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" 23*86736342SJosef Bacik #include "delalloc-space.h" 245d4f98a2SYan Zheng 255d4f98a2SYan Zheng /* 265d4f98a2SYan Zheng * backref_node, mapping_node and tree_block start with this 275d4f98a2SYan Zheng */ 285d4f98a2SYan Zheng struct tree_entry { 295d4f98a2SYan Zheng struct rb_node rb_node; 305d4f98a2SYan Zheng u64 bytenr; 315d4f98a2SYan Zheng }; 325d4f98a2SYan Zheng 335d4f98a2SYan Zheng /* 345d4f98a2SYan Zheng * present a tree block in the backref cache 355d4f98a2SYan Zheng */ 365d4f98a2SYan Zheng struct backref_node { 375d4f98a2SYan Zheng struct rb_node rb_node; 385d4f98a2SYan Zheng u64 bytenr; 393fd0a558SYan, Zheng 403fd0a558SYan, Zheng u64 new_bytenr; 413fd0a558SYan, Zheng /* objectid of tree block owner, can be not uptodate */ 425d4f98a2SYan Zheng u64 owner; 433fd0a558SYan, Zheng /* link to pending, changed or detached list */ 443fd0a558SYan, Zheng struct list_head list; 455d4f98a2SYan Zheng /* list of upper level blocks reference this block */ 465d4f98a2SYan Zheng struct list_head upper; 475d4f98a2SYan Zheng /* list of child blocks in the cache */ 485d4f98a2SYan Zheng struct list_head lower; 495d4f98a2SYan Zheng /* NULL if this node is not tree root */ 505d4f98a2SYan Zheng struct btrfs_root *root; 515d4f98a2SYan Zheng /* extent buffer got by COW the block */ 525d4f98a2SYan Zheng struct extent_buffer *eb; 535d4f98a2SYan Zheng /* level of tree block */ 545d4f98a2SYan Zheng unsigned int level:8; 553fd0a558SYan, Zheng /* is the block in non-reference counted tree */ 563fd0a558SYan, Zheng unsigned int cowonly:1; 573fd0a558SYan, Zheng /* 1 if no child node in the cache */ 585d4f98a2SYan Zheng unsigned int lowest:1; 595d4f98a2SYan Zheng /* is the extent buffer locked */ 605d4f98a2SYan Zheng unsigned int locked:1; 615d4f98a2SYan Zheng /* has the block been processed */ 625d4f98a2SYan Zheng unsigned int processed:1; 635d4f98a2SYan Zheng /* have backrefs of this block been checked */ 645d4f98a2SYan Zheng unsigned int checked:1; 653fd0a558SYan, Zheng /* 663fd0a558SYan, Zheng * 1 if corresponding block has been cowed but some upper 673fd0a558SYan, Zheng * level block pointers may not point to the new location 683fd0a558SYan, Zheng */ 693fd0a558SYan, Zheng unsigned int pending:1; 703fd0a558SYan, Zheng /* 713fd0a558SYan, Zheng * 1 if the backref node isn't connected to any other 723fd0a558SYan, Zheng * backref node. 733fd0a558SYan, Zheng */ 743fd0a558SYan, Zheng unsigned int detached:1; 755d4f98a2SYan Zheng }; 765d4f98a2SYan Zheng 775d4f98a2SYan Zheng /* 785d4f98a2SYan Zheng * present a block pointer in the backref cache 795d4f98a2SYan Zheng */ 805d4f98a2SYan Zheng struct backref_edge { 815d4f98a2SYan Zheng struct list_head list[2]; 825d4f98a2SYan Zheng struct backref_node *node[2]; 835d4f98a2SYan Zheng }; 845d4f98a2SYan Zheng 855d4f98a2SYan Zheng #define LOWER 0 865d4f98a2SYan Zheng #define UPPER 1 870647bf56SWang Shilong #define RELOCATION_RESERVED_NODES 256 885d4f98a2SYan Zheng 895d4f98a2SYan Zheng struct backref_cache { 905d4f98a2SYan Zheng /* red black tree of all backref nodes in the cache */ 915d4f98a2SYan Zheng struct rb_root rb_root; 923fd0a558SYan, Zheng /* for passing backref nodes to btrfs_reloc_cow_block */ 933fd0a558SYan, Zheng struct backref_node *path[BTRFS_MAX_LEVEL]; 943fd0a558SYan, Zheng /* 953fd0a558SYan, Zheng * list of blocks that have been cowed but some block 963fd0a558SYan, Zheng * pointers in upper level blocks may not reflect the 973fd0a558SYan, Zheng * new location 983fd0a558SYan, Zheng */ 995d4f98a2SYan Zheng struct list_head pending[BTRFS_MAX_LEVEL]; 1003fd0a558SYan, Zheng /* list of backref nodes with no child node */ 1013fd0a558SYan, Zheng struct list_head leaves; 1023fd0a558SYan, Zheng /* list of blocks that have been cowed in current transaction */ 1033fd0a558SYan, Zheng struct list_head changed; 1043fd0a558SYan, Zheng /* list of detached backref node. */ 1053fd0a558SYan, Zheng struct list_head detached; 1063fd0a558SYan, Zheng 1073fd0a558SYan, Zheng u64 last_trans; 1083fd0a558SYan, Zheng 1093fd0a558SYan, Zheng int nr_nodes; 1103fd0a558SYan, Zheng int nr_edges; 1115d4f98a2SYan Zheng }; 1125d4f98a2SYan Zheng 1135d4f98a2SYan Zheng /* 1145d4f98a2SYan Zheng * map address of tree root to tree 1155d4f98a2SYan Zheng */ 1165d4f98a2SYan Zheng struct mapping_node { 1175d4f98a2SYan Zheng struct rb_node rb_node; 1185d4f98a2SYan Zheng u64 bytenr; 1195d4f98a2SYan Zheng void *data; 1205d4f98a2SYan Zheng }; 1215d4f98a2SYan Zheng 1225d4f98a2SYan Zheng struct mapping_tree { 1235d4f98a2SYan Zheng struct rb_root rb_root; 1245d4f98a2SYan Zheng spinlock_t lock; 1255d4f98a2SYan Zheng }; 1265d4f98a2SYan Zheng 1275d4f98a2SYan Zheng /* 1285d4f98a2SYan Zheng * present a tree block to process 1295d4f98a2SYan Zheng */ 1305d4f98a2SYan Zheng struct tree_block { 1315d4f98a2SYan Zheng struct rb_node rb_node; 1325d4f98a2SYan Zheng u64 bytenr; 1335d4f98a2SYan Zheng struct btrfs_key key; 1345d4f98a2SYan Zheng unsigned int level:8; 1355d4f98a2SYan Zheng unsigned int key_ready:1; 1365d4f98a2SYan Zheng }; 1375d4f98a2SYan Zheng 1380257bb82SYan, Zheng #define MAX_EXTENTS 128 1390257bb82SYan, Zheng 1400257bb82SYan, Zheng struct file_extent_cluster { 1410257bb82SYan, Zheng u64 start; 1420257bb82SYan, Zheng u64 end; 1430257bb82SYan, Zheng u64 boundary[MAX_EXTENTS]; 1440257bb82SYan, Zheng unsigned int nr; 1450257bb82SYan, Zheng }; 1460257bb82SYan, Zheng 1475d4f98a2SYan Zheng struct reloc_control { 1485d4f98a2SYan Zheng /* block group to relocate */ 1495d4f98a2SYan Zheng struct btrfs_block_group_cache *block_group; 1505d4f98a2SYan Zheng /* extent tree */ 1515d4f98a2SYan Zheng struct btrfs_root *extent_root; 1525d4f98a2SYan Zheng /* inode for moving data */ 1535d4f98a2SYan Zheng struct inode *data_inode; 1543fd0a558SYan, Zheng 1553fd0a558SYan, Zheng struct btrfs_block_rsv *block_rsv; 1563fd0a558SYan, Zheng 1573fd0a558SYan, Zheng struct backref_cache backref_cache; 1583fd0a558SYan, Zheng 1593fd0a558SYan, Zheng struct file_extent_cluster cluster; 1605d4f98a2SYan Zheng /* tree blocks have been processed */ 1615d4f98a2SYan Zheng struct extent_io_tree processed_blocks; 1625d4f98a2SYan Zheng /* map start of tree root to corresponding reloc tree */ 1635d4f98a2SYan Zheng struct mapping_tree reloc_root_tree; 1645d4f98a2SYan Zheng /* list of reloc trees */ 1655d4f98a2SYan Zheng struct list_head reloc_roots; 166d2311e69SQu Wenruo /* list of subvolume trees that get relocated */ 167d2311e69SQu Wenruo struct list_head dirty_subvol_roots; 1683fd0a558SYan, Zheng /* size of metadata reservation for merging reloc trees */ 1693fd0a558SYan, Zheng u64 merging_rsv_size; 1703fd0a558SYan, Zheng /* size of relocated tree nodes */ 1713fd0a558SYan, Zheng u64 nodes_relocated; 1720647bf56SWang Shilong /* reserved size for block group relocation*/ 1730647bf56SWang Shilong u64 reserved_bytes; 1743fd0a558SYan, Zheng 1755d4f98a2SYan Zheng u64 search_start; 1765d4f98a2SYan Zheng u64 extents_found; 1773fd0a558SYan, Zheng 1783fd0a558SYan, Zheng unsigned int stage:8; 1793fd0a558SYan, Zheng unsigned int create_reloc_tree:1; 1803fd0a558SYan, Zheng unsigned int merge_reloc_tree:1; 1815d4f98a2SYan Zheng unsigned int found_file_extent:1; 1825d4f98a2SYan Zheng }; 1835d4f98a2SYan Zheng 1845d4f98a2SYan Zheng /* stages of data relocation */ 1855d4f98a2SYan Zheng #define MOVE_DATA_EXTENTS 0 1865d4f98a2SYan Zheng #define UPDATE_DATA_PTRS 1 1875d4f98a2SYan Zheng 1883fd0a558SYan, Zheng static void remove_backref_node(struct backref_cache *cache, 1893fd0a558SYan, Zheng struct backref_node *node); 1903fd0a558SYan, Zheng static void __mark_block_processed(struct reloc_control *rc, 1913fd0a558SYan, Zheng struct backref_node *node); 1925d4f98a2SYan Zheng 1935d4f98a2SYan Zheng static void mapping_tree_init(struct mapping_tree *tree) 1945d4f98a2SYan Zheng { 1956bef4d31SEric Paris tree->rb_root = RB_ROOT; 1965d4f98a2SYan Zheng spin_lock_init(&tree->lock); 1975d4f98a2SYan Zheng } 1985d4f98a2SYan Zheng 1995d4f98a2SYan Zheng static void backref_cache_init(struct backref_cache *cache) 2005d4f98a2SYan Zheng { 2015d4f98a2SYan Zheng int i; 2026bef4d31SEric Paris cache->rb_root = RB_ROOT; 2035d4f98a2SYan Zheng for (i = 0; i < BTRFS_MAX_LEVEL; i++) 2045d4f98a2SYan Zheng INIT_LIST_HEAD(&cache->pending[i]); 2053fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->changed); 2063fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->detached); 2073fd0a558SYan, Zheng INIT_LIST_HEAD(&cache->leaves); 2085d4f98a2SYan Zheng } 2095d4f98a2SYan Zheng 2103fd0a558SYan, Zheng static void backref_cache_cleanup(struct backref_cache *cache) 2115d4f98a2SYan Zheng { 2123fd0a558SYan, Zheng struct backref_node *node; 2133fd0a558SYan, Zheng int i; 2143fd0a558SYan, Zheng 2153fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 2163fd0a558SYan, Zheng node = list_entry(cache->detached.next, 2173fd0a558SYan, Zheng struct backref_node, list); 2183fd0a558SYan, Zheng remove_backref_node(cache, node); 2193fd0a558SYan, Zheng } 2203fd0a558SYan, Zheng 2213fd0a558SYan, Zheng while (!list_empty(&cache->leaves)) { 2223fd0a558SYan, Zheng node = list_entry(cache->leaves.next, 2233fd0a558SYan, Zheng struct backref_node, lower); 2243fd0a558SYan, Zheng remove_backref_node(cache, node); 2253fd0a558SYan, Zheng } 2263fd0a558SYan, Zheng 2273fd0a558SYan, Zheng cache->last_trans = 0; 2283fd0a558SYan, Zheng 2293fd0a558SYan, Zheng for (i = 0; i < BTRFS_MAX_LEVEL; i++) 230f4907095SLiu Bo ASSERT(list_empty(&cache->pending[i])); 231f4907095SLiu Bo ASSERT(list_empty(&cache->changed)); 232f4907095SLiu Bo ASSERT(list_empty(&cache->detached)); 233f4907095SLiu Bo ASSERT(RB_EMPTY_ROOT(&cache->rb_root)); 234f4907095SLiu Bo ASSERT(!cache->nr_nodes); 235f4907095SLiu Bo ASSERT(!cache->nr_edges); 2363fd0a558SYan, Zheng } 2373fd0a558SYan, Zheng 2383fd0a558SYan, Zheng static struct backref_node *alloc_backref_node(struct backref_cache *cache) 2393fd0a558SYan, Zheng { 2403fd0a558SYan, Zheng struct backref_node *node; 2413fd0a558SYan, Zheng 2423fd0a558SYan, Zheng node = kzalloc(sizeof(*node), GFP_NOFS); 2433fd0a558SYan, Zheng if (node) { 2443fd0a558SYan, Zheng INIT_LIST_HEAD(&node->list); 2455d4f98a2SYan Zheng INIT_LIST_HEAD(&node->upper); 2465d4f98a2SYan Zheng INIT_LIST_HEAD(&node->lower); 2475d4f98a2SYan Zheng RB_CLEAR_NODE(&node->rb_node); 2483fd0a558SYan, Zheng cache->nr_nodes++; 2493fd0a558SYan, Zheng } 2503fd0a558SYan, Zheng return node; 2513fd0a558SYan, Zheng } 2523fd0a558SYan, Zheng 2533fd0a558SYan, Zheng static void free_backref_node(struct backref_cache *cache, 2543fd0a558SYan, Zheng struct backref_node *node) 2553fd0a558SYan, Zheng { 2563fd0a558SYan, Zheng if (node) { 2573fd0a558SYan, Zheng cache->nr_nodes--; 2583fd0a558SYan, Zheng kfree(node); 2593fd0a558SYan, Zheng } 2603fd0a558SYan, Zheng } 2613fd0a558SYan, Zheng 2623fd0a558SYan, Zheng static struct backref_edge *alloc_backref_edge(struct backref_cache *cache) 2633fd0a558SYan, Zheng { 2643fd0a558SYan, Zheng struct backref_edge *edge; 2653fd0a558SYan, Zheng 2663fd0a558SYan, Zheng edge = kzalloc(sizeof(*edge), GFP_NOFS); 2673fd0a558SYan, Zheng if (edge) 2683fd0a558SYan, Zheng cache->nr_edges++; 2693fd0a558SYan, Zheng return edge; 2703fd0a558SYan, Zheng } 2713fd0a558SYan, Zheng 2723fd0a558SYan, Zheng static void free_backref_edge(struct backref_cache *cache, 2733fd0a558SYan, Zheng struct backref_edge *edge) 2743fd0a558SYan, Zheng { 2753fd0a558SYan, Zheng if (edge) { 2763fd0a558SYan, Zheng cache->nr_edges--; 2773fd0a558SYan, Zheng kfree(edge); 2783fd0a558SYan, Zheng } 2795d4f98a2SYan Zheng } 2805d4f98a2SYan Zheng 2815d4f98a2SYan Zheng static struct rb_node *tree_insert(struct rb_root *root, u64 bytenr, 2825d4f98a2SYan Zheng struct rb_node *node) 2835d4f98a2SYan Zheng { 2845d4f98a2SYan Zheng struct rb_node **p = &root->rb_node; 2855d4f98a2SYan Zheng struct rb_node *parent = NULL; 2865d4f98a2SYan Zheng struct tree_entry *entry; 2875d4f98a2SYan Zheng 2885d4f98a2SYan Zheng while (*p) { 2895d4f98a2SYan Zheng parent = *p; 2905d4f98a2SYan Zheng entry = rb_entry(parent, struct tree_entry, rb_node); 2915d4f98a2SYan Zheng 2925d4f98a2SYan Zheng if (bytenr < entry->bytenr) 2935d4f98a2SYan Zheng p = &(*p)->rb_left; 2945d4f98a2SYan Zheng else if (bytenr > entry->bytenr) 2955d4f98a2SYan Zheng p = &(*p)->rb_right; 2965d4f98a2SYan Zheng else 2975d4f98a2SYan Zheng return parent; 2985d4f98a2SYan Zheng } 2995d4f98a2SYan Zheng 3005d4f98a2SYan Zheng rb_link_node(node, parent, p); 3015d4f98a2SYan Zheng rb_insert_color(node, root); 3025d4f98a2SYan Zheng return NULL; 3035d4f98a2SYan Zheng } 3045d4f98a2SYan Zheng 3055d4f98a2SYan Zheng static struct rb_node *tree_search(struct rb_root *root, u64 bytenr) 3065d4f98a2SYan Zheng { 3075d4f98a2SYan Zheng struct rb_node *n = root->rb_node; 3085d4f98a2SYan Zheng struct tree_entry *entry; 3095d4f98a2SYan Zheng 3105d4f98a2SYan Zheng while (n) { 3115d4f98a2SYan Zheng entry = rb_entry(n, struct tree_entry, rb_node); 3125d4f98a2SYan Zheng 3135d4f98a2SYan Zheng if (bytenr < entry->bytenr) 3145d4f98a2SYan Zheng n = n->rb_left; 3155d4f98a2SYan Zheng else if (bytenr > entry->bytenr) 3165d4f98a2SYan Zheng n = n->rb_right; 3175d4f98a2SYan Zheng else 3185d4f98a2SYan Zheng return n; 3195d4f98a2SYan Zheng } 3205d4f98a2SYan Zheng return NULL; 3215d4f98a2SYan Zheng } 3225d4f98a2SYan Zheng 32348a3b636SEric Sandeen static void backref_tree_panic(struct rb_node *rb_node, int errno, u64 bytenr) 32443c04fb1SJeff Mahoney { 32543c04fb1SJeff Mahoney 32643c04fb1SJeff Mahoney struct btrfs_fs_info *fs_info = NULL; 32743c04fb1SJeff Mahoney struct backref_node *bnode = rb_entry(rb_node, struct backref_node, 32843c04fb1SJeff Mahoney rb_node); 32943c04fb1SJeff Mahoney if (bnode->root) 33043c04fb1SJeff Mahoney fs_info = bnode->root->fs_info; 3315d163e0eSJeff Mahoney btrfs_panic(fs_info, errno, 3325d163e0eSJeff Mahoney "Inconsistency in backref cache found at offset %llu", 3335d163e0eSJeff Mahoney bytenr); 33443c04fb1SJeff Mahoney } 33543c04fb1SJeff Mahoney 3365d4f98a2SYan Zheng /* 3375d4f98a2SYan Zheng * walk up backref nodes until reach node presents tree root 3385d4f98a2SYan Zheng */ 3395d4f98a2SYan Zheng static struct backref_node *walk_up_backref(struct backref_node *node, 3405d4f98a2SYan Zheng struct backref_edge *edges[], 3415d4f98a2SYan Zheng int *index) 3425d4f98a2SYan Zheng { 3435d4f98a2SYan Zheng struct backref_edge *edge; 3445d4f98a2SYan Zheng int idx = *index; 3455d4f98a2SYan Zheng 3465d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 3475d4f98a2SYan Zheng edge = list_entry(node->upper.next, 3485d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 3495d4f98a2SYan Zheng edges[idx++] = edge; 3505d4f98a2SYan Zheng node = edge->node[UPPER]; 3515d4f98a2SYan Zheng } 3523fd0a558SYan, Zheng BUG_ON(node->detached); 3535d4f98a2SYan Zheng *index = idx; 3545d4f98a2SYan Zheng return node; 3555d4f98a2SYan Zheng } 3565d4f98a2SYan Zheng 3575d4f98a2SYan Zheng /* 3585d4f98a2SYan Zheng * walk down backref nodes to find start of next reference path 3595d4f98a2SYan Zheng */ 3605d4f98a2SYan Zheng static struct backref_node *walk_down_backref(struct backref_edge *edges[], 3615d4f98a2SYan Zheng int *index) 3625d4f98a2SYan Zheng { 3635d4f98a2SYan Zheng struct backref_edge *edge; 3645d4f98a2SYan Zheng struct backref_node *lower; 3655d4f98a2SYan Zheng int idx = *index; 3665d4f98a2SYan Zheng 3675d4f98a2SYan Zheng while (idx > 0) { 3685d4f98a2SYan Zheng edge = edges[idx - 1]; 3695d4f98a2SYan Zheng lower = edge->node[LOWER]; 3705d4f98a2SYan Zheng if (list_is_last(&edge->list[LOWER], &lower->upper)) { 3715d4f98a2SYan Zheng idx--; 3725d4f98a2SYan Zheng continue; 3735d4f98a2SYan Zheng } 3745d4f98a2SYan Zheng edge = list_entry(edge->list[LOWER].next, 3755d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 3765d4f98a2SYan Zheng edges[idx - 1] = edge; 3775d4f98a2SYan Zheng *index = idx; 3785d4f98a2SYan Zheng return edge->node[UPPER]; 3795d4f98a2SYan Zheng } 3805d4f98a2SYan Zheng *index = 0; 3815d4f98a2SYan Zheng return NULL; 3825d4f98a2SYan Zheng } 3835d4f98a2SYan Zheng 3843fd0a558SYan, Zheng static void unlock_node_buffer(struct backref_node *node) 3855d4f98a2SYan Zheng { 3865d4f98a2SYan Zheng if (node->locked) { 3875d4f98a2SYan Zheng btrfs_tree_unlock(node->eb); 3885d4f98a2SYan Zheng node->locked = 0; 3895d4f98a2SYan Zheng } 3903fd0a558SYan, Zheng } 3913fd0a558SYan, Zheng 3923fd0a558SYan, Zheng static void drop_node_buffer(struct backref_node *node) 3933fd0a558SYan, Zheng { 3943fd0a558SYan, Zheng if (node->eb) { 3953fd0a558SYan, Zheng unlock_node_buffer(node); 3965d4f98a2SYan Zheng free_extent_buffer(node->eb); 3975d4f98a2SYan Zheng node->eb = NULL; 3985d4f98a2SYan Zheng } 3995d4f98a2SYan Zheng } 4005d4f98a2SYan Zheng 4015d4f98a2SYan Zheng static void drop_backref_node(struct backref_cache *tree, 4025d4f98a2SYan Zheng struct backref_node *node) 4035d4f98a2SYan Zheng { 4045d4f98a2SYan Zheng BUG_ON(!list_empty(&node->upper)); 4055d4f98a2SYan Zheng 4065d4f98a2SYan Zheng drop_node_buffer(node); 4073fd0a558SYan, Zheng list_del(&node->list); 4085d4f98a2SYan Zheng list_del(&node->lower); 4093fd0a558SYan, Zheng if (!RB_EMPTY_NODE(&node->rb_node)) 4105d4f98a2SYan Zheng rb_erase(&node->rb_node, &tree->rb_root); 4113fd0a558SYan, Zheng free_backref_node(tree, node); 4125d4f98a2SYan Zheng } 4135d4f98a2SYan Zheng 4145d4f98a2SYan Zheng /* 4155d4f98a2SYan Zheng * remove a backref node from the backref cache 4165d4f98a2SYan Zheng */ 4175d4f98a2SYan Zheng static void remove_backref_node(struct backref_cache *cache, 4185d4f98a2SYan Zheng struct backref_node *node) 4195d4f98a2SYan Zheng { 4205d4f98a2SYan Zheng struct backref_node *upper; 4215d4f98a2SYan Zheng struct backref_edge *edge; 4225d4f98a2SYan Zheng 4235d4f98a2SYan Zheng if (!node) 4245d4f98a2SYan Zheng return; 4255d4f98a2SYan Zheng 4263fd0a558SYan, Zheng BUG_ON(!node->lowest && !node->detached); 4275d4f98a2SYan Zheng while (!list_empty(&node->upper)) { 4285d4f98a2SYan Zheng edge = list_entry(node->upper.next, struct backref_edge, 4295d4f98a2SYan Zheng list[LOWER]); 4305d4f98a2SYan Zheng upper = edge->node[UPPER]; 4315d4f98a2SYan Zheng list_del(&edge->list[LOWER]); 4325d4f98a2SYan Zheng list_del(&edge->list[UPPER]); 4333fd0a558SYan, Zheng free_backref_edge(cache, edge); 4343fd0a558SYan, Zheng 4353fd0a558SYan, Zheng if (RB_EMPTY_NODE(&upper->rb_node)) { 4363fd0a558SYan, Zheng BUG_ON(!list_empty(&node->upper)); 4373fd0a558SYan, Zheng drop_backref_node(cache, node); 4383fd0a558SYan, Zheng node = upper; 4393fd0a558SYan, Zheng node->lowest = 1; 4403fd0a558SYan, Zheng continue; 4413fd0a558SYan, Zheng } 4425d4f98a2SYan Zheng /* 4433fd0a558SYan, Zheng * add the node to leaf node list if no other 4445d4f98a2SYan Zheng * child block cached. 4455d4f98a2SYan Zheng */ 4465d4f98a2SYan Zheng if (list_empty(&upper->lower)) { 4473fd0a558SYan, Zheng list_add_tail(&upper->lower, &cache->leaves); 4485d4f98a2SYan Zheng upper->lowest = 1; 4495d4f98a2SYan Zheng } 4505d4f98a2SYan Zheng } 4513fd0a558SYan, Zheng 4525d4f98a2SYan Zheng drop_backref_node(cache, node); 4535d4f98a2SYan Zheng } 4545d4f98a2SYan Zheng 4553fd0a558SYan, Zheng static void update_backref_node(struct backref_cache *cache, 4563fd0a558SYan, Zheng struct backref_node *node, u64 bytenr) 4573fd0a558SYan, Zheng { 4583fd0a558SYan, Zheng struct rb_node *rb_node; 4593fd0a558SYan, Zheng rb_erase(&node->rb_node, &cache->rb_root); 4603fd0a558SYan, Zheng node->bytenr = bytenr; 4613fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, &node->rb_node); 46243c04fb1SJeff Mahoney if (rb_node) 46343c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, bytenr); 4643fd0a558SYan, Zheng } 4653fd0a558SYan, Zheng 4663fd0a558SYan, Zheng /* 4673fd0a558SYan, Zheng * update backref cache after a transaction commit 4683fd0a558SYan, Zheng */ 4693fd0a558SYan, Zheng static int update_backref_cache(struct btrfs_trans_handle *trans, 4703fd0a558SYan, Zheng struct backref_cache *cache) 4713fd0a558SYan, Zheng { 4723fd0a558SYan, Zheng struct backref_node *node; 4733fd0a558SYan, Zheng int level = 0; 4743fd0a558SYan, Zheng 4753fd0a558SYan, Zheng if (cache->last_trans == 0) { 4763fd0a558SYan, Zheng cache->last_trans = trans->transid; 4773fd0a558SYan, Zheng return 0; 4783fd0a558SYan, Zheng } 4793fd0a558SYan, Zheng 4803fd0a558SYan, Zheng if (cache->last_trans == trans->transid) 4813fd0a558SYan, Zheng return 0; 4823fd0a558SYan, Zheng 4833fd0a558SYan, Zheng /* 4843fd0a558SYan, Zheng * detached nodes are used to avoid unnecessary backref 4853fd0a558SYan, Zheng * lookup. transaction commit changes the extent tree. 4863fd0a558SYan, Zheng * so the detached nodes are no longer useful. 4873fd0a558SYan, Zheng */ 4883fd0a558SYan, Zheng while (!list_empty(&cache->detached)) { 4893fd0a558SYan, Zheng node = list_entry(cache->detached.next, 4903fd0a558SYan, Zheng struct backref_node, list); 4913fd0a558SYan, Zheng remove_backref_node(cache, node); 4923fd0a558SYan, Zheng } 4933fd0a558SYan, Zheng 4943fd0a558SYan, Zheng while (!list_empty(&cache->changed)) { 4953fd0a558SYan, Zheng node = list_entry(cache->changed.next, 4963fd0a558SYan, Zheng struct backref_node, list); 4973fd0a558SYan, Zheng list_del_init(&node->list); 4983fd0a558SYan, Zheng BUG_ON(node->pending); 4993fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 5003fd0a558SYan, Zheng } 5013fd0a558SYan, Zheng 5023fd0a558SYan, Zheng /* 5033fd0a558SYan, Zheng * some nodes can be left in the pending list if there were 5043fd0a558SYan, Zheng * errors during processing the pending nodes. 5053fd0a558SYan, Zheng */ 5063fd0a558SYan, Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 5073fd0a558SYan, Zheng list_for_each_entry(node, &cache->pending[level], list) { 5083fd0a558SYan, Zheng BUG_ON(!node->pending); 5093fd0a558SYan, Zheng if (node->bytenr == node->new_bytenr) 5103fd0a558SYan, Zheng continue; 5113fd0a558SYan, Zheng update_backref_node(cache, node, node->new_bytenr); 5123fd0a558SYan, Zheng } 5133fd0a558SYan, Zheng } 5143fd0a558SYan, Zheng 5153fd0a558SYan, Zheng cache->last_trans = 0; 5163fd0a558SYan, Zheng return 1; 5173fd0a558SYan, Zheng } 5183fd0a558SYan, Zheng 519f2a97a9dSDavid Sterba 5203fd0a558SYan, Zheng static int should_ignore_root(struct btrfs_root *root) 5213fd0a558SYan, Zheng { 5223fd0a558SYan, Zheng struct btrfs_root *reloc_root; 5233fd0a558SYan, Zheng 52427cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 5253fd0a558SYan, Zheng return 0; 5263fd0a558SYan, Zheng 5273fd0a558SYan, Zheng reloc_root = root->reloc_root; 5283fd0a558SYan, Zheng if (!reloc_root) 5293fd0a558SYan, Zheng return 0; 5303fd0a558SYan, Zheng 5313fd0a558SYan, Zheng if (btrfs_root_last_snapshot(&reloc_root->root_item) == 5323fd0a558SYan, Zheng root->fs_info->running_transaction->transid - 1) 5333fd0a558SYan, Zheng return 0; 5343fd0a558SYan, Zheng /* 5353fd0a558SYan, Zheng * if there is reloc tree and it was created in previous 5363fd0a558SYan, Zheng * transaction backref lookup can find the reloc tree, 5373fd0a558SYan, Zheng * so backref node for the fs tree root is useless for 5383fd0a558SYan, Zheng * relocation. 5393fd0a558SYan, Zheng */ 5403fd0a558SYan, Zheng return 1; 5413fd0a558SYan, Zheng } 5425d4f98a2SYan Zheng /* 5435d4f98a2SYan Zheng * find reloc tree by address of tree root 5445d4f98a2SYan Zheng */ 5455d4f98a2SYan Zheng static struct btrfs_root *find_reloc_root(struct reloc_control *rc, 5465d4f98a2SYan Zheng u64 bytenr) 5475d4f98a2SYan Zheng { 5485d4f98a2SYan Zheng struct rb_node *rb_node; 5495d4f98a2SYan Zheng struct mapping_node *node; 5505d4f98a2SYan Zheng struct btrfs_root *root = NULL; 5515d4f98a2SYan Zheng 5525d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 5535d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, bytenr); 5545d4f98a2SYan Zheng if (rb_node) { 5555d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 5565d4f98a2SYan Zheng root = (struct btrfs_root *)node->data; 5575d4f98a2SYan Zheng } 5585d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 5595d4f98a2SYan Zheng return root; 5605d4f98a2SYan Zheng } 5615d4f98a2SYan Zheng 5625d4f98a2SYan Zheng static int is_cowonly_root(u64 root_objectid) 5635d4f98a2SYan Zheng { 5645d4f98a2SYan Zheng if (root_objectid == BTRFS_ROOT_TREE_OBJECTID || 5655d4f98a2SYan Zheng root_objectid == BTRFS_EXTENT_TREE_OBJECTID || 5665d4f98a2SYan Zheng root_objectid == BTRFS_CHUNK_TREE_OBJECTID || 5675d4f98a2SYan Zheng root_objectid == BTRFS_DEV_TREE_OBJECTID || 5685d4f98a2SYan Zheng root_objectid == BTRFS_TREE_LOG_OBJECTID || 56966463748SWang Shilong root_objectid == BTRFS_CSUM_TREE_OBJECTID || 57066463748SWang Shilong root_objectid == BTRFS_UUID_TREE_OBJECTID || 5713e4c5efbSDavid Sterba root_objectid == BTRFS_QUOTA_TREE_OBJECTID || 5723e4c5efbSDavid Sterba root_objectid == BTRFS_FREE_SPACE_TREE_OBJECTID) 5735d4f98a2SYan Zheng return 1; 5745d4f98a2SYan Zheng return 0; 5755d4f98a2SYan Zheng } 5765d4f98a2SYan Zheng 5775d4f98a2SYan Zheng static struct btrfs_root *read_fs_root(struct btrfs_fs_info *fs_info, 5785d4f98a2SYan Zheng u64 root_objectid) 5795d4f98a2SYan Zheng { 5805d4f98a2SYan Zheng struct btrfs_key key; 5815d4f98a2SYan Zheng 5825d4f98a2SYan Zheng key.objectid = root_objectid; 5835d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 5845d4f98a2SYan Zheng if (is_cowonly_root(root_objectid)) 5855d4f98a2SYan Zheng key.offset = 0; 5865d4f98a2SYan Zheng else 5875d4f98a2SYan Zheng key.offset = (u64)-1; 5885d4f98a2SYan Zheng 589c00869f1SMiao Xie return btrfs_get_fs_root(fs_info, &key, false); 5905d4f98a2SYan Zheng } 5915d4f98a2SYan Zheng 5925d4f98a2SYan Zheng static noinline_for_stack 5935d4f98a2SYan Zheng int find_inline_backref(struct extent_buffer *leaf, int slot, 5945d4f98a2SYan Zheng unsigned long *ptr, unsigned long *end) 5955d4f98a2SYan Zheng { 5963173a18fSJosef Bacik struct btrfs_key key; 5975d4f98a2SYan Zheng struct btrfs_extent_item *ei; 5985d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 5995d4f98a2SYan Zheng u32 item_size; 6005d4f98a2SYan Zheng 6013173a18fSJosef Bacik btrfs_item_key_to_cpu(leaf, &key, slot); 6023173a18fSJosef Bacik 6035d4f98a2SYan Zheng item_size = btrfs_item_size_nr(leaf, slot); 604ba3c2b19SNikolay Borisov if (item_size < sizeof(*ei)) { 605ba3c2b19SNikolay Borisov btrfs_print_v0_err(leaf->fs_info); 606ba3c2b19SNikolay Borisov btrfs_handle_fs_error(leaf->fs_info, -EINVAL, NULL); 607ba3c2b19SNikolay Borisov return 1; 608ba3c2b19SNikolay Borisov } 6095d4f98a2SYan Zheng ei = btrfs_item_ptr(leaf, slot, struct btrfs_extent_item); 6105d4f98a2SYan Zheng WARN_ON(!(btrfs_extent_flags(leaf, ei) & 6115d4f98a2SYan Zheng BTRFS_EXTENT_FLAG_TREE_BLOCK)); 6125d4f98a2SYan Zheng 6133173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 6143173a18fSJosef Bacik item_size <= sizeof(*ei) + sizeof(*bi)) { 6155d4f98a2SYan Zheng WARN_ON(item_size < sizeof(*ei) + sizeof(*bi)); 6165d4f98a2SYan Zheng return 1; 6175d4f98a2SYan Zheng } 618d062d13cSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 619d062d13cSJosef Bacik item_size <= sizeof(*ei)) { 620d062d13cSJosef Bacik WARN_ON(item_size < sizeof(*ei)); 621d062d13cSJosef Bacik return 1; 622d062d13cSJosef Bacik } 6235d4f98a2SYan Zheng 6243173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) { 6255d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 6265d4f98a2SYan Zheng *ptr = (unsigned long)(bi + 1); 6273173a18fSJosef Bacik } else { 6283173a18fSJosef Bacik *ptr = (unsigned long)(ei + 1); 6293173a18fSJosef Bacik } 6305d4f98a2SYan Zheng *end = (unsigned long)ei + item_size; 6315d4f98a2SYan Zheng return 0; 6325d4f98a2SYan Zheng } 6335d4f98a2SYan Zheng 6345d4f98a2SYan Zheng /* 6355d4f98a2SYan Zheng * build backref tree for a given tree block. root of the backref tree 6365d4f98a2SYan Zheng * corresponds the tree block, leaves of the backref tree correspond 6375d4f98a2SYan Zheng * roots of b-trees that reference the tree block. 