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++) { 164682fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 164782fa113fSQu Wenruo 16485d4f98a2SYan Zheng cond_resched(); 16495d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, i); 16505d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_DATA_KEY) 16515d4f98a2SYan Zheng continue; 16525d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 16535d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 16545d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 16555d4f98a2SYan Zheng continue; 16565d4f98a2SYan Zheng bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 16575d4f98a2SYan Zheng num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); 16585d4f98a2SYan Zheng if (bytenr == 0) 16595d4f98a2SYan Zheng continue; 16605d4f98a2SYan Zheng if (!in_block_group(bytenr, rc->block_group)) 16615d4f98a2SYan Zheng continue; 16625d4f98a2SYan Zheng 16635d4f98a2SYan Zheng /* 16645d4f98a2SYan Zheng * if we are modifying block in fs tree, wait for readpage 16655d4f98a2SYan Zheng * to complete and drop the extent cache 16665d4f98a2SYan Zheng */ 16675d4f98a2SYan Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 16685d4f98a2SYan Zheng if (first) { 16695d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 16705d4f98a2SYan Zheng first = 0; 16714a0cc7caSNikolay Borisov } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) { 16723fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 16735d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 16745d4f98a2SYan Zheng } 16754a0cc7caSNikolay Borisov if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) { 16765d4f98a2SYan Zheng end = key.offset + 16775d4f98a2SYan Zheng btrfs_file_extent_num_bytes(leaf, fi); 16785d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(key.offset, 16790b246afaSJeff Mahoney fs_info->sectorsize)); 16800b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 16815d4f98a2SYan Zheng end--; 16825d4f98a2SYan Zheng ret = try_lock_extent(&BTRFS_I(inode)->io_tree, 1683d0082371SJeff Mahoney key.offset, end); 16845d4f98a2SYan Zheng if (!ret) 16855d4f98a2SYan Zheng continue; 16865d4f98a2SYan Zheng 1687dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), 1688dcdbc059SNikolay Borisov key.offset, end, 1); 16895d4f98a2SYan Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 1690d0082371SJeff Mahoney key.offset, end); 16915d4f98a2SYan Zheng } 16925d4f98a2SYan Zheng } 16935d4f98a2SYan Zheng 16945d4f98a2SYan Zheng ret = get_new_location(rc->data_inode, &new_bytenr, 16955d4f98a2SYan Zheng bytenr, num_bytes); 169683d4cfd4SJosef Bacik if (ret) { 169783d4cfd4SJosef Bacik /* 169883d4cfd4SJosef Bacik * Don't have to abort since we've not changed anything 169983d4cfd4SJosef Bacik * in the file extent yet. 170083d4cfd4SJosef Bacik */ 170183d4cfd4SJosef Bacik break; 17023fd0a558SYan, Zheng } 17035d4f98a2SYan Zheng 17045d4f98a2SYan Zheng btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr); 17055d4f98a2SYan Zheng dirty = 1; 17065d4f98a2SYan Zheng 17075d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 170882fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 170982fa113fSQu Wenruo num_bytes, parent); 171082fa113fSQu Wenruo ref.real_root = root->root_key.objectid; 171182fa113fSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1712b06c4bf5SFilipe Manana key.objectid, key.offset); 171382fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 171483d4cfd4SJosef Bacik if (ret) { 171566642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 171683d4cfd4SJosef Bacik break; 171783d4cfd4SJosef Bacik } 17185d4f98a2SYan Zheng 1719ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, 1720ffd4bb2aSQu Wenruo num_bytes, parent); 1721ffd4bb2aSQu Wenruo ref.real_root = root->root_key.objectid; 1722ffd4bb2aSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1723b06c4bf5SFilipe Manana key.objectid, key.offset); 1724ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 172583d4cfd4SJosef Bacik if (ret) { 172666642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 172783d4cfd4SJosef Bacik break; 172883d4cfd4SJosef Bacik } 17295d4f98a2SYan Zheng } 17305d4f98a2SYan Zheng if (dirty) 17315d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 17323fd0a558SYan, Zheng if (inode) 17333fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 173483d4cfd4SJosef Bacik return ret; 17355d4f98a2SYan Zheng } 17365d4f98a2SYan Zheng 17375d4f98a2SYan Zheng static noinline_for_stack 17385d4f98a2SYan Zheng int memcmp_node_keys(struct extent_buffer *eb, int slot, 17395d4f98a2SYan Zheng struct btrfs_path *path, int level) 17405d4f98a2SYan Zheng { 17415d4f98a2SYan Zheng struct btrfs_disk_key key1; 17425d4f98a2SYan Zheng struct btrfs_disk_key key2; 17435d4f98a2SYan Zheng btrfs_node_key(eb, &key1, slot); 17445d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &key2, path->slots[level]); 17455d4f98a2SYan Zheng return memcmp(&key1, &key2, sizeof(key1)); 17465d4f98a2SYan Zheng } 17475d4f98a2SYan Zheng 17485d4f98a2SYan Zheng /* 17495d4f98a2SYan Zheng * try to replace tree blocks in fs tree with the new blocks 17505d4f98a2SYan Zheng * in reloc tree. tree blocks haven't been modified since the 17515d4f98a2SYan Zheng * reloc tree was create can be replaced. 17525d4f98a2SYan Zheng * 17535d4f98a2SYan Zheng * if a block was replaced, level of the block + 1 is returned. 17545d4f98a2SYan Zheng * if no block got replaced, 0 is returned. if there are other 17555d4f98a2SYan Zheng * errors, a negative error number is returned. 17565d4f98a2SYan Zheng */ 17573fd0a558SYan, Zheng static noinline_for_stack 17583d0174f7SQu Wenruo int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc, 17595d4f98a2SYan Zheng struct btrfs_root *dest, struct btrfs_root *src, 17605d4f98a2SYan Zheng struct btrfs_path *path, struct btrfs_key *next_key, 17615d4f98a2SYan Zheng int lowest_level, int max_level) 17625d4f98a2SYan Zheng { 17630b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = dest->fs_info; 17645d4f98a2SYan Zheng struct extent_buffer *eb; 17655d4f98a2SYan Zheng struct extent_buffer *parent; 176682fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 17675d4f98a2SYan Zheng struct btrfs_key key; 17685d4f98a2SYan Zheng u64 old_bytenr; 17695d4f98a2SYan Zheng u64 new_bytenr; 17705d4f98a2SYan Zheng u64 old_ptr_gen; 17715d4f98a2SYan Zheng u64 new_ptr_gen; 17725d4f98a2SYan Zheng u64 last_snapshot; 17735d4f98a2SYan Zheng u32 blocksize; 17743fd0a558SYan, Zheng int cow = 0; 17755d4f98a2SYan Zheng int level; 17765d4f98a2SYan Zheng int ret; 17775d4f98a2SYan Zheng int slot; 17785d4f98a2SYan Zheng 17795d4f98a2SYan Zheng BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 17805d4f98a2SYan Zheng BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID); 17815d4f98a2SYan Zheng 17825d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&src->root_item); 17833fd0a558SYan, Zheng again: 17845d4f98a2SYan Zheng slot = path->slots[lowest_level]; 17855d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot); 17865d4f98a2SYan Zheng 17875d4f98a2SYan Zheng eb = btrfs_lock_root_node(dest); 17888bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 17895d4f98a2SYan Zheng level = btrfs_header_level(eb); 17905d4f98a2SYan Zheng 17915d4f98a2SYan Zheng if (level < lowest_level) { 17925d4f98a2SYan Zheng btrfs_tree_unlock(eb); 17935d4f98a2SYan Zheng free_extent_buffer(eb); 17945d4f98a2SYan Zheng return 0; 17955d4f98a2SYan Zheng } 17965d4f98a2SYan Zheng 17973fd0a558SYan, Zheng if (cow) { 17985d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb); 17995d4f98a2SYan Zheng BUG_ON(ret); 18003fd0a558SYan, Zheng } 18018bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 18025d4f98a2SYan Zheng 18035d4f98a2SYan Zheng if (next_key) { 18045d4f98a2SYan Zheng next_key->objectid = (u64)-1; 18055d4f98a2SYan Zheng next_key->type = (u8)-1; 18065d4f98a2SYan Zheng next_key->offset = (u64)-1; 18075d4f98a2SYan Zheng } 18085d4f98a2SYan Zheng 18095d4f98a2SYan Zheng parent = eb; 18105d4f98a2SYan Zheng while (1) { 1811581c1760SQu Wenruo struct btrfs_key first_key; 1812581c1760SQu Wenruo 18135d4f98a2SYan Zheng level = btrfs_header_level(parent); 18145d4f98a2SYan Zheng BUG_ON(level < lowest_level); 18155d4f98a2SYan Zheng 18165d4f98a2SYan Zheng ret = btrfs_bin_search(parent, &key, level, &slot); 1817cbca7d59SFilipe Manana if (ret < 0) 1818cbca7d59SFilipe Manana break; 18195d4f98a2SYan Zheng if (ret && slot > 0) 18205d4f98a2SYan Zheng slot--; 18215d4f98a2SYan Zheng 18225d4f98a2SYan Zheng if (next_key && slot + 1 < btrfs_header_nritems(parent)) 18235d4f98a2SYan Zheng btrfs_node_key_to_cpu(parent, next_key, slot + 1); 18245d4f98a2SYan Zheng 18255d4f98a2SYan Zheng old_bytenr = btrfs_node_blockptr(parent, slot); 18260b246afaSJeff Mahoney blocksize = fs_info->nodesize; 18275d4f98a2SYan Zheng old_ptr_gen = btrfs_node_ptr_generation(parent, slot); 182817515f1bSQu Wenruo btrfs_node_key_to_cpu(parent, &first_key, slot); 18295d4f98a2SYan Zheng 18305d4f98a2SYan Zheng if (level <= max_level) { 18315d4f98a2SYan Zheng eb = path->nodes[level]; 18325d4f98a2SYan Zheng new_bytenr = btrfs_node_blockptr(eb, 18335d4f98a2SYan Zheng path->slots[level]); 18345d4f98a2SYan Zheng new_ptr_gen = btrfs_node_ptr_generation(eb, 18355d4f98a2SYan Zheng path->slots[level]); 18365d4f98a2SYan Zheng } else { 18375d4f98a2SYan Zheng new_bytenr = 0; 18385d4f98a2SYan Zheng new_ptr_gen = 0; 18395d4f98a2SYan Zheng } 18405d4f98a2SYan Zheng 1841fae7f21cSDulshani Gunawardhana if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) { 18425d4f98a2SYan Zheng ret = level; 18435d4f98a2SYan Zheng break; 18445d4f98a2SYan Zheng } 18455d4f98a2SYan Zheng 18465d4f98a2SYan Zheng if (new_bytenr == 0 || old_ptr_gen > last_snapshot || 18475d4f98a2SYan Zheng memcmp_node_keys(parent, slot, path, level)) { 18483fd0a558SYan, Zheng if (level <= lowest_level) { 18495d4f98a2SYan Zheng ret = 0; 18505d4f98a2SYan Zheng break; 18515d4f98a2SYan Zheng } 18525d4f98a2SYan Zheng 1853581c1760SQu Wenruo eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen, 1854581c1760SQu Wenruo level - 1, &first_key); 185564c043deSLiu Bo if (IS_ERR(eb)) { 185664c043deSLiu Bo ret = PTR_ERR(eb); 1857264813acSLiu Bo break; 185864c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 185964c043deSLiu Bo ret = -EIO; 1860416bc658SJosef Bacik free_extent_buffer(eb); 1861379cde74SStefan Behrens break; 1862416bc658SJosef Bacik } 18635d4f98a2SYan Zheng btrfs_tree_lock(eb); 18643fd0a558SYan, Zheng if (cow) { 18655d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, parent, 18665d4f98a2SYan Zheng slot, &eb); 18675d4f98a2SYan Zheng BUG_ON(ret); 18685d4f98a2SYan Zheng } 18698bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 18705d4f98a2SYan Zheng 18715d4f98a2SYan Zheng btrfs_tree_unlock(parent); 18725d4f98a2SYan Zheng free_extent_buffer(parent); 18735d4f98a2SYan Zheng 18745d4f98a2SYan Zheng parent = eb; 18755d4f98a2SYan Zheng continue; 18765d4f98a2SYan Zheng } 18775d4f98a2SYan Zheng 18783fd0a558SYan, Zheng if (!cow) { 18793fd0a558SYan, Zheng btrfs_tree_unlock(parent); 18803fd0a558SYan, Zheng free_extent_buffer(parent); 18813fd0a558SYan, Zheng cow = 1; 18823fd0a558SYan, Zheng goto again; 18833fd0a558SYan, Zheng } 18843fd0a558SYan, Zheng 18855d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 18865d4f98a2SYan Zheng path->slots[level]); 1887b3b4aa74SDavid Sterba btrfs_release_path(path); 18885d4f98a2SYan Zheng 18895d4f98a2SYan Zheng path->lowest_level = level; 18905d4f98a2SYan Zheng ret = btrfs_search_slot(trans, src, &key, path, 0, 1); 18915d4f98a2SYan Zheng path->lowest_level = 0; 18925d4f98a2SYan Zheng BUG_ON(ret); 18935d4f98a2SYan Zheng 18945d4f98a2SYan Zheng /* 1895824d8dffSQu Wenruo * Info qgroup to trace both subtrees. 1896824d8dffSQu Wenruo * 1897824d8dffSQu Wenruo * We must trace both trees. 1898824d8dffSQu Wenruo * 1) Tree reloc subtree 1899824d8dffSQu Wenruo * If not traced, we will leak data numbers 1900824d8dffSQu Wenruo * 2) Fs subtree 1901824d8dffSQu Wenruo * If not traced, we will double count old data 1902f616f5cdSQu Wenruo * 1903f616f5cdSQu Wenruo * We don't scan the subtree right now, but only record 1904f616f5cdSQu Wenruo * the swapped tree blocks. 1905f616f5cdSQu Wenruo * The real subtree rescan is delayed until we have new 1906f616f5cdSQu Wenruo * CoW on the subtree root node before transaction commit. 1907824d8dffSQu Wenruo */ 1908370a11b8SQu Wenruo ret = btrfs_qgroup_add_swapped_blocks(trans, dest, 1909370a11b8SQu Wenruo rc->block_group, parent, slot, 1910370a11b8SQu Wenruo path->nodes[level], path->slots[level], 1911370a11b8SQu Wenruo last_snapshot); 1912370a11b8SQu Wenruo if (ret < 0) 1913370a11b8SQu Wenruo break; 1914824d8dffSQu Wenruo /* 19155d4f98a2SYan Zheng * swap blocks in fs tree and reloc tree. 19165d4f98a2SYan Zheng */ 19175d4f98a2SYan Zheng btrfs_set_node_blockptr(parent, slot, new_bytenr); 19185d4f98a2SYan Zheng btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen); 19195d4f98a2SYan Zheng btrfs_mark_buffer_dirty(parent); 19205d4f98a2SYan Zheng 19215d4f98a2SYan Zheng btrfs_set_node_blockptr(path->nodes[level], 19225d4f98a2SYan Zheng path->slots[level], old_bytenr); 19235d4f98a2SYan Zheng btrfs_set_node_ptr_generation(path->nodes[level], 19245d4f98a2SYan Zheng path->slots[level], old_ptr_gen); 19255d4f98a2SYan Zheng btrfs_mark_buffer_dirty(path->nodes[level]); 19265d4f98a2SYan Zheng 192782fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr, 192882fa113fSQu Wenruo blocksize, path->nodes[level]->start); 192982fa113fSQu Wenruo ref.skip_qgroup = true; 193082fa113fSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid); 193182fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 19325d4f98a2SYan Zheng BUG_ON(ret); 193382fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 193482fa113fSQu Wenruo blocksize, 0); 193582fa113fSQu Wenruo ref.skip_qgroup = true; 193682fa113fSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid); 193782fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 19385d4f98a2SYan Zheng BUG_ON(ret); 19395d4f98a2SYan Zheng 1940ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr, 1941ffd4bb2aSQu Wenruo blocksize, path->nodes[level]->start); 1942ffd4bb2aSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid); 1943ffd4bb2aSQu Wenruo ref.skip_qgroup = true; 1944ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 19455d4f98a2SYan Zheng BUG_ON(ret); 19465d4f98a2SYan Zheng 1947ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr, 1948ffd4bb2aSQu Wenruo blocksize, 0); 1949ffd4bb2aSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid); 1950ffd4bb2aSQu Wenruo ref.skip_qgroup = true; 1951ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 19525d4f98a2SYan Zheng BUG_ON(ret); 19535d4f98a2SYan Zheng 19545d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 19555d4f98a2SYan Zheng 19565d4f98a2SYan Zheng ret = level; 19575d4f98a2SYan Zheng break; 19585d4f98a2SYan Zheng } 19595d4f98a2SYan Zheng btrfs_tree_unlock(parent); 19605d4f98a2SYan Zheng free_extent_buffer(parent); 19615d4f98a2SYan Zheng return ret; 19625d4f98a2SYan Zheng } 19635d4f98a2SYan Zheng 19645d4f98a2SYan Zheng /* 19655d4f98a2SYan Zheng * helper to find next relocated block in reloc tree 19665d4f98a2SYan Zheng */ 19675d4f98a2SYan Zheng static noinline_for_stack 19685d4f98a2SYan Zheng int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 19695d4f98a2SYan Zheng int *level) 19705d4f98a2SYan Zheng { 19715d4f98a2SYan Zheng struct extent_buffer *eb; 19725d4f98a2SYan Zheng int i; 19735d4f98a2SYan Zheng u64 last_snapshot; 19745d4f98a2SYan Zheng u32 nritems; 19755d4f98a2SYan Zheng 19765d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 19775d4f98a2SYan Zheng 19785d4f98a2SYan Zheng for (i = 0; i < *level; i++) { 19795d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 19805d4f98a2SYan Zheng path->nodes[i] = NULL; 19815d4f98a2SYan Zheng } 19825d4f98a2SYan Zheng 19835d4f98a2SYan Zheng for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) { 19845d4f98a2SYan Zheng eb = path->nodes[i]; 19855d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 19865d4f98a2SYan Zheng while (path->slots[i] + 1 < nritems) { 19875d4f98a2SYan Zheng path->slots[i]++; 19885d4f98a2SYan Zheng if (btrfs_node_ptr_generation(eb, path->slots[i]) <= 19895d4f98a2SYan Zheng last_snapshot) 19905d4f98a2SYan Zheng continue; 19915d4f98a2SYan Zheng 19925d4f98a2SYan Zheng *level = i; 19935d4f98a2SYan Zheng return 0; 19945d4f98a2SYan Zheng } 19955d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 19965d4f98a2SYan Zheng path->nodes[i] = NULL; 19975d4f98a2SYan Zheng } 19985d4f98a2SYan Zheng return 1; 19995d4f98a2SYan Zheng } 20005d4f98a2SYan Zheng 20015d4f98a2SYan Zheng /* 20025d4f98a2SYan Zheng * walk down reloc tree to find relocated block of lowest level 20035d4f98a2SYan Zheng */ 20045d4f98a2SYan Zheng static noinline_for_stack 20055d4f98a2SYan Zheng int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 20065d4f98a2SYan Zheng int *level) 20075d4f98a2SYan Zheng { 20082ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 20095d4f98a2SYan Zheng struct extent_buffer *eb = NULL; 20105d4f98a2SYan Zheng int i; 20115d4f98a2SYan Zheng u64 bytenr; 20125d4f98a2SYan Zheng u64 ptr_gen = 0; 20135d4f98a2SYan Zheng u64 last_snapshot; 20145d4f98a2SYan Zheng u32 nritems; 20155d4f98a2SYan Zheng 20165d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 20175d4f98a2SYan Zheng 20185d4f98a2SYan Zheng for (i = *level; i > 0; i--) { 2019581c1760SQu Wenruo struct btrfs_key first_key; 2020581c1760SQu Wenruo 20215d4f98a2SYan Zheng eb = path->nodes[i]; 20225d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 20235d4f98a2SYan Zheng while (path->slots[i] < nritems) { 20245d4f98a2SYan Zheng ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]); 20255d4f98a2SYan Zheng if (ptr_gen > last_snapshot) 20265d4f98a2SYan Zheng break; 20275d4f98a2SYan