Mahoney root = read_fs_root(fs_info, reloc_root->root_key.offset); 25125d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 25135d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 25145d4f98a2SYan Zheng 25155d4f98a2SYan Zheng return btrfs_record_root_in_trans(trans, root); 25165d4f98a2SYan Zheng } 25175d4f98a2SYan Zheng 25183fd0a558SYan, Zheng static noinline_for_stack 25193fd0a558SYan, Zheng struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, 25203fd0a558SYan, Zheng struct reloc_control *rc, 25215d4f98a2SYan Zheng struct backref_node *node, 2522dc4103f9SWang Shilong struct backref_edge *edges[]) 25235d4f98a2SYan Zheng { 25245d4f98a2SYan Zheng struct backref_node *next; 25255d4f98a2SYan Zheng struct btrfs_root *root; 25263fd0a558SYan, Zheng int index = 0; 25273fd0a558SYan, Zheng 25285d4f98a2SYan Zheng next = node; 25295d4f98a2SYan Zheng while (1) { 25305d4f98a2SYan Zheng cond_resched(); 25315d4f98a2SYan Zheng next = walk_up_backref(next, edges, &index); 25325d4f98a2SYan Zheng root = next->root; 25333fd0a558SYan, Zheng BUG_ON(!root); 253427cdeb70SMiao Xie BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state)); 25355d4f98a2SYan Zheng 25365d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 25375d4f98a2SYan Zheng record_reloc_root_in_trans(trans, root); 25385d4f98a2SYan Zheng break; 25395d4f98a2SYan Zheng } 25405d4f98a2SYan Zheng 25415d4f98a2SYan Zheng btrfs_record_root_in_trans(trans, root); 25423fd0a558SYan, Zheng root = root->reloc_root; 25433fd0a558SYan, Zheng 25443fd0a558SYan, Zheng if (next->new_bytenr != root->node->start) { 25453fd0a558SYan, Zheng BUG_ON(next->new_bytenr); 25463fd0a558SYan, Zheng BUG_ON(!list_empty(&next->list)); 25473fd0a558SYan, Zheng next->new_bytenr = root->node->start; 25483fd0a558SYan, Zheng next->root = root; 25493fd0a558SYan, Zheng list_add_tail(&next->list, 25503fd0a558SYan, Zheng &rc->backref_cache.changed); 25513fd0a558SYan, Zheng __mark_block_processed(rc, next); 25525d4f98a2SYan Zheng break; 25535d4f98a2SYan Zheng } 25545d4f98a2SYan Zheng 25553fd0a558SYan, Zheng WARN_ON(1); 25565d4f98a2SYan Zheng root = NULL; 25575d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 25585d4f98a2SYan Zheng if (!next || next->level <= node->level) 25595d4f98a2SYan Zheng break; 25605d4f98a2SYan Zheng } 25613fd0a558SYan, Zheng if (!root) 25623fd0a558SYan, Zheng return NULL; 25635d4f98a2SYan Zheng 25643fd0a558SYan, Zheng next = node; 25653fd0a558SYan, Zheng /* setup backref node path for btrfs_reloc_cow_block */ 25663fd0a558SYan, Zheng while (1) { 25673fd0a558SYan, Zheng rc->backref_cache.path[next->level] = next; 25683fd0a558SYan, Zheng if (--index < 0) 25693fd0a558SYan, Zheng break; 25703fd0a558SYan, Zheng next = edges[index]->node[UPPER]; 25713fd0a558SYan, Zheng } 25725d4f98a2SYan Zheng return root; 25735d4f98a2SYan Zheng } 25745d4f98a2SYan Zheng 25753fd0a558SYan, Zheng /* 25763fd0a558SYan, Zheng * select a tree root for relocation. return NULL if the block 25773fd0a558SYan, Zheng * is reference counted. we should use do_relocation() in this 25783fd0a558SYan, Zheng * case. return a tree root pointer if the block isn't reference 25793fd0a558SYan, Zheng * counted. return -ENOENT if the block is root of reloc tree. 25803fd0a558SYan, Zheng */ 25815d4f98a2SYan Zheng static noinline_for_stack 2582147d256eSZhaolei struct btrfs_root *select_one_root(struct backref_node *node) 25835d4f98a2SYan Zheng { 25843fd0a558SYan, Zheng struct backref_node *next; 25853fd0a558SYan, Zheng struct btrfs_root *root; 25863fd0a558SYan, Zheng struct btrfs_root *fs_root = NULL; 25875d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 25883fd0a558SYan, Zheng int index = 0; 25893fd0a558SYan, Zheng 25903fd0a558SYan, Zheng next = node; 25913fd0a558SYan, Zheng while (1) { 25923fd0a558SYan, Zheng cond_resched(); 25933fd0a558SYan, Zheng next = walk_up_backref(next, edges, &index); 25943fd0a558SYan, Zheng root = next->root; 25953fd0a558SYan, Zheng BUG_ON(!root); 25963fd0a558SYan, Zheng 259725985edcSLucas De Marchi /* no other choice for non-references counted tree */ 259827cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 25993fd0a558SYan, Zheng return root; 26003fd0a558SYan, Zheng 26013fd0a558SYan, Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) 26023fd0a558SYan, Zheng fs_root = root; 26033fd0a558SYan, Zheng 26043fd0a558SYan, Zheng if (next != node) 26053fd0a558SYan, Zheng return NULL; 26063fd0a558SYan, Zheng 26073fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 26083fd0a558SYan, Zheng if (!next || next->level <= node->level) 26093fd0a558SYan, Zheng break; 26103fd0a558SYan, Zheng } 26113fd0a558SYan, Zheng 26123fd0a558SYan, Zheng if (!fs_root) 26133fd0a558SYan, Zheng return ERR_PTR(-ENOENT); 26143fd0a558SYan, Zheng return fs_root; 26155d4f98a2SYan Zheng } 26165d4f98a2SYan Zheng 26175d4f98a2SYan Zheng static noinline_for_stack 26183fd0a558SYan, Zheng u64 calcu_metadata_size(struct reloc_control *rc, 26193fd0a558SYan, Zheng struct backref_node *node, int reserve) 26205d4f98a2SYan Zheng { 26210b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 26223fd0a558SYan, Zheng struct backref_node *next = node; 26233fd0a558SYan, Zheng struct backref_edge *edge; 26243fd0a558SYan, Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 26253fd0a558SYan, Zheng u64 num_bytes = 0; 26263fd0a558SYan, Zheng int index = 0; 26275d4f98a2SYan Zheng 26283fd0a558SYan, Zheng BUG_ON(reserve && node->processed); 26293fd0a558SYan, Zheng 26303fd0a558SYan, Zheng while (next) { 26313fd0a558SYan, Zheng cond_resched(); 26325d4f98a2SYan Zheng while (1) { 26333fd0a558SYan, Zheng if (next->processed && (reserve || next != node)) 26345d4f98a2SYan Zheng break; 26355d4f98a2SYan Zheng 26360b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 26373fd0a558SYan, Zheng 26383fd0a558SYan, Zheng if (list_empty(&next->upper)) 26393fd0a558SYan, Zheng break; 26403fd0a558SYan, Zheng 26413fd0a558SYan, Zheng edge = list_entry(next->upper.next, 26423fd0a558SYan, Zheng struct backref_edge, list[LOWER]); 26433fd0a558SYan, Zheng edges[index++] = edge; 26443fd0a558SYan, Zheng next = edge->node[UPPER]; 26455d4f98a2SYan Zheng } 26463fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 26473fd0a558SYan, Zheng } 26483fd0a558SYan, Zheng return num_bytes; 26493fd0a558SYan, Zheng } 26503fd0a558SYan, Zheng 26513fd0a558SYan, Zheng static int reserve_metadata_space(struct btrfs_trans_handle *trans, 26523fd0a558SYan, Zheng struct reloc_control *rc, 26533fd0a558SYan, Zheng struct backref_node *node) 26543fd0a558SYan, Zheng { 26553fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 2656da17066cSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 26573fd0a558SYan, Zheng u64 num_bytes; 26583fd0a558SYan, Zheng int ret; 26590647bf56SWang Shilong u64 tmp; 26603fd0a558SYan, Zheng 26613fd0a558SYan, Zheng num_bytes = calcu_metadata_size(rc, node, 1) * 2; 26623fd0a558SYan, Zheng 26633fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 26640647bf56SWang Shilong rc->reserved_bytes += num_bytes; 26658ca17f0fSJosef Bacik 26668ca17f0fSJosef Bacik /* 26678ca17f0fSJosef Bacik * We are under a transaction here so we can only do limited flushing. 26688ca17f0fSJosef Bacik * If we get an enospc just kick back -EAGAIN so we know to drop the 26698ca17f0fSJosef Bacik * transaction and try to refill when we can flush all the things. 26708ca17f0fSJosef Bacik */ 26710647bf56SWang Shilong ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes, 26728ca17f0fSJosef Bacik BTRFS_RESERVE_FLUSH_LIMIT); 26733fd0a558SYan, Zheng if (ret) { 2674da17066cSJeff Mahoney tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES; 26750647bf56SWang Shilong while (tmp <= rc->reserved_bytes) 26760647bf56SWang Shilong tmp <<= 1; 26770647bf56SWang Shilong /* 26780647bf56SWang Shilong * only one thread can access block_rsv at this point, 26790647bf56SWang Shilong * so we don't need hold lock to protect block_rsv. 26800647bf56SWang Shilong * we expand more reservation size here to allow enough 268152042d8eSAndrea Gelmini * space for relocation and we will return earlier in 26820647bf56SWang Shilong * enospc case. 26830647bf56SWang Shilong */ 2684da17066cSJeff Mahoney rc->block_rsv->size = tmp + fs_info->nodesize * 26850647bf56SWang Shilong RELOCATION_RESERVED_NODES; 26868ca17f0fSJosef Bacik return -EAGAIN; 26873fd0a558SYan, Zheng } 26883fd0a558SYan, Zheng 26893fd0a558SYan, Zheng return 0; 26903fd0a558SYan, Zheng } 26913fd0a558SYan, Zheng 26925d4f98a2SYan Zheng /* 26935d4f98a2SYan Zheng * relocate a block tree, and then update pointers in upper level 26945d4f98a2SYan Zheng * blocks that reference the block to point to the new location. 26955d4f98a2SYan Zheng * 26965d4f98a2SYan Zheng * if called by link_to_upper, the block has already been relocated. 26975d4f98a2SYan Zheng * in that case this function just updates pointers. 26985d4f98a2SYan Zheng */ 26995d4f98a2SYan Zheng static int do_relocation(struct btrfs_trans_handle *trans, 27003fd0a558SYan, Zheng struct reloc_control *rc, 27015d4f98a2SYan Zheng struct backref_node *node, 27025d4f98a2SYan Zheng struct btrfs_key *key, 27035d4f98a2SYan Zheng struct btrfs_path *path, int lowest) 27045d4f98a2SYan Zheng { 27052ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 27065d4f98a2SYan Zheng struct backref_node *upper; 27075d4f98a2SYan Zheng struct backref_edge *edge; 27085d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 27095d4f98a2SYan Zheng struct btrfs_root *root; 27105d4f98a2SYan Zheng struct extent_buffer *eb; 27115d4f98a2SYan Zheng u32 blocksize; 27125d4f98a2SYan Zheng u64 bytenr; 27135d4f98a2SYan Zheng u64 generation; 27145d4f98a2SYan Zheng int slot; 27155d4f98a2SYan Zheng int ret; 27165d4f98a2SYan Zheng int err = 0; 27175d4f98a2SYan Zheng 27185d4f98a2SYan Zheng BUG_ON(lowest && node->eb); 27195d4f98a2SYan Zheng 27205d4f98a2SYan Zheng path->lowest_level = node->level + 1; 27213fd0a558SYan, Zheng rc->backref_cache.path[node->level] = node; 27225d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) { 2723581c1760SQu Wenruo struct btrfs_key first_key; 2724581c1760SQu Wenruo 27255d4f98a2SYan Zheng cond_resched(); 27265d4f98a2SYan Zheng 27275d4f98a2SYan Zheng upper = edge->node[UPPER]; 2728dc4103f9SWang Shilong root = select_reloc_root(trans, rc, upper, edges); 27293fd0a558SYan, Zheng BUG_ON(!root); 27305d4f98a2SYan Zheng 27313fd0a558SYan, Zheng if (upper->eb && !upper->locked) { 27323fd0a558SYan, Zheng if (!lowest) { 27333fd0a558SYan, Zheng ret = btrfs_bin_search(upper->eb, key, 27343fd0a558SYan, Zheng upper->level, &slot); 2735cbca7d59SFilipe Manana if (ret < 0) { 2736cbca7d59SFilipe Manana err = ret; 2737cbca7d59SFilipe Manana goto next; 2738cbca7d59SFilipe Manana } 27393fd0a558SYan, Zheng BUG_ON(ret); 27403fd0a558SYan, Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 27413fd0a558SYan, Zheng if (node->eb->start == bytenr) 27423fd0a558SYan, Zheng goto next; 27433fd0a558SYan, Zheng } 27445d4f98a2SYan Zheng drop_node_buffer(upper); 27453fd0a558SYan, Zheng } 27465d4f98a2SYan Zheng 27475d4f98a2SYan Zheng if (!upper->eb) { 27485d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 27493561b9dbSLiu Bo if (ret) { 27503561b9dbSLiu Bo if (ret < 0) 27515d4f98a2SYan Zheng err = ret; 27523561b9dbSLiu Bo else 27533561b9dbSLiu Bo err = -ENOENT; 27543561b9dbSLiu Bo 27553561b9dbSLiu Bo btrfs_release_path(path); 27565d4f98a2SYan Zheng break; 27575d4f98a2SYan Zheng } 27585d4f98a2SYan Zheng 27593fd0a558SYan, Zheng if (!upper->eb) { 27603fd0a558SYan, Zheng upper->eb = path->nodes[upper->level]; 27613fd0a558SYan, Zheng path->nodes[upper->level] = NULL; 27623fd0a558SYan, Zheng } else { 27633fd0a558SYan, Zheng BUG_ON(upper->eb != path->nodes[upper->level]); 27643fd0a558SYan, Zheng } 27653fd0a558SYan, Zheng 27663fd0a558SYan, Zheng upper->locked = 1; 27673fd0a558SYan, Zheng path->locks[upper->level] = 0; 27683fd0a558SYan, Zheng 27695d4f98a2SYan Zheng slot = path->slots[upper->level]; 2770b3b4aa74SDavid Sterba btrfs_release_path(path); 27715d4f98a2SYan Zheng } else { 27725d4f98a2SYan Zheng ret = btrfs_bin_search(upper->eb, key, upper->level, 27735d4f98a2SYan Zheng &slot); 2774cbca7d59SFilipe Manana if (ret < 0) { 2775cbca7d59SFilipe Manana err = ret; 2776cbca7d59SFilipe Manana goto next; 2777cbca7d59SFilipe Manana } 27785d4f98a2SYan Zheng BUG_ON(ret); 27795d4f98a2SYan Zheng } 27805d4f98a2SYan Zheng 27815d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 27823fd0a558SYan, Zheng if (lowest) { 27834547f4d8SLiu Bo if (bytenr != node->bytenr) { 27844547f4d8SLiu Bo btrfs_err(root->fs_info, 27854547f4d8SLiu Bo "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu", 27864547f4d8SLiu Bo bytenr, node->bytenr, slot, 27874547f4d8SLiu Bo upper->eb->start); 