6385d4f98a2SYan Zheng * 6395d4f98a2SYan Zheng * the basic idea of this function is check backrefs of a given block 64001327610SNicholas D Steeves * to find upper level blocks that reference the block, and then check 64101327610SNicholas D Steeves * backrefs of these upper level blocks recursively. the recursion stop 6425d4f98a2SYan Zheng * when tree root is reached or backrefs for the block is cached. 6435d4f98a2SYan Zheng * 6445d4f98a2SYan Zheng * NOTE: if we find backrefs for a block are cached, we know backrefs 6455d4f98a2SYan Zheng * for all upper level blocks that directly/indirectly reference the 6465d4f98a2SYan Zheng * block are also cached. 6475d4f98a2SYan Zheng */ 6483fd0a558SYan, Zheng static noinline_for_stack 6493fd0a558SYan, Zheng struct backref_node *build_backref_tree(struct reloc_control *rc, 6505d4f98a2SYan Zheng struct btrfs_key *node_key, 6515d4f98a2SYan Zheng int level, u64 bytenr) 6525d4f98a2SYan Zheng { 6533fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 654fa6ac715SQu Wenruo struct btrfs_path *path1; /* For searching extent root */ 655fa6ac715SQu Wenruo struct btrfs_path *path2; /* For searching parent of TREE_BLOCK_REF */ 6565d4f98a2SYan Zheng struct extent_buffer *eb; 6575d4f98a2SYan Zheng struct btrfs_root *root; 6585d4f98a2SYan Zheng struct backref_node *cur; 6595d4f98a2SYan Zheng struct backref_node *upper; 6605d4f98a2SYan Zheng struct backref_node *lower; 6615d4f98a2SYan Zheng struct backref_node *node = NULL; 6625d4f98a2SYan Zheng struct backref_node *exist = NULL; 6635d4f98a2SYan Zheng struct backref_edge *edge; 6645d4f98a2SYan Zheng struct rb_node *rb_node; 6655d4f98a2SYan Zheng struct btrfs_key key; 6665d4f98a2SYan Zheng unsigned long end; 6675d4f98a2SYan Zheng unsigned long ptr; 668fa6ac715SQu Wenruo LIST_HEAD(list); /* Pending edge list, upper node needs to be checked */ 6693fd0a558SYan, Zheng LIST_HEAD(useless); 6703fd0a558SYan, Zheng int cowonly; 6715d4f98a2SYan Zheng int ret; 6725d4f98a2SYan Zheng int err = 0; 673b6c60c80SJosef Bacik bool need_check = true; 6745d4f98a2SYan Zheng 6755d4f98a2SYan Zheng path1 = btrfs_alloc_path(); 6765d4f98a2SYan Zheng path2 = btrfs_alloc_path(); 6775d4f98a2SYan Zheng if (!path1 || !path2) { 6785d4f98a2SYan Zheng err = -ENOMEM; 6795d4f98a2SYan Zheng goto out; 6805d4f98a2SYan Zheng } 681e4058b54SDavid Sterba path1->reada = READA_FORWARD; 682e4058b54SDavid Sterba path2->reada = READA_FORWARD; 6835d4f98a2SYan Zheng 6843fd0a558SYan, Zheng node = alloc_backref_node(cache); 6855d4f98a2SYan Zheng if (!node) { 6865d4f98a2SYan Zheng err = -ENOMEM; 6875d4f98a2SYan Zheng goto out; 6885d4f98a2SYan Zheng } 6895d4f98a2SYan Zheng 6905d4f98a2SYan Zheng node->bytenr = bytenr; 6915d4f98a2SYan Zheng node->level = level; 6925d4f98a2SYan Zheng node->lowest = 1; 6935d4f98a2SYan Zheng cur = node; 6945d4f98a2SYan Zheng again: 6955d4f98a2SYan Zheng end = 0; 6965d4f98a2SYan Zheng ptr = 0; 6975d4f98a2SYan Zheng key.objectid = cur->bytenr; 6983173a18fSJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 6995d4f98a2SYan Zheng key.offset = (u64)-1; 7005d4f98a2SYan Zheng 7015d4f98a2SYan Zheng path1->search_commit_root = 1; 7025d4f98a2SYan Zheng path1->skip_locking = 1; 7035d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path1, 7045d4f98a2SYan Zheng 0, 0); 7055d4f98a2SYan Zheng if (ret < 0) { 7065d4f98a2SYan Zheng err = ret; 7075d4f98a2SYan Zheng goto out; 7085d4f98a2SYan Zheng } 70975bfb9afSJosef Bacik ASSERT(ret); 71075bfb9afSJosef Bacik ASSERT(path1->slots[0]); 7115d4f98a2SYan Zheng 7125d4f98a2SYan Zheng path1->slots[0]--; 7135d4f98a2SYan Zheng 7145d4f98a2SYan Zheng WARN_ON(cur->checked); 7155d4f98a2SYan Zheng if (!list_empty(&cur->upper)) { 7165d4f98a2SYan Zheng /* 71770f23fd6SJustin P. Mattock * the backref was added previously when processing 7185d4f98a2SYan Zheng * backref of type BTRFS_TREE_BLOCK_REF_KEY 7195d4f98a2SYan Zheng */ 72075bfb9afSJosef Bacik ASSERT(list_is_singular(&cur->upper)); 7215d4f98a2SYan Zheng edge = list_entry(cur->upper.next, struct backref_edge, 7225d4f98a2SYan Zheng list[LOWER]); 72375bfb9afSJosef Bacik ASSERT(list_empty(&edge->list[UPPER])); 7245d4f98a2SYan Zheng exist = edge->node[UPPER]; 7255d4f98a2SYan Zheng /* 7265d4f98a2SYan Zheng * add the upper level block to pending list if we need 7275d4f98a2SYan Zheng * check its backrefs 7285d4f98a2SYan Zheng */ 7295d4f98a2SYan Zheng if (!exist->checked) 7305d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 7315d4f98a2SYan Zheng } else { 7325d4f98a2SYan Zheng exist = NULL; 7335d4f98a2SYan Zheng } 7345d4f98a2SYan Zheng 7355d4f98a2SYan Zheng while (1) { 7365d4f98a2SYan Zheng cond_resched(); 7375d4f98a2SYan Zheng eb = path1->nodes[0]; 7385d4f98a2SYan Zheng 7395d4f98a2SYan Zheng if (ptr >= end) { 7405d4f98a2SYan Zheng if (path1->slots[0] >= btrfs_header_nritems(eb)) { 7415d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path1); 7425d4f98a2SYan Zheng if (ret < 0) { 7435d4f98a2SYan Zheng err = ret; 7445d4f98a2SYan Zheng goto out; 7455d4f98a2SYan Zheng } 7465d4f98a2SYan Zheng if (ret > 0) 7475d4f98a2SYan Zheng break; 7485d4f98a2SYan Zheng eb = path1->nodes[0]; 7495d4f98a2SYan Zheng } 7505d4f98a2SYan Zheng 7515d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &key, path1->slots[0]); 7525d4f98a2SYan Zheng if (key.objectid != cur->bytenr) { 7535d4f98a2SYan Zheng WARN_ON(exist); 7545d4f98a2SYan Zheng break; 7555d4f98a2SYan Zheng } 7565d4f98a2SYan Zheng 7573173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY || 7583173a18fSJosef Bacik key.type == BTRFS_METADATA_ITEM_KEY) { 7595d4f98a2SYan Zheng ret = find_inline_backref(eb, path1->slots[0], 7605d4f98a2SYan Zheng &ptr, &end); 7615d4f98a2SYan Zheng if (ret) 7625d4f98a2SYan Zheng goto next; 7635d4f98a2SYan Zheng } 7645d4f98a2SYan Zheng } 7655d4f98a2SYan Zheng 7665d4f98a2SYan Zheng if (ptr < end) { 7675d4f98a2SYan Zheng /* update key for inline back ref */ 7685d4f98a2SYan Zheng struct btrfs_extent_inline_ref *iref; 7693de28d57SLiu Bo int type; 7705d4f98a2SYan Zheng iref = (struct btrfs_extent_inline_ref *)ptr; 7713de28d57SLiu Bo type = btrfs_get_extent_inline_ref_type(eb, iref, 7723de28d57SLiu Bo BTRFS_REF_TYPE_BLOCK); 7733de28d57SLiu Bo if (type == BTRFS_REF_TYPE_INVALID) { 774af431dcbSSu Yue err = -EUCLEAN; 7753de28d57SLiu Bo goto out; 7763de28d57SLiu Bo } 7773de28d57SLiu Bo key.type = type; 7785d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 7793de28d57SLiu Bo 7805d4f98a2SYan Zheng WARN_ON(key.type != BTRFS_TREE_BLOCK_REF_KEY && 7815d4f98a2SYan Zheng key.type != BTRFS_SHARED_BLOCK_REF_KEY); 7825d4f98a2SYan Zheng } 7835d4f98a2SYan Zheng 784fa6ac715SQu Wenruo /* 785fa6ac715SQu Wenruo * Parent node found and matches current inline ref, no need to 786fa6ac715SQu Wenruo * rebuild this node for this inline ref. 787fa6ac715SQu Wenruo */ 7885d4f98a2SYan Zheng if (exist && 7895d4f98a2SYan Zheng ((key.type == BTRFS_TREE_BLOCK_REF_KEY && 7905d4f98a2SYan Zheng exist->owner == key.offset) || 7915d4f98a2SYan Zheng (key.type == BTRFS_SHARED_BLOCK_REF_KEY && 7925d4f98a2SYan Zheng exist->bytenr == key.offset))) { 7935d4f98a2SYan Zheng exist = NULL; 7945d4f98a2SYan Zheng goto next; 7955d4f98a2SYan Zheng } 7965d4f98a2SYan Zheng 797fa6ac715SQu Wenruo /* SHARED_BLOCK_REF means key.offset is the parent bytenr */ 7985d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_BLOCK_REF_KEY) { 7995d4f98a2SYan Zheng if (key.objectid == key.offset) { 8005d4f98a2SYan Zheng /* 801fa6ac715SQu Wenruo * Only root blocks of reloc trees use backref 802fa6ac715SQu Wenruo * pointing to itself. 8035d4f98a2SYan Zheng */ 8045d4f98a2SYan Zheng root = find_reloc_root(rc, cur->bytenr); 80575bfb9afSJosef Bacik ASSERT(root); 8065d4f98a2SYan Zheng cur->root = root; 8075d4f98a2SYan Zheng break; 8085d4f98a2SYan Zheng } 8095d4f98a2SYan Zheng 8103fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 8115d4f98a2SYan Zheng if (!edge) { 8125d4f98a2SYan Zheng err = -ENOMEM; 8135d4f98a2SYan Zheng goto out; 8145d4f98a2SYan Zheng } 8155d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, key.offset); 8165d4f98a2SYan Zheng if (!rb_node) { 8173fd0a558SYan, Zheng upper = alloc_backref_node(cache); 8185d4f98a2SYan Zheng if (!upper) { 8193fd0a558SYan, Zheng free_backref_edge(cache, edge); 8205d4f98a2SYan Zheng err = -ENOMEM; 8215d4f98a2SYan Zheng goto out; 8225d4f98a2SYan Zheng } 8235d4f98a2SYan Zheng upper->bytenr = key.offset; 8245d4f98a2SYan Zheng upper->level = cur->level + 1; 8255d4f98a2SYan Zheng /* 8265d4f98a2SYan Zheng * backrefs for the upper level block isn't 8275d4f98a2SYan Zheng * cached, add the block to pending list 8285d4f98a2SYan Zheng */ 8295d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 8305d4f98a2SYan Zheng } else { 8315d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 8325d4f98a2SYan Zheng rb_node); 83375bfb9afSJosef Bacik ASSERT(upper->checked); 8345d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 8355d4f98a2SYan Zheng } 8363fd0a558SYan, Zheng list_add_tail(&edge->list[LOWER], &cur->upper); 8375d4f98a2SYan Zheng edge->node[LOWER] = cur; 8383fd0a558SYan, Zheng edge->node[UPPER] = upper; 8395d4f98a2SYan Zheng 8405d4f98a2SYan Zheng goto next; 8416d8ff4e4SDavid Sterba } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { 842ba3c2b19SNikolay Borisov err = -EINVAL; 843ba3c2b19SNikolay Borisov btrfs_print_v0_err(rc->extent_root->fs_info); 844ba3c2b19SNikolay Borisov btrfs_handle_fs_error(rc->extent_root->fs_info, err, 845ba3c2b19SNikolay Borisov NULL); 846ba3c2b19SNikolay Borisov goto out; 8475d4f98a2SYan Zheng } else if (key.type != BTRFS_TREE_BLOCK_REF_KEY) { 8485d4f98a2SYan Zheng goto next; 8495d4f98a2SYan Zheng } 8505d4f98a2SYan Zheng 851fa6ac715SQu Wenruo /* 852fa6ac715SQu Wenruo * key.type == BTRFS_TREE_BLOCK_REF_KEY, inline ref offset 853fa6ac715SQu Wenruo * means the root objectid. We need to search the tree to get 854fa6ac715SQu Wenruo * its parent bytenr. 855fa6ac715SQu Wenruo */ 8565d4f98a2SYan Zheng root = read_fs_root(rc->extent_root->fs_info, key.offset); 8575d4f98a2SYan Zheng if (IS_ERR(root)) { 8585d4f98a2SYan Zheng err = PTR_ERR(root); 8595d4f98a2SYan Zheng goto out; 8605d4f98a2SYan Zheng } 8615d4f98a2SYan Zheng 86227cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 8633fd0a558SYan, Zheng cur->cowonly = 1; 8643fd0a558SYan, Zheng 8655d4f98a2SYan Zheng if (btrfs_root_level(&root->root_item) == cur->level) { 8665d4f98a2SYan Zheng /* tree root */ 86775bfb9afSJosef Bacik ASSERT(btrfs_root_bytenr(&root->root_item) == 8685d4f98a2SYan Zheng cur->bytenr); 8693fd0a558SYan, Zheng if (should_ignore_root(root)) 8703fd0a558SYan, Zheng list_add(&cur->list, &useless); 8713fd0a558SYan, Zheng else 8725d4f98a2SYan Zheng cur->root = root; 8735d4f98a2SYan Zheng break; 8745d4f98a2SYan Zheng } 8755d4f98a2SYan Zheng 8765d4f98a2SYan Zheng level = cur->level + 1; 8775d4f98a2SYan Zheng 878fa6ac715SQu Wenruo /* Search the tree to find parent blocks referring the block. */ 8795d4f98a2SYan Zheng path2->search_commit_root = 1; 8805d4f98a2SYan Zheng path2->skip_locking = 1; 8815d4f98a2SYan Zheng path2->lowest_level = level; 8825d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root, node_key, path2, 0, 0); 8835d4f98a2SYan Zheng path2->lowest_level = 0; 8845d4f98a2SYan Zheng if (ret < 0) { 8855d4f98a2SYan Zheng err = ret; 8865d4f98a2SYan Zheng goto out; 8875d4f98a2SYan Zheng } 88833c66f43SYan Zheng if (ret > 0 && path2->slots[level] > 0) 88933c66f43SYan Zheng path2->slots[level]--; 8905d4f98a2SYan Zheng 8915d4f98a2SYan Zheng eb = path2->nodes[level]; 8923561b9dbSLiu Bo if (btrfs_node_blockptr(eb, path2->slots[level]) != 8933561b9dbSLiu Bo cur->bytenr) { 8943561b9dbSLiu Bo btrfs_err(root->fs_info, 8953561b9dbSLiu Bo "couldn't find block (%llu) (level %d) in tree (%llu) with key (%llu %u %llu)", 8964fd786e6SMisono Tomohiro cur->bytenr, level - 1, 8974fd786e6SMisono Tomohiro root->root_key.objectid, 8983561b9dbSLiu Bo node_key->objectid, node_key->type, 8993561b9dbSLiu Bo node_key->offset); 9003561b9dbSLiu Bo err = -ENOENT; 9013561b9dbSLiu Bo goto out; 9023561b9dbSLiu Bo } 9035d4f98a2SYan Zheng lower = cur; 904b6c60c80SJosef Bacik need_check = true; 905fa6ac715SQu Wenruo 906fa6ac715SQu Wenruo /* Add all nodes and edges in the path */ 9075d4f98a2SYan Zheng for (; level < BTRFS_MAX_LEVEL; level++) { 9085d4f98a2SYan Zheng if (!path2->nodes[level]) { 90975bfb9afSJosef Bacik ASSERT(btrfs_root_bytenr(&root->root_item) == 9105d4f98a2SYan Zheng lower->bytenr); 9113fd0a558SYan, Zheng if (should_ignore_root(root)) 9123fd0a558SYan, Zheng list_add(&lower->list, &useless); 9133fd0a558SYan, Zheng else 9145d4f98a2SYan Zheng lower->root = root; 9155d4f98a2SYan Zheng break; 9165d4f98a2SYan Zheng } 9175d4f98a2SYan Zheng 9183fd0a558SYan, Zheng edge = alloc_backref_edge(cache); 9195d4f98a2SYan Zheng if (!edge) { 9205d4f98a2SYan Zheng err = -ENOMEM; 9215d4f98a2SYan Zheng goto out; 9225d4f98a2SYan Zheng } 9235d4f98a2SYan Zheng 9245d4f98a2SYan Zheng eb = path2->nodes[level]; 9255d4f98a2SYan Zheng rb_node = tree_search(&cache->rb_root, eb->start); 9265d4f98a2SYan Zheng if (!rb_node) { 9273fd0a558SYan, Zheng upper = alloc_backref_node(cache); 9285d4f98a2SYan Zheng if (!upper) { 9293fd0a558SYan, Zheng free_backref_edge(cache, edge); 9305d4f98a2SYan Zheng err = -ENOMEM; 9315d4f98a2SYan Zheng goto out; 9325d4f98a2SYan Zheng } 9335d4f98a2SYan Zheng upper->bytenr = eb->start; 9345d4f98a2SYan Zheng upper->owner = btrfs_header_owner(eb); 9355d4f98a2SYan Zheng upper->level = lower->level + 1; 93627cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, 93727cdeb70SMiao Xie &root->state)) 9383fd0a558SYan, Zheng upper->cowonly = 1; 9395d4f98a2SYan Zheng 9405d4f98a2SYan Zheng /* 9415d4f98a2SYan Zheng * if we know the block isn't shared 9425d4f98a2SYan Zheng * we can void checking its backrefs. 9435d4f98a2SYan Zheng */ 9445d4f98a2SYan Zheng if (btrfs_block_can_be_shared(root, eb)) 9455d4f98a2SYan Zheng upper->checked = 0; 9465d4f98a2SYan Zheng else 9475d4f98a2SYan Zheng upper->checked = 1; 9485d4f98a2SYan Zheng 9495d4f98a2SYan Zheng /* 9505d4f98a2SYan Zheng * add the block to pending list if we 951b6c60c80SJosef Bacik * need check its backrefs, we only do this once 952b6c60c80SJosef Bacik * while walking up a tree as we will catch 953b6c60c80SJosef Bacik * anything else later on. 9545d4f98a2SYan Zheng */ 955b6c60c80SJosef Bacik if (!upper->checked && need_check) { 956b6c60c80SJosef Bacik need_check = false; 9575d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], 9585d4f98a2SYan Zheng &list); 959bbe90514SJosef Bacik } else { 960bbe90514SJosef Bacik if (upper->checked) 961bbe90514SJosef Bacik need_check = true; 9625d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 963bbe90514SJosef Bacik } 9645d4f98a2SYan Zheng } else { 9655d4f98a2SYan Zheng upper = rb_entry(rb_node, struct backref_node, 9665d4f98a2SYan Zheng rb_node); 96775bfb9afSJosef Bacik ASSERT(upper->checked); 9685d4f98a2SYan Zheng INIT_LIST_HEAD(&edge->list[UPPER]); 9693fd0a558SYan, Zheng if (!upper->owner) 9703fd0a558SYan, Zheng upper->owner = btrfs_header_owner(eb); 9715d4f98a2SYan Zheng } 9725d4f98a2SYan Zheng list_add_tail(&edge->list[LOWER], &lower->upper); 9735d4f98a2SYan Zheng edge->node[LOWER] = lower; 9743fd0a558SYan, Zheng edge->node[UPPER] = upper; 9755d4f98a2SYan Zheng 9765d4f98a2SYan Zheng if (rb_node) 9775d4f98a2SYan Zheng break; 9785d4f98a2SYan Zheng lower = upper; 9795d4f98a2SYan Zheng upper = NULL; 9805d4f98a2SYan Zheng } 981b3b4aa74SDavid Sterba btrfs_release_path(path2); 9825d4f98a2SYan Zheng next: 9835d4f98a2SYan Zheng if (ptr < end) { 9845d4f98a2SYan Zheng ptr += btrfs_extent_inline_ref_size(key.type); 9855d4f98a2SYan Zheng if (ptr >= end) { 9865d4f98a2SYan Zheng WARN_ON(ptr > end); 9875d4f98a2SYan Zheng ptr = 0; 9885d4f98a2SYan Zheng end = 0; 9895d4f98a2SYan Zheng } 9905d4f98a2SYan Zheng } 9915d4f98a2SYan Zheng if (ptr >= end) 9925d4f98a2SYan Zheng path1->slots[0]++; 9935d4f98a2SYan Zheng } 994b3b4aa74SDavid Sterba btrfs_release_path(path1); 9955d4f98a2SYan Zheng 9965d4f98a2SYan Zheng cur->checked = 1; 9975d4f98a2SYan Zheng WARN_ON(exist); 9985d4f98a2SYan Zheng 9995d4f98a2SYan Zheng /* the pending list isn't empty, take the first block to process */ 10005d4f98a2SYan Zheng if (!list_empty(&list)) { 10015d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, list[UPPER]); 10025d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 10035d4f98a2SYan Zheng cur = edge->node[UPPER]; 10045d4f98a2SYan Zheng goto again; 10055d4f98a2SYan Zheng } 10065d4f98a2SYan Zheng 10075d4f98a2SYan Zheng /* 10085d4f98a2SYan Zheng * everything goes well, connect backref nodes and insert backref nodes 10095d4f98a2SYan Zheng * into the cache. 10105d4f98a2SYan Zheng */ 101175bfb9afSJosef Bacik ASSERT(node->checked); 10123fd0a558SYan, Zheng cowonly = node->cowonly; 10133fd0a558SYan, Zheng if (!cowonly) { 10143fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, node->bytenr, 10153fd0a558SYan, Zheng &node->rb_node); 101643c04fb1SJeff Mahoney if (rb_node) 101743c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 10183fd0a558SYan, Zheng list_add_tail(&node->lower, &cache->leaves); 10193fd0a558SYan, Zheng } 10205d4f98a2SYan Zheng 10215d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) 10225d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 10235d4f98a2SYan Zheng 10245d4f98a2SYan Zheng while (!list_empty(&list)) { 10255d4f98a2SYan Zheng edge = list_entry(list.next, struct backref_edge, list[UPPER]); 10265d4f98a2SYan Zheng list_del_init(&edge->list[UPPER]); 10275d4f98a2SYan Zheng upper = edge->node[UPPER]; 10283fd0a558SYan, Zheng if (upper->detached) { 10293fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 10303fd0a558SYan, Zheng lower = edge->node[LOWER]; 10313fd0a558SYan, Zheng free_backref_edge(cache, edge); 10323fd0a558SYan, Zheng if (list_empty(&lower->upper)) 10333fd0a558SYan, Zheng list_add(&lower->list, &useless); 10343fd0a558SYan, Zheng continue; 10353fd0a558SYan, Zheng } 10365d4f98a2SYan Zheng 10375d4f98a2SYan Zheng if (!RB_EMPTY_NODE(&upper->rb_node)) { 10385d4f98a2SYan Zheng if (upper->lowest) { 10395d4f98a2SYan Zheng list_del_init(&upper->lower); 10405d4f98a2SYan Zheng upper->lowest = 0; 10415d4f98a2SYan Zheng } 10425d4f98a2SYan Zheng 10435d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 10445d4f98a2SYan Zheng continue; 10455d4f98a2SYan Zheng } 10465d4f98a2SYan Zheng 104775bfb9afSJosef Bacik if (!upper->checked) { 104875bfb9afSJosef Bacik /* 104975bfb9afSJosef Bacik * Still want to blow up for developers since this is a 105075bfb9afSJosef Bacik * logic bug. 105175bfb9afSJosef Bacik */ 105275bfb9afSJosef Bacik ASSERT(0); 105375bfb9afSJosef Bacik err = -EINVAL; 105475bfb9afSJosef Bacik goto out; 105575bfb9afSJosef Bacik } 105675bfb9afSJosef Bacik if (cowonly != upper->cowonly) { 105775bfb9afSJosef Bacik ASSERT(0); 105875bfb9afSJosef Bacik err = -EINVAL; 105975bfb9afSJosef Bacik goto out; 106075bfb9afSJosef Bacik } 106175bfb9afSJosef Bacik 10623fd0a558SYan, Zheng if (!cowonly) { 10635d4f98a2SYan Zheng rb_node = tree_insert(&cache->rb_root, upper->bytenr, 10645d4f98a2SYan Zheng &upper->rb_node); 106543c04fb1SJeff Mahoney if (rb_node) 106643c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 106743c04fb1SJeff Mahoney upper->bytenr); 10683fd0a558SYan, Zheng } 10695d4f98a2SYan Zheng 10705d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &upper->lower); 10715d4f98a2SYan Zheng 10725d4f98a2SYan Zheng list_for_each_entry(edge, &upper->upper, list[LOWER]) 10735d4f98a2SYan Zheng list_add_tail(&edge->list[UPPER], &list); 10745d4f98a2SYan Zheng } 10753fd0a558SYan, Zheng /* 10763fd0a558SYan, Zheng * process useless backref nodes. backref nodes for tree leaves 10773fd0a558SYan, Zheng * are deleted from the cache. backref nodes for upper level 10783fd0a558SYan, Zheng * tree blocks are left in the cache to avoid unnecessary backref 10793fd0a558SYan, Zheng * lookup. 10803fd0a558SYan, Zheng */ 10813fd0a558SYan, Zheng while (!list_empty(&useless)) { 10823fd0a558SYan, Zheng upper = list_entry(useless.next, struct backref_node, list); 10833fd0a558SYan, Zheng list_del_init(&upper->list); 108475bfb9afSJosef Bacik ASSERT(list_empty(&upper->upper)); 10853fd0a558SYan, Zheng if (upper == node) 10863fd0a558SYan, Zheng node = NULL; 10873fd0a558SYan, Zheng if (upper->lowest) { 10883fd0a558SYan, Zheng list_del_init(&upper->lower); 10893fd0a558SYan, Zheng upper->lowest = 0; 10903fd0a558SYan, Zheng } 10913fd0a558SYan, Zheng while (!list_empty(&upper->lower)) { 10923fd0a558SYan, Zheng edge = list_entry(upper->lower.next, 10933fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 10943fd0a558SYan, Zheng list_del(&edge->list[UPPER]); 10953fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 10963fd0a558SYan, Zheng lower = edge->node[LOWER]; 10973fd0a558SYan, Zheng free_backref_edge(cache, edge); 10983fd0a558SYan, Zheng 10993fd0a558SYan, Zheng if (list_empty(&lower->upper)) 11003fd0a558SYan, Zheng list_add(&lower->list, &useless); 11013fd0a558SYan, Zheng } 11023fd0a558SYan, Zheng __mark_block_processed(rc, upper); 11033fd0a558SYan, Zheng if (upper->level > 0) { 11043fd0a558SYan, Zheng list_add(&upper->list, &cache->detached); 11053fd0a558SYan, Zheng upper->detached = 1; 11063fd0a558SYan, Zheng } else { 11073fd0a558SYan, Zheng rb_erase(&upper->rb_node, &cache->rb_root); 11083fd0a558SYan, Zheng free_backref_node(cache, upper); 11093fd0a558SYan, Zheng } 11103fd0a558SYan, Zheng } 11115d4f98a2SYan Zheng out: 11125d4f98a2SYan Zheng btrfs_free_path(path1); 11135d4f98a2SYan Zheng btrfs_free_path(path2); 11145d4f98a2SYan Zheng if (err) { 11153fd0a558SYan, Zheng while (!list_empty(&useless)) { 11163fd0a558SYan, Zheng lower = list_entry(useless.next, 111775bfb9afSJosef Bacik struct backref_node, list); 111875bfb9afSJosef Bacik list_del_init(&lower->list); 11193fd0a558SYan, Zheng } 112075bfb9afSJosef Bacik while (!list_empty(&list)) { 112175bfb9afSJosef Bacik edge = list_first_entry(&list, struct backref_edge, 112275bfb9afSJosef Bacik list[UPPER]); 112375bfb9afSJosef Bacik list_del(&edge->list[UPPER]); 11243fd0a558SYan, Zheng list_del(&edge->list[LOWER]); 112575bfb9afSJosef Bacik lower = edge->node[LOWER]; 11265d4f98a2SYan Zheng upper = edge->node[UPPER]; 11273fd0a558SYan, Zheng free_backref_edge(cache, edge); 112875bfb9afSJosef Bacik 112975bfb9afSJosef Bacik /* 113075bfb9afSJosef Bacik * Lower is no longer linked to any upper backref nodes 113175bfb9afSJosef Bacik * and isn't in the cache, we can free it ourselves. 113275bfb9afSJosef Bacik */ 113375bfb9afSJosef Bacik if (list_empty(&lower->upper) && 113475bfb9afSJosef Bacik RB_EMPTY_NODE(&lower->rb_node)) 113575bfb9afSJosef Bacik list_add(&lower->list, &useless); 113675bfb9afSJosef Bacik 113775bfb9afSJosef Bacik if (!RB_EMPTY_NODE(&upper->rb_node)) 113875bfb9afSJosef Bacik continue; 113975bfb9afSJosef Bacik 114001327610SNicholas D Steeves /* Add this guy's upper edges to the list to process */ 114175bfb9afSJosef Bacik list_for_each_entry(edge, &upper->upper, list[LOWER]) 114275bfb9afSJosef Bacik list_add_tail(&edge->list[UPPER], &list); 114375bfb9afSJosef Bacik if (list_empty(&upper->upper)) 114475bfb9afSJosef Bacik list_add(&upper->list, &useless); 114575bfb9afSJosef Bacik } 114675bfb9afSJosef Bacik 114775bfb9afSJosef Bacik while (!list_empty(&useless)) { 114875bfb9afSJosef Bacik lower = list_entry(useless.next, 114975bfb9afSJosef Bacik struct backref_node, list); 115075bfb9afSJosef Bacik list_del_init(&lower->list); 11510fd8c3daSLiu Bo if (lower == node) 11520fd8c3daSLiu Bo node = NULL; 115375bfb9afSJosef Bacik free_backref_node(cache, lower); 11545d4f98a2SYan Zheng } 11550fd8c3daSLiu Bo 11560fd8c3daSLiu Bo free_backref_node(cache, node); 11575d4f98a2SYan Zheng return ERR_PTR(err); 11585d4f98a2SYan Zheng } 115975bfb9afSJosef Bacik ASSERT(!node || !node->detached); 11605d4f98a2SYan Zheng return node; 11615d4f98a2SYan Zheng } 11625d4f98a2SYan Zheng 11635d4f98a2SYan Zheng /* 11643fd0a558SYan, Zheng * helper to add backref node for the newly created snapshot. 