Zheng path->slots[i]++; 20285d4f98a2SYan Zheng } 20295d4f98a2SYan Zheng if (path->slots[i] >= nritems) { 20305d4f98a2SYan Zheng if (i == *level) 20315d4f98a2SYan Zheng break; 20325d4f98a2SYan Zheng *level = i + 1; 20335d4f98a2SYan Zheng return 0; 20345d4f98a2SYan Zheng } 20355d4f98a2SYan Zheng if (i == 1) { 20365d4f98a2SYan Zheng *level = i; 20375d4f98a2SYan Zheng return 0; 20385d4f98a2SYan Zheng } 20395d4f98a2SYan Zheng 20405d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(eb, path->slots[i]); 2041581c1760SQu Wenruo btrfs_node_key_to_cpu(eb, &first_key, path->slots[i]); 2042581c1760SQu Wenruo eb = read_tree_block(fs_info, bytenr, ptr_gen, i - 1, 2043581c1760SQu Wenruo &first_key); 204464c043deSLiu Bo if (IS_ERR(eb)) { 204564c043deSLiu Bo return PTR_ERR(eb); 204664c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2047416bc658SJosef Bacik free_extent_buffer(eb); 2048416bc658SJosef Bacik return -EIO; 2049416bc658SJosef Bacik } 20505d4f98a2SYan Zheng BUG_ON(btrfs_header_level(eb) != i - 1); 20515d4f98a2SYan Zheng path->nodes[i - 1] = eb; 20525d4f98a2SYan Zheng path->slots[i - 1] = 0; 20535d4f98a2SYan Zheng } 20545d4f98a2SYan Zheng return 1; 20555d4f98a2SYan Zheng } 20565d4f98a2SYan Zheng 20575d4f98a2SYan Zheng /* 20585d4f98a2SYan Zheng * invalidate extent cache for file extents whose key in range of 20595d4f98a2SYan Zheng * [min_key, max_key) 20605d4f98a2SYan Zheng */ 20615d4f98a2SYan Zheng static int invalidate_extent_cache(struct btrfs_root *root, 20625d4f98a2SYan Zheng struct btrfs_key *min_key, 20635d4f98a2SYan Zheng struct btrfs_key *max_key) 20645d4f98a2SYan Zheng { 20650b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 20665d4f98a2SYan Zheng struct inode *inode = NULL; 20675d4f98a2SYan Zheng u64 objectid; 20685d4f98a2SYan Zheng u64 start, end; 206933345d01SLi Zefan u64 ino; 20705d4f98a2SYan Zheng 20715d4f98a2SYan Zheng objectid = min_key->objectid; 20725d4f98a2SYan Zheng while (1) { 20735d4f98a2SYan Zheng cond_resched(); 20745d4f98a2SYan Zheng iput(inode); 20755d4f98a2SYan Zheng 20765d4f98a2SYan Zheng if (objectid > max_key->objectid) 20775d4f98a2SYan Zheng break; 20785d4f98a2SYan Zheng 20795d4f98a2SYan Zheng inode = find_next_inode(root, objectid); 20805d4f98a2SYan Zheng if (!inode) 20815d4f98a2SYan Zheng break; 20824a0cc7caSNikolay Borisov ino = btrfs_ino(BTRFS_I(inode)); 20835d4f98a2SYan Zheng 208433345d01SLi Zefan if (ino > max_key->objectid) { 20855d4f98a2SYan Zheng iput(inode); 20865d4f98a2SYan Zheng break; 20875d4f98a2SYan Zheng } 20885d4f98a2SYan Zheng 208933345d01SLi Zefan objectid = ino + 1; 20905d4f98a2SYan Zheng if (!S_ISREG(inode->i_mode)) 20915d4f98a2SYan Zheng continue; 20925d4f98a2SYan Zheng 209333345d01SLi Zefan if (unlikely(min_key->objectid == ino)) { 20945d4f98a2SYan Zheng if (min_key->type > BTRFS_EXTENT_DATA_KEY) 20955d4f98a2SYan Zheng continue; 20965d4f98a2SYan Zheng if (min_key->type < BTRFS_EXTENT_DATA_KEY) 20975d4f98a2SYan Zheng start = 0; 20985d4f98a2SYan Zheng else { 20995d4f98a2SYan Zheng start = min_key->offset; 21000b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize)); 21015d4f98a2SYan Zheng } 21025d4f98a2SYan Zheng } else { 21035d4f98a2SYan Zheng start = 0; 21045d4f98a2SYan Zheng } 21055d4f98a2SYan Zheng 210633345d01SLi Zefan if (unlikely(max_key->objectid == ino)) { 21075d4f98a2SYan Zheng if (max_key->type < BTRFS_EXTENT_DATA_KEY) 21085d4f98a2SYan Zheng continue; 21095d4f98a2SYan Zheng if (max_key->type > BTRFS_EXTENT_DATA_KEY) { 21105d4f98a2SYan Zheng end = (u64)-1; 21115d4f98a2SYan Zheng } else { 21125d4f98a2SYan Zheng if (max_key->offset == 0) 21135d4f98a2SYan Zheng continue; 21145d4f98a2SYan Zheng end = max_key->offset; 21150b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 21165d4f98a2SYan Zheng end--; 21175d4f98a2SYan Zheng } 21185d4f98a2SYan Zheng } else { 21195d4f98a2SYan Zheng end = (u64)-1; 21205d4f98a2SYan Zheng } 21215d4f98a2SYan Zheng 21225d4f98a2SYan Zheng /* the lock_extent waits for readpage to complete */ 2123d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 2124dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1); 2125d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 21265d4f98a2SYan Zheng } 21275d4f98a2SYan Zheng return 0; 21285d4f98a2SYan Zheng } 21295d4f98a2SYan Zheng 21305d4f98a2SYan Zheng static int find_next_key(struct btrfs_path *path, int level, 21315d4f98a2SYan Zheng struct btrfs_key *key) 21325d4f98a2SYan Zheng 21335d4f98a2SYan Zheng { 21345d4f98a2SYan Zheng while (level < BTRFS_MAX_LEVEL) { 21355d4f98a2SYan Zheng if (!path->nodes[level]) 21365d4f98a2SYan Zheng break; 21375d4f98a2SYan Zheng if (path->slots[level] + 1 < 21385d4f98a2SYan Zheng btrfs_header_nritems(path->nodes[level])) { 21395d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], key, 21405d4f98a2SYan Zheng path->slots[level] + 1); 21415d4f98a2SYan Zheng return 0; 21425d4f98a2SYan Zheng } 21435d4f98a2SYan Zheng level++; 21445d4f98a2SYan Zheng } 21455d4f98a2SYan Zheng return 1; 21465d4f98a2SYan Zheng } 21475d4f98a2SYan Zheng 21485d4f98a2SYan Zheng /* 2149d2311e69SQu Wenruo * Insert current subvolume into reloc_control::dirty_subvol_roots 2150d2311e69SQu Wenruo */ 2151d2311e69SQu Wenruo static void insert_dirty_subvol(struct btrfs_trans_handle *trans, 2152d2311e69SQu Wenruo struct reloc_control *rc, 2153d2311e69SQu Wenruo struct btrfs_root *root) 2154d2311e69SQu Wenruo { 2155d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 2156d2311e69SQu Wenruo struct btrfs_root_item *reloc_root_item; 2157d2311e69SQu Wenruo 2158d2311e69SQu Wenruo /* @root must be a subvolume tree root with a valid reloc tree */ 2159d2311e69SQu Wenruo ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 2160d2311e69SQu Wenruo ASSERT(reloc_root); 2161d2311e69SQu Wenruo 2162d2311e69SQu Wenruo reloc_root_item = &reloc_root->root_item; 2163d2311e69SQu Wenruo memset(&reloc_root_item->drop_progress, 0, 2164d2311e69SQu Wenruo sizeof(reloc_root_item->drop_progress)); 2165d2311e69SQu Wenruo reloc_root_item->drop_level = 0; 2166d2311e69SQu Wenruo btrfs_set_root_refs(reloc_root_item, 0); 2167d2311e69SQu Wenruo btrfs_update_reloc_root(trans, root); 2168d2311e69SQu Wenruo 2169d2311e69SQu Wenruo if (list_empty(&root->reloc_dirty_list)) { 2170d2311e69SQu Wenruo btrfs_grab_fs_root(root); 2171d2311e69SQu Wenruo list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots); 2172d2311e69SQu Wenruo } 2173d2311e69SQu Wenruo } 2174d2311e69SQu Wenruo 2175d2311e69SQu Wenruo static int clean_dirty_subvols(struct reloc_control *rc) 2176d2311e69SQu Wenruo { 2177d2311e69SQu Wenruo struct btrfs_root *root; 2178d2311e69SQu Wenruo struct btrfs_root *next; 2179d2311e69SQu Wenruo int ret = 0; 218030d40577SQu Wenruo int ret2; 2181d2311e69SQu Wenruo 2182d2311e69SQu Wenruo list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots, 2183d2311e69SQu Wenruo reloc_dirty_list) { 218430d40577SQu Wenruo if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 218530d40577SQu Wenruo /* Merged subvolume, cleanup its reloc root */ 2186d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 2187d2311e69SQu Wenruo 2188d2311e69SQu Wenruo clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 2189d2311e69SQu Wenruo list_del_init(&root->reloc_dirty_list); 2190d2311e69SQu Wenruo root->reloc_root = NULL; 2191d2311e69SQu Wenruo if (reloc_root) { 2192d2311e69SQu Wenruo 2193d2311e69SQu Wenruo ret2 = btrfs_drop_snapshot(reloc_root, NULL, 0, 1); 2194d2311e69SQu Wenruo if (ret2 < 0 && !ret) 2195d2311e69SQu Wenruo ret = ret2; 2196d2311e69SQu Wenruo } 2197d2311e69SQu Wenruo btrfs_put_fs_root(root); 219830d40577SQu Wenruo } else { 219930d40577SQu Wenruo /* Orphan reloc tree, just clean it up */ 220030d40577SQu Wenruo ret2 = btrfs_drop_snapshot(root, NULL, 0, 1); 220130d40577SQu Wenruo if (ret2 < 0 && !ret) 220230d40577SQu Wenruo ret = ret2; 220330d40577SQu Wenruo } 2204d2311e69SQu Wenruo } 2205d2311e69SQu Wenruo return ret; 2206d2311e69SQu Wenruo } 2207d2311e69SQu Wenruo 2208d2311e69SQu Wenruo /* 22095d4f98a2SYan Zheng * merge the relocated tree blocks in reloc tree with corresponding 22105d4f98a2SYan Zheng * fs tree. 22115d4f98a2SYan Zheng */ 22125d4f98a2SYan Zheng static noinline_for_stack int merge_reloc_root(struct reloc_control *rc, 22135d4f98a2SYan Zheng struct btrfs_root *root) 22145d4f98a2SYan Zheng { 22150b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 22165d4f98a2SYan Zheng struct btrfs_key key; 22175d4f98a2SYan Zheng struct btrfs_key next_key; 22189e6a0c52SJosef Bacik struct btrfs_trans_handle *trans = NULL; 22195d4f98a2SYan Zheng struct btrfs_root *reloc_root; 22205d4f98a2SYan Zheng struct btrfs_root_item *root_item; 22215d4f98a2SYan Zheng struct btrfs_path *path; 22223fd0a558SYan, Zheng struct extent_buffer *leaf; 22235d4f98a2SYan Zheng int level; 22245d4f98a2SYan Zheng int max_level; 22255d4f98a2SYan Zheng int replaced = 0; 22265d4f98a2SYan Zheng int ret; 22275d4f98a2SYan Zheng int err = 0; 22283fd0a558SYan, Zheng u32 min_reserved; 22295d4f98a2SYan Zheng 22305d4f98a2SYan Zheng path = btrfs_alloc_path(); 22315d4f98a2SYan Zheng if (!path) 22325d4f98a2SYan Zheng return -ENOMEM; 2233e4058b54SDavid Sterba path->reada = READA_FORWARD; 22345d4f98a2SYan Zheng 22355d4f98a2SYan Zheng reloc_root = root->reloc_root; 22365d4f98a2SYan Zheng root_item = &reloc_root->root_item; 22375d4f98a2SYan Zheng 22385d4f98a2SYan Zheng if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { 22395d4f98a2SYan Zheng level = btrfs_root_level(root_item); 22405d4f98a2SYan Zheng extent_buffer_get(reloc_root->node); 22415d4f98a2SYan Zheng path->nodes[level] = reloc_root->node; 22425d4f98a2SYan Zheng path->slots[level] = 0; 22435d4f98a2SYan Zheng } else { 22445d4f98a2SYan Zheng btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); 22455d4f98a2SYan Zheng 22465d4f98a2SYan Zheng level = root_item->drop_level; 22475d4f98a2SYan Zheng BUG_ON(level == 0); 22485d4f98a2SYan Zheng path->lowest_level = level; 22495d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0); 225033c66f43SYan Zheng path->lowest_level = 0; 22515d4f98a2SYan Zheng if (ret < 0) { 22525d4f98a2SYan Zheng btrfs_free_path(path); 22535d4f98a2SYan Zheng return ret; 22545d4f98a2SYan Zheng } 22555d4f98a2SYan Zheng 22565d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &next_key, 22575d4f98a2SYan Zheng path->slots[level]); 22585d4f98a2SYan Zheng WARN_ON(memcmp(&key, &next_key, sizeof(key))); 22595d4f98a2SYan Zheng 22605d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 22615d4f98a2SYan Zheng } 22625d4f98a2SYan Zheng 22630b246afaSJeff Mahoney min_reserved = fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 22645d4f98a2SYan Zheng memset(&next_key, 0, sizeof(next_key)); 22655d4f98a2SYan Zheng 22665d4f98a2SYan Zheng while (1) { 226708e007d2SMiao Xie ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved, 226808e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 22693fd0a558SYan, Zheng if (ret) { 22709e6a0c52SJosef Bacik err = ret; 22719e6a0c52SJosef Bacik goto out; 22723fd0a558SYan, Zheng } 22739e6a0c52SJosef Bacik trans = btrfs_start_transaction(root, 0); 22749e6a0c52SJosef Bacik if (IS_ERR(trans)) { 22759e6a0c52SJosef Bacik err = PTR_ERR(trans); 22769e6a0c52SJosef Bacik trans = NULL; 22779e6a0c52SJosef Bacik goto out; 22789e6a0c52SJosef Bacik } 22799e6a0c52SJosef Bacik trans->block_rsv = rc->block_rsv; 22803fd0a558SYan, Zheng 22813fd0a558SYan, Zheng replaced = 0; 22825d4f98a2SYan Zheng max_level = level; 22835d4f98a2SYan Zheng 22845d4f98a2SYan Zheng ret = walk_down_reloc_tree(reloc_root, path, &level); 22855d4f98a2SYan Zheng if (ret < 0) { 22865d4f98a2SYan Zheng err = ret; 22875d4f98a2SYan Zheng goto out; 22885d4f98a2SYan Zheng } 22895d4f98a2SYan Zheng if (ret > 0) 22905d4f98a2SYan Zheng break; 22915d4f98a2SYan Zheng 22925d4f98a2SYan Zheng if (!find_next_key(path, level, &key) && 22935d4f98a2SYan Zheng btrfs_comp_cpu_keys(&next_key, &key) >= 0) { 22945d4f98a2SYan Zheng ret = 0; 22955d4f98a2SYan Zheng } else { 22963d0174f7SQu Wenruo ret = replace_path(trans, rc, root, reloc_root, path, 22973fd0a558SYan, Zheng &next_key, level, max_level); 22985d4f98a2SYan Zheng } 22995d4f98a2SYan Zheng if (ret < 0) { 23005d4f98a2SYan Zheng err = ret; 23015d4f98a2SYan Zheng goto out; 23025d4f98a2SYan Zheng } 23035d4f98a2SYan Zheng 23045d4f98a2SYan Zheng if (ret > 0) { 23055d4f98a2SYan Zheng level = ret; 23065d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 23075d4f98a2SYan Zheng path->slots[level]); 23085d4f98a2SYan Zheng replaced = 1; 23095d4f98a2SYan Zheng } 23105d4f98a2SYan Zheng 23115d4f98a2SYan Zheng ret = walk_up_reloc_tree(reloc_root, path, &level); 23125d4f98a2SYan Zheng if (ret > 0) 23135d4f98a2SYan Zheng break; 23145d4f98a2SYan Zheng 23155d4f98a2SYan Zheng BUG_ON(level == 0); 23165d4f98a2SYan Zheng /* 23175d4f98a2SYan Zheng * save the merging progress in the drop_progress. 23185d4f98a2SYan Zheng * this is OK since root refs == 1 in this case. 23195d4f98a2SYan Zheng */ 23205d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &root_item->drop_progress, 23215d4f98a2SYan Zheng path->slots[level]); 23225d4f98a2SYan Zheng root_item->drop_level = level; 23235d4f98a2SYan Zheng 23243a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 23259e6a0c52SJosef Bacik trans = NULL; 23265d4f98a2SYan Zheng 23272ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 23285d4f98a2SYan Zheng 23295d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 23305d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 23315d4f98a2SYan Zheng } 23325d4f98a2SYan Zheng 23335d4f98a2SYan Zheng /* 23345d4f98a2SYan Zheng * handle the case only one block in the fs tree need to be 23355d4f98a2SYan Zheng * relocated and the block is tree root. 23365d4f98a2SYan Zheng */ 23375d4f98a2SYan Zheng leaf = btrfs_lock_root_node(root); 23385d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf); 23395d4f98a2SYan Zheng btrfs_tree_unlock(leaf); 23405d4f98a2SYan Zheng free_extent_buffer(leaf); 23415d4f98a2SYan Zheng if (ret < 0) 23425d4f98a2SYan Zheng err = ret; 23435d4f98a2SYan Zheng out: 23445d4f98a2SYan Zheng btrfs_free_path(path); 23455d4f98a2SYan Zheng 2346d2311e69SQu Wenruo if (err == 0) 2347d2311e69SQu Wenruo insert_dirty_subvol(trans, rc, root); 23485d4f98a2SYan Zheng 23499e6a0c52SJosef Bacik if (trans) 23503a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 23515d4f98a2SYan Zheng 23522ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 23535d4f98a2SYan Zheng 23545d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 23555d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 23565d4f98a2SYan Zheng 23575d4f98a2SYan Zheng return err; 23585d4f98a2SYan Zheng } 23595d4f98a2SYan Zheng 23603fd0a558SYan, Zheng static noinline_for_stack 23613fd0a558SYan, Zheng int prepare_to_merge(struct reloc_control *rc, int err) 23625d4f98a2SYan Zheng { 23633fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 23640b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 23653fd0a558SYan, Zheng struct btrfs_root *reloc_root; 23665d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 23673fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 23683fd0a558SYan, Zheng u64 num_bytes = 0; 23693fd0a558SYan, Zheng int ret; 23703fd0a558SYan, Zheng 23710b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 23720b246afaSJeff Mahoney rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 23733fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated * 2; 23740b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 23757585717fSChris Mason 23763fd0a558SYan, Zheng again: 23773fd0a558SYan, Zheng if (!err) { 23783fd0a558SYan, Zheng num_bytes = rc->merging_rsv_size; 237908e007d2SMiao Xie ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes, 238008e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 23813fd0a558SYan, Zheng if (ret) 23823fd0a558SYan, Zheng err = ret; 23833fd0a558SYan, Zheng } 23843fd0a558SYan, Zheng 23857a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 23863612b495STsutomu Itoh if (IS_ERR(trans)) { 23873612b495STsutomu Itoh if (!err) 23882ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 23892ff7e61eSJeff Mahoney num_bytes); 23903612b495STsutomu Itoh return PTR_ERR(trans); 23913612b495STsutomu Itoh } 23923fd0a558SYan, Zheng 23933fd0a558SYan, Zheng if (!err) { 23943fd0a558SYan, Zheng if (num_bytes != rc->merging_rsv_size) { 23953a45bb20SJeff Mahoney btrfs_end_transaction(trans); 23962ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 23972ff7e61eSJeff Mahoney num_bytes); 23983fd0a558SYan, Zheng goto again; 23993fd0a558SYan, Zheng } 24003fd0a558SYan, Zheng } 24013fd0a558SYan, Zheng 24023fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 24033fd0a558SYan, Zheng 24043fd0a558SYan, Zheng while (!list_empty(&rc->reloc_roots)) { 24053fd0a558SYan, Zheng reloc_root = list_entry(rc->reloc_roots.next, 24063fd0a558SYan, Zheng struct btrfs_root, root_list); 24073fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 24083fd0a558SYan, Zheng 24090b246afaSJeff Mahoney root = read_fs_root(fs_info, reloc_root->root_key.offset); 24103fd0a558SYan, Zheng BUG_ON(IS_ERR(root)); 24113fd0a558SYan, Zheng BUG_ON(root->reloc_root != reloc_root); 24123fd0a558SYan, Zheng 24133fd0a558SYan, Zheng /* 24143fd0a558SYan, Zheng * set reference count to 1, so btrfs_recover_relocation 24153fd0a558SYan, Zheng * knows it should resumes merging 24163fd0a558SYan, Zheng */ 24173fd0a558SYan, Zheng if (!err) 24183fd0a558SYan, Zheng btrfs_set_root_refs(&reloc_root->root_item, 1); 24193fd0a558SYan, Zheng btrfs_update_reloc_root(trans, root); 24203fd0a558SYan, Zheng 24213fd0a558SYan, Zheng list_add(&reloc_root->root_list, &reloc_roots); 24223fd0a558SYan, Zheng } 24233fd0a558SYan, Zheng 24243fd0a558SYan, Zheng list_splice(&reloc_roots, &rc->reloc_roots); 24253fd0a558SYan, Zheng 24263fd0a558SYan, Zheng if (!err) 24273a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 