27884547f4d8SLiu Bo err = -EIO; 27894547f4d8SLiu Bo goto next; 27904547f4d8SLiu Bo } 27915d4f98a2SYan Zheng } else { 27923fd0a558SYan, Zheng if (node->eb->start == bytenr) 27933fd0a558SYan, Zheng goto next; 27945d4f98a2SYan Zheng } 27955d4f98a2SYan Zheng 2796da17066cSJeff Mahoney blocksize = root->fs_info->nodesize; 27975d4f98a2SYan Zheng generation = btrfs_node_ptr_generation(upper->eb, slot); 2798581c1760SQu Wenruo btrfs_node_key_to_cpu(upper->eb, &first_key, slot); 2799581c1760SQu Wenruo eb = read_tree_block(fs_info, bytenr, generation, 2800581c1760SQu Wenruo upper->level - 1, &first_key); 280164c043deSLiu Bo if (IS_ERR(eb)) { 280264c043deSLiu Bo err = PTR_ERR(eb); 280364c043deSLiu Bo goto next; 280464c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2805416bc658SJosef Bacik free_extent_buffer(eb); 280697d9a8a4STsutomu Itoh err = -EIO; 280797d9a8a4STsutomu Itoh goto next; 280897d9a8a4STsutomu Itoh } 28095d4f98a2SYan Zheng btrfs_tree_lock(eb); 28108bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 28115d4f98a2SYan Zheng 28125d4f98a2SYan Zheng if (!node->eb) { 28135d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, eb, upper->eb, 28145d4f98a2SYan Zheng slot, &eb); 28153fd0a558SYan, Zheng btrfs_tree_unlock(eb); 28163fd0a558SYan, Zheng free_extent_buffer(eb); 28175d4f98a2SYan Zheng if (ret < 0) { 28185d4f98a2SYan Zheng err = ret; 28193fd0a558SYan, Zheng goto next; 28205d4f98a2SYan Zheng } 28213fd0a558SYan, Zheng BUG_ON(node->eb != eb); 28225d4f98a2SYan Zheng } else { 28235d4f98a2SYan Zheng btrfs_set_node_blockptr(upper->eb, slot, 28245d4f98a2SYan Zheng node->eb->start); 28255d4f98a2SYan Zheng btrfs_set_node_ptr_generation(upper->eb, slot, 28265d4f98a2SYan Zheng trans->transid); 28275d4f98a2SYan Zheng btrfs_mark_buffer_dirty(upper->eb); 28285d4f98a2SYan Zheng 282984f7d8e6SJosef Bacik ret = btrfs_inc_extent_ref(trans, root, 28305d4f98a2SYan Zheng node->eb->start, blocksize, 28315d4f98a2SYan Zheng upper->eb->start, 28325d4f98a2SYan Zheng btrfs_header_owner(upper->eb), 2833b06c4bf5SFilipe Manana node->level, 0); 28345d4f98a2SYan Zheng BUG_ON(ret); 28355d4f98a2SYan Zheng 28365d4f98a2SYan Zheng ret = btrfs_drop_subtree(trans, root, eb, upper->eb); 28375d4f98a2SYan Zheng BUG_ON(ret); 28385d4f98a2SYan Zheng } 28393fd0a558SYan, Zheng next: 28403fd0a558SYan, Zheng if (!upper->pending) 28413fd0a558SYan, Zheng drop_node_buffer(upper); 28423fd0a558SYan, Zheng else 28433fd0a558SYan, Zheng unlock_node_buffer(upper); 28443fd0a558SYan, Zheng if (err) 28453fd0a558SYan, Zheng break; 28465d4f98a2SYan Zheng } 28473fd0a558SYan, Zheng 28483fd0a558SYan, Zheng if (!err && node->pending) { 28493fd0a558SYan, Zheng drop_node_buffer(node); 28503fd0a558SYan, Zheng list_move_tail(&node->list, &rc->backref_cache.changed); 28513fd0a558SYan, Zheng node->pending = 0; 28525d4f98a2SYan Zheng } 28533fd0a558SYan, Zheng 28545d4f98a2SYan Zheng path->lowest_level = 0; 28553fd0a558SYan, Zheng BUG_ON(err == -ENOSPC); 28565d4f98a2SYan Zheng return err; 28575d4f98a2SYan Zheng } 28585d4f98a2SYan Zheng 28595d4f98a2SYan Zheng static int link_to_upper(struct btrfs_trans_handle *trans, 28603fd0a558SYan, Zheng struct reloc_control *rc, 28615d4f98a2SYan Zheng struct backref_node *node, 28625d4f98a2SYan Zheng struct btrfs_path *path) 28635d4f98a2SYan Zheng { 28645d4f98a2SYan Zheng struct btrfs_key key; 28655d4f98a2SYan Zheng 28665d4f98a2SYan Zheng btrfs_node_key_to_cpu(node->eb, &key, 0); 28673fd0a558SYan, Zheng return do_relocation(trans, rc, node, &key, path, 0); 28685d4f98a2SYan Zheng } 28695d4f98a2SYan Zheng 28705d4f98a2SYan Zheng static int finish_pending_nodes(struct btrfs_trans_handle *trans, 28713fd0a558SYan, Zheng struct reloc_control *rc, 28723fd0a558SYan, Zheng struct btrfs_path *path, int err) 28735d4f98a2SYan Zheng { 28743fd0a558SYan, Zheng LIST_HEAD(list); 28753fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 28765d4f98a2SYan Zheng struct backref_node *node; 28775d4f98a2SYan Zheng int level; 28785d4f98a2SYan Zheng int ret; 28795d4f98a2SYan Zheng 28805d4f98a2SYan Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 28815d4f98a2SYan Zheng while (!list_empty(&cache->pending[level])) { 28825d4f98a2SYan Zheng node = list_entry(cache->pending[level].next, 28833fd0a558SYan, Zheng struct backref_node, list); 28843fd0a558SYan, Zheng list_move_tail(&node->list, &list); 28853fd0a558SYan, Zheng BUG_ON(!node->pending); 28865d4f98a2SYan Zheng 28873fd0a558SYan, Zheng if (!err) { 28883fd0a558SYan, Zheng ret = link_to_upper(trans, rc, node, path); 28895d4f98a2SYan Zheng if (ret < 0) 28905d4f98a2SYan Zheng err = ret; 28915d4f98a2SYan Zheng } 28925d4f98a2SYan Zheng } 28933fd0a558SYan, Zheng list_splice_init(&list, &cache->pending[level]); 28943fd0a558SYan, Zheng } 28955d4f98a2SYan Zheng return err; 28965d4f98a2SYan Zheng } 28975d4f98a2SYan Zheng 28985d4f98a2SYan Zheng static void mark_block_processed(struct reloc_control *rc, 28993fd0a558SYan, Zheng u64 bytenr, u32 blocksize) 29003fd0a558SYan, Zheng { 29013fd0a558SYan, Zheng set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1, 2902ceeb0ae7SDavid Sterba EXTENT_DIRTY); 29033fd0a558SYan, Zheng } 29043fd0a558SYan, Zheng 29053fd0a558SYan, Zheng static void __mark_block_processed(struct reloc_control *rc, 29065d4f98a2SYan Zheng struct backref_node *node) 29075d4f98a2SYan Zheng { 29085d4f98a2SYan Zheng u32 blocksize; 29095d4f98a2SYan Zheng if (node->level == 0 || 29105d4f98a2SYan Zheng in_block_group(node->bytenr, rc->block_group)) { 2911da17066cSJeff Mahoney blocksize = rc->extent_root->fs_info->nodesize; 29123fd0a558SYan, Zheng mark_block_processed(rc, node->bytenr, blocksize); 29135d4f98a2SYan Zheng } 29145d4f98a2SYan Zheng node->processed = 1; 29155d4f98a2SYan Zheng } 29165d4f98a2SYan Zheng 29175d4f98a2SYan Zheng /* 29185d4f98a2SYan Zheng * mark a block and all blocks directly/indirectly reference the block 29195d4f98a2SYan Zheng * as processed. 29205d4f98a2SYan Zheng */ 29215d4f98a2SYan Zheng static void update_processed_blocks(struct reloc_control *rc, 29225d4f98a2SYan Zheng struct backref_node *node) 29235d4f98a2SYan Zheng { 29245d4f98a2SYan Zheng struct backref_node *next = node; 29255d4f98a2SYan Zheng struct backref_edge *edge; 29265d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 29275d4f98a2SYan Zheng int index = 0; 29285d4f98a2SYan Zheng 29295d4f98a2SYan Zheng while (next) { 29305d4f98a2SYan Zheng cond_resched(); 29315d4f98a2SYan Zheng while (1) { 29325d4f98a2SYan Zheng if (next->processed) 29335d4f98a2SYan Zheng break; 29345d4f98a2SYan Zheng 29353fd0a558SYan, Zheng __mark_block_processed(rc, next); 29365d4f98a2SYan Zheng 29375d4f98a2SYan Zheng if (list_empty(&next->upper)) 29385d4f98a2SYan Zheng break; 29395d4f98a2SYan Zheng 29405d4f98a2SYan Zheng edge = list_entry(next->upper.next, 29415d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 29425d4f98a2SYan Zheng edges[index++] = edge; 29435d4f98a2SYan Zheng next = edge->node[UPPER]; 29445d4f98a2SYan Zheng } 29455d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 29465d4f98a2SYan Zheng } 29475d4f98a2SYan Zheng } 29485d4f98a2SYan Zheng 29497476dfdaSDavid Sterba static int tree_block_processed(u64 bytenr, struct reloc_control *rc) 29505d4f98a2SYan Zheng { 2951da17066cSJeff Mahoney u32 blocksize = rc->extent_root->fs_info->nodesize; 29527476dfdaSDavid Sterba 29535d4f98a2SYan Zheng if (test_range_bit(&rc->processed_blocks, bytenr, 29549655d298SChris Mason bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL)) 29555d4f98a2SYan Zheng return 1; 29565d4f98a2SYan Zheng return 0; 29575d4f98a2SYan Zheng } 29585d4f98a2SYan Zheng 29592ff7e61eSJeff Mahoney static int get_tree_block_key(struct btrfs_fs_info *fs_info, 29605d4f98a2SYan Zheng struct tree_block *block) 29615d4f98a2SYan Zheng { 29625d4f98a2SYan Zheng struct extent_buffer *eb; 29635d4f98a2SYan Zheng 29645d4f98a2SYan Zheng BUG_ON(block->key_ready); 2965581c1760SQu Wenruo eb = read_tree_block(fs_info, block->bytenr, block->key.offset, 2966581c1760SQu Wenruo block->level, NULL); 296764c043deSLiu Bo if (IS_ERR(eb)) { 296864c043deSLiu Bo return PTR_ERR(eb); 296964c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2970416bc658SJosef Bacik free_extent_buffer(eb); 2971416bc658SJosef Bacik return -EIO; 2972416bc658SJosef Bacik } 29735d4f98a2SYan Zheng if (block->level == 0) 29745d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &block->key, 0); 29755d4f98a2SYan Zheng else 29765d4f98a2SYan Zheng btrfs_node_key_to_cpu(eb, &block->key, 0); 29775d4f98a2SYan Zheng free_extent_buffer(eb); 29785d4f98a2SYan Zheng block->key_ready = 1; 29795d4f98a2SYan Zheng return 0; 29805d4f98a2SYan Zheng } 29815d4f98a2SYan Zheng 29825d4f98a2SYan Zheng /* 29835d4f98a2SYan Zheng * helper function to relocate a tree block 29845d4f98a2SYan Zheng */ 29855d4f98a2SYan Zheng static int relocate_tree_block(struct btrfs_trans_handle *trans, 29865d4f98a2SYan Zheng struct reloc_control *rc, 29875d4f98a2SYan Zheng struct backref_node *node, 29885d4f98a2SYan Zheng struct btrfs_key *key, 29895d4f98a2SYan Zheng struct btrfs_path *path) 29905d4f98a2SYan Zheng { 29915d4f98a2SYan Zheng struct btrfs_root *root; 29923fd0a558SYan, Zheng int ret = 0; 29935d4f98a2SYan Zheng 29943fd0a558SYan, Zheng if (!node) 29955d4f98a2SYan Zheng return 0; 29963fd0a558SYan, Zheng 29973fd0a558SYan, Zheng BUG_ON(node->processed); 2998147d256eSZhaolei root = select_one_root(node); 29993fd0a558SYan, Zheng if (root == ERR_PTR(-ENOENT)) { 30003fd0a558SYan, Zheng update_processed_blocks(rc, node); 30013fd0a558SYan, Zheng goto out; 30025d4f98a2SYan Zheng } 30035d4f98a2SYan Zheng 300427cdeb70SMiao Xie if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 30053fd0a558SYan, Zheng ret = reserve_metadata_space(trans, rc, node); 30063fd0a558SYan, Zheng if (ret) 30075d4f98a2SYan Zheng goto out; 30085d4f98a2SYan Zheng } 30093fd0a558SYan, Zheng 30103fd0a558SYan, Zheng if (root) { 301127cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 30123fd0a558SYan, Zheng BUG_ON(node->new_bytenr); 30133fd0a558SYan, Zheng BUG_ON(!list_empty(&node->list)); 30143fd0a558SYan, Zheng btrfs_record_root_in_trans(trans, root); 30153fd0a558SYan, Zheng root = root->reloc_root; 30163fd0a558SYan, Zheng node->new_bytenr = root->node->start; 30173fd0a558SYan, Zheng node->root = root; 30183fd0a558SYan, Zheng list_add_tail(&node->list, &rc->backref_cache.changed); 30193fd0a558SYan, Zheng } else { 30205d4f98a2SYan Zheng path->lowest_level = node->level; 30215d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 3022b3b4aa74SDavid Sterba btrfs_release_path(path); 30233fd0a558SYan, Zheng if (ret > 0) 30245d4f98a2SYan Zheng ret = 0; 30253fd0a558SYan, Zheng } 30263fd0a558SYan, Zheng if (!ret) 30273fd0a558SYan, Zheng update_processed_blocks(rc, node); 30283fd0a558SYan, Zheng } else { 30293fd0a558SYan, Zheng ret = do_relocation(trans, rc, node, key, path, 1); 30303fd0a558SYan, Zheng } 30315d4f98a2SYan Zheng out: 30320647bf56SWang Shilong if (ret || node->level == 0 || node->cowonly) 30333fd0a558SYan, Zheng remove_backref_node(&rc->backref_cache, node); 30345d4f98a2SYan Zheng return ret; 30355d4f98a2SYan Zheng } 30365d4f98a2SYan Zheng 30375d4f98a2SYan Zheng /* 30385d4f98a2SYan Zheng * relocate a list of blocks 30395d4f98a2SYan Zheng */ 30405d4f98a2SYan Zheng static noinline_for_stack 30415d4f98a2SYan Zheng int relocate_tree_blocks(struct btrfs_trans_handle *trans, 30425d4f98a2SYan Zheng struct reloc_control *rc, struct rb_root *blocks) 30435d4f98a2SYan Zheng { 30442ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 30455d4f98a2SYan Zheng struct backref_node *node; 30465d4f98a2SYan Zheng struct btrfs_path *path; 30475d4f98a2SYan Zheng struct tree_block *block; 304898ff7b94SQu Wenruo struct tree_block *next; 30495d4f98a2SYan Zheng int ret; 30505d4f98a2SYan Zheng int err = 0; 30515d4f98a2SYan Zheng 30525d4f98a2SYan Zheng path = btrfs_alloc_path(); 3053e1a12670SLiu Bo if (!path) { 3054e1a12670SLiu Bo err = -ENOMEM; 305534c2b290SDavid Sterba goto out_free_blocks; 3056e1a12670SLiu Bo } 30575d4f98a2SYan Zheng 305898ff7b94SQu Wenruo /* Kick in readahead for tree blocks with missing keys */ 305998ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 30605d4f98a2SYan Zheng if (!block->key_ready) 30612ff7e61eSJeff Mahoney readahead_tree_block(fs_info, block->bytenr); 30625d4f98a2SYan Zheng } 30635d4f98a2SYan Zheng 306498ff7b94SQu Wenruo /* Get first keys */ 306598ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 306634c2b290SDavid Sterba if (!block->key_ready) { 30672ff7e61eSJeff Mahoney err = get_tree_block_key(fs_info, block); 306834c2b290SDavid Sterba if (err) 306934c2b290SDavid Sterba goto out_free_path; 307034c2b290SDavid Sterba } 30715d4f98a2SYan Zheng } 30725d4f98a2SYan