11653fd0a558SYan, Zheng * the backref node is created by cloning backref node that 11663fd0a558SYan, Zheng * corresponds to root of source tree 11673fd0a558SYan, Zheng */ 11683fd0a558SYan, Zheng static int clone_backref_node(struct btrfs_trans_handle *trans, 11693fd0a558SYan, Zheng struct reloc_control *rc, 11703fd0a558SYan, Zheng struct btrfs_root *src, 11713fd0a558SYan, Zheng struct btrfs_root *dest) 11723fd0a558SYan, Zheng { 11733fd0a558SYan, Zheng struct btrfs_root *reloc_root = src->reloc_root; 11743fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 11753fd0a558SYan, Zheng struct backref_node *node = NULL; 11763fd0a558SYan, Zheng struct backref_node *new_node; 11773fd0a558SYan, Zheng struct backref_edge *edge; 11783fd0a558SYan, Zheng struct backref_edge *new_edge; 11793fd0a558SYan, Zheng struct rb_node *rb_node; 11803fd0a558SYan, Zheng 11813fd0a558SYan, Zheng if (cache->last_trans > 0) 11823fd0a558SYan, Zheng update_backref_cache(trans, cache); 11833fd0a558SYan, Zheng 11843fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, src->commit_root->start); 11853fd0a558SYan, Zheng if (rb_node) { 11863fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, rb_node); 11873fd0a558SYan, Zheng if (node->detached) 11883fd0a558SYan, Zheng node = NULL; 11893fd0a558SYan, Zheng else 11903fd0a558SYan, Zheng BUG_ON(node->new_bytenr != reloc_root->node->start); 11913fd0a558SYan, Zheng } 11923fd0a558SYan, Zheng 11933fd0a558SYan, Zheng if (!node) { 11943fd0a558SYan, Zheng rb_node = tree_search(&cache->rb_root, 11953fd0a558SYan, Zheng reloc_root->commit_root->start); 11963fd0a558SYan, Zheng if (rb_node) { 11973fd0a558SYan, Zheng node = rb_entry(rb_node, struct backref_node, 11983fd0a558SYan, Zheng rb_node); 11993fd0a558SYan, Zheng BUG_ON(node->detached); 12003fd0a558SYan, Zheng } 12013fd0a558SYan, Zheng } 12023fd0a558SYan, Zheng 12033fd0a558SYan, Zheng if (!node) 12043fd0a558SYan, Zheng return 0; 12053fd0a558SYan, Zheng 12063fd0a558SYan, Zheng new_node = alloc_backref_node(cache); 12073fd0a558SYan, Zheng if (!new_node) 12083fd0a558SYan, Zheng return -ENOMEM; 12093fd0a558SYan, Zheng 12103fd0a558SYan, Zheng new_node->bytenr = dest->node->start; 12113fd0a558SYan, Zheng new_node->level = node->level; 12123fd0a558SYan, Zheng new_node->lowest = node->lowest; 12136848ad64SYan, Zheng new_node->checked = 1; 12143fd0a558SYan, Zheng new_node->root = dest; 12153fd0a558SYan, Zheng 12163fd0a558SYan, Zheng if (!node->lowest) { 12173fd0a558SYan, Zheng list_for_each_entry(edge, &node->lower, list[UPPER]) { 12183fd0a558SYan, Zheng new_edge = alloc_backref_edge(cache); 12193fd0a558SYan, Zheng if (!new_edge) 12203fd0a558SYan, Zheng goto fail; 12213fd0a558SYan, Zheng 12223fd0a558SYan, Zheng new_edge->node[UPPER] = new_node; 12233fd0a558SYan, Zheng new_edge->node[LOWER] = edge->node[LOWER]; 12243fd0a558SYan, Zheng list_add_tail(&new_edge->list[UPPER], 12253fd0a558SYan, Zheng &new_node->lower); 12263fd0a558SYan, Zheng } 122776b9e23dSMiao Xie } else { 122876b9e23dSMiao Xie list_add_tail(&new_node->lower, &cache->leaves); 12293fd0a558SYan, Zheng } 12303fd0a558SYan, Zheng 12313fd0a558SYan, Zheng rb_node = tree_insert(&cache->rb_root, new_node->bytenr, 12323fd0a558SYan, Zheng &new_node->rb_node); 123343c04fb1SJeff Mahoney if (rb_node) 123443c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, new_node->bytenr); 12353fd0a558SYan, Zheng 12363fd0a558SYan, Zheng if (!new_node->lowest) { 12373fd0a558SYan, Zheng list_for_each_entry(new_edge, &new_node->lower, list[UPPER]) { 12383fd0a558SYan, Zheng list_add_tail(&new_edge->list[LOWER], 12393fd0a558SYan, Zheng &new_edge->node[LOWER]->upper); 12403fd0a558SYan, Zheng } 12413fd0a558SYan, Zheng } 12423fd0a558SYan, Zheng return 0; 12433fd0a558SYan, Zheng fail: 12443fd0a558SYan, Zheng while (!list_empty(&new_node->lower)) { 12453fd0a558SYan, Zheng new_edge = list_entry(new_node->lower.next, 12463fd0a558SYan, Zheng struct backref_edge, list[UPPER]); 12473fd0a558SYan, Zheng list_del(&new_edge->list[UPPER]); 12483fd0a558SYan, Zheng free_backref_edge(cache, new_edge); 12493fd0a558SYan, Zheng } 12503fd0a558SYan, Zheng free_backref_node(cache, new_node); 12513fd0a558SYan, Zheng return -ENOMEM; 12523fd0a558SYan, Zheng } 12533fd0a558SYan, Zheng 12543fd0a558SYan, Zheng /* 12555d4f98a2SYan Zheng * helper to add 'address of tree root -> reloc tree' mapping 12565d4f98a2SYan Zheng */ 1257ffd7b339SJeff Mahoney static int __must_check __add_reloc_root(struct btrfs_root *root) 12585d4f98a2SYan Zheng { 12590b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 12605d4f98a2SYan Zheng struct rb_node *rb_node; 12615d4f98a2SYan Zheng struct mapping_node *node; 12620b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 12635d4f98a2SYan Zheng 12645d4f98a2SYan Zheng node = kmalloc(sizeof(*node), GFP_NOFS); 1265ffd7b339SJeff Mahoney if (!node) 1266ffd7b339SJeff Mahoney return -ENOMEM; 12675d4f98a2SYan Zheng 12685d4f98a2SYan Zheng node->bytenr = root->node->start; 12695d4f98a2SYan Zheng node->data = root; 12705d4f98a2SYan Zheng 12715d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 12725d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 12735d4f98a2SYan Zheng node->bytenr, &node->rb_node); 12745d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 1275ffd7b339SJeff Mahoney if (rb_node) { 12760b246afaSJeff Mahoney btrfs_panic(fs_info, -EEXIST, 12775d163e0eSJeff Mahoney "Duplicate root found for start=%llu while inserting into relocation tree", 12785d163e0eSJeff Mahoney node->bytenr); 1279ffd7b339SJeff Mahoney } 12805d4f98a2SYan Zheng 12815d4f98a2SYan Zheng list_add_tail(&root->root_list, &rc->reloc_roots); 12825d4f98a2SYan Zheng return 0; 12835d4f98a2SYan Zheng } 12845d4f98a2SYan Zheng 12855d4f98a2SYan Zheng /* 1286c974c464SWang Shilong * helper to delete the 'address of tree root -> reloc tree' 12875d4f98a2SYan Zheng * mapping 12885d4f98a2SYan Zheng */ 1289c974c464SWang Shilong static void __del_reloc_root(struct btrfs_root *root) 12905d4f98a2SYan Zheng { 12910b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 12925d4f98a2SYan Zheng struct rb_node *rb_node; 12935d4f98a2SYan Zheng struct mapping_node *node = NULL; 12940b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 12955d4f98a2SYan Zheng 129665c6e82bSQu Wenruo if (rc && root->node) { 12975d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 12985d4f98a2SYan Zheng rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1299c974c464SWang Shilong root->node->start); 1300c974c464SWang Shilong if (rb_node) { 1301c974c464SWang Shilong node = rb_entry(rb_node, struct mapping_node, rb_node); 1302c974c464SWang Shilong rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 1303c974c464SWang Shilong } 1304c974c464SWang Shilong spin_unlock(&rc->reloc_root_tree.lock); 1305c974c464SWang Shilong if (!node) 1306c974c464SWang Shilong return; 1307c974c464SWang Shilong BUG_ON((struct btrfs_root *)node->data != root); 1308389305b2SQu Wenruo } 1309c974c464SWang Shilong 13100b246afaSJeff Mahoney spin_lock(&fs_info->trans_lock); 1311c974c464SWang Shilong list_del_init(&root->root_list); 13120b246afaSJeff Mahoney spin_unlock(&fs_info->trans_lock); 1313c974c464SWang Shilong kfree(node); 1314c974c464SWang Shilong } 1315c974c464SWang Shilong 1316c974c464SWang Shilong /* 1317c974c464SWang Shilong * helper to update the 'address of tree root -> reloc tree' 1318c974c464SWang Shilong * mapping 1319c974c464SWang Shilong */ 1320c974c464SWang Shilong static int __update_reloc_root(struct btrfs_root *root, u64 new_bytenr) 1321c974c464SWang Shilong { 13220b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 1323c974c464SWang Shilong struct rb_node *rb_node; 1324c974c464SWang Shilong struct mapping_node *node = NULL; 13250b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 1326c974c464SWang Shilong 1327c974c464SWang Shilong spin_lock(&rc->reloc_root_tree.lock); 1328c974c464SWang Shilong rb_node = tree_search(&rc->reloc_root_tree.rb_root, 1329c974c464SWang Shilong root->node->start); 13305d4f98a2SYan Zheng if (rb_node) { 13315d4f98a2SYan Zheng node = rb_entry(rb_node, struct mapping_node, rb_node); 13325d4f98a2SYan Zheng rb_erase(&node->rb_node, &rc->reloc_root_tree.rb_root); 13335d4f98a2SYan Zheng } 13345d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 13355d4f98a2SYan Zheng 13368f71f3e0SLiu Bo if (!node) 13378f71f3e0SLiu Bo return 0; 13385d4f98a2SYan Zheng BUG_ON((struct btrfs_root *)node->data != root); 13395d4f98a2SYan Zheng 13405d4f98a2SYan Zheng spin_lock(&rc->reloc_root_tree.lock); 1341c974c464SWang Shilong node->bytenr = new_bytenr; 13425d4f98a2SYan Zheng rb_node = tree_insert(&rc->reloc_root_tree.rb_root, 13435d4f98a2SYan Zheng node->bytenr, &node->rb_node); 13445d4f98a2SYan Zheng spin_unlock(&rc->reloc_root_tree.lock); 134543c04fb1SJeff Mahoney if (rb_node) 134643c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, node->bytenr); 13475d4f98a2SYan Zheng return 0; 13485d4f98a2SYan Zheng } 13495d4f98a2SYan Zheng 13503fd0a558SYan, Zheng static struct btrfs_root *create_reloc_root(struct btrfs_trans_handle *trans, 13513fd0a558SYan, Zheng struct btrfs_root *root, u64 objectid) 13525d4f98a2SYan Zheng { 13530b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 13545d4f98a2SYan Zheng struct btrfs_root *reloc_root; 13555d4f98a2SYan Zheng struct extent_buffer *eb; 13565d4f98a2SYan Zheng struct btrfs_root_item *root_item; 13575d4f98a2SYan Zheng struct btrfs_key root_key; 13585d4f98a2SYan Zheng int ret; 13595d4f98a2SYan Zheng 13605d4f98a2SYan Zheng root_item = kmalloc(sizeof(*root_item), GFP_NOFS); 13615d4f98a2SYan Zheng BUG_ON(!root_item); 13625d4f98a2SYan Zheng 13635d4f98a2SYan Zheng root_key.objectid = BTRFS_TREE_RELOC_OBJECTID; 13645d4f98a2SYan Zheng root_key.type = BTRFS_ROOT_ITEM_KEY; 13653fd0a558SYan, Zheng root_key.offset = objectid; 13665d4f98a2SYan Zheng 13673fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 1368054570a1SFilipe Manana u64 commit_root_gen; 1369054570a1SFilipe Manana 13703fd0a558SYan, Zheng /* called by btrfs_init_reloc_root */ 13715d4f98a2SYan Zheng ret = btrfs_copy_root(trans, root, root->commit_root, &eb, 13725d4f98a2SYan Zheng BTRFS_TREE_RELOC_OBJECTID); 13735d4f98a2SYan Zheng BUG_ON(ret); 1374054570a1SFilipe Manana /* 1375054570a1SFilipe Manana * Set the last_snapshot field to the generation of the commit 1376054570a1SFilipe Manana * root - like this ctree.c:btrfs_block_can_be_shared() behaves 1377054570a1SFilipe Manana * correctly (returns true) when the relocation root is created 1378054570a1SFilipe Manana * either inside the critical section of a transaction commit 1379054570a1SFilipe Manana * (through transaction.c:qgroup_account_snapshot()) and when 1380054570a1SFilipe Manana * it's created before the transaction commit is started. 1381054570a1SFilipe Manana */ 1382054570a1SFilipe Manana commit_root_gen = btrfs_header_generation(root->commit_root); 1383054570a1SFilipe Manana btrfs_set_root_last_snapshot(&root->root_item, commit_root_gen); 13843fd0a558SYan, Zheng } else { 13853fd0a558SYan, Zheng /* 13863fd0a558SYan, Zheng * called by btrfs_reloc_post_snapshot_hook. 13873fd0a558SYan, Zheng * the source tree is a reloc tree, all tree blocks 13883fd0a558SYan, Zheng * modified after it was created have RELOC flag 13893fd0a558SYan, Zheng * set in their headers. so it's OK to not update 13903fd0a558SYan, Zheng * the 'last_snapshot'. 13913fd0a558SYan, Zheng */ 13923fd0a558SYan, Zheng ret = btrfs_copy_root(trans, root, root->node, &eb, 13933fd0a558SYan, Zheng BTRFS_TREE_RELOC_OBJECTID); 13943fd0a558SYan, Zheng BUG_ON(ret); 13953fd0a558SYan, Zheng } 13963fd0a558SYan, Zheng 13975d4f98a2SYan Zheng memcpy(root_item, &root->root_item, sizeof(*root_item)); 13985d4f98a2SYan Zheng btrfs_set_root_bytenr(root_item, eb->start); 13995d4f98a2SYan Zheng btrfs_set_root_level(root_item, btrfs_header_level(eb)); 14005d4f98a2SYan Zheng btrfs_set_root_generation(root_item, trans->transid); 14013fd0a558SYan, Zheng 14023fd0a558SYan, Zheng if (root->root_key.objectid == objectid) { 14033fd0a558SYan, Zheng btrfs_set_root_refs(root_item, 0); 14043fd0a558SYan, Zheng memset(&root_item->drop_progress, 0, 14053fd0a558SYan, Zheng sizeof(struct btrfs_disk_key)); 14065d4f98a2SYan Zheng root_item->drop_level = 0; 14073fd0a558SYan, Zheng } 14085d4f98a2SYan Zheng 14095d4f98a2SYan Zheng btrfs_tree_unlock(eb); 14105d4f98a2SYan Zheng free_extent_buffer(eb); 14115d4f98a2SYan Zheng 14120b246afaSJeff Mahoney ret = btrfs_insert_root(trans, fs_info->tree_root, 14135d4f98a2SYan Zheng &root_key, root_item); 14145d4f98a2SYan Zheng BUG_ON(ret); 14155d4f98a2SYan Zheng kfree(root_item); 14165d4f98a2SYan Zheng 14170b246afaSJeff Mahoney reloc_root = btrfs_read_fs_root(fs_info->tree_root, &root_key); 14185d4f98a2SYan Zheng BUG_ON(IS_ERR(reloc_root)); 14195d4f98a2SYan Zheng reloc_root->last_trans = trans->transid; 14203fd0a558SYan, Zheng return reloc_root; 14213fd0a558SYan, Zheng } 14223fd0a558SYan, Zheng 14233fd0a558SYan, Zheng /* 14243fd0a558SYan, Zheng * create reloc tree for a given fs tree. reloc tree is just a 14253fd0a558SYan, Zheng * snapshot of the fs tree with special root objectid. 14263fd0a558SYan, Zheng */ 14273fd0a558SYan, Zheng int btrfs_init_reloc_root(struct btrfs_trans_handle *trans, 14283fd0a558SYan, Zheng struct btrfs_root *root) 14293fd0a558SYan, Zheng { 14300b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 14313fd0a558SYan, Zheng struct btrfs_root *reloc_root; 14320b246afaSJeff Mahoney struct reloc_control *rc = fs_info->reloc_ctl; 143320dd2cbfSMiao Xie struct btrfs_block_rsv *rsv; 14343fd0a558SYan, Zheng int clear_rsv = 0; 1435ffd7b339SJeff Mahoney int ret; 14363fd0a558SYan, Zheng 14373fd0a558SYan, Zheng if (root->reloc_root) { 14383fd0a558SYan, Zheng reloc_root = root->reloc_root; 14393fd0a558SYan, Zheng reloc_root->last_trans = trans->transid; 14403fd0a558SYan, Zheng return 0; 14413fd0a558SYan, Zheng } 14423fd0a558SYan, Zheng 14433fd0a558SYan, Zheng if (!rc || !rc->create_reloc_tree || 14443fd0a558SYan, Zheng root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 14453fd0a558SYan, Zheng return 0; 14463fd0a558SYan, Zheng 144720dd2cbfSMiao Xie if (!trans->reloc_reserved) { 144820dd2cbfSMiao Xie rsv = trans->block_rsv; 14493fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 14503fd0a558SYan, Zheng clear_rsv = 1; 14513fd0a558SYan, Zheng } 14523fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root, root->root_key.objectid); 14533fd0a558SYan, Zheng if (clear_rsv) 145420dd2cbfSMiao Xie trans->block_rsv = rsv; 14555d4f98a2SYan Zheng 1456ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 1457ffd7b339SJeff Mahoney BUG_ON(ret < 0); 14585d4f98a2SYan Zheng root->reloc_root = reloc_root; 14595d4f98a2SYan Zheng return 0; 14605d4f98a2SYan Zheng } 14615d4f98a2SYan Zheng 14625d4f98a2SYan Zheng /* 14635d4f98a2SYan Zheng * update root item of reloc tree 14645d4f98a2SYan Zheng */ 14655d4f98a2SYan Zheng int btrfs_update_reloc_root(struct btrfs_trans_handle *trans, 14665d4f98a2SYan Zheng struct btrfs_root *root) 14675d4f98a2SYan Zheng { 14680b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 14695d4f98a2SYan Zheng struct btrfs_root *reloc_root; 14705d4f98a2SYan Zheng struct btrfs_root_item *root_item; 14715d4f98a2SYan Zheng int ret; 14725d4f98a2SYan Zheng 1473d2311e69SQu Wenruo if (test_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state) || 1474d2311e69SQu Wenruo !root->reloc_root) 14757585717fSChris Mason goto out; 14765d4f98a2SYan Zheng 14775d4f98a2SYan Zheng reloc_root = root->reloc_root; 14785d4f98a2SYan Zheng root_item = &reloc_root->root_item; 14795d4f98a2SYan Zheng 1480d2311e69SQu Wenruo /* root->reloc_root will stay until current relocation finished */ 14810b246afaSJeff Mahoney if (fs_info->reloc_ctl->merge_reloc_tree && 14823fd0a558SYan, Zheng btrfs_root_refs(root_item) == 0) { 1483d2311e69SQu Wenruo set_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 1484c974c464SWang Shilong __del_reloc_root(reloc_root); 14855d4f98a2SYan Zheng } 14865d4f98a2SYan Zheng 14875d4f98a2SYan Zheng if (reloc_root->commit_root != reloc_root->node) { 14885d4f98a2SYan Zheng btrfs_set_root_node(root_item, reloc_root->node); 14895d4f98a2SYan Zheng free_extent_buffer(reloc_root->commit_root); 14905d4f98a2SYan Zheng reloc_root->commit_root = btrfs_root_node(reloc_root); 14915d4f98a2SYan Zheng } 14925d4f98a2SYan Zheng 14930b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 14945d4f98a2SYan Zheng &reloc_root->root_key, root_item); 14955d4f98a2SYan Zheng BUG_ON(ret); 14967585717fSChris Mason 14977585717fSChris Mason out: 14985d4f98a2SYan Zheng return 0; 14995d4f98a2SYan Zheng } 15005d4f98a2SYan Zheng 15015d4f98a2SYan Zheng /* 15025d4f98a2SYan Zheng * helper to find first cached inode with inode number >= objectid 15035d4f98a2SYan Zheng * in a subvolume 15045d4f98a2SYan Zheng */ 15055d4f98a2SYan Zheng static struct inode *find_next_inode(struct btrfs_root *root, u64 objectid) 15065d4f98a2SYan Zheng { 15075d4f98a2SYan Zheng struct rb_node *node; 15085d4f98a2SYan Zheng struct rb_node *prev; 15095d4f98a2SYan Zheng struct btrfs_inode *entry; 15105d4f98a2SYan Zheng struct inode *inode; 15115d4f98a2SYan Zheng 15125d4f98a2SYan Zheng spin_lock(&root->inode_lock); 15135d4f98a2SYan Zheng again: 15145d4f98a2SYan Zheng node = root->inode_tree.rb_node; 15155d4f98a2SYan Zheng prev = NULL; 15165d4f98a2SYan Zheng while (node) { 15175d4f98a2SYan Zheng prev = node; 15185d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 15195d4f98a2SYan Zheng 15204a0cc7caSNikolay Borisov if (objectid < btrfs_ino(entry)) 15215d4f98a2SYan Zheng node = node->rb_left; 15224a0cc7caSNikolay Borisov else if (objectid > btrfs_ino(entry)) 15235d4f98a2SYan Zheng node = node->rb_right; 15245d4f98a2SYan Zheng else 15255d4f98a2SYan Zheng break; 15265d4f98a2SYan Zheng } 15275d4f98a2SYan Zheng if (!node) { 15285d4f98a2SYan Zheng while (prev) { 15295d4f98a2SYan Zheng entry = rb_entry(prev, struct btrfs_inode, rb_node); 15304a0cc7caSNikolay Borisov if (objectid <= btrfs_ino(entry)) { 15315d4f98a2SYan Zheng node = prev; 15325d4f98a2SYan Zheng break; 15335d4f98a2SYan Zheng } 15345d4f98a2SYan Zheng prev = rb_next(prev); 15355d4f98a2SYan Zheng } 15365d4f98a2SYan Zheng } 15375d4f98a2SYan Zheng while (node) { 15385d4f98a2SYan Zheng entry = rb_entry(node, struct btrfs_inode, rb_node); 15395d4f98a2SYan Zheng inode = igrab(&entry->vfs_inode); 15405d4f98a2SYan Zheng if (inode) { 15415d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 15425d4f98a2SYan Zheng return inode; 15435d4f98a2SYan Zheng } 15445d4f98a2SYan Zheng 15454a0cc7caSNikolay Borisov objectid = btrfs_ino(entry) + 1; 15465d4f98a2SYan Zheng if (cond_resched_lock(&root->inode_lock)) 15475d4f98a2SYan Zheng goto again; 15485d4f98a2SYan Zheng 15495d4f98a2SYan Zheng node = rb_next(node); 15505d4f98a2SYan Zheng } 15515d4f98a2SYan Zheng spin_unlock(&root->inode_lock); 15525d4f98a2SYan Zheng return NULL; 15535d4f98a2SYan Zheng } 15545d4f98a2SYan Zheng 15555d4f98a2SYan Zheng static int in_block_group(u64 bytenr, 15565d4f98a2SYan Zheng struct btrfs_block_group_cache *block_group) 15575d4f98a2SYan Zheng { 15585d4f98a2SYan Zheng if (bytenr >= block_group->key.objectid && 15595d4f98a2SYan Zheng bytenr < block_group->key.objectid + block_group->key.offset) 15605d4f98a2SYan Zheng return 1; 15615d4f98a2SYan Zheng return 0; 15625d4f98a2SYan Zheng } 15635d4f98a2SYan Zheng 15645d4f98a2SYan Zheng /* 15655d4f98a2SYan Zheng * get new location of data 15665d4f98a2SYan Zheng */ 15675d4f98a2SYan Zheng static int get_new_location(struct inode *reloc_inode, u64 *new_bytenr, 15685d4f98a2SYan Zheng u64 bytenr, u64 num_bytes) 15695d4f98a2SYan Zheng { 15705d4f98a2SYan Zheng struct btrfs_root *root = BTRFS_I(reloc_inode)->root; 15715d4f98a2SYan Zheng struct btrfs_path *path; 15725d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 15735d4f98a2SYan Zheng struct extent_buffer *leaf; 15745d4f98a2SYan Zheng int ret; 15755d4f98a2SYan Zheng 15765d4f98a2SYan Zheng path = btrfs_alloc_path(); 15775d4f98a2SYan Zheng if (!path) 15785d4f98a2SYan Zheng return -ENOMEM; 15795d4f98a2SYan Zheng 15805d4f98a2SYan Zheng bytenr -= BTRFS_I(reloc_inode)->index_cnt; 1581f85b7379SDavid Sterba ret = btrfs_lookup_file_extent(NULL, root, path, 1582f85b7379SDavid Sterba btrfs_ino(BTRFS_I(reloc_inode)), bytenr, 0); 15835d4f98a2SYan Zheng if (ret < 0) 15845d4f98a2SYan Zheng goto out; 15855d4f98a2SYan Zheng if (ret > 0) { 15865d4f98a2SYan Zheng ret = -ENOENT; 15875d4f98a2SYan Zheng goto out; 15885d4f98a2SYan Zheng } 15895d4f98a2SYan Zheng 15905d4f98a2SYan Zheng leaf = path->nodes[0]; 15915d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 15925d4f98a2SYan Zheng struct btrfs_file_extent_item); 15935d4f98a2SYan Zheng 15945d4f98a2SYan Zheng BUG_ON(btrfs_file_extent_offset(leaf, fi) || 15955d4f98a2SYan Zheng btrfs_file_extent_compression(leaf, fi) || 15965d4f98a2SYan Zheng btrfs_file_extent_encryption(leaf, fi) || 15975d4f98a2SYan Zheng btrfs_file_extent_other_encoding(leaf, fi)); 15985d4f98a2SYan Zheng 15995d4f98a2SYan Zheng if (num_bytes != btrfs_file_extent_disk_num_bytes(leaf, fi)) { 160083d4cfd4SJosef Bacik ret = -EINVAL; 16015d4f98a2SYan Zheng goto out; 16025d4f98a2SYan Zheng } 16035d4f98a2SYan Zheng 16045d4f98a2SYan Zheng *new_bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 16055d4f98a2SYan Zheng ret = 0; 16065d4f98a2SYan Zheng out: 16075d4f98a2SYan Zheng btrfs_free_path(path); 16085d4f98a2SYan Zheng return ret; 16095d4f98a2SYan Zheng } 16105d4f98a2SYan Zheng 16115d4f98a2SYan Zheng /* 16125d4f98a2SYan Zheng * update file extent items in the tree leaf to point to 16135d4f98a2SYan Zheng * the new locations. 16145d4f98a2SYan Zheng */ 16153fd0a558SYan, Zheng static noinline_for_stack 16163fd0a558SYan, Zheng int replace_file_extents(struct btrfs_trans_handle *trans, 16175d4f98a2SYan Zheng struct reloc_control *rc, 16185d4f98a2SYan Zheng struct btrfs_root *root, 16193fd0a558SYan, Zheng struct extent_buffer *leaf) 16205d4f98a2SYan Zheng { 16210b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 16225d4f98a2SYan Zheng struct btrfs_key key; 16235d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 16245d4f98a2SYan Zheng struct inode *inode = NULL; 16255d4f98a2SYan Zheng u64 parent; 16265d4f98a2SYan Zheng u64 bytenr; 16273fd0a558SYan, Zheng u64 new_bytenr = 0; 16285d4f98a2SYan Zheng u64 num_bytes; 16295d4f98a2SYan Zheng u64 end; 16305d4f98a2SYan Zheng u32 nritems; 16315d4f98a2SYan Zheng u32 i; 163283d4cfd4SJosef Bacik int ret = 0; 16335d4f98a2SYan Zheng int first = 1; 16345d4f98a2SYan Zheng int dirty = 0; 16355d4f98a2SYan Zheng 16365d4f98a2SYan Zheng if (rc->stage != UPDATE_DATA_PTRS) 16375d4f98a2SYan Zheng return 0; 16385d4f98a2SYan Zheng 16395d4f98a2SYan Zheng /* reloc trees always use full backref */ 16405d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) 16415d4f98a2SYan Zheng parent = leaf->start; 16425d4f98a2SYan Zheng else 16435d4f98a2SYan Zheng parent = 0; 16445d4f98a2SYan Zheng 16455d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 16465d4f98a2SYan Zheng for (i = 0; i < nritems; i++) { 164782fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 164882fa113fSQu Wenruo 16495d4f98a2SYan Zheng cond_resched(); 16505d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, i); 16515d4f98a2SYan Zheng if (key.type != BTRFS_EXTENT_DATA_KEY) 16525d4f98a2SYan Zheng continue; 16535d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, i, struct btrfs_file_extent_item); 16545d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 16555d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 16565d4f98a2SYan Zheng continue; 16575d4f98a2SYan Zheng bytenr = btrfs_file_extent_disk_bytenr(leaf, fi); 16585d4f98a2SYan Zheng num_bytes = btrfs_file_extent_disk_num_bytes(leaf, fi); 16595d4f98a2SYan Zheng if (bytenr == 0) 16605d4f98a2SYan Zheng continue; 16615d4f98a2SYan Zheng if (!in_block_group(bytenr, rc->block_group)) 16625d4f98a2SYan Zheng continue; 16635d4f98a2SYan Zheng 16645d4f98a2SYan Zheng /* 16655d4f98a2SYan Zheng * if we are modifying block in fs tree, wait for readpage 16665d4f98a2SYan Zheng * to complete and drop the extent cache 16675d4f98a2SYan Zheng */ 16685d4f98a2SYan Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 16695d4f98a2SYan Zheng if (first) { 16705d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 16715d4f98a2SYan Zheng first = 0; 16724a0cc7caSNikolay Borisov } else if (inode && btrfs_ino(BTRFS_I(inode)) < key.objectid) { 16733fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 16745d4f98a2SYan Zheng inode = find_next_inode(root, key.objectid); 16755d4f98a2SYan Zheng } 16764a0cc7caSNikolay Borisov if (inode && btrfs_ino(BTRFS_I(inode)) == key.objectid) { 16775d4f98a2SYan Zheng end = key.offset + 16785d4f98a2SYan Zheng btrfs_file_extent_num_bytes(leaf, fi); 16795d4f98a2SYan Zheng WARN_ON(!IS_ALIGNED(key.offset, 16800b246afaSJeff Mahoney fs_info->sectorsize)); 16810b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 16825d4f98a2SYan Zheng end--; 16835d4f98a2SYan Zheng ret = try_lock_extent(&BTRFS_I(inode)->io_tree, 1684d0082371SJeff Mahoney key.offset, end); 16855d4f98a2SYan Zheng if (!ret) 16865d4f98a2SYan Zheng continue; 16875d4f98a2SYan Zheng 1688dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), 1689dcdbc059SNikolay Borisov key.offset, end, 1); 16905d4f98a2SYan Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 1691d0082371SJeff Mahoney key.offset, end); 16925d4f98a2SYan Zheng } 16935d4f98a2SYan Zheng } 16945d4f98a2SYan Zheng 16955d4f98a2SYan Zheng ret = get_new_location(rc->data_inode, &new_bytenr, 16965d4f98a2SYan Zheng bytenr, num_bytes); 169783d4cfd4SJosef Bacik if (ret) { 169883d4cfd4SJosef Bacik /* 169983d4cfd4SJosef Bacik * Don't have to abort since we've not changed anything 170083d4cfd4SJosef Bacik * in the file extent yet. 170183d4cfd4SJosef Bacik */ 170283d4cfd4SJosef Bacik break; 17033fd0a558SYan, Zheng } 17045d4f98a2SYan Zheng 17055d4f98a2SYan Zheng btrfs_set_file_extent_disk_bytenr(leaf, fi, new_bytenr); 17065d4f98a2SYan Zheng dirty = 1; 17075d4f98a2SYan Zheng 17085d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 170982fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 171082fa113fSQu Wenruo num_bytes, parent); 171182fa113fSQu Wenruo ref.real_root = root->root_key.objectid; 171282fa113fSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1713b06c4bf5SFilipe Manana key.objectid, key.offset); 171482fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 171583d4cfd4SJosef Bacik if (ret) { 171666642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 171783d4cfd4SJosef Bacik break; 171883d4cfd4SJosef Bacik } 17195d4f98a2SYan Zheng 1720ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, bytenr, 1721ffd4bb2aSQu Wenruo num_bytes, parent); 1722ffd4bb2aSQu Wenruo ref.real_root = root->root_key.objectid; 1723ffd4bb2aSQu Wenruo btrfs_init_data_ref(&ref, btrfs_header_owner(leaf), 1724b06c4bf5SFilipe Manana key.objectid, key.offset); 1725ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 172683d4cfd4SJosef Bacik if (ret) { 172766642832SJeff Mahoney btrfs_abort_transaction(trans, ret); 172883d4cfd4SJosef Bacik break; 172983d4cfd4SJosef Bacik } 17305d4f98a2SYan Zheng } 17315d4f98a2SYan Zheng if (dirty) 17325d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 17333fd0a558SYan, Zheng if (inode) 17343fd0a558SYan, Zheng btrfs_add_delayed_iput(inode); 173583d4cfd4SJosef Bacik return ret; 17365d4f98a2SYan Zheng } 17375d4f98a2SYan Zheng 17385d4f98a2SYan Zheng static noinline_for_stack 17395d4f98a2SYan Zheng int memcmp_node_keys(struct extent_buffer *eb, int slot, 17405d4f98a2SYan Zheng struct btrfs_path *path, int level) 17415d4f98a2SYan Zheng { 17425d4f98a2SYan Zheng struct btrfs_disk_key key1; 17435d4f98a2SYan Zheng struct btrfs_disk_key key2; 17445d4f98a2SYan Zheng btrfs_node_key(eb, &key1, slot); 17455d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &key2, path->slots[level]); 17465d4f98a2SYan Zheng return memcmp(&key1, &key2, sizeof(key1)); 17475d4f98a2SYan Zheng } 17485d4f98a2SYan Zheng 17495d4f98a2SYan Zheng /* 17505d4f98a2SYan Zheng * try to replace tree blocks in fs tree with the new blocks 17515d4f98a2SYan Zheng * in reloc tree. tree blocks haven't been modified since the 17525d4f98a2SYan Zheng * reloc tree was create can be replaced. 17535d4f98a2SYan Zheng * 17545d4f98a2SYan Zheng * if a block was replaced, level of the block + 1 is returned. 17555d4f98a2SYan Zheng * if no block got replaced, 0 is returned. if there are other 17565d4f98a2SYan Zheng * errors, a negative error number is returned. 