24283fd0a558SYan, Zheng else 24293a45bb20SJeff Mahoney btrfs_end_transaction(trans); 24303fd0a558SYan, Zheng return err; 24313fd0a558SYan, Zheng } 24323fd0a558SYan, Zheng 24333fd0a558SYan, Zheng static noinline_for_stack 2434aca1bba6SLiu Bo void free_reloc_roots(struct list_head *list) 2435aca1bba6SLiu Bo { 2436aca1bba6SLiu Bo struct btrfs_root *reloc_root; 2437aca1bba6SLiu Bo 2438aca1bba6SLiu Bo while (!list_empty(list)) { 2439aca1bba6SLiu Bo reloc_root = list_entry(list->next, struct btrfs_root, 2440aca1bba6SLiu Bo root_list); 2441bb166d72SNaohiro Aota __del_reloc_root(reloc_root); 24426bdf131fSJosef Bacik free_extent_buffer(reloc_root->node); 24436bdf131fSJosef Bacik free_extent_buffer(reloc_root->commit_root); 24446bdf131fSJosef Bacik reloc_root->node = NULL; 24456bdf131fSJosef Bacik reloc_root->commit_root = NULL; 2446aca1bba6SLiu Bo } 2447aca1bba6SLiu Bo } 2448aca1bba6SLiu Bo 2449aca1bba6SLiu Bo static noinline_for_stack 245094404e82SDavid Sterba void merge_reloc_roots(struct reloc_control *rc) 24513fd0a558SYan, Zheng { 24520b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 24535d4f98a2SYan Zheng struct btrfs_root *root; 24545d4f98a2SYan Zheng struct btrfs_root *reloc_root; 24553fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 24563fd0a558SYan, Zheng int found = 0; 2457aca1bba6SLiu Bo int ret = 0; 24583fd0a558SYan, Zheng again: 24593fd0a558SYan, Zheng root = rc->extent_root; 24607585717fSChris Mason 24617585717fSChris Mason /* 24627585717fSChris Mason * this serializes us with btrfs_record_root_in_transaction, 24637585717fSChris Mason * we have to make sure nobody is in the middle of 24647585717fSChris Mason * adding their roots to the list while we are 24657585717fSChris Mason * doing this splice 24667585717fSChris Mason */ 24670b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 24683fd0a558SYan, Zheng list_splice_init(&rc->reloc_roots, &reloc_roots); 24690b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 24705d4f98a2SYan Zheng 24713fd0a558SYan, Zheng while (!list_empty(&reloc_roots)) { 24723fd0a558SYan, Zheng found = 1; 24733fd0a558SYan, Zheng reloc_root = list_entry(reloc_roots.next, 24743fd0a558SYan, Zheng struct btrfs_root, root_list); 24755d4f98a2SYan Zheng 24765d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 24770b246afaSJeff Mahoney root = read_fs_root(fs_info, 24785d4f98a2SYan Zheng reloc_root->root_key.offset); 24795d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 24805d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 24815d4f98a2SYan Zheng 24823fd0a558SYan, Zheng ret = merge_reloc_root(rc, root); 2483b37b39cdSJosef Bacik if (ret) { 248425e293c2SWang Shilong if (list_empty(&reloc_root->root_list)) 248525e293c2SWang Shilong list_add_tail(&reloc_root->root_list, 248625e293c2SWang Shilong &reloc_roots); 2487aca1bba6SLiu Bo goto out; 2488b37b39cdSJosef Bacik } 24893fd0a558SYan, Zheng } else { 24903fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 249130d40577SQu Wenruo /* Don't forget to queue this reloc root for cleanup */ 249230d40577SQu Wenruo list_add_tail(&reloc_root->reloc_dirty_list, 249330d40577SQu Wenruo &rc->dirty_subvol_roots); 24943fd0a558SYan, Zheng } 24955d4f98a2SYan Zheng } 24965d4f98a2SYan Zheng 24973fd0a558SYan, Zheng if (found) { 24983fd0a558SYan, Zheng found = 0; 24993fd0a558SYan, Zheng goto again; 25005d4f98a2SYan Zheng } 2501aca1bba6SLiu Bo out: 2502aca1bba6SLiu Bo if (ret) { 25030b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 2504aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 2505aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 2506467bb1d2SWang Shilong 2507467bb1d2SWang Shilong /* new reloc root may be added */ 25080b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 2509467bb1d2SWang Shilong list_splice_init(&rc->reloc_roots, &reloc_roots); 25100b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2511467bb1d2SWang Shilong if (!list_empty(&reloc_roots)) 2512467bb1d2SWang Shilong free_reloc_roots(&reloc_roots); 2513aca1bba6SLiu Bo } 2514aca1bba6SLiu Bo 25155d4f98a2SYan Zheng BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); 25165d4f98a2SYan Zheng } 25175d4f98a2SYan Zheng 25185d4f98a2SYan Zheng static void free_block_list(struct rb_root *blocks) 25195d4f98a2SYan Zheng { 25205d4f98a2SYan Zheng struct tree_block *block; 25215d4f98a2SYan Zheng struct rb_node *rb_node; 25225d4f98a2SYan Zheng while ((rb_node = rb_first(blocks))) { 25235d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 25245d4f98a2SYan Zheng rb_erase(rb_node, blocks); 25255d4f98a2SYan Zheng kfree(block); 25265d4f98a2SYan Zheng } 25275d4f98a2SYan Zheng } 25285d4f98a2SYan Zheng 25295d4f98a2SYan Zheng static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans, 25305d4f98a2SYan Zheng struct btrfs_root *reloc_root) 25315d4f98a2SYan Zheng { 25320b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = reloc_root->fs_info; 25335d4f98a2SYan Zheng struct btrfs_root *root; 25345d4f98a2SYan Zheng 25355d4f98a2SYan Zheng if (reloc_root->last_trans == trans->transid) 25365d4f98a2SYan Zheng return 0; 25375d4f98a2SYan Zheng 25380b246afaSJeff Mahoney root = read_fs_root(fs_info, reloc_root->root_key.offset); 25395d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 25405d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 25415d4f98a2SYan Zheng 25425d4f98a2SYan Zheng return btrfs_record_root_in_trans(trans, root); 25435d4f98a2SYan Zheng } 25445d4f98a2SYan Zheng 25453fd0a558SYan, Zheng static noinline_for_stack 25463fd0a558SYan, Zheng struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, 25473fd0a558SYan, Zheng struct reloc_control *rc, 25485d4f98a2SYan Zheng struct backref_node *node, 2549dc4103f9SWang Shilong struct backref_edge *edges[]) 25505d4f98a2SYan Zheng { 25515d4f98a2SYan Zheng struct backref_node *next; 25525d4f98a2SYan Zheng struct btrfs_root *root; 25533fd0a558SYan, Zheng int index = 0; 25543fd0a558SYan, Zheng 25555d4f98a2SYan Zheng next = node; 25565d4f98a2SYan Zheng while (1) { 25575d4f98a2SYan Zheng cond_resched(); 25585d4f98a2SYan Zheng next = walk_up_backref(next, edges, &index); 25595d4f98a2SYan Zheng root = next->root; 25603fd0a558SYan, Zheng BUG_ON(!root); 256127cdeb70SMiao Xie BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state)); 25625d4f98a2SYan Zheng 25635d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 25645d4f98a2SYan Zheng record_reloc_root_in_trans(trans, root); 25655d4f98a2SYan Zheng break; 25665d4f98a2SYan Zheng } 25675d4f98a2SYan Zheng 25685d4f98a2SYan Zheng btrfs_record_root_in_trans(trans, root); 25693fd0a558SYan, Zheng root = root->reloc_root; 25703fd0a558SYan, Zheng 25713fd0a558SYan, Zheng if (next->new_bytenr != root->node->start) { 25723fd0a558SYan, Zheng BUG_ON(next->new_bytenr); 25733fd0a558SYan, Zheng BUG_ON(!list_empty(&next->list)); 25743fd0a558SYan, Zheng next->new_bytenr = root->node->start; 25753fd0a558SYan, Zheng next->root = root; 25763fd0a558SYan, Zheng list_add_tail(&next->list, 25773fd0a558SYan, Zheng &rc->backref_cache.changed); 25783fd0a558SYan, Zheng __mark_block_processed(rc, next); 25795d4f98a2SYan Zheng break; 25805d4f98a2SYan Zheng } 25815d4f98a2SYan Zheng 25823fd0a558SYan, Zheng WARN_ON(1); 25835d4f98a2SYan Zheng root = NULL; 25845d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 25855d4f98a2SYan Zheng if (!next || next->level <= node->level) 25865d4f98a2SYan Zheng break; 25875d4f98a2SYan Zheng } 25883fd0a558SYan, Zheng if (!root) 25893fd0a558SYan, Zheng return NULL; 25905d4f98a2SYan Zheng 25913fd0a558SYan, Zheng next = node; 25923fd0a558SYan, Zheng /* setup backref node path for btrfs_reloc_cow_block */ 25933fd0a558SYan, Zheng while (1) { 25943fd0a558SYan, Zheng rc->backref_cache.path[next->level] = next; 25953fd0a558SYan, Zheng if (--index < 0) 25963fd0a558SYan, Zheng break; 25973fd0a558SYan, Zheng next = edges[index]->node[UPPER]; 25983fd0a558SYan, Zheng } 25995d4f98a2SYan Zheng return root; 26005d4f98a2SYan Zheng } 26015d4f98a2SYan Zheng 26023fd0a558SYan, Zheng /* 26033fd0a558SYan, Zheng * select a tree root for relocation. return NULL if the block 26043fd0a558SYan, Zheng * is reference counted. we should use do_relocation() in this 26053fd0a558SYan, Zheng * case. return a tree root pointer if the block isn't reference 26063fd0a558SYan, Zheng * counted. return -ENOENT if the block is root of reloc tree. 26073fd0a558SYan, Zheng */ 26085d4f98a2SYan Zheng static noinline_for_stack 2609147d256eSZhaolei struct btrfs_root *select_one_root(struct backref_node *node) 26105d4f98a2SYan Zheng { 26113fd0a558SYan, Zheng struct backref_node *next; 26123fd0a558SYan, Zheng struct btrfs_root *root; 26133fd0a558SYan, Zheng struct btrfs_root *fs_root = NULL; 26145d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 26153fd0a558SYan, Zheng int index = 0; 26163fd0a558SYan, Zheng 26173fd0a558SYan, Zheng next = node; 26183fd0a558SYan, Zheng while (1) { 26193fd0a558SYan, Zheng cond_resched(); 26203fd0a558SYan, Zheng next = walk_up_backref(next, edges, &index); 26213fd0a558SYan, Zheng root = next->root; 26223fd0a558SYan, Zheng BUG_ON(!root); 26233fd0a558SYan, Zheng 262425985edcSLucas De Marchi /* no other choice for non-references counted tree */ 262527cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 26263fd0a558SYan, Zheng return root; 26273fd0a558SYan, Zheng 26283fd0a558SYan, Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) 26293fd0a558SYan, Zheng fs_root = root; 26303fd0a558SYan, Zheng 26313fd0a558SYan, Zheng if (next != node) 26323fd0a558SYan, Zheng return NULL; 26333fd0a558SYan, Zheng 26343fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 26353fd0a558SYan, Zheng if (!next || next->level <= node->level) 26363fd0a558SYan, Zheng break; 26373fd0a558SYan, Zheng } 26383fd0a558SYan, Zheng 26393fd0a558SYan, Zheng if (!fs_root) 26403fd0a558SYan, Zheng return ERR_PTR(-ENOENT); 26413fd0a558SYan, Zheng return fs_root; 26425d4f98a2SYan Zheng } 26435d4f98a2SYan Zheng 26445d4f98a2SYan Zheng static noinline_for_stack 26453fd0a558SYan, Zheng u64 calcu_metadata_size(struct reloc_control *rc, 26463fd0a558SYan, Zheng struct backref_node *node, int reserve) 26475d4f98a2SYan Zheng { 26480b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 26493fd0a558SYan, Zheng struct backref_node *next = node; 26503fd0a558SYan, Zheng struct backref_edge *edge; 26513fd0a558SYan, Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 26523fd0a558SYan, Zheng u64 num_bytes = 0; 26533fd0a558SYan, Zheng int index = 0; 26545d4f98a2SYan Zheng 26553fd0a558SYan, Zheng BUG_ON(reserve && node->processed); 26563fd0a558SYan, Zheng 26573fd0a558SYan, Zheng while (next) { 26583fd0a558SYan, Zheng cond_resched(); 26595d4f98a2SYan Zheng while (1) { 26603fd0a558SYan, Zheng if (next->processed && (reserve || next != node)) 26615d4f98a2SYan Zheng break; 26625d4f98a2SYan Zheng 26630b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 26643fd0a558SYan, Zheng 26653fd0a558SYan, Zheng if (list_empty(&next->upper)) 26663fd0a558SYan, Zheng break; 26673fd0a558SYan, Zheng 26683fd0a558SYan, Zheng edge = list_entry(next->upper.next, 26693fd0a558SYan, Zheng struct backref_edge, list[LOWER]); 26703fd0a558SYan, Zheng edges[index++] = edge; 26713fd0a558SYan, Zheng next = edge->node[UPPER]; 26725d4f98a2SYan Zheng } 26733fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 26743fd0a558SYan, Zheng } 26753fd0a558SYan, Zheng return num_bytes; 26763fd0a558SYan, Zheng } 26773fd0a558SYan, Zheng 26783fd0a558SYan, Zheng static int reserve_metadata_space(struct btrfs_trans_handle *trans, 26793fd0a558SYan, Zheng struct reloc_control *rc, 26803fd0a558SYan, Zheng struct backref_node *node) 26813fd0a558SYan, Zheng { 26823fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 2683da17066cSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 26843fd0a558SYan, Zheng u64 num_bytes; 26853fd0a558SYan, Zheng int ret; 26860647bf56SWang Shilong u64 tmp; 26873fd0a558SYan, Zheng 26883fd0a558SYan, Zheng num_bytes = calcu_metadata_size(rc, node, 1) * 2; 26893fd0a558SYan, Zheng 26903fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 26910647bf56SWang Shilong rc->reserved_bytes += num_bytes; 26928ca17f0fSJosef Bacik 26938ca17f0fSJosef Bacik /* 26948ca17f0fSJosef Bacik * We are under a transaction here so we can only do limited flushing. 26958ca17f0fSJosef Bacik * If we get an enospc just kick back -EAGAIN so we know to drop the 26968ca17f0fSJosef Bacik * transaction and try to refill when we can flush all the things. 26978ca17f0fSJosef Bacik */ 26980647bf56SWang Shilong ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes, 26998ca17f0fSJosef Bacik BTRFS_RESERVE_FLUSH_LIMIT); 27003fd0a558SYan, Zheng if (ret) { 2701da17066cSJeff Mahoney tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES; 27020647bf56SWang Shilong while (tmp <= rc->reserved_bytes) 27030647bf56SWang Shilong tmp <<= 1; 27040647bf56SWang Shilong /* 27050647bf56SWang Shilong * only one thread can access block_rsv at this point, 27060647bf56SWang Shilong * so we don't need hold lock to protect block_rsv. 27070647bf56SWang Shilong * we expand more reservation size here to allow enough 270852042d8eSAndrea Gelmini * space for relocation and we will return earlier in 27090647bf56SWang Shilong * enospc case. 27100647bf56SWang Shilong */ 2711da17066cSJeff Mahoney rc->block_rsv->size = tmp + fs_info->nodesize * 27120647bf56SWang Shilong RELOCATION_RESERVED_NODES; 27138ca17f0fSJosef Bacik return -EAGAIN; 27143fd0a558SYan, Zheng } 27153fd0a558SYan, Zheng 27163fd0a558SYan, Zheng return 0; 27173fd0a558SYan, Zheng } 27183fd0a558SYan, Zheng 27195d4f98a2SYan Zheng /* 27205d4f98a2SYan Zheng * relocate a block tree, and then update pointers in upper level 27215d4f98a2SYan Zheng * blocks that reference the block to point to the new location. 27225d4f98a2SYan Zheng * 27235d4f98a2SYan Zheng * if called by link_to_upper, the block has already been relocated. 27245d4f98a2SYan Zheng * in that case this function just updates pointers. 27255d4f98a2SYan Zheng */ 27265d4f98a2SYan Zheng static int do_relocation(struct btrfs_trans_handle *trans, 27273fd0a558SYan, Zheng struct reloc_control *rc, 27285d4f98a2SYan Zheng struct backref_node *node, 27295d4f98a2SYan Zheng struct btrfs_key *key, 27305d4f98a2SYan Zheng struct btrfs_path *path, int lowest) 27315d4f98a2SYan Zheng { 27322ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 27335d4f98a2SYan Zheng struct backref_node *upper; 27345d4f98a2SYan Zheng struct backref_edge *edge; 27355d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 27365d4f98a2SYan Zheng struct btrfs_root *root; 27375d4f98a2SYan Zheng struct extent_buffer *eb; 27385d4f98a2SYan Zheng u32 blocksize; 27395d4f98a2SYan Zheng u64 bytenr; 27405d4f98a2SYan Zheng u64 generation; 27415d4f98a2SYan Zheng int slot; 27425d4f98a2SYan Zheng int ret; 27435d4f98a2SYan Zheng int err = 0; 27445d4f98a2SYan Zheng 27455d4f98a2SYan Zheng BUG_ON(lowest && node->eb); 27465d4f98a2SYan Zheng 27475d4f98a2SYan Zheng path->lowest_level = node->level + 1; 27483fd0a558SYan, Zheng rc->backref_cache.path[node->level] = node; 27495d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) { 2750581c1760SQu Wenruo struct btrfs_key first_key; 275182fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 2752581c1760SQu Wenruo 27535d4f98a2SYan Zheng cond_resched(); 27545d4f98a2SYan Zheng 27555d4f98a2SYan Zheng upper = edge->node[UPPER]; 2756dc4103f9SWang Shilong root = select_reloc_root(trans, rc, upper, edges); 27573fd0a558SYan, Zheng BUG_ON(!root); 27585d4f98a2SYan Zheng 27593fd0a558SYan, Zheng if (upper->eb && !upper->locked) { 27603fd0a558SYan, Zheng if (!lowest) { 27613fd0a558SYan, Zheng ret = btrfs_bin_search(upper->eb, key, 27623fd0a558SYan, Zheng upper->level, &slot); 2763cbca7d59SFilipe Manana if (ret < 0) { 2764cbca7d59SFilipe Manana err = ret; 2765cbca7d59SFilipe Manana goto next; 2766cbca7d59SFilipe Manana } 27673fd0a558SYan, Zheng BUG_ON(ret); 27683fd0a558SYan, Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 27693fd0a558SYan, Zheng if (node->eb->start == bytenr) 27703fd0a558SYan, Zheng goto next; 27713fd0a558SYan, Zheng } 27725d4f98a2SYan Zheng drop_node_buffer(upper); 27733fd0a558SYan, Zheng } 27745d4f98a2SYan Zheng 27755d4f98a2SYan Zheng if (!upper->eb) { 27765d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 27773561b9dbSLiu Bo if (ret) { 27783561b9dbSLiu Bo if (ret < 0) 27795d4f98a2SYan Zheng err = ret; 27803561b9dbSLiu Bo else 27813561b9dbSLiu Bo err = -ENOENT; 27823561b9dbSLiu Bo 27833561b9dbSLiu Bo btrfs_release_path(path); 27845d4f98a2SYan Zheng break; 27855d4f98a2SYan Zheng } 27865d4f98a2SYan Zheng 27873fd0a558SYan, Zheng if (!upper->eb) { 27883fd0a558SYan, Zheng upper->eb = path->nodes[upper->level]; 27893fd0a558SYan, Zheng path->nodes[upper->level] = NULL; 27903fd0a558SYan, Zheng } else { 27913fd0a558SYan, Zheng BUG_ON(upper->eb != path->nodes[upper->level]); 27923fd0a558SYan, Zheng } 27933fd0a558SYan, Zheng 27943fd0a558SYan, Zheng upper->locked = 1; 27953fd0a558SYan, Zheng path->locks[upper->level] = 0; 27963fd0a558SYan, Zheng 27975d4f98a2SYan Zheng slot = path->slots[upper->level]; 2798b3b4aa74SDavid Sterba btrfs_release_path(path); 27995d4f98a2SYan Zheng } else { 28005d4f98a2SYan Zheng ret = btrfs_bin_search(upper->eb, key, upper->level, 28015d4f98a2SYan Zheng &slot); 2802cbca7d59SFilipe Manana if (ret < 0) { 2803cbca7d59SFilipe Manana err = ret; 2804cbca7d59SFilipe Manana goto next; 2805cbca7d59SFilipe Manana } 28065d4f98a2SYan Zheng BUG_ON(ret); 28075d4f98a2SYan Zheng } 28085d4f98a2SYan Zheng 28095d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 28103fd0a558SYan, Zheng if (lowest) { 28114547f4d8SLiu Bo if (bytenr != node->bytenr) { 28124547f4d8SLiu Bo btrfs_err(root->fs_info, 28134547f4d8SLiu Bo "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu", 28144547f4d8SLiu Bo bytenr, node->bytenr, slot, 