Zheng 307398ff7b94SQu Wenruo /* Do tree relocation */ 307498ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 30753fd0a558SYan, Zheng node = build_backref_tree(rc, &block->key, 30765d4f98a2SYan Zheng block->level, block->bytenr); 30775d4f98a2SYan Zheng if (IS_ERR(node)) { 30785d4f98a2SYan Zheng err = PTR_ERR(node); 30795d4f98a2SYan Zheng goto out; 30805d4f98a2SYan Zheng } 30815d4f98a2SYan Zheng 30825d4f98a2SYan Zheng ret = relocate_tree_block(trans, rc, node, &block->key, 30835d4f98a2SYan Zheng path); 30845d4f98a2SYan Zheng if (ret < 0) { 308598ff7b94SQu Wenruo if (ret != -EAGAIN || &block->rb_node == rb_first(blocks)) 30865d4f98a2SYan Zheng err = ret; 30875d4f98a2SYan Zheng goto out; 30885d4f98a2SYan Zheng } 30895d4f98a2SYan Zheng } 30905d4f98a2SYan Zheng out: 30913fd0a558SYan, Zheng err = finish_pending_nodes(trans, rc, path, err); 30925d4f98a2SYan Zheng 309334c2b290SDavid Sterba out_free_path: 30945d4f98a2SYan Zheng btrfs_free_path(path); 309534c2b290SDavid Sterba out_free_blocks: 3096e1a12670SLiu Bo free_block_list(blocks); 30975d4f98a2SYan Zheng return err; 30985d4f98a2SYan Zheng } 30995d4f98a2SYan Zheng 31005d4f98a2SYan Zheng static noinline_for_stack 3101efa56464SYan, Zheng int prealloc_file_extent_cluster(struct inode *inode, 3102efa56464SYan, Zheng struct file_extent_cluster *cluster) 3103efa56464SYan, Zheng { 3104efa56464SYan, Zheng u64 alloc_hint = 0; 3105efa56464SYan, Zheng u64 start; 3106efa56464SYan, Zheng u64 end; 3107efa56464SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 3108efa56464SYan, Zheng u64 num_bytes; 3109efa56464SYan, Zheng int nr = 0; 3110efa56464SYan, Zheng int ret = 0; 3111dcb40c19SWang Xiaoguang u64 prealloc_start = cluster->start - offset; 3112dcb40c19SWang Xiaoguang u64 prealloc_end = cluster->end - offset; 311318513091SWang Xiaoguang u64 cur_offset; 3114364ecf36SQu Wenruo struct extent_changeset *data_reserved = NULL; 3115efa56464SYan, Zheng 3116efa56464SYan, Zheng BUG_ON(cluster->start != cluster->boundary[0]); 31175955102cSAl Viro inode_lock(inode); 3118efa56464SYan, Zheng 3119364ecf36SQu Wenruo ret = btrfs_check_data_free_space(inode, &data_reserved, prealloc_start, 3120dcb40c19SWang Xiaoguang prealloc_end + 1 - prealloc_start); 3121efa56464SYan, Zheng if (ret) 3122efa56464SYan, Zheng goto out; 3123efa56464SYan, Zheng 312418513091SWang Xiaoguang cur_offset = prealloc_start; 3125efa56464SYan, Zheng while (nr < cluster->nr) { 3126efa56464SYan, Zheng start = cluster->boundary[nr] - offset; 3127efa56464SYan, Zheng if (nr + 1 < cluster->nr) 3128efa56464SYan, Zheng end = cluster->boundary[nr + 1] - 1 - offset; 3129efa56464SYan, Zheng else 3130efa56464SYan, Zheng end = cluster->end - offset; 3131efa56464SYan, Zheng 3132d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 3133efa56464SYan, Zheng num_bytes = end + 1 - start; 313418513091SWang Xiaoguang if (cur_offset < start) 3135bc42bda2SQu Wenruo btrfs_free_reserved_data_space(inode, data_reserved, 3136bc42bda2SQu Wenruo cur_offset, start - cur_offset); 3137efa56464SYan, Zheng ret = btrfs_prealloc_file_range(inode, 0, start, 3138efa56464SYan, Zheng num_bytes, num_bytes, 3139efa56464SYan, Zheng end + 1, &alloc_hint); 314018513091SWang Xiaoguang cur_offset = end + 1; 3141d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 3142efa56464SYan, Zheng if (ret) 3143efa56464SYan, Zheng break; 3144efa56464SYan, Zheng nr++; 3145efa56464SYan, Zheng } 314618513091SWang Xiaoguang if (cur_offset < prealloc_end) 3147bc42bda2SQu Wenruo btrfs_free_reserved_data_space(inode, data_reserved, 3148bc42bda2SQu Wenruo cur_offset, prealloc_end + 1 - cur_offset); 3149efa56464SYan, Zheng out: 31505955102cSAl Viro inode_unlock(inode); 3151364ecf36SQu Wenruo extent_changeset_free(data_reserved); 3152efa56464SYan, Zheng return ret; 3153efa56464SYan, Zheng } 3154efa56464SYan, Zheng 3155efa56464SYan, Zheng static noinline_for_stack 31560257bb82SYan, Zheng int setup_extent_mapping(struct inode *inode, u64 start, u64 end, 31570257bb82SYan, Zheng u64 block_start) 31585d4f98a2SYan Zheng { 31590b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 31605d4f98a2SYan Zheng struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 31615d4f98a2SYan Zheng struct extent_map *em; 31620257bb82SYan, Zheng int ret = 0; 31635d4f98a2SYan Zheng 3164172ddd60SDavid Sterba em = alloc_extent_map(); 31650257bb82SYan, Zheng if (!em) 31660257bb82SYan, Zheng return -ENOMEM; 31670257bb82SYan, Zheng 31685d4f98a2SYan Zheng em->start = start; 31690257bb82SYan, Zheng em->len = end + 1 - start; 31700257bb82SYan, Zheng em->block_len = em->len; 31710257bb82SYan, Zheng em->block_start = block_start; 31720b246afaSJeff Mahoney em->bdev = fs_info->fs_devices->latest_bdev; 31735d4f98a2SYan Zheng set_bit(EXTENT_FLAG_PINNED, &em->flags); 31745d4f98a2SYan Zheng 3175d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 31765d4f98a2SYan Zheng while (1) { 3177890871beSChris Mason write_lock(&em_tree->lock); 317809a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 3179890871beSChris Mason write_unlock(&em_tree->lock); 31805d4f98a2SYan Zheng if (ret != -EEXIST) { 31815d4f98a2SYan Zheng free_extent_map(em); 31825d4f98a2SYan Zheng break; 31835d4f98a2SYan Zheng } 3184dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0); 31855d4f98a2SYan Zheng } 3186d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 31870257bb82SYan, Zheng return ret; 31880257bb82SYan, Zheng } 31895d4f98a2SYan Zheng 31900257bb82SYan, Zheng static int relocate_file_extent_cluster(struct inode *inode, 31910257bb82SYan, Zheng struct file_extent_cluster *cluster) 31920257bb82SYan, Zheng { 31932ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 31940257bb82SYan, Zheng u64 page_start; 31950257bb82SYan, Zheng u64 page_end; 31960257bb82SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 31970257bb82SYan, Zheng unsigned long index; 31980257bb82SYan, Zheng unsigned long last_index; 31990257bb82SYan, Zheng struct page *page; 32000257bb82SYan, Zheng struct file_ra_state *ra; 32013b16a4e3SJosef Bacik gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 32020257bb82SYan, Zheng int nr = 0; 32030257bb82SYan, Zheng int ret = 0; 32040257bb82SYan, Zheng 32050257bb82SYan, Zheng if (!cluster->nr) 32060257bb82SYan, Zheng return 0; 32070257bb82SYan, Zheng 32080257bb82SYan, Zheng ra = kzalloc(sizeof(*ra), GFP_NOFS); 32090257bb82SYan, Zheng if (!ra) 32100257bb82SYan, Zheng return -ENOMEM; 32110257bb82SYan, Zheng 3212efa56464SYan, Zheng ret = prealloc_file_extent_cluster(inode, cluster); 32130257bb82SYan, Zheng if (ret) 3214efa56464SYan, Zheng goto out; 32150257bb82SYan, Zheng 32160257bb82SYan, Zheng file_ra_state_init(ra, inode->i_mapping); 32170257bb82SYan, Zheng 3218efa56464SYan, Zheng ret = setup_extent_mapping(inode, cluster->start - offset, 3219efa56464SYan, Zheng cluster->end - offset, cluster->start); 3220efa56464SYan, Zheng if (ret) 3221efa56464SYan, Zheng goto out; 3222efa56464SYan, Zheng 322309cbfeafSKirill A. Shutemov index = (cluster->start - offset) >> PAGE_SHIFT; 322409cbfeafSKirill A. Shutemov last_index = (cluster->end - offset) >> PAGE_SHIFT; 32250257bb82SYan, Zheng while (index <= last_index) { 32269f3db423SNikolay Borisov ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), 32279f3db423SNikolay Borisov PAGE_SIZE); 3228efa56464SYan, Zheng if (ret) 3229efa56464SYan, Zheng goto out; 3230efa56464SYan, Zheng 32310257bb82SYan, Zheng page = find_lock_page(inode->i_mapping, index); 32320257bb82SYan, Zheng if (!page) { 32330257bb82SYan, Zheng page_cache_sync_readahead(inode->i_mapping, 32340257bb82SYan, Zheng ra, NULL, index, 32350257bb82SYan, Zheng last_index + 1 - index); 3236a94733d0SJosef Bacik page = find_or_create_page(inode->i_mapping, index, 32373b16a4e3SJosef Bacik mask); 32380257bb82SYan, Zheng if (!page) { 3239691fa059SNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 324043b18595SQu Wenruo PAGE_SIZE, true); 32410257bb82SYan, Zheng ret = -ENOMEM; 3242efa56464SYan, Zheng goto out; 32430257bb82SYan, Zheng } 32440257bb82SYan, Zheng } 32450257bb82SYan, Zheng 32460257bb82SYan, Zheng if (PageReadahead(page)) { 32470257bb82SYan, Zheng page_cache_async_readahead(inode->i_mapping, 32480257bb82SYan, Zheng ra, NULL, page, index, 32490257bb82SYan, Zheng last_index + 1 - index); 32500257bb82SYan, Zheng } 32510257bb82SYan, Zheng 32520257bb82SYan, Zheng if (!PageUptodate(page)) { 32530257bb82SYan, Zheng btrfs_readpage(NULL, page); 32540257bb82SYan, Zheng lock_page(page); 32550257bb82SYan, Zheng if (!PageUptodate(page)) { 32560257bb82SYan, Zheng unlock_page(page); 325709cbfeafSKirill A. Shutemov put_page(page); 3258691fa059SNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 325943b18595SQu Wenruo PAGE_SIZE, true); 32608b62f87bSJosef Bacik btrfs_delalloc_release_extents(BTRFS_I(inode), 326143b18595SQu Wenruo PAGE_SIZE, true); 32620257bb82SYan, Zheng ret = -EIO; 3263efa56464SYan, Zheng goto out; 32640257bb82SYan, Zheng } 32650257bb82SYan, Zheng } 32660257bb82SYan, Zheng 32674eee4fa4SMiao Xie page_start = page_offset(page); 326809cbfeafSKirill A. Shutemov page_end = page_start + PAGE_SIZE - 1; 32690257bb82SYan, Zheng 3270d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end); 32710257bb82SYan, Zheng 32720257bb82SYan, Zheng set_page_extent_mapped(page); 32730257bb82SYan, Zheng 32740257bb82SYan, Zheng if (nr < cluster->nr && 32750257bb82SYan, Zheng page_start + offset == cluster->boundary[nr]) { 32760257bb82SYan, Zheng set_extent_bits(&BTRFS_I(inode)->io_tree, 32770257bb82SYan, Zheng page_start, page_end, 3278ceeb0ae7SDavid Sterba EXTENT_BOUNDARY); 32790257bb82SYan, Zheng nr++; 32800257bb82SYan, Zheng } 32810257bb82SYan, Zheng 3282765f3cebSNikolay Borisov ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0, 3283765f3cebSNikolay Borisov NULL, 0); 3284765f3cebSNikolay Borisov if (ret) { 3285765f3cebSNikolay Borisov unlock_page(page); 3286765f3cebSNikolay Borisov put_page(page); 3287765f3cebSNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 328843b18595SQu Wenruo PAGE_SIZE, true); 3289765f3cebSNikolay Borisov btrfs_delalloc_release_extents(BTRFS_I(inode), 329043b18595SQu Wenruo PAGE_SIZE, true); 3291765f3cebSNikolay Borisov 3292765f3cebSNikolay Borisov clear_extent_bits(&BTRFS_I(inode)->io_tree, 3293765f3cebSNikolay Borisov page_start, page_end, 3294765f3cebSNikolay Borisov EXTENT_LOCKED | EXTENT_BOUNDARY); 3295765f3cebSNikolay Borisov goto out; 3296765f3cebSNikolay Borisov 3297765f3cebSNikolay Borisov } 32980257bb82SYan, Zheng set_page_dirty(page); 32990257bb82SYan, Zheng 33000257bb82SYan, Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 3301d0082371SJeff Mahoney page_start, page_end); 33020257bb82SYan, Zheng unlock_page(page); 330309cbfeafSKirill A. Shutemov put_page(page); 33040257bb82SYan, Zheng 33050257bb82SYan, Zheng index++; 330643b18595SQu Wenruo btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, 330743b18595SQu Wenruo false); 3308efa56464SYan, Zheng balance_dirty_pages_ratelimited(inode->i_mapping); 33092ff7e61eSJeff Mahoney btrfs_throttle(fs_info); 33100257bb82SYan, Zheng } 33110257bb82SYan, Zheng WARN_ON(nr != cluster->nr); 3312efa56464SYan, Zheng out: 33130257bb82SYan, Zheng kfree(ra); 33140257bb82SYan, Zheng return ret; 33150257bb82SYan, Zheng } 33160257bb82SYan, Zheng 33170257bb82SYan, Zheng static noinline_for_stack 33180257bb82SYan, Zheng int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key, 33190257bb82SYan, Zheng struct file_extent_cluster *cluster) 33200257bb82SYan, Zheng { 33210257bb82SYan, Zheng int ret; 33220257bb82SYan, Zheng 33230257bb82SYan, Zheng if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) { 33240257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 33250257bb82SYan, Zheng if (ret) 33260257bb82SYan, Zheng return ret; 33270257bb82SYan, Zheng cluster->nr = 0; 33280257bb82SYan, Zheng } 33290257bb82SYan, Zheng 33300257bb82SYan, Zheng if (!cluster->nr) 33310257bb82SYan, Zheng cluster->start = extent_key->objectid; 33320257bb82SYan, Zheng else 33330257bb82SYan, Zheng BUG_ON(cluster->nr >= MAX_EXTENTS); 33340257bb82SYan, Zheng cluster->end = extent_key->objectid + extent_key->offset - 1; 33350257bb82SYan, Zheng cluster->boundary[cluster->nr] = extent_key->objectid; 33360257bb82SYan, Zheng cluster->nr++; 33370257bb82SYan, Zheng 33380257bb82SYan, Zheng if (cluster->nr >= MAX_EXTENTS) { 33390257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 33400257bb82SYan, Zheng if (ret) 33410257bb82SYan, Zheng return ret; 33420257bb82SYan, Zheng cluster->nr = 0; 33430257bb82SYan, Zheng } 33440257bb82SYan, Zheng return 0; 33455d4f98a2SYan Zheng } 33465d4f98a2SYan Zheng 33475d4f98a2SYan Zheng /* 33485d4f98a2SYan Zheng * helper to add a tree block to the