17575d4f98a2SYan Zheng */ 17583fd0a558SYan, Zheng static noinline_for_stack 17593d0174f7SQu Wenruo int replace_path(struct btrfs_trans_handle *trans, struct reloc_control *rc, 17605d4f98a2SYan Zheng struct btrfs_root *dest, struct btrfs_root *src, 17615d4f98a2SYan Zheng struct btrfs_path *path, struct btrfs_key *next_key, 17625d4f98a2SYan Zheng int lowest_level, int max_level) 17635d4f98a2SYan Zheng { 17640b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = dest->fs_info; 17655d4f98a2SYan Zheng struct extent_buffer *eb; 17665d4f98a2SYan Zheng struct extent_buffer *parent; 176782fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 17685d4f98a2SYan Zheng struct btrfs_key key; 17695d4f98a2SYan Zheng u64 old_bytenr; 17705d4f98a2SYan Zheng u64 new_bytenr; 17715d4f98a2SYan Zheng u64 old_ptr_gen; 17725d4f98a2SYan Zheng u64 new_ptr_gen; 17735d4f98a2SYan Zheng u64 last_snapshot; 17745d4f98a2SYan Zheng u32 blocksize; 17753fd0a558SYan, Zheng int cow = 0; 17765d4f98a2SYan Zheng int level; 17775d4f98a2SYan Zheng int ret; 17785d4f98a2SYan Zheng int slot; 17795d4f98a2SYan Zheng 17805d4f98a2SYan Zheng BUG_ON(src->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 17815d4f98a2SYan Zheng BUG_ON(dest->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID); 17825d4f98a2SYan Zheng 17835d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&src->root_item); 17843fd0a558SYan, Zheng again: 17855d4f98a2SYan Zheng slot = path->slots[lowest_level]; 17865d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[lowest_level], &key, slot); 17875d4f98a2SYan Zheng 17885d4f98a2SYan Zheng eb = btrfs_lock_root_node(dest); 17898bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 17905d4f98a2SYan Zheng level = btrfs_header_level(eb); 17915d4f98a2SYan Zheng 17925d4f98a2SYan Zheng if (level < lowest_level) { 17935d4f98a2SYan Zheng btrfs_tree_unlock(eb); 17945d4f98a2SYan Zheng free_extent_buffer(eb); 17955d4f98a2SYan Zheng return 0; 17965d4f98a2SYan Zheng } 17975d4f98a2SYan Zheng 17983fd0a558SYan, Zheng if (cow) { 17995d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, NULL, 0, &eb); 18005d4f98a2SYan Zheng BUG_ON(ret); 18013fd0a558SYan, Zheng } 18028bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 18035d4f98a2SYan Zheng 18045d4f98a2SYan Zheng if (next_key) { 18055d4f98a2SYan Zheng next_key->objectid = (u64)-1; 18065d4f98a2SYan Zheng next_key->type = (u8)-1; 18075d4f98a2SYan Zheng next_key->offset = (u64)-1; 18085d4f98a2SYan Zheng } 18095d4f98a2SYan Zheng 18105d4f98a2SYan Zheng parent = eb; 18115d4f98a2SYan Zheng while (1) { 1812581c1760SQu Wenruo struct btrfs_key first_key; 1813581c1760SQu Wenruo 18145d4f98a2SYan Zheng level = btrfs_header_level(parent); 18155d4f98a2SYan Zheng BUG_ON(level < lowest_level); 18165d4f98a2SYan Zheng 18175d4f98a2SYan Zheng ret = btrfs_bin_search(parent, &key, level, &slot); 1818cbca7d59SFilipe Manana if (ret < 0) 1819cbca7d59SFilipe Manana break; 18205d4f98a2SYan Zheng if (ret && slot > 0) 18215d4f98a2SYan Zheng slot--; 18225d4f98a2SYan Zheng 18235d4f98a2SYan Zheng if (next_key && slot + 1 < btrfs_header_nritems(parent)) 18245d4f98a2SYan Zheng btrfs_node_key_to_cpu(parent, next_key, slot + 1); 18255d4f98a2SYan Zheng 18265d4f98a2SYan Zheng old_bytenr = btrfs_node_blockptr(parent, slot); 18270b246afaSJeff Mahoney blocksize = fs_info->nodesize; 18285d4f98a2SYan Zheng old_ptr_gen = btrfs_node_ptr_generation(parent, slot); 182917515f1bSQu Wenruo btrfs_node_key_to_cpu(parent, &first_key, slot); 18305d4f98a2SYan Zheng 18315d4f98a2SYan Zheng if (level <= max_level) { 18325d4f98a2SYan Zheng eb = path->nodes[level]; 18335d4f98a2SYan Zheng new_bytenr = btrfs_node_blockptr(eb, 18345d4f98a2SYan Zheng path->slots[level]); 18355d4f98a2SYan Zheng new_ptr_gen = btrfs_node_ptr_generation(eb, 18365d4f98a2SYan Zheng path->slots[level]); 18375d4f98a2SYan Zheng } else { 18385d4f98a2SYan Zheng new_bytenr = 0; 18395d4f98a2SYan Zheng new_ptr_gen = 0; 18405d4f98a2SYan Zheng } 18415d4f98a2SYan Zheng 1842fae7f21cSDulshani Gunawardhana if (WARN_ON(new_bytenr > 0 && new_bytenr == old_bytenr)) { 18435d4f98a2SYan Zheng ret = level; 18445d4f98a2SYan Zheng break; 18455d4f98a2SYan Zheng } 18465d4f98a2SYan Zheng 18475d4f98a2SYan Zheng if (new_bytenr == 0 || old_ptr_gen > last_snapshot || 18485d4f98a2SYan Zheng memcmp_node_keys(parent, slot, path, level)) { 18493fd0a558SYan, Zheng if (level <= lowest_level) { 18505d4f98a2SYan Zheng ret = 0; 18515d4f98a2SYan Zheng break; 18525d4f98a2SYan Zheng } 18535d4f98a2SYan Zheng 1854581c1760SQu Wenruo eb = read_tree_block(fs_info, old_bytenr, old_ptr_gen, 1855581c1760SQu Wenruo level - 1, &first_key); 185664c043deSLiu Bo if (IS_ERR(eb)) { 185764c043deSLiu Bo ret = PTR_ERR(eb); 1858264813acSLiu Bo break; 185964c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 186064c043deSLiu Bo ret = -EIO; 1861416bc658SJosef Bacik free_extent_buffer(eb); 1862379cde74SStefan Behrens break; 1863416bc658SJosef Bacik } 18645d4f98a2SYan Zheng btrfs_tree_lock(eb); 18653fd0a558SYan, Zheng if (cow) { 18665d4f98a2SYan Zheng ret = btrfs_cow_block(trans, dest, eb, parent, 18675d4f98a2SYan Zheng slot, &eb); 18685d4f98a2SYan Zheng BUG_ON(ret); 18695d4f98a2SYan Zheng } 18708bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 18715d4f98a2SYan Zheng 18725d4f98a2SYan Zheng btrfs_tree_unlock(parent); 18735d4f98a2SYan Zheng free_extent_buffer(parent); 18745d4f98a2SYan Zheng 18755d4f98a2SYan Zheng parent = eb; 18765d4f98a2SYan Zheng continue; 18775d4f98a2SYan Zheng } 18785d4f98a2SYan Zheng 18793fd0a558SYan, Zheng if (!cow) { 18803fd0a558SYan, Zheng btrfs_tree_unlock(parent); 18813fd0a558SYan, Zheng free_extent_buffer(parent); 18823fd0a558SYan, Zheng cow = 1; 18833fd0a558SYan, Zheng goto again; 18843fd0a558SYan, Zheng } 18853fd0a558SYan, Zheng 18865d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 18875d4f98a2SYan Zheng path->slots[level]); 1888b3b4aa74SDavid Sterba btrfs_release_path(path); 18895d4f98a2SYan Zheng 18905d4f98a2SYan Zheng path->lowest_level = level; 18915d4f98a2SYan Zheng ret = btrfs_search_slot(trans, src, &key, path, 0, 1); 18925d4f98a2SYan Zheng path->lowest_level = 0; 18935d4f98a2SYan Zheng BUG_ON(ret); 18945d4f98a2SYan Zheng 18955d4f98a2SYan Zheng /* 1896824d8dffSQu Wenruo * Info qgroup to trace both subtrees. 1897824d8dffSQu Wenruo * 1898824d8dffSQu Wenruo * We must trace both trees. 1899824d8dffSQu Wenruo * 1) Tree reloc subtree 1900824d8dffSQu Wenruo * If not traced, we will leak data numbers 1901824d8dffSQu Wenruo * 2) Fs subtree 1902824d8dffSQu Wenruo * If not traced, we will double count old data 1903f616f5cdSQu Wenruo * 1904f616f5cdSQu Wenruo * We don't scan the subtree right now, but only record 1905f616f5cdSQu Wenruo * the swapped tree blocks. 1906f616f5cdSQu Wenruo * The real subtree rescan is delayed until we have new 1907f616f5cdSQu Wenruo * CoW on the subtree root node before transaction commit. 1908824d8dffSQu Wenruo */ 1909370a11b8SQu Wenruo ret = btrfs_qgroup_add_swapped_blocks(trans, dest, 1910370a11b8SQu Wenruo rc->block_group, parent, slot, 1911370a11b8SQu Wenruo path->nodes[level], path->slots[level], 1912370a11b8SQu Wenruo last_snapshot); 1913370a11b8SQu Wenruo if (ret < 0) 1914370a11b8SQu Wenruo break; 1915824d8dffSQu Wenruo /* 19165d4f98a2SYan Zheng * swap blocks in fs tree and reloc tree. 19175d4f98a2SYan Zheng */ 19185d4f98a2SYan Zheng btrfs_set_node_blockptr(parent, slot, new_bytenr); 19195d4f98a2SYan Zheng btrfs_set_node_ptr_generation(parent, slot, new_ptr_gen); 19205d4f98a2SYan Zheng btrfs_mark_buffer_dirty(parent); 19215d4f98a2SYan Zheng 19225d4f98a2SYan Zheng btrfs_set_node_blockptr(path->nodes[level], 19235d4f98a2SYan Zheng path->slots[level], old_bytenr); 19245d4f98a2SYan Zheng btrfs_set_node_ptr_generation(path->nodes[level], 19255d4f98a2SYan Zheng path->slots[level], old_ptr_gen); 19265d4f98a2SYan Zheng btrfs_mark_buffer_dirty(path->nodes[level]); 19275d4f98a2SYan Zheng 192882fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, old_bytenr, 192982fa113fSQu Wenruo blocksize, path->nodes[level]->start); 193082fa113fSQu Wenruo ref.skip_qgroup = true; 193182fa113fSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid); 193282fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 19335d4f98a2SYan Zheng BUG_ON(ret); 193482fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, new_bytenr, 193582fa113fSQu Wenruo blocksize, 0); 193682fa113fSQu Wenruo ref.skip_qgroup = true; 193782fa113fSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid); 193882fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 19395d4f98a2SYan Zheng BUG_ON(ret); 19405d4f98a2SYan Zheng 1941ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, new_bytenr, 1942ffd4bb2aSQu Wenruo blocksize, path->nodes[level]->start); 1943ffd4bb2aSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, src->root_key.objectid); 1944ffd4bb2aSQu Wenruo ref.skip_qgroup = true; 1945ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 19465d4f98a2SYan Zheng BUG_ON(ret); 19475d4f98a2SYan Zheng 1948ffd4bb2aSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_DROP_DELAYED_REF, old_bytenr, 1949ffd4bb2aSQu Wenruo blocksize, 0); 1950ffd4bb2aSQu Wenruo btrfs_init_tree_ref(&ref, level - 1, dest->root_key.objectid); 1951ffd4bb2aSQu Wenruo ref.skip_qgroup = true; 1952ffd4bb2aSQu Wenruo ret = btrfs_free_extent(trans, &ref); 19535d4f98a2SYan Zheng BUG_ON(ret); 19545d4f98a2SYan Zheng 19555d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 19565d4f98a2SYan Zheng 19575d4f98a2SYan Zheng ret = level; 19585d4f98a2SYan Zheng break; 19595d4f98a2SYan Zheng } 19605d4f98a2SYan Zheng btrfs_tree_unlock(parent); 19615d4f98a2SYan Zheng free_extent_buffer(parent); 19625d4f98a2SYan Zheng return ret; 19635d4f98a2SYan Zheng } 19645d4f98a2SYan Zheng 19655d4f98a2SYan Zheng /* 19665d4f98a2SYan Zheng * helper to find next relocated block in reloc tree 19675d4f98a2SYan Zheng */ 19685d4f98a2SYan Zheng static noinline_for_stack 19695d4f98a2SYan Zheng int walk_up_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 19705d4f98a2SYan Zheng int *level) 19715d4f98a2SYan Zheng { 19725d4f98a2SYan Zheng struct extent_buffer *eb; 19735d4f98a2SYan Zheng int i; 19745d4f98a2SYan Zheng u64 last_snapshot; 19755d4f98a2SYan Zheng u32 nritems; 19765d4f98a2SYan Zheng 19775d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 19785d4f98a2SYan Zheng 19795d4f98a2SYan Zheng for (i = 0; i < *level; i++) { 19805d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 19815d4f98a2SYan Zheng path->nodes[i] = NULL; 19825d4f98a2SYan Zheng } 19835d4f98a2SYan Zheng 19845d4f98a2SYan Zheng for (i = *level; i < BTRFS_MAX_LEVEL && path->nodes[i]; i++) { 19855d4f98a2SYan Zheng eb = path->nodes[i]; 19865d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 19875d4f98a2SYan Zheng while (path->slots[i] + 1 < nritems) { 19885d4f98a2SYan Zheng path->slots[i]++; 19895d4f98a2SYan Zheng if (btrfs_node_ptr_generation(eb, path->slots[i]) <= 19905d4f98a2SYan Zheng last_snapshot) 19915d4f98a2SYan Zheng continue; 19925d4f98a2SYan Zheng 19935d4f98a2SYan Zheng *level = i; 19945d4f98a2SYan Zheng return 0; 19955d4f98a2SYan Zheng } 19965d4f98a2SYan Zheng free_extent_buffer(path->nodes[i]); 19975d4f98a2SYan Zheng path->nodes[i] = NULL; 19985d4f98a2SYan Zheng } 19995d4f98a2SYan Zheng return 1; 20005d4f98a2SYan Zheng } 20015d4f98a2SYan Zheng 20025d4f98a2SYan Zheng /* 20035d4f98a2SYan Zheng * walk down reloc tree to find relocated block of lowest level 20045d4f98a2SYan Zheng */ 20055d4f98a2SYan Zheng static noinline_for_stack 20065d4f98a2SYan Zheng int walk_down_reloc_tree(struct btrfs_root *root, struct btrfs_path *path, 20075d4f98a2SYan Zheng int *level) 20085d4f98a2SYan Zheng { 20092ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 20105d4f98a2SYan Zheng struct extent_buffer *eb = NULL; 20115d4f98a2SYan Zheng int i; 20125d4f98a2SYan Zheng u64 bytenr; 20135d4f98a2SYan Zheng u64 ptr_gen = 0; 20145d4f98a2SYan Zheng u64 last_snapshot; 20155d4f98a2SYan Zheng u32 nritems; 20165d4f98a2SYan Zheng 20175d4f98a2SYan Zheng last_snapshot = btrfs_root_last_snapshot(&root->root_item); 20185d4f98a2SYan Zheng 20195d4f98a2SYan Zheng for (i = *level; i > 0; i--) { 2020581c1760SQu Wenruo struct btrfs_key first_key; 2021581c1760SQu Wenruo 20225d4f98a2SYan Zheng eb = path->nodes[i]; 20235d4f98a2SYan Zheng nritems = btrfs_header_nritems(eb); 20245d4f98a2SYan Zheng while (path->slots[i] < nritems) { 20255d4f98a2SYan Zheng ptr_gen = btrfs_node_ptr_generation(eb, path->slots[i]); 20265d4f98a2SYan Zheng if (ptr_gen > last_snapshot) 20275d4f98a2SYan Zheng break; 20285d4f98a2SYan Zheng path->slots[i]++; 20295d4f98a2SYan Zheng } 20305d4f98a2SYan Zheng if (path->slots[i] >= nritems) { 20315d4f98a2SYan Zheng if (i == *level) 20325d4f98a2SYan Zheng break; 20335d4f98a2SYan Zheng *level = i + 1; 20345d4f98a2SYan Zheng return 0; 20355d4f98a2SYan Zheng } 20365d4f98a2SYan Zheng if (i == 1) { 20375d4f98a2SYan Zheng *level = i; 20385d4f98a2SYan Zheng return 0; 20395d4f98a2SYan Zheng } 20405d4f98a2SYan Zheng 20415d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(eb, path->slots[i]); 2042581c1760SQu Wenruo btrfs_node_key_to_cpu(eb, &first_key, path->slots[i]); 2043581c1760SQu Wenruo eb = read_tree_block(fs_info, bytenr, ptr_gen, i - 1, 2044581c1760SQu Wenruo &first_key); 204564c043deSLiu Bo if (IS_ERR(eb)) { 204664c043deSLiu Bo return PTR_ERR(eb); 204764c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2048416bc658SJosef Bacik free_extent_buffer(eb); 2049416bc658SJosef Bacik return -EIO; 2050416bc658SJosef Bacik } 20515d4f98a2SYan Zheng BUG_ON(btrfs_header_level(eb) != i - 1); 20525d4f98a2SYan Zheng path->nodes[i - 1] = eb; 20535d4f98a2SYan Zheng path->slots[i - 1] = 0; 20545d4f98a2SYan Zheng } 20555d4f98a2SYan Zheng return 1; 20565d4f98a2SYan Zheng } 20575d4f98a2SYan Zheng 20585d4f98a2SYan Zheng /* 20595d4f98a2SYan Zheng * invalidate extent cache for file extents whose key in range of 20605d4f98a2SYan Zheng * [min_key, max_key) 20615d4f98a2SYan Zheng */ 20625d4f98a2SYan Zheng static int invalidate_extent_cache(struct btrfs_root *root, 20635d4f98a2SYan Zheng struct btrfs_key *min_key, 20645d4f98a2SYan Zheng struct btrfs_key *max_key) 20655d4f98a2SYan Zheng { 20660b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 20675d4f98a2SYan Zheng struct inode *inode = NULL; 20685d4f98a2SYan Zheng u64 objectid; 20695d4f98a2SYan Zheng u64 start, end; 207033345d01SLi Zefan u64 ino; 20715d4f98a2SYan Zheng 20725d4f98a2SYan Zheng objectid = min_key->objectid; 20735d4f98a2SYan Zheng while (1) { 20745d4f98a2SYan Zheng cond_resched(); 20755d4f98a2SYan Zheng iput(inode); 20765d4f98a2SYan Zheng 20775d4f98a2SYan Zheng if (objectid > max_key->objectid) 20785d4f98a2SYan Zheng break; 20795d4f98a2SYan Zheng 20805d4f98a2SYan Zheng inode = find_next_inode(root, objectid); 20815d4f98a2SYan Zheng if (!inode) 20825d4f98a2SYan Zheng break; 20834a0cc7caSNikolay Borisov ino = btrfs_ino(BTRFS_I(inode)); 20845d4f98a2SYan Zheng 208533345d01SLi Zefan if (ino > max_key->objectid) { 20865d4f98a2SYan Zheng iput(inode); 20875d4f98a2SYan Zheng break; 20885d4f98a2SYan Zheng } 20895d4f98a2SYan Zheng 209033345d01SLi Zefan objectid = ino + 1; 20915d4f98a2SYan Zheng if (!S_ISREG(inode->i_mode)) 20925d4f98a2SYan Zheng continue; 20935d4f98a2SYan Zheng 209433345d01SLi Zefan if (unlikely(min_key->objectid == ino)) { 20955d4f98a2SYan Zheng if (min_key->type > BTRFS_EXTENT_DATA_KEY) 20965d4f98a2SYan Zheng continue; 20975d4f98a2SYan Zheng if (min_key->type < BTRFS_EXTENT_DATA_KEY) 20985d4f98a2SYan Zheng start = 0; 20995d4f98a2SYan Zheng else { 21005d4f98a2SYan Zheng start = min_key->offset; 21010b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(start, fs_info->sectorsize)); 21025d4f98a2SYan Zheng } 21035d4f98a2SYan Zheng } else { 21045d4f98a2SYan Zheng start = 0; 21055d4f98a2SYan Zheng } 21065d4f98a2SYan Zheng 210733345d01SLi Zefan if (unlikely(max_key->objectid == ino)) { 21085d4f98a2SYan Zheng if (max_key->type < BTRFS_EXTENT_DATA_KEY) 21095d4f98a2SYan Zheng continue; 21105d4f98a2SYan Zheng if (max_key->type > BTRFS_EXTENT_DATA_KEY) { 21115d4f98a2SYan Zheng end = (u64)-1; 21125d4f98a2SYan Zheng } else { 21135d4f98a2SYan Zheng if (max_key->offset == 0) 21145d4f98a2SYan Zheng continue; 21155d4f98a2SYan Zheng end = max_key->offset; 21160b246afaSJeff Mahoney WARN_ON(!IS_ALIGNED(end, fs_info->sectorsize)); 21175d4f98a2SYan Zheng end--; 21185d4f98a2SYan Zheng } 21195d4f98a2SYan Zheng } else { 21205d4f98a2SYan Zheng end = (u64)-1; 21215d4f98a2SYan Zheng } 21225d4f98a2SYan Zheng 21235d4f98a2SYan Zheng /* the lock_extent waits for readpage to complete */ 2124d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 2125dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 1); 2126d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 21275d4f98a2SYan Zheng } 21285d4f98a2SYan Zheng return 0; 21295d4f98a2SYan Zheng } 21305d4f98a2SYan Zheng 21315d4f98a2SYan Zheng static int find_next_key(struct btrfs_path *path, int level, 21325d4f98a2SYan Zheng struct btrfs_key *key) 21335d4f98a2SYan Zheng 21345d4f98a2SYan Zheng { 21355d4f98a2SYan Zheng while (level < BTRFS_MAX_LEVEL) { 21365d4f98a2SYan Zheng if (!path->nodes[level]) 21375d4f98a2SYan Zheng break; 21385d4f98a2SYan Zheng if (path->slots[level] + 1 < 21395d4f98a2SYan Zheng btrfs_header_nritems(path->nodes[level])) { 21405d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], key, 21415d4f98a2SYan Zheng path->slots[level] + 1); 21425d4f98a2SYan Zheng return 0; 21435d4f98a2SYan Zheng } 21445d4f98a2SYan Zheng level++; 21455d4f98a2SYan Zheng } 21465d4f98a2SYan Zheng return 1; 21475d4f98a2SYan Zheng } 21485d4f98a2SYan Zheng 21495d4f98a2SYan Zheng /* 2150d2311e69SQu Wenruo * Insert current subvolume into reloc_control::dirty_subvol_roots 2151d2311e69SQu Wenruo */ 2152d2311e69SQu Wenruo static void insert_dirty_subvol(struct btrfs_trans_handle *trans, 2153d2311e69SQu Wenruo struct reloc_control *rc, 2154d2311e69SQu Wenruo struct btrfs_root *root) 2155d2311e69SQu Wenruo { 2156d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 2157d2311e69SQu Wenruo struct btrfs_root_item *reloc_root_item; 2158d2311e69SQu Wenruo 2159d2311e69SQu Wenruo /* @root must be a subvolume tree root with a valid reloc tree */ 2160d2311e69SQu Wenruo ASSERT(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID); 2161d2311e69SQu Wenruo ASSERT(reloc_root); 2162d2311e69SQu Wenruo 2163d2311e69SQu Wenruo reloc_root_item = &reloc_root->root_item; 2164d2311e69SQu Wenruo memset(&reloc_root_item->drop_progress, 0, 2165d2311e69SQu Wenruo sizeof(reloc_root_item->drop_progress)); 2166d2311e69SQu Wenruo reloc_root_item->drop_level = 0; 2167d2311e69SQu Wenruo btrfs_set_root_refs(reloc_root_item, 0); 2168d2311e69SQu Wenruo btrfs_update_reloc_root(trans, root); 2169d2311e69SQu Wenruo 2170d2311e69SQu Wenruo if (list_empty(&root->reloc_dirty_list)) { 2171d2311e69SQu Wenruo btrfs_grab_fs_root(root); 2172d2311e69SQu Wenruo list_add_tail(&root->reloc_dirty_list, &rc->dirty_subvol_roots); 2173d2311e69SQu Wenruo } 2174d2311e69SQu Wenruo } 2175d2311e69SQu Wenruo 2176d2311e69SQu Wenruo static int clean_dirty_subvols(struct reloc_control *rc) 2177d2311e69SQu Wenruo { 2178d2311e69SQu Wenruo struct btrfs_root *root; 2179d2311e69SQu Wenruo struct btrfs_root *next; 2180d2311e69SQu Wenruo int ret = 0; 218130d40577SQu Wenruo int ret2; 2182d2311e69SQu Wenruo 2183d2311e69SQu Wenruo list_for_each_entry_safe(root, next, &rc->dirty_subvol_roots, 2184d2311e69SQu Wenruo reloc_dirty_list) { 218530d40577SQu Wenruo if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) { 218630d40577SQu Wenruo /* Merged subvolume, cleanup its reloc root */ 2187d2311e69SQu Wenruo struct btrfs_root *reloc_root = root->reloc_root; 2188d2311e69SQu Wenruo 2189d2311e69SQu Wenruo clear_bit(BTRFS_ROOT_DEAD_RELOC_TREE, &root->state); 2190d2311e69SQu Wenruo list_del_init(&root->reloc_dirty_list); 2191d2311e69SQu Wenruo root->reloc_root = NULL; 2192d2311e69SQu Wenruo if (reloc_root) { 2193d2311e69SQu Wenruo 2194d2311e69SQu Wenruo ret2 = btrfs_drop_snapshot(reloc_root, NULL, 0, 1); 2195d2311e69SQu Wenruo if (ret2 < 0 && !ret) 2196d2311e69SQu Wenruo ret = ret2; 2197d2311e69SQu Wenruo } 2198d2311e69SQu Wenruo btrfs_put_fs_root(root); 219930d40577SQu Wenruo } else { 220030d40577SQu Wenruo /* Orphan reloc tree, just clean it up */ 220130d40577SQu Wenruo ret2 = btrfs_drop_snapshot(root, NULL, 0, 1); 220230d40577SQu Wenruo if (ret2 < 0 && !ret) 220330d40577SQu Wenruo ret = ret2; 220430d40577SQu Wenruo } 2205d2311e69SQu Wenruo } 2206d2311e69SQu Wenruo return ret; 2207d2311e69SQu Wenruo } 2208d2311e69SQu Wenruo 2209d2311e69SQu Wenruo /* 22105d4f98a2SYan Zheng * merge the relocated tree blocks in reloc tree with corresponding 22115d4f98a2SYan Zheng * fs tree. 22125d4f98a2SYan Zheng */ 22135d4f98a2SYan Zheng static noinline_for_stack int merge_reloc_root(struct reloc_control *rc, 22145d4f98a2SYan Zheng struct btrfs_root *root) 22155d4f98a2SYan Zheng { 22160b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 22175d4f98a2SYan Zheng struct btrfs_key key; 22185d4f98a2SYan Zheng struct btrfs_key next_key; 22199e6a0c52SJosef Bacik struct btrfs_trans_handle *trans = NULL; 22205d4f98a2SYan Zheng struct btrfs_root *reloc_root; 22215d4f98a2SYan Zheng struct btrfs_root_item *root_item; 22225d4f98a2SYan Zheng struct btrfs_path *path; 22233fd0a558SYan, Zheng struct extent_buffer *leaf; 22245d4f98a2SYan Zheng int level; 22255d4f98a2SYan Zheng int max_level; 22265d4f98a2SYan Zheng int replaced = 0; 22275d4f98a2SYan Zheng int ret; 22285d4f98a2SYan Zheng int err = 0; 22293fd0a558SYan, Zheng u32 min_reserved; 22305d4f98a2SYan Zheng 22315d4f98a2SYan Zheng path = btrfs_alloc_path(); 22325d4f98a2SYan Zheng if (!path) 22335d4f98a2SYan Zheng return -ENOMEM; 2234e4058b54SDavid Sterba path->reada = READA_FORWARD; 22355d4f98a2SYan Zheng 22365d4f98a2SYan Zheng reloc_root = root->reloc_root; 22375d4f98a2SYan Zheng root_item = &reloc_root->root_item; 22385d4f98a2SYan Zheng 22395d4f98a2SYan Zheng if (btrfs_disk_key_objectid(&root_item->drop_progress) == 0) { 22405d4f98a2SYan Zheng level = btrfs_root_level(root_item); 22415d4f98a2SYan Zheng extent_buffer_get(reloc_root->node); 22425d4f98a2SYan Zheng path->nodes[level] = reloc_root->node; 22435d4f98a2SYan Zheng path->slots[level] = 0; 22445d4f98a2SYan Zheng } else { 22455d4f98a2SYan Zheng btrfs_disk_key_to_cpu(&key, &root_item->drop_progress); 22465d4f98a2SYan Zheng 22475d4f98a2SYan Zheng level = root_item->drop_level; 22485d4f98a2SYan Zheng BUG_ON(level == 0); 22495d4f98a2SYan Zheng path->lowest_level = level; 22505d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, reloc_root, &key, path, 0, 0); 225133c66f43SYan Zheng path->lowest_level = 0; 22525d4f98a2SYan Zheng if (ret < 0) { 22535d4f98a2SYan Zheng btrfs_free_path(path); 22545d4f98a2SYan Zheng return ret; 22555d4f98a2SYan Zheng } 22565d4f98a2SYan Zheng 22575d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &next_key, 22585d4f98a2SYan Zheng path->slots[level]); 22595d4f98a2SYan Zheng WARN_ON(memcmp(&key, &next_key, sizeof(key))); 22605d4f98a2SYan Zheng 22615d4f98a2SYan Zheng btrfs_unlock_up_safe(path, 0); 22625d4f98a2SYan Zheng } 22635d4f98a2SYan Zheng 22640b246afaSJeff Mahoney min_reserved = fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 22655d4f98a2SYan Zheng memset(&next_key, 0, sizeof(next_key)); 22665d4f98a2SYan Zheng 22675d4f98a2SYan Zheng while (1) { 226808e007d2SMiao Xie ret = btrfs_block_rsv_refill(root, rc->block_rsv, min_reserved, 226908e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 22703fd0a558SYan, Zheng if (ret) { 22719e6a0c52SJosef Bacik err = ret; 22729e6a0c52SJosef Bacik goto out; 22733fd0a558SYan, Zheng } 22749e6a0c52SJosef Bacik trans = btrfs_start_transaction(root, 0); 22759e6a0c52SJosef Bacik if (IS_ERR(trans)) { 22769e6a0c52SJosef Bacik err = PTR_ERR(trans); 22779e6a0c52SJosef Bacik trans = NULL; 22789e6a0c52SJosef Bacik goto out; 22799e6a0c52SJosef Bacik } 22809e6a0c52SJosef Bacik trans->block_rsv = rc->block_rsv; 22813fd0a558SYan, Zheng 22823fd0a558SYan, Zheng replaced = 0; 22835d4f98a2SYan Zheng max_level = level; 22845d4f98a2SYan Zheng 22855d4f98a2SYan Zheng ret = walk_down_reloc_tree(reloc_root, path, &level); 22865d4f98a2SYan Zheng if (ret < 0) { 22875d4f98a2SYan Zheng err = ret; 22885d4f98a2SYan Zheng goto out; 22895d4f98a2SYan Zheng } 22905d4f98a2SYan Zheng if (ret > 0) 22915d4f98a2SYan Zheng break; 22925d4f98a2SYan Zheng 22935d4f98a2SYan Zheng if (!find_next_key(path, level, &key) && 22945d4f98a2SYan Zheng btrfs_comp_cpu_keys(&next_key, &key) >= 0) { 22955d4f98a2SYan Zheng ret = 0; 22965d4f98a2SYan Zheng } else { 22973d0174f7SQu Wenruo ret = replace_path(trans, rc, root, reloc_root, path, 22983fd0a558SYan, Zheng &next_key, level, max_level); 22995d4f98a2SYan Zheng } 23005d4f98a2SYan Zheng if (ret < 0) { 23015d4f98a2SYan Zheng err = ret; 23025d4f98a2SYan Zheng goto out; 23035d4f98a2SYan Zheng } 23045d4f98a2SYan Zheng 23055d4f98a2SYan Zheng if (ret > 0) { 23065d4f98a2SYan Zheng level = ret; 23075d4f98a2SYan Zheng btrfs_node_key_to_cpu(path->nodes[level], &key, 23085d4f98a2SYan Zheng path->slots[level]); 23095d4f98a2SYan Zheng replaced = 1; 23105d4f98a2SYan Zheng } 23115d4f98a2SYan Zheng 23125d4f98a2SYan Zheng ret = walk_up_reloc_tree(reloc_root, path, &level); 23135d4f98a2SYan Zheng if (ret > 0) 23145d4f98a2SYan Zheng break; 23155d4f98a2SYan Zheng 23165d4f98a2SYan Zheng BUG_ON(level == 0); 23175d4f98a2SYan Zheng /* 23185d4f98a2SYan Zheng * save the merging progress in the drop_progress. 