28154547f4d8SLiu Bo upper->eb->start); 28164547f4d8SLiu Bo err = -EIO; 28174547f4d8SLiu Bo goto next; 28184547f4d8SLiu Bo } 28195d4f98a2SYan Zheng } else { 28203fd0a558SYan, Zheng if (node->eb->start == bytenr) 28213fd0a558SYan, Zheng goto next; 28225d4f98a2SYan Zheng } 28235d4f98a2SYan Zheng 2824da17066cSJeff Mahoney blocksize = root->fs_info->nodesize; 28255d4f98a2SYan Zheng generation = btrfs_node_ptr_generation(upper->eb, slot); 2826581c1760SQu Wenruo btrfs_node_key_to_cpu(upper->eb, &first_key, slot); 2827581c1760SQu Wenruo eb = read_tree_block(fs_info, bytenr, generation, 2828581c1760SQu Wenruo upper->level - 1, &first_key); 282964c043deSLiu Bo if (IS_ERR(eb)) { 283064c043deSLiu Bo err = PTR_ERR(eb); 283164c043deSLiu Bo goto next; 283264c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2833416bc658SJosef Bacik free_extent_buffer(eb); 283497d9a8a4STsutomu Itoh err = -EIO; 283597d9a8a4STsutomu Itoh goto next; 283697d9a8a4STsutomu Itoh } 28375d4f98a2SYan Zheng btrfs_tree_lock(eb); 28388bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 28395d4f98a2SYan Zheng 28405d4f98a2SYan Zheng if (!node->eb) { 28415d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, eb, upper->eb, 28425d4f98a2SYan Zheng slot, &eb); 28433fd0a558SYan, Zheng btrfs_tree_unlock(eb); 28443fd0a558SYan, Zheng free_extent_buffer(eb); 28455d4f98a2SYan Zheng if (ret < 0) { 28465d4f98a2SYan Zheng err = ret; 28473fd0a558SYan, Zheng goto next; 28485d4f98a2SYan Zheng } 28493fd0a558SYan, Zheng BUG_ON(node->eb != eb); 28505d4f98a2SYan Zheng } else { 28515d4f98a2SYan Zheng btrfs_set_node_blockptr(upper->eb, slot, 28525d4f98a2SYan Zheng node->eb->start); 28535d4f98a2SYan Zheng btrfs_set_node_ptr_generation(upper->eb, slot, 28545d4f98a2SYan Zheng trans->transid); 28555d4f98a2SYan Zheng btrfs_mark_buffer_dirty(upper->eb); 28565d4f98a2SYan Zheng 285782fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, 28585d4f98a2SYan Zheng node->eb->start, blocksize, 285982fa113fSQu Wenruo upper->eb->start); 286082fa113fSQu Wenruo ref.real_root = root->root_key.objectid; 286182fa113fSQu Wenruo btrfs_init_tree_ref(&ref, node->level, 286282fa113fSQu Wenruo btrfs_header_owner(upper->eb)); 286382fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 28645d4f98a2SYan Zheng BUG_ON(ret); 28655d4f98a2SYan Zheng 28665d4f98a2SYan Zheng ret = btrfs_drop_subtree(trans, root, eb, upper->eb); 28675d4f98a2SYan Zheng BUG_ON(ret); 28685d4f98a2SYan Zheng } 28693fd0a558SYan, Zheng next: 28703fd0a558SYan, Zheng if (!upper->pending) 28713fd0a558SYan, Zheng drop_node_buffer(upper); 28723fd0a558SYan, Zheng else 28733fd0a558SYan, Zheng unlock_node_buffer(upper); 28743fd0a558SYan, Zheng if (err) 28753fd0a558SYan, Zheng break; 28765d4f98a2SYan Zheng } 28773fd0a558SYan, Zheng 28783fd0a558SYan, Zheng if (!err && node->pending) { 28793fd0a558SYan, Zheng drop_node_buffer(node); 28803fd0a558SYan, Zheng list_move_tail(&node->list, &rc->backref_cache.changed); 28813fd0a558SYan, Zheng node->pending = 0; 28825d4f98a2SYan Zheng } 28833fd0a558SYan, Zheng 28845d4f98a2SYan Zheng path->lowest_level = 0; 28853fd0a558SYan, Zheng BUG_ON(err == -ENOSPC); 28865d4f98a2SYan Zheng return err; 28875d4f98a2SYan Zheng } 28885d4f98a2SYan Zheng 28895d4f98a2SYan Zheng static int link_to_upper(struct btrfs_trans_handle *trans, 28903fd0a558SYan, Zheng struct reloc_control *rc, 28915d4f98a2SYan Zheng struct backref_node *node, 28925d4f98a2SYan Zheng struct btrfs_path *path) 28935d4f98a2SYan Zheng { 28945d4f98a2SYan Zheng struct btrfs_key key; 28955d4f98a2SYan Zheng 28965d4f98a2SYan Zheng btrfs_node_key_to_cpu(node->eb, &key, 0); 28973fd0a558SYan, Zheng return do_relocation(trans, rc, node, &key, path, 0); 28985d4f98a2SYan Zheng } 28995d4f98a2SYan Zheng 29005d4f98a2SYan Zheng static int finish_pending_nodes(struct btrfs_trans_handle *trans, 29013fd0a558SYan, Zheng struct reloc_control *rc, 29023fd0a558SYan, Zheng struct btrfs_path *path, int err) 29035d4f98a2SYan Zheng { 29043fd0a558SYan, Zheng LIST_HEAD(list); 29053fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 29065d4f98a2SYan Zheng struct backref_node *node; 29075d4f98a2SYan Zheng int level; 29085d4f98a2SYan Zheng int ret; 29095d4f98a2SYan Zheng 29105d4f98a2SYan Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 29115d4f98a2SYan Zheng while (!list_empty(&cache->pending[level])) { 29125d4f98a2SYan Zheng node = list_entry(cache->pending[level].next, 29133fd0a558SYan, Zheng struct backref_node, list); 29143fd0a558SYan, Zheng list_move_tail(&node->list, &list); 29153fd0a558SYan, Zheng BUG_ON(!node->pending); 29165d4f98a2SYan Zheng 29173fd0a558SYan, Zheng if (!err) { 29183fd0a558SYan, Zheng ret = link_to_upper(trans, rc, node, path); 29195d4f98a2SYan Zheng if (ret < 0) 29205d4f98a2SYan Zheng err = ret; 29215d4f98a2SYan Zheng } 29225d4f98a2SYan Zheng } 29233fd0a558SYan, Zheng list_splice_init(&list, &cache->pending[level]); 29243fd0a558SYan, Zheng } 29255d4f98a2SYan Zheng return err; 29265d4f98a2SYan Zheng } 29275d4f98a2SYan Zheng 29285d4f98a2SYan Zheng static void mark_block_processed(struct reloc_control *rc, 29293fd0a558SYan, Zheng u64 bytenr, u32 blocksize) 29303fd0a558SYan, Zheng { 29313fd0a558SYan, Zheng set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1, 2932ceeb0ae7SDavid Sterba EXTENT_DIRTY); 29333fd0a558SYan, Zheng } 29343fd0a558SYan, Zheng 29353fd0a558SYan, Zheng static void __mark_block_processed(struct reloc_control *rc, 29365d4f98a2SYan Zheng struct backref_node *node) 29375d4f98a2SYan Zheng { 29385d4f98a2SYan Zheng u32 blocksize; 29395d4f98a2SYan Zheng if (node->level == 0 || 29405d4f98a2SYan Zheng in_block_group(node->bytenr, rc->block_group)) { 2941da17066cSJeff Mahoney blocksize = rc->extent_root->fs_info->nodesize; 29423fd0a558SYan, Zheng mark_block_processed(rc, node->bytenr, blocksize); 29435d4f98a2SYan Zheng } 29445d4f98a2SYan Zheng node->processed = 1; 29455d4f98a2SYan Zheng } 29465d4f98a2SYan Zheng 29475d4f98a2SYan Zheng /* 29485d4f98a2SYan Zheng * mark a block and all blocks directly/indirectly reference the block 29495d4f98a2SYan Zheng * as processed. 29505d4f98a2SYan Zheng */ 29515d4f98a2SYan Zheng static void update_processed_blocks(struct reloc_control *rc, 29525d4f98a2SYan Zheng struct backref_node *node) 29535d4f98a2SYan Zheng { 29545d4f98a2SYan Zheng struct backref_node *next = node; 29555d4f98a2SYan Zheng struct backref_edge *edge; 29565d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 29575d4f98a2SYan Zheng int index = 0; 29585d4f98a2SYan Zheng 29595d4f98a2SYan Zheng while (next) { 29605d4f98a2SYan Zheng cond_resched(); 29615d4f98a2SYan Zheng while (1) { 29625d4f98a2SYan Zheng if (next->processed) 29635d4f98a2SYan Zheng break; 29645d4f98a2SYan Zheng 29653fd0a558SYan, Zheng __mark_block_processed(rc, next); 29665d4f98a2SYan Zheng 29675d4f98a2SYan Zheng if (list_empty(&next->upper)) 29685d4f98a2SYan Zheng break; 29695d4f98a2SYan Zheng 29705d4f98a2SYan Zheng edge = list_entry(next->upper.next, 29715d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 29725d4f98a2SYan Zheng edges[index++] = edge; 29735d4f98a2SYan Zheng next = edge->node[UPPER]; 29745d4f98a2SYan Zheng } 29755d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 29765d4f98a2SYan Zheng } 29775d4f98a2SYan Zheng } 29785d4f98a2SYan Zheng 29797476dfdaSDavid Sterba static int tree_block_processed(u64 bytenr, struct reloc_control *rc) 29805d4f98a2SYan Zheng { 2981da17066cSJeff Mahoney u32 blocksize = rc->extent_root->fs_info->nodesize; 29827476dfdaSDavid Sterba 29835d4f98a2SYan Zheng if (test_range_bit(&rc->processed_blocks, bytenr, 29849655d298SChris Mason bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL)) 29855d4f98a2SYan Zheng return 1; 29865d4f98a2SYan Zheng return 0; 29875d4f98a2SYan Zheng } 29885d4f98a2SYan Zheng 29892ff7e61eSJeff Mahoney static int get_tree_block_key(struct btrfs_fs_info *fs_info, 29905d4f98a2SYan Zheng struct tree_block *block) 29915d4f98a2SYan Zheng { 29925d4f98a2SYan Zheng struct extent_buffer *eb; 29935d4f98a2SYan Zheng 29945d4f98a2SYan Zheng BUG_ON(block->key_ready); 2995581c1760SQu Wenruo eb = read_tree_block(fs_info, block->bytenr, block->key.offset, 2996581c1760SQu Wenruo block->level, NULL); 299764c043deSLiu Bo if (IS_ERR(eb)) { 299864c043deSLiu Bo return PTR_ERR(eb); 299964c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 3000416bc658SJosef Bacik free_extent_buffer(eb); 3001416bc658SJosef Bacik return -EIO; 3002416bc658SJosef Bacik } 30035d4f98a2SYan Zheng if (block->level == 0) 30045d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &block->key, 0); 30055d4f98a2SYan Zheng else 30065d4f98a2SYan Zheng btrfs_node_key_to_cpu(eb, &block->key, 0); 30075d4f98a2SYan Zheng free_extent_buffer(eb); 30085d4f98a2SYan Zheng block->key_ready = 1; 30095d4f98a2SYan Zheng return 0; 30105d4f98a2SYan Zheng } 30115d4f98a2SYan Zheng 30125d4f98a2SYan Zheng /* 30135d4f98a2SYan Zheng * helper function to relocate a tree block 30145d4f98a2SYan Zheng */ 30155d4f98a2SYan Zheng static int relocate_tree_block(struct btrfs_trans_handle *trans, 30165d4f98a2SYan Zheng struct reloc_control *rc, 30175d4f98a2SYan Zheng struct backref_node *node, 30185d4f98a2SYan Zheng struct btrfs_key *key, 30195d4f98a2SYan Zheng struct btrfs_path *path) 30205d4f98a2SYan Zheng { 30215d4f98a2SYan Zheng struct btrfs_root *root; 30223fd0a558SYan, Zheng int ret = 0; 30235d4f98a2SYan Zheng 30243fd0a558SYan, Zheng if (!node) 30255d4f98a2SYan Zheng return 0; 30263fd0a558SYan, Zheng 30273fd0a558SYan, Zheng BUG_ON(node->processed); 3028147d256eSZhaolei root = select_one_root(node); 30293fd0a558SYan, Zheng if (root == ERR_PTR(-ENOENT)) { 30303fd0a558SYan, Zheng update_processed_blocks(rc, node); 30313fd0a558SYan, Zheng goto out; 30325d4f98a2SYan Zheng } 30335d4f98a2SYan Zheng 303427cdeb70SMiao Xie if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 30353fd0a558SYan, Zheng ret = reserve_metadata_space(trans, rc, node); 30363fd0a558SYan, Zheng if (ret) 30375d4f98a2SYan Zheng goto out; 30385d4f98a2SYan Zheng } 30393fd0a558SYan, Zheng 30403fd0a558SYan, Zheng if (root) { 304127cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 30423fd0a558SYan, Zheng BUG_ON(node->new_bytenr); 30433fd0a558SYan, Zheng BUG_ON(!list_empty(&node->list)); 30443fd0a558SYan, Zheng btrfs_record_root_in_trans(trans, root); 30453fd0a558SYan, Zheng root = root->reloc_root; 30463fd0a558SYan, Zheng node->new_bytenr = root->node->start; 30473fd0a558SYan, Zheng node->root = root; 30483fd0a558SYan, Zheng list_add_tail(&node->list, &rc->backref_cache.changed); 30493fd0a558SYan, Zheng } else { 30505d4f98a2SYan Zheng path->lowest_level = node->level; 30515d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 3052b3b4aa74SDavid Sterba btrfs_release_path(path); 30533fd0a558SYan, Zheng if (ret > 0) 30545d4f98a2SYan Zheng ret = 0; 30553fd0a558SYan, Zheng } 30563fd0a558SYan, Zheng if (!ret) 30573fd0a558SYan, Zheng update_processed_blocks(rc, node); 30583fd0a558SYan, Zheng } else { 30593fd0a558SYan, Zheng ret = do_relocation(trans, rc, node, key, path, 1); 30603fd0a558SYan, Zheng } 30615d4f98a2SYan Zheng out: 30620647bf56SWang Shilong if (ret || node->level == 0 || node->cowonly) 30633fd0a558SYan, Zheng remove_backref_node(&rc->backref_cache, node); 30645d4f98a2SYan Zheng return ret; 30655d4f98a2SYan Zheng } 30665d4f98a2SYan Zheng 30675d4f98a2SYan Zheng /* 30685d4f98a2SYan Zheng * relocate a list of blocks 30695d4f98a2SYan Zheng */ 30705d4f98a2SYan Zheng static noinline_for_stack 30715d4f98a2SYan Zheng int relocate_tree_blocks(struct btrfs_trans_handle *trans, 30725d4f98a2SYan Zheng struct reloc_control *rc, struct rb_root *blocks) 30735d4f98a2SYan Zheng { 30742ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 30755d4f98a2SYan Zheng struct backref_node *node; 30765d4f98a2SYan Zheng struct btrfs_path *path; 30775d4f98a2SYan Zheng struct tree_block *block; 307898ff7b94SQu Wenruo struct tree_block *next; 30795d4f98a2SYan Zheng int ret; 30805d4f98a2SYan Zheng int err = 0; 30815d4f98a2SYan Zheng 30825d4f98a2SYan Zheng path = btrfs_alloc_path(); 3083e1a12670SLiu Bo if (!path) { 3084e1a12670SLiu Bo err = -ENOMEM; 308534c2b290SDavid Sterba goto out_free_blocks; 3086e1a12670SLiu Bo } 30875d4f98a2SYan Zheng 308898ff7b94SQu Wenruo /* Kick in readahead for tree blocks with missing keys */ 308998ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 30905d4f98a2SYan Zheng if (!block->key_ready) 30912ff7e61eSJeff Mahoney readahead_tree_block(fs_info, block->bytenr); 30925d4f98a2SYan Zheng } 30935d4f98a2SYan Zheng 309498ff7b94SQu Wenruo /* Get first keys */ 309598ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 309634c2b290SDavid Sterba if (!block->key_ready) { 30972ff7e61eSJeff Mahoney err = get_tree_block_key(fs_info, block); 309834c2b290SDavid Sterba if (err) 309934c2b290SDavid Sterba goto out_free_path; 310034c2b290SDavid Sterba } 31015d4f98a2SYan Zheng } 31025d4f98a2SYan Zheng 310398ff7b94SQu Wenruo /* Do tree relocation */ 310498ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 31053fd0a558SYan, Zheng node = build_backref_tree(rc, &block->key, 31065d4f98a2SYan Zheng block->level, block->bytenr); 31075d4f98a2SYan Zheng if (IS_ERR(node)) { 31085d4f98a2SYan Zheng err = PTR_ERR(node); 31095d4f98a2SYan Zheng goto out; 31105d4f98a2SYan Zheng } 31115d4f98a2SYan Zheng 31125d4f98a2SYan Zheng ret = relocate_tree_block(trans, rc, node, &block->key, 31135d4f98a2SYan Zheng path); 31145d4f98a2SYan Zheng if (ret < 0) { 311598ff7b94SQu Wenruo if (ret != -EAGAIN || &block->rb_node == rb_first(blocks)) 31165d4f98a2SYan Zheng err = ret; 31175d4f98a2SYan Zheng goto out; 31185d4f98a2SYan Zheng } 31195d4f98a2SYan Zheng } 31205d4f98a2SYan Zheng out: 31213fd0a558SYan, Zheng err = finish_pending_nodes(trans, rc, path, err); 31225d4f98a2SYan Zheng 312334c2b290SDavid Sterba out_free_path: 31245d4f98a2SYan Zheng btrfs_free_path(path); 312534c2b290SDavid Sterba out_free_blocks: 3126e1a12670SLiu Bo free_block_list(blocks); 31275d4f98a2SYan Zheng return err; 31285d4f98a2SYan Zheng } 31295d4f98a2SYan Zheng 31305d4f98a2SYan Zheng static noinline_for_stack 3131efa56464SYan, Zheng int prealloc_file_extent_cluster(struct inode *inode, 3132efa56464SYan, Zheng struct file_extent_cluster *cluster) 3133efa56464SYan, Zheng { 3134efa56464SYan, Zheng u64 alloc_hint = 0; 3135efa56464SYan, Zheng u64 start; 3136efa56464SYan, Zheng u64 end; 3137efa56464SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 3138efa56464SYan, Zheng u64 num_bytes; 3139efa56464SYan, Zheng int nr = 0; 3140efa56464SYan, Zheng int ret = 0; 3141dcb40c19SWang Xiaoguang u64 prealloc_start = cluster->start - offset; 3142dcb40c19SWang Xiaoguang u64 prealloc_end = cluster->end - offset; 314318513091SWang Xiaoguang u64 cur_offset; 3144364ecf36SQu Wenruo struct extent_changeset *data_reserved = NULL; 3145efa56464SYan, Zheng 3146efa56464SYan, Zheng BUG_ON(cluster->start != cluster->boundary[0]); 31475955102cSAl Viro inode_lock(inode); 3148efa56464SYan, Zheng 3149364ecf36SQu Wenruo ret = btrfs_check_data_free_space(inode, &data_reserved, prealloc_start, 3150dcb40c19SWang Xiaoguang prealloc_end + 1 - prealloc_start); 3151efa56464SYan, Zheng if (ret) 3152efa56464SYan, Zheng goto out; 3153efa56464SYan, Zheng 315418513091SWang Xiaoguang cur_offset = prealloc_start; 3155efa56464SYan, Zheng while (nr < cluster->nr) { 3156efa56464SYan, Zheng start = cluster->boundary[nr] - offset; 3157efa56464SYan, Zheng if (nr + 1 < cluster->nr) 3158efa56464SYan, Zheng end = cluster->boundary[nr + 1] - 1 - offset; 3159efa56464SYan, Zheng else 3160efa56464SYan, Zheng end = cluster->end - offset; 3161efa56464SYan, Zheng 3162d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 3163efa56464SYan, Zheng num_bytes = end + 1 - start; 316418513091SWang Xiaoguang if (cur_offset < start) 3165bc42bda2SQu Wenruo btrfs_free_reserved_data_space(inode, data_reserved, 3166bc42bda2SQu Wenruo cur_offset, start - cur_offset); 3167efa56464SYan, Zheng ret = btrfs_prealloc_file_range(inode, 0, start, 3168efa56464SYan, Zheng num_bytes, num_bytes, 3169efa56464SYan, Zheng end + 1, &alloc_hint); 317018513091SWang Xiaoguang cur_offset = end + 1; 3171d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 3172efa56464SYan, Zheng if (ret) 3173efa56464SYan, Zheng break; 3174efa56464SYan, Zheng nr++; 3175efa56464SYan, Zheng } 317618513091SWang Xiaoguang if (cur_offset < prealloc_end) 3177bc42bda2SQu Wenruo btrfs_free_reserved_data_space(inode, data_reserved, 3178bc42bda2SQu Wenruo cur_offset, prealloc_end + 1 - cur_offset); 3179efa56464SYan, Zheng out: 31805955102cSAl Viro inode_unlock(inode); 3181364ecf36SQu Wenruo extent_changeset_free(data_reserved); 3182efa56464SYan, Zheng return ret; 3183efa56464SYan, Zheng } 3184efa56464SYan, Zheng 3185efa56464SYan, Zheng static noinline_for_stack 31860257bb82SYan, Zheng int setup_extent_mapping(struct inode *inode, u64 start, u64 end, 31870257bb82SYan, Zheng u64 block_start) 31885d4f98a2SYan Zheng { 31890b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 31905d4f98a2SYan Zheng struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 31915d4f98a2SYan Zheng struct extent_map *em; 31920257bb82SYan, Zheng int ret = 0; 31935d4f98a2SYan Zheng 3194172ddd60SDavid Sterba em = alloc_extent_map(); 31950257bb82SYan, Zheng if (!em) 31960257bb82SYan, Zheng return -ENOMEM; 31970257bb82SYan, Zheng 31985d4f98a2SYan Zheng em->start = start; 31990257bb82SYan, Zheng em->len = end + 1 - start; 32000257bb82SYan, Zheng em->block_len = em->len; 32010257bb82SYan, Zheng em->block_start = block_start; 32020b246afaSJeff Mahoney em->bdev = fs_info->fs_devices->latest_bdev; 32035d4f98a2SYan Zheng set_bit(EXTENT_FLAG_PINNED, &em->flags); 32045d4f98a2SYan Zheng 3205d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 32065d4f98a2SYan Zheng while (1) { 3207890871beSChris Mason write_lock(&em_tree->lock); 