list. 33495d4f98a2SYan Zheng * the major work is getting the generation and level of the block 33505d4f98a2SYan Zheng */ 33515d4f98a2SYan Zheng static int add_tree_block(struct reloc_control *rc, 33525d4f98a2SYan Zheng struct btrfs_key *extent_key, 33535d4f98a2SYan Zheng struct btrfs_path *path, 33545d4f98a2SYan Zheng struct rb_root *blocks) 33555d4f98a2SYan Zheng { 33565d4f98a2SYan Zheng struct extent_buffer *eb; 33575d4f98a2SYan Zheng struct btrfs_extent_item *ei; 33585d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 33595d4f98a2SYan Zheng struct tree_block *block; 33605d4f98a2SYan Zheng struct rb_node *rb_node; 33615d4f98a2SYan Zheng u32 item_size; 33625d4f98a2SYan Zheng int level = -1; 33637fdf4b60SWang Shilong u64 generation; 33645d4f98a2SYan Zheng 33655d4f98a2SYan Zheng eb = path->nodes[0]; 33665d4f98a2SYan Zheng item_size = btrfs_item_size_nr(eb, path->slots[0]); 33675d4f98a2SYan Zheng 33683173a18fSJosef Bacik if (extent_key->type == BTRFS_METADATA_ITEM_KEY || 33693173a18fSJosef Bacik item_size >= sizeof(*ei) + sizeof(*bi)) { 33705d4f98a2SYan Zheng ei = btrfs_item_ptr(eb, path->slots[0], 33715d4f98a2SYan Zheng struct btrfs_extent_item); 33723173a18fSJosef Bacik if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) { 33735d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 33745d4f98a2SYan Zheng level = btrfs_tree_block_level(eb, bi); 33755d4f98a2SYan Zheng } else { 33763173a18fSJosef Bacik level = (int)extent_key->offset; 33773173a18fSJosef Bacik } 33783173a18fSJosef Bacik generation = btrfs_extent_generation(eb, ei); 33796d8ff4e4SDavid Sterba } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) { 3380ba3c2b19SNikolay Borisov btrfs_print_v0_err(eb->fs_info); 3381ba3c2b19SNikolay Borisov btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL); 3382ba3c2b19SNikolay Borisov return -EINVAL; 33833173a18fSJosef Bacik } else { 33845d4f98a2SYan Zheng BUG(); 33855d4f98a2SYan Zheng } 33865d4f98a2SYan Zheng 3387b3b4aa74SDavid Sterba btrfs_release_path(path); 33885d4f98a2SYan Zheng 33895d4f98a2SYan Zheng BUG_ON(level == -1); 33905d4f98a2SYan Zheng 33915d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 33925d4f98a2SYan Zheng if (!block) 33935d4f98a2SYan Zheng return -ENOMEM; 33945d4f98a2SYan Zheng 33955d4f98a2SYan Zheng block->bytenr = extent_key->objectid; 3396da17066cSJeff Mahoney block->key.objectid = rc->extent_root->fs_info->nodesize; 33975d4f98a2SYan Zheng block->key.offset = generation; 33985d4f98a2SYan Zheng block->level = level; 33995d4f98a2SYan Zheng block->key_ready = 0; 34005d4f98a2SYan Zheng 34015d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, &block->rb_node); 340243c04fb1SJeff Mahoney if (rb_node) 340343c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, block->bytenr); 34045d4f98a2SYan Zheng 34055d4f98a2SYan Zheng return 0; 34065d4f98a2SYan Zheng } 34075d4f98a2SYan Zheng 34085d4f98a2SYan Zheng /* 34095d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY 34105d4f98a2SYan Zheng */ 34115d4f98a2SYan Zheng static int __add_tree_block(struct reloc_control *rc, 34125d4f98a2SYan Zheng u64 bytenr, u32 blocksize, 34135d4f98a2SYan Zheng struct rb_root *blocks) 34145d4f98a2SYan Zheng { 34150b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 34165d4f98a2SYan Zheng struct btrfs_path *path; 34175d4f98a2SYan Zheng struct btrfs_key key; 34185d4f98a2SYan Zheng int ret; 34190b246afaSJeff Mahoney bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA); 34205d4f98a2SYan Zheng 34217476dfdaSDavid Sterba if (tree_block_processed(bytenr, rc)) 34225d4f98a2SYan Zheng return 0; 34235d4f98a2SYan Zheng 34245d4f98a2SYan Zheng if (tree_search(blocks, bytenr)) 34255d4f98a2SYan Zheng return 0; 34265d4f98a2SYan Zheng 34275d4f98a2SYan Zheng path = btrfs_alloc_path(); 34285d4f98a2SYan Zheng if (!path) 34295d4f98a2SYan Zheng return -ENOMEM; 3430aee68ee5SJosef Bacik again: 34315d4f98a2SYan Zheng key.objectid = bytenr; 3432aee68ee5SJosef Bacik if (skinny) { 3433aee68ee5SJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 3434aee68ee5SJosef Bacik key.offset = (u64)-1; 3435aee68ee5SJosef Bacik } else { 34365d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 34375d4f98a2SYan Zheng key.offset = blocksize; 3438aee68ee5SJosef Bacik } 34395d4f98a2SYan Zheng 34405d4f98a2SYan Zheng path->search_commit_root = 1; 34415d4f98a2SYan Zheng path->skip_locking = 1; 34425d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0); 34435d4f98a2SYan Zheng if (ret < 0) 34445d4f98a2SYan Zheng goto out; 34455d4f98a2SYan Zheng 3446aee68ee5SJosef Bacik if (ret > 0 && skinny) { 3447aee68ee5SJosef Bacik if (path->slots[0]) { 3448aee68ee5SJosef Bacik path->slots[0]--; 3449aee68ee5SJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, 3450aee68ee5SJosef Bacik path->slots[0]); 34513173a18fSJosef Bacik if (key.objectid == bytenr && 3452aee68ee5SJosef Bacik (key.type == BTRFS_METADATA_ITEM_KEY || 3453aee68ee5SJosef Bacik (key.type == BTRFS_EXTENT_ITEM_KEY && 3454aee68ee5SJosef Bacik key.offset == blocksize))) 34553173a18fSJosef Bacik ret = 0; 34563173a18fSJosef Bacik } 3457aee68ee5SJosef Bacik 3458aee68ee5SJosef Bacik if (ret) { 3459aee68ee5SJosef Bacik skinny = false; 3460aee68ee5SJosef Bacik btrfs_release_path(path); 3461aee68ee5SJosef Bacik goto again; 3462aee68ee5SJosef Bacik } 3463aee68ee5SJosef Bacik } 3464cdccee99SLiu Bo if (ret) { 3465cdccee99SLiu Bo ASSERT(ret == 1); 3466cdccee99SLiu Bo btrfs_print_leaf(path->nodes[0]); 3467cdccee99SLiu Bo btrfs_err(fs_info, 3468cdccee99SLiu Bo "tree block extent item (%llu) is not found in extent tree", 3469cdccee99SLiu Bo bytenr); 3470cdccee99SLiu Bo WARN_ON(1); 3471cdccee99SLiu Bo ret = -EINVAL; 3472cdccee99SLiu Bo goto out; 3473cdccee99SLiu Bo } 34743173a18fSJosef Bacik 34755d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, blocks); 34765d4f98a2SYan Zheng out: 34775d4f98a2SYan Zheng btrfs_free_path(path); 34785d4f98a2SYan Zheng return ret; 34795d4f98a2SYan Zheng } 34805d4f98a2SYan Zheng 34815d4f98a2SYan Zheng /* 34825d4f98a2SYan Zheng * helper to check if the block use full backrefs for pointers in it 34835d4f98a2SYan Zheng */ 34845d4f98a2SYan Zheng static int block_use_full_backref(struct reloc_control *rc, 34855d4f98a2SYan Zheng struct extent_buffer *eb) 34865d4f98a2SYan Zheng { 34875d4f98a2SYan Zheng u64 flags; 34885d4f98a2SYan Zheng int ret; 34895d4f98a2SYan Zheng 34905d4f98a2SYan Zheng if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) || 34915d4f98a2SYan Zheng btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV) 34925d4f98a2SYan Zheng return 1; 34935d4f98a2SYan Zheng 34942ff7e61eSJeff Mahoney ret = btrfs_lookup_extent_info(NULL, rc->extent_root->fs_info, 34953173a18fSJosef Bacik eb->start, btrfs_header_level(eb), 1, 34963173a18fSJosef Bacik NULL, &flags); 34975d4f98a2SYan Zheng BUG_ON(ret); 34985d4f98a2SYan Zheng 34995d4f98a2SYan Zheng if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) 35005d4f98a2SYan Zheng ret = 1; 35015d4f98a2SYan Zheng else 35025d4f98a2SYan Zheng ret = 0; 35035d4f98a2SYan Zheng return ret; 35045d4f98a2SYan Zheng } 35055d4f98a2SYan Zheng 35060af3d00bSJosef Bacik static int delete_block_group_cache(struct btrfs_fs_info *fs_info, 35071bbc621eSChris Mason struct btrfs_block_group_cache *block_group, 35081bbc621eSChris Mason struct inode *inode, 35091bbc621eSChris Mason u64 ino) 35100af3d00bSJosef Bacik { 35110af3d00bSJosef Bacik struct btrfs_key key; 35120af3d00bSJosef Bacik struct btrfs_root *root = fs_info->tree_root; 35130af3d00bSJosef Bacik struct btrfs_trans_handle *trans; 35140af3d00bSJosef Bacik int ret = 0; 35150af3d00bSJosef Bacik 35160af3d00bSJosef Bacik if (inode) 35170af3d00bSJosef Bacik goto truncate; 35180af3d00bSJosef Bacik 35190af3d00bSJosef Bacik key.objectid = ino; 35200af3d00bSJosef Bacik key.type = BTRFS_INODE_ITEM_KEY; 35210af3d00bSJosef Bacik key.offset = 0; 35220af3d00bSJosef Bacik 35230af3d00bSJosef Bacik inode = btrfs_iget(fs_info->sb, &key, root, NULL); 35242e19f1f9SAl Viro if (IS_ERR(inode)) 35250af3d00bSJosef Bacik return -ENOENT; 35260af3d00bSJosef Bacik 35270af3d00bSJosef Bacik truncate: 35282ff7e61eSJeff Mahoney ret = btrfs_check_trunc_cache_free_space(fs_info, 35297b61cd92SMiao Xie &fs_info->global_block_rsv); 35307b61cd92SMiao Xie if (ret) 35317b61cd92SMiao Xie goto out; 35327b61cd92SMiao Xie 35337a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 35340af3d00bSJosef Bacik if (IS_ERR(trans)) { 35353612b495STsutomu Itoh ret = PTR_ERR(trans); 35360af3d00bSJosef Bacik goto out; 35370af3d00bSJosef Bacik } 35380af3d00bSJosef Bacik 353977ab86bfSJeff Mahoney ret = btrfs_truncate_free_space_cache(trans, block_group, inode); 35400af3d00bSJosef Bacik 35413a45bb20SJeff Mahoney btrfs_end_transaction(trans); 35422ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 35430af3d00bSJosef Bacik out: 35440af3d00bSJosef Bacik iput(inode); 35450af3d00bSJosef Bacik return ret; 35460af3d00bSJosef Bacik } 35470af3d00bSJosef Bacik 35485d4f98a2SYan Zheng /* 35495d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY 35505d4f98a2SYan Zheng * this function scans fs tree to find blocks reference the data extent 35515d4f98a2SYan Zheng */ 35525d4f98a2SYan Zheng static int find_data_references(struct reloc_control *rc, 35535d4f98a2SYan Zheng struct btrfs_key *extent_key, 35545d4f98a2SYan Zheng struct extent_buffer *leaf, 35555d4f98a2SYan Zheng struct btrfs_extent_data_ref *ref, 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 tree_block *block; 35615d4f98a2SYan Zheng struct btrfs_root *root; 35625d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 35635d4f98a2SYan Zheng struct rb_node *rb_node; 35645d4f98a2SYan Zheng struct btrfs_key key; 35655d4f98a2SYan Zheng u64 ref_root; 35665d4f98a2SYan Zheng u64 ref_objectid; 35675d4f98a2SYan Zheng u64 ref_offset; 35685d4f98a2SYan Zheng u32 ref_count; 35695d4f98a2SYan Zheng u32 nritems; 35705d4f98a2SYan Zheng int err = 0; 35715d4f98a2SYan Zheng int added = 0; 35725d4f98a2SYan Zheng int counted; 35735d4f98a2SYan Zheng int ret; 35745d4f98a2SYan Zheng 35755d4f98a2SYan Zheng ref_root = btrfs_extent_data_ref_root(leaf, ref); 35765d4f98a2SYan Zheng ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref); 35775d4f98a2SYan Zheng ref_offset = btrfs_extent_data_ref_offset(leaf, ref); 35785d4f98a2SYan Zheng ref_count = btrfs_extent_data_ref_count(leaf, ref); 35795d4f98a2SYan Zheng 35800af3d00bSJosef Bacik /* 35810af3d00bSJosef Bacik * This is an extent belonging to the free space cache, lets just delete 35820af3d00bSJosef Bacik * it and redo the search. 35830af3d00bSJosef Bacik */ 35840af3d00bSJosef Bacik if (ref_root == BTRFS_ROOT_TREE_OBJECTID) { 35850b246afaSJeff Mahoney ret = delete_block_group_cache(fs_info, rc->block_group, 35860af3d00bSJosef Bacik NULL, ref_objectid); 35870af3d00bSJosef Bacik if (ret != -ENOENT) 35880af3d00bSJosef Bacik return ret; 35890af3d00bSJosef Bacik ret = 0; 35900af3d00bSJosef Bacik } 35910af3d00bSJosef Bacik 35920af3d00bSJosef Bacik path = btrfs_alloc_path(); 35930af3d00bSJosef Bacik if (!path) 35940af3d00bSJosef Bacik return -ENOMEM; 3595e4058b54SDavid Sterba path->reada = READA_FORWARD; 35960af3d00bSJosef Bacik 35970b246afaSJeff Mahoney root = read_fs_root(fs_info, ref_root); 35985d4f98a2SYan Zheng if (IS_ERR(root)) { 35995d4f98a2SYan Zheng err = PTR_ERR(root); 36005d4f98a2SYan Zheng goto out; 36015d4f98a2SYan Zheng } 36025d4f98a2SYan Zheng 36035d4f98a2SYan Zheng key.objectid = ref_objectid; 36045d4f98a2SYan Zheng key.type = BTRFS_EXTENT_DATA_KEY; 360584850e8dSYan, Zheng if (ref_offset > ((u64)-1 << 32)) 360684850e8dSYan, Zheng key.offset = 0; 360784850e8dSYan, Zheng else 360884850e8dSYan, Zheng key.offset = ref_offset; 36095d4f98a2SYan Zheng 36105d4f98a2SYan Zheng path->search_commit_root = 1; 36115d4f98a2SYan Zheng path->skip_locking = 1; 36125d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 36135d4f98a2SYan Zheng if (ret < 0) { 36145d4f98a2SYan Zheng err = ret; 36155d4f98a2SYan Zheng goto out; 36165d4f98a2SYan Zheng } 36175d4f98a2SYan Zheng 36185d4f98a2SYan Zheng leaf = path->nodes[0]; 36195d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 36205d4f98a2SYan Zheng /* 36215d4f98a2SYan Zheng * the references in tree blocks that use full backrefs 36225d4f98a2SYan Zheng * are not counted in 36235d4f98a2SYan Zheng */ 36245d4f98a2SYan Zheng if (block_use_full_backref(rc, leaf)) 36255d4f98a2SYan Zheng counted = 0; 36265d4f98a2SYan Zheng else 36275d4f98a2SYan Zheng counted = 1; 36285d4f98a2SYan Zheng rb_node = tree_search(blocks, leaf->start); 36295d4f98a2SYan Zheng if (rb_node) { 36305d4f98a2SYan Zheng if (counted) 36315d4f98a2SYan Zheng added = 1; 36325d4f98a2SYan Zheng else 36335d4f98a2SYan Zheng path->slots[0] = nritems; 36345d4f98a2SYan