23195d4f98a2SYan Zheng * this is OK since root refs == 1 in this case. 23205d4f98a2SYan Zheng */ 23215d4f98a2SYan Zheng btrfs_node_key(path->nodes[level], &root_item->drop_progress, 23225d4f98a2SYan Zheng path->slots[level]); 23235d4f98a2SYan Zheng root_item->drop_level = level; 23245d4f98a2SYan Zheng 23253a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 23269e6a0c52SJosef Bacik trans = NULL; 23275d4f98a2SYan Zheng 23282ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 23295d4f98a2SYan Zheng 23305d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 23315d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 23325d4f98a2SYan Zheng } 23335d4f98a2SYan Zheng 23345d4f98a2SYan Zheng /* 23355d4f98a2SYan Zheng * handle the case only one block in the fs tree need to be 23365d4f98a2SYan Zheng * relocated and the block is tree root. 23375d4f98a2SYan Zheng */ 23385d4f98a2SYan Zheng leaf = btrfs_lock_root_node(root); 23395d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, leaf, NULL, 0, &leaf); 23405d4f98a2SYan Zheng btrfs_tree_unlock(leaf); 23415d4f98a2SYan Zheng free_extent_buffer(leaf); 23425d4f98a2SYan Zheng if (ret < 0) 23435d4f98a2SYan Zheng err = ret; 23445d4f98a2SYan Zheng out: 23455d4f98a2SYan Zheng btrfs_free_path(path); 23465d4f98a2SYan Zheng 2347d2311e69SQu Wenruo if (err == 0) 2348d2311e69SQu Wenruo insert_dirty_subvol(trans, rc, root); 23495d4f98a2SYan Zheng 23509e6a0c52SJosef Bacik if (trans) 23513a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 23525d4f98a2SYan Zheng 23532ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 23545d4f98a2SYan Zheng 23555d4f98a2SYan Zheng if (replaced && rc->stage == UPDATE_DATA_PTRS) 23565d4f98a2SYan Zheng invalidate_extent_cache(root, &key, &next_key); 23575d4f98a2SYan Zheng 23585d4f98a2SYan Zheng return err; 23595d4f98a2SYan Zheng } 23605d4f98a2SYan Zheng 23613fd0a558SYan, Zheng static noinline_for_stack 23623fd0a558SYan, Zheng int prepare_to_merge(struct reloc_control *rc, int err) 23635d4f98a2SYan Zheng { 23643fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 23650b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 23663fd0a558SYan, Zheng struct btrfs_root *reloc_root; 23675d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 23683fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 23693fd0a558SYan, Zheng u64 num_bytes = 0; 23703fd0a558SYan, Zheng int ret; 23713fd0a558SYan, Zheng 23720b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 23730b246afaSJeff Mahoney rc->merging_rsv_size += fs_info->nodesize * (BTRFS_MAX_LEVEL - 1) * 2; 23743fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated * 2; 23750b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 23767585717fSChris Mason 23773fd0a558SYan, Zheng again: 23783fd0a558SYan, Zheng if (!err) { 23793fd0a558SYan, Zheng num_bytes = rc->merging_rsv_size; 238008e007d2SMiao Xie ret = btrfs_block_rsv_add(root, rc->block_rsv, num_bytes, 238108e007d2SMiao Xie BTRFS_RESERVE_FLUSH_ALL); 23823fd0a558SYan, Zheng if (ret) 23833fd0a558SYan, Zheng err = ret; 23843fd0a558SYan, Zheng } 23853fd0a558SYan, Zheng 23867a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 23873612b495STsutomu Itoh if (IS_ERR(trans)) { 23883612b495STsutomu Itoh if (!err) 23892ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 23902ff7e61eSJeff Mahoney num_bytes); 23913612b495STsutomu Itoh return PTR_ERR(trans); 23923612b495STsutomu Itoh } 23933fd0a558SYan, Zheng 23943fd0a558SYan, Zheng if (!err) { 23953fd0a558SYan, Zheng if (num_bytes != rc->merging_rsv_size) { 23963a45bb20SJeff Mahoney btrfs_end_transaction(trans); 23972ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, 23982ff7e61eSJeff Mahoney num_bytes); 23993fd0a558SYan, Zheng goto again; 24003fd0a558SYan, Zheng } 24013fd0a558SYan, Zheng } 24023fd0a558SYan, Zheng 24033fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 24043fd0a558SYan, Zheng 24053fd0a558SYan, Zheng while (!list_empty(&rc->reloc_roots)) { 24063fd0a558SYan, Zheng reloc_root = list_entry(rc->reloc_roots.next, 24073fd0a558SYan, Zheng struct btrfs_root, root_list); 24083fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 24093fd0a558SYan, Zheng 24100b246afaSJeff Mahoney root = read_fs_root(fs_info, reloc_root->root_key.offset); 24113fd0a558SYan, Zheng BUG_ON(IS_ERR(root)); 24123fd0a558SYan, Zheng BUG_ON(root->reloc_root != reloc_root); 24133fd0a558SYan, Zheng 24143fd0a558SYan, Zheng /* 24153fd0a558SYan, Zheng * set reference count to 1, so btrfs_recover_relocation 24163fd0a558SYan, Zheng * knows it should resumes merging 24173fd0a558SYan, Zheng */ 24183fd0a558SYan, Zheng if (!err) 24193fd0a558SYan, Zheng btrfs_set_root_refs(&reloc_root->root_item, 1); 24203fd0a558SYan, Zheng btrfs_update_reloc_root(trans, root); 24213fd0a558SYan, Zheng 24223fd0a558SYan, Zheng list_add(&reloc_root->root_list, &reloc_roots); 24233fd0a558SYan, Zheng } 24243fd0a558SYan, Zheng 24253fd0a558SYan, Zheng list_splice(&reloc_roots, &rc->reloc_roots); 24263fd0a558SYan, Zheng 24273fd0a558SYan, Zheng if (!err) 24283a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 24293fd0a558SYan, Zheng else 24303a45bb20SJeff Mahoney btrfs_end_transaction(trans); 24313fd0a558SYan, Zheng return err; 24323fd0a558SYan, Zheng } 24333fd0a558SYan, Zheng 24343fd0a558SYan, Zheng static noinline_for_stack 2435aca1bba6SLiu Bo void free_reloc_roots(struct list_head *list) 2436aca1bba6SLiu Bo { 2437aca1bba6SLiu Bo struct btrfs_root *reloc_root; 2438aca1bba6SLiu Bo 2439aca1bba6SLiu Bo while (!list_empty(list)) { 2440aca1bba6SLiu Bo reloc_root = list_entry(list->next, struct btrfs_root, 2441aca1bba6SLiu Bo root_list); 2442bb166d72SNaohiro Aota __del_reloc_root(reloc_root); 24436bdf131fSJosef Bacik free_extent_buffer(reloc_root->node); 24446bdf131fSJosef Bacik free_extent_buffer(reloc_root->commit_root); 24456bdf131fSJosef Bacik reloc_root->node = NULL; 24466bdf131fSJosef Bacik reloc_root->commit_root = NULL; 2447aca1bba6SLiu Bo } 2448aca1bba6SLiu Bo } 2449aca1bba6SLiu Bo 2450aca1bba6SLiu Bo static noinline_for_stack 245194404e82SDavid Sterba void merge_reloc_roots(struct reloc_control *rc) 24523fd0a558SYan, Zheng { 24530b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 24545d4f98a2SYan Zheng struct btrfs_root *root; 24555d4f98a2SYan Zheng struct btrfs_root *reloc_root; 24563fd0a558SYan, Zheng LIST_HEAD(reloc_roots); 24573fd0a558SYan, Zheng int found = 0; 2458aca1bba6SLiu Bo int ret = 0; 24593fd0a558SYan, Zheng again: 24603fd0a558SYan, Zheng root = rc->extent_root; 24617585717fSChris Mason 24627585717fSChris Mason /* 24637585717fSChris Mason * this serializes us with btrfs_record_root_in_transaction, 24647585717fSChris Mason * we have to make sure nobody is in the middle of 24657585717fSChris Mason * adding their roots to the list while we are 24667585717fSChris Mason * doing this splice 24677585717fSChris Mason */ 24680b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 24693fd0a558SYan, Zheng list_splice_init(&rc->reloc_roots, &reloc_roots); 24700b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 24715d4f98a2SYan Zheng 24723fd0a558SYan, Zheng while (!list_empty(&reloc_roots)) { 24733fd0a558SYan, Zheng found = 1; 24743fd0a558SYan, Zheng reloc_root = list_entry(reloc_roots.next, 24753fd0a558SYan, Zheng struct btrfs_root, root_list); 24765d4f98a2SYan Zheng 24775d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 24780b246afaSJeff Mahoney root = read_fs_root(fs_info, 24795d4f98a2SYan Zheng reloc_root->root_key.offset); 24805d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 24815d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 24825d4f98a2SYan Zheng 24833fd0a558SYan, Zheng ret = merge_reloc_root(rc, root); 2484b37b39cdSJosef Bacik if (ret) { 248525e293c2SWang Shilong if (list_empty(&reloc_root->root_list)) 248625e293c2SWang Shilong list_add_tail(&reloc_root->root_list, 248725e293c2SWang Shilong &reloc_roots); 2488aca1bba6SLiu Bo goto out; 2489b37b39cdSJosef Bacik } 24903fd0a558SYan, Zheng } else { 24913fd0a558SYan, Zheng list_del_init(&reloc_root->root_list); 249230d40577SQu Wenruo /* Don't forget to queue this reloc root for cleanup */ 249330d40577SQu Wenruo list_add_tail(&reloc_root->reloc_dirty_list, 249430d40577SQu Wenruo &rc->dirty_subvol_roots); 24953fd0a558SYan, Zheng } 24965d4f98a2SYan Zheng } 24975d4f98a2SYan Zheng 24983fd0a558SYan, Zheng if (found) { 24993fd0a558SYan, Zheng found = 0; 25003fd0a558SYan, Zheng goto again; 25015d4f98a2SYan Zheng } 2502aca1bba6SLiu Bo out: 2503aca1bba6SLiu Bo if (ret) { 25040b246afaSJeff Mahoney btrfs_handle_fs_error(fs_info, ret, NULL); 2505aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 2506aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 2507467bb1d2SWang Shilong 2508467bb1d2SWang Shilong /* new reloc root may be added */ 25090b246afaSJeff Mahoney mutex_lock(&fs_info->reloc_mutex); 2510467bb1d2SWang Shilong list_splice_init(&rc->reloc_roots, &reloc_roots); 25110b246afaSJeff Mahoney mutex_unlock(&fs_info->reloc_mutex); 2512467bb1d2SWang Shilong if (!list_empty(&reloc_roots)) 2513467bb1d2SWang Shilong free_reloc_roots(&reloc_roots); 2514aca1bba6SLiu Bo } 2515aca1bba6SLiu Bo 25165d4f98a2SYan Zheng BUG_ON(!RB_EMPTY_ROOT(&rc->reloc_root_tree.rb_root)); 25175d4f98a2SYan Zheng } 25185d4f98a2SYan Zheng 25195d4f98a2SYan Zheng static void free_block_list(struct rb_root *blocks) 25205d4f98a2SYan Zheng { 25215d4f98a2SYan Zheng struct tree_block *block; 25225d4f98a2SYan Zheng struct rb_node *rb_node; 25235d4f98a2SYan Zheng while ((rb_node = rb_first(blocks))) { 25245d4f98a2SYan Zheng block = rb_entry(rb_node, struct tree_block, rb_node); 25255d4f98a2SYan Zheng rb_erase(rb_node, blocks); 25265d4f98a2SYan Zheng kfree(block); 25275d4f98a2SYan Zheng } 25285d4f98a2SYan Zheng } 25295d4f98a2SYan Zheng 25305d4f98a2SYan Zheng static int record_reloc_root_in_trans(struct btrfs_trans_handle *trans, 25315d4f98a2SYan Zheng struct btrfs_root *reloc_root) 25325d4f98a2SYan Zheng { 25330b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = reloc_root->fs_info; 25345d4f98a2SYan Zheng struct btrfs_root *root; 25355d4f98a2SYan Zheng 25365d4f98a2SYan Zheng if (reloc_root->last_trans == trans->transid) 25375d4f98a2SYan Zheng return 0; 25385d4f98a2SYan Zheng 25390b246afaSJeff Mahoney root = read_fs_root(fs_info, reloc_root->root_key.offset); 25405d4f98a2SYan Zheng BUG_ON(IS_ERR(root)); 25415d4f98a2SYan Zheng BUG_ON(root->reloc_root != reloc_root); 25425d4f98a2SYan Zheng 25435d4f98a2SYan Zheng return btrfs_record_root_in_trans(trans, root); 25445d4f98a2SYan Zheng } 25455d4f98a2SYan Zheng 25463fd0a558SYan, Zheng static noinline_for_stack 25473fd0a558SYan, Zheng struct btrfs_root *select_reloc_root(struct btrfs_trans_handle *trans, 25483fd0a558SYan, Zheng struct reloc_control *rc, 25495d4f98a2SYan Zheng struct backref_node *node, 2550dc4103f9SWang Shilong struct backref_edge *edges[]) 25515d4f98a2SYan Zheng { 25525d4f98a2SYan Zheng struct backref_node *next; 25535d4f98a2SYan Zheng struct btrfs_root *root; 25543fd0a558SYan, Zheng int index = 0; 25553fd0a558SYan, Zheng 25565d4f98a2SYan Zheng next = node; 25575d4f98a2SYan Zheng while (1) { 25585d4f98a2SYan Zheng cond_resched(); 25595d4f98a2SYan Zheng next = walk_up_backref(next, edges, &index); 25605d4f98a2SYan Zheng root = next->root; 25613fd0a558SYan, Zheng BUG_ON(!root); 256227cdeb70SMiao Xie BUG_ON(!test_bit(BTRFS_ROOT_REF_COWS, &root->state)); 25635d4f98a2SYan Zheng 25645d4f98a2SYan Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 25655d4f98a2SYan Zheng record_reloc_root_in_trans(trans, root); 25665d4f98a2SYan Zheng break; 25675d4f98a2SYan Zheng } 25685d4f98a2SYan Zheng 25695d4f98a2SYan Zheng btrfs_record_root_in_trans(trans, root); 25703fd0a558SYan, Zheng root = root->reloc_root; 25713fd0a558SYan, Zheng 25723fd0a558SYan, Zheng if (next->new_bytenr != root->node->start) { 25733fd0a558SYan, Zheng BUG_ON(next->new_bytenr); 25743fd0a558SYan, Zheng BUG_ON(!list_empty(&next->list)); 25753fd0a558SYan, Zheng next->new_bytenr = root->node->start; 25763fd0a558SYan, Zheng next->root = root; 25773fd0a558SYan, Zheng list_add_tail(&next->list, 25783fd0a558SYan, Zheng &rc->backref_cache.changed); 25793fd0a558SYan, Zheng __mark_block_processed(rc, next); 25805d4f98a2SYan Zheng break; 25815d4f98a2SYan Zheng } 25825d4f98a2SYan Zheng 25833fd0a558SYan, Zheng WARN_ON(1); 25845d4f98a2SYan Zheng root = NULL; 25855d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 25865d4f98a2SYan Zheng if (!next || next->level <= node->level) 25875d4f98a2SYan Zheng break; 25885d4f98a2SYan Zheng } 25893fd0a558SYan, Zheng if (!root) 25903fd0a558SYan, Zheng return NULL; 25915d4f98a2SYan Zheng 25923fd0a558SYan, Zheng next = node; 25933fd0a558SYan, Zheng /* setup backref node path for btrfs_reloc_cow_block */ 25943fd0a558SYan, Zheng while (1) { 25953fd0a558SYan, Zheng rc->backref_cache.path[next->level] = next; 25963fd0a558SYan, Zheng if (--index < 0) 25973fd0a558SYan, Zheng break; 25983fd0a558SYan, Zheng next = edges[index]->node[UPPER]; 25993fd0a558SYan, Zheng } 26005d4f98a2SYan Zheng return root; 26015d4f98a2SYan Zheng } 26025d4f98a2SYan Zheng 26033fd0a558SYan, Zheng /* 26043fd0a558SYan, Zheng * select a tree root for relocation. return NULL if the block 26053fd0a558SYan, Zheng * is reference counted. we should use do_relocation() in this 26063fd0a558SYan, Zheng * case. return a tree root pointer if the block isn't reference 26073fd0a558SYan, Zheng * counted. return -ENOENT if the block is root of reloc tree. 26083fd0a558SYan, Zheng */ 26095d4f98a2SYan Zheng static noinline_for_stack 2610147d256eSZhaolei struct btrfs_root *select_one_root(struct backref_node *node) 26115d4f98a2SYan Zheng { 26123fd0a558SYan, Zheng struct backref_node *next; 26133fd0a558SYan, Zheng struct btrfs_root *root; 26143fd0a558SYan, Zheng struct btrfs_root *fs_root = NULL; 26155d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 26163fd0a558SYan, Zheng int index = 0; 26173fd0a558SYan, Zheng 26183fd0a558SYan, Zheng next = node; 26193fd0a558SYan, Zheng while (1) { 26203fd0a558SYan, Zheng cond_resched(); 26213fd0a558SYan, Zheng next = walk_up_backref(next, edges, &index); 26223fd0a558SYan, Zheng root = next->root; 26233fd0a558SYan, Zheng BUG_ON(!root); 26243fd0a558SYan, Zheng 262525985edcSLucas De Marchi /* no other choice for non-references counted tree */ 262627cdeb70SMiao Xie if (!test_bit(BTRFS_ROOT_REF_COWS, &root->state)) 26273fd0a558SYan, Zheng return root; 26283fd0a558SYan, Zheng 26293fd0a558SYan, Zheng if (root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID) 26303fd0a558SYan, Zheng fs_root = root; 26313fd0a558SYan, Zheng 26323fd0a558SYan, Zheng if (next != node) 26333fd0a558SYan, Zheng return NULL; 26343fd0a558SYan, Zheng 26353fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 26363fd0a558SYan, Zheng if (!next || next->level <= node->level) 26373fd0a558SYan, Zheng break; 26383fd0a558SYan, Zheng } 26393fd0a558SYan, Zheng 26403fd0a558SYan, Zheng if (!fs_root) 26413fd0a558SYan, Zheng return ERR_PTR(-ENOENT); 26423fd0a558SYan, Zheng return fs_root; 26435d4f98a2SYan Zheng } 26445d4f98a2SYan Zheng 26455d4f98a2SYan Zheng static noinline_for_stack 26463fd0a558SYan, Zheng u64 calcu_metadata_size(struct reloc_control *rc, 26473fd0a558SYan, Zheng struct backref_node *node, int reserve) 26485d4f98a2SYan Zheng { 26490b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 26503fd0a558SYan, Zheng struct backref_node *next = node; 26513fd0a558SYan, Zheng struct backref_edge *edge; 26523fd0a558SYan, Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 26533fd0a558SYan, Zheng u64 num_bytes = 0; 26543fd0a558SYan, Zheng int index = 0; 26555d4f98a2SYan Zheng 26563fd0a558SYan, Zheng BUG_ON(reserve && node->processed); 26573fd0a558SYan, Zheng 26583fd0a558SYan, Zheng while (next) { 26593fd0a558SYan, Zheng cond_resched(); 26605d4f98a2SYan Zheng while (1) { 26613fd0a558SYan, Zheng if (next->processed && (reserve || next != node)) 26625d4f98a2SYan Zheng break; 26635d4f98a2SYan Zheng 26640b246afaSJeff Mahoney num_bytes += fs_info->nodesize; 26653fd0a558SYan, Zheng 26663fd0a558SYan, Zheng if (list_empty(&next->upper)) 26673fd0a558SYan, Zheng break; 26683fd0a558SYan, Zheng 26693fd0a558SYan, Zheng edge = list_entry(next->upper.next, 26703fd0a558SYan, Zheng struct backref_edge, list[LOWER]); 26713fd0a558SYan, Zheng edges[index++] = edge; 26723fd0a558SYan, Zheng next = edge->node[UPPER]; 26735d4f98a2SYan Zheng } 26743fd0a558SYan, Zheng next = walk_down_backref(edges, &index); 26753fd0a558SYan, Zheng } 26763fd0a558SYan, Zheng return num_bytes; 26773fd0a558SYan, Zheng } 26783fd0a558SYan, Zheng 26793fd0a558SYan, Zheng static int reserve_metadata_space(struct btrfs_trans_handle *trans, 26803fd0a558SYan, Zheng struct reloc_control *rc, 26813fd0a558SYan, Zheng struct backref_node *node) 26823fd0a558SYan, Zheng { 26833fd0a558SYan, Zheng struct btrfs_root *root = rc->extent_root; 2684da17066cSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 26853fd0a558SYan, Zheng u64 num_bytes; 26863fd0a558SYan, Zheng int ret; 26870647bf56SWang Shilong u64 tmp; 26883fd0a558SYan, Zheng 26893fd0a558SYan, Zheng num_bytes = calcu_metadata_size(rc, node, 1) * 2; 26903fd0a558SYan, Zheng 26913fd0a558SYan, Zheng trans->block_rsv = rc->block_rsv; 26920647bf56SWang Shilong rc->reserved_bytes += num_bytes; 26938ca17f0fSJosef Bacik 26948ca17f0fSJosef Bacik /* 26958ca17f0fSJosef Bacik * We are under a transaction here so we can only do limited flushing. 26968ca17f0fSJosef Bacik * If we get an enospc just kick back -EAGAIN so we know to drop the 26978ca17f0fSJosef Bacik * transaction and try to refill when we can flush all the things. 26988ca17f0fSJosef Bacik */ 26990647bf56SWang Shilong ret = btrfs_block_rsv_refill(root, rc->block_rsv, num_bytes, 27008ca17f0fSJosef Bacik BTRFS_RESERVE_FLUSH_LIMIT); 27013fd0a558SYan, Zheng if (ret) { 2702da17066cSJeff Mahoney tmp = fs_info->nodesize * RELOCATION_RESERVED_NODES; 27030647bf56SWang Shilong while (tmp <= rc->reserved_bytes) 27040647bf56SWang Shilong tmp <<= 1; 27050647bf56SWang Shilong /* 27060647bf56SWang Shilong * only one thread can access block_rsv at this point, 27070647bf56SWang Shilong * so we don't need hold lock to protect block_rsv. 27080647bf56SWang Shilong * we expand more reservation size here to allow enough 270952042d8eSAndrea Gelmini * space for relocation and we will return earlier in 27100647bf56SWang Shilong * enospc case. 27110647bf56SWang Shilong */ 2712da17066cSJeff Mahoney rc->block_rsv->size = tmp + fs_info->nodesize * 27130647bf56SWang Shilong RELOCATION_RESERVED_NODES; 27148ca17f0fSJosef Bacik return -EAGAIN; 27153fd0a558SYan, Zheng } 27163fd0a558SYan, Zheng 27173fd0a558SYan, Zheng return 0; 27183fd0a558SYan, Zheng } 27193fd0a558SYan, Zheng 27205d4f98a2SYan Zheng /* 27215d4f98a2SYan Zheng * relocate a block tree, and then update pointers in upper level 27225d4f98a2SYan Zheng * blocks that reference the block to point to the new location. 27235d4f98a2SYan Zheng * 27245d4f98a2SYan Zheng * if called by link_to_upper, the block has already been relocated. 27255d4f98a2SYan Zheng * in that case this function just updates pointers. 27265d4f98a2SYan Zheng */ 27275d4f98a2SYan Zheng static int do_relocation(struct btrfs_trans_handle *trans, 27283fd0a558SYan, Zheng struct reloc_control *rc, 27295d4f98a2SYan Zheng struct backref_node *node, 27305d4f98a2SYan Zheng struct btrfs_key *key, 27315d4f98a2SYan Zheng struct btrfs_path *path, int lowest) 27325d4f98a2SYan Zheng { 27332ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 27345d4f98a2SYan Zheng struct backref_node *upper; 27355d4f98a2SYan Zheng struct backref_edge *edge; 27365d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 27375d4f98a2SYan Zheng struct btrfs_root *root; 27385d4f98a2SYan Zheng struct extent_buffer *eb; 27395d4f98a2SYan Zheng u32 blocksize; 27405d4f98a2SYan Zheng u64 bytenr; 27415d4f98a2SYan Zheng u64 generation; 27425d4f98a2SYan Zheng int slot; 27435d4f98a2SYan Zheng int ret; 27445d4f98a2SYan Zheng int err = 0; 27455d4f98a2SYan Zheng 27465d4f98a2SYan Zheng BUG_ON(lowest && node->eb); 27475d4f98a2SYan Zheng 27485d4f98a2SYan Zheng path->lowest_level = node->level + 1; 27493fd0a558SYan, Zheng rc->backref_cache.path[node->level] = node; 27505d4f98a2SYan Zheng list_for_each_entry(edge, &node->upper, list[LOWER]) { 2751581c1760SQu Wenruo struct btrfs_key first_key; 275282fa113fSQu Wenruo struct btrfs_ref ref = { 0 }; 2753581c1760SQu Wenruo 27545d4f98a2SYan Zheng cond_resched(); 27555d4f98a2SYan Zheng 27565d4f98a2SYan Zheng upper = edge->node[UPPER]; 2757dc4103f9SWang Shilong root = select_reloc_root(trans, rc, upper, edges); 27583fd0a558SYan, Zheng BUG_ON(!root); 27595d4f98a2SYan Zheng 27603fd0a558SYan, Zheng if (upper->eb && !upper->locked) { 27613fd0a558SYan, Zheng if (!lowest) { 27623fd0a558SYan, Zheng ret = btrfs_bin_search(upper->eb, key, 27633fd0a558SYan, Zheng upper->level, &slot); 2764cbca7d59SFilipe Manana if (ret < 0) { 2765cbca7d59SFilipe Manana err = ret; 2766cbca7d59SFilipe Manana goto next; 2767cbca7d59SFilipe Manana } 27683fd0a558SYan, Zheng BUG_ON(ret); 27693fd0a558SYan, Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 27703fd0a558SYan, Zheng if (node->eb->start == bytenr) 27713fd0a558SYan, Zheng goto next; 27723fd0a558SYan, Zheng } 27735d4f98a2SYan Zheng drop_node_buffer(upper); 27743fd0a558SYan, Zheng } 27755d4f98a2SYan Zheng 27765d4f98a2SYan Zheng if (!upper->eb) { 27775d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 27783561b9dbSLiu Bo if (ret) { 27793561b9dbSLiu Bo if (ret < 0) 27805d4f98a2SYan Zheng err = ret; 27813561b9dbSLiu Bo else 27823561b9dbSLiu Bo err = -ENOENT; 27833561b9dbSLiu Bo 27843561b9dbSLiu Bo btrfs_release_path(path); 27855d4f98a2SYan Zheng break; 27865d4f98a2SYan Zheng } 27875d4f98a2SYan Zheng 27883fd0a558SYan, Zheng if (!upper->eb) { 27893fd0a558SYan, Zheng upper->eb = path->nodes[upper->level]; 27903fd0a558SYan, Zheng path->nodes[upper->level] = NULL; 27913fd0a558SYan, Zheng } else { 27923fd0a558SYan, Zheng BUG_ON(upper->eb != path->nodes[upper->level]); 27933fd0a558SYan, Zheng } 27943fd0a558SYan, Zheng 27953fd0a558SYan, Zheng upper->locked = 1; 27963fd0a558SYan, Zheng path->locks[upper->level] = 0; 27973fd0a558SYan, Zheng 27985d4f98a2SYan Zheng slot = path->slots[upper->level]; 2799b3b4aa74SDavid Sterba btrfs_release_path(path); 28005d4f98a2SYan Zheng } else { 28015d4f98a2SYan Zheng ret = btrfs_bin_search(upper->eb, key, upper->level, 28025d4f98a2SYan Zheng &slot); 2803cbca7d59SFilipe Manana if (ret < 0) { 2804cbca7d59SFilipe Manana err = ret; 2805cbca7d59SFilipe Manana goto next; 2806cbca7d59SFilipe Manana } 28075d4f98a2SYan Zheng BUG_ON(ret); 28085d4f98a2SYan Zheng } 28095d4f98a2SYan Zheng 28105d4f98a2SYan Zheng bytenr = btrfs_node_blockptr(upper->eb, slot); 28113fd0a558SYan, Zheng if (lowest) { 28124547f4d8SLiu Bo if (bytenr != node->bytenr) { 28134547f4d8SLiu Bo btrfs_err(root->fs_info, 28144547f4d8SLiu Bo "lowest leaf/node mismatch: bytenr %llu node->bytenr %llu slot %d upper %llu", 28154547f4d8SLiu Bo bytenr, node->bytenr, slot, 28164547f4d8SLiu Bo upper->eb->start); 28174547f4d8SLiu Bo err = -EIO; 28184547f4d8SLiu Bo goto next; 28194547f4d8SLiu Bo } 28205d4f98a2SYan Zheng } else { 28213fd0a558SYan, Zheng if (node->eb->start == bytenr) 28223fd0a558SYan, Zheng goto next; 28235d4f98a2SYan Zheng } 28245d4f98a2SYan Zheng 2825da17066cSJeff Mahoney blocksize = root->fs_info->nodesize; 28265d4f98a2SYan Zheng generation = btrfs_node_ptr_generation(upper->eb, slot); 2827581c1760SQu Wenruo btrfs_node_key_to_cpu(upper->eb, &first_key, slot); 2828581c1760SQu Wenruo eb = read_tree_block(fs_info, bytenr, generation, 2829581c1760SQu Wenruo upper->level - 1, &first_key); 283064c043deSLiu Bo if (IS_ERR(eb)) { 283164c043deSLiu Bo err = PTR_ERR(eb); 283264c043deSLiu Bo goto next; 283364c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 2834416bc658SJosef Bacik free_extent_buffer(eb); 283597d9a8a4STsutomu Itoh err = -EIO; 283697d9a8a4STsutomu Itoh goto next; 283797d9a8a4STsutomu Itoh } 28385d4f98a2SYan Zheng btrfs_tree_lock(eb); 28398bead258SDavid Sterba btrfs_set_lock_blocking_write(eb); 28405d4f98a2SYan Zheng 28415d4f98a2SYan Zheng if (!node->eb) { 28425d4f98a2SYan Zheng ret = btrfs_cow_block(trans, root, eb, upper->eb, 28435d4f98a2SYan Zheng slot, &eb); 28443fd0a558SYan, Zheng btrfs_tree_unlock(eb); 28453fd0a558SYan, Zheng free_extent_buffer(eb); 28465d4f98a2SYan Zheng if (ret < 0) { 28475d4f98a2SYan Zheng err = ret; 28483fd0a558SYan, Zheng goto next; 28495d4f98a2SYan Zheng } 28503fd0a558SYan, Zheng BUG_ON(node->eb != eb); 28515d4f98a2SYan Zheng } else { 28525d4f98a2SYan Zheng btrfs_set_node_blockptr(upper->eb, slot, 28535d4f98a2SYan Zheng node->eb->start); 28545d4f98a2SYan Zheng btrfs_set_node_ptr_generation(upper->eb, slot, 28555d4f98a2SYan Zheng trans->transid); 28565d4f98a2SYan Zheng btrfs_mark_buffer_dirty(upper->eb); 28575d4f98a2SYan Zheng 285882fa113fSQu Wenruo btrfs_init_generic_ref(&ref, BTRFS_ADD_DELAYED_REF, 28595d4f98a2SYan Zheng node->eb->start, blocksize, 286082fa113fSQu Wenruo upper->eb->start); 286182fa113fSQu Wenruo ref.real_root = root->root_key.objectid; 286282fa113fSQu Wenruo btrfs_init_tree_ref(&ref, node->level, 286382fa113fSQu Wenruo btrfs_header_owner(upper->eb)); 286482fa113fSQu Wenruo ret = btrfs_inc_extent_ref(trans, &ref); 28655d4f98a2SYan Zheng BUG_ON(ret); 28665d4f98a2SYan Zheng 28675d4f98a2SYan Zheng ret = btrfs_drop_subtree(trans, root, eb, upper->eb); 28685d4f98a2SYan Zheng BUG_ON(ret); 28695d4f98a2SYan Zheng } 28703fd0a558SYan, Zheng next: 28713fd0a558SYan, Zheng if (!upper->pending) 28723fd0a558SYan, Zheng drop_node_buffer(upper); 28733fd0a558SYan, Zheng else 28743fd0a558SYan, Zheng unlock_node_buffer(upper); 28753fd0a558SYan, Zheng if (err) 28763fd0a558SYan, Zheng break; 28775d4f98a2SYan Zheng } 28783fd0a558SYan, Zheng 28793fd0a558SYan, Zheng if (!err && node->pending) { 28803fd0a558SYan, Zheng drop_node_buffer(node); 28813fd0a558SYan, Zheng list_move_tail(&node->list, &rc->backref_cache.changed); 28823fd0a558SYan, Zheng node->pending = 0; 28835d4f98a2SYan Zheng } 28843fd0a558SYan, Zheng 28855d4f98a2SYan Zheng path->lowest_level = 0; 28863fd0a558SYan, Zheng BUG_ON(err == -ENOSPC); 28875d4f98a2SYan Zheng return err; 28885d4f98a2SYan Zheng } 28895d4f98a2SYan Zheng 28905d4f98a2SYan Zheng static int link_to_upper(struct btrfs_trans_handle *trans, 28913fd0a558SYan, Zheng struct reloc_control *rc, 28925d4f98a2SYan Zheng struct backref_node *node, 28935d4f98a2SYan Zheng struct btrfs_path *path) 28945d4f98a2SYan Zheng { 28955d4f98a2SYan Zheng struct btrfs_key key; 28965d4f98a2SYan Zheng 28975d4f98a2SYan Zheng btrfs_node_key_to_cpu(node->eb, &key, 0); 28983fd0a558SYan, Zheng return do_relocation(trans, rc, node, &key, path, 0); 28995d4f98a2SYan Zheng } 29005d4f98a2SYan Zheng 29015d4f98a2SYan Zheng static int finish_pending_nodes(struct btrfs_trans_handle *trans, 29023fd0a558SYan, Zheng struct reloc_control *rc, 29033fd0a558SYan, Zheng struct btrfs_path *path, int err) 29045d4f98a2SYan Zheng { 29053fd0a558SYan, Zheng LIST_HEAD(list); 29063fd0a558SYan, Zheng struct backref_cache *cache = &rc->backref_cache; 29075d4f98a2SYan Zheng struct backref_node *node; 29085d4f98a2SYan Zheng int level; 29095d4f98a2SYan Zheng int ret; 29105d4f98a2SYan Zheng 29115d4f98a2SYan Zheng for (level = 0; level < BTRFS_MAX_LEVEL; level++) { 29125d4f98a2SYan Zheng while (!list_empty(&cache->pending[level])) { 29135d4f98a2SYan Zheng node = list_entry(cache->pending[level].next, 29143fd0a558SYan, Zheng struct backref_node, list); 29153fd0a558SYan, Zheng list_move_tail(&node->list, &list); 29163fd0a558SYan, Zheng BUG_ON(!node->pending); 29175d4f98a2SYan Zheng 29183fd0a558SYan, Zheng if (!err) { 29193fd0a558SYan, Zheng ret = link_to_upper(trans, rc, node, path); 29205d4f98a2SYan Zheng if (ret < 0) 29215d4f98a2SYan Zheng err = ret; 29225d4f98a2SYan Zheng } 29235d4f98a2SYan Zheng } 29243fd0a558SYan, Zheng list_splice_init(&list, &cache->pending[level]); 29253fd0a558SYan, Zheng } 29265d4f98a2SYan Zheng return err; 29275d4f98a2SYan Zheng } 29285d4f98a2SYan Zheng 29295d4f98a2SYan Zheng static void mark_block_processed(struct reloc_control *rc, 29303fd0a558SYan, Zheng u64 bytenr, u32 blocksize) 29313fd0a558SYan, Zheng { 29323fd0a558SYan, Zheng set_extent_bits(&rc->processed_blocks, bytenr, bytenr + blocksize - 1, 2933ceeb0ae7SDavid Sterba EXTENT_DIRTY); 29343fd0a558SYan, Zheng } 29353fd0a558SYan, Zheng 29363fd0a558SYan, Zheng static void __mark_block_processed(struct reloc_control *rc, 29375d4f98a2SYan Zheng struct backref_node *node) 29385d4f98a2SYan Zheng { 29395d4f98a2SYan Zheng u32 blocksize; 29405d4f98a2SYan Zheng if (node->level == 0 || 29415d4f98a2SYan Zheng in_block_group(node->bytenr, rc->block_group)) { 2942da17066cSJeff Mahoney blocksize = rc->extent_root->fs_info->nodesize; 29433fd0a558SYan, Zheng mark_block_processed(rc, node->bytenr, blocksize); 29445d4f98a2SYan Zheng } 29455d4f98a2SYan Zheng node->processed = 1; 29465d4f98a2SYan Zheng } 29475d4f98a2SYan Zheng 29485d4f98a2SYan Zheng /* 29495d4f98a2SYan Zheng * mark a block and all blocks directly/indirectly reference the block 29505d4f98a2SYan Zheng * as processed. 