320809a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 3209890871beSChris Mason write_unlock(&em_tree->lock); 32105d4f98a2SYan Zheng if (ret != -EEXIST) { 32115d4f98a2SYan Zheng free_extent_map(em); 32125d4f98a2SYan Zheng break; 32135d4f98a2SYan Zheng } 3214dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0); 32155d4f98a2SYan Zheng } 3216d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 32170257bb82SYan, Zheng return ret; 32180257bb82SYan, Zheng } 32195d4f98a2SYan Zheng 32200257bb82SYan, Zheng static int relocate_file_extent_cluster(struct inode *inode, 32210257bb82SYan, Zheng struct file_extent_cluster *cluster) 32220257bb82SYan, Zheng { 32232ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 32240257bb82SYan, Zheng u64 page_start; 32250257bb82SYan, Zheng u64 page_end; 32260257bb82SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 32270257bb82SYan, Zheng unsigned long index; 32280257bb82SYan, Zheng unsigned long last_index; 32290257bb82SYan, Zheng struct page *page; 32300257bb82SYan, Zheng struct file_ra_state *ra; 32313b16a4e3SJosef Bacik gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 32320257bb82SYan, Zheng int nr = 0; 32330257bb82SYan, Zheng int ret = 0; 32340257bb82SYan, Zheng 32350257bb82SYan, Zheng if (!cluster->nr) 32360257bb82SYan, Zheng return 0; 32370257bb82SYan, Zheng 32380257bb82SYan, Zheng ra = kzalloc(sizeof(*ra), GFP_NOFS); 32390257bb82SYan, Zheng if (!ra) 32400257bb82SYan, Zheng return -ENOMEM; 32410257bb82SYan, Zheng 3242efa56464SYan, Zheng ret = prealloc_file_extent_cluster(inode, cluster); 32430257bb82SYan, Zheng if (ret) 3244efa56464SYan, Zheng goto out; 32450257bb82SYan, Zheng 32460257bb82SYan, Zheng file_ra_state_init(ra, inode->i_mapping); 32470257bb82SYan, Zheng 3248efa56464SYan, Zheng ret = setup_extent_mapping(inode, cluster->start - offset, 3249efa56464SYan, Zheng cluster->end - offset, cluster->start); 3250efa56464SYan, Zheng if (ret) 3251efa56464SYan, Zheng goto out; 3252efa56464SYan, Zheng 325309cbfeafSKirill A. Shutemov index = (cluster->start - offset) >> PAGE_SHIFT; 325409cbfeafSKirill A. Shutemov last_index = (cluster->end - offset) >> PAGE_SHIFT; 32550257bb82SYan, Zheng while (index <= last_index) { 32569f3db423SNikolay Borisov ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), 32579f3db423SNikolay Borisov PAGE_SIZE); 3258efa56464SYan, Zheng if (ret) 3259efa56464SYan, Zheng goto out; 3260efa56464SYan, Zheng 32610257bb82SYan, Zheng page = find_lock_page(inode->i_mapping, index); 32620257bb82SYan, Zheng if (!page) { 32630257bb82SYan, Zheng page_cache_sync_readahead(inode->i_mapping, 32640257bb82SYan, Zheng ra, NULL, index, 32650257bb82SYan, Zheng last_index + 1 - index); 3266a94733d0SJosef Bacik page = find_or_create_page(inode->i_mapping, index, 32673b16a4e3SJosef Bacik mask); 32680257bb82SYan, Zheng if (!page) { 3269691fa059SNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 327043b18595SQu Wenruo PAGE_SIZE, true); 32710257bb82SYan, Zheng ret = -ENOMEM; 3272efa56464SYan, Zheng goto out; 32730257bb82SYan, Zheng } 32740257bb82SYan, Zheng } 32750257bb82SYan, Zheng 32760257bb82SYan, Zheng if (PageReadahead(page)) { 32770257bb82SYan, Zheng page_cache_async_readahead(inode->i_mapping, 32780257bb82SYan, Zheng ra, NULL, page, index, 32790257bb82SYan, Zheng last_index + 1 - index); 32800257bb82SYan, Zheng } 32810257bb82SYan, Zheng 32820257bb82SYan, Zheng if (!PageUptodate(page)) { 32830257bb82SYan, Zheng btrfs_readpage(NULL, page); 32840257bb82SYan, Zheng lock_page(page); 32850257bb82SYan, Zheng if (!PageUptodate(page)) { 32860257bb82SYan, Zheng unlock_page(page); 328709cbfeafSKirill A. Shutemov put_page(page); 3288691fa059SNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 328943b18595SQu Wenruo PAGE_SIZE, true); 32908b62f87bSJosef Bacik btrfs_delalloc_release_extents(BTRFS_I(inode), 329143b18595SQu Wenruo PAGE_SIZE, true); 32920257bb82SYan, Zheng ret = -EIO; 3293efa56464SYan, Zheng goto out; 32940257bb82SYan, Zheng } 32950257bb82SYan, Zheng } 32960257bb82SYan, Zheng 32974eee4fa4SMiao Xie page_start = page_offset(page); 329809cbfeafSKirill A. Shutemov page_end = page_start + PAGE_SIZE - 1; 32990257bb82SYan, Zheng 3300d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end); 33010257bb82SYan, Zheng 33020257bb82SYan, Zheng set_page_extent_mapped(page); 33030257bb82SYan, Zheng 33040257bb82SYan, Zheng if (nr < cluster->nr && 33050257bb82SYan, Zheng page_start + offset == cluster->boundary[nr]) { 33060257bb82SYan, Zheng set_extent_bits(&BTRFS_I(inode)->io_tree, 33070257bb82SYan, Zheng page_start, page_end, 3308ceeb0ae7SDavid Sterba EXTENT_BOUNDARY); 33090257bb82SYan, Zheng nr++; 33100257bb82SYan, Zheng } 33110257bb82SYan, Zheng 3312765f3cebSNikolay Borisov ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0, 3313765f3cebSNikolay Borisov NULL, 0); 3314765f3cebSNikolay Borisov if (ret) { 3315765f3cebSNikolay Borisov unlock_page(page); 3316765f3cebSNikolay Borisov put_page(page); 3317765f3cebSNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 331843b18595SQu Wenruo PAGE_SIZE, true); 3319765f3cebSNikolay Borisov btrfs_delalloc_release_extents(BTRFS_I(inode), 332043b18595SQu Wenruo PAGE_SIZE, true); 3321765f3cebSNikolay Borisov 3322765f3cebSNikolay Borisov clear_extent_bits(&BTRFS_I(inode)->io_tree, 3323765f3cebSNikolay Borisov page_start, page_end, 3324765f3cebSNikolay Borisov EXTENT_LOCKED | EXTENT_BOUNDARY); 3325765f3cebSNikolay Borisov goto out; 3326765f3cebSNikolay Borisov 3327765f3cebSNikolay Borisov } 33280257bb82SYan, Zheng set_page_dirty(page); 33290257bb82SYan, Zheng 33300257bb82SYan, Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 3331d0082371SJeff Mahoney page_start, page_end); 33320257bb82SYan, Zheng unlock_page(page); 333309cbfeafSKirill A. Shutemov put_page(page); 33340257bb82SYan, Zheng 33350257bb82SYan, Zheng index++; 333643b18595SQu Wenruo btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, 333743b18595SQu Wenruo false); 3338efa56464SYan, Zheng balance_dirty_pages_ratelimited(inode->i_mapping); 33392ff7e61eSJeff Mahoney btrfs_throttle(fs_info); 33400257bb82SYan, Zheng } 33410257bb82SYan, Zheng WARN_ON(nr != cluster->nr); 3342efa56464SYan, Zheng out: 33430257bb82SYan, Zheng kfree(ra); 33440257bb82SYan, Zheng return ret; 33450257bb82SYan, Zheng } 33460257bb82SYan, Zheng 33470257bb82SYan, Zheng static noinline_for_stack 33480257bb82SYan, Zheng int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key, 33490257bb82SYan, Zheng struct file_extent_cluster *cluster) 33500257bb82SYan, Zheng { 33510257bb82SYan, Zheng int ret; 33520257bb82SYan, Zheng 33530257bb82SYan, Zheng if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) { 33540257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 33550257bb82SYan, Zheng if (ret) 33560257bb82SYan, Zheng return ret; 33570257bb82SYan, Zheng cluster->nr = 0; 33580257bb82SYan, Zheng } 33590257bb82SYan, Zheng 33600257bb82SYan, Zheng if (!cluster->nr) 33610257bb82SYan, Zheng cluster->start = extent_key->objectid; 33620257bb82SYan, Zheng else 33630257bb82SYan, Zheng BUG_ON(cluster->nr >= MAX_EXTENTS); 33640257bb82SYan, Zheng cluster->end = extent_key->objectid + extent_key->offset - 1; 33650257bb82SYan, Zheng cluster->boundary[cluster->nr] = extent_key->objectid; 33660257bb82SYan, Zheng cluster->nr++; 33670257bb82SYan, Zheng 33680257bb82SYan, Zheng if (cluster->nr >= MAX_EXTENTS) { 33690257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 33700257bb82SYan, Zheng if (ret) 33710257bb82SYan, Zheng return ret; 33720257bb82SYan, Zheng cluster->nr = 0; 33730257bb82SYan, Zheng } 33740257bb82SYan, Zheng return 0; 33755d4f98a2SYan Zheng } 33765d4f98a2SYan Zheng 33775d4f98a2SYan Zheng /* 33785d4f98a2SYan Zheng * helper to add a tree block to the list. 33795d4f98a2SYan Zheng * the major work is getting the generation and level of the block 33805d4f98a2SYan Zheng */ 33815d4f98a2SYan Zheng static int add_tree_block(struct reloc_control *rc, 33825d4f98a2SYan Zheng struct btrfs_key *extent_key, 33835d4f98a2SYan Zheng struct btrfs_path *path, 33845d4f98a2SYan Zheng struct rb_root *blocks) 33855d4f98a2SYan Zheng { 33865d4f98a2SYan Zheng struct extent_buffer *eb; 33875d4f98a2SYan Zheng struct btrfs_extent_item *ei; 33885d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 33895d4f98a2SYan Zheng struct tree_block *block; 33905d4f98a2SYan Zheng struct rb_node *rb_node; 33915d4f98a2SYan Zheng u32 item_size; 33925d4f98a2SYan Zheng int level = -1; 33937fdf4b60SWang Shilong u64 generation; 33945d4f98a2SYan Zheng 33955d4f98a2SYan Zheng eb = path->nodes[0]; 33965d4f98a2SYan Zheng item_size = btrfs_item_size_nr(eb, path->slots[0]); 33975d4f98a2SYan Zheng 33983173a18fSJosef Bacik if (extent_key->type == BTRFS_METADATA_ITEM_KEY || 33993173a18fSJosef Bacik item_size >= sizeof(*ei) + sizeof(*bi)) { 34005d4f98a2SYan Zheng ei = btrfs_item_ptr(eb, path->slots[0], 34015d4f98a2SYan Zheng struct btrfs_extent_item); 34023173a18fSJosef Bacik if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) { 34035d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 34045d4f98a2SYan Zheng level = btrfs_tree_block_level(eb, bi); 34055d4f98a2SYan Zheng } else { 34063173a18fSJosef Bacik level = (int)extent_key->offset; 34073173a18fSJosef Bacik } 34083173a18fSJosef Bacik generation = btrfs_extent_generation(eb, ei); 34096d8ff4e4SDavid Sterba } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) { 3410ba3c2b19SNikolay Borisov btrfs_print_v0_err(eb->fs_info); 3411ba3c2b19SNikolay Borisov btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL); 3412ba3c2b19SNikolay Borisov return -EINVAL; 34133173a18fSJosef Bacik } else { 34145d4f98a2SYan Zheng BUG(); 34155d4f98a2SYan Zheng } 34165d4f98a2SYan Zheng 3417b3b4aa74SDavid Sterba btrfs_release_path(path); 34185d4f98a2SYan Zheng 34195d4f98a2SYan Zheng BUG_ON(level == -1); 34205d4f98a2SYan Zheng 34215d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 34225d4f98a2SYan Zheng if (!block) 34235d4f98a2SYan Zheng return -ENOMEM; 34245d4f98a2SYan Zheng 34255d4f98a2SYan Zheng block->bytenr = extent_key->objectid; 3426da17066cSJeff Mahoney block->key.objectid = rc->extent_root->fs_info->nodesize; 34275d4f98a2SYan Zheng block->key.offset = generation; 34285d4f98a2SYan Zheng block->level = level; 34295d4f98a2SYan Zheng block->key_ready = 0; 34305d4f98a2SYan Zheng 34315d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, &block->rb_node); 343243c04fb1SJeff Mahoney if (rb_node) 343343c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, block->bytenr); 34345d4f98a2SYan Zheng 34355d4f98a2SYan Zheng return 0; 34365d4f98a2SYan Zheng } 34375d4f98a2SYan Zheng 34385d4f98a2SYan Zheng /* 34395d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY 34405d4f98a2SYan Zheng */ 34415d4f98a2SYan Zheng static int __add_tree_block(struct reloc_control *rc, 34425d4f98a2SYan Zheng u64 bytenr, u32 blocksize, 34435d4f98a2SYan Zheng struct rb_root *blocks) 34445d4f98a2SYan Zheng { 34450b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 34465d4f98a2SYan Zheng struct btrfs_path *path; 34475d4f98a2SYan Zheng struct btrfs_key key; 34485d4f98a2SYan Zheng int ret; 34490b246afaSJeff Mahoney bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA); 34505d4f98a2SYan Zheng 34517476dfdaSDavid Sterba if (tree_block_processed(bytenr, rc)) 34525d4f98a2SYan Zheng return 0; 34535d4f98a2SYan Zheng 34545d4f98a2SYan Zheng if (tree_search(blocks, bytenr)) 34555d4f98a2SYan Zheng return 0; 34565d4f98a2SYan Zheng 34575d4f98a2SYan Zheng path = btrfs_alloc_path(); 34585d4f98a2SYan Zheng if (!path) 34595d4f98a2SYan Zheng return -ENOMEM; 3460aee68ee5SJosef Bacik again: 34615d4f98a2SYan Zheng key.objectid = bytenr; 3462aee68ee5SJosef Bacik if (skinny) { 3463aee68ee5SJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 3464aee68ee5SJosef Bacik key.offset = (u64)-1; 3465aee68ee5SJosef Bacik } else { 34665d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 34675d4f98a2SYan Zheng key.offset = blocksize; 3468aee68ee5SJosef Bacik } 34695d4f98a2SYan Zheng 34705d4f98a2SYan Zheng path->search_commit_root = 1; 34715d4f98a2SYan Zheng path->skip_locking = 1; 34725d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0); 34735d4f98a2SYan Zheng if (ret < 0) 34745d4f98a2SYan Zheng goto out; 34755d4f98a2SYan Zheng 3476aee68ee5SJosef Bacik if (ret > 0 && skinny) { 3477aee68ee5SJosef Bacik if (path->slots[0]) { 3478aee68ee5SJosef Bacik path->slots[0]--; 3479aee68ee5SJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, 3480aee68ee5SJosef Bacik path->slots[0]); 34813173a18fSJosef Bacik if (key.objectid == bytenr && 3482aee68ee5SJosef Bacik (key.type == BTRFS_METADATA_ITEM_KEY || 3483aee68ee5SJosef Bacik (key.type == BTRFS_EXTENT_ITEM_KEY && 3484aee68ee5SJosef Bacik key.offset == blocksize))) 34853173a18fSJosef Bacik ret = 0; 34863173a18fSJosef Bacik } 3487aee68ee5SJosef Bacik 3488aee68ee5SJosef Bacik if (ret) { 3489aee68ee5SJosef Bacik skinny = false; 3490aee68ee5SJosef Bacik btrfs_release_path(path); 3491aee68ee5SJosef Bacik goto again; 3492aee68ee5SJosef Bacik } 3493aee68ee5SJosef Bacik } 3494cdccee99SLiu Bo if (ret) { 3495cdccee99SLiu Bo ASSERT(ret == 1); 3496cdccee99SLiu Bo btrfs_print_leaf(path->nodes[0]); 3497cdccee99SLiu Bo btrfs_err(fs_info, 3498cdccee99SLiu Bo "tree block extent item (%llu) is not found in extent tree", 3499cdccee99SLiu Bo bytenr); 3500cdccee99SLiu Bo WARN_ON(1); 3501cdccee99SLiu Bo ret = -EINVAL; 3502cdccee99SLiu Bo goto out; 3503cdccee99SLiu Bo } 35043173a18fSJosef Bacik 35055d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, blocks); 35065d4f98a2SYan Zheng out: 35075d4f98a2SYan Zheng btrfs_free_path(path); 35085d4f98a2SYan Zheng return ret; 35095d4f98a2SYan Zheng } 35105d4f98a2SYan Zheng 35115d4f98a2SYan Zheng /* 35125d4f98a2SYan Zheng * helper to check if the block use full backrefs for pointers in it 35135d4f98a2SYan Zheng */ 35145d4f98a2SYan Zheng static int block_use_full_backref(struct reloc_control *rc, 35155d4f98a2SYan Zheng struct extent_buffer *eb) 35165d4f98a2SYan Zheng { 35175d4f98a2SYan Zheng u64 flags; 35185d4f98a2SYan Zheng int ret; 35195d4f98a2SYan Zheng 35205d4f98a2SYan Zheng if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) || 35215d4f98a2SYan Zheng btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV) 35225d4f98a2SYan Zheng return 1; 35235d4f98a2SYan Zheng 35242ff7e61eSJeff Mahoney ret = btrfs_lookup_extent_info(NULL, rc->extent_root->fs_info, 35253173a18fSJosef Bacik eb->start, btrfs_header_level(eb), 1, 35263173a18fSJosef Bacik NULL, &flags); 35275d4f98a2SYan Zheng BUG_ON(ret); 35285d4f98a2SYan Zheng 35295d4f98a2SYan Zheng if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) 35305d4f98a2SYan Zheng ret = 1; 35315d4f98a2SYan Zheng else 35325d4f98a2SYan Zheng ret = 0; 35335d4f98a2SYan Zheng return ret; 35345d4f98a2SYan Zheng } 35355d4f98a2SYan Zheng 35360af3d00bSJosef Bacik static int delete_block_group_cache(struct btrfs_fs_info *fs_info, 35371bbc621eSChris Mason struct btrfs_block_group_cache *block_group, 35381bbc621eSChris Mason struct inode *inode, 35391bbc621eSChris Mason u64 ino) 35400af3d00bSJosef Bacik { 35410af3d00bSJosef Bacik struct btrfs_key key; 35420af3d00bSJosef Bacik struct btrfs_root *root = fs_info->tree_root; 35430af3d00bSJosef Bacik struct btrfs_trans_handle *trans; 35440af3d00bSJosef Bacik int ret = 0; 35450af3d00bSJosef Bacik 35460af3d00bSJosef Bacik if (inode) 35470af3d00bSJosef Bacik goto truncate; 35480af3d00bSJosef Bacik 35490af3d00bSJosef Bacik key.objectid = ino; 35500af3d00bSJosef Bacik key.type = BTRFS_INODE_ITEM_KEY; 35510af3d00bSJosef Bacik key.offset = 0; 35520af3d00bSJosef Bacik 35530af3d00bSJosef Bacik inode = btrfs_iget(fs_info->sb, &key, root, NULL); 35542e19f1f9SAl Viro if (IS_ERR(inode)) 35550af3d00bSJosef Bacik return -ENOENT; 35560af3d00bSJosef Bacik 35570af3d00bSJosef Bacik truncate: 35582ff7e61eSJeff Mahoney ret = btrfs_check_trunc_cache_free_space(fs_info, 35597b61cd92SMiao Xie &fs_info->global_block_rsv); 35607b61cd92SMiao Xie if (ret) 35617b61cd92SMiao Xie goto out; 35627b61cd92SMiao Xie 35637a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 35640af3d00bSJosef Bacik if (IS_ERR(trans)) { 35653612b495STsutomu Itoh ret = PTR_ERR(trans); 35660af3d00bSJosef Bacik goto out; 35670af3d00bSJosef Bacik } 35680af3d00bSJosef Bacik 356977ab86bfSJeff Mahoney ret = btrfs_truncate_free_space_cache(trans, block_group, inode); 35700af3d00bSJosef Bacik 35713a45bb20SJeff Mahoney btrfs_end_transaction(trans); 35722ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 35730af3d00bSJosef Bacik out: 35740af3d00bSJosef Bacik iput(inode); 35750af3d00bSJosef Bacik return ret; 35760af3d00bSJosef Bacik } 35770af3d00bSJosef Bacik 35785d4f98a2SYan Zheng /* 35795d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY 35805d4f98a2SYan Zheng * this function scans fs tree to find blocks reference the data extent 35815d4f98a2SYan Zheng */ 35825d4f98a2SYan Zheng static int find_data_references(struct reloc_control *rc, 35835d4f98a2SYan Zheng struct btrfs_key *extent_key, 35845d4f98a2SYan Zheng struct extent_buffer *leaf, 35855d4f98a2SYan Zheng struct btrfs_extent_data_ref *ref, 35865d4f98a2SYan Zheng struct rb_root *blocks) 35875d4f98a2SYan Zheng { 35880b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 35895d4f98a2SYan Zheng struct btrfs_path *path; 35905d4f98a2SYan Zheng struct tree_block *block; 35915d4f98a2SYan Zheng struct btrfs_root *root; 35925d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 35935d4f98a2SYan Zheng struct rb_node *rb_node; 35945d4f98a2SYan Zheng struct btrfs_key key; 35955d4f98a2SYan Zheng u64 ref_root; 35965d4f98a2SYan Zheng u64 ref_objectid; 35975d4f98a2SYan Zheng u64 ref_offset; 35985d4f98a2SYan Zheng u32 ref_count; 35995d4f98a2SYan Zheng u32 nritems; 36005d4f98a2SYan Zheng int err = 0; 36015d4f98a2SYan Zheng int added = 0; 36025d4f98a2SYan Zheng int counted; 36035d4f98a2SYan Zheng int ret; 36045d4f98a2SYan Zheng 36055d4f98a2SYan Zheng ref_root = btrfs_extent_data_ref_root(leaf, ref); 36065d4f98a2SYan Zheng ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref); 36075d4f98a2SYan Zheng ref_offset = btrfs_extent_data_ref_offset(leaf, ref); 36085d4f98a2SYan Zheng ref_count = btrfs_extent_data_ref_count(leaf, ref); 36095d4f98a2SYan Zheng 36100af3d00bSJosef Bacik /* 36110af3d00bSJosef Bacik * This is an extent belonging to the free space cache, lets just delete 36120af3d00bSJosef Bacik * it and redo the search. 