Zheng } 36355d4f98a2SYan Zheng 36365d4f98a2SYan Zheng while (ref_count > 0) { 36375d4f98a2SYan Zheng while (path->slots[0] >= nritems) { 36385d4f98a2SYan Zheng ret = btrfs_next_leaf(root, path); 36395d4f98a2SYan Zheng if (ret < 0) { 36405d4f98a2SYan Zheng err = ret; 36415d4f98a2SYan Zheng goto out; 36425d4f98a2SYan Zheng } 3643fae7f21cSDulshani Gunawardhana if (WARN_ON(ret > 0)) 36445d4f98a2SYan Zheng goto out; 36455d4f98a2SYan Zheng 36465d4f98a2SYan Zheng leaf = path->nodes[0]; 36475d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 36485d4f98a2SYan Zheng added = 0; 36495d4f98a2SYan Zheng 36505d4f98a2SYan Zheng if (block_use_full_backref(rc, leaf)) 36515d4f98a2SYan Zheng counted = 0; 36525d4f98a2SYan Zheng else 36535d4f98a2SYan Zheng counted = 1; 36545d4f98a2SYan Zheng rb_node = tree_search(blocks, leaf->start); 36555d4f98a2SYan Zheng if (rb_node) { 36565d4f98a2SYan Zheng if (counted) 36575d4f98a2SYan Zheng added = 1; 36585d4f98a2SYan Zheng else 36595d4f98a2SYan Zheng path->slots[0] = nritems; 36605d4f98a2SYan Zheng } 36615d4f98a2SYan Zheng } 36625d4f98a2SYan Zheng 36635d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 3664fae7f21cSDulshani Gunawardhana if (WARN_ON(key.objectid != ref_objectid || 3665fae7f21cSDulshani Gunawardhana key.type != BTRFS_EXTENT_DATA_KEY)) 36665d4f98a2SYan Zheng break; 36675d4f98a2SYan Zheng 36685d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 36695d4f98a2SYan Zheng struct btrfs_file_extent_item); 36705d4f98a2SYan Zheng 36715d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 36725d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 36735d4f98a2SYan Zheng goto next; 36745d4f98a2SYan Zheng 36755d4f98a2SYan Zheng if (btrfs_file_extent_disk_bytenr(leaf, fi) != 36765d4f98a2SYan Zheng extent_key->objectid) 36775d4f98a2SYan Zheng goto next; 36785d4f98a2SYan Zheng 36795d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 36805d4f98a2SYan Zheng if (key.offset != ref_offset) 36815d4f98a2SYan Zheng goto next; 36825d4f98a2SYan Zheng 36835d4f98a2SYan Zheng if (counted) 36845d4f98a2SYan Zheng ref_count--; 36855d4f98a2SYan Zheng if (added) 36865d4f98a2SYan Zheng goto next; 36875d4f98a2SYan Zheng 36887476dfdaSDavid Sterba if (!tree_block_processed(leaf->start, rc)) { 36895d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 36905d4f98a2SYan Zheng if (!block) { 36915d4f98a2SYan Zheng err = -ENOMEM; 36925d4f98a2SYan Zheng break; 36935d4f98a2SYan Zheng } 36945d4f98a2SYan Zheng block->bytenr = leaf->start; 36955d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &block->key, 0); 36965d4f98a2SYan Zheng block->level = 0; 36975d4f98a2SYan Zheng block->key_ready = 1; 36985d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, 36995d4f98a2SYan Zheng &block->rb_node); 370043c04fb1SJeff Mahoney if (rb_node) 370143c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 370243c04fb1SJeff Mahoney block->bytenr); 37035d4f98a2SYan Zheng } 37045d4f98a2SYan Zheng if (counted) 37055d4f98a2SYan Zheng added = 1; 37065d4f98a2SYan Zheng else 37075d4f98a2SYan Zheng path->slots[0] = nritems; 37085d4f98a2SYan Zheng next: 37095d4f98a2SYan Zheng path->slots[0]++; 37105d4f98a2SYan Zheng 37115d4f98a2SYan Zheng } 37125d4f98a2SYan Zheng out: 37135d4f98a2SYan Zheng btrfs_free_path(path); 37145d4f98a2SYan Zheng return err; 37155d4f98a2SYan Zheng } 37165d4f98a2SYan Zheng 37175d4f98a2SYan Zheng /* 37182c016dc2SLiu Bo * helper to find all tree blocks that reference a given data extent 37195d4f98a2SYan Zheng */ 37205d4f98a2SYan Zheng static noinline_for_stack 37215d4f98a2SYan Zheng int add_data_references(struct reloc_control *rc, 37225d4f98a2SYan Zheng struct btrfs_key *extent_key, 37235d4f98a2SYan Zheng struct btrfs_path *path, 37245d4f98a2SYan Zheng struct rb_root *blocks) 37255d4f98a2SYan Zheng { 37265d4f98a2SYan Zheng struct btrfs_key key; 37275d4f98a2SYan Zheng struct extent_buffer *eb; 37285d4f98a2SYan Zheng struct btrfs_extent_data_ref *dref; 37295d4f98a2SYan Zheng struct btrfs_extent_inline_ref *iref; 37305d4f98a2SYan Zheng unsigned long ptr; 37315d4f98a2SYan Zheng unsigned long end; 3732da17066cSJeff Mahoney u32 blocksize = rc->extent_root->fs_info->nodesize; 3733647f63bdSFilipe David Borba Manana int ret = 0; 37345d4f98a2SYan Zheng int err = 0; 37355d4f98a2SYan Zheng 37365d4f98a2SYan Zheng eb = path->nodes[0]; 37375d4f98a2SYan Zheng ptr = btrfs_item_ptr_offset(eb, path->slots[0]); 37385d4f98a2SYan Zheng end = ptr + btrfs_item_size_nr(eb, path->slots[0]); 37395d4f98a2SYan Zheng ptr += sizeof(struct btrfs_extent_item); 37405d4f98a2SYan Zheng 37415d4f98a2SYan Zheng while (ptr < end) { 37425d4f98a2SYan Zheng iref = (struct btrfs_extent_inline_ref *)ptr; 37433de28d57SLiu Bo key.type = btrfs_get_extent_inline_ref_type(eb, iref, 37443de28d57SLiu Bo BTRFS_REF_TYPE_DATA); 37455d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY) { 37465d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 37475d4f98a2SYan Zheng ret = __add_tree_block(rc, key.offset, blocksize, 37485d4f98a2SYan Zheng blocks); 37495d4f98a2SYan Zheng } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { 37505d4f98a2SYan Zheng dref = (struct btrfs_extent_data_ref *)(&iref->offset); 37515d4f98a2SYan Zheng ret = find_data_references(rc, extent_key, 37525d4f98a2SYan Zheng eb, dref, blocks); 37535d4f98a2SYan Zheng } else { 3754af431dcbSSu Yue ret = -EUCLEAN; 3755b14c55a1SLiu Bo btrfs_err(rc->extent_root->fs_info, 3756b14c55a1SLiu Bo "extent %llu slot %d has an invalid inline ref type", 3757b14c55a1SLiu Bo eb->start, path->slots[0]); 37585d4f98a2SYan Zheng } 3759647f63bdSFilipe David Borba Manana if (ret) { 3760647f63bdSFilipe David Borba Manana err = ret; 3761647f63bdSFilipe David Borba Manana goto out; 3762647f63bdSFilipe David Borba Manana } 37635d4f98a2SYan Zheng ptr += btrfs_extent_inline_ref_size(key.type); 37645d4f98a2SYan Zheng } 37655d4f98a2SYan Zheng WARN_ON(ptr > end); 37665d4f98a2SYan Zheng 37675d4f98a2SYan Zheng while (1) { 37685d4f98a2SYan Zheng cond_resched(); 37695d4f98a2SYan Zheng eb = path->nodes[0]; 37705d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(eb)) { 37715d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 37725d4f98a2SYan Zheng if (ret < 0) { 37735d4f98a2SYan Zheng err = ret; 37745d4f98a2SYan Zheng break; 37755d4f98a2SYan Zheng } 37765d4f98a2SYan Zheng if (ret > 0) 37775d4f98a2SYan Zheng break; 37785d4f98a2SYan Zheng eb = path->nodes[0]; 37795d4f98a2SYan Zheng } 37805d4f98a2SYan Zheng 37815d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &key, path->slots[0]); 37825d4f98a2SYan Zheng if (key.objectid != extent_key->objectid) 37835d4f98a2SYan Zheng break; 37845d4f98a2SYan Zheng 37855d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY) { 37865d4f98a2SYan Zheng ret = __add_tree_block(rc, key.offset, blocksize, 37875d4f98a2SYan Zheng blocks); 37885d4f98a2SYan Zheng } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { 37895d4f98a2SYan Zheng dref = btrfs_item_ptr(eb, path->slots[0], 37905d4f98a2SYan Zheng struct btrfs_extent_data_ref); 37915d4f98a2SYan Zheng ret = find_data_references(rc, extent_key, 37925d4f98a2SYan Zheng eb, dref, blocks); 37936d8ff4e4SDavid Sterba } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { 3794ba3c2b19SNikolay Borisov btrfs_print_v0_err(eb->fs_info); 3795ba3c2b19SNikolay Borisov btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL); 3796ba3c2b19SNikolay Borisov ret = -EINVAL; 37975d4f98a2SYan Zheng } else { 37985d4f98a2SYan Zheng ret = 0; 37995d4f98a2SYan Zheng } 38005d4f98a2SYan Zheng if (ret) { 38015d4f98a2SYan Zheng err = ret; 38025d4f98a2SYan Zheng break; 38035d4f98a2SYan Zheng } 38045d4f98a2SYan Zheng path->slots[0]++; 38055d4f98a2SYan Zheng } 3806647f63bdSFilipe David Borba Manana out: 3807b3b4aa74SDavid Sterba btrfs_release_path(path); 38085d4f98a2SYan Zheng if (err) 38095d4f98a2SYan Zheng free_block_list(blocks); 38105d4f98a2SYan Zheng return err; 38115d4f98a2SYan Zheng } 38125d4f98a2SYan Zheng 38135d4f98a2SYan Zheng /* 38142c016dc2SLiu Bo * helper to find next unprocessed extent 38155d4f98a2SYan Zheng */ 38165d4f98a2SYan Zheng static noinline_for_stack 3817147d256eSZhaolei int find_next_extent(struct reloc_control *rc, struct btrfs_path *path, 38183fd0a558SYan, Zheng struct btrfs_key *extent_key) 38195d4f98a2SYan Zheng { 38200b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 38215d4f98a2SYan Zheng struct btrfs_key key; 38225d4f98a2SYan Zheng struct extent_buffer *leaf; 38235d4f98a2SYan Zheng u64 start, end, last; 38245d4f98a2SYan Zheng int ret; 38255d4f98a2SYan Zheng 38265d4f98a2SYan Zheng last = rc->block_group->key.objectid + rc->block_group->key.offset; 38275d4f98a2SYan Zheng while (1) { 38285d4f98a2SYan Zheng cond_resched(); 38295d4f98a2SYan Zheng if (rc->search_start >= last) { 38305d4f98a2SYan Zheng ret = 1; 38315d4f98a2SYan Zheng break; 38325d4f98a2SYan Zheng } 38335d4f98a2SYan Zheng 38345d4f98a2SYan Zheng key.objectid = rc->search_start; 38355d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 38365d4f98a2SYan Zheng key.offset = 0; 38375d4f98a2SYan Zheng 38385d4f98a2SYan Zheng path->search_commit_root = 1; 38395d4f98a2SYan Zheng path->skip_locking = 1; 38405d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 38415d4f98a2SYan Zheng 0, 0); 38425d4f98a2SYan Zheng if (ret < 0) 38435d4f98a2SYan Zheng break; 38445d4f98a2SYan Zheng next: 38455d4f98a2SYan Zheng leaf = path->nodes[0]; 38465d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(leaf)) { 38475d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 38485d4f98a2SYan Zheng if (ret != 0) 38495d4f98a2SYan Zheng break; 38505d4f98a2SYan Zheng leaf = path->nodes[0]; 38515d4f98a2SYan Zheng } 38525d4f98a2SYan Zheng 38535d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 38545d4f98a2SYan Zheng if (key.objectid >= last) { 38555d4f98a2SYan Zheng ret = 1; 38565d4f98a2SYan Zheng break; 38575d4f98a2SYan Zheng } 38585d4f98a2SYan Zheng 38593173a18fSJosef Bacik if (key.type != BTRFS_EXTENT_ITEM_KEY && 38603173a18fSJosef Bacik key.type != BTRFS_METADATA_ITEM_KEY) { 38613173a18fSJosef Bacik path->slots[0]++; 38623173a18fSJosef Bacik goto next; 38633173a18fSJosef Bacik } 38643173a18fSJosef Bacik 38653173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 38665d4f98a2SYan Zheng key.objectid + key.offset <= rc->search_start) { 38675d4f98a2SYan Zheng path->slots[0]++; 38685d4f98a2SYan Zheng goto next; 38695d4f98a2SYan Zheng } 38705d4f98a2SYan Zheng 38713173a18fSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 38720b246afaSJeff Mahoney key.objectid + fs_info->nodesize <= 38733173a18fSJosef Bacik rc->search_start) { 38743173a18fSJosef Bacik path->slots[0]++; 38753173a18fSJosef Bacik goto next; 38763173a18fSJosef Bacik } 38773173a18fSJosef Bacik 38785d4f98a2SYan Zheng ret = find_first_extent_bit(&rc->processed_blocks, 38795d4f98a2SYan Zheng key.objectid, &start, &end, 3880e6138876SJosef Bacik EXTENT_DIRTY, NULL); 38815d4f98a2SYan Zheng 38825d4f98a2SYan Zheng if (ret == 0 && start <= key.objectid) { 3883b3b4aa74SDavid Sterba btrfs_release_path(path); 38845d4f98a2SYan Zheng rc->search_start = end + 1; 38855d4f98a2SYan Zheng } else { 38863173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) 38875d4f98a2SYan Zheng rc->search_start = key.objectid + key.offset; 38883173a18fSJosef Bacik else 38893173a18fSJosef Bacik rc->search_start = key.objectid + 38900b246afaSJeff Mahoney fs_info->nodesize; 38913fd0a558SYan, Zheng memcpy(extent_key, &key, sizeof(key)); 38925d4f98a2SYan Zheng return 0; 38935d4f98a2SYan Zheng } 38945d4f98a2SYan Zheng } 3895b3b4aa74SDavid Sterba btrfs_release_path(path); 38965d4f98a2SYan Zheng return ret; 38975d4f98a2SYan Zheng } 38985d4f98a2SYan Zheng 38995d4f98a2SYan Zheng static void set_reloc_control(struct reloc_control *rc) 39005d4f98a2SYan Zheng { 39015d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 39027585717fSChris Mason 39037585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 39045d4f98a2SYan Zheng fs_info->reloc_ctl = rc; 39057585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 39065d4f98a2SYan Zheng } 39075d4f98a2SYan Zheng 39085d4f98a2SYan Zheng static void unset_reloc_control(struct reloc_control *rc) 39095d4f98a2SYan Zheng { 39105d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 39117585717fSChris Mason 39127585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 39135d4f98a2SYan Zheng fs_info->reloc_ctl = NULL; 39147585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 39155d4f98a2SYan Zheng } 39165d4f98a2SYan Zheng 39175d4f98a2SYan Zheng static int check_extent_flags(u64 flags) 39185d4f98a2SYan Zheng { 39195d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 39205d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 39215d4f98a2SYan Zheng return 1; 39225d4f98a2SYan Zheng if (!