29515d4f98a2SYan Zheng */ 29525d4f98a2SYan Zheng static void update_processed_blocks(struct reloc_control *rc, 29535d4f98a2SYan Zheng struct backref_node *node) 29545d4f98a2SYan Zheng { 29555d4f98a2SYan Zheng struct backref_node *next = node; 29565d4f98a2SYan Zheng struct backref_edge *edge; 29575d4f98a2SYan Zheng struct backref_edge *edges[BTRFS_MAX_LEVEL - 1]; 29585d4f98a2SYan Zheng int index = 0; 29595d4f98a2SYan Zheng 29605d4f98a2SYan Zheng while (next) { 29615d4f98a2SYan Zheng cond_resched(); 29625d4f98a2SYan Zheng while (1) { 29635d4f98a2SYan Zheng if (next->processed) 29645d4f98a2SYan Zheng break; 29655d4f98a2SYan Zheng 29663fd0a558SYan, Zheng __mark_block_processed(rc, next); 29675d4f98a2SYan Zheng 29685d4f98a2SYan Zheng if (list_empty(&next->upper)) 29695d4f98a2SYan Zheng break; 29705d4f98a2SYan Zheng 29715d4f98a2SYan Zheng edge = list_entry(next->upper.next, 29725d4f98a2SYan Zheng struct backref_edge, list[LOWER]); 29735d4f98a2SYan Zheng edges[index++] = edge; 29745d4f98a2SYan Zheng next = edge->node[UPPER]; 29755d4f98a2SYan Zheng } 29765d4f98a2SYan Zheng next = walk_down_backref(edges, &index); 29775d4f98a2SYan Zheng } 29785d4f98a2SYan Zheng } 29795d4f98a2SYan Zheng 29807476dfdaSDavid Sterba static int tree_block_processed(u64 bytenr, struct reloc_control *rc) 29815d4f98a2SYan Zheng { 2982da17066cSJeff Mahoney u32 blocksize = rc->extent_root->fs_info->nodesize; 29837476dfdaSDavid Sterba 29845d4f98a2SYan Zheng if (test_range_bit(&rc->processed_blocks, bytenr, 29859655d298SChris Mason bytenr + blocksize - 1, EXTENT_DIRTY, 1, NULL)) 29865d4f98a2SYan Zheng return 1; 29875d4f98a2SYan Zheng return 0; 29885d4f98a2SYan Zheng } 29895d4f98a2SYan Zheng 29902ff7e61eSJeff Mahoney static int get_tree_block_key(struct btrfs_fs_info *fs_info, 29915d4f98a2SYan Zheng struct tree_block *block) 29925d4f98a2SYan Zheng { 29935d4f98a2SYan Zheng struct extent_buffer *eb; 29945d4f98a2SYan Zheng 29955d4f98a2SYan Zheng BUG_ON(block->key_ready); 2996581c1760SQu Wenruo eb = read_tree_block(fs_info, block->bytenr, block->key.offset, 2997581c1760SQu Wenruo block->level, NULL); 299864c043deSLiu Bo if (IS_ERR(eb)) { 299964c043deSLiu Bo return PTR_ERR(eb); 300064c043deSLiu Bo } else if (!extent_buffer_uptodate(eb)) { 3001416bc658SJosef Bacik free_extent_buffer(eb); 3002416bc658SJosef Bacik return -EIO; 3003416bc658SJosef Bacik } 30045d4f98a2SYan Zheng if (block->level == 0) 30055d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &block->key, 0); 30065d4f98a2SYan Zheng else 30075d4f98a2SYan Zheng btrfs_node_key_to_cpu(eb, &block->key, 0); 30085d4f98a2SYan Zheng free_extent_buffer(eb); 30095d4f98a2SYan Zheng block->key_ready = 1; 30105d4f98a2SYan Zheng return 0; 30115d4f98a2SYan Zheng } 30125d4f98a2SYan Zheng 30135d4f98a2SYan Zheng /* 30145d4f98a2SYan Zheng * helper function to relocate a tree block 30155d4f98a2SYan Zheng */ 30165d4f98a2SYan Zheng static int relocate_tree_block(struct btrfs_trans_handle *trans, 30175d4f98a2SYan Zheng struct reloc_control *rc, 30185d4f98a2SYan Zheng struct backref_node *node, 30195d4f98a2SYan Zheng struct btrfs_key *key, 30205d4f98a2SYan Zheng struct btrfs_path *path) 30215d4f98a2SYan Zheng { 30225d4f98a2SYan Zheng struct btrfs_root *root; 30233fd0a558SYan, Zheng int ret = 0; 30245d4f98a2SYan Zheng 30253fd0a558SYan, Zheng if (!node) 30265d4f98a2SYan Zheng return 0; 30273fd0a558SYan, Zheng 30283fd0a558SYan, Zheng BUG_ON(node->processed); 3029147d256eSZhaolei root = select_one_root(node); 30303fd0a558SYan, Zheng if (root == ERR_PTR(-ENOENT)) { 30313fd0a558SYan, Zheng update_processed_blocks(rc, node); 30323fd0a558SYan, Zheng goto out; 30335d4f98a2SYan Zheng } 30345d4f98a2SYan Zheng 303527cdeb70SMiao Xie if (!root || test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 30363fd0a558SYan, Zheng ret = reserve_metadata_space(trans, rc, node); 30373fd0a558SYan, Zheng if (ret) 30385d4f98a2SYan Zheng goto out; 30395d4f98a2SYan Zheng } 30403fd0a558SYan, Zheng 30413fd0a558SYan, Zheng if (root) { 304227cdeb70SMiao Xie if (test_bit(BTRFS_ROOT_REF_COWS, &root->state)) { 30433fd0a558SYan, Zheng BUG_ON(node->new_bytenr); 30443fd0a558SYan, Zheng BUG_ON(!list_empty(&node->list)); 30453fd0a558SYan, Zheng btrfs_record_root_in_trans(trans, root); 30463fd0a558SYan, Zheng root = root->reloc_root; 30473fd0a558SYan, Zheng node->new_bytenr = root->node->start; 30483fd0a558SYan, Zheng node->root = root; 30493fd0a558SYan, Zheng list_add_tail(&node->list, &rc->backref_cache.changed); 30503fd0a558SYan, Zheng } else { 30515d4f98a2SYan Zheng path->lowest_level = node->level; 30525d4f98a2SYan Zheng ret = btrfs_search_slot(trans, root, key, path, 0, 1); 3053b3b4aa74SDavid Sterba btrfs_release_path(path); 30543fd0a558SYan, Zheng if (ret > 0) 30555d4f98a2SYan Zheng ret = 0; 30563fd0a558SYan, Zheng } 30573fd0a558SYan, Zheng if (!ret) 30583fd0a558SYan, Zheng update_processed_blocks(rc, node); 30593fd0a558SYan, Zheng } else { 30603fd0a558SYan, Zheng ret = do_relocation(trans, rc, node, key, path, 1); 30613fd0a558SYan, Zheng } 30625d4f98a2SYan Zheng out: 30630647bf56SWang Shilong if (ret || node->level == 0 || node->cowonly) 30643fd0a558SYan, Zheng remove_backref_node(&rc->backref_cache, node); 30655d4f98a2SYan Zheng return ret; 30665d4f98a2SYan Zheng } 30675d4f98a2SYan Zheng 30685d4f98a2SYan Zheng /* 30695d4f98a2SYan Zheng * relocate a list of blocks 30705d4f98a2SYan Zheng */ 30715d4f98a2SYan Zheng static noinline_for_stack 30725d4f98a2SYan Zheng int relocate_tree_blocks(struct btrfs_trans_handle *trans, 30735d4f98a2SYan Zheng struct reloc_control *rc, struct rb_root *blocks) 30745d4f98a2SYan Zheng { 30752ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 30765d4f98a2SYan Zheng struct backref_node *node; 30775d4f98a2SYan Zheng struct btrfs_path *path; 30785d4f98a2SYan Zheng struct tree_block *block; 307998ff7b94SQu Wenruo struct tree_block *next; 30805d4f98a2SYan Zheng int ret; 30815d4f98a2SYan Zheng int err = 0; 30825d4f98a2SYan Zheng 30835d4f98a2SYan Zheng path = btrfs_alloc_path(); 3084e1a12670SLiu Bo if (!path) { 3085e1a12670SLiu Bo err = -ENOMEM; 308634c2b290SDavid Sterba goto out_free_blocks; 3087e1a12670SLiu Bo } 30885d4f98a2SYan Zheng 308998ff7b94SQu Wenruo /* Kick in readahead for tree blocks with missing keys */ 309098ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 30915d4f98a2SYan Zheng if (!block->key_ready) 30922ff7e61eSJeff Mahoney readahead_tree_block(fs_info, block->bytenr); 30935d4f98a2SYan Zheng } 30945d4f98a2SYan Zheng 309598ff7b94SQu Wenruo /* Get first keys */ 309698ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 309734c2b290SDavid Sterba if (!block->key_ready) { 30982ff7e61eSJeff Mahoney err = get_tree_block_key(fs_info, block); 309934c2b290SDavid Sterba if (err) 310034c2b290SDavid Sterba goto out_free_path; 310134c2b290SDavid Sterba } 31025d4f98a2SYan Zheng } 31035d4f98a2SYan Zheng 310498ff7b94SQu Wenruo /* Do tree relocation */ 310598ff7b94SQu Wenruo rbtree_postorder_for_each_entry_safe(block, next, blocks, rb_node) { 31063fd0a558SYan, Zheng node = build_backref_tree(rc, &block->key, 31075d4f98a2SYan Zheng block->level, block->bytenr); 31085d4f98a2SYan Zheng if (IS_ERR(node)) { 31095d4f98a2SYan Zheng err = PTR_ERR(node); 31105d4f98a2SYan Zheng goto out; 31115d4f98a2SYan Zheng } 31125d4f98a2SYan Zheng 31135d4f98a2SYan Zheng ret = relocate_tree_block(trans, rc, node, &block->key, 31145d4f98a2SYan Zheng path); 31155d4f98a2SYan Zheng if (ret < 0) { 311698ff7b94SQu Wenruo if (ret != -EAGAIN || &block->rb_node == rb_first(blocks)) 31175d4f98a2SYan Zheng err = ret; 31185d4f98a2SYan Zheng goto out; 31195d4f98a2SYan Zheng } 31205d4f98a2SYan Zheng } 31215d4f98a2SYan Zheng out: 31223fd0a558SYan, Zheng err = finish_pending_nodes(trans, rc, path, err); 31235d4f98a2SYan Zheng 312434c2b290SDavid Sterba out_free_path: 31255d4f98a2SYan Zheng btrfs_free_path(path); 312634c2b290SDavid Sterba out_free_blocks: 3127e1a12670SLiu Bo free_block_list(blocks); 31285d4f98a2SYan Zheng return err; 31295d4f98a2SYan Zheng } 31305d4f98a2SYan Zheng 31315d4f98a2SYan Zheng static noinline_for_stack 3132efa56464SYan, Zheng int prealloc_file_extent_cluster(struct inode *inode, 3133efa56464SYan, Zheng struct file_extent_cluster *cluster) 3134efa56464SYan, Zheng { 3135efa56464SYan, Zheng u64 alloc_hint = 0; 3136efa56464SYan, Zheng u64 start; 3137efa56464SYan, Zheng u64 end; 3138efa56464SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 3139efa56464SYan, Zheng u64 num_bytes; 3140efa56464SYan, Zheng int nr = 0; 3141efa56464SYan, Zheng int ret = 0; 3142dcb40c19SWang Xiaoguang u64 prealloc_start = cluster->start - offset; 3143dcb40c19SWang Xiaoguang u64 prealloc_end = cluster->end - offset; 314418513091SWang Xiaoguang u64 cur_offset; 3145364ecf36SQu Wenruo struct extent_changeset *data_reserved = NULL; 3146efa56464SYan, Zheng 3147efa56464SYan, Zheng BUG_ON(cluster->start != cluster->boundary[0]); 31485955102cSAl Viro inode_lock(inode); 3149efa56464SYan, Zheng 3150364ecf36SQu Wenruo ret = btrfs_check_data_free_space(inode, &data_reserved, prealloc_start, 3151dcb40c19SWang Xiaoguang prealloc_end + 1 - prealloc_start); 3152efa56464SYan, Zheng if (ret) 3153efa56464SYan, Zheng goto out; 3154efa56464SYan, Zheng 315518513091SWang Xiaoguang cur_offset = prealloc_start; 3156efa56464SYan, Zheng while (nr < cluster->nr) { 3157efa56464SYan, Zheng start = cluster->boundary[nr] - offset; 3158efa56464SYan, Zheng if (nr + 1 < cluster->nr) 3159efa56464SYan, Zheng end = cluster->boundary[nr + 1] - 1 - offset; 3160efa56464SYan, Zheng else 3161efa56464SYan, Zheng end = cluster->end - offset; 3162efa56464SYan, Zheng 3163d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 3164efa56464SYan, Zheng num_bytes = end + 1 - start; 316518513091SWang Xiaoguang if (cur_offset < start) 3166bc42bda2SQu Wenruo btrfs_free_reserved_data_space(inode, data_reserved, 3167bc42bda2SQu Wenruo cur_offset, start - cur_offset); 3168efa56464SYan, Zheng ret = btrfs_prealloc_file_range(inode, 0, start, 3169efa56464SYan, Zheng num_bytes, num_bytes, 3170efa56464SYan, Zheng end + 1, &alloc_hint); 317118513091SWang Xiaoguang cur_offset = end + 1; 3172d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 3173efa56464SYan, Zheng if (ret) 3174efa56464SYan, Zheng break; 3175efa56464SYan, Zheng nr++; 3176efa56464SYan, Zheng } 317718513091SWang Xiaoguang if (cur_offset < prealloc_end) 3178bc42bda2SQu Wenruo btrfs_free_reserved_data_space(inode, data_reserved, 3179bc42bda2SQu Wenruo cur_offset, prealloc_end + 1 - cur_offset); 3180efa56464SYan, Zheng out: 31815955102cSAl Viro inode_unlock(inode); 3182364ecf36SQu Wenruo extent_changeset_free(data_reserved); 3183efa56464SYan, Zheng return ret; 3184efa56464SYan, Zheng } 3185efa56464SYan, Zheng 3186efa56464SYan, Zheng static noinline_for_stack 31870257bb82SYan, Zheng int setup_extent_mapping(struct inode *inode, u64 start, u64 end, 31880257bb82SYan, Zheng u64 block_start) 31895d4f98a2SYan Zheng { 31900b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 31915d4f98a2SYan Zheng struct extent_map_tree *em_tree = &BTRFS_I(inode)->extent_tree; 31925d4f98a2SYan Zheng struct extent_map *em; 31930257bb82SYan, Zheng int ret = 0; 31945d4f98a2SYan Zheng 3195172ddd60SDavid Sterba em = alloc_extent_map(); 31960257bb82SYan, Zheng if (!em) 31970257bb82SYan, Zheng return -ENOMEM; 31980257bb82SYan, Zheng 31995d4f98a2SYan Zheng em->start = start; 32000257bb82SYan, Zheng em->len = end + 1 - start; 32010257bb82SYan, Zheng em->block_len = em->len; 32020257bb82SYan, Zheng em->block_start = block_start; 32030b246afaSJeff Mahoney em->bdev = fs_info->fs_devices->latest_bdev; 32045d4f98a2SYan Zheng set_bit(EXTENT_FLAG_PINNED, &em->flags); 32055d4f98a2SYan Zheng 3206d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, start, end); 32075d4f98a2SYan Zheng while (1) { 3208890871beSChris Mason write_lock(&em_tree->lock); 320909a2a8f9SJosef Bacik ret = add_extent_mapping(em_tree, em, 0); 3210890871beSChris Mason write_unlock(&em_tree->lock); 32115d4f98a2SYan Zheng if (ret != -EEXIST) { 32125d4f98a2SYan Zheng free_extent_map(em); 32135d4f98a2SYan Zheng break; 32145d4f98a2SYan Zheng } 3215dcdbc059SNikolay Borisov btrfs_drop_extent_cache(BTRFS_I(inode), start, end, 0); 32165d4f98a2SYan Zheng } 3217d0082371SJeff Mahoney unlock_extent(&BTRFS_I(inode)->io_tree, start, end); 32180257bb82SYan, Zheng return ret; 32190257bb82SYan, Zheng } 32205d4f98a2SYan Zheng 32210257bb82SYan, Zheng static int relocate_file_extent_cluster(struct inode *inode, 32220257bb82SYan, Zheng struct file_extent_cluster *cluster) 32230257bb82SYan, Zheng { 32242ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 32250257bb82SYan, Zheng u64 page_start; 32260257bb82SYan, Zheng u64 page_end; 32270257bb82SYan, Zheng u64 offset = BTRFS_I(inode)->index_cnt; 32280257bb82SYan, Zheng unsigned long index; 32290257bb82SYan, Zheng unsigned long last_index; 32300257bb82SYan, Zheng struct page *page; 32310257bb82SYan, Zheng struct file_ra_state *ra; 32323b16a4e3SJosef Bacik gfp_t mask = btrfs_alloc_write_mask(inode->i_mapping); 32330257bb82SYan, Zheng int nr = 0; 32340257bb82SYan, Zheng int ret = 0; 32350257bb82SYan, Zheng 32360257bb82SYan, Zheng if (!cluster->nr) 32370257bb82SYan, Zheng return 0; 32380257bb82SYan, Zheng 32390257bb82SYan, Zheng ra = kzalloc(sizeof(*ra), GFP_NOFS); 32400257bb82SYan, Zheng if (!ra) 32410257bb82SYan, Zheng return -ENOMEM; 32420257bb82SYan, Zheng 3243efa56464SYan, Zheng ret = prealloc_file_extent_cluster(inode, cluster); 32440257bb82SYan, Zheng if (ret) 3245efa56464SYan, Zheng goto out; 32460257bb82SYan, Zheng 32470257bb82SYan, Zheng file_ra_state_init(ra, inode->i_mapping); 32480257bb82SYan, Zheng 3249efa56464SYan, Zheng ret = setup_extent_mapping(inode, cluster->start - offset, 3250efa56464SYan, Zheng cluster->end - offset, cluster->start); 3251efa56464SYan, Zheng if (ret) 3252efa56464SYan, Zheng goto out; 3253efa56464SYan, Zheng 325409cbfeafSKirill A. Shutemov index = (cluster->start - offset) >> PAGE_SHIFT; 325509cbfeafSKirill A. Shutemov last_index = (cluster->end - offset) >> PAGE_SHIFT; 32560257bb82SYan, Zheng while (index <= last_index) { 32579f3db423SNikolay Borisov ret = btrfs_delalloc_reserve_metadata(BTRFS_I(inode), 32589f3db423SNikolay Borisov PAGE_SIZE); 3259efa56464SYan, Zheng if (ret) 3260efa56464SYan, Zheng goto out; 3261efa56464SYan, Zheng 32620257bb82SYan, Zheng page = find_lock_page(inode->i_mapping, index); 32630257bb82SYan, Zheng if (!page) { 32640257bb82SYan, Zheng page_cache_sync_readahead(inode->i_mapping, 32650257bb82SYan, Zheng ra, NULL, index, 32660257bb82SYan, Zheng last_index + 1 - index); 3267a94733d0SJosef Bacik page = find_or_create_page(inode->i_mapping, index, 32683b16a4e3SJosef Bacik mask); 32690257bb82SYan, Zheng if (!page) { 3270691fa059SNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 327143b18595SQu Wenruo PAGE_SIZE, true); 32720257bb82SYan, Zheng ret = -ENOMEM; 3273efa56464SYan, Zheng goto out; 32740257bb82SYan, Zheng } 32750257bb82SYan, Zheng } 32760257bb82SYan, Zheng 32770257bb82SYan, Zheng if (PageReadahead(page)) { 32780257bb82SYan, Zheng page_cache_async_readahead(inode->i_mapping, 32790257bb82SYan, Zheng ra, NULL, page, index, 32800257bb82SYan, Zheng last_index + 1 - index); 32810257bb82SYan, Zheng } 32820257bb82SYan, Zheng 32830257bb82SYan, Zheng if (!PageUptodate(page)) { 32840257bb82SYan, Zheng btrfs_readpage(NULL, page); 32850257bb82SYan, Zheng lock_page(page); 32860257bb82SYan, Zheng if (!PageUptodate(page)) { 32870257bb82SYan, Zheng unlock_page(page); 328809cbfeafSKirill A. Shutemov put_page(page); 3289691fa059SNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 329043b18595SQu Wenruo PAGE_SIZE, true); 32918b62f87bSJosef Bacik btrfs_delalloc_release_extents(BTRFS_I(inode), 329243b18595SQu Wenruo PAGE_SIZE, true); 32930257bb82SYan, Zheng ret = -EIO; 3294efa56464SYan, Zheng goto out; 32950257bb82SYan, Zheng } 32960257bb82SYan, Zheng } 32970257bb82SYan, Zheng 32984eee4fa4SMiao Xie page_start = page_offset(page); 329909cbfeafSKirill A. Shutemov page_end = page_start + PAGE_SIZE - 1; 33000257bb82SYan, Zheng 3301d0082371SJeff Mahoney lock_extent(&BTRFS_I(inode)->io_tree, page_start, page_end); 33020257bb82SYan, Zheng 33030257bb82SYan, Zheng set_page_extent_mapped(page); 33040257bb82SYan, Zheng 33050257bb82SYan, Zheng if (nr < cluster->nr && 33060257bb82SYan, Zheng page_start + offset == cluster->boundary[nr]) { 33070257bb82SYan, Zheng set_extent_bits(&BTRFS_I(inode)->io_tree, 33080257bb82SYan, Zheng page_start, page_end, 3309ceeb0ae7SDavid Sterba EXTENT_BOUNDARY); 33100257bb82SYan, Zheng nr++; 33110257bb82SYan, Zheng } 33120257bb82SYan, Zheng 3313765f3cebSNikolay Borisov ret = btrfs_set_extent_delalloc(inode, page_start, page_end, 0, 3314765f3cebSNikolay Borisov NULL, 0); 3315765f3cebSNikolay Borisov if (ret) { 3316765f3cebSNikolay Borisov unlock_page(page); 3317765f3cebSNikolay Borisov put_page(page); 3318765f3cebSNikolay Borisov btrfs_delalloc_release_metadata(BTRFS_I(inode), 331943b18595SQu Wenruo PAGE_SIZE, true); 3320765f3cebSNikolay Borisov btrfs_delalloc_release_extents(BTRFS_I(inode), 332143b18595SQu Wenruo PAGE_SIZE, true); 3322765f3cebSNikolay Borisov 3323765f3cebSNikolay Borisov clear_extent_bits(&BTRFS_I(inode)->io_tree, 3324765f3cebSNikolay Borisov page_start, page_end, 3325765f3cebSNikolay Borisov EXTENT_LOCKED | EXTENT_BOUNDARY); 3326765f3cebSNikolay Borisov goto out; 3327765f3cebSNikolay Borisov 3328765f3cebSNikolay Borisov } 33290257bb82SYan, Zheng set_page_dirty(page); 33300257bb82SYan, Zheng 33310257bb82SYan, Zheng unlock_extent(&BTRFS_I(inode)->io_tree, 3332d0082371SJeff Mahoney page_start, page_end); 33330257bb82SYan, Zheng unlock_page(page); 333409cbfeafSKirill A. Shutemov put_page(page); 33350257bb82SYan, Zheng 33360257bb82SYan, Zheng index++; 333743b18595SQu Wenruo btrfs_delalloc_release_extents(BTRFS_I(inode), PAGE_SIZE, 333843b18595SQu Wenruo false); 3339efa56464SYan, Zheng balance_dirty_pages_ratelimited(inode->i_mapping); 33402ff7e61eSJeff Mahoney btrfs_throttle(fs_info); 33410257bb82SYan, Zheng } 33420257bb82SYan, Zheng WARN_ON(nr != cluster->nr); 3343efa56464SYan, Zheng out: 33440257bb82SYan, Zheng kfree(ra); 33450257bb82SYan, Zheng return ret; 33460257bb82SYan, Zheng } 33470257bb82SYan, Zheng 33480257bb82SYan, Zheng static noinline_for_stack 33490257bb82SYan, Zheng int relocate_data_extent(struct inode *inode, struct btrfs_key *extent_key, 33500257bb82SYan, Zheng struct file_extent_cluster *cluster) 33510257bb82SYan, Zheng { 33520257bb82SYan, Zheng int ret; 33530257bb82SYan, Zheng 33540257bb82SYan, Zheng if (cluster->nr > 0 && extent_key->objectid != cluster->end + 1) { 33550257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 33560257bb82SYan, Zheng if (ret) 33570257bb82SYan, Zheng return ret; 33580257bb82SYan, Zheng cluster->nr = 0; 33590257bb82SYan, Zheng } 33600257bb82SYan, Zheng 33610257bb82SYan, Zheng if (!cluster->nr) 33620257bb82SYan, Zheng cluster->start = extent_key->objectid; 33630257bb82SYan, Zheng else 33640257bb82SYan, Zheng BUG_ON(cluster->nr >= MAX_EXTENTS); 33650257bb82SYan, Zheng cluster->end = extent_key->objectid + extent_key->offset - 1; 33660257bb82SYan, Zheng cluster->boundary[cluster->nr] = extent_key->objectid; 33670257bb82SYan, Zheng cluster->nr++; 33680257bb82SYan, Zheng 33690257bb82SYan, Zheng if (cluster->nr >= MAX_EXTENTS) { 33700257bb82SYan, Zheng ret = relocate_file_extent_cluster(inode, cluster); 33710257bb82SYan, Zheng if (ret) 33720257bb82SYan, Zheng return ret; 33730257bb82SYan, Zheng cluster->nr = 0; 33740257bb82SYan, Zheng } 33750257bb82SYan, Zheng return 0; 33765d4f98a2SYan Zheng } 33775d4f98a2SYan Zheng 33785d4f98a2SYan Zheng /* 33795d4f98a2SYan Zheng * helper to add a tree block to the list. 33805d4f98a2SYan Zheng * the major work is getting the generation and level of the block 33815d4f98a2SYan Zheng */ 33825d4f98a2SYan Zheng static int add_tree_block(struct reloc_control *rc, 33835d4f98a2SYan Zheng struct btrfs_key *extent_key, 33845d4f98a2SYan Zheng struct btrfs_path *path, 33855d4f98a2SYan Zheng struct rb_root *blocks) 33865d4f98a2SYan Zheng { 33875d4f98a2SYan Zheng struct extent_buffer *eb; 33885d4f98a2SYan Zheng struct btrfs_extent_item *ei; 33895d4f98a2SYan Zheng struct btrfs_tree_block_info *bi; 33905d4f98a2SYan Zheng struct tree_block *block; 33915d4f98a2SYan Zheng struct rb_node *rb_node; 33925d4f98a2SYan Zheng u32 item_size; 33935d4f98a2SYan Zheng int level = -1; 33947fdf4b60SWang Shilong u64 generation; 33955d4f98a2SYan Zheng 33965d4f98a2SYan Zheng eb = path->nodes[0]; 33975d4f98a2SYan Zheng item_size = btrfs_item_size_nr(eb, path->slots[0]); 33985d4f98a2SYan Zheng 33993173a18fSJosef Bacik if (extent_key->type == BTRFS_METADATA_ITEM_KEY || 34003173a18fSJosef Bacik item_size >= sizeof(*ei) + sizeof(*bi)) { 34015d4f98a2SYan Zheng ei = btrfs_item_ptr(eb, path->slots[0], 34025d4f98a2SYan Zheng struct btrfs_extent_item); 34033173a18fSJosef Bacik if (extent_key->type == BTRFS_EXTENT_ITEM_KEY) { 34045d4f98a2SYan Zheng bi = (struct btrfs_tree_block_info *)(ei + 1); 34055d4f98a2SYan Zheng level = btrfs_tree_block_level(eb, bi); 34065d4f98a2SYan Zheng } else { 34073173a18fSJosef Bacik level = (int)extent_key->offset; 34083173a18fSJosef Bacik } 34093173a18fSJosef Bacik generation = btrfs_extent_generation(eb, ei); 34106d8ff4e4SDavid Sterba } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) { 3411ba3c2b19SNikolay Borisov btrfs_print_v0_err(eb->fs_info); 3412ba3c2b19SNikolay Borisov btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL); 3413ba3c2b19SNikolay Borisov return -EINVAL; 34143173a18fSJosef Bacik } else { 34155d4f98a2SYan Zheng BUG(); 34165d4f98a2SYan Zheng } 34175d4f98a2SYan Zheng 3418b3b4aa74SDavid Sterba btrfs_release_path(path); 34195d4f98a2SYan Zheng 34205d4f98a2SYan Zheng BUG_ON(level == -1); 34215d4f98a2SYan Zheng 34225d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 34235d4f98a2SYan Zheng if (!block) 34245d4f98a2SYan Zheng return -ENOMEM; 34255d4f98a2SYan Zheng 34265d4f98a2SYan Zheng block->bytenr = extent_key->objectid; 3427da17066cSJeff Mahoney block->key.objectid = rc->extent_root->fs_info->nodesize; 34285d4f98a2SYan Zheng block->key.offset = generation; 34295d4f98a2SYan Zheng block->level = level; 34305d4f98a2SYan Zheng block->key_ready = 0; 34315d4f98a2SYan Zheng 34325d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, &block->rb_node); 343343c04fb1SJeff Mahoney if (rb_node) 343443c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, block->bytenr); 34355d4f98a2SYan Zheng 34365d4f98a2SYan Zheng return 0; 34375d4f98a2SYan Zheng } 34385d4f98a2SYan Zheng 34395d4f98a2SYan Zheng /* 34405d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_SHARED_DATA_REF_KEY 34415d4f98a2SYan Zheng */ 34425d4f98a2SYan Zheng static int __add_tree_block(struct reloc_control *rc, 34435d4f98a2SYan Zheng u64 bytenr, u32 blocksize, 34445d4f98a2SYan Zheng struct rb_root *blocks) 34455d4f98a2SYan Zheng { 34460b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 34475d4f98a2SYan Zheng struct btrfs_path *path; 34485d4f98a2SYan Zheng struct btrfs_key key; 34495d4f98a2SYan Zheng int ret; 34500b246afaSJeff Mahoney bool skinny = btrfs_fs_incompat(fs_info, SKINNY_METADATA); 34515d4f98a2SYan Zheng 34527476dfdaSDavid Sterba if (tree_block_processed(bytenr, rc)) 34535d4f98a2SYan Zheng return 0; 34545d4f98a2SYan Zheng 34555d4f98a2SYan Zheng if (tree_search(blocks, bytenr)) 34565d4f98a2SYan Zheng return 0; 34575d4f98a2SYan Zheng 34585d4f98a2SYan Zheng path = btrfs_alloc_path(); 34595d4f98a2SYan Zheng if (!path) 34605d4f98a2SYan Zheng return -ENOMEM; 3461aee68ee5SJosef Bacik again: 34625d4f98a2SYan Zheng key.objectid = bytenr; 3463aee68ee5SJosef Bacik if (skinny) { 3464aee68ee5SJosef Bacik key.type = BTRFS_METADATA_ITEM_KEY; 3465aee68ee5SJosef Bacik key.offset = (u64)-1; 3466aee68ee5SJosef Bacik } else { 34675d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 34685d4f98a2SYan Zheng key.offset = blocksize; 3469aee68ee5SJosef Bacik } 34705d4f98a2SYan Zheng 34715d4f98a2SYan Zheng path->search_commit_root = 1; 34725d4f98a2SYan Zheng path->skip_locking = 1; 34735d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 0, 0); 34745d4f98a2SYan Zheng if (ret < 0) 34755d4f98a2SYan Zheng goto out; 34765d4f98a2SYan Zheng 3477aee68ee5SJosef Bacik if (ret > 0 && skinny) { 3478aee68ee5SJosef Bacik if (path->slots[0]) { 3479aee68ee5SJosef Bacik path->slots[0]--; 3480aee68ee5SJosef Bacik btrfs_item_key_to_cpu(path->nodes[0], &key, 3481aee68ee5SJosef Bacik path->slots[0]); 34823173a18fSJosef Bacik if (key.objectid == bytenr && 3483aee68ee5SJosef Bacik (key.type == BTRFS_METADATA_ITEM_KEY || 3484aee68ee5SJosef Bacik (key.type == BTRFS_EXTENT_ITEM_KEY && 3485aee68ee5SJosef Bacik key.offset == blocksize))) 34863173a18fSJosef Bacik ret = 0; 34873173a18fSJosef Bacik } 3488aee68ee5SJosef Bacik 3489aee68ee5SJosef Bacik if (ret) { 3490aee68ee5SJosef Bacik skinny = false; 3491aee68ee5SJosef Bacik btrfs_release_path(path); 3492aee68ee5SJosef Bacik goto again; 3493aee68ee5SJosef Bacik } 3494aee68ee5SJosef Bacik } 3495cdccee99SLiu Bo if (ret) { 3496cdccee99SLiu Bo ASSERT(ret == 1); 3497cdccee99SLiu Bo btrfs_print_leaf(path->nodes[0]); 3498cdccee99SLiu Bo btrfs_err(fs_info, 3499cdccee99SLiu Bo "tree block extent item (%llu) is not found in extent tree", 3500cdccee99SLiu Bo bytenr); 3501cdccee99SLiu Bo WARN_ON(1); 3502cdccee99SLiu Bo ret = -EINVAL; 3503cdccee99SLiu Bo goto out; 3504cdccee99SLiu Bo } 35053173a18fSJosef Bacik 35065d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, blocks); 35075d4f98a2SYan Zheng out: 35085d4f98a2SYan Zheng btrfs_free_path(path); 35095d4f98a2SYan Zheng return ret; 35105d4f98a2SYan Zheng } 35115d4f98a2SYan Zheng 35125d4f98a2SYan Zheng /* 35135d4f98a2SYan Zheng * helper to check if the block use full backrefs for pointers in it 35145d4f98a2SYan Zheng */ 35155d4f98a2SYan Zheng static int block_use_full_backref(struct reloc_control *rc, 35165d4f98a2SYan Zheng struct extent_buffer *eb) 35175d4f98a2SYan Zheng { 35185d4f98a2SYan Zheng u64 flags; 35195d4f98a2SYan Zheng int ret; 35205d4f98a2SYan Zheng 35215d4f98a2SYan Zheng if (btrfs_header_flag(eb, BTRFS_HEADER_FLAG_RELOC) || 35225d4f98a2SYan Zheng btrfs_header_backref_rev(eb) < BTRFS_MIXED_BACKREF_REV) 35235d4f98a2SYan Zheng return 1; 35245d4f98a2SYan Zheng 35252ff7e61eSJeff Mahoney ret = btrfs_lookup_extent_info(NULL, rc->extent_root->fs_info, 35263173a18fSJosef Bacik eb->start, btrfs_header_level(eb), 1, 35273173a18fSJosef Bacik NULL, &flags); 35285d4f98a2SYan Zheng BUG_ON(ret); 35295d4f98a2SYan Zheng 35305d4f98a2SYan Zheng if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) 35315d4f98a2SYan Zheng ret = 1; 35325d4f98a2SYan Zheng else 35335d4f98a2SYan Zheng ret = 0; 35345d4f98a2SYan Zheng return ret; 35355d4f98a2SYan Zheng } 35365d4f98a2SYan Zheng 35370af3d00bSJosef Bacik static int delete_block_group_cache(struct btrfs_fs_info *fs_info, 35381bbc621eSChris Mason struct btrfs_block_group_cache *block_group, 35391bbc621eSChris Mason struct inode *inode, 35401bbc621eSChris Mason u64 ino) 35410af3d00bSJosef Bacik { 35420af3d00bSJosef Bacik struct btrfs_key key; 35430af3d00bSJosef Bacik struct btrfs_root *root = fs_info->tree_root; 35440af3d00bSJosef Bacik struct btrfs_trans_handle *trans; 35450af3d00bSJosef Bacik int ret = 0; 35460af3d00bSJosef Bacik 35470af3d00bSJosef Bacik if (inode) 35480af3d00bSJosef Bacik goto truncate; 35490af3d00bSJosef Bacik 35500af3d00bSJosef Bacik key.objectid = ino; 35510af3d00bSJosef Bacik key.type = BTRFS_INODE_ITEM_KEY; 35520af3d00bSJosef Bacik key.offset = 0; 35530af3d00bSJosef Bacik 35540af3d00bSJosef Bacik inode = btrfs_iget(fs_info->sb, &key, root, NULL); 35552e19f1f9SAl Viro if (IS_ERR(inode)) 35560af3d00bSJosef Bacik return -ENOENT; 35570af3d00bSJosef Bacik 35580af3d00bSJosef Bacik truncate: 35592ff7e61eSJeff Mahoney ret = btrfs_check_trunc_cache_free_space(fs_info, 35607b61cd92SMiao Xie &fs_info->global_block_rsv); 35617b61cd92SMiao Xie if (ret) 35627b61cd92SMiao Xie goto out; 35637b61cd92SMiao Xie 35647a7eaa40SJosef Bacik trans = btrfs_join_transaction(root); 35650af3d00bSJosef Bacik if (IS_ERR(trans)) { 35663612b495STsutomu Itoh ret = PTR_ERR(trans); 35670af3d00bSJosef Bacik goto out; 35680af3d00bSJosef Bacik } 35690af3d00bSJosef Bacik 357077ab86bfSJeff Mahoney ret = btrfs_truncate_free_space_cache(trans, block_group, inode); 35710af3d00bSJosef Bacik 35723a45bb20SJeff Mahoney btrfs_end_transaction(trans); 35732ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 35740af3d00bSJosef Bacik out: 35750af3d00bSJosef Bacik iput(inode); 35760af3d00bSJosef Bacik return ret; 35770af3d00bSJosef Bacik } 35780af3d00bSJosef Bacik 35795d4f98a2SYan Zheng /* 35805d4f98a2SYan Zheng * helper to add tree blocks for backref of type BTRFS_EXTENT_DATA_REF_KEY 35815d4f98a2SYan Zheng * this function scans fs tree to find blocks reference the data extent 35825d4f98a2SYan Zheng */ 35835d4f98a2SYan Zheng static int find_data_references(struct reloc_control *rc, 35845d4f98a2SYan Zheng struct btrfs_key *extent_key, 35855d4f98a2SYan Zheng struct extent_buffer *leaf, 35865d4f98a2SYan Zheng struct btrfs_extent_data_ref *ref, 35875d4f98a2SYan Zheng struct rb_root *blocks) 35885d4f98a2SYan Zheng { 35890b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 35905d4f98a2SYan Zheng struct btrfs_path *path; 35915d4f98a2SYan Zheng struct tree_block *block; 35925d4f98a2SYan Zheng struct btrfs_root *root; 35935d4f98a2SYan Zheng struct btrfs_file_extent_item *fi; 35945d4f98a2SYan Zheng struct rb_node *rb_node; 35955d4f98a2SYan Zheng struct btrfs_key key; 35965d4f98a2SYan Zheng u64 ref_root; 35975d4f98a2SYan Zheng u64 ref_objectid; 35985d4f98a2SYan Zheng u64 ref_offset; 35995d4f98a2SYan Zheng u32 ref_count; 36005d4f98a2SYan Zheng u32 nritems; 36015d4f98a2SYan Zheng int err = 0; 36025d4f98a2SYan Zheng int added = 0; 36035d4f98a2SYan Zheng int counted; 36045d4f98a2SYan Zheng int ret; 36055d4f98a2SYan Zheng 36065d4f98a2SYan Zheng ref_root = btrfs_extent_data_ref_root(leaf, ref); 36075d4f98a2SYan Zheng ref_objectid = btrfs_extent_data_ref_objectid(leaf, ref); 36085d4f98a2SYan Zheng ref_offset = btrfs_extent_data_ref_offset(leaf, ref); 36095d4f98a2SYan Zheng ref_count = btrfs_extent_data_ref_count(leaf, ref); 36105d4f98a2SYan Zheng 36110af3d00bSJosef Bacik /* 36120af3d00bSJosef Bacik * This is an extent belonging to the free space cache, lets just delete 36130af3d00bSJosef Bacik * it and redo the search. 