36130af3d00bSJosef Bacik */ 36140af3d00bSJosef Bacik if (ref_root == BTRFS_ROOT_TREE_OBJECTID) { 36150b246afaSJeff Mahoney ret = delete_block_group_cache(fs_info, rc->block_group, 36160af3d00bSJosef Bacik NULL, ref_objectid); 36170af3d00bSJosef Bacik if (ret != -ENOENT) 36180af3d00bSJosef Bacik return ret; 36190af3d00bSJosef Bacik ret = 0; 36200af3d00bSJosef Bacik } 36210af3d00bSJosef Bacik 36220af3d00bSJosef Bacik path = btrfs_alloc_path(); 36230af3d00bSJosef Bacik if (!path) 36240af3d00bSJosef Bacik return -ENOMEM; 3625e4058b54SDavid Sterba path->reada = READA_FORWARD; 36260af3d00bSJosef Bacik 36270b246afaSJeff Mahoney root = read_fs_root(fs_info, ref_root); 36285d4f98a2SYan Zheng if (IS_ERR(root)) { 36295d4f98a2SYan Zheng err = PTR_ERR(root); 36305d4f98a2SYan Zheng goto out; 36315d4f98a2SYan Zheng } 36325d4f98a2SYan Zheng 36335d4f98a2SYan Zheng key.objectid = ref_objectid; 36345d4f98a2SYan Zheng key.type = BTRFS_EXTENT_DATA_KEY; 363584850e8dSYan, Zheng if (ref_offset > ((u64)-1 << 32)) 363684850e8dSYan, Zheng key.offset = 0; 363784850e8dSYan, Zheng else 363884850e8dSYan, Zheng key.offset = ref_offset; 36395d4f98a2SYan Zheng 36405d4f98a2SYan Zheng path->search_commit_root = 1; 36415d4f98a2SYan Zheng path->skip_locking = 1; 36425d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 36435d4f98a2SYan Zheng if (ret < 0) { 36445d4f98a2SYan Zheng err = ret; 36455d4f98a2SYan Zheng goto out; 36465d4f98a2SYan Zheng } 36475d4f98a2SYan Zheng 36485d4f98a2SYan Zheng leaf = path->nodes[0]; 36495d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 36505d4f98a2SYan Zheng /* 36515d4f98a2SYan Zheng * the references in tree blocks that use full backrefs 36525d4f98a2SYan Zheng * are not counted in 36535d4f98a2SYan Zheng */ 36545d4f98a2SYan Zheng if (block_use_full_backref(rc, leaf)) 36555d4f98a2SYan Zheng counted = 0; 36565d4f98a2SYan Zheng else 36575d4f98a2SYan Zheng counted = 1; 36585d4f98a2SYan Zheng rb_node = tree_search(blocks, leaf->start); 36595d4f98a2SYan Zheng if (rb_node) { 36605d4f98a2SYan Zheng if (counted) 36615d4f98a2SYan Zheng added = 1; 36625d4f98a2SYan Zheng else 36635d4f98a2SYan Zheng path->slots[0] = nritems; 36645d4f98a2SYan Zheng } 36655d4f98a2SYan Zheng 36665d4f98a2SYan Zheng while (ref_count > 0) { 36675d4f98a2SYan Zheng while (path->slots[0] >= nritems) { 36685d4f98a2SYan Zheng ret = btrfs_next_leaf(root, path); 36695d4f98a2SYan Zheng if (ret < 0) { 36705d4f98a2SYan Zheng err = ret; 36715d4f98a2SYan Zheng goto out; 36725d4f98a2SYan Zheng } 3673fae7f21cSDulshani Gunawardhana if (WARN_ON(ret > 0)) 36745d4f98a2SYan Zheng goto out; 36755d4f98a2SYan Zheng 36765d4f98a2SYan Zheng leaf = path->nodes[0]; 36775d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 36785d4f98a2SYan Zheng added = 0; 36795d4f98a2SYan Zheng 36805d4f98a2SYan Zheng if (block_use_full_backref(rc, leaf)) 36815d4f98a2SYan Zheng counted = 0; 36825d4f98a2SYan Zheng else 36835d4f98a2SYan Zheng counted = 1; 36845d4f98a2SYan Zheng rb_node = tree_search(blocks, leaf->start); 36855d4f98a2SYan Zheng if (rb_node) { 36865d4f98a2SYan Zheng if (counted) 36875d4f98a2SYan Zheng added = 1; 36885d4f98a2SYan Zheng else 36895d4f98a2SYan Zheng path->slots[0] = nritems; 36905d4f98a2SYan Zheng } 36915d4f98a2SYan Zheng } 36925d4f98a2SYan Zheng 36935d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 3694fae7f21cSDulshani Gunawardhana if (WARN_ON(key.objectid != ref_objectid || 3695fae7f21cSDulshani Gunawardhana key.type != BTRFS_EXTENT_DATA_KEY)) 36965d4f98a2SYan Zheng break; 36975d4f98a2SYan Zheng 36985d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 36995d4f98a2SYan Zheng struct btrfs_file_extent_item); 37005d4f98a2SYan Zheng 37015d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 37025d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 37035d4f98a2SYan Zheng goto next; 37045d4f98a2SYan Zheng 37055d4f98a2SYan Zheng if (btrfs_file_extent_disk_bytenr(leaf, fi) != 37065d4f98a2SYan Zheng extent_key->objectid) 37075d4f98a2SYan Zheng goto next; 37085d4f98a2SYan Zheng 37095d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 37105d4f98a2SYan Zheng if (key.offset != ref_offset) 37115d4f98a2SYan Zheng goto next; 37125d4f98a2SYan Zheng 37135d4f98a2SYan Zheng if (counted) 37145d4f98a2SYan Zheng ref_count--; 37155d4f98a2SYan Zheng if (added) 37165d4f98a2SYan Zheng goto next; 37175d4f98a2SYan Zheng 37187476dfdaSDavid Sterba if (!tree_block_processed(leaf->start, rc)) { 37195d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 37205d4f98a2SYan Zheng if (!block) { 37215d4f98a2SYan Zheng err = -ENOMEM; 37225d4f98a2SYan Zheng break; 37235d4f98a2SYan Zheng } 37245d4f98a2SYan Zheng block->bytenr = leaf->start; 37255d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &block->key, 0); 37265d4f98a2SYan Zheng block->level = 0; 37275d4f98a2SYan Zheng block->key_ready = 1; 37285d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, 37295d4f98a2SYan Zheng &block->rb_node); 373043c04fb1SJeff Mahoney if (rb_node) 373143c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 373243c04fb1SJeff Mahoney block->bytenr); 37335d4f98a2SYan Zheng } 37345d4f98a2SYan Zheng if (counted) 37355d4f98a2SYan Zheng added = 1; 37365d4f98a2SYan Zheng else 37375d4f98a2SYan Zheng path->slots[0] = nritems; 37385d4f98a2SYan Zheng next: 37395d4f98a2SYan Zheng path->slots[0]++; 37405d4f98a2SYan Zheng 37415d4f98a2SYan Zheng } 37425d4f98a2SYan Zheng out: 37435d4f98a2SYan Zheng btrfs_free_path(path); 37445d4f98a2SYan Zheng return err; 37455d4f98a2SYan Zheng } 37465d4f98a2SYan Zheng 37475d4f98a2SYan Zheng /* 37482c016dc2SLiu Bo * helper to find all tree blocks that reference a given data extent 37495d4f98a2SYan Zheng */ 37505d4f98a2SYan Zheng static noinline_for_stack 37515d4f98a2SYan Zheng int add_data_references(struct reloc_control *rc, 37525d4f98a2SYan Zheng struct btrfs_key *extent_key, 37535d4f98a2SYan Zheng struct btrfs_path *path, 37545d4f98a2SYan Zheng struct rb_root *blocks) 37555d4f98a2SYan Zheng { 37565d4f98a2SYan Zheng struct btrfs_key key; 37575d4f98a2SYan Zheng struct extent_buffer *eb; 37585d4f98a2SYan Zheng struct btrfs_extent_data_ref *dref; 37595d4f98a2SYan Zheng struct btrfs_extent_inline_ref *iref; 37605d4f98a2SYan Zheng unsigned long ptr; 37615d4f98a2SYan Zheng unsigned long end; 3762da17066cSJeff Mahoney u32 blocksize = rc->extent_root->fs_info->nodesize; 3763647f63bdSFilipe David Borba Manana int ret = 0; 37645d4f98a2SYan Zheng int err = 0; 37655d4f98a2SYan Zheng 37665d4f98a2SYan Zheng eb = path->nodes[0]; 37675d4f98a2SYan Zheng ptr = btrfs_item_ptr_offset(eb, path->slots[0]); 37685d4f98a2SYan Zheng end = ptr + btrfs_item_size_nr(eb, path->slots[0]); 37695d4f98a2SYan Zheng ptr += sizeof(struct btrfs_extent_item); 37705d4f98a2SYan Zheng 37715d4f98a2SYan Zheng while (ptr < end) { 37725d4f98a2SYan Zheng iref = (struct btrfs_extent_inline_ref *)ptr; 37733de28d57SLiu Bo key.type = btrfs_get_extent_inline_ref_type(eb, iref, 37743de28d57SLiu Bo BTRFS_REF_TYPE_DATA); 37755d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY) { 37765d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 37775d4f98a2SYan Zheng ret = __add_tree_block(rc, key.offset, blocksize, 37785d4f98a2SYan Zheng blocks); 37795d4f98a2SYan Zheng } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { 37805d4f98a2SYan Zheng dref = (struct btrfs_extent_data_ref *)(&iref->offset); 37815d4f98a2SYan Zheng ret = find_data_references(rc, extent_key, 37825d4f98a2SYan Zheng eb, dref, blocks); 37835d4f98a2SYan Zheng } else { 3784af431dcbSSu Yue ret = -EUCLEAN; 3785b14c55a1SLiu Bo btrfs_err(rc->extent_root->fs_info, 3786b14c55a1SLiu Bo "extent %llu slot %d has an invalid inline ref type", 3787b14c55a1SLiu Bo eb->start, path->slots[0]); 37885d4f98a2SYan Zheng } 3789647f63bdSFilipe David Borba Manana if (ret) { 3790647f63bdSFilipe David Borba Manana err = ret; 3791647f63bdSFilipe David Borba Manana goto out; 3792647f63bdSFilipe David Borba Manana } 37935d4f98a2SYan Zheng ptr += btrfs_extent_inline_ref_size(key.type); 37945d4f98a2SYan Zheng } 37955d4f98a2SYan Zheng WARN_ON(ptr > end); 37965d4f98a2SYan Zheng 37975d4f98a2SYan Zheng while (1) { 37985d4f98a2SYan Zheng cond_resched(); 37995d4f98a2SYan Zheng eb = path->nodes[0]; 38005d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(eb)) { 38015d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 38025d4f98a2SYan Zheng if (ret < 0) { 38035d4f98a2SYan Zheng err = ret; 38045d4f98a2SYan Zheng break; 38055d4f98a2SYan Zheng } 38065d4f98a2SYan Zheng if (ret > 0) 38075d4f98a2SYan Zheng break; 38085d4f98a2SYan Zheng eb = path->nodes[0]; 38095d4f98a2SYan Zheng } 38105d4f98a2SYan Zheng 38115d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &key, path->slots[0]); 38125d4f98a2SYan Zheng if (key.objectid != extent_key->objectid) 38135d4f98a2SYan Zheng break; 38145d4f98a2SYan Zheng 38155d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY) { 38165d4f98a2SYan Zheng ret = __add_tree_block(rc, key.offset, blocksize, 38175d4f98a2SYan Zheng blocks); 38185d4f98a2SYan Zheng } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { 38195d4f98a2SYan Zheng dref = btrfs_item_ptr(eb, path->slots[0], 38205d4f98a2SYan Zheng struct btrfs_extent_data_ref); 38215d4f98a2SYan Zheng ret = find_data_references(rc, extent_key, 38225d4f98a2SYan Zheng eb, dref, blocks); 38236d8ff4e4SDavid Sterba } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { 3824ba3c2b19SNikolay Borisov btrfs_print_v0_err(eb->fs_info); 3825ba3c2b19SNikolay Borisov btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL); 3826ba3c2b19SNikolay Borisov ret = -EINVAL; 38275d4f98a2SYan Zheng } else { 38285d4f98a2SYan Zheng ret = 0; 38295d4f98a2SYan Zheng } 38305d4f98a2SYan Zheng if (ret) { 38315d4f98a2SYan Zheng err = ret; 38325d4f98a2SYan Zheng break; 38335d4f98a2SYan Zheng } 38345d4f98a2SYan Zheng path->slots[0]++; 38355d4f98a2SYan Zheng } 3836647f63bdSFilipe David Borba Manana out: 3837b3b4aa74SDavid Sterba btrfs_release_path(path); 38385d4f98a2SYan Zheng if (err) 38395d4f98a2SYan Zheng free_block_list(blocks); 38405d4f98a2SYan Zheng return err; 38415d4f98a2SYan Zheng } 38425d4f98a2SYan Zheng 38435d4f98a2SYan Zheng /* 38442c016dc2SLiu Bo * helper to find next unprocessed extent 38455d4f98a2SYan Zheng */ 38465d4f98a2SYan Zheng static noinline_for_stack 3847147d256eSZhaolei int find_next_extent(struct reloc_control *rc, struct btrfs_path *path, 38483fd0a558SYan, Zheng struct btrfs_key *extent_key) 38495d4f98a2SYan Zheng { 38500b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 38515d4f98a2SYan Zheng struct btrfs_key key; 38525d4f98a2SYan Zheng struct extent_buffer *leaf; 38535d4f98a2SYan Zheng u64 start, end, last; 38545d4f98a2SYan Zheng int ret; 38555d4f98a2SYan Zheng 38565d4f98a2SYan Zheng last = rc->block_group->key.objectid + rc->block_group->key.offset; 38575d4f98a2SYan Zheng while (1) { 38585d4f98a2SYan Zheng cond_resched(); 38595d4f98a2SYan Zheng if (rc->search_start >= last) { 38605d4f98a2SYan Zheng ret = 1; 38615d4f98a2SYan Zheng break; 38625d4f98a2SYan Zheng } 38635d4f98a2SYan Zheng 38645d4f98a2SYan Zheng key.objectid = rc->search_start; 38655d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 38665d4f98a2SYan Zheng key.offset = 0; 38675d4f98a2SYan Zheng 38685d4f98a2SYan Zheng path->search_commit_root = 1; 38695d4f98a2SYan Zheng path->skip_locking = 1; 38705d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 38715d4f98a2SYan Zheng 0, 0); 38725d4f98a2SYan Zheng if (ret < 0) 38735d4f98a2SYan Zheng break; 38745d4f98a2SYan Zheng next: 38755d4f98a2SYan Zheng leaf = path->nodes[0]; 38765d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(leaf)) { 38775d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 38785d4f98a2SYan Zheng if (ret != 0) 38795d4f98a2SYan Zheng break; 38805d4f98a2SYan Zheng leaf = path->nodes[0]; 38815d4f98a2SYan Zheng } 38825d4f98a2SYan Zheng 38835d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 38845d4f98a2SYan Zheng if (key.objectid >= last) { 38855d4f98a2SYan Zheng ret = 1; 38865d4f98a2SYan Zheng break; 38875d4f98a2SYan Zheng } 38885d4f98a2SYan Zheng 38893173a18fSJosef Bacik if (key.type != BTRFS_EXTENT_ITEM_KEY && 38903173a18fSJosef Bacik key.type != BTRFS_METADATA_ITEM_KEY) { 38913173a18fSJosef Bacik path->slots[0]++; 38923173a18fSJosef Bacik goto next; 38933173a18fSJosef Bacik } 38943173a18fSJosef Bacik 38953173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 38965d4f98a2SYan Zheng key.objectid + key.offset <= rc->search_start) { 38975d4f98a2SYan Zheng path->slots[0]++; 38985d4f98a2SYan Zheng goto next; 38995d4f98a2SYan Zheng } 39005d4f98a2SYan Zheng 39013173a18fSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 39020b246afaSJeff Mahoney key.objectid + fs_info->nodesize <= 39033173a18fSJosef Bacik rc->search_start) { 39043173a18fSJosef Bacik path->slots[0]++; 39053173a18fSJosef Bacik goto next; 39063173a18fSJosef Bacik } 39073173a18fSJosef Bacik 39085d4f98a2SYan Zheng ret = find_first_extent_bit(&rc->processed_blocks, 39095d4f98a2SYan Zheng key.objectid, &start, &end, 3910e6138876SJosef Bacik EXTENT_DIRTY, NULL); 39115d4f98a2SYan Zheng 39125d4f98a2SYan Zheng if (ret == 0 && start <= key.objectid) { 3913b3b4aa74SDavid Sterba btrfs_release_path(path); 39145d4f98a2SYan Zheng rc->search_start = end + 1; 39155d4f98a2SYan Zheng } else { 39163173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) 39175d4f98a2SYan Zheng rc->search_start = key.objectid + key.offset; 39183173a18fSJosef Bacik else 39193173a18fSJosef Bacik rc->search_start = key.objectid + 39200b246afaSJeff Mahoney fs_info->nodesize; 39213fd0a558SYan, Zheng memcpy(extent_key, &key, sizeof(key)); 39225d4f98a2SYan Zheng return 0; 39235d4f98a2SYan Zheng } 39245d4f98a2SYan Zheng } 3925b3b4aa74SDavid Sterba btrfs_release_path(path); 39265d4f98a2SYan Zheng return ret; 39275d4f98a2SYan Zheng } 39285d4f98a2SYan Zheng 39295d4f98a2SYan Zheng static void set_reloc_control(struct reloc_control *rc) 39305d4f98a2SYan Zheng { 39315d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 39327585717fSChris Mason 39337585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 39345d4f98a2SYan Zheng fs_info->reloc_ctl = rc; 39357585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 39365d4f98a2SYan Zheng } 39375d4f98a2SYan Zheng 39385d4f98a2SYan Zheng static void unset_reloc_control(struct reloc_control *rc) 39395d4f98a2SYan Zheng { 39405d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 39417585717fSChris Mason 39427585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 39435d4f98a2SYan Zheng fs_info->reloc_ctl = NULL; 39447585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 39455d4f98a2SYan Zheng } 39465d4f98a2SYan Zheng 39475d4f98a2SYan Zheng static int check_extent_flags(u64 flags) 39485d4f98a2SYan Zheng { 39495d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 39505d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 39515d4f98a2SYan Zheng return 1; 39525d4f98a2SYan Zheng if (!(flags & BTRFS_EXTENT_FLAG_DATA) && 39535d4f98a2SYan Zheng !