(flags & BTRFS_EXTENT_FLAG_DATA) && 39235d4f98a2SYan Zheng !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 39245d4f98a2SYan Zheng return 1; 39255d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 39265d4f98a2SYan Zheng (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) 39275d4f98a2SYan Zheng return 1; 39285d4f98a2SYan Zheng return 0; 39295d4f98a2SYan Zheng } 39305d4f98a2SYan Zheng 39313fd0a558SYan, Zheng static noinline_for_stack 39323fd0a558SYan, Zheng int prepare_to_relocate(struct reloc_control *rc) 39333fd0a558SYan, Zheng { 39343fd0a558SYan, Zheng struct btrfs_trans_handle *trans; 3935ac2fabacSJosef Bacik int ret; 39363fd0a558SYan, Zheng 39372ff7e61eSJeff Mahoney rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info, 393866d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 39393fd0a558SYan, Zheng if (!rc->block_rsv) 39403fd0a558SYan, Zheng return -ENOMEM; 39413fd0a558SYan, Zheng 39423fd0a558SYan, Zheng memset(&rc->cluster, 0, sizeof(rc->cluster)); 39433fd0a558SYan, Zheng rc->search_start = rc->block_group->key.objectid; 39443fd0a558SYan, Zheng rc->extents_found = 0; 39453fd0a558SYan, Zheng rc->nodes_relocated = 0; 39463fd0a558SYan, Zheng rc->merging_rsv_size = 0; 39470647bf56SWang Shilong rc->reserved_bytes = 0; 3948da17066cSJeff Mahoney rc->block_rsv->size = rc->extent_root->fs_info->nodesize * 39490647bf56SWang Shilong RELOCATION_RESERVED_NODES; 3950ac2fabacSJosef Bacik ret = btrfs_block_rsv_refill(rc->extent_root, 3951ac2fabacSJosef Bacik rc->block_rsv, rc->block_rsv->size, 3952ac2fabacSJosef Bacik BTRFS_RESERVE_FLUSH_ALL); 3953ac2fabacSJosef Bacik if (ret) 3954ac2fabacSJosef Bacik return ret; 39553fd0a558SYan, Zheng 39563fd0a558SYan, Zheng rc->create_reloc_tree = 1; 39573fd0a558SYan, Zheng set_reloc_control(rc); 39583fd0a558SYan, Zheng 39597a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 396028818947SLiu Bo if (IS_ERR(trans)) { 396128818947SLiu Bo unset_reloc_control(rc); 396228818947SLiu Bo /* 396328818947SLiu Bo * extent tree is not a ref_cow tree and has no reloc_root to 396428818947SLiu Bo * cleanup. And callers are responsible to free the above 396528818947SLiu Bo * block rsv. 396628818947SLiu Bo */ 396728818947SLiu Bo return PTR_ERR(trans); 396828818947SLiu Bo } 39693a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 39703fd0a558SYan, Zheng return 0; 39713fd0a558SYan, Zheng } 397276dda93cSYan, Zheng 39735d4f98a2SYan Zheng static noinline_for_stack int relocate_block_group(struct reloc_control *rc) 39745d4f98a2SYan Zheng { 39752ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 39765d4f98a2SYan Zheng struct rb_root blocks = RB_ROOT; 39775d4f98a2SYan Zheng struct btrfs_key key; 39785d4f98a2SYan Zheng struct btrfs_trans_handle *trans = NULL; 39795d4f98a2SYan Zheng struct btrfs_path *path; 39805d4f98a2SYan Zheng struct btrfs_extent_item *ei; 39815d4f98a2SYan Zheng u64 flags; 39825d4f98a2SYan Zheng u32 item_size; 39835d4f98a2SYan Zheng int ret; 39845d4f98a2SYan Zheng int err = 0; 3985c87f08caSChris Mason int progress = 0; 39865d4f98a2SYan Zheng 39875d4f98a2SYan Zheng path = btrfs_alloc_path(); 39883fd0a558SYan, Zheng if (!path) 39895d4f98a2SYan Zheng return -ENOMEM; 3990e4058b54SDavid Sterba path->reada = READA_FORWARD; 39913fd0a558SYan, Zheng 39923fd0a558SYan, Zheng ret = prepare_to_relocate(rc); 39933fd0a558SYan, Zheng if (ret) { 39943fd0a558SYan, Zheng err = ret; 39953fd0a558SYan, Zheng goto out_free; 39962423fdfbSJiri Slaby } 39975d4f98a2SYan Zheng 39985d4f98a2SYan Zheng while (1) { 39990647bf56SWang Shilong rc->reserved_bytes = 0; 40000647bf56SWang Shilong ret = btrfs_block_rsv_refill(rc->extent_root, 40010647bf56SWang Shilong rc->block_rsv, rc->block_rsv->size, 40020647bf56SWang Shilong BTRFS_RESERVE_FLUSH_ALL); 40030647bf56SWang Shilong if (ret) { 40040647bf56SWang Shilong err = ret; 40050647bf56SWang Shilong break; 40060647bf56SWang Shilong } 4007c87f08caSChris Mason progress++; 4008a22285a6SYan, Zheng trans = btrfs_start_transaction(rc->extent_root, 0); 40090f788c58SLiu Bo if (IS_ERR(trans)) { 40100f788c58SLiu Bo err = PTR_ERR(trans); 40110f788c58SLiu Bo trans = NULL; 40120f788c58SLiu Bo break; 40130f788c58SLiu Bo } 4014c87f08caSChris Mason restart: 40153fd0a558SYan, Zheng if (update_backref_cache(trans, &rc->backref_cache)) { 40163a45bb20SJeff Mahoney btrfs_end_transaction(trans); 401742a657f5SPan Bian trans = NULL; 40183fd0a558SYan, Zheng continue; 40193fd0a558SYan, Zheng } 40203fd0a558SYan, Zheng 4021147d256eSZhaolei ret = find_next_extent(rc, path, &key); 40225d4f98a2SYan Zheng if (ret < 0) 40235d4f98a2SYan Zheng err = ret; 40245d4f98a2SYan Zheng if (ret != 0) 40255d4f98a2SYan Zheng break; 40265d4f98a2SYan Zheng 40275d4f98a2SYan Zheng rc->extents_found++; 40285d4f98a2SYan Zheng 40295d4f98a2SYan Zheng ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 40305d4f98a2SYan Zheng struct btrfs_extent_item); 40313fd0a558SYan, Zheng item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); 40325d4f98a2SYan Zheng if (item_size >= sizeof(*ei)) { 40335d4f98a2SYan Zheng flags = btrfs_extent_flags(path->nodes[0], ei); 40345d4f98a2SYan Zheng ret = check_extent_flags(flags); 40355d4f98a2SYan Zheng BUG_ON(ret); 40366d8ff4e4SDavid Sterba } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) { 4037ba3c2b19SNikolay Borisov err = -EINVAL; 4038ba3c2b19SNikolay Borisov btrfs_print_v0_err(trans->fs_info); 4039ba3c2b19SNikolay Borisov btrfs_abort_transaction(trans, err); 4040ba3c2b19SNikolay Borisov break; 40415d4f98a2SYan Zheng } else { 40425d4f98a2SYan Zheng BUG(); 40435d4f98a2SYan Zheng } 40445d4f98a2SYan Zheng 40455d4f98a2SYan Zheng if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 40465d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, &blocks); 40475d4f98a2SYan Zheng } else if (rc->stage == UPDATE_DATA_PTRS && 40485d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 40495d4f98a2SYan Zheng ret = add_data_references(rc, &key, path, &blocks); 40505d4f98a2SYan Zheng } else { 4051b3b4aa74SDavid Sterba btrfs_release_path(path); 40525d4f98a2SYan Zheng ret = 0; 40535d4f98a2SYan Zheng } 40545d4f98a2SYan Zheng if (ret < 0) { 40553fd0a558SYan, Zheng err = ret; 40565d4f98a2SYan Zheng break; 40575d4f98a2SYan Zheng } 40585d4f98a2SYan Zheng 40595d4f98a2SYan Zheng if (!RB_EMPTY_ROOT(&blocks)) { 40605d4f98a2SYan Zheng ret = relocate_tree_blocks(trans, rc, &blocks); 40615d4f98a2SYan Zheng if (ret < 0) { 40621708cc57SWang Shilong /* 40631708cc57SWang Shilong * if we fail to relocate tree blocks, force to update 40641708cc57SWang Shilong * backref cache when committing transaction. 40651708cc57SWang Shilong */ 40661708cc57SWang Shilong rc->backref_cache.last_trans = trans->transid - 1; 40671708cc57SWang Shilong 40683fd0a558SYan, Zheng if (ret != -EAGAIN) { 40695d4f98a2SYan Zheng err = ret; 40705d4f98a2SYan Zheng break; 40715d4f98a2SYan Zheng } 40723fd0a558SYan, Zheng rc->extents_found--; 40733fd0a558SYan, Zheng rc->search_start = key.objectid; 40743fd0a558SYan, Zheng } 40755d4f98a2SYan Zheng } 40765d4f98a2SYan Zheng 40773a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 40782ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 40793fd0a558SYan, Zheng trans = NULL; 40805d4f98a2SYan Zheng 40815d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && 40825d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 40835d4f98a2SYan Zheng rc->found_file_extent = 1; 40840257bb82SYan, Zheng ret = relocate_data_extent(rc->data_inode, 40853fd0a558SYan, Zheng &key, &rc->cluster); 40865d4f98a2SYan Zheng if (ret < 0) { 40875d4f98a2SYan Zheng err = ret; 40885d4f98a2SYan Zheng break; 40895d4f98a2SYan Zheng } 40905d4f98a2SYan Zheng } 40915d4f98a2SYan Zheng } 4092c87f08caSChris Mason if (trans && progress && err == -ENOSPC) { 409343a7e99dSNikolay Borisov ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags); 40949689457bSShilong Wang if (ret == 1) { 4095c87f08caSChris Mason err = 0; 4096c87f08caSChris Mason progress = 0; 4097c87f08caSChris Mason goto restart; 4098c87f08caSChris Mason } 4099c87f08caSChris Mason } 41003fd0a558SYan, Zheng 4101b3b4aa74SDavid Sterba btrfs_release_path(path); 410291166212SDavid Sterba clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY); 41035d4f98a2SYan Zheng 41045d4f98a2SYan Zheng if (trans) { 41053a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 41062ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 41075d4f98a2SYan Zheng } 41085d4f98a2SYan Zheng 41090257bb82SYan, Zheng if (!err) { 41103fd0a558SYan, Zheng ret = relocate_file_extent_cluster(rc->data_inode, 41113fd0a558SYan, Zheng &rc->cluster); 41120257bb82SYan, Zheng if (ret < 0) 41130257bb82SYan, Zheng err = ret; 41140257bb82SYan, Zheng } 41150257bb82SYan, Zheng 41163fd0a558SYan, Zheng rc->create_reloc_tree = 0; 41173fd0a558SYan, Zheng set_reloc_control(rc); 41180257bb82SYan, Zheng 41193fd0a558SYan, Zheng backref_cache_cleanup(&rc->backref_cache); 41202ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1); 41215d4f98a2SYan Zheng 41223fd0a558SYan, Zheng err = prepare_to_merge(rc, err); 41235d4f98a2SYan Zheng 41245d4f98a2SYan Zheng merge_reloc_roots(rc); 41255d4f98a2SYan Zheng 41263fd0a558SYan, Zheng rc->merge_reloc_tree = 0; 41275d4f98a2SYan Zheng unset_reloc_control(rc); 41282ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1); 41295d4f98a2SYan Zheng 41305d4f98a2SYan Zheng /* get rid of pinned extents */ 41317a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 413262b99540SQu Wenruo if (IS_ERR(trans)) { 41333612b495STsutomu Itoh err = PTR_ERR(trans); 413462b99540SQu Wenruo goto out_free; 413562b99540SQu Wenruo } 41363a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 4137d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 4138d2311e69SQu Wenruo if (ret < 0 && !err) 4139d2311e69SQu Wenruo err = ret; 41403fd0a558SYan, Zheng out_free: 41412ff7e61eSJeff Mahoney btrfs_free_block_rsv(fs_info, rc->block_rsv); 41423fd0a558SYan, Zheng btrfs_free_path(path); 41435d4f98a2SYan Zheng return err; 41445d4f98a2SYan Zheng } 41455d4f98a2SYan Zheng 41465d4f98a2SYan Zheng static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 41470257bb82SYan, Zheng struct btrfs_root *root, u64 objectid) 41485d4f98a2SYan Zheng { 41495d4f98a2SYan Zheng struct btrfs_path *path; 41505d4f98a2SYan Zheng struct btrfs_inode_item *item; 41515d4f98a2SYan Zheng struct extent_buffer *leaf; 41525d4f98a2SYan Zheng int ret; 41535d4f98a2SYan Zheng 41545d4f98a2SYan Zheng path = btrfs_alloc_path(); 41555d4f98a2SYan Zheng if (!path) 41565d4f98a2SYan Zheng return -ENOMEM; 41575d4f98a2SYan Zheng 41585d4f98a2SYan Zheng ret = btrfs_insert_empty_inode(trans, root, path, objectid); 41595d4f98a2SYan Zheng if (ret) 41605d4f98a2SYan Zheng goto out; 41615d4f98a2SYan Zheng 41625d4f98a2SYan Zheng leaf = path->nodes[0]; 41635d4f98a2SYan Zheng item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); 4164b159fa28SDavid Sterba memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 41655d4f98a2SYan Zheng btrfs_set_inode_generation(leaf, item, 1); 41660257bb82SYan, Zheng btrfs_set_inode_size(leaf, item, 0); 41675d4f98a2SYan Zheng btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); 41683fd0a558SYan, Zheng btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS | 41693fd0a558SYan, Zheng BTRFS_INODE_PREALLOC); 41705d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 41715d4f98a2SYan Zheng out: 41725d4f98a2SYan Zheng btrfs_free_path(path); 41735d4f98a2SYan Zheng return ret; 41745d4f98a2SYan Zheng } 41755d4f98a2SYan Zheng 41765d4f98a2SYan Zheng /* 41775d4f98a2SYan Zheng * helper to create inode for data relocation. 