36140af3d00bSJosef Bacik */ 36150af3d00bSJosef Bacik if (ref_root == BTRFS_ROOT_TREE_OBJECTID) { 36160b246afaSJeff Mahoney ret = delete_block_group_cache(fs_info, rc->block_group, 36170af3d00bSJosef Bacik NULL, ref_objectid); 36180af3d00bSJosef Bacik if (ret != -ENOENT) 36190af3d00bSJosef Bacik return ret; 36200af3d00bSJosef Bacik ret = 0; 36210af3d00bSJosef Bacik } 36220af3d00bSJosef Bacik 36230af3d00bSJosef Bacik path = btrfs_alloc_path(); 36240af3d00bSJosef Bacik if (!path) 36250af3d00bSJosef Bacik return -ENOMEM; 3626e4058b54SDavid Sterba path->reada = READA_FORWARD; 36270af3d00bSJosef Bacik 36280b246afaSJeff Mahoney root = read_fs_root(fs_info, ref_root); 36295d4f98a2SYan Zheng if (IS_ERR(root)) { 36305d4f98a2SYan Zheng err = PTR_ERR(root); 36315d4f98a2SYan Zheng goto out; 36325d4f98a2SYan Zheng } 36335d4f98a2SYan Zheng 36345d4f98a2SYan Zheng key.objectid = ref_objectid; 36355d4f98a2SYan Zheng key.type = BTRFS_EXTENT_DATA_KEY; 363684850e8dSYan, Zheng if (ref_offset > ((u64)-1 << 32)) 363784850e8dSYan, Zheng key.offset = 0; 363884850e8dSYan, Zheng else 363984850e8dSYan, Zheng key.offset = ref_offset; 36405d4f98a2SYan Zheng 36415d4f98a2SYan Zheng path->search_commit_root = 1; 36425d4f98a2SYan Zheng path->skip_locking = 1; 36435d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); 36445d4f98a2SYan Zheng if (ret < 0) { 36455d4f98a2SYan Zheng err = ret; 36465d4f98a2SYan Zheng goto out; 36475d4f98a2SYan Zheng } 36485d4f98a2SYan Zheng 36495d4f98a2SYan Zheng leaf = path->nodes[0]; 36505d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 36515d4f98a2SYan Zheng /* 36525d4f98a2SYan Zheng * the references in tree blocks that use full backrefs 36535d4f98a2SYan Zheng * are not counted in 36545d4f98a2SYan Zheng */ 36555d4f98a2SYan Zheng if (block_use_full_backref(rc, leaf)) 36565d4f98a2SYan Zheng counted = 0; 36575d4f98a2SYan Zheng else 36585d4f98a2SYan Zheng counted = 1; 36595d4f98a2SYan Zheng rb_node = tree_search(blocks, leaf->start); 36605d4f98a2SYan Zheng if (rb_node) { 36615d4f98a2SYan Zheng if (counted) 36625d4f98a2SYan Zheng added = 1; 36635d4f98a2SYan Zheng else 36645d4f98a2SYan Zheng path->slots[0] = nritems; 36655d4f98a2SYan Zheng } 36665d4f98a2SYan Zheng 36675d4f98a2SYan Zheng while (ref_count > 0) { 36685d4f98a2SYan Zheng while (path->slots[0] >= nritems) { 36695d4f98a2SYan Zheng ret = btrfs_next_leaf(root, path); 36705d4f98a2SYan Zheng if (ret < 0) { 36715d4f98a2SYan Zheng err = ret; 36725d4f98a2SYan Zheng goto out; 36735d4f98a2SYan Zheng } 3674fae7f21cSDulshani Gunawardhana if (WARN_ON(ret > 0)) 36755d4f98a2SYan Zheng goto out; 36765d4f98a2SYan Zheng 36775d4f98a2SYan Zheng leaf = path->nodes[0]; 36785d4f98a2SYan Zheng nritems = btrfs_header_nritems(leaf); 36795d4f98a2SYan Zheng added = 0; 36805d4f98a2SYan Zheng 36815d4f98a2SYan Zheng if (block_use_full_backref(rc, leaf)) 36825d4f98a2SYan Zheng counted = 0; 36835d4f98a2SYan Zheng else 36845d4f98a2SYan Zheng counted = 1; 36855d4f98a2SYan Zheng rb_node = tree_search(blocks, leaf->start); 36865d4f98a2SYan Zheng if (rb_node) { 36875d4f98a2SYan Zheng if (counted) 36885d4f98a2SYan Zheng added = 1; 36895d4f98a2SYan Zheng else 36905d4f98a2SYan Zheng path->slots[0] = nritems; 36915d4f98a2SYan Zheng } 36925d4f98a2SYan Zheng } 36935d4f98a2SYan Zheng 36945d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 3695fae7f21cSDulshani Gunawardhana if (WARN_ON(key.objectid != ref_objectid || 3696fae7f21cSDulshani Gunawardhana key.type != BTRFS_EXTENT_DATA_KEY)) 36975d4f98a2SYan Zheng break; 36985d4f98a2SYan Zheng 36995d4f98a2SYan Zheng fi = btrfs_item_ptr(leaf, path->slots[0], 37005d4f98a2SYan Zheng struct btrfs_file_extent_item); 37015d4f98a2SYan Zheng 37025d4f98a2SYan Zheng if (btrfs_file_extent_type(leaf, fi) == 37035d4f98a2SYan Zheng BTRFS_FILE_EXTENT_INLINE) 37045d4f98a2SYan Zheng goto next; 37055d4f98a2SYan Zheng 37065d4f98a2SYan Zheng if (btrfs_file_extent_disk_bytenr(leaf, fi) != 37075d4f98a2SYan Zheng extent_key->objectid) 37085d4f98a2SYan Zheng goto next; 37095d4f98a2SYan Zheng 37105d4f98a2SYan Zheng key.offset -= btrfs_file_extent_offset(leaf, fi); 37115d4f98a2SYan Zheng if (key.offset != ref_offset) 37125d4f98a2SYan Zheng goto next; 37135d4f98a2SYan Zheng 37145d4f98a2SYan Zheng if (counted) 37155d4f98a2SYan Zheng ref_count--; 37165d4f98a2SYan Zheng if (added) 37175d4f98a2SYan Zheng goto next; 37185d4f98a2SYan Zheng 37197476dfdaSDavid Sterba if (!tree_block_processed(leaf->start, rc)) { 37205d4f98a2SYan Zheng block = kmalloc(sizeof(*block), GFP_NOFS); 37215d4f98a2SYan Zheng if (!block) { 37225d4f98a2SYan Zheng err = -ENOMEM; 37235d4f98a2SYan Zheng break; 37245d4f98a2SYan Zheng } 37255d4f98a2SYan Zheng block->bytenr = leaf->start; 37265d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &block->key, 0); 37275d4f98a2SYan Zheng block->level = 0; 37285d4f98a2SYan Zheng block->key_ready = 1; 37295d4f98a2SYan Zheng rb_node = tree_insert(blocks, block->bytenr, 37305d4f98a2SYan Zheng &block->rb_node); 373143c04fb1SJeff Mahoney if (rb_node) 373243c04fb1SJeff Mahoney backref_tree_panic(rb_node, -EEXIST, 373343c04fb1SJeff Mahoney block->bytenr); 37345d4f98a2SYan Zheng } 37355d4f98a2SYan Zheng if (counted) 37365d4f98a2SYan Zheng added = 1; 37375d4f98a2SYan Zheng else 37385d4f98a2SYan Zheng path->slots[0] = nritems; 37395d4f98a2SYan Zheng next: 37405d4f98a2SYan Zheng path->slots[0]++; 37415d4f98a2SYan Zheng 37425d4f98a2SYan Zheng } 37435d4f98a2SYan Zheng out: 37445d4f98a2SYan Zheng btrfs_free_path(path); 37455d4f98a2SYan Zheng return err; 37465d4f98a2SYan Zheng } 37475d4f98a2SYan Zheng 37485d4f98a2SYan Zheng /* 37492c016dc2SLiu Bo * helper to find all tree blocks that reference a given data extent 37505d4f98a2SYan Zheng */ 37515d4f98a2SYan Zheng static noinline_for_stack 37525d4f98a2SYan Zheng int add_data_references(struct reloc_control *rc, 37535d4f98a2SYan Zheng struct btrfs_key *extent_key, 37545d4f98a2SYan Zheng struct btrfs_path *path, 37555d4f98a2SYan Zheng struct rb_root *blocks) 37565d4f98a2SYan Zheng { 37575d4f98a2SYan Zheng struct btrfs_key key; 37585d4f98a2SYan Zheng struct extent_buffer *eb; 37595d4f98a2SYan Zheng struct btrfs_extent_data_ref *dref; 37605d4f98a2SYan Zheng struct btrfs_extent_inline_ref *iref; 37615d4f98a2SYan Zheng unsigned long ptr; 37625d4f98a2SYan Zheng unsigned long end; 3763da17066cSJeff Mahoney u32 blocksize = rc->extent_root->fs_info->nodesize; 3764647f63bdSFilipe David Borba Manana int ret = 0; 37655d4f98a2SYan Zheng int err = 0; 37665d4f98a2SYan Zheng 37675d4f98a2SYan Zheng eb = path->nodes[0]; 37685d4f98a2SYan Zheng ptr = btrfs_item_ptr_offset(eb, path->slots[0]); 37695d4f98a2SYan Zheng end = ptr + btrfs_item_size_nr(eb, path->slots[0]); 37705d4f98a2SYan Zheng ptr += sizeof(struct btrfs_extent_item); 37715d4f98a2SYan Zheng 37725d4f98a2SYan Zheng while (ptr < end) { 37735d4f98a2SYan Zheng iref = (struct btrfs_extent_inline_ref *)ptr; 37743de28d57SLiu Bo key.type = btrfs_get_extent_inline_ref_type(eb, iref, 37753de28d57SLiu Bo BTRFS_REF_TYPE_DATA); 37765d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY) { 37775d4f98a2SYan Zheng key.offset = btrfs_extent_inline_ref_offset(eb, iref); 37785d4f98a2SYan Zheng ret = __add_tree_block(rc, key.offset, blocksize, 37795d4f98a2SYan Zheng blocks); 37805d4f98a2SYan Zheng } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { 37815d4f98a2SYan Zheng dref = (struct btrfs_extent_data_ref *)(&iref->offset); 37825d4f98a2SYan Zheng ret = find_data_references(rc, extent_key, 37835d4f98a2SYan Zheng eb, dref, blocks); 37845d4f98a2SYan Zheng } else { 3785af431dcbSSu Yue ret = -EUCLEAN; 3786b14c55a1SLiu Bo btrfs_err(rc->extent_root->fs_info, 3787b14c55a1SLiu Bo "extent %llu slot %d has an invalid inline ref type", 3788b14c55a1SLiu Bo eb->start, path->slots[0]); 37895d4f98a2SYan Zheng } 3790647f63bdSFilipe David Borba Manana if (ret) { 3791647f63bdSFilipe David Borba Manana err = ret; 3792647f63bdSFilipe David Borba Manana goto out; 3793647f63bdSFilipe David Borba Manana } 37945d4f98a2SYan Zheng ptr += btrfs_extent_inline_ref_size(key.type); 37955d4f98a2SYan Zheng } 37965d4f98a2SYan Zheng WARN_ON(ptr > end); 37975d4f98a2SYan Zheng 37985d4f98a2SYan Zheng while (1) { 37995d4f98a2SYan Zheng cond_resched(); 38005d4f98a2SYan Zheng eb = path->nodes[0]; 38015d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(eb)) { 38025d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 38035d4f98a2SYan Zheng if (ret < 0) { 38045d4f98a2SYan Zheng err = ret; 38055d4f98a2SYan Zheng break; 38065d4f98a2SYan Zheng } 38075d4f98a2SYan Zheng if (ret > 0) 38085d4f98a2SYan Zheng break; 38095d4f98a2SYan Zheng eb = path->nodes[0]; 38105d4f98a2SYan Zheng } 38115d4f98a2SYan Zheng 38125d4f98a2SYan Zheng btrfs_item_key_to_cpu(eb, &key, path->slots[0]); 38135d4f98a2SYan Zheng if (key.objectid != extent_key->objectid) 38145d4f98a2SYan Zheng break; 38155d4f98a2SYan Zheng 38165d4f98a2SYan Zheng if (key.type == BTRFS_SHARED_DATA_REF_KEY) { 38175d4f98a2SYan Zheng ret = __add_tree_block(rc, key.offset, blocksize, 38185d4f98a2SYan Zheng blocks); 38195d4f98a2SYan Zheng } else if (key.type == BTRFS_EXTENT_DATA_REF_KEY) { 38205d4f98a2SYan Zheng dref = btrfs_item_ptr(eb, path->slots[0], 38215d4f98a2SYan Zheng struct btrfs_extent_data_ref); 38225d4f98a2SYan Zheng ret = find_data_references(rc, extent_key, 38235d4f98a2SYan Zheng eb, dref, blocks); 38246d8ff4e4SDavid Sterba } else if (unlikely(key.type == BTRFS_EXTENT_REF_V0_KEY)) { 3825ba3c2b19SNikolay Borisov btrfs_print_v0_err(eb->fs_info); 3826ba3c2b19SNikolay Borisov btrfs_handle_fs_error(eb->fs_info, -EINVAL, NULL); 3827ba3c2b19SNikolay Borisov ret = -EINVAL; 38285d4f98a2SYan Zheng } else { 38295d4f98a2SYan Zheng ret = 0; 38305d4f98a2SYan Zheng } 38315d4f98a2SYan Zheng if (ret) { 38325d4f98a2SYan Zheng err = ret; 38335d4f98a2SYan Zheng break; 38345d4f98a2SYan Zheng } 38355d4f98a2SYan Zheng path->slots[0]++; 38365d4f98a2SYan Zheng } 3837647f63bdSFilipe David Borba Manana out: 3838b3b4aa74SDavid Sterba btrfs_release_path(path); 38395d4f98a2SYan Zheng if (err) 38405d4f98a2SYan Zheng free_block_list(blocks); 38415d4f98a2SYan Zheng return err; 38425d4f98a2SYan Zheng } 38435d4f98a2SYan Zheng 38445d4f98a2SYan Zheng /* 38452c016dc2SLiu Bo * helper to find next unprocessed extent 38465d4f98a2SYan Zheng */ 38475d4f98a2SYan Zheng static noinline_for_stack 3848147d256eSZhaolei int find_next_extent(struct reloc_control *rc, struct btrfs_path *path, 38493fd0a558SYan, Zheng struct btrfs_key *extent_key) 38505d4f98a2SYan Zheng { 38510b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 38525d4f98a2SYan Zheng struct btrfs_key key; 38535d4f98a2SYan Zheng struct extent_buffer *leaf; 38545d4f98a2SYan Zheng u64 start, end, last; 38555d4f98a2SYan Zheng int ret; 38565d4f98a2SYan Zheng 38575d4f98a2SYan Zheng last = rc->block_group->key.objectid + rc->block_group->key.offset; 38585d4f98a2SYan Zheng while (1) { 38595d4f98a2SYan Zheng cond_resched(); 38605d4f98a2SYan Zheng if (rc->search_start >= last) { 38615d4f98a2SYan Zheng ret = 1; 38625d4f98a2SYan Zheng break; 38635d4f98a2SYan Zheng } 38645d4f98a2SYan Zheng 38655d4f98a2SYan Zheng key.objectid = rc->search_start; 38665d4f98a2SYan Zheng key.type = BTRFS_EXTENT_ITEM_KEY; 38675d4f98a2SYan Zheng key.offset = 0; 38685d4f98a2SYan Zheng 38695d4f98a2SYan Zheng path->search_commit_root = 1; 38705d4f98a2SYan Zheng path->skip_locking = 1; 38715d4f98a2SYan Zheng ret = btrfs_search_slot(NULL, rc->extent_root, &key, path, 38725d4f98a2SYan Zheng 0, 0); 38735d4f98a2SYan Zheng if (ret < 0) 38745d4f98a2SYan Zheng break; 38755d4f98a2SYan Zheng next: 38765d4f98a2SYan Zheng leaf = path->nodes[0]; 38775d4f98a2SYan Zheng if (path->slots[0] >= btrfs_header_nritems(leaf)) { 38785d4f98a2SYan Zheng ret = btrfs_next_leaf(rc->extent_root, path); 38795d4f98a2SYan Zheng if (ret != 0) 38805d4f98a2SYan Zheng break; 38815d4f98a2SYan Zheng leaf = path->nodes[0]; 38825d4f98a2SYan Zheng } 38835d4f98a2SYan Zheng 38845d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 38855d4f98a2SYan Zheng if (key.objectid >= last) { 38865d4f98a2SYan Zheng ret = 1; 38875d4f98a2SYan Zheng break; 38885d4f98a2SYan Zheng } 38895d4f98a2SYan Zheng 38903173a18fSJosef Bacik if (key.type != BTRFS_EXTENT_ITEM_KEY && 38913173a18fSJosef Bacik key.type != BTRFS_METADATA_ITEM_KEY) { 38923173a18fSJosef Bacik path->slots[0]++; 38933173a18fSJosef Bacik goto next; 38943173a18fSJosef Bacik } 38953173a18fSJosef Bacik 38963173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY && 38975d4f98a2SYan Zheng key.objectid + key.offset <= rc->search_start) { 38985d4f98a2SYan Zheng path->slots[0]++; 38995d4f98a2SYan Zheng goto next; 39005d4f98a2SYan Zheng } 39015d4f98a2SYan Zheng 39023173a18fSJosef Bacik if (key.type == BTRFS_METADATA_ITEM_KEY && 39030b246afaSJeff Mahoney key.objectid + fs_info->nodesize <= 39043173a18fSJosef Bacik rc->search_start) { 39053173a18fSJosef Bacik path->slots[0]++; 39063173a18fSJosef Bacik goto next; 39073173a18fSJosef Bacik } 39083173a18fSJosef Bacik 39095d4f98a2SYan Zheng ret = find_first_extent_bit(&rc->processed_blocks, 39105d4f98a2SYan Zheng key.objectid, &start, &end, 3911e6138876SJosef Bacik EXTENT_DIRTY, NULL); 39125d4f98a2SYan Zheng 39135d4f98a2SYan Zheng if (ret == 0 && start <= key.objectid) { 3914b3b4aa74SDavid Sterba btrfs_release_path(path); 39155d4f98a2SYan Zheng rc->search_start = end + 1; 39165d4f98a2SYan Zheng } else { 39173173a18fSJosef Bacik if (key.type == BTRFS_EXTENT_ITEM_KEY) 39185d4f98a2SYan Zheng rc->search_start = key.objectid + key.offset; 39193173a18fSJosef Bacik else 39203173a18fSJosef Bacik rc->search_start = key.objectid + 39210b246afaSJeff Mahoney fs_info->nodesize; 39223fd0a558SYan, Zheng memcpy(extent_key, &key, sizeof(key)); 39235d4f98a2SYan Zheng return 0; 39245d4f98a2SYan Zheng } 39255d4f98a2SYan Zheng } 3926b3b4aa74SDavid Sterba btrfs_release_path(path); 39275d4f98a2SYan Zheng return ret; 39285d4f98a2SYan Zheng } 39295d4f98a2SYan Zheng 39305d4f98a2SYan Zheng static void set_reloc_control(struct reloc_control *rc) 39315d4f98a2SYan Zheng { 39325d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 39337585717fSChris Mason 39347585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 39355d4f98a2SYan Zheng fs_info->reloc_ctl = rc; 39367585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 39375d4f98a2SYan Zheng } 39385d4f98a2SYan Zheng 39395d4f98a2SYan Zheng static void unset_reloc_control(struct reloc_control *rc) 39405d4f98a2SYan Zheng { 39415d4f98a2SYan Zheng struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 39427585717fSChris Mason 39437585717fSChris Mason mutex_lock(&fs_info->reloc_mutex); 39445d4f98a2SYan Zheng fs_info->reloc_ctl = NULL; 39457585717fSChris Mason mutex_unlock(&fs_info->reloc_mutex); 39465d4f98a2SYan Zheng } 39475d4f98a2SYan Zheng 39485d4f98a2SYan Zheng static int check_extent_flags(u64 flags) 39495d4f98a2SYan Zheng { 39505d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 39515d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 39525d4f98a2SYan Zheng return 1; 39535d4f98a2SYan Zheng if (!(flags & BTRFS_EXTENT_FLAG_DATA) && 39545d4f98a2SYan Zheng !(flags & BTRFS_EXTENT_FLAG_TREE_BLOCK)) 39555d4f98a2SYan Zheng return 1; 39565d4f98a2SYan Zheng if ((flags & BTRFS_EXTENT_FLAG_DATA) && 39575d4f98a2SYan Zheng (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) 39585d4f98a2SYan Zheng return 1; 39595d4f98a2SYan Zheng return 0; 39605d4f98a2SYan Zheng } 39615d4f98a2SYan Zheng 39623fd0a558SYan, Zheng static noinline_for_stack 39633fd0a558SYan, Zheng int prepare_to_relocate(struct reloc_control *rc) 39643fd0a558SYan, Zheng { 39653fd0a558SYan, Zheng struct btrfs_trans_handle *trans; 3966ac2fabacSJosef Bacik int ret; 39673fd0a558SYan, Zheng 39682ff7e61eSJeff Mahoney rc->block_rsv = btrfs_alloc_block_rsv(rc->extent_root->fs_info, 396966d8f3ddSMiao Xie BTRFS_BLOCK_RSV_TEMP); 39703fd0a558SYan, Zheng if (!rc->block_rsv) 39713fd0a558SYan, Zheng return -ENOMEM; 39723fd0a558SYan, Zheng 39733fd0a558SYan, Zheng memset(&rc->cluster, 0, sizeof(rc->cluster)); 39743fd0a558SYan, Zheng rc->search_start = rc->block_group->key.objectid; 39753fd0a558SYan, Zheng rc->extents_found = 0; 39763fd0a558SYan, Zheng rc->nodes_relocated = 0; 39773fd0a558SYan, Zheng rc->merging_rsv_size = 0; 39780647bf56SWang Shilong rc->reserved_bytes = 0; 3979da17066cSJeff Mahoney rc->block_rsv->size = rc->extent_root->fs_info->nodesize * 39800647bf56SWang Shilong RELOCATION_RESERVED_NODES; 3981ac2fabacSJosef Bacik ret = btrfs_block_rsv_refill(rc->extent_root, 3982ac2fabacSJosef Bacik rc->block_rsv, rc->block_rsv->size, 3983ac2fabacSJosef Bacik BTRFS_RESERVE_FLUSH_ALL); 3984ac2fabacSJosef Bacik if (ret) 3985ac2fabacSJosef Bacik return ret; 39863fd0a558SYan, Zheng 39873fd0a558SYan, Zheng rc->create_reloc_tree = 1; 39883fd0a558SYan, Zheng set_reloc_control(rc); 39893fd0a558SYan, Zheng 39907a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 399128818947SLiu Bo if (IS_ERR(trans)) { 399228818947SLiu Bo unset_reloc_control(rc); 399328818947SLiu Bo /* 399428818947SLiu Bo * extent tree is not a ref_cow tree and has no reloc_root to 399528818947SLiu Bo * cleanup. And callers are responsible to free the above 399628818947SLiu Bo * block rsv. 399728818947SLiu Bo */ 399828818947SLiu Bo return PTR_ERR(trans); 399928818947SLiu Bo } 40003a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 40013fd0a558SYan, Zheng return 0; 40023fd0a558SYan, Zheng } 400376dda93cSYan, Zheng 40045d4f98a2SYan Zheng static noinline_for_stack int relocate_block_group(struct reloc_control *rc) 40055d4f98a2SYan Zheng { 40062ff7e61eSJeff Mahoney struct btrfs_fs_info *fs_info = rc->extent_root->fs_info; 40075d4f98a2SYan Zheng struct rb_root blocks = RB_ROOT; 40085d4f98a2SYan Zheng struct btrfs_key key; 40095d4f98a2SYan Zheng struct btrfs_trans_handle *trans = NULL; 40105d4f98a2SYan Zheng struct btrfs_path *path; 40115d4f98a2SYan Zheng struct btrfs_extent_item *ei; 40125d4f98a2SYan Zheng u64 flags; 40135d4f98a2SYan Zheng u32 item_size; 40145d4f98a2SYan Zheng int ret; 40155d4f98a2SYan Zheng int err = 0; 4016c87f08caSChris Mason int progress = 0; 40175d4f98a2SYan Zheng 40185d4f98a2SYan Zheng path = btrfs_alloc_path(); 40193fd0a558SYan, Zheng if (!path) 40205d4f98a2SYan Zheng return -ENOMEM; 4021e4058b54SDavid Sterba path->reada = READA_FORWARD; 40223fd0a558SYan, Zheng 40233fd0a558SYan, Zheng ret = prepare_to_relocate(rc); 40243fd0a558SYan, Zheng if (ret) { 40253fd0a558SYan, Zheng err = ret; 40263fd0a558SYan, Zheng goto out_free; 40272423fdfbSJiri Slaby } 40285d4f98a2SYan Zheng 40295d4f98a2SYan Zheng while (1) { 40300647bf56SWang Shilong rc->reserved_bytes = 0; 40310647bf56SWang Shilong ret = btrfs_block_rsv_refill(rc->extent_root, 40320647bf56SWang Shilong rc->block_rsv, rc->block_rsv->size, 40330647bf56SWang Shilong BTRFS_RESERVE_FLUSH_ALL); 40340647bf56SWang Shilong if (ret) { 40350647bf56SWang Shilong err = ret; 40360647bf56SWang Shilong break; 40370647bf56SWang Shilong } 4038c87f08caSChris Mason progress++; 4039a22285a6SYan, Zheng trans = btrfs_start_transaction(rc->extent_root, 0); 40400f788c58SLiu Bo if (IS_ERR(trans)) { 40410f788c58SLiu Bo err = PTR_ERR(trans); 40420f788c58SLiu Bo trans = NULL; 40430f788c58SLiu Bo break; 40440f788c58SLiu Bo } 4045c87f08caSChris Mason restart: 40463fd0a558SYan, Zheng if (update_backref_cache(trans, &rc->backref_cache)) { 40473a45bb20SJeff Mahoney btrfs_end_transaction(trans); 404842a657f5SPan Bian trans = NULL; 40493fd0a558SYan, Zheng continue; 40503fd0a558SYan, Zheng } 40513fd0a558SYan, Zheng 4052147d256eSZhaolei ret = find_next_extent(rc, path, &key); 40535d4f98a2SYan Zheng if (ret < 0) 40545d4f98a2SYan Zheng err = ret; 40555d4f98a2SYan Zheng if (ret != 0) 40565d4f98a2SYan Zheng break; 40575d4f98a2SYan Zheng 40585d4f98a2SYan Zheng rc->extents_found++; 40595d4f98a2SYan Zheng 40605d4f98a2SYan Zheng ei = btrfs_item_ptr(path->nodes[0], path->slots[0], 40615d4f98a2SYan Zheng struct btrfs_extent_item); 40623fd0a558SYan, Zheng item_size = btrfs_item_size_nr(path->nodes[0], path->slots[0]); 40635d4f98a2SYan Zheng if (item_size >= sizeof(*ei)) { 40645d4f98a2SYan Zheng flags = btrfs_extent_flags(path->nodes[0], ei); 40655d4f98a2SYan Zheng ret = check_extent_flags(flags); 40665d4f98a2SYan Zheng BUG_ON(ret); 40676d8ff4e4SDavid Sterba } else if (unlikely(item_size == sizeof(struct btrfs_extent_item_v0))) { 4068ba3c2b19SNikolay Borisov err = -EINVAL; 4069ba3c2b19SNikolay Borisov btrfs_print_v0_err(trans->fs_info); 4070ba3c2b19SNikolay Borisov btrfs_abort_transaction(trans, err); 4071ba3c2b19SNikolay Borisov break; 40725d4f98a2SYan Zheng } else { 40735d4f98a2SYan Zheng BUG(); 40745d4f98a2SYan Zheng } 40755d4f98a2SYan Zheng 40765d4f98a2SYan Zheng if (flags & BTRFS_EXTENT_FLAG_TREE_BLOCK) { 40775d4f98a2SYan Zheng ret = add_tree_block(rc, &key, path, &blocks); 40785d4f98a2SYan Zheng } else if (rc->stage == UPDATE_DATA_PTRS && 40795d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 40805d4f98a2SYan Zheng ret = add_data_references(rc, &key, path, &blocks); 40815d4f98a2SYan Zheng } else { 4082b3b4aa74SDavid Sterba btrfs_release_path(path); 40835d4f98a2SYan Zheng ret = 0; 40845d4f98a2SYan Zheng } 40855d4f98a2SYan Zheng if (ret < 0) { 40863fd0a558SYan, Zheng err = ret; 40875d4f98a2SYan Zheng break; 40885d4f98a2SYan Zheng } 40895d4f98a2SYan Zheng 40905d4f98a2SYan Zheng if (!RB_EMPTY_ROOT(&blocks)) { 40915d4f98a2SYan Zheng ret = relocate_tree_blocks(trans, rc, &blocks); 40925d4f98a2SYan Zheng if (ret < 0) { 40931708cc57SWang Shilong /* 40941708cc57SWang Shilong * if we fail to relocate tree blocks, force to update 40951708cc57SWang Shilong * backref cache when committing transaction. 40961708cc57SWang Shilong */ 40971708cc57SWang Shilong rc->backref_cache.last_trans = trans->transid - 1; 40981708cc57SWang Shilong 40993fd0a558SYan, Zheng if (ret != -EAGAIN) { 41005d4f98a2SYan Zheng err = ret; 41015d4f98a2SYan Zheng break; 41025d4f98a2SYan Zheng } 41033fd0a558SYan, Zheng rc->extents_found--; 41043fd0a558SYan, Zheng rc->search_start = key.objectid; 41053fd0a558SYan, Zheng } 41065d4f98a2SYan Zheng } 41075d4f98a2SYan Zheng 41083a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 41092ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 41103fd0a558SYan, Zheng trans = NULL; 41115d4f98a2SYan Zheng 41125d4f98a2SYan Zheng if (rc->stage == MOVE_DATA_EXTENTS && 41135d4f98a2SYan Zheng (flags & BTRFS_EXTENT_FLAG_DATA)) { 41145d4f98a2SYan Zheng rc->found_file_extent = 1; 41150257bb82SYan, Zheng ret = relocate_data_extent(rc->data_inode, 41163fd0a558SYan, Zheng &key, &rc->cluster); 41175d4f98a2SYan Zheng if (ret < 0) { 41185d4f98a2SYan Zheng err = ret; 41195d4f98a2SYan Zheng break; 41205d4f98a2SYan Zheng } 41215d4f98a2SYan Zheng } 41225d4f98a2SYan Zheng } 4123c87f08caSChris Mason if (trans && progress && err == -ENOSPC) { 412443a7e99dSNikolay Borisov ret = btrfs_force_chunk_alloc(trans, rc->block_group->flags); 41259689457bSShilong Wang if (ret == 1) { 4126c87f08caSChris Mason err = 0; 4127c87f08caSChris Mason progress = 0; 4128c87f08caSChris Mason goto restart; 4129c87f08caSChris Mason } 4130c87f08caSChris Mason } 41313fd0a558SYan, Zheng 4132b3b4aa74SDavid Sterba btrfs_release_path(path); 413391166212SDavid Sterba clear_extent_bits(&rc->processed_blocks, 0, (u64)-1, EXTENT_DIRTY); 41345d4f98a2SYan Zheng 41355d4f98a2SYan Zheng if (trans) { 41363a45bb20SJeff Mahoney btrfs_end_transaction_throttle(trans); 41372ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 41385d4f98a2SYan Zheng } 41395d4f98a2SYan Zheng 41400257bb82SYan, Zheng if (!err) { 41413fd0a558SYan, Zheng ret = relocate_file_extent_cluster(rc->data_inode, 41423fd0a558SYan, Zheng &rc->cluster); 41430257bb82SYan, Zheng if (ret < 0) 41440257bb82SYan, Zheng err = ret; 41450257bb82SYan, Zheng } 41460257bb82SYan, Zheng 41473fd0a558SYan, Zheng rc->create_reloc_tree = 0; 41483fd0a558SYan, Zheng set_reloc_control(rc); 41490257bb82SYan, Zheng 41503fd0a558SYan, Zheng backref_cache_cleanup(&rc->backref_cache); 41512ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1); 41525d4f98a2SYan Zheng 41533fd0a558SYan, Zheng err = prepare_to_merge(rc, err); 41545d4f98a2SYan Zheng 41555d4f98a2SYan Zheng merge_reloc_roots(rc); 41565d4f98a2SYan Zheng 41573fd0a558SYan, Zheng rc->merge_reloc_tree = 0; 41585d4f98a2SYan Zheng unset_reloc_control(rc); 41592ff7e61eSJeff Mahoney btrfs_block_rsv_release(fs_info, rc->block_rsv, (u64)-1); 41605d4f98a2SYan Zheng 41615d4f98a2SYan Zheng /* get rid of pinned extents */ 41627a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 416362b99540SQu Wenruo if (IS_ERR(trans)) { 41643612b495STsutomu Itoh err = PTR_ERR(trans); 416562b99540SQu Wenruo goto out_free; 416662b99540SQu Wenruo } 41673a45bb20SJeff Mahoney btrfs_commit_transaction(trans); 4168d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 4169d2311e69SQu Wenruo if (ret < 0 && !err) 4170d2311e69SQu Wenruo err = ret; 41713fd0a558SYan, Zheng out_free: 41722ff7e61eSJeff Mahoney btrfs_free_block_rsv(fs_info, rc->block_rsv); 41733fd0a558SYan, Zheng btrfs_free_path(path); 41745d4f98a2SYan Zheng return err; 41755d4f98a2SYan Zheng } 41765d4f98a2SYan Zheng 41775d4f98a2SYan Zheng static int __insert_orphan_inode(struct btrfs_trans_handle *trans, 41780257bb82SYan, Zheng struct btrfs_root *root, u64 objectid) 41795d4f98a2SYan Zheng { 41805d4f98a2SYan Zheng struct btrfs_path *path; 41815d4f98a2SYan Zheng struct btrfs_inode_item *item; 41825d4f98a2SYan Zheng struct extent_buffer *leaf; 41835d4f98a2SYan Zheng int ret; 41845d4f98a2SYan Zheng 41855d4f98a2SYan Zheng path = btrfs_alloc_path(); 41865d4f98a2SYan Zheng if (!path) 41875d4f98a2SYan Zheng return -ENOMEM; 41885d4f98a2SYan Zheng 41895d4f98a2SYan Zheng ret = btrfs_insert_empty_inode(trans, root, path, objectid); 41905d4f98a2SYan Zheng if (ret) 41915d4f98a2SYan Zheng goto out; 41925d4f98a2SYan Zheng 41935d4f98a2SYan Zheng leaf = path->nodes[0]; 41945d4f98a2SYan Zheng item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_inode_item); 4195b159fa28SDavid Sterba memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); 41965d4f98a2SYan Zheng btrfs_set_inode_generation(leaf, item, 1); 41970257bb82SYan, Zheng btrfs_set_inode_size(leaf, item, 0); 41985d4f98a2SYan Zheng btrfs_set_inode_mode(leaf, item, S_IFREG | 0600); 41993fd0a558SYan, Zheng btrfs_set_inode_flags(leaf, item, BTRFS_INODE_NOCOMPRESS | 42003fd0a558SYan, Zheng BTRFS_INODE_PREALLOC); 42015d4f98a2SYan Zheng btrfs_mark_buffer_dirty(leaf); 42025d4f98a2SYan Zheng out: 42035d4f98a2SYan Zheng btrfs_free_path(path); 42045d4f98a2SYan Zheng return ret; 42055d4f98a2SYan Zheng } 42065d4f98a2SYan Zheng 42075d4f98a2SYan Zheng /* 42085d4f98a2SYan Zheng * helper to create inode for data relocation. 