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 39545d4f98a2SYan Zheng return 1; 39555d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 39565d4f98a2SYan Zheng (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) 39575d4f98a2SYan Zheng return 1; 39585d4f98a2SYan Zheng return 0; 39595d4f98a2SYan Zheng } 39605d4f98a2SYan Zheng 39613fd0a558SYan, Zheng static noinline_for_stack 39623fd0a558SYan, Zheng int prepare_to_relocate(struct reloc_control *rc) 39633fd0a558SYan, Zheng { 39643fd0a558SYan, Zheng struct btrfs_trans_handle *trans; 3965ac2fabacSJosef Bacik int ret; 39663fd0a558SYan, Zheng 39672ff7e61eSJeff Mahoney rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info, 396866d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 39693fd0a558SYan, Zheng if (!rc->block_rsv) 39703fd0a558SYan, Zheng return -ENOMEM; 39713fd0a558SYan, Zheng 39723fd0a558SYan, Zheng memset(&rc->cluster, 0, sizeof(rc->cluster)); 39733fd0a558SYan, Zheng rc->search_start = rc->block_group->key.objectid; 39743fd0a558SYan, Zheng rc->extents_found = 0; 39753fd0a558SYan, Zheng rc->nodes_relocated = 0; 39763fd0a558SYan, Zheng rc->merging_rsv_size = 0; 39770647bf56SWang Shilong rc->reserved_bytes = 0; 3978da17066cSJeff Mahoney rc->block_rsv->size = rc->extent_root->fs_info->nodesize * 39790647bf56SWang Shilong RELOCATION_RESERVED_NODES; 3980ac2fabacSJosef Bacik ret = btrfs_block_rsv_refill(rc->extent_root, 3981ac2fabacSJosef Bacik rc->block_rsv, rc->block_rsv->size, 3982ac2fabacSJosef Bacik BTRFS_RESERVE_FLUSH_ALL); 3983ac2fabacSJosef Bacik if (ret) 3984ac2fabacSJosef Bacik return ret; 39853fd0a558SYan, Zheng 39863fd0a558SYan, Zheng rc->create_reloc_tree = 1; 39873fd0a558SYan, Zheng set_reloc_control(rc); 39883fd0a558SYan, Zheng 39897a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 399028818947SLiu Bo if (IS_ERR(trans)) { 399128818947SLiu Bo unset_reloc_control(rc); 399228818947SLiu Bo /* 399328818947SLiu Bo * extent tree is not a ref_cow tree and has no reloc_root to 399428818947SLiu Bo * cleanup. And callers are responsible to free the above 399528818947SLiu Bo * block rsv. 399628818947SLiu Bo */ 399728818947SLiu Bo return PTR_ERR(trans); 399828818947SLiu Bo } 39993a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 40003fd0a558SYan, Zheng return 0; 40013fd0a558SYan, Zheng } 400276dda93cSYan, Zheng 40035d4f98a2SYan Zheng static noinline_for_stack int relocate_block_group(struct reloc_control *rc) 40045d4f98a2SYan Zheng { 40052ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 40065d4f98a2SYan Zheng struct rb_root blocks = RB_ROOT; 40075d4f98a2SYan Zheng struct btrfs_key key; 40085d4f98a2SYan Zheng struct btrfs_trans_handle *trans = NULL; 40095d4f98a2SYan Zheng struct btrfs_path *path; 40105d4f98a2SYan Zheng struct btrfs_extent_item *ei; 40115d4f98a2SYan Zheng u64 flags; 40125d4f98a2SYan Zheng u32 item_size; 40135d4f98a2SYan Zheng int ret; 40145d4f98a2SYan Zheng int err = 0; 4015c87f08caSChris Mason int progress = 0; 40165d4f98a2SYan Zheng 40175d4f98a2SYan Zheng path = btrfs_alloc_path(); 40183fd0a558SYan, Zheng if (!path) 40195d4f98a2SYan Zheng return -ENOMEM; 4020e4058b54SDavid Sterba path->reada = READA_FORWARD; 40213fd0a558SYan, Zheng 40223fd0a558SYan, Zheng ret = prepare_to_relocate(rc); 40233fd0a558SYan, Zheng if (ret) { 40243fd0a558SYan, Zheng err = ret; 40253fd0a558SYan, Zheng goto out_free; 40262423fdfbSJiri Slaby } 40275d4f98a2SYan Zheng 40285d4f98a2SYan Zheng while (1) { 40290647bf56SWang Shilong rc->reserved_bytes = 0; 40300647bf56SWang Shilong ret = btrfs_block_rsv_refill(rc->extent_root, 40310647bf56SWang Shilong rc->block_rsv, rc->block_rsv->size, 40320647bf56SWang Shilong BTRFS_RESERVE_FLUSH_ALL); 40330647bf56SWang Shilong if (ret) { 40340647bf56SWang Shilong err = ret; 40350647bf56SWang Shilong break; 40360647bf56SWang Shilong } 4037c87f08caSChris Mason progress++; 4038a22285a6SYan, Zheng trans = btrfs_start_transaction(rc->extent_root, 0); 40390f788c58SLiu Bo if (IS_ERR(trans)) { 40400f788c58SLiu Bo err = PTR_ERR(trans); 40410f788c58SLiu Bo trans = NULL; 40420f788c58SLiu Bo break; 40430f788c58SLiu Bo } 4044c87f08caSChris Mason restart: 40453fd0a558SYan, Zheng if (update_backref_cache(trans, &rc->backref_cache)) { 40463a45bb20SJeff Mahoney btrfs_end_transaction(trans); 404742a657f5SPan Bian trans = NULL; 40483fd0a558SYan, Zheng continue; 40493fd0a558SYan, Zheng } 40503fd0a558SYan, Zheng 4051147d256eSZhaolei ret = find_next_extent(rc, path, &key); 40525d4f98a2SYan Zheng if (ret < 0) 40535d4f98a2SYan Zheng err = ret; 40545d4f98a2SYan Zheng if (ret != 0) 40555d4f98a2SYan Zheng break; 40565d4f98a2SYan Zheng 40575d4f98a2SYan Zheng rc->extents_found++; 40585d4f98a2SYan Zheng 40595d4f98a2SYan Zheng ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 40605d4f98a2SYan Zheng struct btrfs_extent_item); 40613fd0a558SYan, Zheng item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); 40625d4f98a2SYan Zheng if (item_size >= sizeof(*ei)) { 40635d4f98a2SYan Zheng flags = btrfs_extent_flags(path->nodes[0], ei); 40645d4f98a2SYan Zheng ret = check_extent_flags(flags); 40655d4f98a2SYan Zheng BUG_ON(ret); 40666d8ff4e4SDavid Sterba } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) { 4067ba3c2b19SNikolay Borisov err = -EINVAL; 4068ba3c2b19SNikolay Borisov btrfs_print_v0_err(trans->fs_info); 4069ba3c2b19SNikolay Borisov btrfs_abort_transaction(trans, err); 4070ba3c2b19SNikolay Borisov break; 40715d4f98a2SYan Zheng } else { 40725d4f98a2SYan Zheng BUG(); 40735d4f98a2SYan Zheng } 40745d4f98a2SYan Zheng 40755d4f98a2SYan Zheng if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 40765d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, &blocks); 40775d4f98a2SYan Zheng } else if (rc->stage == UPDATE_DATA_PTRS && 40785d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 40795d4f98a2SYan Zheng ret = add_data_references(rc, &key, path, &blocks); 40805d4f98a2SYan Zheng } else { 4081b3b4aa74SDavid Sterba btrfs_release_path(path); 40825d4f98a2SYan Zheng ret = 0; 40835d4f98a2SYan Zheng } 40845d4f98a2SYan Zheng if (ret < 0) { 40853fd0a558SYan, Zheng err = ret; 40865d4f98a2SYan Zheng break; 40875d4f98a2SYan Zheng } 40885d4f98a2SYan Zheng 40895d4f98a2SYan Zheng if (!RB_EMPTY_ROOT(&blocks)) { 40905d4f98a2SYan Zheng ret = relocate_tree_blocks(trans, rc, &blocks); 40915d4f98a2SYan Zheng if (ret < 0) { 40921708cc57SWang Shilong /* 40931708cc57SWang Shilong * if we fail to relocate tree blocks, force to update 40941708cc57SWang Shilong * backref cache when committing transaction. 40951708cc57SWang Shilong */ 40961708cc57SWang Shilong rc->backref_cache.last_trans = trans->transid - 1; 40971708cc57SWang Shilong 40983fd0a558SYan, Zheng if (ret != -EAGAIN) { 40995d4f98a2SYan Zheng err = ret; 41005d4f98a2SYan Zheng break; 41015d4f98a2SYan Zheng } 41023fd0a558SYan, Zheng rc->extents_found--; 41033fd0a558SYan, Zheng rc->search_start = key.objectid; 41043fd0a558SYan, Zheng } 41055d4f98a2SYan Zheng } 41065d4f98a2SYan Zheng 41073a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 41082ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 41093fd0a558SYan, Zheng trans = NULL; 41105d4f98a2SYan Zheng 41115d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && 41125d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 41135d4f98a2SYan Zheng rc->found_file_extent = 1; 41140257bb82SYan, Zheng ret = relocate_data_extent(rc->data_inode, 41153fd0a558SYan, Zheng &key, &rc->cluster); 41165d4f98a2SYan Zheng if (ret < 0) { 41175d4f98a2SYan Zheng err = ret; 41185d4f98a2SYan Zheng break; 41195d4f98a2SYan Zheng } 41205d4f98a2SYan Zheng } 41215d4f98a2SYan Zheng } 4122c87f08caSChris Mason if (trans && progress && err == -ENOSPC) { 412343a7e99dSNikolay Borisov ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags); 41249689457bSShilong Wang if (ret == 1) { 4125c87f08caSChris Mason err = 0; 4126c87f08caSChris Mason progress = 0; 4127c87f08caSChris Mason goto restart; 4128c87f08caSChris Mason } 4129c87f08caSChris Mason } 41303fd0a558SYan, Zheng 4131b3b4aa74SDavid Sterba btrfs_release_path(path); 413291166212SDavid Sterba clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY); 41335d4f98a2SYan Zheng 41345d4f98a2SYan Zheng if (trans) { 41353a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 41362ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 41375d4f98a2SYan Zheng } 41385d4f98a2SYan Zheng 41390257bb82SYan, Zheng if (!err) { 41403fd0a558SYan, Zheng ret = relocate_file_extent_cluster(rc->data_inode, 41413fd0a558SYan, Zheng &rc->cluster); 41420257bb82SYan, Zheng if (ret < 0) 41430257bb82SYan, Zheng err = ret; 41440257bb82SYan, Zheng } 41450257bb82SYan, Zheng 41463fd0a558SYan, Zheng rc->create_reloc_tree = 0; 41473fd0a558SYan, Zheng set_reloc_control(rc); 41480257bb82SYan, Zheng 41493fd0a558SYan, Zheng backref_cache_cleanup(&rc->backref_cache); 41502ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1); 41515d4f98a2SYan Zheng 41523fd0a558SYan, Zheng err = prepare_to_merge(rc, err); 41535d4f98a2SYan Zheng 41545d4f98a2SYan Zheng merge_reloc_roots(rc); 41555d4f98a2SYan Zheng 41563fd0a558SYan, Zheng rc->merge_reloc_tree = 0; 41575d4f98a2SYan Zheng unset_reloc_control(rc); 41582ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1); 41595d4f98a2SYan Zheng 41605d4f98a2SYan Zheng /* get rid of pinned extents */ 41617a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 416262b99540SQu Wenruo if (IS_ERR(trans)) { 41633612b495STsutomu Itoh err = PTR_ERR(trans); 416462b99540SQu Wenruo goto out_free; 416562b99540SQu Wenruo } 41663a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 4167d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 4168d2311e69SQu Wenruo if (ret < 0 && !err) 4169d2311e69SQu Wenruo err = ret; 41703fd0a558SYan, Zheng out_free: 41712ff7e61eSJeff Mahoney btrfs_free_block_rsv(fs_info, rc->block_rsv); 41723fd0a558SYan, Zheng btrfs_free_path(path); 41735d4f98a2SYan Zheng return err; 41745d4f98a2SYan Zheng } 41755d4f98a2SYan Zheng 41765d4f98a2SYan Zheng static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 41770257bb82SYan, Zheng struct btrfs_root *root, u64 objectid) 41785d4f98a2SYan Zheng { 41795d4f98a2SYan Zheng struct btrfs_path *path; 41805d4f98a2SYan Zheng struct btrfs_inode_item *item; 41815d4f98a2SYan Zheng struct extent_buffer *leaf; 41825d4f98a2SYan Zheng int ret; 41835d4f98a2SYan Zheng 41845d4f98a2SYan Zheng path = btrfs_alloc_path(); 41855d4f98a2SYan Zheng if (!path) 41865d4f98a2SYan Zheng return -ENOMEM; 41875d4f98a2SYan Zheng 41885d4f98a2SYan Zheng ret = btrfs_insert_empty_inode(trans, root, path, objectid); 41895d4f98a2SYan Zheng if (ret) 41905d4f98a2SYan Zheng goto out; 41915d4f98a2SYan Zheng 41925d4f98a2SYan Zheng leaf = path->nodes[0]; 41935d4f98a2SYan Zheng item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); 4194b159fa28SDavid Sterba memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 41955d4f98a2SYan Zheng btrfs_set_inode_generation(leaf, item, 1); 41960257bb82SYan, Zheng btrfs_set_inode_size(leaf, item, 0); 41975d4f98a2SYan Zheng btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); 41983fd0a558SYan, Zheng btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS | 41993fd0a558SYan, Zheng BTRFS_INODE_PREALLOC); 42005d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 42015d4f98a2SYan Zheng out: 42025d4f98a2SYan Zheng btrfs_free_path(path); 42035d4f98a2SYan Zheng return ret; 42045d4f98a2SYan Zheng } 42055d4f98a2SYan Zheng 42065d4f98a2SYan Zheng /* 42075d4f98a2SYan Zheng * helper to create inode for data relocation. 42085d4f98a2SYan Zheng * the inode is in data relocation tree and its link count is 0 42095d4f98a2SYan Zheng */ 42103fd0a558SYan, Zheng static noinline_for_stack 42113fd0a558SYan, Zheng struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, 42125d4f98a2SYan Zheng struct btrfs_block_group_cache *group) 42135d4f98a2SYan Zheng { 42145d4f98a2SYan Zheng struct inode *inode = NULL; 42155d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 42165d4f98a2SYan Zheng struct btrfs_root *root; 42175d4f98a2SYan Zheng struct btrfs_key key; 42184624900dSZhaolei u64 objectid; 42195d4f98a2SYan Zheng int err = 0; 42205d4f98a2SYan Zheng 42215d4f98a2SYan Zheng root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 42225d4f98a2SYan Zheng if (IS_ERR(root)) 42235d4f98a2SYan Zheng return ERR_CAST(root); 42245d4f98a2SYan Zheng 4225a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 6); 42263fd0a558SYan, Zheng if (IS_ERR(trans)) 42273fd0a558SYan, Zheng return ERR_CAST(trans); 42285d4f98a2SYan Zheng 4229581bb050SLi Zefan err = btrfs_find_free_objectid(root, &objectid); 42305d4f98a2SYan Zheng if (err) 42315d4f98a2SYan Zheng goto out; 42325d4f98a2SYan Zheng 42330257bb82SYan, Zheng err = __insert_orphan_inode(trans, root, objectid); 42345d4f98a2SYan Zheng BUG_ON(err); 42355d4f98a2SYan Zheng 42365d4f98a2SYan Zheng key.objectid = objectid; 42375d4f98a2SYan Zheng key.type = BTRFS_INODE_ITEM_KEY; 42385d4f98a2SYan Zheng key.offset = 0; 42390b246afaSJeff Mahoney inode = btrfs_iget(fs_info->sb, &key, root, NULL); 42402e19f1f9SAl Viro BUG_ON(IS_ERR(inode)); 42415d4f98a2SYan Zheng BTRFS_I(inode)->index_cnt = group->key.objectid; 42425d4f98a2SYan Zheng 424373f2e545SNikolay Borisov err = btrfs_orphan_add(trans, BTRFS_I(inode)); 42445d4f98a2SYan Zheng out: 42453a45bb20SJeff Mahoney btrfs_end_transaction(trans); 42462ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 42475d4f98a2SYan Zheng if (err) { 42485d4f98a2SYan Zheng if (inode) 42495d4f98a2SYan Zheng iput(inode); 42505d4f98a2SYan Zheng inode = ERR_PTR(err); 42515d4f98a2SYan Zheng } 42525d4f98a2SYan Zheng return inode; 42535d4f98a2SYan Zheng } 42545d4f98a2SYan Zheng 4255c258d6e3SQu Wenruo static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info) 42563fd0a558SYan, Zheng { 42573fd0a558SYan, Zheng struct reloc_control *rc; 42583fd0a558SYan, Zheng 42593fd0a558SYan, Zheng rc = kzalloc(sizeof(*rc), GFP_NOFS); 42603fd0a558SYan, Zheng if (!rc) 42613fd0a558SYan, Zheng return NULL; 42623fd0a558SYan, Zheng 42633fd0a558SYan, Zheng INIT_LIST_HEAD(&rc->reloc_roots); 4264d2311e69SQu Wenruo INIT_LIST_HEAD(&rc->dirty_subvol_roots); 42653fd0a558SYan, Zheng backref_cache_init(&rc->backref_cache); 42663fd0a558SYan, Zheng mapping_tree_init(&rc->reloc_root_tree); 426743eb5f29SQu Wenruo extent_io_tree_init(fs_info, &rc->processed_blocks, 426843eb5f29SQu Wenruo IO_TREE_RELOC_BLOCKS, NULL); 42693fd0a558SYan, Zheng return rc; 42703fd0a558SYan, Zheng } 42713fd0a558SYan, Zheng 42725d4f98a2SYan Zheng /* 4273ebce0e01SAdam Borowski * Print the block group being relocated 4274ebce0e01SAdam Borowski */ 4275ebce0e01SAdam Borowski static void describe_relocation(struct btrfs_fs_info *fs_info, 4276ebce0e01SAdam Borowski struct btrfs_block_group_cache *block_group) 4277ebce0e01SAdam Borowski { 4278f89e09cfSAnand Jain char buf[128] = {'\0'}; 4279ebce0e01SAdam Borowski 4280f89e09cfSAnand Jain btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf)); 4281ebce0e01SAdam Borowski 4282ebce0e01SAdam Borowski btrfs_info(fs_info, 4283ebce0e01SAdam Borowski "relocating block group %llu flags %s", 4284f89e09cfSAnand Jain block_group->key.objectid, buf); 4285ebce0e01SAdam Borowski } 4286ebce0e01SAdam Borowski 4287ebce0e01SAdam Borowski /* 42885d4f98a2SYan Zheng * function to relocate all extents in a block group. 42895d4f98a2SYan Zheng */ 42906bccf3abSJeff Mahoney int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) 42915d4f98a2SYan Zheng { 4292eede2bf3SOmar Sandoval struct btrfs_block_group_cache *bg; 42936bccf3abSJeff Mahoney struct btrfs_root *extent_root = fs_info->extent_root; 42945d4f98a2SYan Zheng struct reloc_control *rc; 42950af3d00bSJosef Bacik struct inode *inode; 42960af3d00bSJosef Bacik struct btrfs_path *path; 42975d4f98a2SYan Zheng int ret; 4298f0486c68SYan, Zheng int rw = 0; 42995d4f98a2SYan Zheng int err = 0; 43005d4f98a2SYan Zheng 4301eede2bf3SOmar Sandoval bg = btrfs_lookup_block_group(fs_info, group_start); 4302eede2bf3SOmar Sandoval if (!bg) 4303eede2bf3SOmar Sandoval return -ENOENT; 4304eede2bf3SOmar Sandoval 4305eede2bf3SOmar Sandoval if (btrfs_pinned_by_swapfile(fs_info, bg)) { 4306eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 4307eede2bf3SOmar Sandoval return -ETXTBSY; 4308eede2bf3SOmar Sandoval } 4309eede2bf3SOmar Sandoval 4310c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 4311eede2bf3SOmar Sandoval if (!rc) { 4312eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 43135d4f98a2SYan Zheng return -ENOMEM; 4314eede2bf3SOmar Sandoval } 43155d4f98a2SYan Zheng 4316f0486c68SYan, Zheng rc->extent_root = extent_root; 4317eede2bf3SOmar Sandoval rc->block_group = bg; 43185d4f98a2SYan Zheng 4319c83488afSNikolay Borisov ret = btrfs_inc_block_group_ro(rc->block_group); 4320f0486c68SYan, Zheng if (ret) { 4321f0486c68SYan, Zheng err = ret; 4322f0486c68SYan, Zheng goto out; 4323f0486c68SYan, Zheng } 4324f0486c68SYan, Zheng rw = 1; 4325f0486c68SYan, Zheng 43260af3d00bSJosef Bacik path = btrfs_alloc_path(); 43270af3d00bSJosef Bacik if (!path) { 43280af3d00bSJosef Bacik err = -ENOMEM; 43290af3d00bSJosef Bacik goto out; 43300af3d00bSJosef Bacik } 43310af3d00bSJosef Bacik 43327949f339SDavid Sterba inode = lookup_free_space_inode(rc->block_group, path); 43330af3d00bSJosef Bacik btrfs_free_path(path); 43340af3d00bSJosef Bacik 43350af3d00bSJosef Bacik if (!IS_ERR(inode)) 43361bbc621eSChris Mason ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0); 43370af3d00bSJosef Bacik else 43380af3d00bSJosef Bacik ret = PTR_ERR(inode); 43390af3d00bSJosef Bacik 43400af3d00bSJosef Bacik if (ret && ret != -ENOENT) { 43410af3d00bSJosef Bacik err = ret; 43420af3d00bSJosef Bacik goto out; 43430af3d00bSJosef Bacik } 43440af3d00bSJosef Bacik 43455d4f98a2SYan Zheng rc->data_inode = create_reloc_inode(fs_info, rc->block_group); 43465d4f98a2SYan Zheng if (IS_ERR(rc->data_inode)) { 43475d4f98a2SYan Zheng err = PTR_ERR(rc->data_inode); 43485d4f98a2SYan Zheng rc->data_inode = NULL; 43495d4f98a2SYan Zheng goto out; 43505d4f98a2SYan Zheng } 43515d4f98a2SYan Zheng 43520b246afaSJeff Mahoney describe_relocation(fs_info, rc->block_group); 43535d4f98a2SYan Zheng 43549cfa3e34SFilipe Manana btrfs_wait_block_group_reservations(rc->block_group); 4355f78c436cSFilipe Manana btrfs_wait_nocow_writers(rc->block_group); 43566374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 4357578def7cSFilipe Manana rc->block_group->key.objectid, 4358578def7cSFilipe Manana rc->block_group->key.offset); 43595d4f98a2SYan Zheng 43605d4f98a2SYan Zheng while (1) { 436176dda93cSYan, Zheng mutex_lock(&fs_info->cleaner_mutex); 43625d4f98a2SYan Zheng ret = relocate_block_group(rc); 436376dda93cSYan, Zheng mutex_unlock(&fs_info->cleaner_mutex); 4364ff612ba7SJosef Bacik if (ret < 0) 43655d4f98a2SYan Zheng err = ret; 4366ff612ba7SJosef Bacik 4367ff612ba7SJosef Bacik /* 4368ff612ba7SJosef Bacik * We may have gotten ENOSPC after we already dirtied some 4369ff612ba7SJosef Bacik * extents. If writeout happens while we're relocating a 4370ff612ba7SJosef Bacik * different block group we could end up hitting the 4371ff612ba7SJosef Bacik * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in 4372ff612ba7SJosef Bacik * btrfs_reloc_cow_block. Make sure we write everything out 4373ff612ba7SJosef Bacik * properly so we don't trip over this problem, and then break 4374ff612ba7SJosef Bacik * out of the loop if we hit an error. 