41785d4f98a2SYan Zheng * the inode is in data relocation tree and its link count is 0 41795d4f98a2SYan Zheng */ 41803fd0a558SYan, Zheng static noinline_for_stack 41813fd0a558SYan, Zheng struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, 41825d4f98a2SYan Zheng struct btrfs_block_group_cache *group) 41835d4f98a2SYan Zheng { 41845d4f98a2SYan Zheng struct inode *inode = NULL; 41855d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 41865d4f98a2SYan Zheng struct btrfs_root *root; 41875d4f98a2SYan Zheng struct btrfs_key key; 41884624900dSZhaolei u64 objectid; 41895d4f98a2SYan Zheng int err = 0; 41905d4f98a2SYan Zheng 41915d4f98a2SYan Zheng root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 41925d4f98a2SYan Zheng if (IS_ERR(root)) 41935d4f98a2SYan Zheng return ERR_CAST(root); 41945d4f98a2SYan Zheng 4195a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 6); 41963fd0a558SYan, Zheng if (IS_ERR(trans)) 41973fd0a558SYan, Zheng return ERR_CAST(trans); 41985d4f98a2SYan Zheng 4199581bb050SLi Zefan err = btrfs_find_free_objectid(root, &objectid); 42005d4f98a2SYan Zheng if (err) 42015d4f98a2SYan Zheng goto out; 42025d4f98a2SYan Zheng 42030257bb82SYan, Zheng err = __insert_orphan_inode(trans, root, objectid); 42045d4f98a2SYan Zheng BUG_ON(err); 42055d4f98a2SYan Zheng 42065d4f98a2SYan Zheng key.objectid = objectid; 42075d4f98a2SYan Zheng key.type = BTRFS_INODE_ITEM_KEY; 42085d4f98a2SYan Zheng key.offset = 0; 42090b246afaSJeff Mahoney inode = btrfs_iget(fs_info->sb, &key, root, NULL); 42102e19f1f9SAl Viro BUG_ON(IS_ERR(inode)); 42115d4f98a2SYan Zheng BTRFS_I(inode)->index_cnt = group->key.objectid; 42125d4f98a2SYan Zheng 421373f2e545SNikolay Borisov err = btrfs_orphan_add(trans, BTRFS_I(inode)); 42145d4f98a2SYan Zheng out: 42153a45bb20SJeff Mahoney btrfs_end_transaction(trans); 42162ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 42175d4f98a2SYan Zheng if (err) { 42185d4f98a2SYan Zheng if (inode) 42195d4f98a2SYan Zheng iput(inode); 42205d4f98a2SYan Zheng inode = ERR_PTR(err); 42215d4f98a2SYan Zheng } 42225d4f98a2SYan Zheng return inode; 42235d4f98a2SYan Zheng } 42245d4f98a2SYan Zheng 4225c258d6e3SQu Wenruo static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info) 42263fd0a558SYan, Zheng { 42273fd0a558SYan, Zheng struct reloc_control *rc; 42283fd0a558SYan, Zheng 42293fd0a558SYan, Zheng rc = kzalloc(sizeof(*rc), GFP_NOFS); 42303fd0a558SYan, Zheng if (!rc) 42313fd0a558SYan, Zheng return NULL; 42323fd0a558SYan, Zheng 42333fd0a558SYan, Zheng INIT_LIST_HEAD(&rc->reloc_roots); 4234d2311e69SQu Wenruo INIT_LIST_HEAD(&rc->dirty_subvol_roots); 42353fd0a558SYan, Zheng backref_cache_init(&rc->backref_cache); 42363fd0a558SYan, Zheng mapping_tree_init(&rc->reloc_root_tree); 423743eb5f29SQu Wenruo extent_io_tree_init(fs_info, &rc->processed_blocks, 423843eb5f29SQu Wenruo IO_TREE_RELOC_BLOCKS, NULL); 42393fd0a558SYan, Zheng return rc; 42403fd0a558SYan, Zheng } 42413fd0a558SYan, Zheng 42425d4f98a2SYan Zheng /* 4243ebce0e01SAdam Borowski * Print the block group being relocated 4244ebce0e01SAdam Borowski */ 4245ebce0e01SAdam Borowski static void describe_relocation(struct btrfs_fs_info *fs_info, 4246ebce0e01SAdam Borowski struct btrfs_block_group_cache *block_group) 4247ebce0e01SAdam Borowski { 4248f89e09cfSAnand Jain char buf[128] = {'\0'}; 4249ebce0e01SAdam Borowski 4250f89e09cfSAnand Jain btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf)); 4251ebce0e01SAdam Borowski 4252ebce0e01SAdam Borowski btrfs_info(fs_info, 4253ebce0e01SAdam Borowski "relocating block group %llu flags %s", 4254f89e09cfSAnand Jain block_group->key.objectid, buf); 4255ebce0e01SAdam Borowski } 4256ebce0e01SAdam Borowski 4257ebce0e01SAdam Borowski /* 42585d4f98a2SYan Zheng * function to relocate all extents in a block group. 42595d4f98a2SYan Zheng */ 42606bccf3abSJeff Mahoney int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) 42615d4f98a2SYan Zheng { 4262eede2bf3SOmar Sandoval struct btrfs_block_group_cache *bg; 42636bccf3abSJeff Mahoney struct btrfs_root *extent_root = fs_info->extent_root; 42645d4f98a2SYan Zheng struct reloc_control *rc; 42650af3d00bSJosef Bacik struct inode *inode; 42660af3d00bSJosef Bacik struct btrfs_path *path; 42675d4f98a2SYan Zheng int ret; 4268f0486c68SYan, Zheng int rw = 0; 42695d4f98a2SYan Zheng int err = 0; 42705d4f98a2SYan Zheng 4271eede2bf3SOmar Sandoval bg = btrfs_lookup_block_group(fs_info, group_start); 4272eede2bf3SOmar Sandoval if (!bg) 4273eede2bf3SOmar Sandoval return -ENOENT; 4274eede2bf3SOmar Sandoval 4275eede2bf3SOmar Sandoval if (btrfs_pinned_by_swapfile(fs_info, bg)) { 4276eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 4277eede2bf3SOmar Sandoval return -ETXTBSY; 4278eede2bf3SOmar Sandoval } 4279eede2bf3SOmar Sandoval 4280c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 4281eede2bf3SOmar Sandoval if (!rc) { 4282eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 42835d4f98a2SYan Zheng return -ENOMEM; 4284eede2bf3SOmar Sandoval } 42855d4f98a2SYan Zheng 4286f0486c68SYan, Zheng rc->extent_root = extent_root; 4287eede2bf3SOmar Sandoval rc->block_group = bg; 42885d4f98a2SYan Zheng 4289c83488afSNikolay Borisov ret = btrfs_inc_block_group_ro(rc->block_group); 4290f0486c68SYan, Zheng if (ret) { 4291f0486c68SYan, Zheng err = ret; 4292f0486c68SYan, Zheng goto out; 4293f0486c68SYan, Zheng } 4294f0486c68SYan, Zheng rw = 1; 4295f0486c68SYan, Zheng 42960af3d00bSJosef Bacik path = btrfs_alloc_path(); 42970af3d00bSJosef Bacik if (!path) { 42980af3d00bSJosef Bacik err = -ENOMEM; 42990af3d00bSJosef Bacik goto out; 43000af3d00bSJosef Bacik } 43010af3d00bSJosef Bacik 4302*7949f339SDavid Sterba inode = lookup_free_space_inode(rc->block_group, path); 43030af3d00bSJosef Bacik btrfs_free_path(path); 43040af3d00bSJosef Bacik 43050af3d00bSJosef Bacik if (!IS_ERR(inode)) 43061bbc621eSChris Mason ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0); 43070af3d00bSJosef Bacik else 43080af3d00bSJosef Bacik ret = PTR_ERR(inode); 43090af3d00bSJosef Bacik 43100af3d00bSJosef Bacik if (ret && ret != -ENOENT) { 43110af3d00bSJosef Bacik err = ret; 43120af3d00bSJosef Bacik goto out; 43130af3d00bSJosef Bacik } 43140af3d00bSJosef Bacik 43155d4f98a2SYan Zheng rc->data_inode = create_reloc_inode(fs_info, rc->block_group); 43165d4f98a2SYan Zheng if (IS_ERR(rc->data_inode)) { 43175d4f98a2SYan Zheng err = PTR_ERR(rc->data_inode); 43185d4f98a2SYan Zheng rc->data_inode = NULL; 43195d4f98a2SYan Zheng goto out; 43205d4f98a2SYan Zheng } 43215d4f98a2SYan Zheng 43220b246afaSJeff Mahoney describe_relocation(fs_info, rc->block_group); 43235d4f98a2SYan Zheng 43249cfa3e34SFilipe Manana btrfs_wait_block_group_reservations(rc->block_group); 4325f78c436cSFilipe Manana btrfs_wait_nocow_writers(rc->block_group); 43266374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 4327578def7cSFilipe Manana rc->block_group->key.objectid, 4328578def7cSFilipe Manana rc->block_group->key.offset); 43295d4f98a2SYan Zheng 43305d4f98a2SYan Zheng while (1) { 433176dda93cSYan, Zheng mutex_lock(&fs_info->cleaner_mutex); 43325d4f98a2SYan Zheng ret = relocate_block_group(rc); 433376dda93cSYan, Zheng mutex_unlock(&fs_info->cleaner_mutex); 4334ff612ba7SJosef Bacik if (ret < 0) 43355d4f98a2SYan Zheng err = ret; 4336ff612ba7SJosef Bacik 4337ff612ba7SJosef Bacik /* 4338ff612ba7SJosef Bacik * We may have gotten ENOSPC after we already dirtied some 4339ff612ba7SJosef Bacik * extents. If writeout happens while we're relocating a 4340ff612ba7SJosef Bacik * different block group we could end up hitting the 4341ff612ba7SJosef Bacik * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in 4342ff612ba7SJosef Bacik * btrfs_reloc_cow_block. Make sure we write everything out 4343ff612ba7SJosef Bacik * properly so we don't trip over this problem, and then break 4344ff612ba7SJosef Bacik * out of the loop if we hit an error. 4345ff612ba7SJosef Bacik */ 4346ff612ba7SJosef Bacik if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) { 4347ff612ba7SJosef Bacik ret = btrfs_wait_ordered_range(rc->data_inode, 0, 4348ff612ba7SJosef Bacik (u64)-1); 4349ff612ba7SJosef Bacik if (ret) 4350ff612ba7SJosef Bacik err = ret; 4351ff612ba7SJosef Bacik invalidate_mapping_pages(rc->data_inode->i_mapping, 4352ff612ba7SJosef Bacik 0, -1); 4353ff612ba7SJosef Bacik rc->stage = UPDATE_DATA_PTRS; 43545d4f98a2SYan Zheng } 43555d4f98a2SYan Zheng 4356ff612ba7SJosef Bacik if (err < 0) 4357ff612ba7SJosef Bacik goto out; 4358ff612ba7SJosef Bacik 43595d4f98a2SYan Zheng if (rc->extents_found == 0) 43605d4f98a2SYan Zheng break; 43615d4f98a2SYan Zheng 43620b246afaSJeff Mahoney btrfs_info(fs_info, "found %llu extents", rc->extents_found); 43635d4f98a2SYan Zheng 43645d4f98a2SYan Zheng } 43655d4f98a2SYan Zheng 43665d4f98a2SYan Zheng WARN_ON(rc->block_group->pinned > 0); 43675d4f98a2SYan Zheng WARN_ON(rc->block_group->reserved > 0); 43685d4f98a2SYan Zheng WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0); 43695d4f98a2SYan Zheng out: 4370f0486c68SYan, Zheng if (err && rw) 43712ff7e61eSJeff Mahoney btrfs_dec_block_group_ro(rc->block_group); 43725d4f98a2SYan Zheng iput(rc->data_inode); 43735d4f98a2SYan Zheng btrfs_put_block_group(rc->block_group); 43745d4f98a2SYan Zheng kfree(rc); 43755d4f98a2SYan Zheng return err; 43765d4f98a2SYan Zheng } 43775d4f98a2SYan Zheng 437876dda93cSYan, Zheng static noinline_for_stack int mark_garbage_root(struct btrfs_root *root) 437976dda93cSYan, Zheng { 43800b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 438176dda93cSYan, Zheng struct btrfs_trans_handle *trans; 438279787eaaSJeff Mahoney int ret, err; 438376dda93cSYan, Zheng 43840b246afaSJeff Mahoney trans = btrfs_start_transaction(fs_info->tree_root, 0); 438579787eaaSJeff Mahoney if (IS_ERR(trans)) 438679787eaaSJeff Mahoney return PTR_ERR(trans); 438776dda93cSYan, Zheng 438876dda93cSYan, Zheng memset(&root->root_item.drop_progress, 0, 438976dda93cSYan, Zheng sizeof(root->root_item.drop_progress)); 439076dda93cSYan, Zheng root->root_item.drop_level = 0; 439176dda93cSYan, Zheng btrfs_set_root_refs(&root->root_item, 0); 43920b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 439376dda93cSYan, Zheng &root->root_key, &root->root_item); 439476dda93cSYan, Zheng 43953a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 439679787eaaSJeff Mahoney if (err) 439779787eaaSJeff Mahoney return err; 439879787eaaSJeff Mahoney return ret; 439976dda93cSYan, Zheng } 440076dda93cSYan, Zheng 44015d4f98a2SYan Zheng /* 44025d4f98a2SYan Zheng * recover relocation interrupted by system crash. 44035d4f98a2SYan Zheng * 44045d4f98a2SYan Zheng * this function resumes merging reloc trees with corresponding fs trees. 44055d4f98a2SYan Zheng * this is important for keeping the sharing of tree blocks 44065d4f98a2SYan Zheng */ 44075d4f98a2SYan Zheng int btrfs_recover_relocation(struct btrfs_root *root) 44085d4f98a2SYan Zheng { 44090b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 44105d4f98a2SYan Zheng LIST_HEAD(reloc_roots); 44115d4f98a2SYan Zheng struct btrfs_key key; 44125d4f98a2SYan Zheng struct btrfs_root *fs_root; 44135d4f98a2SYan Zheng struct btrfs_root *reloc_root; 44145d4f98a2SYan Zheng struct btrfs_path *path; 44155d4f98a2SYan Zheng struct extent_buffer *leaf; 44165d4f98a2SYan Zheng struct reloc_control *rc = NULL; 44175d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 44185d4f98a2SYan Zheng int ret; 44195d4f98a2SYan Zheng int err = 0; 44205d4f98a2SYan Zheng 44215d4f98a2SYan Zheng path = btrfs_alloc_path(); 44225d4f98a2SYan Zheng if (!path) 44235d4f98a2SYan Zheng return -ENOMEM; 4424e4058b54SDavid Sterba path->reada = READA_BACK; 44255d4f98a2SYan Zheng 44265d4f98a2SYan Zheng key.objectid = BTRFS_TREE_RELOC_OBJECTID; 44275d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 44285d4f98a2SYan Zheng key.offset = (u64)-1; 44295d4f98a2SYan Zheng 44305d4f98a2SYan Zheng while (1) { 44310b246afaSJeff Mahoney ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, 44325d4f98a2SYan Zheng path, 0, 0); 44335d4f98a2SYan Zheng if (ret < 0) { 44345d4f98a2SYan Zheng err = ret; 44355d4f98a2SYan Zheng goto out; 44365d4f98a2SYan Zheng } 44375d4f98a2SYan Zheng if (ret > 0) { 44385d4f98a2SYan Zheng if (path->slots[0] == 0) 44395d4f98a2SYan Zheng break; 44405d4f98a2SYan Zheng path->slots[0]--; 44415d4f98a2SYan Zheng } 44425d4f98a2SYan Zheng leaf = path->nodes[0]; 44435d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 4444b3b4aa74SDavid Sterba btrfs_release_path(path); 44455d4f98a2SYan Zheng 44465d4f98a2SYan Zheng if (key.objectid != BTRFS_TREE_RELOC_OBJECTID || 44475d4f98a2SYan Zheng key.type != BTRFS_ROOT_ITEM_KEY) 44485d4f98a2SYan Zheng break; 44495d4f98a2SYan Zheng 4450cb517eabSMiao Xie reloc_root = btrfs_read_fs_root(root, &key); 44515d4f98a2SYan Zheng if (IS_ERR(reloc_root)) { 44525d4f98a2SYan Zheng err = PTR_ERR(reloc_root); 44535d4f98a2SYan Zheng goto out; 44545d4f98a2SYan Zheng } 44555d4f98a2SYan Zheng 44565d4f98a2SYan Zheng list_add(&reloc_root->root_list, &reloc_roots); 44575d4f98a2SYan Zheng 44585d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 44590b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, 44605d4f98a2SYan Zheng reloc_root->root_key.offset); 44615d4f98a2SYan Zheng if (IS_ERR(fs_root)) { 446276dda93cSYan, Zheng ret = PTR_ERR(fs_root); 446376dda93cSYan, Zheng if (ret != -ENOENT) { 446476dda93cSYan, Zheng err = ret; 44655d4f98a2SYan Zheng goto out; 44665d4f98a2SYan Zheng } 446779787eaaSJeff Mahoney ret = mark_garbage_root(reloc_root); 446879787eaaSJeff Mahoney if (ret < 0) { 446979787eaaSJeff Mahoney err = ret; 447079787eaaSJeff Mahoney goto out; 447179787eaaSJeff Mahoney } 447276dda93cSYan, Zheng } 44735d4f98a2SYan Zheng } 44745d4f98a2SYan Zheng 44755d4f98a2SYan Zheng if (key.offset == 0) 44765d4f98a2SYan Zheng break; 44775d4f98a2SYan