42095d4f98a2SYan Zheng * the inode is in data relocation tree and its link count is 0 42105d4f98a2SYan Zheng */ 42113fd0a558SYan, Zheng static noinline_for_stack 42123fd0a558SYan, Zheng struct inode *create_reloc_inode(struct btrfs_fs_info *fs_info, 42135d4f98a2SYan Zheng struct btrfs_block_group_cache *group) 42145d4f98a2SYan Zheng { 42155d4f98a2SYan Zheng struct inode *inode = NULL; 42165d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 42175d4f98a2SYan Zheng struct btrfs_root *root; 42185d4f98a2SYan Zheng struct btrfs_key key; 42194624900dSZhaolei u64 objectid; 42205d4f98a2SYan Zheng int err = 0; 42215d4f98a2SYan Zheng 42225d4f98a2SYan Zheng root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 42235d4f98a2SYan Zheng if (IS_ERR(root)) 42245d4f98a2SYan Zheng return ERR_CAST(root); 42255d4f98a2SYan Zheng 4226a22285a6SYan, Zheng trans = btrfs_start_transaction(root, 6); 42273fd0a558SYan, Zheng if (IS_ERR(trans)) 42283fd0a558SYan, Zheng return ERR_CAST(trans); 42295d4f98a2SYan Zheng 4230581bb050SLi Zefan err = btrfs_find_free_objectid(root, &objectid); 42315d4f98a2SYan Zheng if (err) 42325d4f98a2SYan Zheng goto out; 42335d4f98a2SYan Zheng 42340257bb82SYan, Zheng err = __insert_orphan_inode(trans, root, objectid); 42355d4f98a2SYan Zheng BUG_ON(err); 42365d4f98a2SYan Zheng 42375d4f98a2SYan Zheng key.objectid = objectid; 42385d4f98a2SYan Zheng key.type = BTRFS_INODE_ITEM_KEY; 42395d4f98a2SYan Zheng key.offset = 0; 42400b246afaSJeff Mahoney inode = btrfs_iget(fs_info->sb, &key, root, NULL); 42412e19f1f9SAl Viro BUG_ON(IS_ERR(inode)); 42425d4f98a2SYan Zheng BTRFS_I(inode)->index_cnt = group->key.objectid; 42435d4f98a2SYan Zheng 424473f2e545SNikolay Borisov err = btrfs_orphan_add(trans, BTRFS_I(inode)); 42455d4f98a2SYan Zheng out: 42463a45bb20SJeff Mahoney btrfs_end_transaction(trans); 42472ff7e61eSJeff Mahoney btrfs_btree_balance_dirty(fs_info); 42485d4f98a2SYan Zheng if (err) { 42495d4f98a2SYan Zheng if (inode) 42505d4f98a2SYan Zheng iput(inode); 42515d4f98a2SYan Zheng inode = ERR_PTR(err); 42525d4f98a2SYan Zheng } 42535d4f98a2SYan Zheng return inode; 42545d4f98a2SYan Zheng } 42555d4f98a2SYan Zheng 4256c258d6e3SQu Wenruo static struct reloc_control *alloc_reloc_control(struct btrfs_fs_info *fs_info) 42573fd0a558SYan, Zheng { 42583fd0a558SYan, Zheng struct reloc_control *rc; 42593fd0a558SYan, Zheng 42603fd0a558SYan, Zheng rc = kzalloc(sizeof(*rc), GFP_NOFS); 42613fd0a558SYan, Zheng if (!rc) 42623fd0a558SYan, Zheng return NULL; 42633fd0a558SYan, Zheng 42643fd0a558SYan, Zheng INIT_LIST_HEAD(&rc->reloc_roots); 4265d2311e69SQu Wenruo INIT_LIST_HEAD(&rc->dirty_subvol_roots); 42663fd0a558SYan, Zheng backref_cache_init(&rc->backref_cache); 42673fd0a558SYan, Zheng mapping_tree_init(&rc->reloc_root_tree); 426843eb5f29SQu Wenruo extent_io_tree_init(fs_info, &rc->processed_blocks, 426943eb5f29SQu Wenruo IO_TREE_RELOC_BLOCKS, NULL); 42703fd0a558SYan, Zheng return rc; 42713fd0a558SYan, Zheng } 42723fd0a558SYan, Zheng 42735d4f98a2SYan Zheng /* 4274ebce0e01SAdam Borowski * Print the block group being relocated 4275ebce0e01SAdam Borowski */ 4276ebce0e01SAdam Borowski static void describe_relocation(struct btrfs_fs_info *fs_info, 4277ebce0e01SAdam Borowski struct btrfs_block_group_cache *block_group) 4278ebce0e01SAdam Borowski { 4279f89e09cfSAnand Jain char buf[128] = {'\0'}; 4280ebce0e01SAdam Borowski 4281f89e09cfSAnand Jain btrfs_describe_block_groups(block_group->flags, buf, sizeof(buf)); 4282ebce0e01SAdam Borowski 4283ebce0e01SAdam Borowski btrfs_info(fs_info, 4284ebce0e01SAdam Borowski "relocating block group %llu flags %s", 4285f89e09cfSAnand Jain block_group->key.objectid, buf); 4286ebce0e01SAdam Borowski } 4287ebce0e01SAdam Borowski 4288ebce0e01SAdam Borowski /* 42895d4f98a2SYan Zheng * function to relocate all extents in a block group. 42905d4f98a2SYan Zheng */ 42916bccf3abSJeff Mahoney int btrfs_relocate_block_group(struct btrfs_fs_info *fs_info, u64 group_start) 42925d4f98a2SYan Zheng { 4293eede2bf3SOmar Sandoval struct btrfs_block_group_cache *bg; 42946bccf3abSJeff Mahoney struct btrfs_root *extent_root = fs_info->extent_root; 42955d4f98a2SYan Zheng struct reloc_control *rc; 42960af3d00bSJosef Bacik struct inode *inode; 42970af3d00bSJosef Bacik struct btrfs_path *path; 42985d4f98a2SYan Zheng int ret; 4299f0486c68SYan, Zheng int rw = 0; 43005d4f98a2SYan Zheng int err = 0; 43015d4f98a2SYan Zheng 4302eede2bf3SOmar Sandoval bg = btrfs_lookup_block_group(fs_info, group_start); 4303eede2bf3SOmar Sandoval if (!bg) 4304eede2bf3SOmar Sandoval return -ENOENT; 4305eede2bf3SOmar Sandoval 4306eede2bf3SOmar Sandoval if (btrfs_pinned_by_swapfile(fs_info, bg)) { 4307eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 4308eede2bf3SOmar Sandoval return -ETXTBSY; 4309eede2bf3SOmar Sandoval } 4310eede2bf3SOmar Sandoval 4311c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 4312eede2bf3SOmar Sandoval if (!rc) { 4313eede2bf3SOmar Sandoval btrfs_put_block_group(bg); 43145d4f98a2SYan Zheng return -ENOMEM; 4315eede2bf3SOmar Sandoval } 43165d4f98a2SYan Zheng 4317f0486c68SYan, Zheng rc->extent_root = extent_root; 4318eede2bf3SOmar Sandoval rc->block_group = bg; 43195d4f98a2SYan Zheng 4320c83488afSNikolay Borisov ret = btrfs_inc_block_group_ro(rc->block_group); 4321f0486c68SYan, Zheng if (ret) { 4322f0486c68SYan, Zheng err = ret; 4323f0486c68SYan, Zheng goto out; 4324f0486c68SYan, Zheng } 4325f0486c68SYan, Zheng rw = 1; 4326f0486c68SYan, Zheng 43270af3d00bSJosef Bacik path = btrfs_alloc_path(); 43280af3d00bSJosef Bacik if (!path) { 43290af3d00bSJosef Bacik err = -ENOMEM; 43300af3d00bSJosef Bacik goto out; 43310af3d00bSJosef Bacik } 43320af3d00bSJosef Bacik 43337949f339SDavid Sterba inode = lookup_free_space_inode(rc->block_group, path); 43340af3d00bSJosef Bacik btrfs_free_path(path); 43350af3d00bSJosef Bacik 43360af3d00bSJosef Bacik if (!IS_ERR(inode)) 43371bbc621eSChris Mason ret = delete_block_group_cache(fs_info, rc->block_group, inode, 0); 43380af3d00bSJosef Bacik else 43390af3d00bSJosef Bacik ret = PTR_ERR(inode); 43400af3d00bSJosef Bacik 43410af3d00bSJosef Bacik if (ret && ret != -ENOENT) { 43420af3d00bSJosef Bacik err = ret; 43430af3d00bSJosef Bacik goto out; 43440af3d00bSJosef Bacik } 43450af3d00bSJosef Bacik 43465d4f98a2SYan Zheng rc->data_inode = create_reloc_inode(fs_info, rc->block_group); 43475d4f98a2SYan Zheng if (IS_ERR(rc->data_inode)) { 43485d4f98a2SYan Zheng err = PTR_ERR(rc->data_inode); 43495d4f98a2SYan Zheng rc->data_inode = NULL; 43505d4f98a2SYan Zheng goto out; 43515d4f98a2SYan Zheng } 43525d4f98a2SYan Zheng 43530b246afaSJeff Mahoney describe_relocation(fs_info, rc->block_group); 43545d4f98a2SYan Zheng 43559cfa3e34SFilipe Manana btrfs_wait_block_group_reservations(rc->block_group); 4356f78c436cSFilipe Manana btrfs_wait_nocow_writers(rc->block_group); 43576374e57aSChris Mason btrfs_wait_ordered_roots(fs_info, U64_MAX, 4358578def7cSFilipe Manana rc->block_group->key.objectid, 4359578def7cSFilipe Manana rc->block_group->key.offset); 43605d4f98a2SYan Zheng 43615d4f98a2SYan Zheng while (1) { 436276dda93cSYan, Zheng mutex_lock(&fs_info->cleaner_mutex); 43635d4f98a2SYan Zheng ret = relocate_block_group(rc); 436476dda93cSYan, Zheng mutex_unlock(&fs_info->cleaner_mutex); 4365ff612ba7SJosef Bacik if (ret < 0) 43665d4f98a2SYan Zheng err = ret; 4367ff612ba7SJosef Bacik 4368ff612ba7SJosef Bacik /* 4369ff612ba7SJosef Bacik * We may have gotten ENOSPC after we already dirtied some 4370ff612ba7SJosef Bacik * extents. If writeout happens while we're relocating a 4371ff612ba7SJosef Bacik * different block group we could end up hitting the 4372ff612ba7SJosef Bacik * BUG_ON(rc->stage == UPDATE_DATA_PTRS) in 4373ff612ba7SJosef Bacik * btrfs_reloc_cow_block. Make sure we write everything out 4374ff612ba7SJosef Bacik * properly so we don't trip over this problem, and then break 4375ff612ba7SJosef Bacik * out of the loop if we hit an error. 4376ff612ba7SJosef Bacik */ 4377ff612ba7SJosef Bacik if (rc->stage == MOVE_DATA_EXTENTS && rc->found_file_extent) { 4378ff612ba7SJosef Bacik ret = btrfs_wait_ordered_range(rc->data_inode, 0, 4379ff612ba7SJosef Bacik (u64)-1); 4380ff612ba7SJosef Bacik if (ret) 4381ff612ba7SJosef Bacik err = ret; 4382ff612ba7SJosef Bacik invalidate_mapping_pages(rc->data_inode->i_mapping, 4383ff612ba7SJosef Bacik 0, -1); 4384ff612ba7SJosef Bacik rc->stage = UPDATE_DATA_PTRS; 43855d4f98a2SYan Zheng } 43865d4f98a2SYan Zheng 4387ff612ba7SJosef Bacik if (err < 0) 4388ff612ba7SJosef Bacik goto out; 4389ff612ba7SJosef Bacik 43905d4f98a2SYan Zheng if (rc->extents_found == 0) 43915d4f98a2SYan Zheng break; 43925d4f98a2SYan Zheng 43930b246afaSJeff Mahoney btrfs_info(fs_info, "found %llu extents", rc->extents_found); 43945d4f98a2SYan Zheng 43955d4f98a2SYan Zheng } 43965d4f98a2SYan Zheng 43975d4f98a2SYan Zheng WARN_ON(rc->block_group->pinned > 0); 43985d4f98a2SYan Zheng WARN_ON(rc->block_group->reserved > 0); 43995d4f98a2SYan Zheng WARN_ON(btrfs_block_group_used(&rc->block_group->item) > 0); 44005d4f98a2SYan Zheng out: 4401f0486c68SYan, Zheng if (err && rw) 44022ff7e61eSJeff Mahoney btrfs_dec_block_group_ro(rc->block_group); 44035d4f98a2SYan Zheng iput(rc->data_inode); 44045d4f98a2SYan Zheng btrfs_put_block_group(rc->block_group); 44055d4f98a2SYan Zheng kfree(rc); 44065d4f98a2SYan Zheng return err; 44075d4f98a2SYan Zheng } 44085d4f98a2SYan Zheng 440976dda93cSYan, Zheng static noinline_for_stack int mark_garbage_root(struct btrfs_root *root) 441076dda93cSYan, Zheng { 44110b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 441276dda93cSYan, Zheng struct btrfs_trans_handle *trans; 441379787eaaSJeff Mahoney int ret, err; 441476dda93cSYan, Zheng 44150b246afaSJeff Mahoney trans = btrfs_start_transaction(fs_info->tree_root, 0); 441679787eaaSJeff Mahoney if (IS_ERR(trans)) 441779787eaaSJeff Mahoney return PTR_ERR(trans); 441876dda93cSYan, Zheng 441976dda93cSYan, Zheng memset(&root->root_item.drop_progress, 0, 442076dda93cSYan, Zheng sizeof(root->root_item.drop_progress)); 442176dda93cSYan, Zheng root->root_item.drop_level = 0; 442276dda93cSYan, Zheng btrfs_set_root_refs(&root->root_item, 0); 44230b246afaSJeff Mahoney ret = btrfs_update_root(trans, fs_info->tree_root, 442476dda93cSYan, Zheng &root->root_key, &root->root_item); 442576dda93cSYan, Zheng 44263a45bb20SJeff Mahoney err = btrfs_end_transaction(trans); 442779787eaaSJeff Mahoney if (err) 442879787eaaSJeff Mahoney return err; 442979787eaaSJeff Mahoney return ret; 443076dda93cSYan, Zheng } 443176dda93cSYan, Zheng 44325d4f98a2SYan Zheng /* 44335d4f98a2SYan Zheng * recover relocation interrupted by system crash. 44345d4f98a2SYan Zheng * 44355d4f98a2SYan Zheng * this function resumes merging reloc trees with corresponding fs trees. 44365d4f98a2SYan Zheng * this is important for keeping the sharing of tree blocks 44375d4f98a2SYan Zheng */ 44385d4f98a2SYan Zheng int btrfs_recover_relocation(struct btrfs_root *root) 44395d4f98a2SYan Zheng { 44400b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 44415d4f98a2SYan Zheng LIST_HEAD(reloc_roots); 44425d4f98a2SYan Zheng struct btrfs_key key; 44435d4f98a2SYan Zheng struct btrfs_root *fs_root; 44445d4f98a2SYan Zheng struct btrfs_root *reloc_root; 44455d4f98a2SYan Zheng struct btrfs_path *path; 44465d4f98a2SYan Zheng struct extent_buffer *leaf; 44475d4f98a2SYan Zheng struct reloc_control *rc = NULL; 44485d4f98a2SYan Zheng struct btrfs_trans_handle *trans; 44495d4f98a2SYan Zheng int ret; 44505d4f98a2SYan Zheng int err = 0; 44515d4f98a2SYan Zheng 44525d4f98a2SYan Zheng path = btrfs_alloc_path(); 44535d4f98a2SYan Zheng if (!path) 44545d4f98a2SYan Zheng return -ENOMEM; 4455e4058b54SDavid Sterba path->reada = READA_BACK; 44565d4f98a2SYan Zheng 44575d4f98a2SYan Zheng key.objectid = BTRFS_TREE_RELOC_OBJECTID; 44585d4f98a2SYan Zheng key.type = BTRFS_ROOT_ITEM_KEY; 44595d4f98a2SYan Zheng key.offset = (u64)-1; 44605d4f98a2SYan Zheng 44615d4f98a2SYan Zheng while (1) { 44620b246afaSJeff Mahoney ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, 44635d4f98a2SYan Zheng path, 0, 0); 44645d4f98a2SYan Zheng if (ret < 0) { 44655d4f98a2SYan Zheng err = ret; 44665d4f98a2SYan Zheng goto out; 44675d4f98a2SYan Zheng } 44685d4f98a2SYan Zheng if (ret > 0) { 44695d4f98a2SYan Zheng if (path->slots[0] == 0) 44705d4f98a2SYan Zheng break; 44715d4f98a2SYan Zheng path->slots[0]--; 44725d4f98a2SYan Zheng } 44735d4f98a2SYan Zheng leaf = path->nodes[0]; 44745d4f98a2SYan Zheng btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); 4475b3b4aa74SDavid Sterba btrfs_release_path(path); 44765d4f98a2SYan Zheng 44775d4f98a2SYan Zheng if (key.objectid != BTRFS_TREE_RELOC_OBJECTID || 44785d4f98a2SYan Zheng key.type != BTRFS_ROOT_ITEM_KEY) 44795d4f98a2SYan Zheng break; 44805d4f98a2SYan Zheng 4481cb517eabSMiao Xie reloc_root = btrfs_read_fs_root(root, &key); 44825d4f98a2SYan Zheng if (IS_ERR(reloc_root)) { 44835d4f98a2SYan Zheng err = PTR_ERR(reloc_root); 44845d4f98a2SYan Zheng goto out; 44855d4f98a2SYan Zheng } 44865d4f98a2SYan Zheng 44875d4f98a2SYan Zheng list_add(&reloc_root->root_list, &reloc_roots); 44885d4f98a2SYan Zheng 44895d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) > 0) { 44900b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, 44915d4f98a2SYan Zheng reloc_root->root_key.offset); 44925d4f98a2SYan Zheng if (IS_ERR(fs_root)) { 449376dda93cSYan, Zheng ret = PTR_ERR(fs_root); 449476dda93cSYan, Zheng if (ret != -ENOENT) { 449576dda93cSYan, Zheng err = ret; 44965d4f98a2SYan Zheng goto out; 44975d4f98a2SYan Zheng } 449879787eaaSJeff Mahoney ret = mark_garbage_root(reloc_root); 449979787eaaSJeff Mahoney if (ret < 0) { 450079787eaaSJeff Mahoney err = ret; 450179787eaaSJeff Mahoney goto out; 450279787eaaSJeff Mahoney } 450376dda93cSYan, Zheng } 45045d4f98a2SYan Zheng } 45055d4f98a2SYan Zheng 45065d4f98a2SYan Zheng if (key.offset == 0) 45075d4f98a2SYan Zheng break; 45085d4f98a2SYan Zheng 45095d4f98a2SYan Zheng key.offset--; 45105d4f98a2SYan Zheng } 4511b3b4aa74SDavid Sterba btrfs_release_path(path); 45125d4f98a2SYan Zheng 45135d4f98a2SYan Zheng if (list_empty(&reloc_roots)) 45145d4f98a2SYan Zheng goto out; 45155d4f98a2SYan Zheng 4516c258d6e3SQu Wenruo rc = alloc_reloc_control(fs_info); 45175d4f98a2SYan Zheng if (!rc) { 45185d4f98a2SYan Zheng err = -ENOMEM; 45195d4f98a2SYan Zheng goto out; 45205d4f98a2SYan Zheng } 45215d4f98a2SYan Zheng 45220b246afaSJeff Mahoney rc->extent_root = fs_info->extent_root; 45235d4f98a2SYan Zheng 45245d4f98a2SYan Zheng set_reloc_control(rc); 45255d4f98a2SYan Zheng 45267a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 45273612b495STsutomu Itoh if (IS_ERR(trans)) { 45283612b495STsutomu Itoh unset_reloc_control(rc); 45293612b495STsutomu Itoh err = PTR_ERR(trans); 45303612b495STsutomu Itoh goto out_free; 45313612b495STsutomu Itoh } 45323fd0a558SYan, Zheng 45333fd0a558SYan, Zheng rc->merge_reloc_tree = 1; 45343fd0a558SYan, Zheng 45355d4f98a2SYan Zheng while (!list_empty(&reloc_roots)) { 45365d4f98a2SYan Zheng reloc_root = list_entry(reloc_roots.next, 45375d4f98a2SYan Zheng struct btrfs_root, root_list); 45385d4f98a2SYan Zheng list_del(&reloc_root->root_list); 45395d4f98a2SYan Zheng 45405d4f98a2SYan Zheng if (btrfs_root_refs(&reloc_root->root_item) == 0) { 45415d4f98a2SYan Zheng list_add_tail(&reloc_root->root_list, 45425d4f98a2SYan Zheng &rc->reloc_roots); 45435d4f98a2SYan Zheng continue; 45445d4f98a2SYan Zheng } 45455d4f98a2SYan Zheng 45460b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, reloc_root->root_key.offset); 454779787eaaSJeff Mahoney if (IS_ERR(fs_root)) { 454879787eaaSJeff Mahoney err = PTR_ERR(fs_root); 454979787eaaSJeff Mahoney goto out_free; 455079787eaaSJeff Mahoney } 45515d4f98a2SYan Zheng 4552ffd7b339SJeff Mahoney err = __add_reloc_root(reloc_root); 455379787eaaSJeff Mahoney BUG_ON(err < 0); /* -ENOMEM or logic error */ 45545d4f98a2SYan Zheng fs_root->reloc_root = reloc_root; 45555d4f98a2SYan Zheng } 45565d4f98a2SYan Zheng 45573a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 455879787eaaSJeff Mahoney if (err) 455979787eaaSJeff Mahoney goto out_free; 45605d4f98a2SYan Zheng 45615d4f98a2SYan Zheng merge_reloc_roots(rc); 45625d4f98a2SYan Zheng 45635d4f98a2SYan Zheng unset_reloc_control(rc); 45645d4f98a2SYan Zheng 45657a7eaa40SJosef Bacik trans = btrfs_join_transaction(rc->extent_root); 456662b99540SQu Wenruo if (IS_ERR(trans)) { 45673612b495STsutomu Itoh err = PTR_ERR(trans); 456862b99540SQu Wenruo goto out_free; 456962b99540SQu Wenruo } 45703a45bb20SJeff Mahoney err = btrfs_commit_transaction(trans); 4571d2311e69SQu Wenruo 4572d2311e69SQu Wenruo ret = clean_dirty_subvols(rc); 4573d2311e69SQu Wenruo if (ret < 0 && !err) 4574d2311e69SQu Wenruo err = ret; 45753612b495STsutomu Itoh out_free: 45765d4f98a2SYan Zheng kfree(rc); 45773612b495STsutomu Itoh out: 4578aca1bba6SLiu Bo if (!list_empty(&reloc_roots)) 4579aca1bba6SLiu Bo free_reloc_roots(&reloc_roots); 4580aca1bba6SLiu Bo 45815d4f98a2SYan Zheng btrfs_free_path(path); 45825d4f98a2SYan Zheng 45835d4f98a2SYan Zheng if (err == 0) { 45845d4f98a2SYan Zheng /* cleanup orphan inode in data relocation tree */ 45850b246afaSJeff Mahoney fs_root = read_fs_root(fs_info, BTRFS_DATA_RELOC_TREE_OBJECTID); 45865d4f98a2SYan Zheng if (IS_ERR(fs_root)) 45875d4f98a2SYan Zheng err = PTR_ERR(fs_root); 4588d7ce5843SMiao Xie else 458966b4ffd1SJosef Bacik err = btrfs_orphan_cleanup(fs_root); 45905d4f98a2SYan Zheng } 45915d4f98a2SYan Zheng return err; 45925d4f98a2SYan Zheng } 45935d4f98a2SYan Zheng 45945d4f98a2SYan Zheng /* 45955d4f98a2SYan Zheng * helper to add ordered checksum for data relocation. 45965d4f98a2SYan Zheng * 45975d4f98a2SYan Zheng * cloning checksum properly handles the nodatasum extents. 45985d4f98a2SYan Zheng * it also saves CPU time to re-calculate the checksum. 45995d4f98a2SYan Zheng */ 46005d4f98a2SYan Zheng int btrfs_reloc_clone_csums(struct inode *inode, u64 file_pos, u64 len) 46015d4f98a2SYan Zheng { 46020b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = btrfs_sb(inode->i_sb); 46035d4f98a2SYan Zheng struct btrfs_ordered_sum *sums; 46045d4f98a2SYan Zheng struct btrfs_ordered_extent *ordered; 46055d4f98a2SYan Zheng int ret; 46065d4f98a2SYan Zheng u64 disk_bytenr; 46074577b014SJosef Bacik u64 new_bytenr; 46085d4f98a2SYan Zheng LIST_HEAD(list); 46095d4f98a2SYan Zheng 46105d4f98a2SYan Zheng ordered = btrfs_lookup_ordered_extent(inode, file_pos); 46115d4f98a2SYan Zheng BUG_ON(ordered->file_offset != file_pos || ordered->len != len); 46125d4f98a2SYan Zheng 46135d4f98a2SYan Zheng disk_bytenr = file_pos + BTRFS_I(inode)->index_cnt; 46140b246afaSJeff Mahoney ret = btrfs_lookup_csums_range(fs_info->csum_root, disk_bytenr, 4615a2de733cSArne Jansen disk_bytenr + len - 1, &list, 0); 461679787eaaSJeff Mahoney if (ret) 461779787eaaSJeff Mahoney goto out; 46185d4f98a2SYan Zheng 46195d4f98a2SYan Zheng while (!list_empty(&list)) { 46205d4f98a2SYan Zheng sums = list_entry(list.next, struct btrfs_ordered_sum, list); 46215d4f98a2SYan Zheng list_del_init(&sums->list); 46225d4f98a2SYan Zheng 46234577b014SJosef Bacik /* 46244577b014SJosef Bacik * We need to offset the new_bytenr based on where the csum is. 46254577b014SJosef Bacik * We need to do this because we will read in entire prealloc 46264577b014SJosef Bacik * extents but we may have written to say the middle of the 46274577b014SJosef Bacik * prealloc extent, so we need to make sure the csum goes with 46284577b014SJosef Bacik * the right disk offset. 46294577b014SJosef Bacik * 46304577b014SJosef Bacik * We can do this because the data reloc inode refers strictly 46314577b014SJosef Bacik * to the on disk bytes, so we don't have to worry about 46324577b014SJosef Bacik * disk_len vs real len like with real inodes since it's all 46334577b014SJosef Bacik * disk length. 46344577b014SJosef Bacik */ 46354577b014SJosef Bacik new_bytenr = ordered->start + (sums->bytenr - disk_bytenr); 46364577b014SJosef Bacik sums->bytenr = new_bytenr; 46375d4f98a2SYan Zheng 4638f9756261SNikolay Borisov btrfs_add_ordered_sum(ordered, sums); 46395d4f98a2SYan Zheng } 464079787eaaSJeff Mahoney out: 46415d4f98a2SYan Zheng btrfs_put_ordered_extent(ordered); 4642411fc6bcSAndi Kleen return ret; 46435d4f98a2SYan Zheng } 46443fd0a558SYan, Zheng 464583d4cfd4SJosef Bacik int btrfs_reloc_cow_block(struct btrfs_trans_handle *trans, 46463fd0a558SYan, Zheng struct btrfs_root *root, struct extent_buffer *buf, 46473fd0a558SYan, Zheng struct extent_buffer *cow) 46483fd0a558SYan, Zheng { 46490b246afaSJeff Mahoney struct btrfs_fs_info *fs_info = root->fs_info; 46503fd0a558SYan, Zheng struct reloc_control *rc; 46513fd0a558SYan, Zheng struct backref_node *node; 46523fd0a558SYan, Zheng int first_cow = 0; 46533fd0a558SYan, Zheng int level; 465483d4cfd4SJosef Bacik int ret = 0; 46553fd0a558SYan, Zheng 46560b246afaSJeff Mahoney rc = fs_info->reloc_ctl; 46573fd0a558SYan, Zheng if (!rc) 465883d4cfd4SJosef Bacik return 0; 46593fd0a558SYan, Zheng 46603fd0a558SYan, Zheng BUG_ON(rc->stage == UPDATE_DATA_PTRS && 46613fd0a558SYan, Zheng root->root_key.objectid == BTRFS_DATA_RELOC_TREE_OBJECTID); 46623fd0a558SYan, Zheng 4663c974c464SWang Shilong if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) { 4664c974c464SWang Shilong if (buf == root->node) 4665c974c464SWang Shilong __update_reloc_root(root, cow->start); 4666c974c464SWang Shilong } 4667c974c464SWang Shilong 46683fd0a558SYan, Zheng level = btrfs_header_level(buf); 46693fd0a558SYan, Zheng if (btrfs_header_generation(buf) <= 46703fd0a558SYan, Zheng btrfs_root_last_snapshot(&root->root_item)) 46713fd0a558SYan, Zheng first_cow = 1; 46723fd0a558SYan, Zheng 46733fd0a558SYan, Zheng if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID && 46743fd0a558SYan, Zheng rc->create_reloc_tree) { 46753fd0a558SYan, Zheng WARN_ON(!first_cow && level == 0); 46763fd0a558SYan, Zheng 46773fd0a558SYan, Zheng node = rc->backref_cache.path[level]; 46783fd0a558SYan, Zheng BUG_ON(node->bytenr != buf->start && 46793fd0a558SYan, Zheng node->new_bytenr != buf->start); 46803fd0a558SYan, Zheng 46813fd0a558SYan, Zheng drop_node_buffer(node); 46823fd0a558SYan, Zheng extent_buffer_get(cow); 46833fd0a558SYan, Zheng node->eb = cow; 46843fd0a558SYan, Zheng node->new_bytenr = cow->start; 46853fd0a558SYan, Zheng 46863fd0a558SYan, Zheng if (!node->pending) { 46873fd0a558SYan, Zheng list_move_tail(&node->list, 46883fd0a558SYan, Zheng &rc->backref_cache.pending[level]); 46893fd0a558SYan, Zheng node->pending = 1; 46903fd0a558SYan, Zheng } 46913fd0a558SYan, Zheng 46923fd0a558SYan, Zheng if (first_cow) 46933fd0a558SYan, Zheng __mark_block_processed(rc, node); 46943fd0a558SYan, Zheng 46953fd0a558SYan, Zheng if (first_cow && level > 0) 46963fd0a558SYan, Zheng rc->nodes_relocated += buf->len; 46973fd0a558SYan, Zheng } 46983fd0a558SYan, Zheng 469983d4cfd4SJosef Bacik if (level == 0 && first_cow && rc->stage == UPDATE_DATA_PTRS) 47003fd0a558SYan, Zheng ret = replace_file_extents(trans, rc, root, cow); 470183d4cfd4SJosef Bacik return ret; 47023fd0a558SYan, Zheng } 47033fd0a558SYan, Zheng 47043fd0a558SYan, Zheng /* 47053fd0a558SYan, Zheng * called before creating snapshot. it calculates metadata reservation 470601327610SNicholas D Steeves * required for relocating tree blocks in the snapshot 47073fd0a558SYan, Zheng */ 4708147d256eSZhaolei void btrfs_reloc_pre_snapshot(struct btrfs_pending_snapshot *pending, 47093fd0a558SYan, Zheng u64 *bytes_to_reserve) 47103fd0a558SYan, Zheng { 471110995c04SQu Wenruo struct btrfs_root *root = pending->root; 471210995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 47133fd0a558SYan, Zheng 471410995c04SQu Wenruo if (!root->reloc_root || !rc) 47153fd0a558SYan, Zheng return; 47163fd0a558SYan, Zheng 47173fd0a558SYan, Zheng if (!rc->merge_reloc_tree) 47183fd0a558SYan, Zheng return; 47193fd0a558SYan, Zheng 47203fd0a558SYan, Zheng root = root->reloc_root; 47213fd0a558SYan, Zheng BUG_ON(btrfs_root_refs(&root->root_item) == 0); 47223fd0a558SYan, Zheng /* 47233fd0a558SYan, Zheng * relocation is in the stage of merging trees. the space 47243fd0a558SYan, Zheng * used by merging a reloc tree is twice the size of 47253fd0a558SYan, Zheng * relocated tree nodes in the worst case. half for cowing 47263fd0a558SYan, Zheng * the reloc tree, half for cowing the fs tree. the space 47273fd0a558SYan, Zheng * used by cowing the reloc tree will be freed after the 47283fd0a558SYan, Zheng * tree is dropped. if we create snapshot, cowing the fs 47293fd0a558SYan, Zheng * tree may use more space than it frees. so we need 47303fd0a558SYan, Zheng * reserve extra space. 47313fd0a558SYan, Zheng */ 47323fd0a558SYan, Zheng *bytes_to_reserve += rc->nodes_relocated; 47333fd0a558SYan, Zheng } 47343fd0a558SYan, Zheng 47353fd0a558SYan, Zheng /* 47363fd0a558SYan, Zheng * called after snapshot is created. migrate block reservation 47373fd0a558SYan, Zheng * and create reloc root for the newly created snapshot 47383fd0a558SYan, Zheng */ 473949b25e05SJeff Mahoney int btrfs_reloc_post_snapshot(struct btrfs_trans_handle *trans, 47403fd0a558SYan, Zheng struct btrfs_pending_snapshot *pending) 47413fd0a558SYan, Zheng { 47423fd0a558SYan, Zheng struct btrfs_root *root = pending->root; 47433fd0a558SYan, Zheng struct btrfs_root *reloc_root; 47443fd0a558SYan, Zheng struct btrfs_root *new_root; 474510995c04SQu Wenruo struct reloc_control *rc = root->fs_info->reloc_ctl; 47463fd0a558SYan, Zheng int ret; 47473fd0a558SYan, Zheng 474810995c04SQu Wenruo if (!root->reloc_root || !rc) 474949b25e05SJeff Mahoney return 0; 47503fd0a558SYan, Zheng 47513fd0a558SYan, Zheng rc = root->fs_info->reloc_ctl; 47523fd0a558SYan, Zheng rc->merging_rsv_size += rc->nodes_relocated; 47533fd0a558SYan, Zheng 47543fd0a558SYan, Zheng if (rc->merge_reloc_tree) { 47553fd0a558SYan, Zheng ret = btrfs_block_rsv_migrate(&pending->block_rsv, 47563fd0a558SYan, Zheng rc->block_rsv, 47573a584174SLu Fengqi rc->nodes_relocated, true); 475849b25e05SJeff Mahoney if (ret) 475949b25e05SJeff Mahoney return ret; 47603fd0a558SYan, Zheng } 47613fd0a558SYan, Zheng 47623fd0a558SYan, Zheng new_root = pending->snap; 47633fd0a558SYan, Zheng reloc_root = create_reloc_root(trans, root->reloc_root, 47643fd0a558SYan, Zheng new_root->root_key.objectid); 476549b25e05SJeff Mahoney if (IS_ERR(reloc_root)) 476649b25e05SJeff Mahoney return PTR_ERR(reloc_root); 47673fd0a558SYan, Zheng 4768ffd7b339SJeff Mahoney ret = __add_reloc_root(reloc_root); 4769ffd7b339SJeff Mahoney BUG_ON(ret < 0); 47703fd0a558SYan, Zheng new_root->reloc_root = reloc_root; 47713fd0a558SYan, Zheng 477249b25e05SJeff Mahoney if (rc->create_reloc_tree) 47733fd0a558SYan, Zheng ret = clone_backref_node(trans, rc, root, reloc_root); 477449b25e05SJeff Mahoney return ret; 47753fd0a558SYan, Zheng } 4776