4375ff612ba7SJosef Bacik */ 4376ff612ba7SJosef Bacik if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) { 4377ff612ba7SJosef Bacik ret = btrfs_wait_ordered_range(rc->data_inode, 0, 4378ff612ba7SJosef Bacik (u64)-1); 4379ff612ba7SJosef Bacik if (ret) 4380ff612ba7SJosef Bacik err = ret; 4381ff612ba7SJosef Bacik invalidate_mapping_pages(rc->data_inode->i_mapping, 4382ff612ba7SJosef Bacik 0, -1); 4383ff612ba7SJosef Bacik rc->stage = UPDATE_DATA_PTRS; 43845d4f98a2SYan Zheng } 43855d4f98a2SYan Zheng 4386ff612ba7SJosef Bacik if (err < 0) 4387ff612ba7SJosef Bacik goto out; 4388ff612ba7SJosef Bacik 43895d4f98a2SYan Zheng if (rc->extents_found == 0) 43905d4f98a2SYan Zheng break; 43915d4f98a2SYan Zheng 43920b246afaSJeff Mahoney btrfs_info(fs_info, "found %llu extents", rc->extents_found); 43935d4f98a2SYan Zheng 43945d4f98a2SYan Zheng } 43955d4f98a2SYan Zheng 43965d4f98a2SYan Zheng WARN_ON(rc->block_group->pinned > 0); 43975d4f98a2SYan Zheng WARN_ON(rc->block_group->reserved > 0); 43985d4f98a2SYan Zheng WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0); 43995d4f98a2SYan Zheng out: 4400f0486c68SYan, Zheng if (err && rw) 44012ff7e61eSJeff Mahoney btrfs_dec_block_group_ro(rc->block_group); 44025d4f98a2SYan Zheng iput(rc->data_inode); 44035d4f98a2SYan Zheng btrfs_put_block_group(rc->block_group); 44045d4f98a2SYan Zheng kfree(rc); 44055d4f98a2SYan Zheng return err; 44065d4f98a2SYan Zheng } 44075d4f98a2SYan Zheng 440876dda93cSYan, Zheng static noinline_for_stack int mark_garbage_root(struct btrfs_root *root) 440976dda93cSYan, Zheng { 44100b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 441176dda93cSYan, Zheng struct btrfs_trans_handle *trans; 441279787eaaSJeff Mahoney int ret, err; 441376dda93cSYan, Zheng 44140b246afaSJeff Mahoney trans = btrfs_start_transaction(fs_info->tree_root, 0); 441579787eaaSJeff Mahoney if (IS_ERR(trans)) 441679787eaaSJeff Mahoney return PTR_ERR(trans); 441776dda93cSYan, Zheng 441876dda93cSYan, Zheng memset(&root->root_item.drop_progress, 0, 441976dda93cSYan, Zheng sizeof(root->root_item.drop_progress)); 442076dda93cSYan, Zheng root->root_item.drop_level = 0; 442176dda93cSYan, Zheng btrfs_set_root_refs(&root->root_item, 0); 44220b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 442376dda93cSYan, Zheng &root->root_key, &root->root_item); 442476dda93cSYan, Zheng 44253a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 442679787eaaSJeff Mahoney if (err) 442779787eaaSJeff Mahoney return err; 442879787eaaSJeff Mahoney return ret; 442976dda93cSYan, Zheng } 443076dda93cSYan, Zheng 44315d4f98a2SYan Zheng /* 44325d4f98a2SYan Zheng * recover relocation interrupted by system crash. 44335d4f98a2SYan Zheng * 44345d4f98a2SYan Zheng * this function resumes merging reloc trees with corresponding fs trees. 44355d4f98a2SYan Zheng * this is important for keeping the sharing of tree blocks 44365d4f98a2SYan Zheng */ 44375d4f98a2SYan Zheng int btrfs_recover_relocation(struct btrfs_root *root) 44385d4f98a2SYan Zheng { 44390b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 44405d4f98a2SYan Zheng LIST_HEAD(reloc_roots); 44415d4f98a2SYan Zheng struct btrfs_key key; 44425d4f98a2SYan Zheng struct btrfs_root *fs_root; 44435d4f98a2SYan Zheng struct btrfs_root *reloc_root; 44445d4f98a2SYan Zheng struct btrfs_path *path; 44455d4f98a2SYan Zheng struct extent_buffer *leaf; 44465d4f98a2SYan Zheng struct reloc_control *rc = NULL; 44475d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 44485d4f98a2SYan Zheng int ret; 44495d4f98a2SYan Zheng int err = 0; 44505d4f98a2SYan Zheng 44515d4f98a2SYan Zheng path = btrfs_alloc_path(); 44525d4f98a2SYan Zheng if (!path) 44535d4f98a2SYan Zheng return -ENOMEM; 4454e4058b54SDavid Sterba path->reada = READA_BACK; 44555d4f98a2SYan Zheng 44565d4f98a2SYan Zheng key.objectid = BTRFS_TREE_RELOC_OBJECTID; 44575d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 44585d4f98a2SYan Zheng key.offset = (u64)-1; 44595d4f98a2SYan Zheng 44605d4f98a2SYan Zheng while (1) { 44610b246afaSJeff Mahoney ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, 44625d4f98a2SYan Zheng path, 0, 0); 44635d4f98a2SYan Zheng if (ret < 0) { 44645d4f98a2SYan Zheng err = ret; 44655d4f98a2SYan Zheng goto out; 44665d4f98a2SYan Zheng } 44675d4f98a2SYan Zheng if (ret > 0) { 44685d4f98a2SYan Zheng if (path->slots[0] == 0) 44695d4f98a2SYan Zheng break; 44705d4f98a2SYan Zheng path->slots[0]--; 44715d4f98a2SYan Zheng } 44725d4f98a2SYan Zheng leaf = path->nodes[0]; 44735d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 4474b3b4aa74SDavid Sterba btrfs_release_path(path); 44755d4f98a2SYan Zheng 44765d4f98a2SYan Zheng if (key.objectid != BTRFS_TREE_RELOC_OBJECTID || 44775d4f98a2SYan Zheng key.type != BTRFS_ROOT_ITEM_KEY) 44785d4f98a2SYan Zheng break; 44795d4f98a2SYan Zheng 4480cb517eabSMiao Xie reloc_root = btrfs_read_fs_root(root, &key); 44815d4f98a2SYan Zheng if (IS_ERR(reloc_root)) { 44825d4f98a2SYan Zheng err = PTR_ERR(reloc_root); 44835d4f98a2SYan Zheng goto out; 44845d4f98a2SYan Zheng } 44855d4f98a2SYan Zheng 44865d4f98a2SYan Zheng list_add(&reloc_root->root_list, &reloc_roots); 44875d4f98a2SYan Zheng 44885d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 44890b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, 44905d4f98a2SYan Zheng reloc_root->root_key.offset); 44915d4f98a2SYan Zheng if (IS_ERR(fs_root)) { 449276dda93cSYan, Zheng ret = PTR_ERR(fs_root); 449376dda93cSYan, Zheng if (ret != -ENOENT) { 449476dda93cSYan, Zheng err = ret; 44955d4f98a2SYan Zheng goto out; 44965d4f98a2SYan Zheng } 449779787eaaSJeff Mahoney ret = mark_garbage_root(reloc_root); 449879787eaaSJeff Mahoney if (ret < 0) { 449979787eaaSJeff Mahoney err = ret; 450079787eaaSJeff Mahoney goto out; 450179787eaaSJeff Mahoney } 450276dda93cSYan, Zheng } 45035d4f98a2SYan Zheng } 45045d4f98a2SYan Zheng 45055d4f98a2SYan Zheng if (key.offset == 0) 45065d4f98a2SYan Zheng break; 45075d4f98a2SYan Zheng 45085d4f98a2SYan Zheng key.offset--; 45095d4f98a2SYan Zheng } 4510b3b4aa74SDavid Sterba btrfs_release_path(path); 45115d4f98a2SYan Zheng 45125d4f98a2SYan Zheng if (list_empty(&reloc_roots)) 45135d4f98a2SYan Zheng goto out; 45145d4f98a2SYan Zheng 4515c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 45165d4f98a2SYan Zheng if (!rc) { 45175d4f98a2SYan Zheng err = -ENOMEM; 45185d4f98a2SYan Zheng goto out; 45195d4f98a2SYan Zheng } 45205d4f98a2SYan Zheng 45210b246afaSJeff Mahoney rc->extent_root = fs_info->extent_root; 45225d4f98a2SYan Zheng 45235d4f98a2SYan Zheng set_reloc_control(rc); 45245d4f98a2SYan Zheng 45257a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 45263612b495STsutomu Itoh if (IS_ERR(trans)) { 45273612b495STsutomu Itoh unset_reloc_control(rc); 45283612b495STsutomu Itoh err = PTR_ERR(trans); 45293612b495STsutomu Itoh goto out_free; 45303612b495STsutomu Itoh } 45313fd0a558SYan, Zheng 45323fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 45333fd0a558SYan, Zheng 45345d4f98a2SYan Zheng while (!list_empty(&reloc_roots)) { 45355d4f98a2SYan Zheng reloc_root = list_entry(reloc_roots.next, 45365d4f98a2SYan Zheng struct btrfs_root, root_list); 45375d4f98a2SYan Zheng list_del(&reloc_root->root_list); 45385d4f98a2SYan Zheng 45395d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) == 0) { 45405d4f98a2SYan Zheng list_add_tail(&reloc_root->root_list, 45415d4f98a2SYan Zheng &rc->reloc_roots); 45425d4f98a2SYan Zheng continue; 45435d4f98a2SYan Zheng } 45445d4f98a2SYan Zheng 45450b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, reloc_root->root_key.offset); 454679787eaaSJeff Mahoney if (IS_ERR(fs_root)) { 454779787eaaSJeff Mahoney err = PTR_ERR(fs_root); 454879787eaaSJeff Mahoney goto out_free; 454979787eaaSJeff Mahoney } 45505d4f98a2SYan Zheng 4551ffd7b339SJeff Mahoney err = __add_reloc_root(reloc_root); 455279787eaaSJeff Mahoney BUG_ON(err < 0); /* -ENOMEM or logic error */ 45535d4f98a2SYan Zheng fs_root->reloc_root = reloc_root; 45545d4f98a2SYan Zheng } 45555d4f98a2SYan Zheng 45563a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 455779787eaaSJeff Mahoney if (err) 455879787eaaSJeff Mahoney goto out_free; 45595d4f98a2SYan Zheng 45605d4f98a2SYan Zheng merge_reloc_roots(rc); 45615d4f98a2SYan Zheng 45625d4f98a2SYan Zheng unset_reloc_control(rc); 45635d4f98a2SYan Zheng 45647a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 456562b99540SQu Wenruo if (IS_ERR(trans)) { 45663612b495STsutomu Itoh err = PTR_ERR(trans); 456762b99540SQu Wenruo goto out_free; 456862b99540SQu Wenruo } 45693a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 4570d2311e69SQu Wenruo 4571d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 4572d2311e69SQu Wenruo if (ret < 0 && !err) 4573d2311e69SQu Wenruo err = ret; 45743612b495STsutomu Itoh out_free: 45755d4f98a2SYan Zheng kfree(rc); 45763612b495STsutomu Itoh out: 4577aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 4578aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 4579aca1bba6SLiu Bo 45805d4f98a2SYan Zheng btrfs_free_path(path); 45815d4f98a2SYan Zheng 45825d4f98a2SYan Zheng if (err == 0) { 45835d4f98a2SYan Zheng /* cleanup orphan inode in data relocation tree */ 45840b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 45855d4f98a2SYan Zheng if (IS_ERR(fs_root)) 45865d4f98a2SYan Zheng err = PTR_ERR(fs_root); 4587d7ce5843SMiao Xie else 458866b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_root); 45895d4f98a2SYan Zheng } 45905d4f98a2SYan Zheng return err; 45915d4f98a2SYan Zheng } 45925d4f98a2SYan Zheng 45935d4f98a2SYan Zheng /* 45945d4f98a2SYan Zheng * helper to add ordered checksum for data relocation. 45955d4f98a2SYan Zheng * 45965d4f98a2SYan Zheng * cloning checksum properly handles the nodatasum extents. 45975d4f98a2SYan Zheng * it also saves CPU time to re-calculate the checksum. 45985d4f98a2SYan Zheng */ 45995d4f98a2SYan Zheng int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len) 46005d4f98a2SYan Zheng { 46010b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 46025d4f98a2SYan Zheng struct btrfs_ordered_sum *sums; 46035d4f98a2SYan Zheng struct btrfs_ordered_extent *ordered; 46045d4f98a2SYan Zheng int ret; 46055d4f98a2SYan Zheng u64 disk_bytenr; 46064577b014SJosef Bacik u64 new_bytenr; 46075d4f98a2SYan Zheng LIST_HEAD(list); 46085d4f98a2SYan Zheng 46095d4f98a2SYan Zheng ordered = btrfs_lookup_ordered_extent(inode, file_pos); 46105d4f98a2SYan Zheng BUG_ON(ordered->file_offset != file_pos || ordered->len != len); 46115d4f98a2SYan Zheng 46125d4f98a2SYan Zheng disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt; 46130b246afaSJeff Mahoney ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr, 4614a2de733cSArne Jansen disk_bytenr + len - 1, &list, 0); 461579787eaaSJeff Mahoney if (ret) 461679787eaaSJeff Mahoney goto out; 46175d4f98a2SYan Zheng 46185d4f98a2SYan Zheng while (!list_empty(&list)) { 46195d4f98a2SYan Zheng sums = list_entry(list.next, struct btrfs_ordered_sum, list); 46205d4f98a2SYan Zheng list_del_init(&sums->list); 46215d4f98a2SYan Zheng 46224577b014SJosef Bacik /* 46234577b014SJosef Bacik * We need to offset the new_bytenr based on where the csum is. 46244577b014SJosef Bacik * We need to do this because we will read in entire prealloc 46254577b014SJosef Bacik * extents but we may have written to say the middle of the 46264577b014SJosef Bacik * prealloc extent, so we need to make sure the csum goes with 46274577b014SJosef Bacik * the right disk offset. 46284577b014SJosef Bacik * 46294577b014SJosef Bacik * We can do this because the data reloc inode refers strictly 46304577b014SJosef Bacik * to the on disk bytes, so we don't have to worry about 46314577b014SJosef Bacik * disk_len vs real len like with real inodes since it's all 46324577b014SJosef Bacik * disk length. 46334577b014SJosef Bacik */ 46344577b014SJosef Bacik new_bytenr = ordered->start + (sums->bytenr - disk_bytenr); 46354577b014SJosef Bacik sums->bytenr = new_bytenr; 46365d4f98a2SYan Zheng 4637f9756261SNikolay Borisov btrfs_add_ordered_sum(ordered, sums); 46385d4f98a2SYan Zheng } 463979787eaaSJeff Mahoney out: 46405d4f98a2SYan Zheng btrfs_put_ordered_extent(ordered); 4641411fc6bcSAndi Kleen return ret; 46425d4f98a2SYan Zheng } 46433fd0a558SYan, Zheng 464483d4cfd4SJosef Bacik int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans, 46453fd0a558SYan, Zheng struct btrfs_root *root, struct extent_buffer *buf, 46463fd0a558SYan, Zheng struct extent_buffer *cow) 46473fd0a558SYan, Zheng { 46480b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 46493fd0a558SYan, Zheng struct reloc_control *rc; 46503fd0a558SYan, Zheng struct backref_node *node; 46513fd0a558SYan, Zheng int first_cow = 0; 46523fd0a558SYan, Zheng int level; 465383d4cfd4SJosef Bacik int ret = 0; 46543fd0a558SYan, Zheng 46550b246afaSJeff Mahoney rc = fs_info->reloc_ctl; 46563fd0a558SYan, Zheng if (!rc) 465783d4cfd4SJosef Bacik return 0; 46583fd0a558SYan, Zheng 46593fd0a558SYan, Zheng BUG_ON(rc->stage == UPDATE_DATA_PTRS && 46603fd0a558SYan, Zheng root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID); 46613fd0a558SYan, Zheng 4662c974c464SWang Shilong if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 4663c974c464SWang Shilong if (buf == root->node) 4664c974c464SWang Shilong __update_reloc_root(root, cow->start); 4665c974c464SWang Shilong } 4666c974c464SWang Shilong 46673fd0a558SYan, Zheng level = btrfs_header_level(buf); 46683fd0a558SYan, Zheng if (btrfs_header_generation(buf) <= 46693fd0a558SYan, Zheng btrfs_root_last_snapshot(&root->root_item)) 46703fd0a558SYan, Zheng first_cow = 1; 46713fd0a558SYan, Zheng 46723fd0a558SYan, Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID && 46733fd0a558SYan, Zheng rc->create_reloc_tree) { 46743fd0a558SYan, Zheng WARN_ON(!first_cow && level == 0); 46753fd0a558SYan, Zheng 46763fd0a558SYan, Zheng node = rc->backref_cache.path[level]; 46773fd0a558SYan, Zheng BUG_ON(node->bytenr != buf->start && 46783fd0a558SYan, Zheng node->new_bytenr != buf->start); 46793fd0a558SYan, Zheng 46803fd0a558SYan, Zheng drop_node_buffer(node); 46813fd0a558SYan, Zheng extent_buffer_get(cow); 46823fd0a558SYan, Zheng node->eb = cow; 46833fd0a558SYan, Zheng node->new_bytenr = cow->start; 46843fd0a558SYan, Zheng 46853fd0a558SYan, Zheng if (!node->pending) { 46863fd0a558SYan, Zheng list_move_tail(&node->list, 46873fd0a558SYan, Zheng &rc->backref_cache.pending[level]); 46883fd0a558SYan, Zheng node->pending = 1; 46893fd0a558SYan, Zheng } 46903fd0a558SYan, Zheng 46913fd0a558SYan, Zheng if (first_cow) 46923fd0a558SYan, Zheng __mark_block_processed(rc, node); 46933fd0a558SYan, Zheng 46943fd0a558SYan, Zheng if (first_cow && level > 0) 46953fd0a558SYan, Zheng rc->nodes_relocated += buf->len; 46963fd0a558SYan, Zheng } 46973fd0a558SYan, Zheng 469883d4cfd4SJosef Bacik if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) 46993fd0a558SYan, Zheng ret = replace_file_extents(trans, rc, root, cow); 470083d4cfd4SJosef Bacik return ret; 47013fd0a558SYan, Zheng } 47023fd0a558SYan, Zheng 47033fd0a558SYan, Zheng /* 47043fd0a558SYan, Zheng * called before creating snapshot. it calculates metadata reservation 470501327610SNicholas D Steeves * required for relocating tree blocks in the snapshot 47063fd0a558SYan, Zheng */ 4707147d256eSZhaolei void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending, 47083fd0a558SYan, Zheng u64 *bytes_to_reserve) 47093fd0a558SYan, Zheng { 471010995c04SQu Wenruo struct btrfs_root *root = pending->root; 471110995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 47123fd0a558SYan, Zheng 471310995c04SQu Wenruo if (!root->reloc_root || !rc) 47143fd0a558SYan, Zheng return; 47153fd0a558SYan, Zheng 47163fd0a558SYan, Zheng if (!rc->merge_reloc_tree) 47173fd0a558SYan, Zheng return; 47183fd0a558SYan, Zheng 47193fd0a558SYan, Zheng root = root->reloc_root; 47203fd0a558SYan, Zheng BUG_ON(btrfs_root_refs(&root->root_item) == 0); 47213fd0a558SYan, Zheng /* 47223fd0a558SYan, Zheng * relocation is in the stage of merging trees. the space 47233fd0a558SYan, Zheng * used by merging a reloc tree is twice the size of 47243fd0a558SYan, Zheng * relocated tree nodes in the worst case. half for cowing 47253fd0a558SYan, Zheng * the reloc tree, half for cowing the fs tree. the space 47263fd0a558SYan, Zheng * used by cowing the reloc tree will be freed after the 47273fd0a558SYan, Zheng * tree is dropped. if we create snapshot, cowing the fs 47283fd0a558SYan, Zheng * tree may use more space than it frees. so we need 47293fd0a558SYan, Zheng * reserve extra space. 47303fd0a558SYan, Zheng */ 47313fd0a558SYan, Zheng *bytes_to_reserve += rc->nodes_relocated; 47323fd0a558SYan, Zheng } 47333fd0a558SYan, Zheng 47343fd0a558SYan, Zheng /* 47353fd0a558SYan, Zheng * called after snapshot is created. migrate block reservation 47363fd0a558SYan, Zheng * and create reloc root for the newly created snapshot 47373fd0a558SYan, Zheng */ 473849b25e05SJeff Mahoney int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans, 47393fd0a558SYan, Zheng struct btrfs_pending_snapshot *pending) 47403fd0a558SYan, Zheng { 47413fd0a558SYan, Zheng struct btrfs_root *root = pending->root; 47423fd0a558SYan, Zheng struct btrfs_root *reloc_root; 47433fd0a558SYan, Zheng struct btrfs_root *new_root; 474410995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 47453fd0a558SYan, Zheng int ret; 47463fd0a558SYan, Zheng 474710995c04SQu Wenruo if (!root->reloc_root || !rc) 474849b25e05SJeff Mahoney return 0; 47493fd0a558SYan, Zheng 47503fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 47513fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated; 47523fd0a558SYan, Zheng 47533fd0a558SYan, Zheng if (rc->merge_reloc_tree) { 47543fd0a558SYan, Zheng ret = btrfs_block_rsv_migrate(&pending->block_rsv, 47553fd0a558SYan, Zheng rc->block_rsv, 47563a584174SLu Fengqi rc->nodes_relocated, true); 475749b25e05SJeff Mahoney if (ret) 475849b25e05SJeff Mahoney return ret; 47593fd0a558SYan, Zheng } 47603fd0a558SYan, Zheng 47613fd0a558SYan, Zheng new_root = pending->snap; 47623fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root->reloc_root, 47633fd0a558SYan, Zheng new_root->root_key.objectid); 476449b25e05SJeff Mahoney if (IS_ERR(reloc_root)) 476549b25e05SJeff Mahoney return PTR_ERR(reloc_root); 47663fd0a558SYan, Zheng 4767ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 4768ffd7b339SJeff Mahoney BUG_ON(ret < 0); 47693fd0a558SYan, Zheng new_root->reloc_root = reloc_root; 47703fd0a558SYan, Zheng 477149b25e05SJeff Mahoney if (rc->create_reloc_tree) 47723fd0a558SYan, Zheng ret = clone_backref_node(trans, rc, root, reloc_root); 477349b25e05SJeff Mahoney return ret; 47743fd0a558SYan, Zheng } 4775