Zheng 44785d4f98a2SYan Zheng key.offset--; 44795d4f98a2SYan Zheng } 4480b3b4aa74SDavid Sterba btrfs_release_path(path); 44815d4f98a2SYan Zheng 44825d4f98a2SYan Zheng if (list_empty(&reloc_roots)) 44835d4f98a2SYan Zheng goto out; 44845d4f98a2SYan Zheng 4485c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 44865d4f98a2SYan Zheng if (!rc) { 44875d4f98a2SYan Zheng err = -ENOMEM; 44885d4f98a2SYan Zheng goto out; 44895d4f98a2SYan Zheng } 44905d4f98a2SYan Zheng 44910b246afaSJeff Mahoney rc->extent_root = fs_info->extent_root; 44925d4f98a2SYan Zheng 44935d4f98a2SYan Zheng set_reloc_control(rc); 44945d4f98a2SYan Zheng 44957a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 44963612b495STsutomu Itoh if (IS_ERR(trans)) { 44973612b495STsutomu Itoh unset_reloc_control(rc); 44983612b495STsutomu Itoh err = PTR_ERR(trans); 44993612b495STsutomu Itoh goto out_free; 45003612b495STsutomu Itoh } 45013fd0a558SYan, Zheng 45023fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 45033fd0a558SYan, Zheng 45045d4f98a2SYan Zheng while (!list_empty(&reloc_roots)) { 45055d4f98a2SYan Zheng reloc_root = list_entry(reloc_roots.next, 45065d4f98a2SYan Zheng struct btrfs_root, root_list); 45075d4f98a2SYan Zheng list_del(&reloc_root->root_list); 45085d4f98a2SYan Zheng 45095d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) == 0) { 45105d4f98a2SYan Zheng list_add_tail(&reloc_root->root_list, 45115d4f98a2SYan Zheng &rc->reloc_roots); 45125d4f98a2SYan Zheng continue; 45135d4f98a2SYan Zheng } 45145d4f98a2SYan Zheng 45150b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, reloc_root->root_key.offset); 451679787eaaSJeff Mahoney if (IS_ERR(fs_root)) { 451779787eaaSJeff Mahoney err = PTR_ERR(fs_root); 451879787eaaSJeff Mahoney goto out_free; 451979787eaaSJeff Mahoney } 45205d4f98a2SYan Zheng 4521ffd7b339SJeff Mahoney err = __add_reloc_root(reloc_root); 452279787eaaSJeff Mahoney BUG_ON(err < 0); /* -ENOMEM or logic error */ 45235d4f98a2SYan Zheng fs_root->reloc_root = reloc_root; 45245d4f98a2SYan Zheng } 45255d4f98a2SYan Zheng 45263a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 452779787eaaSJeff Mahoney if (err) 452879787eaaSJeff Mahoney goto out_free; 45295d4f98a2SYan Zheng 45305d4f98a2SYan Zheng merge_reloc_roots(rc); 45315d4f98a2SYan Zheng 45325d4f98a2SYan Zheng unset_reloc_control(rc); 45335d4f98a2SYan Zheng 45347a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 453562b99540SQu Wenruo if (IS_ERR(trans)) { 45363612b495STsutomu Itoh err = PTR_ERR(trans); 453762b99540SQu Wenruo goto out_free; 453862b99540SQu Wenruo } 45393a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 4540d2311e69SQu Wenruo 4541d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 4542d2311e69SQu Wenruo if (ret < 0 && !err) 4543d2311e69SQu Wenruo err = ret; 45443612b495STsutomu Itoh out_free: 45455d4f98a2SYan Zheng kfree(rc); 45463612b495STsutomu Itoh out: 4547aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 4548aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 4549aca1bba6SLiu Bo 45505d4f98a2SYan Zheng btrfs_free_path(path); 45515d4f98a2SYan Zheng 45525d4f98a2SYan Zheng if (err == 0) { 45535d4f98a2SYan Zheng /* cleanup orphan inode in data relocation tree */ 45540b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 45555d4f98a2SYan Zheng if (IS_ERR(fs_root)) 45565d4f98a2SYan Zheng err = PTR_ERR(fs_root); 4557d7ce5843SMiao Xie else 455866b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_root); 45595d4f98a2SYan Zheng } 45605d4f98a2SYan Zheng return err; 45615d4f98a2SYan Zheng } 45625d4f98a2SYan Zheng 45635d4f98a2SYan Zheng /* 45645d4f98a2SYan Zheng * helper to add ordered checksum for data relocation. 45655d4f98a2SYan Zheng * 45665d4f98a2SYan Zheng * cloning checksum properly handles the nodatasum extents. 45675d4f98a2SYan Zheng * it also saves CPU time to re-calculate the checksum. 45685d4f98a2SYan Zheng */ 45695d4f98a2SYan Zheng int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len) 45705d4f98a2SYan Zheng { 45710b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 45725d4f98a2SYan Zheng struct btrfs_ordered_sum *sums; 45735d4f98a2SYan Zheng struct btrfs_ordered_extent *ordered; 45745d4f98a2SYan Zheng int ret; 45755d4f98a2SYan Zheng u64 disk_bytenr; 45764577b014SJosef Bacik u64 new_bytenr; 45775d4f98a2SYan Zheng LIST_HEAD(list); 45785d4f98a2SYan Zheng 45795d4f98a2SYan Zheng ordered = btrfs_lookup_ordered_extent(inode, file_pos); 45805d4f98a2SYan Zheng BUG_ON(ordered->file_offset != file_pos || ordered->len != len); 45815d4f98a2SYan Zheng 45825d4f98a2SYan Zheng disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt; 45830b246afaSJeff Mahoney ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr, 4584a2de733cSArne Jansen disk_bytenr + len - 1, &list, 0); 458579787eaaSJeff Mahoney if (ret) 458679787eaaSJeff Mahoney goto out; 45875d4f98a2SYan Zheng 45885d4f98a2SYan Zheng while (!list_empty(&list)) { 45895d4f98a2SYan Zheng sums = list_entry(list.next, struct btrfs_ordered_sum, list); 45905d4f98a2SYan Zheng list_del_init(&sums->list); 45915d4f98a2SYan Zheng 45924577b014SJosef Bacik /* 45934577b014SJosef Bacik * We need to offset the new_bytenr based on where the csum is. 45944577b014SJosef Bacik * We need to do this because we will read in entire prealloc 45954577b014SJosef Bacik * extents but we may have written to say the middle of the 45964577b014SJosef Bacik * prealloc extent, so we need to make sure the csum goes with 45974577b014SJosef Bacik * the right disk offset. 45984577b014SJosef Bacik * 45994577b014SJosef Bacik * We can do this because the data reloc inode refers strictly 46004577b014SJosef Bacik * to the on disk bytes, so we don't have to worry about 46014577b014SJosef Bacik * disk_len vs real len like with real inodes since it's all 46024577b014SJosef Bacik * disk length. 46034577b014SJosef Bacik */ 46044577b014SJosef Bacik new_bytenr = ordered->start + (sums->bytenr - disk_bytenr); 46054577b014SJosef Bacik sums->bytenr = new_bytenr; 46065d4f98a2SYan Zheng 4607f9756261SNikolay Borisov btrfs_add_ordered_sum(ordered, sums); 46085d4f98a2SYan Zheng } 460979787eaaSJeff Mahoney out: 46105d4f98a2SYan Zheng btrfs_put_ordered_extent(ordered); 4611411fc6bcSAndi Kleen return ret; 46125d4f98a2SYan Zheng } 46133fd0a558SYan, Zheng 461483d4cfd4SJosef Bacik int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans, 46153fd0a558SYan, Zheng struct btrfs_root *root, struct extent_buffer *buf, 46163fd0a558SYan, Zheng struct extent_buffer *cow) 46173fd0a558SYan, Zheng { 46180b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 46193fd0a558SYan, Zheng struct reloc_control *rc; 46203fd0a558SYan, Zheng struct backref_node *node; 46213fd0a558SYan, Zheng int first_cow = 0; 46223fd0a558SYan, Zheng int level; 462383d4cfd4SJosef Bacik int ret = 0; 46243fd0a558SYan, Zheng 46250b246afaSJeff Mahoney rc = fs_info->reloc_ctl; 46263fd0a558SYan, Zheng if (!rc) 462783d4cfd4SJosef Bacik return 0; 46283fd0a558SYan, Zheng 46293fd0a558SYan, Zheng BUG_ON(rc->stage == UPDATE_DATA_PTRS && 46303fd0a558SYan, Zheng root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID); 46313fd0a558SYan, Zheng 4632c974c464SWang Shilong if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 4633c974c464SWang Shilong if (buf == root->node) 4634c974c464SWang Shilong __update_reloc_root(root, cow->start); 4635c974c464SWang Shilong } 4636c974c464SWang Shilong 46373fd0a558SYan, Zheng level = btrfs_header_level(buf); 46383fd0a558SYan, Zheng if (btrfs_header_generation(buf) <= 46393fd0a558SYan, Zheng btrfs_root_last_snapshot(&root->root_item)) 46403fd0a558SYan, Zheng first_cow = 1; 46413fd0a558SYan, Zheng 46423fd0a558SYan, Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID && 46433fd0a558SYan, Zheng rc->create_reloc_tree) { 46443fd0a558SYan, Zheng WARN_ON(!first_cow && level == 0); 46453fd0a558SYan, Zheng 46463fd0a558SYan, Zheng node = rc->backref_cache.path[level]; 46473fd0a558SYan, Zheng BUG_ON(node->bytenr != buf->start && 46483fd0a558SYan, Zheng node->new_bytenr != buf->start); 46493fd0a558SYan, Zheng 46503fd0a558SYan, Zheng drop_node_buffer(node); 46513fd0a558SYan, Zheng extent_buffer_get(cow); 46523fd0a558SYan, Zheng node->eb = cow; 46533fd0a558SYan, Zheng node->new_bytenr = cow->start; 46543fd0a558SYan, Zheng 46553fd0a558SYan, Zheng if (!node->pending) { 46563fd0a558SYan, Zheng list_move_tail(&node->list, 46573fd0a558SYan, Zheng &rc->backref_cache.pending[level]); 46583fd0a558SYan, Zheng node->pending = 1; 46593fd0a558SYan, Zheng } 46603fd0a558SYan, Zheng 46613fd0a558SYan, Zheng if (first_cow) 46623fd0a558SYan, Zheng __mark_block_processed(rc, node); 46633fd0a558SYan, Zheng 46643fd0a558SYan, Zheng if (first_cow && level > 0) 46653fd0a558SYan, Zheng rc->nodes_relocated += buf->len; 46663fd0a558SYan, Zheng } 46673fd0a558SYan, Zheng 466883d4cfd4SJosef Bacik if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) 46693fd0a558SYan, Zheng ret = replace_file_extents(trans, rc, root, cow); 467083d4cfd4SJosef Bacik return ret; 46713fd0a558SYan, Zheng } 46723fd0a558SYan, Zheng 46733fd0a558SYan, Zheng /* 46743fd0a558SYan, Zheng * called before creating snapshot. it calculates metadata reservation 467501327610SNicholas D Steeves * required for relocating tree blocks in the snapshot 46763fd0a558SYan, Zheng */ 4677147d256eSZhaolei void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending, 46783fd0a558SYan, Zheng u64 *bytes_to_reserve) 46793fd0a558SYan, Zheng { 468010995c04SQu Wenruo struct btrfs_root *root = pending->root; 468110995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 46823fd0a558SYan, Zheng 468310995c04SQu Wenruo if (!root->reloc_root || !rc) 46843fd0a558SYan, Zheng return; 46853fd0a558SYan, Zheng 46863fd0a558SYan, Zheng if (!rc->merge_reloc_tree) 46873fd0a558SYan, Zheng return; 46883fd0a558SYan, Zheng 46893fd0a558SYan, Zheng root = root->reloc_root; 46903fd0a558SYan, Zheng BUG_ON(btrfs_root_refs(&root->root_item) == 0); 46913fd0a558SYan, Zheng /* 46923fd0a558SYan, Zheng * relocation is in the stage of merging trees. the space 46933fd0a558SYan, Zheng * used by merging a reloc tree is twice the size of 46943fd0a558SYan, Zheng * relocated tree nodes in the worst case. half for cowing 46953fd0a558SYan, Zheng * the reloc tree, half for cowing the fs tree. the space 46963fd0a558SYan, Zheng * used by cowing the reloc tree will be freed after the 46973fd0a558SYan, Zheng * tree is dropped. if we create snapshot, cowing the fs 46983fd0a558SYan, Zheng * tree may use more space than it frees. so we need 46993fd0a558SYan, Zheng * reserve extra space. 47003fd0a558SYan, Zheng */ 47013fd0a558SYan, Zheng *bytes_to_reserve += rc->nodes_relocated; 47023fd0a558SYan, Zheng } 47033fd0a558SYan, Zheng 47043fd0a558SYan, Zheng /* 47053fd0a558SYan, Zheng * called after snapshot is created. migrate block reservation 47063fd0a558SYan, Zheng * and create reloc root for the newly created snapshot 47073fd0a558SYan, Zheng */ 470849b25e05SJeff Mahoney int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans, 47093fd0a558SYan, Zheng struct btrfs_pending_snapshot *pending) 47103fd0a558SYan, Zheng { 47113fd0a558SYan, Zheng struct btrfs_root *root = pending->root; 47123fd0a558SYan, Zheng struct btrfs_root *reloc_root; 47133fd0a558SYan, Zheng struct btrfs_root *new_root; 471410995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 47153fd0a558SYan, Zheng int ret; 47163fd0a558SYan, Zheng 471710995c04SQu Wenruo if (!root->reloc_root || !rc) 471849b25e05SJeff Mahoney return 0; 47193fd0a558SYan, Zheng 47203fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 47213fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated; 47223fd0a558SYan, Zheng 47233fd0a558SYan, Zheng if (rc->merge_reloc_tree) { 47243fd0a558SYan, Zheng ret = btrfs_block_rsv_migrate(&pending->block_rsv, 47253fd0a558SYan, Zheng rc->block_rsv, 47263a584174SLu Fengqi rc->nodes_relocated, true); 472749b25e05SJeff Mahoney if (ret) 472849b25e05SJeff Mahoney return ret; 47293fd0a558SYan, Zheng } 47303fd0a558SYan, Zheng 47313fd0a558SYan, Zheng new_root = pending->snap; 47323fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root->reloc_root, 47333fd0a558SYan, Zheng new_root->root_key.objectid); 473449b25e05SJeff Mahoney if (IS_ERR(reloc_root)) 473549b25e05SJeff Mahoney return PTR_ERR(reloc_root); 47363fd0a558SYan, Zheng 4737ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 4738ffd7b339SJeff Mahoney BUG_ON(ret < 0); 47393fd0a558SYan, Zheng new_root->reloc_root = reloc_root; 47403fd0a558SYan, Zheng 474149b25e05SJeff Mahoney if (rc->create_reloc_tree) 47423fd0a558SYan, Zheng ret = clone_backref_node(trans, rc, root, reloc_root); 474349b25e05SJeff Mahoney return ret; 47443fd0a558SYan, Zheng } 4745